1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Linux Socket Filter - Kernel level socket filtering
4 *
5 * Based on the design of the Berkeley Packet Filter. The new
6 * internal format has been designed by PLUMgrid:
7 *
8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
9 *
10 * Authors:
11 *
12 * Jay Schulist <jschlst@samba.org>
13 * Alexei Starovoitov <ast@plumgrid.com>
14 * Daniel Borkmann <dborkman@redhat.com>
15 *
16 * Andi Kleen - Fix a few bad bugs and races.
17 * Kris Katterjohn - Added many additional checks in bpf_check_classic()
18 */
19
20 #include <linux/atomic.h>
21 #include <linux/bpf_verifier.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
24 #include <linux/mm.h>
25 #include <linux/fcntl.h>
26 #include <linux/socket.h>
27 #include <linux/sock_diag.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_packet.h>
32 #include <linux/if_arp.h>
33 #include <linux/gfp.h>
34 #include <net/inet_common.h>
35 #include <net/ip.h>
36 #include <net/protocol.h>
37 #include <net/netlink.h>
38 #include <linux/skbuff.h>
39 #include <linux/skmsg.h>
40 #include <net/sock.h>
41 #include <net/flow_dissector.h>
42 #include <linux/errno.h>
43 #include <linux/timer.h>
44 #include <linux/uaccess.h>
45 #include <asm/unaligned.h>
46 #include <linux/filter.h>
47 #include <linux/ratelimit.h>
48 #include <linux/seccomp.h>
49 #include <linux/if_vlan.h>
50 #include <linux/bpf.h>
51 #include <linux/btf.h>
52 #include <net/sch_generic.h>
53 #include <net/cls_cgroup.h>
54 #include <net/dst_metadata.h>
55 #include <net/dst.h>
56 #include <net/sock_reuseport.h>
57 #include <net/busy_poll.h>
58 #include <net/tcp.h>
59 #include <net/xfrm.h>
60 #include <net/udp.h>
61 #include <linux/bpf_trace.h>
62 #include <net/xdp_sock.h>
63 #include <linux/inetdevice.h>
64 #include <net/inet_hashtables.h>
65 #include <net/inet6_hashtables.h>
66 #include <net/ip_fib.h>
67 #include <net/nexthop.h>
68 #include <net/flow.h>
69 #include <net/arp.h>
70 #include <net/ipv6.h>
71 #include <net/net_namespace.h>
72 #include <linux/seg6_local.h>
73 #include <net/seg6.h>
74 #include <net/seg6_local.h>
75 #include <net/lwtunnel.h>
76 #include <net/ipv6_stubs.h>
77 #include <net/bpf_sk_storage.h>
78 #include <net/transp_v6.h>
79 #include <linux/btf_ids.h>
80 #include <net/tls.h>
81 #include <net/xdp.h>
82 #include <net/mptcp.h>
83 #include <net/netfilter/nf_conntrack_bpf.h>
84 #include <net/netkit.h>
85 #include <linux/un.h>
86 #include <net/xdp_sock_drv.h>
87
88 #include "dev.h"
89
90 /* Keep the struct bpf_fib_lookup small so that it fits into a cacheline */
91 static_assert(sizeof(struct bpf_fib_lookup) == 64, "struct bpf_fib_lookup size check");
92
93 static const struct bpf_func_proto *
94 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog);
95
copy_bpf_fprog_from_user(struct sock_fprog * dst,sockptr_t src,int len)96 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
97 {
98 if (in_compat_syscall()) {
99 struct compat_sock_fprog f32;
100
101 if (len != sizeof(f32))
102 return -EINVAL;
103 if (copy_from_sockptr(&f32, src, sizeof(f32)))
104 return -EFAULT;
105 memset(dst, 0, sizeof(*dst));
106 dst->len = f32.len;
107 dst->filter = compat_ptr(f32.filter);
108 } else {
109 if (len != sizeof(*dst))
110 return -EINVAL;
111 if (copy_from_sockptr(dst, src, sizeof(*dst)))
112 return -EFAULT;
113 }
114
115 return 0;
116 }
117 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
118
119 /**
120 * sk_filter_trim_cap - run a packet through a socket filter
121 * @sk: sock associated with &sk_buff
122 * @skb: buffer to filter
123 * @cap: limit on how short the eBPF program may trim the packet
124 *
125 * Run the eBPF program and then cut skb->data to correct size returned by
126 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
127 * than pkt_len we keep whole skb->data. This is the socket level
128 * wrapper to bpf_prog_run. It returns 0 if the packet should
129 * be accepted or -EPERM if the packet should be tossed.
130 *
131 */
sk_filter_trim_cap(struct sock * sk,struct sk_buff * skb,unsigned int cap)132 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
133 {
134 int err;
135 struct sk_filter *filter;
136
137 /*
138 * If the skb was allocated from pfmemalloc reserves, only
139 * allow SOCK_MEMALLOC sockets to use it as this socket is
140 * helping free memory
141 */
142 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
143 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
144 return -ENOMEM;
145 }
146 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
147 if (err)
148 return err;
149
150 err = security_sock_rcv_skb(sk, skb);
151 if (err)
152 return err;
153
154 rcu_read_lock();
155 filter = rcu_dereference(sk->sk_filter);
156 if (filter) {
157 struct sock *save_sk = skb->sk;
158 unsigned int pkt_len;
159
160 skb->sk = sk;
161 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
162 skb->sk = save_sk;
163 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
164 }
165 rcu_read_unlock();
166
167 return err;
168 }
169 EXPORT_SYMBOL(sk_filter_trim_cap);
170
BPF_CALL_1(bpf_skb_get_pay_offset,struct sk_buff *,skb)171 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
172 {
173 return skb_get_poff(skb);
174 }
175
BPF_CALL_3(bpf_skb_get_nlattr,struct sk_buff *,skb,u32,a,u32,x)176 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
177 {
178 struct nlattr *nla;
179
180 if (skb_is_nonlinear(skb))
181 return 0;
182
183 if (skb->len < sizeof(struct nlattr))
184 return 0;
185
186 if (a > skb->len - sizeof(struct nlattr))
187 return 0;
188
189 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
190 if (nla)
191 return (void *) nla - (void *) skb->data;
192
193 return 0;
194 }
195
BPF_CALL_3(bpf_skb_get_nlattr_nest,struct sk_buff *,skb,u32,a,u32,x)196 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
197 {
198 struct nlattr *nla;
199
200 if (skb_is_nonlinear(skb))
201 return 0;
202
203 if (skb->len < sizeof(struct nlattr))
204 return 0;
205
206 if (a > skb->len - sizeof(struct nlattr))
207 return 0;
208
209 nla = (struct nlattr *) &skb->data[a];
210 if (!nla_ok(nla, skb->len - a))
211 return 0;
212
213 nla = nla_find_nested(nla, x);
214 if (nla)
215 return (void *) nla - (void *) skb->data;
216
217 return 0;
218 }
219
BPF_CALL_4(bpf_skb_load_helper_8,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)220 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
221 data, int, headlen, int, offset)
222 {
223 u8 tmp, *ptr;
224 const int len = sizeof(tmp);
225
226 if (offset >= 0) {
227 if (headlen - offset >= len)
228 return *(u8 *)(data + offset);
229 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
230 return tmp;
231 } else {
232 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
233 if (likely(ptr))
234 return *(u8 *)ptr;
235 }
236
237 return -EFAULT;
238 }
239
BPF_CALL_2(bpf_skb_load_helper_8_no_cache,const struct sk_buff *,skb,int,offset)240 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
241 int, offset)
242 {
243 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
244 offset);
245 }
246
BPF_CALL_4(bpf_skb_load_helper_16,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)247 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
248 data, int, headlen, int, offset)
249 {
250 __be16 tmp, *ptr;
251 const int len = sizeof(tmp);
252
253 if (offset >= 0) {
254 if (headlen - offset >= len)
255 return get_unaligned_be16(data + offset);
256 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
257 return be16_to_cpu(tmp);
258 } else {
259 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
260 if (likely(ptr))
261 return get_unaligned_be16(ptr);
262 }
263
264 return -EFAULT;
265 }
266
BPF_CALL_2(bpf_skb_load_helper_16_no_cache,const struct sk_buff *,skb,int,offset)267 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
268 int, offset)
269 {
270 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
271 offset);
272 }
273
BPF_CALL_4(bpf_skb_load_helper_32,const struct sk_buff *,skb,const void *,data,int,headlen,int,offset)274 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
275 data, int, headlen, int, offset)
276 {
277 __be32 tmp, *ptr;
278 const int len = sizeof(tmp);
279
280 if (likely(offset >= 0)) {
281 if (headlen - offset >= len)
282 return get_unaligned_be32(data + offset);
283 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
284 return be32_to_cpu(tmp);
285 } else {
286 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
287 if (likely(ptr))
288 return get_unaligned_be32(ptr);
289 }
290
291 return -EFAULT;
292 }
293
BPF_CALL_2(bpf_skb_load_helper_32_no_cache,const struct sk_buff *,skb,int,offset)294 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
295 int, offset)
296 {
297 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
298 offset);
299 }
300
convert_skb_access(int skb_field,int dst_reg,int src_reg,struct bpf_insn * insn_buf)301 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
302 struct bpf_insn *insn_buf)
303 {
304 struct bpf_insn *insn = insn_buf;
305
306 switch (skb_field) {
307 case SKF_AD_MARK:
308 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
309
310 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
311 offsetof(struct sk_buff, mark));
312 break;
313
314 case SKF_AD_PKTTYPE:
315 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
316 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
317 #ifdef __BIG_ENDIAN_BITFIELD
318 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
319 #endif
320 break;
321
322 case SKF_AD_QUEUE:
323 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
324
325 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
326 offsetof(struct sk_buff, queue_mapping));
327 break;
328
329 case SKF_AD_VLAN_TAG:
330 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
331
332 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
333 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
334 offsetof(struct sk_buff, vlan_tci));
335 break;
336 case SKF_AD_VLAN_TAG_PRESENT:
337 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
338 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
339 offsetof(struct sk_buff, vlan_all));
340 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
341 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1);
342 break;
343 }
344
345 return insn - insn_buf;
346 }
347
convert_bpf_extensions(struct sock_filter * fp,struct bpf_insn ** insnp)348 static bool convert_bpf_extensions(struct sock_filter *fp,
349 struct bpf_insn **insnp)
350 {
351 struct bpf_insn *insn = *insnp;
352 u32 cnt;
353
354 switch (fp->k) {
355 case SKF_AD_OFF + SKF_AD_PROTOCOL:
356 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
357
358 /* A = *(u16 *) (CTX + offsetof(protocol)) */
359 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
360 offsetof(struct sk_buff, protocol));
361 /* A = ntohs(A) [emitting a nop or swap16] */
362 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
363 break;
364
365 case SKF_AD_OFF + SKF_AD_PKTTYPE:
366 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
367 insn += cnt - 1;
368 break;
369
370 case SKF_AD_OFF + SKF_AD_IFINDEX:
371 case SKF_AD_OFF + SKF_AD_HATYPE:
372 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
373 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
374
375 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
376 BPF_REG_TMP, BPF_REG_CTX,
377 offsetof(struct sk_buff, dev));
378 /* if (tmp != 0) goto pc + 1 */
379 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
380 *insn++ = BPF_EXIT_INSN();
381 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
382 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
383 offsetof(struct net_device, ifindex));
384 else
385 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
386 offsetof(struct net_device, type));
387 break;
388
389 case SKF_AD_OFF + SKF_AD_MARK:
390 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
391 insn += cnt - 1;
392 break;
393
394 case SKF_AD_OFF + SKF_AD_RXHASH:
395 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
396
397 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
398 offsetof(struct sk_buff, hash));
399 break;
400
401 case SKF_AD_OFF + SKF_AD_QUEUE:
402 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
403 insn += cnt - 1;
404 break;
405
406 case SKF_AD_OFF + SKF_AD_VLAN_TAG:
407 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
408 BPF_REG_A, BPF_REG_CTX, insn);
409 insn += cnt - 1;
410 break;
411
412 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
413 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
414 BPF_REG_A, BPF_REG_CTX, insn);
415 insn += cnt - 1;
416 break;
417
418 case SKF_AD_OFF + SKF_AD_VLAN_TPID:
419 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
420
421 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
422 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
423 offsetof(struct sk_buff, vlan_proto));
424 /* A = ntohs(A) [emitting a nop or swap16] */
425 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
426 break;
427
428 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
429 case SKF_AD_OFF + SKF_AD_NLATTR:
430 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
431 case SKF_AD_OFF + SKF_AD_CPU:
432 case SKF_AD_OFF + SKF_AD_RANDOM:
433 /* arg1 = CTX */
434 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
435 /* arg2 = A */
436 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
437 /* arg3 = X */
438 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
439 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
440 switch (fp->k) {
441 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
442 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
443 break;
444 case SKF_AD_OFF + SKF_AD_NLATTR:
445 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
446 break;
447 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
448 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
449 break;
450 case SKF_AD_OFF + SKF_AD_CPU:
451 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
452 break;
453 case SKF_AD_OFF + SKF_AD_RANDOM:
454 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
455 bpf_user_rnd_init_once();
456 break;
457 }
458 break;
459
460 case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
461 /* A ^= X */
462 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
463 break;
464
465 default:
466 /* This is just a dummy call to avoid letting the compiler
467 * evict __bpf_call_base() as an optimization. Placed here
468 * where no-one bothers.
469 */
470 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
471 return false;
472 }
473
474 *insnp = insn;
475 return true;
476 }
477
convert_bpf_ld_abs(struct sock_filter * fp,struct bpf_insn ** insnp)478 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
479 {
480 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
481 int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
482 bool endian = BPF_SIZE(fp->code) == BPF_H ||
483 BPF_SIZE(fp->code) == BPF_W;
484 bool indirect = BPF_MODE(fp->code) == BPF_IND;
485 const int ip_align = NET_IP_ALIGN;
486 struct bpf_insn *insn = *insnp;
487 int offset = fp->k;
488
489 if (!indirect &&
490 ((unaligned_ok && offset >= 0) ||
491 (!unaligned_ok && offset >= 0 &&
492 offset + ip_align >= 0 &&
493 offset + ip_align % size == 0))) {
494 bool ldx_off_ok = offset <= S16_MAX;
495
496 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
497 if (offset)
498 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
499 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
500 size, 2 + endian + (!ldx_off_ok * 2));
501 if (ldx_off_ok) {
502 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
503 BPF_REG_D, offset);
504 } else {
505 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
506 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
507 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
508 BPF_REG_TMP, 0);
509 }
510 if (endian)
511 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
512 *insn++ = BPF_JMP_A(8);
513 }
514
515 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
516 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
517 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
518 if (!indirect) {
519 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
520 } else {
521 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
522 if (fp->k)
523 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
524 }
525
526 switch (BPF_SIZE(fp->code)) {
527 case BPF_B:
528 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
529 break;
530 case BPF_H:
531 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
532 break;
533 case BPF_W:
534 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
535 break;
536 default:
537 return false;
538 }
539
540 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
541 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
542 *insn = BPF_EXIT_INSN();
543
544 *insnp = insn;
545 return true;
546 }
547
548 /**
549 * bpf_convert_filter - convert filter program
550 * @prog: the user passed filter program
551 * @len: the length of the user passed filter program
552 * @new_prog: allocated 'struct bpf_prog' or NULL
553 * @new_len: pointer to store length of converted program
554 * @seen_ld_abs: bool whether we've seen ld_abs/ind
555 *
556 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
557 * style extended BPF (eBPF).
558 * Conversion workflow:
559 *
560 * 1) First pass for calculating the new program length:
561 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
562 *
563 * 2) 2nd pass to remap in two passes: 1st pass finds new
564 * jump offsets, 2nd pass remapping:
565 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
566 */
bpf_convert_filter(struct sock_filter * prog,int len,struct bpf_prog * new_prog,int * new_len,bool * seen_ld_abs)567 static int bpf_convert_filter(struct sock_filter *prog, int len,
568 struct bpf_prog *new_prog, int *new_len,
569 bool *seen_ld_abs)
570 {
571 int new_flen = 0, pass = 0, target, i, stack_off;
572 struct bpf_insn *new_insn, *first_insn = NULL;
573 struct sock_filter *fp;
574 int *addrs = NULL;
575 u8 bpf_src;
576
577 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
578 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
579
580 if (len <= 0 || len > BPF_MAXINSNS)
581 return -EINVAL;
582
583 if (new_prog) {
584 first_insn = new_prog->insnsi;
585 addrs = kcalloc(len, sizeof(*addrs),
586 GFP_KERNEL | __GFP_NOWARN);
587 if (!addrs)
588 return -ENOMEM;
589 }
590
591 do_pass:
592 new_insn = first_insn;
593 fp = prog;
594
595 /* Classic BPF related prologue emission. */
596 if (new_prog) {
597 /* Classic BPF expects A and X to be reset first. These need
598 * to be guaranteed to be the first two instructions.
599 */
600 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
601 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
602
603 /* All programs must keep CTX in callee saved BPF_REG_CTX.
604 * In eBPF case it's done by the compiler, here we need to
605 * do this ourself. Initial CTX is present in BPF_REG_ARG1.
606 */
607 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
608 if (*seen_ld_abs) {
609 /* For packet access in classic BPF, cache skb->data
610 * in callee-saved BPF R8 and skb->len - skb->data_len
611 * (headlen) in BPF R9. Since classic BPF is read-only
612 * on CTX, we only need to cache it once.
613 */
614 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
615 BPF_REG_D, BPF_REG_CTX,
616 offsetof(struct sk_buff, data));
617 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
618 offsetof(struct sk_buff, len));
619 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
620 offsetof(struct sk_buff, data_len));
621 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
622 }
623 } else {
624 new_insn += 3;
625 }
626
627 for (i = 0; i < len; fp++, i++) {
628 struct bpf_insn tmp_insns[32] = { };
629 struct bpf_insn *insn = tmp_insns;
630
631 if (addrs)
632 addrs[i] = new_insn - first_insn;
633
634 switch (fp->code) {
635 /* All arithmetic insns and skb loads map as-is. */
636 case BPF_ALU | BPF_ADD | BPF_X:
637 case BPF_ALU | BPF_ADD | BPF_K:
638 case BPF_ALU | BPF_SUB | BPF_X:
639 case BPF_ALU | BPF_SUB | BPF_K:
640 case BPF_ALU | BPF_AND | BPF_X:
641 case BPF_ALU | BPF_AND | BPF_K:
642 case BPF_ALU | BPF_OR | BPF_X:
643 case BPF_ALU | BPF_OR | BPF_K:
644 case BPF_ALU | BPF_LSH | BPF_X:
645 case BPF_ALU | BPF_LSH | BPF_K:
646 case BPF_ALU | BPF_RSH | BPF_X:
647 case BPF_ALU | BPF_RSH | BPF_K:
648 case BPF_ALU | BPF_XOR | BPF_X:
649 case BPF_ALU | BPF_XOR | BPF_K:
650 case BPF_ALU | BPF_MUL | BPF_X:
651 case BPF_ALU | BPF_MUL | BPF_K:
652 case BPF_ALU | BPF_DIV | BPF_X:
653 case BPF_ALU | BPF_DIV | BPF_K:
654 case BPF_ALU | BPF_MOD | BPF_X:
655 case BPF_ALU | BPF_MOD | BPF_K:
656 case BPF_ALU | BPF_NEG:
657 case BPF_LD | BPF_ABS | BPF_W:
658 case BPF_LD | BPF_ABS | BPF_H:
659 case BPF_LD | BPF_ABS | BPF_B:
660 case BPF_LD | BPF_IND | BPF_W:
661 case BPF_LD | BPF_IND | BPF_H:
662 case BPF_LD | BPF_IND | BPF_B:
663 /* Check for overloaded BPF extension and
664 * directly convert it if found, otherwise
665 * just move on with mapping.
666 */
667 if (BPF_CLASS(fp->code) == BPF_LD &&
668 BPF_MODE(fp->code) == BPF_ABS &&
669 convert_bpf_extensions(fp, &insn))
670 break;
671 if (BPF_CLASS(fp->code) == BPF_LD &&
672 convert_bpf_ld_abs(fp, &insn)) {
673 *seen_ld_abs = true;
674 break;
675 }
676
677 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
678 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
679 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
680 /* Error with exception code on div/mod by 0.
681 * For cBPF programs, this was always return 0.
682 */
683 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
684 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
685 *insn++ = BPF_EXIT_INSN();
686 }
687
688 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
689 break;
690
691 /* Jump transformation cannot use BPF block macros
692 * everywhere as offset calculation and target updates
693 * require a bit more work than the rest, i.e. jump
694 * opcodes map as-is, but offsets need adjustment.
695 */
696
697 #define BPF_EMIT_JMP \
698 do { \
699 const s32 off_min = S16_MIN, off_max = S16_MAX; \
700 s32 off; \
701 \
702 if (target >= len || target < 0) \
703 goto err; \
704 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \
705 /* Adjust pc relative offset for 2nd or 3rd insn. */ \
706 off -= insn - tmp_insns; \
707 /* Reject anything not fitting into insn->off. */ \
708 if (off < off_min || off > off_max) \
709 goto err; \
710 insn->off = off; \
711 } while (0)
712
713 case BPF_JMP | BPF_JA:
714 target = i + fp->k + 1;
715 insn->code = fp->code;
716 BPF_EMIT_JMP;
717 break;
718
719 case BPF_JMP | BPF_JEQ | BPF_K:
720 case BPF_JMP | BPF_JEQ | BPF_X:
721 case BPF_JMP | BPF_JSET | BPF_K:
722 case BPF_JMP | BPF_JSET | BPF_X:
723 case BPF_JMP | BPF_JGT | BPF_K:
724 case BPF_JMP | BPF_JGT | BPF_X:
725 case BPF_JMP | BPF_JGE | BPF_K:
726 case BPF_JMP | BPF_JGE | BPF_X:
727 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
728 /* BPF immediates are signed, zero extend
729 * immediate into tmp register and use it
730 * in compare insn.
731 */
732 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
733
734 insn->dst_reg = BPF_REG_A;
735 insn->src_reg = BPF_REG_TMP;
736 bpf_src = BPF_X;
737 } else {
738 insn->dst_reg = BPF_REG_A;
739 insn->imm = fp->k;
740 bpf_src = BPF_SRC(fp->code);
741 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
742 }
743
744 /* Common case where 'jump_false' is next insn. */
745 if (fp->jf == 0) {
746 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
747 target = i + fp->jt + 1;
748 BPF_EMIT_JMP;
749 break;
750 }
751
752 /* Convert some jumps when 'jump_true' is next insn. */
753 if (fp->jt == 0) {
754 switch (BPF_OP(fp->code)) {
755 case BPF_JEQ:
756 insn->code = BPF_JMP | BPF_JNE | bpf_src;
757 break;
758 case BPF_JGT:
759 insn->code = BPF_JMP | BPF_JLE | bpf_src;
760 break;
761 case BPF_JGE:
762 insn->code = BPF_JMP | BPF_JLT | bpf_src;
763 break;
764 default:
765 goto jmp_rest;
766 }
767
768 target = i + fp->jf + 1;
769 BPF_EMIT_JMP;
770 break;
771 }
772 jmp_rest:
773 /* Other jumps are mapped into two insns: Jxx and JA. */
774 target = i + fp->jt + 1;
775 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
776 BPF_EMIT_JMP;
777 insn++;
778
779 insn->code = BPF_JMP | BPF_JA;
780 target = i + fp->jf + 1;
781 BPF_EMIT_JMP;
782 break;
783
784 /* ldxb 4 * ([14] & 0xf) is remapped into 6 insns. */
785 case BPF_LDX | BPF_MSH | BPF_B: {
786 struct sock_filter tmp = {
787 .code = BPF_LD | BPF_ABS | BPF_B,
788 .k = fp->k,
789 };
790
791 *seen_ld_abs = true;
792
793 /* X = A */
794 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
795 /* A = BPF_R0 = *(u8 *) (skb->data + K) */
796 convert_bpf_ld_abs(&tmp, &insn);
797 insn++;
798 /* A &= 0xf */
799 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
800 /* A <<= 2 */
801 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
802 /* tmp = X */
803 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
804 /* X = A */
805 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
806 /* A = tmp */
807 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
808 break;
809 }
810 /* RET_K is remapped into 2 insns. RET_A case doesn't need an
811 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
812 */
813 case BPF_RET | BPF_A:
814 case BPF_RET | BPF_K:
815 if (BPF_RVAL(fp->code) == BPF_K)
816 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
817 0, fp->k);
818 *insn = BPF_EXIT_INSN();
819 break;
820
821 /* Store to stack. */
822 case BPF_ST:
823 case BPF_STX:
824 stack_off = fp->k * 4 + 4;
825 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
826 BPF_ST ? BPF_REG_A : BPF_REG_X,
827 -stack_off);
828 /* check_load_and_stores() verifies that classic BPF can
829 * load from stack only after write, so tracking
830 * stack_depth for ST|STX insns is enough
831 */
832 if (new_prog && new_prog->aux->stack_depth < stack_off)
833 new_prog->aux->stack_depth = stack_off;
834 break;
835
836 /* Load from stack. */
837 case BPF_LD | BPF_MEM:
838 case BPF_LDX | BPF_MEM:
839 stack_off = fp->k * 4 + 4;
840 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
841 BPF_REG_A : BPF_REG_X, BPF_REG_FP,
842 -stack_off);
843 break;
844
845 /* A = K or X = K */
846 case BPF_LD | BPF_IMM:
847 case BPF_LDX | BPF_IMM:
848 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
849 BPF_REG_A : BPF_REG_X, fp->k);
850 break;
851
852 /* X = A */
853 case BPF_MISC | BPF_TAX:
854 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
855 break;
856
857 /* A = X */
858 case BPF_MISC | BPF_TXA:
859 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
860 break;
861
862 /* A = skb->len or X = skb->len */
863 case BPF_LD | BPF_W | BPF_LEN:
864 case BPF_LDX | BPF_W | BPF_LEN:
865 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
866 BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
867 offsetof(struct sk_buff, len));
868 break;
869
870 /* Access seccomp_data fields. */
871 case BPF_LDX | BPF_ABS | BPF_W:
872 /* A = *(u32 *) (ctx + K) */
873 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
874 break;
875
876 /* Unknown instruction. */
877 default:
878 goto err;
879 }
880
881 insn++;
882 if (new_prog)
883 memcpy(new_insn, tmp_insns,
884 sizeof(*insn) * (insn - tmp_insns));
885 new_insn += insn - tmp_insns;
886 }
887
888 if (!new_prog) {
889 /* Only calculating new length. */
890 *new_len = new_insn - first_insn;
891 if (*seen_ld_abs)
892 *new_len += 4; /* Prologue bits. */
893 return 0;
894 }
895
896 pass++;
897 if (new_flen != new_insn - first_insn) {
898 new_flen = new_insn - first_insn;
899 if (pass > 2)
900 goto err;
901 goto do_pass;
902 }
903
904 kfree(addrs);
905 BUG_ON(*new_len != new_flen);
906 return 0;
907 err:
908 kfree(addrs);
909 return -EINVAL;
910 }
911
912 /* Security:
913 *
914 * As we dont want to clear mem[] array for each packet going through
915 * __bpf_prog_run(), we check that filter loaded by user never try to read
916 * a cell if not previously written, and we check all branches to be sure
917 * a malicious user doesn't try to abuse us.
918 */
check_load_and_stores(const struct sock_filter * filter,int flen)919 static int check_load_and_stores(const struct sock_filter *filter, int flen)
920 {
921 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
922 int pc, ret = 0;
923
924 BUILD_BUG_ON(BPF_MEMWORDS > 16);
925
926 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
927 if (!masks)
928 return -ENOMEM;
929
930 memset(masks, 0xff, flen * sizeof(*masks));
931
932 for (pc = 0; pc < flen; pc++) {
933 memvalid &= masks[pc];
934
935 switch (filter[pc].code) {
936 case BPF_ST:
937 case BPF_STX:
938 memvalid |= (1 << filter[pc].k);
939 break;
940 case BPF_LD | BPF_MEM:
941 case BPF_LDX | BPF_MEM:
942 if (!(memvalid & (1 << filter[pc].k))) {
943 ret = -EINVAL;
944 goto error;
945 }
946 break;
947 case BPF_JMP | BPF_JA:
948 /* A jump must set masks on target */
949 masks[pc + 1 + filter[pc].k] &= memvalid;
950 memvalid = ~0;
951 break;
952 case BPF_JMP | BPF_JEQ | BPF_K:
953 case BPF_JMP | BPF_JEQ | BPF_X:
954 case BPF_JMP | BPF_JGE | BPF_K:
955 case BPF_JMP | BPF_JGE | BPF_X:
956 case BPF_JMP | BPF_JGT | BPF_K:
957 case BPF_JMP | BPF_JGT | BPF_X:
958 case BPF_JMP | BPF_JSET | BPF_K:
959 case BPF_JMP | BPF_JSET | BPF_X:
960 /* A jump must set masks on targets */
961 masks[pc + 1 + filter[pc].jt] &= memvalid;
962 masks[pc + 1 + filter[pc].jf] &= memvalid;
963 memvalid = ~0;
964 break;
965 }
966 }
967 error:
968 kfree(masks);
969 return ret;
970 }
971
chk_code_allowed(u16 code_to_probe)972 static bool chk_code_allowed(u16 code_to_probe)
973 {
974 static const bool codes[] = {
975 /* 32 bit ALU operations */
976 [BPF_ALU | BPF_ADD | BPF_K] = true,
977 [BPF_ALU | BPF_ADD | BPF_X] = true,
978 [BPF_ALU | BPF_SUB | BPF_K] = true,
979 [BPF_ALU | BPF_SUB | BPF_X] = true,
980 [BPF_ALU | BPF_MUL | BPF_K] = true,
981 [BPF_ALU | BPF_MUL | BPF_X] = true,
982 [BPF_ALU | BPF_DIV | BPF_K] = true,
983 [BPF_ALU | BPF_DIV | BPF_X] = true,
984 [BPF_ALU | BPF_MOD | BPF_K] = true,
985 [BPF_ALU | BPF_MOD | BPF_X] = true,
986 [BPF_ALU | BPF_AND | BPF_K] = true,
987 [BPF_ALU | BPF_AND | BPF_X] = true,
988 [BPF_ALU | BPF_OR | BPF_K] = true,
989 [BPF_ALU | BPF_OR | BPF_X] = true,
990 [BPF_ALU | BPF_XOR | BPF_K] = true,
991 [BPF_ALU | BPF_XOR | BPF_X] = true,
992 [BPF_ALU | BPF_LSH | BPF_K] = true,
993 [BPF_ALU | BPF_LSH | BPF_X] = true,
994 [BPF_ALU | BPF_RSH | BPF_K] = true,
995 [BPF_ALU | BPF_RSH | BPF_X] = true,
996 [BPF_ALU | BPF_NEG] = true,
997 /* Load instructions */
998 [BPF_LD | BPF_W | BPF_ABS] = true,
999 [BPF_LD | BPF_H | BPF_ABS] = true,
1000 [BPF_LD | BPF_B | BPF_ABS] = true,
1001 [BPF_LD | BPF_W | BPF_LEN] = true,
1002 [BPF_LD | BPF_W | BPF_IND] = true,
1003 [BPF_LD | BPF_H | BPF_IND] = true,
1004 [BPF_LD | BPF_B | BPF_IND] = true,
1005 [BPF_LD | BPF_IMM] = true,
1006 [BPF_LD | BPF_MEM] = true,
1007 [BPF_LDX | BPF_W | BPF_LEN] = true,
1008 [BPF_LDX | BPF_B | BPF_MSH] = true,
1009 [BPF_LDX | BPF_IMM] = true,
1010 [BPF_LDX | BPF_MEM] = true,
1011 /* Store instructions */
1012 [BPF_ST] = true,
1013 [BPF_STX] = true,
1014 /* Misc instructions */
1015 [BPF_MISC | BPF_TAX] = true,
1016 [BPF_MISC | BPF_TXA] = true,
1017 /* Return instructions */
1018 [BPF_RET | BPF_K] = true,
1019 [BPF_RET | BPF_A] = true,
1020 /* Jump instructions */
1021 [BPF_JMP | BPF_JA] = true,
1022 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1023 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1024 [BPF_JMP | BPF_JGE | BPF_K] = true,
1025 [BPF_JMP | BPF_JGE | BPF_X] = true,
1026 [BPF_JMP | BPF_JGT | BPF_K] = true,
1027 [BPF_JMP | BPF_JGT | BPF_X] = true,
1028 [BPF_JMP | BPF_JSET | BPF_K] = true,
1029 [BPF_JMP | BPF_JSET | BPF_X] = true,
1030 };
1031
1032 if (code_to_probe >= ARRAY_SIZE(codes))
1033 return false;
1034
1035 return codes[code_to_probe];
1036 }
1037
bpf_check_basics_ok(const struct sock_filter * filter,unsigned int flen)1038 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1039 unsigned int flen)
1040 {
1041 if (filter == NULL)
1042 return false;
1043 if (flen == 0 || flen > BPF_MAXINSNS)
1044 return false;
1045
1046 return true;
1047 }
1048
1049 /**
1050 * bpf_check_classic - verify socket filter code
1051 * @filter: filter to verify
1052 * @flen: length of filter
1053 *
1054 * Check the user's filter code. If we let some ugly
1055 * filter code slip through kaboom! The filter must contain
1056 * no references or jumps that are out of range, no illegal
1057 * instructions, and must end with a RET instruction.
1058 *
1059 * All jumps are forward as they are not signed.
1060 *
1061 * Returns 0 if the rule set is legal or -EINVAL if not.
1062 */
bpf_check_classic(const struct sock_filter * filter,unsigned int flen)1063 static int bpf_check_classic(const struct sock_filter *filter,
1064 unsigned int flen)
1065 {
1066 bool anc_found;
1067 int pc;
1068
1069 /* Check the filter code now */
1070 for (pc = 0; pc < flen; pc++) {
1071 const struct sock_filter *ftest = &filter[pc];
1072
1073 /* May we actually operate on this code? */
1074 if (!chk_code_allowed(ftest->code))
1075 return -EINVAL;
1076
1077 /* Some instructions need special checks */
1078 switch (ftest->code) {
1079 case BPF_ALU | BPF_DIV | BPF_K:
1080 case BPF_ALU | BPF_MOD | BPF_K:
1081 /* Check for division by zero */
1082 if (ftest->k == 0)
1083 return -EINVAL;
1084 break;
1085 case BPF_ALU | BPF_LSH | BPF_K:
1086 case BPF_ALU | BPF_RSH | BPF_K:
1087 if (ftest->k >= 32)
1088 return -EINVAL;
1089 break;
1090 case BPF_LD | BPF_MEM:
1091 case BPF_LDX | BPF_MEM:
1092 case BPF_ST:
1093 case BPF_STX:
1094 /* Check for invalid memory addresses */
1095 if (ftest->k >= BPF_MEMWORDS)
1096 return -EINVAL;
1097 break;
1098 case BPF_JMP | BPF_JA:
1099 /* Note, the large ftest->k might cause loops.
1100 * Compare this with conditional jumps below,
1101 * where offsets are limited. --ANK (981016)
1102 */
1103 if (ftest->k >= (unsigned int)(flen - pc - 1))
1104 return -EINVAL;
1105 break;
1106 case BPF_JMP | BPF_JEQ | BPF_K:
1107 case BPF_JMP | BPF_JEQ | BPF_X:
1108 case BPF_JMP | BPF_JGE | BPF_K:
1109 case BPF_JMP | BPF_JGE | BPF_X:
1110 case BPF_JMP | BPF_JGT | BPF_K:
1111 case BPF_JMP | BPF_JGT | BPF_X:
1112 case BPF_JMP | BPF_JSET | BPF_K:
1113 case BPF_JMP | BPF_JSET | BPF_X:
1114 /* Both conditionals must be safe */
1115 if (pc + ftest->jt + 1 >= flen ||
1116 pc + ftest->jf + 1 >= flen)
1117 return -EINVAL;
1118 break;
1119 case BPF_LD | BPF_W | BPF_ABS:
1120 case BPF_LD | BPF_H | BPF_ABS:
1121 case BPF_LD | BPF_B | BPF_ABS:
1122 anc_found = false;
1123 if (bpf_anc_helper(ftest) & BPF_ANC)
1124 anc_found = true;
1125 /* Ancillary operation unknown or unsupported */
1126 if (anc_found == false && ftest->k >= SKF_AD_OFF)
1127 return -EINVAL;
1128 }
1129 }
1130
1131 /* Last instruction must be a RET code */
1132 switch (filter[flen - 1].code) {
1133 case BPF_RET | BPF_K:
1134 case BPF_RET | BPF_A:
1135 return check_load_and_stores(filter, flen);
1136 }
1137
1138 return -EINVAL;
1139 }
1140
bpf_prog_store_orig_filter(struct bpf_prog * fp,const struct sock_fprog * fprog)1141 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1142 const struct sock_fprog *fprog)
1143 {
1144 unsigned int fsize = bpf_classic_proglen(fprog);
1145 struct sock_fprog_kern *fkprog;
1146
1147 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1148 if (!fp->orig_prog)
1149 return -ENOMEM;
1150
1151 fkprog = fp->orig_prog;
1152 fkprog->len = fprog->len;
1153
1154 fkprog->filter = kmemdup(fp->insns, fsize,
1155 GFP_KERNEL | __GFP_NOWARN);
1156 if (!fkprog->filter) {
1157 kfree(fp->orig_prog);
1158 return -ENOMEM;
1159 }
1160
1161 return 0;
1162 }
1163
bpf_release_orig_filter(struct bpf_prog * fp)1164 static void bpf_release_orig_filter(struct bpf_prog *fp)
1165 {
1166 struct sock_fprog_kern *fprog = fp->orig_prog;
1167
1168 if (fprog) {
1169 kfree(fprog->filter);
1170 kfree(fprog);
1171 }
1172 }
1173
__bpf_prog_release(struct bpf_prog * prog)1174 static void __bpf_prog_release(struct bpf_prog *prog)
1175 {
1176 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1177 bpf_prog_put(prog);
1178 } else {
1179 bpf_release_orig_filter(prog);
1180 bpf_prog_free(prog);
1181 }
1182 }
1183
__sk_filter_release(struct sk_filter * fp)1184 static void __sk_filter_release(struct sk_filter *fp)
1185 {
1186 __bpf_prog_release(fp->prog);
1187 kfree(fp);
1188 }
1189
1190 /**
1191 * sk_filter_release_rcu - Release a socket filter by rcu_head
1192 * @rcu: rcu_head that contains the sk_filter to free
1193 */
sk_filter_release_rcu(struct rcu_head * rcu)1194 static void sk_filter_release_rcu(struct rcu_head *rcu)
1195 {
1196 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1197
1198 __sk_filter_release(fp);
1199 }
1200
1201 /**
1202 * sk_filter_release - release a socket filter
1203 * @fp: filter to remove
1204 *
1205 * Remove a filter from a socket and release its resources.
1206 */
sk_filter_release(struct sk_filter * fp)1207 static void sk_filter_release(struct sk_filter *fp)
1208 {
1209 if (refcount_dec_and_test(&fp->refcnt))
1210 call_rcu(&fp->rcu, sk_filter_release_rcu);
1211 }
1212
sk_filter_uncharge(struct sock * sk,struct sk_filter * fp)1213 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1214 {
1215 u32 filter_size = bpf_prog_size(fp->prog->len);
1216
1217 atomic_sub(filter_size, &sk->sk_omem_alloc);
1218 sk_filter_release(fp);
1219 }
1220
1221 /* try to charge the socket memory if there is space available
1222 * return true on success
1223 */
__sk_filter_charge(struct sock * sk,struct sk_filter * fp)1224 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1225 {
1226 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1227 u32 filter_size = bpf_prog_size(fp->prog->len);
1228
1229 /* same check as in sock_kmalloc() */
1230 if (filter_size <= optmem_max &&
1231 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1232 atomic_add(filter_size, &sk->sk_omem_alloc);
1233 return true;
1234 }
1235 return false;
1236 }
1237
sk_filter_charge(struct sock * sk,struct sk_filter * fp)1238 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1239 {
1240 if (!refcount_inc_not_zero(&fp->refcnt))
1241 return false;
1242
1243 if (!__sk_filter_charge(sk, fp)) {
1244 sk_filter_release(fp);
1245 return false;
1246 }
1247 return true;
1248 }
1249
bpf_migrate_filter(struct bpf_prog * fp)1250 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1251 {
1252 struct sock_filter *old_prog;
1253 struct bpf_prog *old_fp;
1254 int err, new_len, old_len = fp->len;
1255 bool seen_ld_abs = false;
1256
1257 /* We are free to overwrite insns et al right here as it won't be used at
1258 * this point in time anymore internally after the migration to the eBPF
1259 * instruction representation.
1260 */
1261 BUILD_BUG_ON(sizeof(struct sock_filter) !=
1262 sizeof(struct bpf_insn));
1263
1264 /* Conversion cannot happen on overlapping memory areas,
1265 * so we need to keep the user BPF around until the 2nd
1266 * pass. At this time, the user BPF is stored in fp->insns.
1267 */
1268 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1269 GFP_KERNEL | __GFP_NOWARN);
1270 if (!old_prog) {
1271 err = -ENOMEM;
1272 goto out_err;
1273 }
1274
1275 /* 1st pass: calculate the new program length. */
1276 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1277 &seen_ld_abs);
1278 if (err)
1279 goto out_err_free;
1280
1281 /* Expand fp for appending the new filter representation. */
1282 old_fp = fp;
1283 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1284 if (!fp) {
1285 /* The old_fp is still around in case we couldn't
1286 * allocate new memory, so uncharge on that one.
1287 */
1288 fp = old_fp;
1289 err = -ENOMEM;
1290 goto out_err_free;
1291 }
1292
1293 fp->len = new_len;
1294
1295 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1296 err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1297 &seen_ld_abs);
1298 if (err)
1299 /* 2nd bpf_convert_filter() can fail only if it fails
1300 * to allocate memory, remapping must succeed. Note,
1301 * that at this time old_fp has already been released
1302 * by krealloc().
1303 */
1304 goto out_err_free;
1305
1306 fp = bpf_prog_select_runtime(fp, &err);
1307 if (err)
1308 goto out_err_free;
1309
1310 kfree(old_prog);
1311 return fp;
1312
1313 out_err_free:
1314 kfree(old_prog);
1315 out_err:
1316 __bpf_prog_release(fp);
1317 return ERR_PTR(err);
1318 }
1319
bpf_prepare_filter(struct bpf_prog * fp,bpf_aux_classic_check_t trans)1320 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1321 bpf_aux_classic_check_t trans)
1322 {
1323 int err;
1324
1325 fp->bpf_func = NULL;
1326 fp->jited = 0;
1327
1328 err = bpf_check_classic(fp->insns, fp->len);
1329 if (err) {
1330 __bpf_prog_release(fp);
1331 return ERR_PTR(err);
1332 }
1333
1334 /* There might be additional checks and transformations
1335 * needed on classic filters, f.e. in case of seccomp.
1336 */
1337 if (trans) {
1338 err = trans(fp->insns, fp->len);
1339 if (err) {
1340 __bpf_prog_release(fp);
1341 return ERR_PTR(err);
1342 }
1343 }
1344
1345 /* Probe if we can JIT compile the filter and if so, do
1346 * the compilation of the filter.
1347 */
1348 bpf_jit_compile(fp);
1349
1350 /* JIT compiler couldn't process this filter, so do the eBPF translation
1351 * for the optimized interpreter.
1352 */
1353 if (!fp->jited)
1354 fp = bpf_migrate_filter(fp);
1355
1356 return fp;
1357 }
1358
1359 /**
1360 * bpf_prog_create - create an unattached filter
1361 * @pfp: the unattached filter that is created
1362 * @fprog: the filter program
1363 *
1364 * Create a filter independent of any socket. We first run some
1365 * sanity checks on it to make sure it does not explode on us later.
1366 * If an error occurs or there is insufficient memory for the filter
1367 * a negative errno code is returned. On success the return is zero.
1368 */
bpf_prog_create(struct bpf_prog ** pfp,struct sock_fprog_kern * fprog)1369 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1370 {
1371 unsigned int fsize = bpf_classic_proglen(fprog);
1372 struct bpf_prog *fp;
1373
1374 /* Make sure new filter is there and in the right amounts. */
1375 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1376 return -EINVAL;
1377
1378 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1379 if (!fp)
1380 return -ENOMEM;
1381
1382 memcpy(fp->insns, fprog->filter, fsize);
1383
1384 fp->len = fprog->len;
1385 /* Since unattached filters are not copied back to user
1386 * space through sk_get_filter(), we do not need to hold
1387 * a copy here, and can spare us the work.
1388 */
1389 fp->orig_prog = NULL;
1390
1391 /* bpf_prepare_filter() already takes care of freeing
1392 * memory in case something goes wrong.
1393 */
1394 fp = bpf_prepare_filter(fp, NULL);
1395 if (IS_ERR(fp))
1396 return PTR_ERR(fp);
1397
1398 *pfp = fp;
1399 return 0;
1400 }
1401 EXPORT_SYMBOL_GPL(bpf_prog_create);
1402
1403 /**
1404 * bpf_prog_create_from_user - create an unattached filter from user buffer
1405 * @pfp: the unattached filter that is created
1406 * @fprog: the filter program
1407 * @trans: post-classic verifier transformation handler
1408 * @save_orig: save classic BPF program
1409 *
1410 * This function effectively does the same as bpf_prog_create(), only
1411 * that it builds up its insns buffer from user space provided buffer.
1412 * It also allows for passing a bpf_aux_classic_check_t handler.
1413 */
bpf_prog_create_from_user(struct bpf_prog ** pfp,struct sock_fprog * fprog,bpf_aux_classic_check_t trans,bool save_orig)1414 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1415 bpf_aux_classic_check_t trans, bool save_orig)
1416 {
1417 unsigned int fsize = bpf_classic_proglen(fprog);
1418 struct bpf_prog *fp;
1419 int err;
1420
1421 /* Make sure new filter is there and in the right amounts. */
1422 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1423 return -EINVAL;
1424
1425 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1426 if (!fp)
1427 return -ENOMEM;
1428
1429 if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1430 __bpf_prog_free(fp);
1431 return -EFAULT;
1432 }
1433
1434 fp->len = fprog->len;
1435 fp->orig_prog = NULL;
1436
1437 if (save_orig) {
1438 err = bpf_prog_store_orig_filter(fp, fprog);
1439 if (err) {
1440 __bpf_prog_free(fp);
1441 return -ENOMEM;
1442 }
1443 }
1444
1445 /* bpf_prepare_filter() already takes care of freeing
1446 * memory in case something goes wrong.
1447 */
1448 fp = bpf_prepare_filter(fp, trans);
1449 if (IS_ERR(fp))
1450 return PTR_ERR(fp);
1451
1452 *pfp = fp;
1453 return 0;
1454 }
1455 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1456
bpf_prog_destroy(struct bpf_prog * fp)1457 void bpf_prog_destroy(struct bpf_prog *fp)
1458 {
1459 __bpf_prog_release(fp);
1460 }
1461 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1462
__sk_attach_prog(struct bpf_prog * prog,struct sock * sk)1463 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1464 {
1465 struct sk_filter *fp, *old_fp;
1466
1467 fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1468 if (!fp)
1469 return -ENOMEM;
1470
1471 fp->prog = prog;
1472
1473 if (!__sk_filter_charge(sk, fp)) {
1474 kfree(fp);
1475 return -ENOMEM;
1476 }
1477 refcount_set(&fp->refcnt, 1);
1478
1479 old_fp = rcu_dereference_protected(sk->sk_filter,
1480 lockdep_sock_is_held(sk));
1481 rcu_assign_pointer(sk->sk_filter, fp);
1482
1483 if (old_fp)
1484 sk_filter_uncharge(sk, old_fp);
1485
1486 return 0;
1487 }
1488
1489 static
__get_filter(struct sock_fprog * fprog,struct sock * sk)1490 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1491 {
1492 unsigned int fsize = bpf_classic_proglen(fprog);
1493 struct bpf_prog *prog;
1494 int err;
1495
1496 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1497 return ERR_PTR(-EPERM);
1498
1499 /* Make sure new filter is there and in the right amounts. */
1500 if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1501 return ERR_PTR(-EINVAL);
1502
1503 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1504 if (!prog)
1505 return ERR_PTR(-ENOMEM);
1506
1507 if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1508 __bpf_prog_free(prog);
1509 return ERR_PTR(-EFAULT);
1510 }
1511
1512 prog->len = fprog->len;
1513
1514 err = bpf_prog_store_orig_filter(prog, fprog);
1515 if (err) {
1516 __bpf_prog_free(prog);
1517 return ERR_PTR(-ENOMEM);
1518 }
1519
1520 /* bpf_prepare_filter() already takes care of freeing
1521 * memory in case something goes wrong.
1522 */
1523 return bpf_prepare_filter(prog, NULL);
1524 }
1525
1526 /**
1527 * sk_attach_filter - attach a socket filter
1528 * @fprog: the filter program
1529 * @sk: the socket to use
1530 *
1531 * Attach the user's filter code. We first run some sanity checks on
1532 * it to make sure it does not explode on us later. If an error
1533 * occurs or there is insufficient memory for the filter a negative
1534 * errno code is returned. On success the return is zero.
1535 */
sk_attach_filter(struct sock_fprog * fprog,struct sock * sk)1536 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1537 {
1538 struct bpf_prog *prog = __get_filter(fprog, sk);
1539 int err;
1540
1541 if (IS_ERR(prog))
1542 return PTR_ERR(prog);
1543
1544 err = __sk_attach_prog(prog, sk);
1545 if (err < 0) {
1546 __bpf_prog_release(prog);
1547 return err;
1548 }
1549
1550 return 0;
1551 }
1552 EXPORT_SYMBOL_GPL(sk_attach_filter);
1553
sk_reuseport_attach_filter(struct sock_fprog * fprog,struct sock * sk)1554 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1555 {
1556 struct bpf_prog *prog = __get_filter(fprog, sk);
1557 int err, optmem_max;
1558
1559 if (IS_ERR(prog))
1560 return PTR_ERR(prog);
1561
1562 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1563 if (bpf_prog_size(prog->len) > optmem_max)
1564 err = -ENOMEM;
1565 else
1566 err = reuseport_attach_prog(sk, prog);
1567
1568 if (err)
1569 __bpf_prog_release(prog);
1570
1571 return err;
1572 }
1573
__get_bpf(u32 ufd,struct sock * sk)1574 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1575 {
1576 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1577 return ERR_PTR(-EPERM);
1578
1579 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1580 }
1581
sk_attach_bpf(u32 ufd,struct sock * sk)1582 int sk_attach_bpf(u32 ufd, struct sock *sk)
1583 {
1584 struct bpf_prog *prog = __get_bpf(ufd, sk);
1585 int err;
1586
1587 if (IS_ERR(prog))
1588 return PTR_ERR(prog);
1589
1590 err = __sk_attach_prog(prog, sk);
1591 if (err < 0) {
1592 bpf_prog_put(prog);
1593 return err;
1594 }
1595
1596 return 0;
1597 }
1598
sk_reuseport_attach_bpf(u32 ufd,struct sock * sk)1599 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1600 {
1601 struct bpf_prog *prog;
1602 int err, optmem_max;
1603
1604 if (sock_flag(sk, SOCK_FILTER_LOCKED))
1605 return -EPERM;
1606
1607 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1608 if (PTR_ERR(prog) == -EINVAL)
1609 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1610 if (IS_ERR(prog))
1611 return PTR_ERR(prog);
1612
1613 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1614 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1615 * bpf prog (e.g. sockmap). It depends on the
1616 * limitation imposed by bpf_prog_load().
1617 * Hence, sysctl_optmem_max is not checked.
1618 */
1619 if ((sk->sk_type != SOCK_STREAM &&
1620 sk->sk_type != SOCK_DGRAM) ||
1621 (sk->sk_protocol != IPPROTO_UDP &&
1622 sk->sk_protocol != IPPROTO_TCP) ||
1623 (sk->sk_family != AF_INET &&
1624 sk->sk_family != AF_INET6)) {
1625 err = -ENOTSUPP;
1626 goto err_prog_put;
1627 }
1628 } else {
1629 /* BPF_PROG_TYPE_SOCKET_FILTER */
1630 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
1631 if (bpf_prog_size(prog->len) > optmem_max) {
1632 err = -ENOMEM;
1633 goto err_prog_put;
1634 }
1635 }
1636
1637 err = reuseport_attach_prog(sk, prog);
1638 err_prog_put:
1639 if (err)
1640 bpf_prog_put(prog);
1641
1642 return err;
1643 }
1644
sk_reuseport_prog_free(struct bpf_prog * prog)1645 void sk_reuseport_prog_free(struct bpf_prog *prog)
1646 {
1647 if (!prog)
1648 return;
1649
1650 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1651 bpf_prog_put(prog);
1652 else
1653 bpf_prog_destroy(prog);
1654 }
1655
1656 struct bpf_scratchpad {
1657 union {
1658 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1659 u8 buff[MAX_BPF_STACK];
1660 };
1661 };
1662
1663 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1664
__bpf_try_make_writable(struct sk_buff * skb,unsigned int write_len)1665 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1666 unsigned int write_len)
1667 {
1668 return skb_ensure_writable(skb, write_len);
1669 }
1670
bpf_try_make_writable(struct sk_buff * skb,unsigned int write_len)1671 static inline int bpf_try_make_writable(struct sk_buff *skb,
1672 unsigned int write_len)
1673 {
1674 int err = __bpf_try_make_writable(skb, write_len);
1675
1676 bpf_compute_data_pointers(skb);
1677 return err;
1678 }
1679
bpf_try_make_head_writable(struct sk_buff * skb)1680 static int bpf_try_make_head_writable(struct sk_buff *skb)
1681 {
1682 return bpf_try_make_writable(skb, skb_headlen(skb));
1683 }
1684
bpf_push_mac_rcsum(struct sk_buff * skb)1685 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1686 {
1687 if (skb_at_tc_ingress(skb))
1688 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1689 }
1690
bpf_pull_mac_rcsum(struct sk_buff * skb)1691 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1692 {
1693 if (skb_at_tc_ingress(skb))
1694 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1695 }
1696
BPF_CALL_5(bpf_skb_store_bytes,struct sk_buff *,skb,u32,offset,const void *,from,u32,len,u64,flags)1697 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1698 const void *, from, u32, len, u64, flags)
1699 {
1700 void *ptr;
1701
1702 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1703 return -EINVAL;
1704 if (unlikely(offset > INT_MAX))
1705 return -EFAULT;
1706 if (unlikely(bpf_try_make_writable(skb, offset + len)))
1707 return -EFAULT;
1708
1709 ptr = skb->data + offset;
1710 if (flags & BPF_F_RECOMPUTE_CSUM)
1711 __skb_postpull_rcsum(skb, ptr, len, offset);
1712
1713 memcpy(ptr, from, len);
1714
1715 if (flags & BPF_F_RECOMPUTE_CSUM)
1716 __skb_postpush_rcsum(skb, ptr, len, offset);
1717 if (flags & BPF_F_INVALIDATE_HASH)
1718 skb_clear_hash(skb);
1719
1720 return 0;
1721 }
1722
1723 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1724 .func = bpf_skb_store_bytes,
1725 .gpl_only = false,
1726 .ret_type = RET_INTEGER,
1727 .arg1_type = ARG_PTR_TO_CTX,
1728 .arg2_type = ARG_ANYTHING,
1729 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1730 .arg4_type = ARG_CONST_SIZE,
1731 .arg5_type = ARG_ANYTHING,
1732 };
1733
__bpf_skb_store_bytes(struct sk_buff * skb,u32 offset,const void * from,u32 len,u64 flags)1734 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1735 u32 len, u64 flags)
1736 {
1737 return ____bpf_skb_store_bytes(skb, offset, from, len, flags);
1738 }
1739
BPF_CALL_4(bpf_skb_load_bytes,const struct sk_buff *,skb,u32,offset,void *,to,u32,len)1740 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1741 void *, to, u32, len)
1742 {
1743 void *ptr;
1744
1745 if (unlikely(offset > INT_MAX))
1746 goto err_clear;
1747
1748 ptr = skb_header_pointer(skb, offset, len, to);
1749 if (unlikely(!ptr))
1750 goto err_clear;
1751 if (ptr != to)
1752 memcpy(to, ptr, len);
1753
1754 return 0;
1755 err_clear:
1756 memset(to, 0, len);
1757 return -EFAULT;
1758 }
1759
1760 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1761 .func = bpf_skb_load_bytes,
1762 .gpl_only = false,
1763 .ret_type = RET_INTEGER,
1764 .arg1_type = ARG_PTR_TO_CTX,
1765 .arg2_type = ARG_ANYTHING,
1766 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1767 .arg4_type = ARG_CONST_SIZE,
1768 };
1769
__bpf_skb_load_bytes(const struct sk_buff * skb,u32 offset,void * to,u32 len)1770 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len)
1771 {
1772 return ____bpf_skb_load_bytes(skb, offset, to, len);
1773 }
1774
BPF_CALL_4(bpf_flow_dissector_load_bytes,const struct bpf_flow_dissector *,ctx,u32,offset,void *,to,u32,len)1775 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1776 const struct bpf_flow_dissector *, ctx, u32, offset,
1777 void *, to, u32, len)
1778 {
1779 void *ptr;
1780
1781 if (unlikely(offset > 0xffff))
1782 goto err_clear;
1783
1784 if (unlikely(!ctx->skb))
1785 goto err_clear;
1786
1787 ptr = skb_header_pointer(ctx->skb, offset, len, to);
1788 if (unlikely(!ptr))
1789 goto err_clear;
1790 if (ptr != to)
1791 memcpy(to, ptr, len);
1792
1793 return 0;
1794 err_clear:
1795 memset(to, 0, len);
1796 return -EFAULT;
1797 }
1798
1799 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1800 .func = bpf_flow_dissector_load_bytes,
1801 .gpl_only = false,
1802 .ret_type = RET_INTEGER,
1803 .arg1_type = ARG_PTR_TO_CTX,
1804 .arg2_type = ARG_ANYTHING,
1805 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1806 .arg4_type = ARG_CONST_SIZE,
1807 };
1808
BPF_CALL_5(bpf_skb_load_bytes_relative,const struct sk_buff *,skb,u32,offset,void *,to,u32,len,u32,start_header)1809 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1810 u32, offset, void *, to, u32, len, u32, start_header)
1811 {
1812 u8 *end = skb_tail_pointer(skb);
1813 u8 *start, *ptr;
1814
1815 if (unlikely(offset > 0xffff))
1816 goto err_clear;
1817
1818 switch (start_header) {
1819 case BPF_HDR_START_MAC:
1820 if (unlikely(!skb_mac_header_was_set(skb)))
1821 goto err_clear;
1822 start = skb_mac_header(skb);
1823 break;
1824 case BPF_HDR_START_NET:
1825 start = skb_network_header(skb);
1826 break;
1827 default:
1828 goto err_clear;
1829 }
1830
1831 ptr = start + offset;
1832
1833 if (likely(ptr + len <= end)) {
1834 memcpy(to, ptr, len);
1835 return 0;
1836 }
1837
1838 err_clear:
1839 memset(to, 0, len);
1840 return -EFAULT;
1841 }
1842
1843 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1844 .func = bpf_skb_load_bytes_relative,
1845 .gpl_only = false,
1846 .ret_type = RET_INTEGER,
1847 .arg1_type = ARG_PTR_TO_CTX,
1848 .arg2_type = ARG_ANYTHING,
1849 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
1850 .arg4_type = ARG_CONST_SIZE,
1851 .arg5_type = ARG_ANYTHING,
1852 };
1853
BPF_CALL_2(bpf_skb_pull_data,struct sk_buff *,skb,u32,len)1854 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1855 {
1856 /* Idea is the following: should the needed direct read/write
1857 * test fail during runtime, we can pull in more data and redo
1858 * again, since implicitly, we invalidate previous checks here.
1859 *
1860 * Or, since we know how much we need to make read/writeable,
1861 * this can be done once at the program beginning for direct
1862 * access case. By this we overcome limitations of only current
1863 * headroom being accessible.
1864 */
1865 return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1866 }
1867
1868 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1869 .func = bpf_skb_pull_data,
1870 .gpl_only = false,
1871 .ret_type = RET_INTEGER,
1872 .arg1_type = ARG_PTR_TO_CTX,
1873 .arg2_type = ARG_ANYTHING,
1874 };
1875
BPF_CALL_1(bpf_sk_fullsock,struct sock *,sk)1876 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1877 {
1878 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1879 }
1880
1881 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1882 .func = bpf_sk_fullsock,
1883 .gpl_only = false,
1884 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
1885 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
1886 };
1887
sk_skb_try_make_writable(struct sk_buff * skb,unsigned int write_len)1888 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1889 unsigned int write_len)
1890 {
1891 return __bpf_try_make_writable(skb, write_len);
1892 }
1893
BPF_CALL_2(sk_skb_pull_data,struct sk_buff *,skb,u32,len)1894 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1895 {
1896 /* Idea is the following: should the needed direct read/write
1897 * test fail during runtime, we can pull in more data and redo
1898 * again, since implicitly, we invalidate previous checks here.
1899 *
1900 * Or, since we know how much we need to make read/writeable,
1901 * this can be done once at the program beginning for direct
1902 * access case. By this we overcome limitations of only current
1903 * headroom being accessible.
1904 */
1905 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1906 }
1907
1908 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1909 .func = sk_skb_pull_data,
1910 .gpl_only = false,
1911 .ret_type = RET_INTEGER,
1912 .arg1_type = ARG_PTR_TO_CTX,
1913 .arg2_type = ARG_ANYTHING,
1914 };
1915
BPF_CALL_5(bpf_l3_csum_replace,struct sk_buff *,skb,u32,offset,u64,from,u64,to,u64,flags)1916 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1917 u64, from, u64, to, u64, flags)
1918 {
1919 __sum16 *ptr;
1920
1921 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1922 return -EINVAL;
1923 if (unlikely(offset > 0xffff || offset & 1))
1924 return -EFAULT;
1925 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1926 return -EFAULT;
1927
1928 ptr = (__sum16 *)(skb->data + offset);
1929 switch (flags & BPF_F_HDR_FIELD_MASK) {
1930 case 0:
1931 if (unlikely(from != 0))
1932 return -EINVAL;
1933
1934 csum_replace_by_diff(ptr, to);
1935 break;
1936 case 2:
1937 csum_replace2(ptr, from, to);
1938 break;
1939 case 4:
1940 csum_replace4(ptr, from, to);
1941 break;
1942 default:
1943 return -EINVAL;
1944 }
1945
1946 return 0;
1947 }
1948
1949 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1950 .func = bpf_l3_csum_replace,
1951 .gpl_only = false,
1952 .ret_type = RET_INTEGER,
1953 .arg1_type = ARG_PTR_TO_CTX,
1954 .arg2_type = ARG_ANYTHING,
1955 .arg3_type = ARG_ANYTHING,
1956 .arg4_type = ARG_ANYTHING,
1957 .arg5_type = ARG_ANYTHING,
1958 };
1959
BPF_CALL_5(bpf_l4_csum_replace,struct sk_buff *,skb,u32,offset,u64,from,u64,to,u64,flags)1960 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1961 u64, from, u64, to, u64, flags)
1962 {
1963 bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1964 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1965 bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1966 __sum16 *ptr;
1967
1968 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1969 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1970 return -EINVAL;
1971 if (unlikely(offset > 0xffff || offset & 1))
1972 return -EFAULT;
1973 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1974 return -EFAULT;
1975
1976 ptr = (__sum16 *)(skb->data + offset);
1977 if (is_mmzero && !do_mforce && !*ptr)
1978 return 0;
1979
1980 switch (flags & BPF_F_HDR_FIELD_MASK) {
1981 case 0:
1982 if (unlikely(from != 0))
1983 return -EINVAL;
1984
1985 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1986 break;
1987 case 2:
1988 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1989 break;
1990 case 4:
1991 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1992 break;
1993 default:
1994 return -EINVAL;
1995 }
1996
1997 if (is_mmzero && !*ptr)
1998 *ptr = CSUM_MANGLED_0;
1999 return 0;
2000 }
2001
2002 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
2003 .func = bpf_l4_csum_replace,
2004 .gpl_only = false,
2005 .ret_type = RET_INTEGER,
2006 .arg1_type = ARG_PTR_TO_CTX,
2007 .arg2_type = ARG_ANYTHING,
2008 .arg3_type = ARG_ANYTHING,
2009 .arg4_type = ARG_ANYTHING,
2010 .arg5_type = ARG_ANYTHING,
2011 };
2012
BPF_CALL_5(bpf_csum_diff,__be32 *,from,u32,from_size,__be32 *,to,u32,to_size,__wsum,seed)2013 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
2014 __be32 *, to, u32, to_size, __wsum, seed)
2015 {
2016 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
2017 u32 diff_size = from_size + to_size;
2018 int i, j = 0;
2019
2020 /* This is quite flexible, some examples:
2021 *
2022 * from_size == 0, to_size > 0, seed := csum --> pushing data
2023 * from_size > 0, to_size == 0, seed := csum --> pulling data
2024 * from_size > 0, to_size > 0, seed := 0 --> diffing data
2025 *
2026 * Even for diffing, from_size and to_size don't need to be equal.
2027 */
2028 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2029 diff_size > sizeof(sp->diff)))
2030 return -EINVAL;
2031
2032 for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2033 sp->diff[j] = ~from[i];
2034 for (i = 0; i < to_size / sizeof(__be32); i++, j++)
2035 sp->diff[j] = to[i];
2036
2037 return csum_partial(sp->diff, diff_size, seed);
2038 }
2039
2040 static const struct bpf_func_proto bpf_csum_diff_proto = {
2041 .func = bpf_csum_diff,
2042 .gpl_only = false,
2043 .pkt_access = true,
2044 .ret_type = RET_INTEGER,
2045 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2046 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
2047 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2048 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
2049 .arg5_type = ARG_ANYTHING,
2050 };
2051
BPF_CALL_2(bpf_csum_update,struct sk_buff *,skb,__wsum,csum)2052 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2053 {
2054 /* The interface is to be used in combination with bpf_csum_diff()
2055 * for direct packet writes. csum rotation for alignment as well
2056 * as emulating csum_sub() can be done from the eBPF program.
2057 */
2058 if (skb->ip_summed == CHECKSUM_COMPLETE)
2059 return (skb->csum = csum_add(skb->csum, csum));
2060
2061 return -ENOTSUPP;
2062 }
2063
2064 static const struct bpf_func_proto bpf_csum_update_proto = {
2065 .func = bpf_csum_update,
2066 .gpl_only = false,
2067 .ret_type = RET_INTEGER,
2068 .arg1_type = ARG_PTR_TO_CTX,
2069 .arg2_type = ARG_ANYTHING,
2070 };
2071
BPF_CALL_2(bpf_csum_level,struct sk_buff *,skb,u64,level)2072 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2073 {
2074 /* The interface is to be used in combination with bpf_skb_adjust_room()
2075 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2076 * is passed as flags, for example.
2077 */
2078 switch (level) {
2079 case BPF_CSUM_LEVEL_INC:
2080 __skb_incr_checksum_unnecessary(skb);
2081 break;
2082 case BPF_CSUM_LEVEL_DEC:
2083 __skb_decr_checksum_unnecessary(skb);
2084 break;
2085 case BPF_CSUM_LEVEL_RESET:
2086 __skb_reset_checksum_unnecessary(skb);
2087 break;
2088 case BPF_CSUM_LEVEL_QUERY:
2089 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2090 skb->csum_level : -EACCES;
2091 default:
2092 return -EINVAL;
2093 }
2094
2095 return 0;
2096 }
2097
2098 static const struct bpf_func_proto bpf_csum_level_proto = {
2099 .func = bpf_csum_level,
2100 .gpl_only = false,
2101 .ret_type = RET_INTEGER,
2102 .arg1_type = ARG_PTR_TO_CTX,
2103 .arg2_type = ARG_ANYTHING,
2104 };
2105
__bpf_rx_skb(struct net_device * dev,struct sk_buff * skb)2106 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2107 {
2108 return dev_forward_skb_nomtu(dev, skb);
2109 }
2110
__bpf_rx_skb_no_mac(struct net_device * dev,struct sk_buff * skb)2111 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2112 struct sk_buff *skb)
2113 {
2114 int ret = ____dev_forward_skb(dev, skb, false);
2115
2116 if (likely(!ret)) {
2117 skb->dev = dev;
2118 ret = netif_rx(skb);
2119 }
2120
2121 return ret;
2122 }
2123
__bpf_tx_skb(struct net_device * dev,struct sk_buff * skb)2124 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2125 {
2126 int ret;
2127
2128 if (dev_xmit_recursion()) {
2129 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2130 kfree_skb(skb);
2131 return -ENETDOWN;
2132 }
2133
2134 skb->dev = dev;
2135 skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb));
2136 skb_clear_tstamp(skb);
2137
2138 dev_xmit_recursion_inc();
2139 ret = dev_queue_xmit(skb);
2140 dev_xmit_recursion_dec();
2141
2142 return ret;
2143 }
2144
__bpf_redirect_no_mac(struct sk_buff * skb,struct net_device * dev,u32 flags)2145 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2146 u32 flags)
2147 {
2148 unsigned int mlen = skb_network_offset(skb);
2149
2150 if (unlikely(skb->len <= mlen)) {
2151 kfree_skb(skb);
2152 return -ERANGE;
2153 }
2154
2155 if (mlen) {
2156 __skb_pull(skb, mlen);
2157
2158 /* At ingress, the mac header has already been pulled once.
2159 * At egress, skb_pospull_rcsum has to be done in case that
2160 * the skb is originated from ingress (i.e. a forwarded skb)
2161 * to ensure that rcsum starts at net header.
2162 */
2163 if (!skb_at_tc_ingress(skb))
2164 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2165 }
2166 skb_pop_mac_header(skb);
2167 skb_reset_mac_len(skb);
2168 return flags & BPF_F_INGRESS ?
2169 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2170 }
2171
__bpf_redirect_common(struct sk_buff * skb,struct net_device * dev,u32 flags)2172 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2173 u32 flags)
2174 {
2175 /* Verify that a link layer header is carried */
2176 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2177 kfree_skb(skb);
2178 return -ERANGE;
2179 }
2180
2181 bpf_push_mac_rcsum(skb);
2182 return flags & BPF_F_INGRESS ?
2183 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2184 }
2185
__bpf_redirect(struct sk_buff * skb,struct net_device * dev,u32 flags)2186 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2187 u32 flags)
2188 {
2189 if (dev_is_mac_header_xmit(dev))
2190 return __bpf_redirect_common(skb, dev, flags);
2191 else
2192 return __bpf_redirect_no_mac(skb, dev, flags);
2193 }
2194
2195 #if IS_ENABLED(CONFIG_IPV6)
bpf_out_neigh_v6(struct net * net,struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2196 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2197 struct net_device *dev, struct bpf_nh_params *nh)
2198 {
2199 u32 hh_len = LL_RESERVED_SPACE(dev);
2200 const struct in6_addr *nexthop;
2201 struct dst_entry *dst = NULL;
2202 struct neighbour *neigh;
2203
2204 if (dev_xmit_recursion()) {
2205 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2206 goto out_drop;
2207 }
2208
2209 skb->dev = dev;
2210 skb_clear_tstamp(skb);
2211
2212 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2213 skb = skb_expand_head(skb, hh_len);
2214 if (!skb)
2215 return -ENOMEM;
2216 }
2217
2218 rcu_read_lock();
2219 if (!nh) {
2220 dst = skb_dst(skb);
2221 nexthop = rt6_nexthop(dst_rt6_info(dst),
2222 &ipv6_hdr(skb)->daddr);
2223 } else {
2224 nexthop = &nh->ipv6_nh;
2225 }
2226 neigh = ip_neigh_gw6(dev, nexthop);
2227 if (likely(!IS_ERR(neigh))) {
2228 int ret;
2229
2230 sock_confirm_neigh(skb, neigh);
2231 local_bh_disable();
2232 dev_xmit_recursion_inc();
2233 ret = neigh_output(neigh, skb, false);
2234 dev_xmit_recursion_dec();
2235 local_bh_enable();
2236 rcu_read_unlock();
2237 return ret;
2238 }
2239 rcu_read_unlock_bh();
2240 if (dst)
2241 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2242 out_drop:
2243 kfree_skb(skb);
2244 return -ENETDOWN;
2245 }
2246
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2247 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2248 struct bpf_nh_params *nh)
2249 {
2250 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2251 struct net *net = dev_net(dev);
2252 int err, ret = NET_XMIT_DROP;
2253
2254 if (!nh) {
2255 struct dst_entry *dst;
2256 struct flowi6 fl6 = {
2257 .flowi6_flags = FLOWI_FLAG_ANYSRC,
2258 .flowi6_mark = skb->mark,
2259 .flowlabel = ip6_flowinfo(ip6h),
2260 .flowi6_oif = dev->ifindex,
2261 .flowi6_proto = ip6h->nexthdr,
2262 .daddr = ip6h->daddr,
2263 .saddr = ip6h->saddr,
2264 };
2265
2266 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2267 if (IS_ERR(dst))
2268 goto out_drop;
2269
2270 skb_dst_set(skb, dst);
2271 } else if (nh->nh_family != AF_INET6) {
2272 goto out_drop;
2273 }
2274
2275 err = bpf_out_neigh_v6(net, skb, dev, nh);
2276 if (unlikely(net_xmit_eval(err)))
2277 dev->stats.tx_errors++;
2278 else
2279 ret = NET_XMIT_SUCCESS;
2280 goto out_xmit;
2281 out_drop:
2282 dev->stats.tx_errors++;
2283 kfree_skb(skb);
2284 out_xmit:
2285 return ret;
2286 }
2287 #else
__bpf_redirect_neigh_v6(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2288 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2289 struct bpf_nh_params *nh)
2290 {
2291 kfree_skb(skb);
2292 return NET_XMIT_DROP;
2293 }
2294 #endif /* CONFIG_IPV6 */
2295
2296 #if IS_ENABLED(CONFIG_INET)
bpf_out_neigh_v4(struct net * net,struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2297 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2298 struct net_device *dev, struct bpf_nh_params *nh)
2299 {
2300 u32 hh_len = LL_RESERVED_SPACE(dev);
2301 struct neighbour *neigh;
2302 bool is_v6gw = false;
2303
2304 if (dev_xmit_recursion()) {
2305 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2306 goto out_drop;
2307 }
2308
2309 skb->dev = dev;
2310 skb_clear_tstamp(skb);
2311
2312 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2313 skb = skb_expand_head(skb, hh_len);
2314 if (!skb)
2315 return -ENOMEM;
2316 }
2317
2318 rcu_read_lock();
2319 if (!nh) {
2320 struct rtable *rt = skb_rtable(skb);
2321
2322 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2323 } else if (nh->nh_family == AF_INET6) {
2324 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2325 is_v6gw = true;
2326 } else if (nh->nh_family == AF_INET) {
2327 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2328 } else {
2329 rcu_read_unlock();
2330 goto out_drop;
2331 }
2332
2333 if (likely(!IS_ERR(neigh))) {
2334 int ret;
2335
2336 sock_confirm_neigh(skb, neigh);
2337 local_bh_disable();
2338 dev_xmit_recursion_inc();
2339 ret = neigh_output(neigh, skb, is_v6gw);
2340 dev_xmit_recursion_dec();
2341 local_bh_enable();
2342 rcu_read_unlock();
2343 return ret;
2344 }
2345 rcu_read_unlock();
2346 out_drop:
2347 kfree_skb(skb);
2348 return -ENETDOWN;
2349 }
2350
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2351 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2352 struct bpf_nh_params *nh)
2353 {
2354 const struct iphdr *ip4h = ip_hdr(skb);
2355 struct net *net = dev_net(dev);
2356 int err, ret = NET_XMIT_DROP;
2357
2358 if (!nh) {
2359 struct flowi4 fl4 = {
2360 .flowi4_flags = FLOWI_FLAG_ANYSRC,
2361 .flowi4_mark = skb->mark,
2362 .flowi4_tos = RT_TOS(ip4h->tos),
2363 .flowi4_oif = dev->ifindex,
2364 .flowi4_proto = ip4h->protocol,
2365 .daddr = ip4h->daddr,
2366 .saddr = ip4h->saddr,
2367 };
2368 struct rtable *rt;
2369
2370 rt = ip_route_output_flow(net, &fl4, NULL);
2371 if (IS_ERR(rt))
2372 goto out_drop;
2373 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2374 ip_rt_put(rt);
2375 goto out_drop;
2376 }
2377
2378 skb_dst_set(skb, &rt->dst);
2379 }
2380
2381 err = bpf_out_neigh_v4(net, skb, dev, nh);
2382 if (unlikely(net_xmit_eval(err)))
2383 dev->stats.tx_errors++;
2384 else
2385 ret = NET_XMIT_SUCCESS;
2386 goto out_xmit;
2387 out_drop:
2388 dev->stats.tx_errors++;
2389 kfree_skb(skb);
2390 out_xmit:
2391 return ret;
2392 }
2393 #else
__bpf_redirect_neigh_v4(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2394 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2395 struct bpf_nh_params *nh)
2396 {
2397 kfree_skb(skb);
2398 return NET_XMIT_DROP;
2399 }
2400 #endif /* CONFIG_INET */
2401
__bpf_redirect_neigh(struct sk_buff * skb,struct net_device * dev,struct bpf_nh_params * nh)2402 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2403 struct bpf_nh_params *nh)
2404 {
2405 struct ethhdr *ethh = eth_hdr(skb);
2406
2407 if (unlikely(skb->mac_header >= skb->network_header))
2408 goto out;
2409 bpf_push_mac_rcsum(skb);
2410 if (is_multicast_ether_addr(ethh->h_dest))
2411 goto out;
2412
2413 skb_pull(skb, sizeof(*ethh));
2414 skb_unset_mac_header(skb);
2415 skb_reset_network_header(skb);
2416
2417 if (skb->protocol == htons(ETH_P_IP))
2418 return __bpf_redirect_neigh_v4(skb, dev, nh);
2419 else if (skb->protocol == htons(ETH_P_IPV6))
2420 return __bpf_redirect_neigh_v6(skb, dev, nh);
2421 out:
2422 kfree_skb(skb);
2423 return -ENOTSUPP;
2424 }
2425
2426 /* Internal, non-exposed redirect flags. */
2427 enum {
2428 BPF_F_NEIGH = (1ULL << 1),
2429 BPF_F_PEER = (1ULL << 2),
2430 BPF_F_NEXTHOP = (1ULL << 3),
2431 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2432 };
2433
BPF_CALL_3(bpf_clone_redirect,struct sk_buff *,skb,u32,ifindex,u64,flags)2434 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2435 {
2436 struct net_device *dev;
2437 struct sk_buff *clone;
2438 int ret;
2439
2440 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2441 return -EINVAL;
2442
2443 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2444 if (unlikely(!dev))
2445 return -EINVAL;
2446
2447 clone = skb_clone(skb, GFP_ATOMIC);
2448 if (unlikely(!clone))
2449 return -ENOMEM;
2450
2451 /* For direct write, we need to keep the invariant that the skbs
2452 * we're dealing with need to be uncloned. Should uncloning fail
2453 * here, we need to free the just generated clone to unclone once
2454 * again.
2455 */
2456 ret = bpf_try_make_head_writable(skb);
2457 if (unlikely(ret)) {
2458 kfree_skb(clone);
2459 return -ENOMEM;
2460 }
2461
2462 return __bpf_redirect(clone, dev, flags);
2463 }
2464
2465 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2466 .func = bpf_clone_redirect,
2467 .gpl_only = false,
2468 .ret_type = RET_INTEGER,
2469 .arg1_type = ARG_PTR_TO_CTX,
2470 .arg2_type = ARG_ANYTHING,
2471 .arg3_type = ARG_ANYTHING,
2472 };
2473
2474 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2475 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2476
skb_get_peer_dev(struct net_device * dev)2477 static struct net_device *skb_get_peer_dev(struct net_device *dev)
2478 {
2479 const struct net_device_ops *ops = dev->netdev_ops;
2480
2481 if (likely(ops->ndo_get_peer_dev))
2482 return INDIRECT_CALL_1(ops->ndo_get_peer_dev,
2483 netkit_peer_dev, dev);
2484 return NULL;
2485 }
2486
skb_do_redirect(struct sk_buff * skb)2487 int skb_do_redirect(struct sk_buff *skb)
2488 {
2489 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2490 struct net *net = dev_net(skb->dev);
2491 struct net_device *dev;
2492 u32 flags = ri->flags;
2493
2494 dev = dev_get_by_index_rcu(net, ri->tgt_index);
2495 ri->tgt_index = 0;
2496 ri->flags = 0;
2497 if (unlikely(!dev))
2498 goto out_drop;
2499 if (flags & BPF_F_PEER) {
2500 if (unlikely(!skb_at_tc_ingress(skb)))
2501 goto out_drop;
2502 dev = skb_get_peer_dev(dev);
2503 if (unlikely(!dev ||
2504 !(dev->flags & IFF_UP) ||
2505 net_eq(net, dev_net(dev))))
2506 goto out_drop;
2507 skb->dev = dev;
2508 dev_sw_netstats_rx_add(dev, skb->len);
2509 return -EAGAIN;
2510 }
2511 return flags & BPF_F_NEIGH ?
2512 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2513 &ri->nh : NULL) :
2514 __bpf_redirect(skb, dev, flags);
2515 out_drop:
2516 kfree_skb(skb);
2517 return -EINVAL;
2518 }
2519
BPF_CALL_2(bpf_redirect,u32,ifindex,u64,flags)2520 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2521 {
2522 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2523
2524 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2525 return TC_ACT_SHOT;
2526
2527 ri->flags = flags;
2528 ri->tgt_index = ifindex;
2529
2530 return TC_ACT_REDIRECT;
2531 }
2532
2533 static const struct bpf_func_proto bpf_redirect_proto = {
2534 .func = bpf_redirect,
2535 .gpl_only = false,
2536 .ret_type = RET_INTEGER,
2537 .arg1_type = ARG_ANYTHING,
2538 .arg2_type = ARG_ANYTHING,
2539 };
2540
BPF_CALL_2(bpf_redirect_peer,u32,ifindex,u64,flags)2541 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2542 {
2543 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2544
2545 if (unlikely(flags))
2546 return TC_ACT_SHOT;
2547
2548 ri->flags = BPF_F_PEER;
2549 ri->tgt_index = ifindex;
2550
2551 return TC_ACT_REDIRECT;
2552 }
2553
2554 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2555 .func = bpf_redirect_peer,
2556 .gpl_only = false,
2557 .ret_type = RET_INTEGER,
2558 .arg1_type = ARG_ANYTHING,
2559 .arg2_type = ARG_ANYTHING,
2560 };
2561
BPF_CALL_4(bpf_redirect_neigh,u32,ifindex,struct bpf_redir_neigh *,params,int,plen,u64,flags)2562 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2563 int, plen, u64, flags)
2564 {
2565 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2566
2567 if (unlikely((plen && plen < sizeof(*params)) || flags))
2568 return TC_ACT_SHOT;
2569
2570 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2571 ri->tgt_index = ifindex;
2572
2573 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2574 if (plen)
2575 memcpy(&ri->nh, params, sizeof(ri->nh));
2576
2577 return TC_ACT_REDIRECT;
2578 }
2579
2580 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2581 .func = bpf_redirect_neigh,
2582 .gpl_only = false,
2583 .ret_type = RET_INTEGER,
2584 .arg1_type = ARG_ANYTHING,
2585 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2586 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
2587 .arg4_type = ARG_ANYTHING,
2588 };
2589
BPF_CALL_2(bpf_msg_apply_bytes,struct sk_msg *,msg,u32,bytes)2590 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2591 {
2592 msg->apply_bytes = bytes;
2593 return 0;
2594 }
2595
2596 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2597 .func = bpf_msg_apply_bytes,
2598 .gpl_only = false,
2599 .ret_type = RET_INTEGER,
2600 .arg1_type = ARG_PTR_TO_CTX,
2601 .arg2_type = ARG_ANYTHING,
2602 };
2603
BPF_CALL_2(bpf_msg_cork_bytes,struct sk_msg *,msg,u32,bytes)2604 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2605 {
2606 msg->cork_bytes = bytes;
2607 return 0;
2608 }
2609
sk_msg_reset_curr(struct sk_msg * msg)2610 static void sk_msg_reset_curr(struct sk_msg *msg)
2611 {
2612 u32 i = msg->sg.start;
2613 u32 len = 0;
2614
2615 do {
2616 len += sk_msg_elem(msg, i)->length;
2617 sk_msg_iter_var_next(i);
2618 if (len >= msg->sg.size)
2619 break;
2620 } while (i != msg->sg.end);
2621
2622 msg->sg.curr = i;
2623 msg->sg.copybreak = 0;
2624 }
2625
2626 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2627 .func = bpf_msg_cork_bytes,
2628 .gpl_only = false,
2629 .ret_type = RET_INTEGER,
2630 .arg1_type = ARG_PTR_TO_CTX,
2631 .arg2_type = ARG_ANYTHING,
2632 };
2633
BPF_CALL_4(bpf_msg_pull_data,struct sk_msg *,msg,u32,start,u32,end,u64,flags)2634 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2635 u32, end, u64, flags)
2636 {
2637 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2638 u32 first_sge, last_sge, i, shift, bytes_sg_total;
2639 struct scatterlist *sge;
2640 u8 *raw, *to, *from;
2641 struct page *page;
2642
2643 if (unlikely(flags || end <= start))
2644 return -EINVAL;
2645
2646 /* First find the starting scatterlist element */
2647 i = msg->sg.start;
2648 do {
2649 offset += len;
2650 len = sk_msg_elem(msg, i)->length;
2651 if (start < offset + len)
2652 break;
2653 sk_msg_iter_var_next(i);
2654 } while (i != msg->sg.end);
2655
2656 if (unlikely(start >= offset + len))
2657 return -EINVAL;
2658
2659 first_sge = i;
2660 /* The start may point into the sg element so we need to also
2661 * account for the headroom.
2662 */
2663 bytes_sg_total = start - offset + bytes;
2664 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2665 goto out;
2666
2667 /* At this point we need to linearize multiple scatterlist
2668 * elements or a single shared page. Either way we need to
2669 * copy into a linear buffer exclusively owned by BPF. Then
2670 * place the buffer in the scatterlist and fixup the original
2671 * entries by removing the entries now in the linear buffer
2672 * and shifting the remaining entries. For now we do not try
2673 * to copy partial entries to avoid complexity of running out
2674 * of sg_entry slots. The downside is reading a single byte
2675 * will copy the entire sg entry.
2676 */
2677 do {
2678 copy += sk_msg_elem(msg, i)->length;
2679 sk_msg_iter_var_next(i);
2680 if (bytes_sg_total <= copy)
2681 break;
2682 } while (i != msg->sg.end);
2683 last_sge = i;
2684
2685 if (unlikely(bytes_sg_total > copy))
2686 return -EINVAL;
2687
2688 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2689 get_order(copy));
2690 if (unlikely(!page))
2691 return -ENOMEM;
2692
2693 raw = page_address(page);
2694 i = first_sge;
2695 do {
2696 sge = sk_msg_elem(msg, i);
2697 from = sg_virt(sge);
2698 len = sge->length;
2699 to = raw + poffset;
2700
2701 memcpy(to, from, len);
2702 poffset += len;
2703 sge->length = 0;
2704 put_page(sg_page(sge));
2705
2706 sk_msg_iter_var_next(i);
2707 } while (i != last_sge);
2708
2709 sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2710
2711 /* To repair sg ring we need to shift entries. If we only
2712 * had a single entry though we can just replace it and
2713 * be done. Otherwise walk the ring and shift the entries.
2714 */
2715 WARN_ON_ONCE(last_sge == first_sge);
2716 shift = last_sge > first_sge ?
2717 last_sge - first_sge - 1 :
2718 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2719 if (!shift)
2720 goto out;
2721
2722 i = first_sge;
2723 sk_msg_iter_var_next(i);
2724 do {
2725 u32 move_from;
2726
2727 if (i + shift >= NR_MSG_FRAG_IDS)
2728 move_from = i + shift - NR_MSG_FRAG_IDS;
2729 else
2730 move_from = i + shift;
2731 if (move_from == msg->sg.end)
2732 break;
2733
2734 msg->sg.data[i] = msg->sg.data[move_from];
2735 msg->sg.data[move_from].length = 0;
2736 msg->sg.data[move_from].page_link = 0;
2737 msg->sg.data[move_from].offset = 0;
2738 sk_msg_iter_var_next(i);
2739 } while (1);
2740
2741 msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2742 msg->sg.end - shift + NR_MSG_FRAG_IDS :
2743 msg->sg.end - shift;
2744 out:
2745 sk_msg_reset_curr(msg);
2746 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2747 msg->data_end = msg->data + bytes;
2748 return 0;
2749 }
2750
2751 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2752 .func = bpf_msg_pull_data,
2753 .gpl_only = false,
2754 .ret_type = RET_INTEGER,
2755 .arg1_type = ARG_PTR_TO_CTX,
2756 .arg2_type = ARG_ANYTHING,
2757 .arg3_type = ARG_ANYTHING,
2758 .arg4_type = ARG_ANYTHING,
2759 };
2760
BPF_CALL_4(bpf_msg_push_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2761 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2762 u32, len, u64, flags)
2763 {
2764 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2765 u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2766 u8 *raw, *to, *from;
2767 struct page *page;
2768
2769 if (unlikely(flags))
2770 return -EINVAL;
2771
2772 if (unlikely(len == 0))
2773 return 0;
2774
2775 /* First find the starting scatterlist element */
2776 i = msg->sg.start;
2777 do {
2778 offset += l;
2779 l = sk_msg_elem(msg, i)->length;
2780
2781 if (start < offset + l)
2782 break;
2783 sk_msg_iter_var_next(i);
2784 } while (i != msg->sg.end);
2785
2786 if (start >= offset + l)
2787 return -EINVAL;
2788
2789 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2790
2791 /* If no space available will fallback to copy, we need at
2792 * least one scatterlist elem available to push data into
2793 * when start aligns to the beginning of an element or two
2794 * when it falls inside an element. We handle the start equals
2795 * offset case because its the common case for inserting a
2796 * header.
2797 */
2798 if (!space || (space == 1 && start != offset))
2799 copy = msg->sg.data[i].length;
2800
2801 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2802 get_order(copy + len));
2803 if (unlikely(!page))
2804 return -ENOMEM;
2805
2806 if (copy) {
2807 int front, back;
2808
2809 raw = page_address(page);
2810
2811 psge = sk_msg_elem(msg, i);
2812 front = start - offset;
2813 back = psge->length - front;
2814 from = sg_virt(psge);
2815
2816 if (front)
2817 memcpy(raw, from, front);
2818
2819 if (back) {
2820 from += front;
2821 to = raw + front + len;
2822
2823 memcpy(to, from, back);
2824 }
2825
2826 put_page(sg_page(psge));
2827 } else if (start - offset) {
2828 psge = sk_msg_elem(msg, i);
2829 rsge = sk_msg_elem_cpy(msg, i);
2830
2831 psge->length = start - offset;
2832 rsge.length -= psge->length;
2833 rsge.offset += start;
2834
2835 sk_msg_iter_var_next(i);
2836 sg_unmark_end(psge);
2837 sg_unmark_end(&rsge);
2838 sk_msg_iter_next(msg, end);
2839 }
2840
2841 /* Slot(s) to place newly allocated data */
2842 new = i;
2843
2844 /* Shift one or two slots as needed */
2845 if (!copy) {
2846 sge = sk_msg_elem_cpy(msg, i);
2847
2848 sk_msg_iter_var_next(i);
2849 sg_unmark_end(&sge);
2850 sk_msg_iter_next(msg, end);
2851
2852 nsge = sk_msg_elem_cpy(msg, i);
2853 if (rsge.length) {
2854 sk_msg_iter_var_next(i);
2855 nnsge = sk_msg_elem_cpy(msg, i);
2856 }
2857
2858 while (i != msg->sg.end) {
2859 msg->sg.data[i] = sge;
2860 sge = nsge;
2861 sk_msg_iter_var_next(i);
2862 if (rsge.length) {
2863 nsge = nnsge;
2864 nnsge = sk_msg_elem_cpy(msg, i);
2865 } else {
2866 nsge = sk_msg_elem_cpy(msg, i);
2867 }
2868 }
2869 }
2870
2871 /* Place newly allocated data buffer */
2872 sk_mem_charge(msg->sk, len);
2873 msg->sg.size += len;
2874 __clear_bit(new, msg->sg.copy);
2875 sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2876 if (rsge.length) {
2877 get_page(sg_page(&rsge));
2878 sk_msg_iter_var_next(new);
2879 msg->sg.data[new] = rsge;
2880 }
2881
2882 sk_msg_reset_curr(msg);
2883 sk_msg_compute_data_pointers(msg);
2884 return 0;
2885 }
2886
2887 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2888 .func = bpf_msg_push_data,
2889 .gpl_only = false,
2890 .ret_type = RET_INTEGER,
2891 .arg1_type = ARG_PTR_TO_CTX,
2892 .arg2_type = ARG_ANYTHING,
2893 .arg3_type = ARG_ANYTHING,
2894 .arg4_type = ARG_ANYTHING,
2895 };
2896
sk_msg_shift_left(struct sk_msg * msg,int i)2897 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2898 {
2899 int prev;
2900
2901 do {
2902 prev = i;
2903 sk_msg_iter_var_next(i);
2904 msg->sg.data[prev] = msg->sg.data[i];
2905 } while (i != msg->sg.end);
2906
2907 sk_msg_iter_prev(msg, end);
2908 }
2909
sk_msg_shift_right(struct sk_msg * msg,int i)2910 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2911 {
2912 struct scatterlist tmp, sge;
2913
2914 sk_msg_iter_next(msg, end);
2915 sge = sk_msg_elem_cpy(msg, i);
2916 sk_msg_iter_var_next(i);
2917 tmp = sk_msg_elem_cpy(msg, i);
2918
2919 while (i != msg->sg.end) {
2920 msg->sg.data[i] = sge;
2921 sk_msg_iter_var_next(i);
2922 sge = tmp;
2923 tmp = sk_msg_elem_cpy(msg, i);
2924 }
2925 }
2926
BPF_CALL_4(bpf_msg_pop_data,struct sk_msg *,msg,u32,start,u32,len,u64,flags)2927 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2928 u32, len, u64, flags)
2929 {
2930 u32 i = 0, l = 0, space, offset = 0;
2931 u64 last = start + len;
2932 int pop;
2933
2934 if (unlikely(flags))
2935 return -EINVAL;
2936
2937 /* First find the starting scatterlist element */
2938 i = msg->sg.start;
2939 do {
2940 offset += l;
2941 l = sk_msg_elem(msg, i)->length;
2942
2943 if (start < offset + l)
2944 break;
2945 sk_msg_iter_var_next(i);
2946 } while (i != msg->sg.end);
2947
2948 /* Bounds checks: start and pop must be inside message */
2949 if (start >= offset + l || last >= msg->sg.size)
2950 return -EINVAL;
2951
2952 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2953
2954 pop = len;
2955 /* --------------| offset
2956 * -| start |-------- len -------|
2957 *
2958 * |----- a ----|-------- pop -------|----- b ----|
2959 * |______________________________________________| length
2960 *
2961 *
2962 * a: region at front of scatter element to save
2963 * b: region at back of scatter element to save when length > A + pop
2964 * pop: region to pop from element, same as input 'pop' here will be
2965 * decremented below per iteration.
2966 *
2967 * Two top-level cases to handle when start != offset, first B is non
2968 * zero and second B is zero corresponding to when a pop includes more
2969 * than one element.
2970 *
2971 * Then if B is non-zero AND there is no space allocate space and
2972 * compact A, B regions into page. If there is space shift ring to
2973 * the right free'ing the next element in ring to place B, leaving
2974 * A untouched except to reduce length.
2975 */
2976 if (start != offset) {
2977 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2978 int a = start;
2979 int b = sge->length - pop - a;
2980
2981 sk_msg_iter_var_next(i);
2982
2983 if (pop < sge->length - a) {
2984 if (space) {
2985 sge->length = a;
2986 sk_msg_shift_right(msg, i);
2987 nsge = sk_msg_elem(msg, i);
2988 get_page(sg_page(sge));
2989 sg_set_page(nsge,
2990 sg_page(sge),
2991 b, sge->offset + pop + a);
2992 } else {
2993 struct page *page, *orig;
2994 u8 *to, *from;
2995
2996 page = alloc_pages(__GFP_NOWARN |
2997 __GFP_COMP | GFP_ATOMIC,
2998 get_order(a + b));
2999 if (unlikely(!page))
3000 return -ENOMEM;
3001
3002 sge->length = a;
3003 orig = sg_page(sge);
3004 from = sg_virt(sge);
3005 to = page_address(page);
3006 memcpy(to, from, a);
3007 memcpy(to + a, from + a + pop, b);
3008 sg_set_page(sge, page, a + b, 0);
3009 put_page(orig);
3010 }
3011 pop = 0;
3012 } else if (pop >= sge->length - a) {
3013 pop -= (sge->length - a);
3014 sge->length = a;
3015 }
3016 }
3017
3018 /* From above the current layout _must_ be as follows,
3019 *
3020 * -| offset
3021 * -| start
3022 *
3023 * |---- pop ---|---------------- b ------------|
3024 * |____________________________________________| length
3025 *
3026 * Offset and start of the current msg elem are equal because in the
3027 * previous case we handled offset != start and either consumed the
3028 * entire element and advanced to the next element OR pop == 0.
3029 *
3030 * Two cases to handle here are first pop is less than the length
3031 * leaving some remainder b above. Simply adjust the element's layout
3032 * in this case. Or pop >= length of the element so that b = 0. In this
3033 * case advance to next element decrementing pop.
3034 */
3035 while (pop) {
3036 struct scatterlist *sge = sk_msg_elem(msg, i);
3037
3038 if (pop < sge->length) {
3039 sge->length -= pop;
3040 sge->offset += pop;
3041 pop = 0;
3042 } else {
3043 pop -= sge->length;
3044 sk_msg_shift_left(msg, i);
3045 }
3046 sk_msg_iter_var_next(i);
3047 }
3048
3049 sk_mem_uncharge(msg->sk, len - pop);
3050 msg->sg.size -= (len - pop);
3051 sk_msg_reset_curr(msg);
3052 sk_msg_compute_data_pointers(msg);
3053 return 0;
3054 }
3055
3056 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3057 .func = bpf_msg_pop_data,
3058 .gpl_only = false,
3059 .ret_type = RET_INTEGER,
3060 .arg1_type = ARG_PTR_TO_CTX,
3061 .arg2_type = ARG_ANYTHING,
3062 .arg3_type = ARG_ANYTHING,
3063 .arg4_type = ARG_ANYTHING,
3064 };
3065
3066 #ifdef CONFIG_CGROUP_NET_CLASSID
BPF_CALL_0(bpf_get_cgroup_classid_curr)3067 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3068 {
3069 return __task_get_classid(current);
3070 }
3071
3072 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3073 .func = bpf_get_cgroup_classid_curr,
3074 .gpl_only = false,
3075 .ret_type = RET_INTEGER,
3076 };
3077
BPF_CALL_1(bpf_skb_cgroup_classid,const struct sk_buff *,skb)3078 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3079 {
3080 struct sock *sk = skb_to_full_sk(skb);
3081
3082 if (!sk || !sk_fullsock(sk))
3083 return 0;
3084
3085 return sock_cgroup_classid(&sk->sk_cgrp_data);
3086 }
3087
3088 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3089 .func = bpf_skb_cgroup_classid,
3090 .gpl_only = false,
3091 .ret_type = RET_INTEGER,
3092 .arg1_type = ARG_PTR_TO_CTX,
3093 };
3094 #endif
3095
BPF_CALL_1(bpf_get_cgroup_classid,const struct sk_buff *,skb)3096 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3097 {
3098 return task_get_classid(skb);
3099 }
3100
3101 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3102 .func = bpf_get_cgroup_classid,
3103 .gpl_only = false,
3104 .ret_type = RET_INTEGER,
3105 .arg1_type = ARG_PTR_TO_CTX,
3106 };
3107
BPF_CALL_1(bpf_get_route_realm,const struct sk_buff *,skb)3108 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3109 {
3110 return dst_tclassid(skb);
3111 }
3112
3113 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3114 .func = bpf_get_route_realm,
3115 .gpl_only = false,
3116 .ret_type = RET_INTEGER,
3117 .arg1_type = ARG_PTR_TO_CTX,
3118 };
3119
BPF_CALL_1(bpf_get_hash_recalc,struct sk_buff *,skb)3120 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3121 {
3122 /* If skb_clear_hash() was called due to mangling, we can
3123 * trigger SW recalculation here. Later access to hash
3124 * can then use the inline skb->hash via context directly
3125 * instead of calling this helper again.
3126 */
3127 return skb_get_hash(skb);
3128 }
3129
3130 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3131 .func = bpf_get_hash_recalc,
3132 .gpl_only = false,
3133 .ret_type = RET_INTEGER,
3134 .arg1_type = ARG_PTR_TO_CTX,
3135 };
3136
BPF_CALL_1(bpf_set_hash_invalid,struct sk_buff *,skb)3137 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3138 {
3139 /* After all direct packet write, this can be used once for
3140 * triggering a lazy recalc on next skb_get_hash() invocation.
3141 */
3142 skb_clear_hash(skb);
3143 return 0;
3144 }
3145
3146 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3147 .func = bpf_set_hash_invalid,
3148 .gpl_only = false,
3149 .ret_type = RET_INTEGER,
3150 .arg1_type = ARG_PTR_TO_CTX,
3151 };
3152
BPF_CALL_2(bpf_set_hash,struct sk_buff *,skb,u32,hash)3153 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3154 {
3155 /* Set user specified hash as L4(+), so that it gets returned
3156 * on skb_get_hash() call unless BPF prog later on triggers a
3157 * skb_clear_hash().
3158 */
3159 __skb_set_sw_hash(skb, hash, true);
3160 return 0;
3161 }
3162
3163 static const struct bpf_func_proto bpf_set_hash_proto = {
3164 .func = bpf_set_hash,
3165 .gpl_only = false,
3166 .ret_type = RET_INTEGER,
3167 .arg1_type = ARG_PTR_TO_CTX,
3168 .arg2_type = ARG_ANYTHING,
3169 };
3170
BPF_CALL_3(bpf_skb_vlan_push,struct sk_buff *,skb,__be16,vlan_proto,u16,vlan_tci)3171 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3172 u16, vlan_tci)
3173 {
3174 int ret;
3175
3176 if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3177 vlan_proto != htons(ETH_P_8021AD)))
3178 vlan_proto = htons(ETH_P_8021Q);
3179
3180 bpf_push_mac_rcsum(skb);
3181 ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3182 bpf_pull_mac_rcsum(skb);
3183
3184 bpf_compute_data_pointers(skb);
3185 return ret;
3186 }
3187
3188 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3189 .func = bpf_skb_vlan_push,
3190 .gpl_only = false,
3191 .ret_type = RET_INTEGER,
3192 .arg1_type = ARG_PTR_TO_CTX,
3193 .arg2_type = ARG_ANYTHING,
3194 .arg3_type = ARG_ANYTHING,
3195 };
3196
BPF_CALL_1(bpf_skb_vlan_pop,struct sk_buff *,skb)3197 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3198 {
3199 int ret;
3200
3201 bpf_push_mac_rcsum(skb);
3202 ret = skb_vlan_pop(skb);
3203 bpf_pull_mac_rcsum(skb);
3204
3205 bpf_compute_data_pointers(skb);
3206 return ret;
3207 }
3208
3209 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3210 .func = bpf_skb_vlan_pop,
3211 .gpl_only = false,
3212 .ret_type = RET_INTEGER,
3213 .arg1_type = ARG_PTR_TO_CTX,
3214 };
3215
bpf_skb_generic_push(struct sk_buff * skb,u32 off,u32 len)3216 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3217 {
3218 /* Caller already did skb_cow() with len as headroom,
3219 * so no need to do it here.
3220 */
3221 skb_push(skb, len);
3222 memmove(skb->data, skb->data + len, off);
3223 memset(skb->data + off, 0, len);
3224
3225 /* No skb_postpush_rcsum(skb, skb->data + off, len)
3226 * needed here as it does not change the skb->csum
3227 * result for checksum complete when summing over
3228 * zeroed blocks.
3229 */
3230 return 0;
3231 }
3232
bpf_skb_generic_pop(struct sk_buff * skb,u32 off,u32 len)3233 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3234 {
3235 void *old_data;
3236
3237 /* skb_ensure_writable() is not needed here, as we're
3238 * already working on an uncloned skb.
3239 */
3240 if (unlikely(!pskb_may_pull(skb, off + len)))
3241 return -ENOMEM;
3242
3243 old_data = skb->data;
3244 __skb_pull(skb, len);
3245 skb_postpull_rcsum(skb, old_data + off, len);
3246 memmove(skb->data, old_data, off);
3247
3248 return 0;
3249 }
3250
bpf_skb_net_hdr_push(struct sk_buff * skb,u32 off,u32 len)3251 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3252 {
3253 bool trans_same = skb->transport_header == skb->network_header;
3254 int ret;
3255
3256 /* There's no need for __skb_push()/__skb_pull() pair to
3257 * get to the start of the mac header as we're guaranteed
3258 * to always start from here under eBPF.
3259 */
3260 ret = bpf_skb_generic_push(skb, off, len);
3261 if (likely(!ret)) {
3262 skb->mac_header -= len;
3263 skb->network_header -= len;
3264 if (trans_same)
3265 skb->transport_header = skb->network_header;
3266 }
3267
3268 return ret;
3269 }
3270
bpf_skb_net_hdr_pop(struct sk_buff * skb,u32 off,u32 len)3271 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3272 {
3273 bool trans_same = skb->transport_header == skb->network_header;
3274 int ret;
3275
3276 /* Same here, __skb_push()/__skb_pull() pair not needed. */
3277 ret = bpf_skb_generic_pop(skb, off, len);
3278 if (likely(!ret)) {
3279 skb->mac_header += len;
3280 skb->network_header += len;
3281 if (trans_same)
3282 skb->transport_header = skb->network_header;
3283 }
3284
3285 return ret;
3286 }
3287
bpf_skb_proto_4_to_6(struct sk_buff * skb)3288 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3289 {
3290 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3291 u32 off = skb_mac_header_len(skb);
3292 int ret;
3293
3294 ret = skb_cow(skb, len_diff);
3295 if (unlikely(ret < 0))
3296 return ret;
3297
3298 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3299 if (unlikely(ret < 0))
3300 return ret;
3301
3302 if (skb_is_gso(skb)) {
3303 struct skb_shared_info *shinfo = skb_shinfo(skb);
3304
3305 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3306 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3307 shinfo->gso_type &= ~SKB_GSO_TCPV4;
3308 shinfo->gso_type |= SKB_GSO_TCPV6;
3309 }
3310 }
3311
3312 skb->protocol = htons(ETH_P_IPV6);
3313 skb_clear_hash(skb);
3314
3315 return 0;
3316 }
3317
bpf_skb_proto_6_to_4(struct sk_buff * skb)3318 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3319 {
3320 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3321 u32 off = skb_mac_header_len(skb);
3322 int ret;
3323
3324 ret = skb_unclone(skb, GFP_ATOMIC);
3325 if (unlikely(ret < 0))
3326 return ret;
3327
3328 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3329 if (unlikely(ret < 0))
3330 return ret;
3331
3332 if (skb_is_gso(skb)) {
3333 struct skb_shared_info *shinfo = skb_shinfo(skb);
3334
3335 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3336 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3337 shinfo->gso_type &= ~SKB_GSO_TCPV6;
3338 shinfo->gso_type |= SKB_GSO_TCPV4;
3339 }
3340 }
3341
3342 skb->protocol = htons(ETH_P_IP);
3343 skb_clear_hash(skb);
3344
3345 return 0;
3346 }
3347
bpf_skb_proto_xlat(struct sk_buff * skb,__be16 to_proto)3348 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3349 {
3350 __be16 from_proto = skb->protocol;
3351
3352 if (from_proto == htons(ETH_P_IP) &&
3353 to_proto == htons(ETH_P_IPV6))
3354 return bpf_skb_proto_4_to_6(skb);
3355
3356 if (from_proto == htons(ETH_P_IPV6) &&
3357 to_proto == htons(ETH_P_IP))
3358 return bpf_skb_proto_6_to_4(skb);
3359
3360 return -ENOTSUPP;
3361 }
3362
BPF_CALL_3(bpf_skb_change_proto,struct sk_buff *,skb,__be16,proto,u64,flags)3363 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3364 u64, flags)
3365 {
3366 int ret;
3367
3368 if (unlikely(flags))
3369 return -EINVAL;
3370
3371 /* General idea is that this helper does the basic groundwork
3372 * needed for changing the protocol, and eBPF program fills the
3373 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3374 * and other helpers, rather than passing a raw buffer here.
3375 *
3376 * The rationale is to keep this minimal and without a need to
3377 * deal with raw packet data. F.e. even if we would pass buffers
3378 * here, the program still needs to call the bpf_lX_csum_replace()
3379 * helpers anyway. Plus, this way we keep also separation of
3380 * concerns, since f.e. bpf_skb_store_bytes() should only take
3381 * care of stores.
3382 *
3383 * Currently, additional options and extension header space are
3384 * not supported, but flags register is reserved so we can adapt
3385 * that. For offloads, we mark packet as dodgy, so that headers
3386 * need to be verified first.
3387 */
3388 ret = bpf_skb_proto_xlat(skb, proto);
3389 bpf_compute_data_pointers(skb);
3390 return ret;
3391 }
3392
3393 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3394 .func = bpf_skb_change_proto,
3395 .gpl_only = false,
3396 .ret_type = RET_INTEGER,
3397 .arg1_type = ARG_PTR_TO_CTX,
3398 .arg2_type = ARG_ANYTHING,
3399 .arg3_type = ARG_ANYTHING,
3400 };
3401
BPF_CALL_2(bpf_skb_change_type,struct sk_buff *,skb,u32,pkt_type)3402 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3403 {
3404 /* We only allow a restricted subset to be changed for now. */
3405 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3406 !skb_pkt_type_ok(pkt_type)))
3407 return -EINVAL;
3408
3409 skb->pkt_type = pkt_type;
3410 return 0;
3411 }
3412
3413 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3414 .func = bpf_skb_change_type,
3415 .gpl_only = false,
3416 .ret_type = RET_INTEGER,
3417 .arg1_type = ARG_PTR_TO_CTX,
3418 .arg2_type = ARG_ANYTHING,
3419 };
3420
bpf_skb_net_base_len(const struct sk_buff * skb)3421 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3422 {
3423 switch (skb->protocol) {
3424 case htons(ETH_P_IP):
3425 return sizeof(struct iphdr);
3426 case htons(ETH_P_IPV6):
3427 return sizeof(struct ipv6hdr);
3428 default:
3429 return ~0U;
3430 }
3431 }
3432
3433 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3434 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3435
3436 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3437 BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3438
3439 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \
3440 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3441 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3442 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3443 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3444 BPF_F_ADJ_ROOM_ENCAP_L2( \
3445 BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3446 BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3447
bpf_skb_net_grow(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3448 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3449 u64 flags)
3450 {
3451 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3452 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3453 u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3454 unsigned int gso_type = SKB_GSO_DODGY;
3455 int ret;
3456
3457 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3458 /* udp gso_size delineates datagrams, only allow if fixed */
3459 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3460 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3461 return -ENOTSUPP;
3462 }
3463
3464 ret = skb_cow_head(skb, len_diff);
3465 if (unlikely(ret < 0))
3466 return ret;
3467
3468 if (encap) {
3469 if (skb->protocol != htons(ETH_P_IP) &&
3470 skb->protocol != htons(ETH_P_IPV6))
3471 return -ENOTSUPP;
3472
3473 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3474 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3475 return -EINVAL;
3476
3477 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3478 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3479 return -EINVAL;
3480
3481 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3482 inner_mac_len < ETH_HLEN)
3483 return -EINVAL;
3484
3485 if (skb->encapsulation)
3486 return -EALREADY;
3487
3488 mac_len = skb->network_header - skb->mac_header;
3489 inner_net = skb->network_header;
3490 if (inner_mac_len > len_diff)
3491 return -EINVAL;
3492 inner_trans = skb->transport_header;
3493 }
3494
3495 ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3496 if (unlikely(ret < 0))
3497 return ret;
3498
3499 if (encap) {
3500 skb->inner_mac_header = inner_net - inner_mac_len;
3501 skb->inner_network_header = inner_net;
3502 skb->inner_transport_header = inner_trans;
3503
3504 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3505 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3506 else
3507 skb_set_inner_protocol(skb, skb->protocol);
3508
3509 skb->encapsulation = 1;
3510 skb_set_network_header(skb, mac_len);
3511
3512 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3513 gso_type |= SKB_GSO_UDP_TUNNEL;
3514 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3515 gso_type |= SKB_GSO_GRE;
3516 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3517 gso_type |= SKB_GSO_IPXIP6;
3518 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3519 gso_type |= SKB_GSO_IPXIP4;
3520
3521 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3522 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3523 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3524 sizeof(struct ipv6hdr) :
3525 sizeof(struct iphdr);
3526
3527 skb_set_transport_header(skb, mac_len + nh_len);
3528 }
3529
3530 /* Match skb->protocol to new outer l3 protocol */
3531 if (skb->protocol == htons(ETH_P_IP) &&
3532 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3533 skb->protocol = htons(ETH_P_IPV6);
3534 else if (skb->protocol == htons(ETH_P_IPV6) &&
3535 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3536 skb->protocol = htons(ETH_P_IP);
3537 }
3538
3539 if (skb_is_gso(skb)) {
3540 struct skb_shared_info *shinfo = skb_shinfo(skb);
3541
3542 /* Due to header grow, MSS needs to be downgraded. */
3543 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3544 skb_decrease_gso_size(shinfo, len_diff);
3545
3546 /* Header must be checked, and gso_segs recomputed. */
3547 shinfo->gso_type |= gso_type;
3548 shinfo->gso_segs = 0;
3549 }
3550
3551 return 0;
3552 }
3553
bpf_skb_net_shrink(struct sk_buff * skb,u32 off,u32 len_diff,u64 flags)3554 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3555 u64 flags)
3556 {
3557 int ret;
3558
3559 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3560 BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3561 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3562 return -EINVAL;
3563
3564 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3565 /* udp gso_size delineates datagrams, only allow if fixed */
3566 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3567 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3568 return -ENOTSUPP;
3569 }
3570
3571 ret = skb_unclone(skb, GFP_ATOMIC);
3572 if (unlikely(ret < 0))
3573 return ret;
3574
3575 ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3576 if (unlikely(ret < 0))
3577 return ret;
3578
3579 /* Match skb->protocol to new outer l3 protocol */
3580 if (skb->protocol == htons(ETH_P_IP) &&
3581 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3582 skb->protocol = htons(ETH_P_IPV6);
3583 else if (skb->protocol == htons(ETH_P_IPV6) &&
3584 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3585 skb->protocol = htons(ETH_P_IP);
3586
3587 if (skb_is_gso(skb)) {
3588 struct skb_shared_info *shinfo = skb_shinfo(skb);
3589
3590 /* Due to header shrink, MSS can be upgraded. */
3591 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3592 skb_increase_gso_size(shinfo, len_diff);
3593
3594 /* Header must be checked, and gso_segs recomputed. */
3595 shinfo->gso_type |= SKB_GSO_DODGY;
3596 shinfo->gso_segs = 0;
3597 }
3598
3599 return 0;
3600 }
3601
3602 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3603
BPF_CALL_4(sk_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3604 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3605 u32, mode, u64, flags)
3606 {
3607 u32 len_diff_abs = abs(len_diff);
3608 bool shrink = len_diff < 0;
3609 int ret = 0;
3610
3611 if (unlikely(flags || mode))
3612 return -EINVAL;
3613 if (unlikely(len_diff_abs > 0xfffU))
3614 return -EFAULT;
3615
3616 if (!shrink) {
3617 ret = skb_cow(skb, len_diff);
3618 if (unlikely(ret < 0))
3619 return ret;
3620 __skb_push(skb, len_diff_abs);
3621 memset(skb->data, 0, len_diff_abs);
3622 } else {
3623 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3624 return -ENOMEM;
3625 __skb_pull(skb, len_diff_abs);
3626 }
3627 if (tls_sw_has_ctx_rx(skb->sk)) {
3628 struct strp_msg *rxm = strp_msg(skb);
3629
3630 rxm->full_len += len_diff;
3631 }
3632 return ret;
3633 }
3634
3635 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3636 .func = sk_skb_adjust_room,
3637 .gpl_only = false,
3638 .ret_type = RET_INTEGER,
3639 .arg1_type = ARG_PTR_TO_CTX,
3640 .arg2_type = ARG_ANYTHING,
3641 .arg3_type = ARG_ANYTHING,
3642 .arg4_type = ARG_ANYTHING,
3643 };
3644
BPF_CALL_4(bpf_skb_adjust_room,struct sk_buff *,skb,s32,len_diff,u32,mode,u64,flags)3645 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3646 u32, mode, u64, flags)
3647 {
3648 u32 len_cur, len_diff_abs = abs(len_diff);
3649 u32 len_min = bpf_skb_net_base_len(skb);
3650 u32 len_max = BPF_SKB_MAX_LEN;
3651 __be16 proto = skb->protocol;
3652 bool shrink = len_diff < 0;
3653 u32 off;
3654 int ret;
3655
3656 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3657 BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3658 return -EINVAL;
3659 if (unlikely(len_diff_abs > 0xfffU))
3660 return -EFAULT;
3661 if (unlikely(proto != htons(ETH_P_IP) &&
3662 proto != htons(ETH_P_IPV6)))
3663 return -ENOTSUPP;
3664
3665 off = skb_mac_header_len(skb);
3666 switch (mode) {
3667 case BPF_ADJ_ROOM_NET:
3668 off += bpf_skb_net_base_len(skb);
3669 break;
3670 case BPF_ADJ_ROOM_MAC:
3671 break;
3672 default:
3673 return -ENOTSUPP;
3674 }
3675
3676 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3677 if (!shrink)
3678 return -EINVAL;
3679
3680 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3681 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3682 len_min = sizeof(struct iphdr);
3683 break;
3684 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3685 len_min = sizeof(struct ipv6hdr);
3686 break;
3687 default:
3688 return -EINVAL;
3689 }
3690 }
3691
3692 len_cur = skb->len - skb_network_offset(skb);
3693 if ((shrink && (len_diff_abs >= len_cur ||
3694 len_cur - len_diff_abs < len_min)) ||
3695 (!shrink && (skb->len + len_diff_abs > len_max &&
3696 !skb_is_gso(skb))))
3697 return -ENOTSUPP;
3698
3699 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3700 bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3701 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3702 __skb_reset_checksum_unnecessary(skb);
3703
3704 bpf_compute_data_pointers(skb);
3705 return ret;
3706 }
3707
3708 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3709 .func = bpf_skb_adjust_room,
3710 .gpl_only = false,
3711 .ret_type = RET_INTEGER,
3712 .arg1_type = ARG_PTR_TO_CTX,
3713 .arg2_type = ARG_ANYTHING,
3714 .arg3_type = ARG_ANYTHING,
3715 .arg4_type = ARG_ANYTHING,
3716 };
3717
__bpf_skb_min_len(const struct sk_buff * skb)3718 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3719 {
3720 u32 min_len = skb_network_offset(skb);
3721
3722 if (skb_transport_header_was_set(skb))
3723 min_len = skb_transport_offset(skb);
3724 if (skb->ip_summed == CHECKSUM_PARTIAL)
3725 min_len = skb_checksum_start_offset(skb) +
3726 skb->csum_offset + sizeof(__sum16);
3727 return min_len;
3728 }
3729
bpf_skb_grow_rcsum(struct sk_buff * skb,unsigned int new_len)3730 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3731 {
3732 unsigned int old_len = skb->len;
3733 int ret;
3734
3735 ret = __skb_grow_rcsum(skb, new_len);
3736 if (!ret)
3737 memset(skb->data + old_len, 0, new_len - old_len);
3738 return ret;
3739 }
3740
bpf_skb_trim_rcsum(struct sk_buff * skb,unsigned int new_len)3741 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3742 {
3743 return __skb_trim_rcsum(skb, new_len);
3744 }
3745
__bpf_skb_change_tail(struct sk_buff * skb,u32 new_len,u64 flags)3746 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3747 u64 flags)
3748 {
3749 u32 max_len = BPF_SKB_MAX_LEN;
3750 u32 min_len = __bpf_skb_min_len(skb);
3751 int ret;
3752
3753 if (unlikely(flags || new_len > max_len || new_len < min_len))
3754 return -EINVAL;
3755 if (skb->encapsulation)
3756 return -ENOTSUPP;
3757
3758 /* The basic idea of this helper is that it's performing the
3759 * needed work to either grow or trim an skb, and eBPF program
3760 * rewrites the rest via helpers like bpf_skb_store_bytes(),
3761 * bpf_lX_csum_replace() and others rather than passing a raw
3762 * buffer here. This one is a slow path helper and intended
3763 * for replies with control messages.
3764 *
3765 * Like in bpf_skb_change_proto(), we want to keep this rather
3766 * minimal and without protocol specifics so that we are able
3767 * to separate concerns as in bpf_skb_store_bytes() should only
3768 * be the one responsible for writing buffers.
3769 *
3770 * It's really expected to be a slow path operation here for
3771 * control message replies, so we're implicitly linearizing,
3772 * uncloning and drop offloads from the skb by this.
3773 */
3774 ret = __bpf_try_make_writable(skb, skb->len);
3775 if (!ret) {
3776 if (new_len > skb->len)
3777 ret = bpf_skb_grow_rcsum(skb, new_len);
3778 else if (new_len < skb->len)
3779 ret = bpf_skb_trim_rcsum(skb, new_len);
3780 if (!ret && skb_is_gso(skb))
3781 skb_gso_reset(skb);
3782 }
3783 return ret;
3784 }
3785
BPF_CALL_3(bpf_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3786 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3787 u64, flags)
3788 {
3789 int ret = __bpf_skb_change_tail(skb, new_len, flags);
3790
3791 bpf_compute_data_pointers(skb);
3792 return ret;
3793 }
3794
3795 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3796 .func = bpf_skb_change_tail,
3797 .gpl_only = false,
3798 .ret_type = RET_INTEGER,
3799 .arg1_type = ARG_PTR_TO_CTX,
3800 .arg2_type = ARG_ANYTHING,
3801 .arg3_type = ARG_ANYTHING,
3802 };
3803
BPF_CALL_3(sk_skb_change_tail,struct sk_buff *,skb,u32,new_len,u64,flags)3804 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3805 u64, flags)
3806 {
3807 return __bpf_skb_change_tail(skb, new_len, flags);
3808 }
3809
3810 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3811 .func = sk_skb_change_tail,
3812 .gpl_only = false,
3813 .ret_type = RET_INTEGER,
3814 .arg1_type = ARG_PTR_TO_CTX,
3815 .arg2_type = ARG_ANYTHING,
3816 .arg3_type = ARG_ANYTHING,
3817 };
3818
__bpf_skb_change_head(struct sk_buff * skb,u32 head_room,u64 flags)3819 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3820 u64 flags)
3821 {
3822 u32 max_len = BPF_SKB_MAX_LEN;
3823 u32 new_len = skb->len + head_room;
3824 int ret;
3825
3826 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3827 new_len < skb->len))
3828 return -EINVAL;
3829
3830 ret = skb_cow(skb, head_room);
3831 if (likely(!ret)) {
3832 /* Idea for this helper is that we currently only
3833 * allow to expand on mac header. This means that
3834 * skb->protocol network header, etc, stay as is.
3835 * Compared to bpf_skb_change_tail(), we're more
3836 * flexible due to not needing to linearize or
3837 * reset GSO. Intention for this helper is to be
3838 * used by an L3 skb that needs to push mac header
3839 * for redirection into L2 device.
3840 */
3841 __skb_push(skb, head_room);
3842 memset(skb->data, 0, head_room);
3843 skb_reset_mac_header(skb);
3844 skb_reset_mac_len(skb);
3845 }
3846
3847 return ret;
3848 }
3849
BPF_CALL_3(bpf_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3850 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3851 u64, flags)
3852 {
3853 int ret = __bpf_skb_change_head(skb, head_room, flags);
3854
3855 bpf_compute_data_pointers(skb);
3856 return ret;
3857 }
3858
3859 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3860 .func = bpf_skb_change_head,
3861 .gpl_only = false,
3862 .ret_type = RET_INTEGER,
3863 .arg1_type = ARG_PTR_TO_CTX,
3864 .arg2_type = ARG_ANYTHING,
3865 .arg3_type = ARG_ANYTHING,
3866 };
3867
BPF_CALL_3(sk_skb_change_head,struct sk_buff *,skb,u32,head_room,u64,flags)3868 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3869 u64, flags)
3870 {
3871 return __bpf_skb_change_head(skb, head_room, flags);
3872 }
3873
3874 static const struct bpf_func_proto sk_skb_change_head_proto = {
3875 .func = sk_skb_change_head,
3876 .gpl_only = false,
3877 .ret_type = RET_INTEGER,
3878 .arg1_type = ARG_PTR_TO_CTX,
3879 .arg2_type = ARG_ANYTHING,
3880 .arg3_type = ARG_ANYTHING,
3881 };
3882
BPF_CALL_1(bpf_xdp_get_buff_len,struct xdp_buff *,xdp)3883 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3884 {
3885 return xdp_get_buff_len(xdp);
3886 }
3887
3888 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3889 .func = bpf_xdp_get_buff_len,
3890 .gpl_only = false,
3891 .ret_type = RET_INTEGER,
3892 .arg1_type = ARG_PTR_TO_CTX,
3893 };
3894
3895 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3896
3897 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3898 .func = bpf_xdp_get_buff_len,
3899 .gpl_only = false,
3900 .arg1_type = ARG_PTR_TO_BTF_ID,
3901 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0],
3902 };
3903
xdp_get_metalen(const struct xdp_buff * xdp)3904 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3905 {
3906 return xdp_data_meta_unsupported(xdp) ? 0 :
3907 xdp->data - xdp->data_meta;
3908 }
3909
BPF_CALL_2(bpf_xdp_adjust_head,struct xdp_buff *,xdp,int,offset)3910 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3911 {
3912 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3913 unsigned long metalen = xdp_get_metalen(xdp);
3914 void *data_start = xdp_frame_end + metalen;
3915 void *data = xdp->data + offset;
3916
3917 if (unlikely(data < data_start ||
3918 data > xdp->data_end - ETH_HLEN))
3919 return -EINVAL;
3920
3921 if (metalen)
3922 memmove(xdp->data_meta + offset,
3923 xdp->data_meta, metalen);
3924 xdp->data_meta += offset;
3925 xdp->data = data;
3926
3927 return 0;
3928 }
3929
3930 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3931 .func = bpf_xdp_adjust_head,
3932 .gpl_only = false,
3933 .ret_type = RET_INTEGER,
3934 .arg1_type = ARG_PTR_TO_CTX,
3935 .arg2_type = ARG_ANYTHING,
3936 };
3937
bpf_xdp_copy_buf(struct xdp_buff * xdp,unsigned long off,void * buf,unsigned long len,bool flush)3938 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3939 void *buf, unsigned long len, bool flush)
3940 {
3941 unsigned long ptr_len, ptr_off = 0;
3942 skb_frag_t *next_frag, *end_frag;
3943 struct skb_shared_info *sinfo;
3944 void *src, *dst;
3945 u8 *ptr_buf;
3946
3947 if (likely(xdp->data_end - xdp->data >= off + len)) {
3948 src = flush ? buf : xdp->data + off;
3949 dst = flush ? xdp->data + off : buf;
3950 memcpy(dst, src, len);
3951 return;
3952 }
3953
3954 sinfo = xdp_get_shared_info_from_buff(xdp);
3955 end_frag = &sinfo->frags[sinfo->nr_frags];
3956 next_frag = &sinfo->frags[0];
3957
3958 ptr_len = xdp->data_end - xdp->data;
3959 ptr_buf = xdp->data;
3960
3961 while (true) {
3962 if (off < ptr_off + ptr_len) {
3963 unsigned long copy_off = off - ptr_off;
3964 unsigned long copy_len = min(len, ptr_len - copy_off);
3965
3966 src = flush ? buf : ptr_buf + copy_off;
3967 dst = flush ? ptr_buf + copy_off : buf;
3968 memcpy(dst, src, copy_len);
3969
3970 off += copy_len;
3971 len -= copy_len;
3972 buf += copy_len;
3973 }
3974
3975 if (!len || next_frag == end_frag)
3976 break;
3977
3978 ptr_off += ptr_len;
3979 ptr_buf = skb_frag_address(next_frag);
3980 ptr_len = skb_frag_size(next_frag);
3981 next_frag++;
3982 }
3983 }
3984
bpf_xdp_pointer(struct xdp_buff * xdp,u32 offset,u32 len)3985 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3986 {
3987 u32 size = xdp->data_end - xdp->data;
3988 struct skb_shared_info *sinfo;
3989 void *addr = xdp->data;
3990 int i;
3991
3992 if (unlikely(offset > 0xffff || len > 0xffff))
3993 return ERR_PTR(-EFAULT);
3994
3995 if (unlikely(offset + len > xdp_get_buff_len(xdp)))
3996 return ERR_PTR(-EINVAL);
3997
3998 if (likely(offset < size)) /* linear area */
3999 goto out;
4000
4001 sinfo = xdp_get_shared_info_from_buff(xdp);
4002 offset -= size;
4003 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
4004 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
4005
4006 if (offset < frag_size) {
4007 addr = skb_frag_address(&sinfo->frags[i]);
4008 size = frag_size;
4009 break;
4010 }
4011 offset -= frag_size;
4012 }
4013 out:
4014 return offset + len <= size ? addr + offset : NULL;
4015 }
4016
BPF_CALL_4(bpf_xdp_load_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4017 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
4018 void *, buf, u32, len)
4019 {
4020 void *ptr;
4021
4022 ptr = bpf_xdp_pointer(xdp, offset, len);
4023 if (IS_ERR(ptr))
4024 return PTR_ERR(ptr);
4025
4026 if (!ptr)
4027 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
4028 else
4029 memcpy(buf, ptr, len);
4030
4031 return 0;
4032 }
4033
4034 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
4035 .func = bpf_xdp_load_bytes,
4036 .gpl_only = false,
4037 .ret_type = RET_INTEGER,
4038 .arg1_type = ARG_PTR_TO_CTX,
4039 .arg2_type = ARG_ANYTHING,
4040 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4041 .arg4_type = ARG_CONST_SIZE,
4042 };
4043
__bpf_xdp_load_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4044 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4045 {
4046 return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4047 }
4048
BPF_CALL_4(bpf_xdp_store_bytes,struct xdp_buff *,xdp,u32,offset,void *,buf,u32,len)4049 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4050 void *, buf, u32, len)
4051 {
4052 void *ptr;
4053
4054 ptr = bpf_xdp_pointer(xdp, offset, len);
4055 if (IS_ERR(ptr))
4056 return PTR_ERR(ptr);
4057
4058 if (!ptr)
4059 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4060 else
4061 memcpy(ptr, buf, len);
4062
4063 return 0;
4064 }
4065
4066 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4067 .func = bpf_xdp_store_bytes,
4068 .gpl_only = false,
4069 .ret_type = RET_INTEGER,
4070 .arg1_type = ARG_PTR_TO_CTX,
4071 .arg2_type = ARG_ANYTHING,
4072 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
4073 .arg4_type = ARG_CONST_SIZE,
4074 };
4075
__bpf_xdp_store_bytes(struct xdp_buff * xdp,u32 offset,void * buf,u32 len)4076 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4077 {
4078 return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4079 }
4080
bpf_xdp_frags_increase_tail(struct xdp_buff * xdp,int offset)4081 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4082 {
4083 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4084 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4085 struct xdp_rxq_info *rxq = xdp->rxq;
4086 unsigned int tailroom;
4087
4088 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4089 return -EOPNOTSUPP;
4090
4091 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4092 if (unlikely(offset > tailroom))
4093 return -EINVAL;
4094
4095 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4096 skb_frag_size_add(frag, offset);
4097 sinfo->xdp_frags_size += offset;
4098 if (rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL)
4099 xsk_buff_get_tail(xdp)->data_end += offset;
4100
4101 return 0;
4102 }
4103
bpf_xdp_shrink_data_zc(struct xdp_buff * xdp,int shrink,struct xdp_mem_info * mem_info,bool release)4104 static void bpf_xdp_shrink_data_zc(struct xdp_buff *xdp, int shrink,
4105 struct xdp_mem_info *mem_info, bool release)
4106 {
4107 struct xdp_buff *zc_frag = xsk_buff_get_tail(xdp);
4108
4109 if (release) {
4110 xsk_buff_del_tail(zc_frag);
4111 __xdp_return(NULL, mem_info, false, zc_frag);
4112 } else {
4113 zc_frag->data_end -= shrink;
4114 }
4115 }
4116
bpf_xdp_shrink_data(struct xdp_buff * xdp,skb_frag_t * frag,int shrink)4117 static bool bpf_xdp_shrink_data(struct xdp_buff *xdp, skb_frag_t *frag,
4118 int shrink)
4119 {
4120 struct xdp_mem_info *mem_info = &xdp->rxq->mem;
4121 bool release = skb_frag_size(frag) == shrink;
4122
4123 if (mem_info->type == MEM_TYPE_XSK_BUFF_POOL) {
4124 bpf_xdp_shrink_data_zc(xdp, shrink, mem_info, release);
4125 goto out;
4126 }
4127
4128 if (release) {
4129 struct page *page = skb_frag_page(frag);
4130
4131 __xdp_return(page_address(page), mem_info, false, NULL);
4132 }
4133
4134 out:
4135 return release;
4136 }
4137
bpf_xdp_frags_shrink_tail(struct xdp_buff * xdp,int offset)4138 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4139 {
4140 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4141 int i, n_frags_free = 0, len_free = 0;
4142
4143 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4144 return -EINVAL;
4145
4146 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4147 skb_frag_t *frag = &sinfo->frags[i];
4148 int shrink = min_t(int, offset, skb_frag_size(frag));
4149
4150 len_free += shrink;
4151 offset -= shrink;
4152 if (bpf_xdp_shrink_data(xdp, frag, shrink)) {
4153 n_frags_free++;
4154 } else {
4155 skb_frag_size_sub(frag, shrink);
4156 break;
4157 }
4158 }
4159 sinfo->nr_frags -= n_frags_free;
4160 sinfo->xdp_frags_size -= len_free;
4161
4162 if (unlikely(!sinfo->nr_frags)) {
4163 xdp_buff_clear_frags_flag(xdp);
4164 xdp->data_end -= offset;
4165 }
4166
4167 return 0;
4168 }
4169
BPF_CALL_2(bpf_xdp_adjust_tail,struct xdp_buff *,xdp,int,offset)4170 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4171 {
4172 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4173 void *data_end = xdp->data_end + offset;
4174
4175 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4176 if (offset < 0)
4177 return bpf_xdp_frags_shrink_tail(xdp, -offset);
4178
4179 return bpf_xdp_frags_increase_tail(xdp, offset);
4180 }
4181
4182 /* Notice that xdp_data_hard_end have reserved some tailroom */
4183 if (unlikely(data_end > data_hard_end))
4184 return -EINVAL;
4185
4186 if (unlikely(data_end < xdp->data + ETH_HLEN))
4187 return -EINVAL;
4188
4189 /* Clear memory area on grow, can contain uninit kernel memory */
4190 if (offset > 0)
4191 memset(xdp->data_end, 0, offset);
4192
4193 xdp->data_end = data_end;
4194
4195 return 0;
4196 }
4197
4198 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4199 .func = bpf_xdp_adjust_tail,
4200 .gpl_only = false,
4201 .ret_type = RET_INTEGER,
4202 .arg1_type = ARG_PTR_TO_CTX,
4203 .arg2_type = ARG_ANYTHING,
4204 };
4205
BPF_CALL_2(bpf_xdp_adjust_meta,struct xdp_buff *,xdp,int,offset)4206 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4207 {
4208 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4209 void *meta = xdp->data_meta + offset;
4210 unsigned long metalen = xdp->data - meta;
4211
4212 if (xdp_data_meta_unsupported(xdp))
4213 return -ENOTSUPP;
4214 if (unlikely(meta < xdp_frame_end ||
4215 meta > xdp->data))
4216 return -EINVAL;
4217 if (unlikely(xdp_metalen_invalid(metalen)))
4218 return -EACCES;
4219
4220 xdp->data_meta = meta;
4221
4222 return 0;
4223 }
4224
4225 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4226 .func = bpf_xdp_adjust_meta,
4227 .gpl_only = false,
4228 .ret_type = RET_INTEGER,
4229 .arg1_type = ARG_PTR_TO_CTX,
4230 .arg2_type = ARG_ANYTHING,
4231 };
4232
4233 /**
4234 * DOC: xdp redirect
4235 *
4236 * XDP_REDIRECT works by a three-step process, implemented in the functions
4237 * below:
4238 *
4239 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4240 * of the redirect and store it (along with some other metadata) in a per-CPU
4241 * struct bpf_redirect_info.
4242 *
4243 * 2. When the program returns the XDP_REDIRECT return code, the driver will
4244 * call xdp_do_redirect() which will use the information in struct
4245 * bpf_redirect_info to actually enqueue the frame into a map type-specific
4246 * bulk queue structure.
4247 *
4248 * 3. Before exiting its NAPI poll loop, the driver will call
4249 * xdp_do_flush(), which will flush all the different bulk queues,
4250 * thus completing the redirect. Note that xdp_do_flush() must be
4251 * called before napi_complete_done() in the driver, as the
4252 * XDP_REDIRECT logic relies on being inside a single NAPI instance
4253 * through to the xdp_do_flush() call for RCU protection of all
4254 * in-kernel data structures.
4255 */
4256 /*
4257 * Pointers to the map entries will be kept around for this whole sequence of
4258 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4259 * the core code; instead, the RCU protection relies on everything happening
4260 * inside a single NAPI poll sequence, which means it's between a pair of calls
4261 * to local_bh_disable()/local_bh_enable().
4262 *
4263 * The map entries are marked as __rcu and the map code makes sure to
4264 * dereference those pointers with rcu_dereference_check() in a way that works
4265 * for both sections that to hold an rcu_read_lock() and sections that are
4266 * called from NAPI without a separate rcu_read_lock(). The code below does not
4267 * use RCU annotations, but relies on those in the map code.
4268 */
xdp_do_flush(void)4269 void xdp_do_flush(void)
4270 {
4271 __dev_flush();
4272 __cpu_map_flush();
4273 __xsk_map_flush();
4274 }
4275 EXPORT_SYMBOL_GPL(xdp_do_flush);
4276
4277 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL)
xdp_do_check_flushed(struct napi_struct * napi)4278 void xdp_do_check_flushed(struct napi_struct *napi)
4279 {
4280 bool ret;
4281
4282 ret = dev_check_flush();
4283 ret |= cpu_map_check_flush();
4284 ret |= xsk_map_check_flush();
4285
4286 WARN_ONCE(ret, "Missing xdp_do_flush() invocation after NAPI by %ps\n",
4287 napi->poll);
4288 }
4289 #endif
4290
bpf_clear_redirect_map(struct bpf_map * map)4291 void bpf_clear_redirect_map(struct bpf_map *map)
4292 {
4293 struct bpf_redirect_info *ri;
4294 int cpu;
4295
4296 for_each_possible_cpu(cpu) {
4297 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4298 /* Avoid polluting remote cacheline due to writes if
4299 * not needed. Once we pass this test, we need the
4300 * cmpxchg() to make sure it hasn't been changed in
4301 * the meantime by remote CPU.
4302 */
4303 if (unlikely(READ_ONCE(ri->map) == map))
4304 cmpxchg(&ri->map, map, NULL);
4305 }
4306 }
4307
4308 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4309 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4310
xdp_master_redirect(struct xdp_buff * xdp)4311 u32 xdp_master_redirect(struct xdp_buff *xdp)
4312 {
4313 struct net_device *master, *slave;
4314 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4315
4316 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4317 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4318 if (slave && slave != xdp->rxq->dev) {
4319 /* The target device is different from the receiving device, so
4320 * redirect it to the new device.
4321 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4322 * drivers to unmap the packet from their rx ring.
4323 */
4324 ri->tgt_index = slave->ifindex;
4325 ri->map_id = INT_MAX;
4326 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4327 return XDP_REDIRECT;
4328 }
4329 return XDP_TX;
4330 }
4331 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4332
__xdp_do_redirect_xsk(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4333 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4334 struct net_device *dev,
4335 struct xdp_buff *xdp,
4336 struct bpf_prog *xdp_prog)
4337 {
4338 enum bpf_map_type map_type = ri->map_type;
4339 void *fwd = ri->tgt_value;
4340 u32 map_id = ri->map_id;
4341 int err;
4342
4343 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4344 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4345
4346 err = __xsk_map_redirect(fwd, xdp);
4347 if (unlikely(err))
4348 goto err;
4349
4350 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4351 return 0;
4352 err:
4353 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4354 return err;
4355 }
4356
__xdp_do_redirect_frame(struct bpf_redirect_info * ri,struct net_device * dev,struct xdp_frame * xdpf,struct bpf_prog * xdp_prog)4357 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4358 struct net_device *dev,
4359 struct xdp_frame *xdpf,
4360 struct bpf_prog *xdp_prog)
4361 {
4362 enum bpf_map_type map_type = ri->map_type;
4363 void *fwd = ri->tgt_value;
4364 u32 map_id = ri->map_id;
4365 u32 flags = ri->flags;
4366 struct bpf_map *map;
4367 int err;
4368
4369 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4370 ri->flags = 0;
4371 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4372
4373 if (unlikely(!xdpf)) {
4374 err = -EOVERFLOW;
4375 goto err;
4376 }
4377
4378 switch (map_type) {
4379 case BPF_MAP_TYPE_DEVMAP:
4380 fallthrough;
4381 case BPF_MAP_TYPE_DEVMAP_HASH:
4382 if (unlikely(flags & BPF_F_BROADCAST)) {
4383 map = READ_ONCE(ri->map);
4384
4385 /* The map pointer is cleared when the map is being torn
4386 * down by bpf_clear_redirect_map()
4387 */
4388 if (unlikely(!map)) {
4389 err = -ENOENT;
4390 break;
4391 }
4392
4393 WRITE_ONCE(ri->map, NULL);
4394 err = dev_map_enqueue_multi(xdpf, dev, map,
4395 flags & BPF_F_EXCLUDE_INGRESS);
4396 } else {
4397 err = dev_map_enqueue(fwd, xdpf, dev);
4398 }
4399 break;
4400 case BPF_MAP_TYPE_CPUMAP:
4401 err = cpu_map_enqueue(fwd, xdpf, dev);
4402 break;
4403 case BPF_MAP_TYPE_UNSPEC:
4404 if (map_id == INT_MAX) {
4405 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4406 if (unlikely(!fwd)) {
4407 err = -EINVAL;
4408 break;
4409 }
4410 err = dev_xdp_enqueue(fwd, xdpf, dev);
4411 break;
4412 }
4413 fallthrough;
4414 default:
4415 err = -EBADRQC;
4416 }
4417
4418 if (unlikely(err))
4419 goto err;
4420
4421 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4422 return 0;
4423 err:
4424 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4425 return err;
4426 }
4427
xdp_do_redirect(struct net_device * dev,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4428 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4429 struct bpf_prog *xdp_prog)
4430 {
4431 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4432 enum bpf_map_type map_type = ri->map_type;
4433
4434 if (map_type == BPF_MAP_TYPE_XSKMAP)
4435 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4436
4437 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4438 xdp_prog);
4439 }
4440 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4441
xdp_do_redirect_frame(struct net_device * dev,struct xdp_buff * xdp,struct xdp_frame * xdpf,struct bpf_prog * xdp_prog)4442 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4443 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4444 {
4445 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4446 enum bpf_map_type map_type = ri->map_type;
4447
4448 if (map_type == BPF_MAP_TYPE_XSKMAP)
4449 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4450
4451 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4452 }
4453 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4454
xdp_do_generic_redirect_map(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog,void * fwd,enum bpf_map_type map_type,u32 map_id,u32 flags)4455 static int xdp_do_generic_redirect_map(struct net_device *dev,
4456 struct sk_buff *skb,
4457 struct xdp_buff *xdp,
4458 struct bpf_prog *xdp_prog, void *fwd,
4459 enum bpf_map_type map_type, u32 map_id,
4460 u32 flags)
4461 {
4462 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4463 struct bpf_map *map;
4464 int err;
4465
4466 switch (map_type) {
4467 case BPF_MAP_TYPE_DEVMAP:
4468 fallthrough;
4469 case BPF_MAP_TYPE_DEVMAP_HASH:
4470 if (unlikely(flags & BPF_F_BROADCAST)) {
4471 map = READ_ONCE(ri->map);
4472
4473 /* The map pointer is cleared when the map is being torn
4474 * down by bpf_clear_redirect_map()
4475 */
4476 if (unlikely(!map)) {
4477 err = -ENOENT;
4478 break;
4479 }
4480
4481 WRITE_ONCE(ri->map, NULL);
4482 err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4483 flags & BPF_F_EXCLUDE_INGRESS);
4484 } else {
4485 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4486 }
4487 if (unlikely(err))
4488 goto err;
4489 break;
4490 case BPF_MAP_TYPE_XSKMAP:
4491 err = xsk_generic_rcv(fwd, xdp);
4492 if (err)
4493 goto err;
4494 consume_skb(skb);
4495 break;
4496 case BPF_MAP_TYPE_CPUMAP:
4497 err = cpu_map_generic_redirect(fwd, skb);
4498 if (unlikely(err))
4499 goto err;
4500 break;
4501 default:
4502 err = -EBADRQC;
4503 goto err;
4504 }
4505
4506 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4507 return 0;
4508 err:
4509 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4510 return err;
4511 }
4512
xdp_do_generic_redirect(struct net_device * dev,struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4513 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4514 struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4515 {
4516 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4517 enum bpf_map_type map_type = ri->map_type;
4518 void *fwd = ri->tgt_value;
4519 u32 map_id = ri->map_id;
4520 u32 flags = ri->flags;
4521 int err;
4522
4523 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4524 ri->flags = 0;
4525 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4526
4527 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4528 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4529 if (unlikely(!fwd)) {
4530 err = -EINVAL;
4531 goto err;
4532 }
4533
4534 err = xdp_ok_fwd_dev(fwd, skb->len);
4535 if (unlikely(err))
4536 goto err;
4537
4538 skb->dev = fwd;
4539 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4540 generic_xdp_tx(skb, xdp_prog);
4541 return 0;
4542 }
4543
4544 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id, flags);
4545 err:
4546 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4547 return err;
4548 }
4549
BPF_CALL_2(bpf_xdp_redirect,u32,ifindex,u64,flags)4550 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4551 {
4552 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4553
4554 if (unlikely(flags))
4555 return XDP_ABORTED;
4556
4557 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4558 * by map_idr) is used for ifindex based XDP redirect.
4559 */
4560 ri->tgt_index = ifindex;
4561 ri->map_id = INT_MAX;
4562 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4563
4564 return XDP_REDIRECT;
4565 }
4566
4567 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4568 .func = bpf_xdp_redirect,
4569 .gpl_only = false,
4570 .ret_type = RET_INTEGER,
4571 .arg1_type = ARG_ANYTHING,
4572 .arg2_type = ARG_ANYTHING,
4573 };
4574
BPF_CALL_3(bpf_xdp_redirect_map,struct bpf_map *,map,u64,key,u64,flags)4575 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4576 u64, flags)
4577 {
4578 return map->ops->map_redirect(map, key, flags);
4579 }
4580
4581 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4582 .func = bpf_xdp_redirect_map,
4583 .gpl_only = false,
4584 .ret_type = RET_INTEGER,
4585 .arg1_type = ARG_CONST_MAP_PTR,
4586 .arg2_type = ARG_ANYTHING,
4587 .arg3_type = ARG_ANYTHING,
4588 };
4589
bpf_skb_copy(void * dst_buff,const void * skb,unsigned long off,unsigned long len)4590 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4591 unsigned long off, unsigned long len)
4592 {
4593 void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4594
4595 if (unlikely(!ptr))
4596 return len;
4597 if (ptr != dst_buff)
4598 memcpy(dst_buff, ptr, len);
4599
4600 return 0;
4601 }
4602
BPF_CALL_5(bpf_skb_event_output,struct sk_buff *,skb,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)4603 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4604 u64, flags, void *, meta, u64, meta_size)
4605 {
4606 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4607
4608 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4609 return -EINVAL;
4610 if (unlikely(!skb || skb_size > skb->len))
4611 return -EFAULT;
4612
4613 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4614 bpf_skb_copy);
4615 }
4616
4617 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4618 .func = bpf_skb_event_output,
4619 .gpl_only = true,
4620 .ret_type = RET_INTEGER,
4621 .arg1_type = ARG_PTR_TO_CTX,
4622 .arg2_type = ARG_CONST_MAP_PTR,
4623 .arg3_type = ARG_ANYTHING,
4624 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4625 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4626 };
4627
4628 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4629
4630 const struct bpf_func_proto bpf_skb_output_proto = {
4631 .func = bpf_skb_event_output,
4632 .gpl_only = true,
4633 .ret_type = RET_INTEGER,
4634 .arg1_type = ARG_PTR_TO_BTF_ID,
4635 .arg1_btf_id = &bpf_skb_output_btf_ids[0],
4636 .arg2_type = ARG_CONST_MAP_PTR,
4637 .arg3_type = ARG_ANYTHING,
4638 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4639 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
4640 };
4641
bpf_tunnel_key_af(u64 flags)4642 static unsigned short bpf_tunnel_key_af(u64 flags)
4643 {
4644 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4645 }
4646
BPF_CALL_4(bpf_skb_get_tunnel_key,struct sk_buff *,skb,struct bpf_tunnel_key *,to,u32,size,u64,flags)4647 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4648 u32, size, u64, flags)
4649 {
4650 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4651 u8 compat[sizeof(struct bpf_tunnel_key)];
4652 void *to_orig = to;
4653 int err;
4654
4655 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4656 BPF_F_TUNINFO_FLAGS)))) {
4657 err = -EINVAL;
4658 goto err_clear;
4659 }
4660 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4661 err = -EPROTO;
4662 goto err_clear;
4663 }
4664 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4665 err = -EINVAL;
4666 switch (size) {
4667 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4668 case offsetof(struct bpf_tunnel_key, tunnel_label):
4669 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4670 goto set_compat;
4671 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4672 /* Fixup deprecated structure layouts here, so we have
4673 * a common path later on.
4674 */
4675 if (ip_tunnel_info_af(info) != AF_INET)
4676 goto err_clear;
4677 set_compat:
4678 to = (struct bpf_tunnel_key *)compat;
4679 break;
4680 default:
4681 goto err_clear;
4682 }
4683 }
4684
4685 to->tunnel_id = be64_to_cpu(info->key.tun_id);
4686 to->tunnel_tos = info->key.tos;
4687 to->tunnel_ttl = info->key.ttl;
4688 if (flags & BPF_F_TUNINFO_FLAGS)
4689 to->tunnel_flags = ip_tunnel_flags_to_be16(info->key.tun_flags);
4690 else
4691 to->tunnel_ext = 0;
4692
4693 if (flags & BPF_F_TUNINFO_IPV6) {
4694 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4695 sizeof(to->remote_ipv6));
4696 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4697 sizeof(to->local_ipv6));
4698 to->tunnel_label = be32_to_cpu(info->key.label);
4699 } else {
4700 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4701 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4702 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4703 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4704 to->tunnel_label = 0;
4705 }
4706
4707 if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4708 memcpy(to_orig, to, size);
4709
4710 return 0;
4711 err_clear:
4712 memset(to_orig, 0, size);
4713 return err;
4714 }
4715
4716 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4717 .func = bpf_skb_get_tunnel_key,
4718 .gpl_only = false,
4719 .ret_type = RET_INTEGER,
4720 .arg1_type = ARG_PTR_TO_CTX,
4721 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4722 .arg3_type = ARG_CONST_SIZE,
4723 .arg4_type = ARG_ANYTHING,
4724 };
4725
BPF_CALL_3(bpf_skb_get_tunnel_opt,struct sk_buff *,skb,u8 *,to,u32,size)4726 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4727 {
4728 const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4729 int err;
4730
4731 if (unlikely(!info ||
4732 !ip_tunnel_is_options_present(info->key.tun_flags))) {
4733 err = -ENOENT;
4734 goto err_clear;
4735 }
4736 if (unlikely(size < info->options_len)) {
4737 err = -ENOMEM;
4738 goto err_clear;
4739 }
4740
4741 ip_tunnel_info_opts_get(to, info);
4742 if (size > info->options_len)
4743 memset(to + info->options_len, 0, size - info->options_len);
4744
4745 return info->options_len;
4746 err_clear:
4747 memset(to, 0, size);
4748 return err;
4749 }
4750
4751 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4752 .func = bpf_skb_get_tunnel_opt,
4753 .gpl_only = false,
4754 .ret_type = RET_INTEGER,
4755 .arg1_type = ARG_PTR_TO_CTX,
4756 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
4757 .arg3_type = ARG_CONST_SIZE,
4758 };
4759
4760 static struct metadata_dst __percpu *md_dst;
4761
BPF_CALL_4(bpf_skb_set_tunnel_key,struct sk_buff *,skb,const struct bpf_tunnel_key *,from,u32,size,u64,flags)4762 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4763 const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4764 {
4765 struct metadata_dst *md = this_cpu_ptr(md_dst);
4766 u8 compat[sizeof(struct bpf_tunnel_key)];
4767 struct ip_tunnel_info *info;
4768
4769 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4770 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4771 BPF_F_NO_TUNNEL_KEY)))
4772 return -EINVAL;
4773 if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4774 switch (size) {
4775 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4776 case offsetof(struct bpf_tunnel_key, tunnel_label):
4777 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4778 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4779 /* Fixup deprecated structure layouts here, so we have
4780 * a common path later on.
4781 */
4782 memcpy(compat, from, size);
4783 memset(compat + size, 0, sizeof(compat) - size);
4784 from = (const struct bpf_tunnel_key *) compat;
4785 break;
4786 default:
4787 return -EINVAL;
4788 }
4789 }
4790 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4791 from->tunnel_ext))
4792 return -EINVAL;
4793
4794 skb_dst_drop(skb);
4795 dst_hold((struct dst_entry *) md);
4796 skb_dst_set(skb, (struct dst_entry *) md);
4797
4798 info = &md->u.tun_info;
4799 memset(info, 0, sizeof(*info));
4800 info->mode = IP_TUNNEL_INFO_TX;
4801
4802 __set_bit(IP_TUNNEL_NOCACHE_BIT, info->key.tun_flags);
4803 __assign_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, info->key.tun_flags,
4804 flags & BPF_F_DONT_FRAGMENT);
4805 __assign_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags,
4806 !(flags & BPF_F_ZERO_CSUM_TX));
4807 __assign_bit(IP_TUNNEL_SEQ_BIT, info->key.tun_flags,
4808 flags & BPF_F_SEQ_NUMBER);
4809 __assign_bit(IP_TUNNEL_KEY_BIT, info->key.tun_flags,
4810 !(flags & BPF_F_NO_TUNNEL_KEY));
4811
4812 info->key.tun_id = cpu_to_be64(from->tunnel_id);
4813 info->key.tos = from->tunnel_tos;
4814 info->key.ttl = from->tunnel_ttl;
4815
4816 if (flags & BPF_F_TUNINFO_IPV6) {
4817 info->mode |= IP_TUNNEL_INFO_IPV6;
4818 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4819 sizeof(from->remote_ipv6));
4820 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4821 sizeof(from->local_ipv6));
4822 info->key.label = cpu_to_be32(from->tunnel_label) &
4823 IPV6_FLOWLABEL_MASK;
4824 } else {
4825 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4826 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4827 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4828 }
4829
4830 return 0;
4831 }
4832
4833 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4834 .func = bpf_skb_set_tunnel_key,
4835 .gpl_only = false,
4836 .ret_type = RET_INTEGER,
4837 .arg1_type = ARG_PTR_TO_CTX,
4838 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4839 .arg3_type = ARG_CONST_SIZE,
4840 .arg4_type = ARG_ANYTHING,
4841 };
4842
BPF_CALL_3(bpf_skb_set_tunnel_opt,struct sk_buff *,skb,const u8 *,from,u32,size)4843 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4844 const u8 *, from, u32, size)
4845 {
4846 struct ip_tunnel_info *info = skb_tunnel_info(skb);
4847 const struct metadata_dst *md = this_cpu_ptr(md_dst);
4848 IP_TUNNEL_DECLARE_FLAGS(present) = { };
4849
4850 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4851 return -EINVAL;
4852 if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4853 return -ENOMEM;
4854
4855 ip_tunnel_set_options_present(present);
4856 ip_tunnel_info_opts_set(info, from, size, present);
4857
4858 return 0;
4859 }
4860
4861 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4862 .func = bpf_skb_set_tunnel_opt,
4863 .gpl_only = false,
4864 .ret_type = RET_INTEGER,
4865 .arg1_type = ARG_PTR_TO_CTX,
4866 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
4867 .arg3_type = ARG_CONST_SIZE,
4868 };
4869
4870 static const struct bpf_func_proto *
bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)4871 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4872 {
4873 if (!md_dst) {
4874 struct metadata_dst __percpu *tmp;
4875
4876 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4877 METADATA_IP_TUNNEL,
4878 GFP_KERNEL);
4879 if (!tmp)
4880 return NULL;
4881 if (cmpxchg(&md_dst, NULL, tmp))
4882 metadata_dst_free_percpu(tmp);
4883 }
4884
4885 switch (which) {
4886 case BPF_FUNC_skb_set_tunnel_key:
4887 return &bpf_skb_set_tunnel_key_proto;
4888 case BPF_FUNC_skb_set_tunnel_opt:
4889 return &bpf_skb_set_tunnel_opt_proto;
4890 default:
4891 return NULL;
4892 }
4893 }
4894
BPF_CALL_3(bpf_skb_under_cgroup,struct sk_buff *,skb,struct bpf_map *,map,u32,idx)4895 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4896 u32, idx)
4897 {
4898 struct bpf_array *array = container_of(map, struct bpf_array, map);
4899 struct cgroup *cgrp;
4900 struct sock *sk;
4901
4902 sk = skb_to_full_sk(skb);
4903 if (!sk || !sk_fullsock(sk))
4904 return -ENOENT;
4905 if (unlikely(idx >= array->map.max_entries))
4906 return -E2BIG;
4907
4908 cgrp = READ_ONCE(array->ptrs[idx]);
4909 if (unlikely(!cgrp))
4910 return -EAGAIN;
4911
4912 return sk_under_cgroup_hierarchy(sk, cgrp);
4913 }
4914
4915 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4916 .func = bpf_skb_under_cgroup,
4917 .gpl_only = false,
4918 .ret_type = RET_INTEGER,
4919 .arg1_type = ARG_PTR_TO_CTX,
4920 .arg2_type = ARG_CONST_MAP_PTR,
4921 .arg3_type = ARG_ANYTHING,
4922 };
4923
4924 #ifdef CONFIG_SOCK_CGROUP_DATA
__bpf_sk_cgroup_id(struct sock * sk)4925 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4926 {
4927 struct cgroup *cgrp;
4928
4929 sk = sk_to_full_sk(sk);
4930 if (!sk || !sk_fullsock(sk))
4931 return 0;
4932
4933 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4934 return cgroup_id(cgrp);
4935 }
4936
BPF_CALL_1(bpf_skb_cgroup_id,const struct sk_buff *,skb)4937 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4938 {
4939 return __bpf_sk_cgroup_id(skb->sk);
4940 }
4941
4942 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4943 .func = bpf_skb_cgroup_id,
4944 .gpl_only = false,
4945 .ret_type = RET_INTEGER,
4946 .arg1_type = ARG_PTR_TO_CTX,
4947 };
4948
__bpf_sk_ancestor_cgroup_id(struct sock * sk,int ancestor_level)4949 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4950 int ancestor_level)
4951 {
4952 struct cgroup *ancestor;
4953 struct cgroup *cgrp;
4954
4955 sk = sk_to_full_sk(sk);
4956 if (!sk || !sk_fullsock(sk))
4957 return 0;
4958
4959 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4960 ancestor = cgroup_ancestor(cgrp, ancestor_level);
4961 if (!ancestor)
4962 return 0;
4963
4964 return cgroup_id(ancestor);
4965 }
4966
BPF_CALL_2(bpf_skb_ancestor_cgroup_id,const struct sk_buff *,skb,int,ancestor_level)4967 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4968 ancestor_level)
4969 {
4970 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4971 }
4972
4973 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4974 .func = bpf_skb_ancestor_cgroup_id,
4975 .gpl_only = false,
4976 .ret_type = RET_INTEGER,
4977 .arg1_type = ARG_PTR_TO_CTX,
4978 .arg2_type = ARG_ANYTHING,
4979 };
4980
BPF_CALL_1(bpf_sk_cgroup_id,struct sock *,sk)4981 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4982 {
4983 return __bpf_sk_cgroup_id(sk);
4984 }
4985
4986 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4987 .func = bpf_sk_cgroup_id,
4988 .gpl_only = false,
4989 .ret_type = RET_INTEGER,
4990 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4991 };
4992
BPF_CALL_2(bpf_sk_ancestor_cgroup_id,struct sock *,sk,int,ancestor_level)4993 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4994 {
4995 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4996 }
4997
4998 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4999 .func = bpf_sk_ancestor_cgroup_id,
5000 .gpl_only = false,
5001 .ret_type = RET_INTEGER,
5002 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5003 .arg2_type = ARG_ANYTHING,
5004 };
5005 #endif
5006
bpf_xdp_copy(void * dst,const void * ctx,unsigned long off,unsigned long len)5007 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
5008 unsigned long off, unsigned long len)
5009 {
5010 struct xdp_buff *xdp = (struct xdp_buff *)ctx;
5011
5012 bpf_xdp_copy_buf(xdp, off, dst, len, false);
5013 return 0;
5014 }
5015
BPF_CALL_5(bpf_xdp_event_output,struct xdp_buff *,xdp,struct bpf_map *,map,u64,flags,void *,meta,u64,meta_size)5016 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
5017 u64, flags, void *, meta, u64, meta_size)
5018 {
5019 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
5020
5021 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
5022 return -EINVAL;
5023
5024 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
5025 return -EFAULT;
5026
5027 return bpf_event_output(map, flags, meta, meta_size, xdp,
5028 xdp_size, bpf_xdp_copy);
5029 }
5030
5031 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
5032 .func = bpf_xdp_event_output,
5033 .gpl_only = true,
5034 .ret_type = RET_INTEGER,
5035 .arg1_type = ARG_PTR_TO_CTX,
5036 .arg2_type = ARG_CONST_MAP_PTR,
5037 .arg3_type = ARG_ANYTHING,
5038 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5039 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5040 };
5041
5042 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
5043
5044 const struct bpf_func_proto bpf_xdp_output_proto = {
5045 .func = bpf_xdp_event_output,
5046 .gpl_only = true,
5047 .ret_type = RET_INTEGER,
5048 .arg1_type = ARG_PTR_TO_BTF_ID,
5049 .arg1_btf_id = &bpf_xdp_output_btf_ids[0],
5050 .arg2_type = ARG_CONST_MAP_PTR,
5051 .arg3_type = ARG_ANYTHING,
5052 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5053 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
5054 };
5055
BPF_CALL_1(bpf_get_socket_cookie,struct sk_buff *,skb)5056 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
5057 {
5058 return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
5059 }
5060
5061 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
5062 .func = bpf_get_socket_cookie,
5063 .gpl_only = false,
5064 .ret_type = RET_INTEGER,
5065 .arg1_type = ARG_PTR_TO_CTX,
5066 };
5067
BPF_CALL_1(bpf_get_socket_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5068 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5069 {
5070 return __sock_gen_cookie(ctx->sk);
5071 }
5072
5073 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
5074 .func = bpf_get_socket_cookie_sock_addr,
5075 .gpl_only = false,
5076 .ret_type = RET_INTEGER,
5077 .arg1_type = ARG_PTR_TO_CTX,
5078 };
5079
BPF_CALL_1(bpf_get_socket_cookie_sock,struct sock *,ctx)5080 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
5081 {
5082 return __sock_gen_cookie(ctx);
5083 }
5084
5085 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
5086 .func = bpf_get_socket_cookie_sock,
5087 .gpl_only = false,
5088 .ret_type = RET_INTEGER,
5089 .arg1_type = ARG_PTR_TO_CTX,
5090 };
5091
BPF_CALL_1(bpf_get_socket_ptr_cookie,struct sock *,sk)5092 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
5093 {
5094 return sk ? sock_gen_cookie(sk) : 0;
5095 }
5096
5097 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
5098 .func = bpf_get_socket_ptr_cookie,
5099 .gpl_only = false,
5100 .ret_type = RET_INTEGER,
5101 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5102 };
5103
BPF_CALL_1(bpf_get_socket_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5104 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5105 {
5106 return __sock_gen_cookie(ctx->sk);
5107 }
5108
5109 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5110 .func = bpf_get_socket_cookie_sock_ops,
5111 .gpl_only = false,
5112 .ret_type = RET_INTEGER,
5113 .arg1_type = ARG_PTR_TO_CTX,
5114 };
5115
__bpf_get_netns_cookie(struct sock * sk)5116 static u64 __bpf_get_netns_cookie(struct sock *sk)
5117 {
5118 const struct net *net = sk ? sock_net(sk) : &init_net;
5119
5120 return net->net_cookie;
5121 }
5122
BPF_CALL_1(bpf_get_netns_cookie_sock,struct sock *,ctx)5123 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5124 {
5125 return __bpf_get_netns_cookie(ctx);
5126 }
5127
5128 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5129 .func = bpf_get_netns_cookie_sock,
5130 .gpl_only = false,
5131 .ret_type = RET_INTEGER,
5132 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5133 };
5134
BPF_CALL_1(bpf_get_netns_cookie_sock_addr,struct bpf_sock_addr_kern *,ctx)5135 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5136 {
5137 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5138 }
5139
5140 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5141 .func = bpf_get_netns_cookie_sock_addr,
5142 .gpl_only = false,
5143 .ret_type = RET_INTEGER,
5144 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5145 };
5146
BPF_CALL_1(bpf_get_netns_cookie_sock_ops,struct bpf_sock_ops_kern *,ctx)5147 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5148 {
5149 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5150 }
5151
5152 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5153 .func = bpf_get_netns_cookie_sock_ops,
5154 .gpl_only = false,
5155 .ret_type = RET_INTEGER,
5156 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5157 };
5158
BPF_CALL_1(bpf_get_netns_cookie_sk_msg,struct sk_msg *,ctx)5159 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5160 {
5161 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5162 }
5163
5164 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5165 .func = bpf_get_netns_cookie_sk_msg,
5166 .gpl_only = false,
5167 .ret_type = RET_INTEGER,
5168 .arg1_type = ARG_PTR_TO_CTX_OR_NULL,
5169 };
5170
BPF_CALL_1(bpf_get_socket_uid,struct sk_buff *,skb)5171 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5172 {
5173 struct sock *sk = sk_to_full_sk(skb->sk);
5174 kuid_t kuid;
5175
5176 if (!sk || !sk_fullsock(sk))
5177 return overflowuid;
5178 kuid = sock_net_uid(sock_net(sk), sk);
5179 return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5180 }
5181
5182 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5183 .func = bpf_get_socket_uid,
5184 .gpl_only = false,
5185 .ret_type = RET_INTEGER,
5186 .arg1_type = ARG_PTR_TO_CTX,
5187 };
5188
sol_socket_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5189 static int sol_socket_sockopt(struct sock *sk, int optname,
5190 char *optval, int *optlen,
5191 bool getopt)
5192 {
5193 switch (optname) {
5194 case SO_REUSEADDR:
5195 case SO_SNDBUF:
5196 case SO_RCVBUF:
5197 case SO_KEEPALIVE:
5198 case SO_PRIORITY:
5199 case SO_REUSEPORT:
5200 case SO_RCVLOWAT:
5201 case SO_MARK:
5202 case SO_MAX_PACING_RATE:
5203 case SO_BINDTOIFINDEX:
5204 case SO_TXREHASH:
5205 if (*optlen != sizeof(int))
5206 return -EINVAL;
5207 break;
5208 case SO_BINDTODEVICE:
5209 break;
5210 default:
5211 return -EINVAL;
5212 }
5213
5214 if (getopt) {
5215 if (optname == SO_BINDTODEVICE)
5216 return -EINVAL;
5217 return sk_getsockopt(sk, SOL_SOCKET, optname,
5218 KERNEL_SOCKPTR(optval),
5219 KERNEL_SOCKPTR(optlen));
5220 }
5221
5222 return sk_setsockopt(sk, SOL_SOCKET, optname,
5223 KERNEL_SOCKPTR(optval), *optlen);
5224 }
5225
bpf_sol_tcp_setsockopt(struct sock * sk,int optname,char * optval,int optlen)5226 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5227 char *optval, int optlen)
5228 {
5229 struct tcp_sock *tp = tcp_sk(sk);
5230 unsigned long timeout;
5231 int val;
5232
5233 if (optlen != sizeof(int))
5234 return -EINVAL;
5235
5236 val = *(int *)optval;
5237
5238 /* Only some options are supported */
5239 switch (optname) {
5240 case TCP_BPF_IW:
5241 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5242 return -EINVAL;
5243 tcp_snd_cwnd_set(tp, val);
5244 break;
5245 case TCP_BPF_SNDCWND_CLAMP:
5246 if (val <= 0)
5247 return -EINVAL;
5248 tp->snd_cwnd_clamp = val;
5249 tp->snd_ssthresh = val;
5250 break;
5251 case TCP_BPF_DELACK_MAX:
5252 timeout = usecs_to_jiffies(val);
5253 if (timeout > TCP_DELACK_MAX ||
5254 timeout < TCP_TIMEOUT_MIN)
5255 return -EINVAL;
5256 inet_csk(sk)->icsk_delack_max = timeout;
5257 break;
5258 case TCP_BPF_RTO_MIN:
5259 timeout = usecs_to_jiffies(val);
5260 if (timeout > TCP_RTO_MIN ||
5261 timeout < TCP_TIMEOUT_MIN)
5262 return -EINVAL;
5263 inet_csk(sk)->icsk_rto_min = timeout;
5264 break;
5265 default:
5266 return -EINVAL;
5267 }
5268
5269 return 0;
5270 }
5271
sol_tcp_sockopt_congestion(struct sock * sk,char * optval,int * optlen,bool getopt)5272 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5273 int *optlen, bool getopt)
5274 {
5275 struct tcp_sock *tp;
5276 int ret;
5277
5278 if (*optlen < 2)
5279 return -EINVAL;
5280
5281 if (getopt) {
5282 if (!inet_csk(sk)->icsk_ca_ops)
5283 return -EINVAL;
5284 /* BPF expects NULL-terminated tcp-cc string */
5285 optval[--(*optlen)] = '\0';
5286 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5287 KERNEL_SOCKPTR(optval),
5288 KERNEL_SOCKPTR(optlen));
5289 }
5290
5291 /* "cdg" is the only cc that alloc a ptr
5292 * in inet_csk_ca area. The bpf-tcp-cc may
5293 * overwrite this ptr after switching to cdg.
5294 */
5295 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5296 return -ENOTSUPP;
5297
5298 /* It stops this looping
5299 *
5300 * .init => bpf_setsockopt(tcp_cc) => .init =>
5301 * bpf_setsockopt(tcp_cc)" => .init => ....
5302 *
5303 * The second bpf_setsockopt(tcp_cc) is not allowed
5304 * in order to break the loop when both .init
5305 * are the same bpf prog.
5306 *
5307 * This applies even the second bpf_setsockopt(tcp_cc)
5308 * does not cause a loop. This limits only the first
5309 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5310 * pick a fallback cc (eg. peer does not support ECN)
5311 * and the second '.init' cannot fallback to
5312 * another.
5313 */
5314 tp = tcp_sk(sk);
5315 if (tp->bpf_chg_cc_inprogress)
5316 return -EBUSY;
5317
5318 tp->bpf_chg_cc_inprogress = 1;
5319 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5320 KERNEL_SOCKPTR(optval), *optlen);
5321 tp->bpf_chg_cc_inprogress = 0;
5322 return ret;
5323 }
5324
sol_tcp_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5325 static int sol_tcp_sockopt(struct sock *sk, int optname,
5326 char *optval, int *optlen,
5327 bool getopt)
5328 {
5329 if (sk->sk_protocol != IPPROTO_TCP)
5330 return -EINVAL;
5331
5332 switch (optname) {
5333 case TCP_NODELAY:
5334 case TCP_MAXSEG:
5335 case TCP_KEEPIDLE:
5336 case TCP_KEEPINTVL:
5337 case TCP_KEEPCNT:
5338 case TCP_SYNCNT:
5339 case TCP_WINDOW_CLAMP:
5340 case TCP_THIN_LINEAR_TIMEOUTS:
5341 case TCP_USER_TIMEOUT:
5342 case TCP_NOTSENT_LOWAT:
5343 case TCP_SAVE_SYN:
5344 if (*optlen != sizeof(int))
5345 return -EINVAL;
5346 break;
5347 case TCP_CONGESTION:
5348 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5349 case TCP_SAVED_SYN:
5350 if (*optlen < 1)
5351 return -EINVAL;
5352 break;
5353 default:
5354 if (getopt)
5355 return -EINVAL;
5356 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5357 }
5358
5359 if (getopt) {
5360 if (optname == TCP_SAVED_SYN) {
5361 struct tcp_sock *tp = tcp_sk(sk);
5362
5363 if (!tp->saved_syn ||
5364 *optlen > tcp_saved_syn_len(tp->saved_syn))
5365 return -EINVAL;
5366 memcpy(optval, tp->saved_syn->data, *optlen);
5367 /* It cannot free tp->saved_syn here because it
5368 * does not know if the user space still needs it.
5369 */
5370 return 0;
5371 }
5372
5373 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5374 KERNEL_SOCKPTR(optval),
5375 KERNEL_SOCKPTR(optlen));
5376 }
5377
5378 return do_tcp_setsockopt(sk, SOL_TCP, optname,
5379 KERNEL_SOCKPTR(optval), *optlen);
5380 }
5381
sol_ip_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5382 static int sol_ip_sockopt(struct sock *sk, int optname,
5383 char *optval, int *optlen,
5384 bool getopt)
5385 {
5386 if (sk->sk_family != AF_INET)
5387 return -EINVAL;
5388
5389 switch (optname) {
5390 case IP_TOS:
5391 if (*optlen != sizeof(int))
5392 return -EINVAL;
5393 break;
5394 default:
5395 return -EINVAL;
5396 }
5397
5398 if (getopt)
5399 return do_ip_getsockopt(sk, SOL_IP, optname,
5400 KERNEL_SOCKPTR(optval),
5401 KERNEL_SOCKPTR(optlen));
5402
5403 return do_ip_setsockopt(sk, SOL_IP, optname,
5404 KERNEL_SOCKPTR(optval), *optlen);
5405 }
5406
sol_ipv6_sockopt(struct sock * sk,int optname,char * optval,int * optlen,bool getopt)5407 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5408 char *optval, int *optlen,
5409 bool getopt)
5410 {
5411 if (sk->sk_family != AF_INET6)
5412 return -EINVAL;
5413
5414 switch (optname) {
5415 case IPV6_TCLASS:
5416 case IPV6_AUTOFLOWLABEL:
5417 if (*optlen != sizeof(int))
5418 return -EINVAL;
5419 break;
5420 default:
5421 return -EINVAL;
5422 }
5423
5424 if (getopt)
5425 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5426 KERNEL_SOCKPTR(optval),
5427 KERNEL_SOCKPTR(optlen));
5428
5429 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5430 KERNEL_SOCKPTR(optval), *optlen);
5431 }
5432
__bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5433 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5434 char *optval, int optlen)
5435 {
5436 if (!sk_fullsock(sk))
5437 return -EINVAL;
5438
5439 if (level == SOL_SOCKET)
5440 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5441 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5442 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5443 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5444 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5445 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5446 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5447
5448 return -EINVAL;
5449 }
5450
_bpf_setsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5451 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5452 char *optval, int optlen)
5453 {
5454 if (sk_fullsock(sk))
5455 sock_owned_by_me(sk);
5456 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5457 }
5458
__bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5459 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5460 char *optval, int optlen)
5461 {
5462 int err, saved_optlen = optlen;
5463
5464 if (!sk_fullsock(sk)) {
5465 err = -EINVAL;
5466 goto done;
5467 }
5468
5469 if (level == SOL_SOCKET)
5470 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5471 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5472 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5473 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5474 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5475 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5476 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5477 else
5478 err = -EINVAL;
5479
5480 done:
5481 if (err)
5482 optlen = 0;
5483 if (optlen < saved_optlen)
5484 memset(optval + optlen, 0, saved_optlen - optlen);
5485 return err;
5486 }
5487
_bpf_getsockopt(struct sock * sk,int level,int optname,char * optval,int optlen)5488 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5489 char *optval, int optlen)
5490 {
5491 if (sk_fullsock(sk))
5492 sock_owned_by_me(sk);
5493 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5494 }
5495
BPF_CALL_5(bpf_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5496 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5497 int, optname, char *, optval, int, optlen)
5498 {
5499 return _bpf_setsockopt(sk, level, optname, optval, optlen);
5500 }
5501
5502 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5503 .func = bpf_sk_setsockopt,
5504 .gpl_only = false,
5505 .ret_type = RET_INTEGER,
5506 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5507 .arg2_type = ARG_ANYTHING,
5508 .arg3_type = ARG_ANYTHING,
5509 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5510 .arg5_type = ARG_CONST_SIZE,
5511 };
5512
BPF_CALL_5(bpf_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5513 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5514 int, optname, char *, optval, int, optlen)
5515 {
5516 return _bpf_getsockopt(sk, level, optname, optval, optlen);
5517 }
5518
5519 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5520 .func = bpf_sk_getsockopt,
5521 .gpl_only = false,
5522 .ret_type = RET_INTEGER,
5523 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5524 .arg2_type = ARG_ANYTHING,
5525 .arg3_type = ARG_ANYTHING,
5526 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5527 .arg5_type = ARG_CONST_SIZE,
5528 };
5529
BPF_CALL_5(bpf_unlocked_sk_setsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5530 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5531 int, optname, char *, optval, int, optlen)
5532 {
5533 return __bpf_setsockopt(sk, level, optname, optval, optlen);
5534 }
5535
5536 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5537 .func = bpf_unlocked_sk_setsockopt,
5538 .gpl_only = false,
5539 .ret_type = RET_INTEGER,
5540 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5541 .arg2_type = ARG_ANYTHING,
5542 .arg3_type = ARG_ANYTHING,
5543 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5544 .arg5_type = ARG_CONST_SIZE,
5545 };
5546
BPF_CALL_5(bpf_unlocked_sk_getsockopt,struct sock *,sk,int,level,int,optname,char *,optval,int,optlen)5547 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5548 int, optname, char *, optval, int, optlen)
5549 {
5550 return __bpf_getsockopt(sk, level, optname, optval, optlen);
5551 }
5552
5553 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5554 .func = bpf_unlocked_sk_getsockopt,
5555 .gpl_only = false,
5556 .ret_type = RET_INTEGER,
5557 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5558 .arg2_type = ARG_ANYTHING,
5559 .arg3_type = ARG_ANYTHING,
5560 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5561 .arg5_type = ARG_CONST_SIZE,
5562 };
5563
BPF_CALL_5(bpf_sock_addr_setsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5564 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5565 int, level, int, optname, char *, optval, int, optlen)
5566 {
5567 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5568 }
5569
5570 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5571 .func = bpf_sock_addr_setsockopt,
5572 .gpl_only = false,
5573 .ret_type = RET_INTEGER,
5574 .arg1_type = ARG_PTR_TO_CTX,
5575 .arg2_type = ARG_ANYTHING,
5576 .arg3_type = ARG_ANYTHING,
5577 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5578 .arg5_type = ARG_CONST_SIZE,
5579 };
5580
BPF_CALL_5(bpf_sock_addr_getsockopt,struct bpf_sock_addr_kern *,ctx,int,level,int,optname,char *,optval,int,optlen)5581 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5582 int, level, int, optname, char *, optval, int, optlen)
5583 {
5584 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5585 }
5586
5587 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5588 .func = bpf_sock_addr_getsockopt,
5589 .gpl_only = false,
5590 .ret_type = RET_INTEGER,
5591 .arg1_type = ARG_PTR_TO_CTX,
5592 .arg2_type = ARG_ANYTHING,
5593 .arg3_type = ARG_ANYTHING,
5594 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5595 .arg5_type = ARG_CONST_SIZE,
5596 };
5597
BPF_CALL_5(bpf_sock_ops_setsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5598 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5599 int, level, int, optname, char *, optval, int, optlen)
5600 {
5601 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5602 }
5603
5604 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5605 .func = bpf_sock_ops_setsockopt,
5606 .gpl_only = false,
5607 .ret_type = RET_INTEGER,
5608 .arg1_type = ARG_PTR_TO_CTX,
5609 .arg2_type = ARG_ANYTHING,
5610 .arg3_type = ARG_ANYTHING,
5611 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5612 .arg5_type = ARG_CONST_SIZE,
5613 };
5614
bpf_sock_ops_get_syn(struct bpf_sock_ops_kern * bpf_sock,int optname,const u8 ** start)5615 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5616 int optname, const u8 **start)
5617 {
5618 struct sk_buff *syn_skb = bpf_sock->syn_skb;
5619 const u8 *hdr_start;
5620 int ret;
5621
5622 if (syn_skb) {
5623 /* sk is a request_sock here */
5624
5625 if (optname == TCP_BPF_SYN) {
5626 hdr_start = syn_skb->data;
5627 ret = tcp_hdrlen(syn_skb);
5628 } else if (optname == TCP_BPF_SYN_IP) {
5629 hdr_start = skb_network_header(syn_skb);
5630 ret = skb_network_header_len(syn_skb) +
5631 tcp_hdrlen(syn_skb);
5632 } else {
5633 /* optname == TCP_BPF_SYN_MAC */
5634 hdr_start = skb_mac_header(syn_skb);
5635 ret = skb_mac_header_len(syn_skb) +
5636 skb_network_header_len(syn_skb) +
5637 tcp_hdrlen(syn_skb);
5638 }
5639 } else {
5640 struct sock *sk = bpf_sock->sk;
5641 struct saved_syn *saved_syn;
5642
5643 if (sk->sk_state == TCP_NEW_SYN_RECV)
5644 /* synack retransmit. bpf_sock->syn_skb will
5645 * not be available. It has to resort to
5646 * saved_syn (if it is saved).
5647 */
5648 saved_syn = inet_reqsk(sk)->saved_syn;
5649 else
5650 saved_syn = tcp_sk(sk)->saved_syn;
5651
5652 if (!saved_syn)
5653 return -ENOENT;
5654
5655 if (optname == TCP_BPF_SYN) {
5656 hdr_start = saved_syn->data +
5657 saved_syn->mac_hdrlen +
5658 saved_syn->network_hdrlen;
5659 ret = saved_syn->tcp_hdrlen;
5660 } else if (optname == TCP_BPF_SYN_IP) {
5661 hdr_start = saved_syn->data +
5662 saved_syn->mac_hdrlen;
5663 ret = saved_syn->network_hdrlen +
5664 saved_syn->tcp_hdrlen;
5665 } else {
5666 /* optname == TCP_BPF_SYN_MAC */
5667
5668 /* TCP_SAVE_SYN may not have saved the mac hdr */
5669 if (!saved_syn->mac_hdrlen)
5670 return -ENOENT;
5671
5672 hdr_start = saved_syn->data;
5673 ret = saved_syn->mac_hdrlen +
5674 saved_syn->network_hdrlen +
5675 saved_syn->tcp_hdrlen;
5676 }
5677 }
5678
5679 *start = hdr_start;
5680 return ret;
5681 }
5682
BPF_CALL_5(bpf_sock_ops_getsockopt,struct bpf_sock_ops_kern *,bpf_sock,int,level,int,optname,char *,optval,int,optlen)5683 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5684 int, level, int, optname, char *, optval, int, optlen)
5685 {
5686 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5687 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5688 int ret, copy_len = 0;
5689 const u8 *start;
5690
5691 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5692 if (ret > 0) {
5693 copy_len = ret;
5694 if (optlen < copy_len) {
5695 copy_len = optlen;
5696 ret = -ENOSPC;
5697 }
5698
5699 memcpy(optval, start, copy_len);
5700 }
5701
5702 /* Zero out unused buffer at the end */
5703 memset(optval + copy_len, 0, optlen - copy_len);
5704
5705 return ret;
5706 }
5707
5708 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5709 }
5710
5711 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5712 .func = bpf_sock_ops_getsockopt,
5713 .gpl_only = false,
5714 .ret_type = RET_INTEGER,
5715 .arg1_type = ARG_PTR_TO_CTX,
5716 .arg2_type = ARG_ANYTHING,
5717 .arg3_type = ARG_ANYTHING,
5718 .arg4_type = ARG_PTR_TO_UNINIT_MEM,
5719 .arg5_type = ARG_CONST_SIZE,
5720 };
5721
BPF_CALL_2(bpf_sock_ops_cb_flags_set,struct bpf_sock_ops_kern *,bpf_sock,int,argval)5722 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5723 int, argval)
5724 {
5725 struct sock *sk = bpf_sock->sk;
5726 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5727
5728 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5729 return -EINVAL;
5730
5731 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5732
5733 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5734 }
5735
5736 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5737 .func = bpf_sock_ops_cb_flags_set,
5738 .gpl_only = false,
5739 .ret_type = RET_INTEGER,
5740 .arg1_type = ARG_PTR_TO_CTX,
5741 .arg2_type = ARG_ANYTHING,
5742 };
5743
5744 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5745 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5746
BPF_CALL_3(bpf_bind,struct bpf_sock_addr_kern *,ctx,struct sockaddr *,addr,int,addr_len)5747 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5748 int, addr_len)
5749 {
5750 #ifdef CONFIG_INET
5751 struct sock *sk = ctx->sk;
5752 u32 flags = BIND_FROM_BPF;
5753 int err;
5754
5755 err = -EINVAL;
5756 if (addr_len < offsetofend(struct sockaddr, sa_family))
5757 return err;
5758 if (addr->sa_family == AF_INET) {
5759 if (addr_len < sizeof(struct sockaddr_in))
5760 return err;
5761 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5762 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5763 return __inet_bind(sk, addr, addr_len, flags);
5764 #if IS_ENABLED(CONFIG_IPV6)
5765 } else if (addr->sa_family == AF_INET6) {
5766 if (addr_len < SIN6_LEN_RFC2133)
5767 return err;
5768 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5769 flags |= BIND_FORCE_ADDRESS_NO_PORT;
5770 /* ipv6_bpf_stub cannot be NULL, since it's called from
5771 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5772 */
5773 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5774 #endif /* CONFIG_IPV6 */
5775 }
5776 #endif /* CONFIG_INET */
5777
5778 return -EAFNOSUPPORT;
5779 }
5780
5781 static const struct bpf_func_proto bpf_bind_proto = {
5782 .func = bpf_bind,
5783 .gpl_only = false,
5784 .ret_type = RET_INTEGER,
5785 .arg1_type = ARG_PTR_TO_CTX,
5786 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
5787 .arg3_type = ARG_CONST_SIZE,
5788 };
5789
5790 #ifdef CONFIG_XFRM
5791
5792 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5793 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5794
5795 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5796 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5797
5798 #endif
5799
BPF_CALL_5(bpf_skb_get_xfrm_state,struct sk_buff *,skb,u32,index,struct bpf_xfrm_state *,to,u32,size,u64,flags)5800 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5801 struct bpf_xfrm_state *, to, u32, size, u64, flags)
5802 {
5803 const struct sec_path *sp = skb_sec_path(skb);
5804 const struct xfrm_state *x;
5805
5806 if (!sp || unlikely(index >= sp->len || flags))
5807 goto err_clear;
5808
5809 x = sp->xvec[index];
5810
5811 if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5812 goto err_clear;
5813
5814 to->reqid = x->props.reqid;
5815 to->spi = x->id.spi;
5816 to->family = x->props.family;
5817 to->ext = 0;
5818
5819 if (to->family == AF_INET6) {
5820 memcpy(to->remote_ipv6, x->props.saddr.a6,
5821 sizeof(to->remote_ipv6));
5822 } else {
5823 to->remote_ipv4 = x->props.saddr.a4;
5824 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5825 }
5826
5827 return 0;
5828 err_clear:
5829 memset(to, 0, size);
5830 return -EINVAL;
5831 }
5832
5833 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5834 .func = bpf_skb_get_xfrm_state,
5835 .gpl_only = false,
5836 .ret_type = RET_INTEGER,
5837 .arg1_type = ARG_PTR_TO_CTX,
5838 .arg2_type = ARG_ANYTHING,
5839 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
5840 .arg4_type = ARG_CONST_SIZE,
5841 .arg5_type = ARG_ANYTHING,
5842 };
5843 #endif
5844
5845 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
bpf_fib_set_fwd_params(struct bpf_fib_lookup * params,u32 mtu)5846 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5847 {
5848 params->h_vlan_TCI = 0;
5849 params->h_vlan_proto = 0;
5850 if (mtu)
5851 params->mtu_result = mtu; /* union with tot_len */
5852
5853 return 0;
5854 }
5855 #endif
5856
5857 #if IS_ENABLED(CONFIG_INET)
bpf_ipv4_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)5858 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5859 u32 flags, bool check_mtu)
5860 {
5861 struct fib_nh_common *nhc;
5862 struct in_device *in_dev;
5863 struct neighbour *neigh;
5864 struct net_device *dev;
5865 struct fib_result res;
5866 struct flowi4 fl4;
5867 u32 mtu = 0;
5868 int err;
5869
5870 dev = dev_get_by_index_rcu(net, params->ifindex);
5871 if (unlikely(!dev))
5872 return -ENODEV;
5873
5874 /* verify forwarding is enabled on this interface */
5875 in_dev = __in_dev_get_rcu(dev);
5876 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5877 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5878
5879 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5880 fl4.flowi4_iif = 1;
5881 fl4.flowi4_oif = params->ifindex;
5882 } else {
5883 fl4.flowi4_iif = params->ifindex;
5884 fl4.flowi4_oif = 0;
5885 }
5886 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5887 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5888 fl4.flowi4_flags = 0;
5889
5890 fl4.flowi4_proto = params->l4_protocol;
5891 fl4.daddr = params->ipv4_dst;
5892 fl4.saddr = params->ipv4_src;
5893 fl4.fl4_sport = params->sport;
5894 fl4.fl4_dport = params->dport;
5895 fl4.flowi4_multipath_hash = 0;
5896
5897 if (flags & BPF_FIB_LOOKUP_DIRECT) {
5898 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5899 struct fib_table *tb;
5900
5901 if (flags & BPF_FIB_LOOKUP_TBID) {
5902 tbid = params->tbid;
5903 /* zero out for vlan output */
5904 params->tbid = 0;
5905 }
5906
5907 tb = fib_get_table(net, tbid);
5908 if (unlikely(!tb))
5909 return BPF_FIB_LKUP_RET_NOT_FWDED;
5910
5911 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5912 } else {
5913 if (flags & BPF_FIB_LOOKUP_MARK)
5914 fl4.flowi4_mark = params->mark;
5915 else
5916 fl4.flowi4_mark = 0;
5917 fl4.flowi4_secid = 0;
5918 fl4.flowi4_tun_key.tun_id = 0;
5919 fl4.flowi4_uid = sock_net_uid(net, NULL);
5920
5921 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5922 }
5923
5924 if (err) {
5925 /* map fib lookup errors to RTN_ type */
5926 if (err == -EINVAL)
5927 return BPF_FIB_LKUP_RET_BLACKHOLE;
5928 if (err == -EHOSTUNREACH)
5929 return BPF_FIB_LKUP_RET_UNREACHABLE;
5930 if (err == -EACCES)
5931 return BPF_FIB_LKUP_RET_PROHIBIT;
5932
5933 return BPF_FIB_LKUP_RET_NOT_FWDED;
5934 }
5935
5936 if (res.type != RTN_UNICAST)
5937 return BPF_FIB_LKUP_RET_NOT_FWDED;
5938
5939 if (fib_info_num_path(res.fi) > 1)
5940 fib_select_path(net, &res, &fl4, NULL);
5941
5942 if (check_mtu) {
5943 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5944 if (params->tot_len > mtu) {
5945 params->mtu_result = mtu; /* union with tot_len */
5946 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5947 }
5948 }
5949
5950 nhc = res.nhc;
5951
5952 /* do not handle lwt encaps right now */
5953 if (nhc->nhc_lwtstate)
5954 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5955
5956 dev = nhc->nhc_dev;
5957
5958 params->rt_metric = res.fi->fib_priority;
5959 params->ifindex = dev->ifindex;
5960
5961 if (flags & BPF_FIB_LOOKUP_SRC)
5962 params->ipv4_src = fib_result_prefsrc(net, &res);
5963
5964 /* xdp and cls_bpf programs are run in RCU-bh so
5965 * rcu_read_lock_bh is not needed here
5966 */
5967 if (likely(nhc->nhc_gw_family != AF_INET6)) {
5968 if (nhc->nhc_gw_family)
5969 params->ipv4_dst = nhc->nhc_gw.ipv4;
5970 } else {
5971 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5972
5973 params->family = AF_INET6;
5974 *dst = nhc->nhc_gw.ipv6;
5975 }
5976
5977 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5978 goto set_fwd_params;
5979
5980 if (likely(nhc->nhc_gw_family != AF_INET6))
5981 neigh = __ipv4_neigh_lookup_noref(dev,
5982 (__force u32)params->ipv4_dst);
5983 else
5984 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5985
5986 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5987 return BPF_FIB_LKUP_RET_NO_NEIGH;
5988 memcpy(params->dmac, neigh->ha, ETH_ALEN);
5989 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5990
5991 set_fwd_params:
5992 return bpf_fib_set_fwd_params(params, mtu);
5993 }
5994 #endif
5995
5996 #if IS_ENABLED(CONFIG_IPV6)
bpf_ipv6_fib_lookup(struct net * net,struct bpf_fib_lookup * params,u32 flags,bool check_mtu)5997 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5998 u32 flags, bool check_mtu)
5999 {
6000 struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
6001 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
6002 struct fib6_result res = {};
6003 struct neighbour *neigh;
6004 struct net_device *dev;
6005 struct inet6_dev *idev;
6006 struct flowi6 fl6;
6007 int strict = 0;
6008 int oif, err;
6009 u32 mtu = 0;
6010
6011 /* link local addresses are never forwarded */
6012 if (rt6_need_strict(dst) || rt6_need_strict(src))
6013 return BPF_FIB_LKUP_RET_NOT_FWDED;
6014
6015 dev = dev_get_by_index_rcu(net, params->ifindex);
6016 if (unlikely(!dev))
6017 return -ENODEV;
6018
6019 idev = __in6_dev_get_safely(dev);
6020 if (unlikely(!idev || !READ_ONCE(idev->cnf.forwarding)))
6021 return BPF_FIB_LKUP_RET_FWD_DISABLED;
6022
6023 if (flags & BPF_FIB_LOOKUP_OUTPUT) {
6024 fl6.flowi6_iif = 1;
6025 oif = fl6.flowi6_oif = params->ifindex;
6026 } else {
6027 oif = fl6.flowi6_iif = params->ifindex;
6028 fl6.flowi6_oif = 0;
6029 strict = RT6_LOOKUP_F_HAS_SADDR;
6030 }
6031 fl6.flowlabel = params->flowinfo;
6032 fl6.flowi6_scope = 0;
6033 fl6.flowi6_flags = 0;
6034 fl6.mp_hash = 0;
6035
6036 fl6.flowi6_proto = params->l4_protocol;
6037 fl6.daddr = *dst;
6038 fl6.saddr = *src;
6039 fl6.fl6_sport = params->sport;
6040 fl6.fl6_dport = params->dport;
6041
6042 if (flags & BPF_FIB_LOOKUP_DIRECT) {
6043 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
6044 struct fib6_table *tb;
6045
6046 if (flags & BPF_FIB_LOOKUP_TBID) {
6047 tbid = params->tbid;
6048 /* zero out for vlan output */
6049 params->tbid = 0;
6050 }
6051
6052 tb = ipv6_stub->fib6_get_table(net, tbid);
6053 if (unlikely(!tb))
6054 return BPF_FIB_LKUP_RET_NOT_FWDED;
6055
6056 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
6057 strict);
6058 } else {
6059 if (flags & BPF_FIB_LOOKUP_MARK)
6060 fl6.flowi6_mark = params->mark;
6061 else
6062 fl6.flowi6_mark = 0;
6063 fl6.flowi6_secid = 0;
6064 fl6.flowi6_tun_key.tun_id = 0;
6065 fl6.flowi6_uid = sock_net_uid(net, NULL);
6066
6067 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
6068 }
6069
6070 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
6071 res.f6i == net->ipv6.fib6_null_entry))
6072 return BPF_FIB_LKUP_RET_NOT_FWDED;
6073
6074 switch (res.fib6_type) {
6075 /* only unicast is forwarded */
6076 case RTN_UNICAST:
6077 break;
6078 case RTN_BLACKHOLE:
6079 return BPF_FIB_LKUP_RET_BLACKHOLE;
6080 case RTN_UNREACHABLE:
6081 return BPF_FIB_LKUP_RET_UNREACHABLE;
6082 case RTN_PROHIBIT:
6083 return BPF_FIB_LKUP_RET_PROHIBIT;
6084 default:
6085 return BPF_FIB_LKUP_RET_NOT_FWDED;
6086 }
6087
6088 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
6089 fl6.flowi6_oif != 0, NULL, strict);
6090
6091 if (check_mtu) {
6092 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
6093 if (params->tot_len > mtu) {
6094 params->mtu_result = mtu; /* union with tot_len */
6095 return BPF_FIB_LKUP_RET_FRAG_NEEDED;
6096 }
6097 }
6098
6099 if (res.nh->fib_nh_lws)
6100 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
6101
6102 if (res.nh->fib_nh_gw_family)
6103 *dst = res.nh->fib_nh_gw6;
6104
6105 dev = res.nh->fib_nh_dev;
6106 params->rt_metric = res.f6i->fib6_metric;
6107 params->ifindex = dev->ifindex;
6108
6109 if (flags & BPF_FIB_LOOKUP_SRC) {
6110 if (res.f6i->fib6_prefsrc.plen) {
6111 *src = res.f6i->fib6_prefsrc.addr;
6112 } else {
6113 err = ipv6_bpf_stub->ipv6_dev_get_saddr(net, dev,
6114 &fl6.daddr, 0,
6115 src);
6116 if (err)
6117 return BPF_FIB_LKUP_RET_NO_SRC_ADDR;
6118 }
6119 }
6120
6121 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6122 goto set_fwd_params;
6123
6124 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6125 * not needed here.
6126 */
6127 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6128 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6129 return BPF_FIB_LKUP_RET_NO_NEIGH;
6130 memcpy(params->dmac, neigh->ha, ETH_ALEN);
6131 memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6132
6133 set_fwd_params:
6134 return bpf_fib_set_fwd_params(params, mtu);
6135 }
6136 #endif
6137
6138 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6139 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID | \
6140 BPF_FIB_LOOKUP_SRC | BPF_FIB_LOOKUP_MARK)
6141
BPF_CALL_4(bpf_xdp_fib_lookup,struct xdp_buff *,ctx,struct bpf_fib_lookup *,params,int,plen,u32,flags)6142 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6143 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6144 {
6145 if (plen < sizeof(*params))
6146 return -EINVAL;
6147
6148 if (flags & ~BPF_FIB_LOOKUP_MASK)
6149 return -EINVAL;
6150
6151 switch (params->family) {
6152 #if IS_ENABLED(CONFIG_INET)
6153 case AF_INET:
6154 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6155 flags, true);
6156 #endif
6157 #if IS_ENABLED(CONFIG_IPV6)
6158 case AF_INET6:
6159 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6160 flags, true);
6161 #endif
6162 }
6163 return -EAFNOSUPPORT;
6164 }
6165
6166 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6167 .func = bpf_xdp_fib_lookup,
6168 .gpl_only = true,
6169 .ret_type = RET_INTEGER,
6170 .arg1_type = ARG_PTR_TO_CTX,
6171 .arg2_type = ARG_PTR_TO_MEM,
6172 .arg3_type = ARG_CONST_SIZE,
6173 .arg4_type = ARG_ANYTHING,
6174 };
6175
BPF_CALL_4(bpf_skb_fib_lookup,struct sk_buff *,skb,struct bpf_fib_lookup *,params,int,plen,u32,flags)6176 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6177 struct bpf_fib_lookup *, params, int, plen, u32, flags)
6178 {
6179 struct net *net = dev_net(skb->dev);
6180 int rc = -EAFNOSUPPORT;
6181 bool check_mtu = false;
6182
6183 if (plen < sizeof(*params))
6184 return -EINVAL;
6185
6186 if (flags & ~BPF_FIB_LOOKUP_MASK)
6187 return -EINVAL;
6188
6189 if (params->tot_len)
6190 check_mtu = true;
6191
6192 switch (params->family) {
6193 #if IS_ENABLED(CONFIG_INET)
6194 case AF_INET:
6195 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6196 break;
6197 #endif
6198 #if IS_ENABLED(CONFIG_IPV6)
6199 case AF_INET6:
6200 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6201 break;
6202 #endif
6203 }
6204
6205 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6206 struct net_device *dev;
6207
6208 /* When tot_len isn't provided by user, check skb
6209 * against MTU of FIB lookup resulting net_device
6210 */
6211 dev = dev_get_by_index_rcu(net, params->ifindex);
6212 if (!is_skb_forwardable(dev, skb))
6213 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6214
6215 params->mtu_result = dev->mtu; /* union with tot_len */
6216 }
6217
6218 return rc;
6219 }
6220
6221 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6222 .func = bpf_skb_fib_lookup,
6223 .gpl_only = true,
6224 .ret_type = RET_INTEGER,
6225 .arg1_type = ARG_PTR_TO_CTX,
6226 .arg2_type = ARG_PTR_TO_MEM,
6227 .arg3_type = ARG_CONST_SIZE,
6228 .arg4_type = ARG_ANYTHING,
6229 };
6230
__dev_via_ifindex(struct net_device * dev_curr,u32 ifindex)6231 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6232 u32 ifindex)
6233 {
6234 struct net *netns = dev_net(dev_curr);
6235
6236 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6237 if (ifindex == 0)
6238 return dev_curr;
6239
6240 return dev_get_by_index_rcu(netns, ifindex);
6241 }
6242
BPF_CALL_5(bpf_skb_check_mtu,struct sk_buff *,skb,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6243 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6244 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6245 {
6246 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6247 struct net_device *dev = skb->dev;
6248 int skb_len, dev_len;
6249 int mtu;
6250
6251 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6252 return -EINVAL;
6253
6254 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6255 return -EINVAL;
6256
6257 dev = __dev_via_ifindex(dev, ifindex);
6258 if (unlikely(!dev))
6259 return -ENODEV;
6260
6261 mtu = READ_ONCE(dev->mtu);
6262
6263 dev_len = mtu + dev->hard_header_len;
6264
6265 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6266 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6267
6268 skb_len += len_diff; /* minus result pass check */
6269 if (skb_len <= dev_len) {
6270 ret = BPF_MTU_CHK_RET_SUCCESS;
6271 goto out;
6272 }
6273 /* At this point, skb->len exceed MTU, but as it include length of all
6274 * segments, it can still be below MTU. The SKB can possibly get
6275 * re-segmented in transmit path (see validate_xmit_skb). Thus, user
6276 * must choose if segs are to be MTU checked.
6277 */
6278 if (skb_is_gso(skb)) {
6279 ret = BPF_MTU_CHK_RET_SUCCESS;
6280
6281 if (flags & BPF_MTU_CHK_SEGS &&
6282 !skb_gso_validate_network_len(skb, mtu))
6283 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6284 }
6285 out:
6286 /* BPF verifier guarantees valid pointer */
6287 *mtu_len = mtu;
6288
6289 return ret;
6290 }
6291
BPF_CALL_5(bpf_xdp_check_mtu,struct xdp_buff *,xdp,u32,ifindex,u32 *,mtu_len,s32,len_diff,u64,flags)6292 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6293 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6294 {
6295 struct net_device *dev = xdp->rxq->dev;
6296 int xdp_len = xdp->data_end - xdp->data;
6297 int ret = BPF_MTU_CHK_RET_SUCCESS;
6298 int mtu, dev_len;
6299
6300 /* XDP variant doesn't support multi-buffer segment check (yet) */
6301 if (unlikely(flags))
6302 return -EINVAL;
6303
6304 dev = __dev_via_ifindex(dev, ifindex);
6305 if (unlikely(!dev))
6306 return -ENODEV;
6307
6308 mtu = READ_ONCE(dev->mtu);
6309
6310 /* Add L2-header as dev MTU is L3 size */
6311 dev_len = mtu + dev->hard_header_len;
6312
6313 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6314 if (*mtu_len)
6315 xdp_len = *mtu_len + dev->hard_header_len;
6316
6317 xdp_len += len_diff; /* minus result pass check */
6318 if (xdp_len > dev_len)
6319 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6320
6321 /* BPF verifier guarantees valid pointer */
6322 *mtu_len = mtu;
6323
6324 return ret;
6325 }
6326
6327 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6328 .func = bpf_skb_check_mtu,
6329 .gpl_only = true,
6330 .ret_type = RET_INTEGER,
6331 .arg1_type = ARG_PTR_TO_CTX,
6332 .arg2_type = ARG_ANYTHING,
6333 .arg3_type = ARG_PTR_TO_INT,
6334 .arg4_type = ARG_ANYTHING,
6335 .arg5_type = ARG_ANYTHING,
6336 };
6337
6338 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6339 .func = bpf_xdp_check_mtu,
6340 .gpl_only = true,
6341 .ret_type = RET_INTEGER,
6342 .arg1_type = ARG_PTR_TO_CTX,
6343 .arg2_type = ARG_ANYTHING,
6344 .arg3_type = ARG_PTR_TO_INT,
6345 .arg4_type = ARG_ANYTHING,
6346 .arg5_type = ARG_ANYTHING,
6347 };
6348
6349 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
bpf_push_seg6_encap(struct sk_buff * skb,u32 type,void * hdr,u32 len)6350 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6351 {
6352 int err;
6353 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6354
6355 if (!seg6_validate_srh(srh, len, false))
6356 return -EINVAL;
6357
6358 switch (type) {
6359 case BPF_LWT_ENCAP_SEG6_INLINE:
6360 if (skb->protocol != htons(ETH_P_IPV6))
6361 return -EBADMSG;
6362
6363 err = seg6_do_srh_inline(skb, srh);
6364 break;
6365 case BPF_LWT_ENCAP_SEG6:
6366 skb_reset_inner_headers(skb);
6367 skb->encapsulation = 1;
6368 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6369 break;
6370 default:
6371 return -EINVAL;
6372 }
6373
6374 bpf_compute_data_pointers(skb);
6375 if (err)
6376 return err;
6377
6378 skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6379
6380 return seg6_lookup_nexthop(skb, NULL, 0);
6381 }
6382 #endif /* CONFIG_IPV6_SEG6_BPF */
6383
6384 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
bpf_push_ip_encap(struct sk_buff * skb,void * hdr,u32 len,bool ingress)6385 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6386 bool ingress)
6387 {
6388 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6389 }
6390 #endif
6391
BPF_CALL_4(bpf_lwt_in_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6392 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6393 u32, len)
6394 {
6395 switch (type) {
6396 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6397 case BPF_LWT_ENCAP_SEG6:
6398 case BPF_LWT_ENCAP_SEG6_INLINE:
6399 return bpf_push_seg6_encap(skb, type, hdr, len);
6400 #endif
6401 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6402 case BPF_LWT_ENCAP_IP:
6403 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6404 #endif
6405 default:
6406 return -EINVAL;
6407 }
6408 }
6409
BPF_CALL_4(bpf_lwt_xmit_push_encap,struct sk_buff *,skb,u32,type,void *,hdr,u32,len)6410 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6411 void *, hdr, u32, len)
6412 {
6413 switch (type) {
6414 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6415 case BPF_LWT_ENCAP_IP:
6416 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6417 #endif
6418 default:
6419 return -EINVAL;
6420 }
6421 }
6422
6423 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6424 .func = bpf_lwt_in_push_encap,
6425 .gpl_only = false,
6426 .ret_type = RET_INTEGER,
6427 .arg1_type = ARG_PTR_TO_CTX,
6428 .arg2_type = ARG_ANYTHING,
6429 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6430 .arg4_type = ARG_CONST_SIZE
6431 };
6432
6433 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6434 .func = bpf_lwt_xmit_push_encap,
6435 .gpl_only = false,
6436 .ret_type = RET_INTEGER,
6437 .arg1_type = ARG_PTR_TO_CTX,
6438 .arg2_type = ARG_ANYTHING,
6439 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6440 .arg4_type = ARG_CONST_SIZE
6441 };
6442
6443 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
BPF_CALL_4(bpf_lwt_seg6_store_bytes,struct sk_buff *,skb,u32,offset,const void *,from,u32,len)6444 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6445 const void *, from, u32, len)
6446 {
6447 struct seg6_bpf_srh_state *srh_state =
6448 this_cpu_ptr(&seg6_bpf_srh_states);
6449 struct ipv6_sr_hdr *srh = srh_state->srh;
6450 void *srh_tlvs, *srh_end, *ptr;
6451 int srhoff = 0;
6452
6453 if (srh == NULL)
6454 return -EINVAL;
6455
6456 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6457 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6458
6459 ptr = skb->data + offset;
6460 if (ptr >= srh_tlvs && ptr + len <= srh_end)
6461 srh_state->valid = false;
6462 else if (ptr < (void *)&srh->flags ||
6463 ptr + len > (void *)&srh->segments)
6464 return -EFAULT;
6465
6466 if (unlikely(bpf_try_make_writable(skb, offset + len)))
6467 return -EFAULT;
6468 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6469 return -EINVAL;
6470 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6471
6472 memcpy(skb->data + offset, from, len);
6473 return 0;
6474 }
6475
6476 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6477 .func = bpf_lwt_seg6_store_bytes,
6478 .gpl_only = false,
6479 .ret_type = RET_INTEGER,
6480 .arg1_type = ARG_PTR_TO_CTX,
6481 .arg2_type = ARG_ANYTHING,
6482 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6483 .arg4_type = ARG_CONST_SIZE
6484 };
6485
bpf_update_srh_state(struct sk_buff * skb)6486 static void bpf_update_srh_state(struct sk_buff *skb)
6487 {
6488 struct seg6_bpf_srh_state *srh_state =
6489 this_cpu_ptr(&seg6_bpf_srh_states);
6490 int srhoff = 0;
6491
6492 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6493 srh_state->srh = NULL;
6494 } else {
6495 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6496 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6497 srh_state->valid = true;
6498 }
6499 }
6500
BPF_CALL_4(bpf_lwt_seg6_action,struct sk_buff *,skb,u32,action,void *,param,u32,param_len)6501 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6502 u32, action, void *, param, u32, param_len)
6503 {
6504 struct seg6_bpf_srh_state *srh_state =
6505 this_cpu_ptr(&seg6_bpf_srh_states);
6506 int hdroff = 0;
6507 int err;
6508
6509 switch (action) {
6510 case SEG6_LOCAL_ACTION_END_X:
6511 if (!seg6_bpf_has_valid_srh(skb))
6512 return -EBADMSG;
6513 if (param_len != sizeof(struct in6_addr))
6514 return -EINVAL;
6515 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6516 case SEG6_LOCAL_ACTION_END_T:
6517 if (!seg6_bpf_has_valid_srh(skb))
6518 return -EBADMSG;
6519 if (param_len != sizeof(int))
6520 return -EINVAL;
6521 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6522 case SEG6_LOCAL_ACTION_END_DT6:
6523 if (!seg6_bpf_has_valid_srh(skb))
6524 return -EBADMSG;
6525 if (param_len != sizeof(int))
6526 return -EINVAL;
6527
6528 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6529 return -EBADMSG;
6530 if (!pskb_pull(skb, hdroff))
6531 return -EBADMSG;
6532
6533 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6534 skb_reset_network_header(skb);
6535 skb_reset_transport_header(skb);
6536 skb->encapsulation = 0;
6537
6538 bpf_compute_data_pointers(skb);
6539 bpf_update_srh_state(skb);
6540 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6541 case SEG6_LOCAL_ACTION_END_B6:
6542 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6543 return -EBADMSG;
6544 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6545 param, param_len);
6546 if (!err)
6547 bpf_update_srh_state(skb);
6548
6549 return err;
6550 case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6551 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6552 return -EBADMSG;
6553 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6554 param, param_len);
6555 if (!err)
6556 bpf_update_srh_state(skb);
6557
6558 return err;
6559 default:
6560 return -EINVAL;
6561 }
6562 }
6563
6564 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6565 .func = bpf_lwt_seg6_action,
6566 .gpl_only = false,
6567 .ret_type = RET_INTEGER,
6568 .arg1_type = ARG_PTR_TO_CTX,
6569 .arg2_type = ARG_ANYTHING,
6570 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6571 .arg4_type = ARG_CONST_SIZE
6572 };
6573
BPF_CALL_3(bpf_lwt_seg6_adjust_srh,struct sk_buff *,skb,u32,offset,s32,len)6574 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6575 s32, len)
6576 {
6577 struct seg6_bpf_srh_state *srh_state =
6578 this_cpu_ptr(&seg6_bpf_srh_states);
6579 struct ipv6_sr_hdr *srh = srh_state->srh;
6580 void *srh_end, *srh_tlvs, *ptr;
6581 struct ipv6hdr *hdr;
6582 int srhoff = 0;
6583 int ret;
6584
6585 if (unlikely(srh == NULL))
6586 return -EINVAL;
6587
6588 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6589 ((srh->first_segment + 1) << 4));
6590 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6591 srh_state->hdrlen);
6592 ptr = skb->data + offset;
6593
6594 if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6595 return -EFAULT;
6596 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6597 return -EFAULT;
6598
6599 if (len > 0) {
6600 ret = skb_cow_head(skb, len);
6601 if (unlikely(ret < 0))
6602 return ret;
6603
6604 ret = bpf_skb_net_hdr_push(skb, offset, len);
6605 } else {
6606 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6607 }
6608
6609 bpf_compute_data_pointers(skb);
6610 if (unlikely(ret < 0))
6611 return ret;
6612
6613 hdr = (struct ipv6hdr *)skb->data;
6614 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6615
6616 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6617 return -EINVAL;
6618 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6619 srh_state->hdrlen += len;
6620 srh_state->valid = false;
6621 return 0;
6622 }
6623
6624 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6625 .func = bpf_lwt_seg6_adjust_srh,
6626 .gpl_only = false,
6627 .ret_type = RET_INTEGER,
6628 .arg1_type = ARG_PTR_TO_CTX,
6629 .arg2_type = ARG_ANYTHING,
6630 .arg3_type = ARG_ANYTHING,
6631 };
6632 #endif /* CONFIG_IPV6_SEG6_BPF */
6633
6634 #ifdef CONFIG_INET
sk_lookup(struct net * net,struct bpf_sock_tuple * tuple,int dif,int sdif,u8 family,u8 proto)6635 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6636 int dif, int sdif, u8 family, u8 proto)
6637 {
6638 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6639 bool refcounted = false;
6640 struct sock *sk = NULL;
6641
6642 if (family == AF_INET) {
6643 __be32 src4 = tuple->ipv4.saddr;
6644 __be32 dst4 = tuple->ipv4.daddr;
6645
6646 if (proto == IPPROTO_TCP)
6647 sk = __inet_lookup(net, hinfo, NULL, 0,
6648 src4, tuple->ipv4.sport,
6649 dst4, tuple->ipv4.dport,
6650 dif, sdif, &refcounted);
6651 else
6652 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6653 dst4, tuple->ipv4.dport,
6654 dif, sdif, net->ipv4.udp_table, NULL);
6655 #if IS_ENABLED(CONFIG_IPV6)
6656 } else {
6657 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6658 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6659
6660 if (proto == IPPROTO_TCP)
6661 sk = __inet6_lookup(net, hinfo, NULL, 0,
6662 src6, tuple->ipv6.sport,
6663 dst6, ntohs(tuple->ipv6.dport),
6664 dif, sdif, &refcounted);
6665 else if (likely(ipv6_bpf_stub))
6666 sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6667 src6, tuple->ipv6.sport,
6668 dst6, tuple->ipv6.dport,
6669 dif, sdif,
6670 net->ipv4.udp_table, NULL);
6671 #endif
6672 }
6673
6674 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6675 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6676 sk = NULL;
6677 }
6678 return sk;
6679 }
6680
6681 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6682 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6683 */
6684 static struct sock *
__bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,struct net * caller_net,u32 ifindex,u8 proto,u64 netns_id,u64 flags,int sdif)6685 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6686 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6687 u64 flags, int sdif)
6688 {
6689 struct sock *sk = NULL;
6690 struct net *net;
6691 u8 family;
6692
6693 if (len == sizeof(tuple->ipv4))
6694 family = AF_INET;
6695 else if (len == sizeof(tuple->ipv6))
6696 family = AF_INET6;
6697 else
6698 return NULL;
6699
6700 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6701 goto out;
6702
6703 if (sdif < 0) {
6704 if (family == AF_INET)
6705 sdif = inet_sdif(skb);
6706 else
6707 sdif = inet6_sdif(skb);
6708 }
6709
6710 if ((s32)netns_id < 0) {
6711 net = caller_net;
6712 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6713 } else {
6714 net = get_net_ns_by_id(caller_net, netns_id);
6715 if (unlikely(!net))
6716 goto out;
6717 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6718 put_net(net);
6719 }
6720
6721 out:
6722 return sk;
6723 }
6724
6725 static struct sock *
__bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,struct net * caller_net,u32 ifindex,u8 proto,u64 netns_id,u64 flags,int sdif)6726 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6727 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6728 u64 flags, int sdif)
6729 {
6730 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6731 ifindex, proto, netns_id, flags,
6732 sdif);
6733
6734 if (sk) {
6735 struct sock *sk2 = sk_to_full_sk(sk);
6736
6737 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6738 * sock refcnt is decremented to prevent a request_sock leak.
6739 */
6740 if (!sk_fullsock(sk2))
6741 sk2 = NULL;
6742 if (sk2 != sk) {
6743 sock_gen_put(sk);
6744 /* Ensure there is no need to bump sk2 refcnt */
6745 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6746 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6747 return NULL;
6748 }
6749 sk = sk2;
6750 }
6751 }
6752
6753 return sk;
6754 }
6755
6756 static struct sock *
bpf_skc_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6757 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6758 u8 proto, u64 netns_id, u64 flags)
6759 {
6760 struct net *caller_net;
6761 int ifindex;
6762
6763 if (skb->dev) {
6764 caller_net = dev_net(skb->dev);
6765 ifindex = skb->dev->ifindex;
6766 } else {
6767 caller_net = sock_net(skb->sk);
6768 ifindex = 0;
6769 }
6770
6771 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6772 netns_id, flags, -1);
6773 }
6774
6775 static struct sock *
bpf_sk_lookup(struct sk_buff * skb,struct bpf_sock_tuple * tuple,u32 len,u8 proto,u64 netns_id,u64 flags)6776 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6777 u8 proto, u64 netns_id, u64 flags)
6778 {
6779 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6780 flags);
6781
6782 if (sk) {
6783 struct sock *sk2 = sk_to_full_sk(sk);
6784
6785 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6786 * sock refcnt is decremented to prevent a request_sock leak.
6787 */
6788 if (!sk_fullsock(sk2))
6789 sk2 = NULL;
6790 if (sk2 != sk) {
6791 sock_gen_put(sk);
6792 /* Ensure there is no need to bump sk2 refcnt */
6793 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6794 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6795 return NULL;
6796 }
6797 sk = sk2;
6798 }
6799 }
6800
6801 return sk;
6802 }
6803
BPF_CALL_5(bpf_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6804 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6805 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6806 {
6807 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6808 netns_id, flags);
6809 }
6810
6811 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6812 .func = bpf_skc_lookup_tcp,
6813 .gpl_only = false,
6814 .pkt_access = true,
6815 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6816 .arg1_type = ARG_PTR_TO_CTX,
6817 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6818 .arg3_type = ARG_CONST_SIZE,
6819 .arg4_type = ARG_ANYTHING,
6820 .arg5_type = ARG_ANYTHING,
6821 };
6822
BPF_CALL_5(bpf_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6823 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6824 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6825 {
6826 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6827 netns_id, flags);
6828 }
6829
6830 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6831 .func = bpf_sk_lookup_tcp,
6832 .gpl_only = false,
6833 .pkt_access = true,
6834 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6835 .arg1_type = ARG_PTR_TO_CTX,
6836 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6837 .arg3_type = ARG_CONST_SIZE,
6838 .arg4_type = ARG_ANYTHING,
6839 .arg5_type = ARG_ANYTHING,
6840 };
6841
BPF_CALL_5(bpf_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6842 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6843 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6844 {
6845 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6846 netns_id, flags);
6847 }
6848
6849 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6850 .func = bpf_sk_lookup_udp,
6851 .gpl_only = false,
6852 .pkt_access = true,
6853 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6854 .arg1_type = ARG_PTR_TO_CTX,
6855 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6856 .arg3_type = ARG_CONST_SIZE,
6857 .arg4_type = ARG_ANYTHING,
6858 .arg5_type = ARG_ANYTHING,
6859 };
6860
BPF_CALL_5(bpf_tc_skc_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6861 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6862 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6863 {
6864 struct net_device *dev = skb->dev;
6865 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6866 struct net *caller_net = dev_net(dev);
6867
6868 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6869 ifindex, IPPROTO_TCP, netns_id,
6870 flags, sdif);
6871 }
6872
6873 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6874 .func = bpf_tc_skc_lookup_tcp,
6875 .gpl_only = false,
6876 .pkt_access = true,
6877 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6878 .arg1_type = ARG_PTR_TO_CTX,
6879 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6880 .arg3_type = ARG_CONST_SIZE,
6881 .arg4_type = ARG_ANYTHING,
6882 .arg5_type = ARG_ANYTHING,
6883 };
6884
BPF_CALL_5(bpf_tc_sk_lookup_tcp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6885 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6886 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6887 {
6888 struct net_device *dev = skb->dev;
6889 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6890 struct net *caller_net = dev_net(dev);
6891
6892 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6893 ifindex, IPPROTO_TCP, netns_id,
6894 flags, sdif);
6895 }
6896
6897 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6898 .func = bpf_tc_sk_lookup_tcp,
6899 .gpl_only = false,
6900 .pkt_access = true,
6901 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6902 .arg1_type = ARG_PTR_TO_CTX,
6903 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6904 .arg3_type = ARG_CONST_SIZE,
6905 .arg4_type = ARG_ANYTHING,
6906 .arg5_type = ARG_ANYTHING,
6907 };
6908
BPF_CALL_5(bpf_tc_sk_lookup_udp,struct sk_buff *,skb,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)6909 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6910 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6911 {
6912 struct net_device *dev = skb->dev;
6913 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6914 struct net *caller_net = dev_net(dev);
6915
6916 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6917 ifindex, IPPROTO_UDP, netns_id,
6918 flags, sdif);
6919 }
6920
6921 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6922 .func = bpf_tc_sk_lookup_udp,
6923 .gpl_only = false,
6924 .pkt_access = true,
6925 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6926 .arg1_type = ARG_PTR_TO_CTX,
6927 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6928 .arg3_type = ARG_CONST_SIZE,
6929 .arg4_type = ARG_ANYTHING,
6930 .arg5_type = ARG_ANYTHING,
6931 };
6932
BPF_CALL_1(bpf_sk_release,struct sock *,sk)6933 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6934 {
6935 if (sk && sk_is_refcounted(sk))
6936 sock_gen_put(sk);
6937 return 0;
6938 }
6939
6940 static const struct bpf_func_proto bpf_sk_release_proto = {
6941 .func = bpf_sk_release,
6942 .gpl_only = false,
6943 .ret_type = RET_INTEGER,
6944 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6945 };
6946
BPF_CALL_5(bpf_xdp_sk_lookup_udp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6947 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6948 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6949 {
6950 struct net_device *dev = ctx->rxq->dev;
6951 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6952 struct net *caller_net = dev_net(dev);
6953
6954 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6955 ifindex, IPPROTO_UDP, netns_id,
6956 flags, sdif);
6957 }
6958
6959 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6960 .func = bpf_xdp_sk_lookup_udp,
6961 .gpl_only = false,
6962 .pkt_access = true,
6963 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
6964 .arg1_type = ARG_PTR_TO_CTX,
6965 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6966 .arg3_type = ARG_CONST_SIZE,
6967 .arg4_type = ARG_ANYTHING,
6968 .arg5_type = ARG_ANYTHING,
6969 };
6970
BPF_CALL_5(bpf_xdp_skc_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6971 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6972 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6973 {
6974 struct net_device *dev = ctx->rxq->dev;
6975 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6976 struct net *caller_net = dev_net(dev);
6977
6978 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6979 ifindex, IPPROTO_TCP, netns_id,
6980 flags, sdif);
6981 }
6982
6983 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6984 .func = bpf_xdp_skc_lookup_tcp,
6985 .gpl_only = false,
6986 .pkt_access = true,
6987 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6988 .arg1_type = ARG_PTR_TO_CTX,
6989 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
6990 .arg3_type = ARG_CONST_SIZE,
6991 .arg4_type = ARG_ANYTHING,
6992 .arg5_type = ARG_ANYTHING,
6993 };
6994
BPF_CALL_5(bpf_xdp_sk_lookup_tcp,struct xdp_buff *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u32,netns_id,u64,flags)6995 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6996 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6997 {
6998 struct net_device *dev = ctx->rxq->dev;
6999 int ifindex = dev->ifindex, sdif = dev_sdif(dev);
7000 struct net *caller_net = dev_net(dev);
7001
7002 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
7003 ifindex, IPPROTO_TCP, netns_id,
7004 flags, sdif);
7005 }
7006
7007 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
7008 .func = bpf_xdp_sk_lookup_tcp,
7009 .gpl_only = false,
7010 .pkt_access = true,
7011 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7012 .arg1_type = ARG_PTR_TO_CTX,
7013 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7014 .arg3_type = ARG_CONST_SIZE,
7015 .arg4_type = ARG_ANYTHING,
7016 .arg5_type = ARG_ANYTHING,
7017 };
7018
BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7019 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7020 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7021 {
7022 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
7023 sock_net(ctx->sk), 0,
7024 IPPROTO_TCP, netns_id, flags,
7025 -1);
7026 }
7027
7028 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
7029 .func = bpf_sock_addr_skc_lookup_tcp,
7030 .gpl_only = false,
7031 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL,
7032 .arg1_type = ARG_PTR_TO_CTX,
7033 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7034 .arg3_type = ARG_CONST_SIZE,
7035 .arg4_type = ARG_ANYTHING,
7036 .arg5_type = ARG_ANYTHING,
7037 };
7038
BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7039 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
7040 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7041 {
7042 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7043 sock_net(ctx->sk), 0, IPPROTO_TCP,
7044 netns_id, flags, -1);
7045 }
7046
7047 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
7048 .func = bpf_sock_addr_sk_lookup_tcp,
7049 .gpl_only = false,
7050 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7051 .arg1_type = ARG_PTR_TO_CTX,
7052 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7053 .arg3_type = ARG_CONST_SIZE,
7054 .arg4_type = ARG_ANYTHING,
7055 .arg5_type = ARG_ANYTHING,
7056 };
7057
BPF_CALL_5(bpf_sock_addr_sk_lookup_udp,struct bpf_sock_addr_kern *,ctx,struct bpf_sock_tuple *,tuple,u32,len,u64,netns_id,u64,flags)7058 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
7059 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
7060 {
7061 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
7062 sock_net(ctx->sk), 0, IPPROTO_UDP,
7063 netns_id, flags, -1);
7064 }
7065
7066 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
7067 .func = bpf_sock_addr_sk_lookup_udp,
7068 .gpl_only = false,
7069 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7070 .arg1_type = ARG_PTR_TO_CTX,
7071 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7072 .arg3_type = ARG_CONST_SIZE,
7073 .arg4_type = ARG_ANYTHING,
7074 .arg5_type = ARG_ANYTHING,
7075 };
7076
bpf_tcp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7077 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7078 struct bpf_insn_access_aux *info)
7079 {
7080 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
7081 icsk_retransmits))
7082 return false;
7083
7084 if (off % size != 0)
7085 return false;
7086
7087 switch (off) {
7088 case offsetof(struct bpf_tcp_sock, bytes_received):
7089 case offsetof(struct bpf_tcp_sock, bytes_acked):
7090 return size == sizeof(__u64);
7091 default:
7092 return size == sizeof(__u32);
7093 }
7094 }
7095
bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)7096 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
7097 const struct bpf_insn *si,
7098 struct bpf_insn *insn_buf,
7099 struct bpf_prog *prog, u32 *target_size)
7100 {
7101 struct bpf_insn *insn = insn_buf;
7102
7103 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \
7104 do { \
7105 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \
7106 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7107 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
7108 si->dst_reg, si->src_reg, \
7109 offsetof(struct tcp_sock, FIELD)); \
7110 } while (0)
7111
7112 #define BPF_INET_SOCK_GET_COMMON(FIELD) \
7113 do { \
7114 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \
7115 FIELD) > \
7116 sizeof_field(struct bpf_tcp_sock, FIELD)); \
7117 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
7118 struct inet_connection_sock, \
7119 FIELD), \
7120 si->dst_reg, si->src_reg, \
7121 offsetof( \
7122 struct inet_connection_sock, \
7123 FIELD)); \
7124 } while (0)
7125
7126 BTF_TYPE_EMIT(struct bpf_tcp_sock);
7127
7128 switch (si->off) {
7129 case offsetof(struct bpf_tcp_sock, rtt_min):
7130 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7131 sizeof(struct minmax));
7132 BUILD_BUG_ON(sizeof(struct minmax) <
7133 sizeof(struct minmax_sample));
7134
7135 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7136 offsetof(struct tcp_sock, rtt_min) +
7137 offsetof(struct minmax_sample, v));
7138 break;
7139 case offsetof(struct bpf_tcp_sock, snd_cwnd):
7140 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7141 break;
7142 case offsetof(struct bpf_tcp_sock, srtt_us):
7143 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7144 break;
7145 case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7146 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7147 break;
7148 case offsetof(struct bpf_tcp_sock, rcv_nxt):
7149 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7150 break;
7151 case offsetof(struct bpf_tcp_sock, snd_nxt):
7152 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7153 break;
7154 case offsetof(struct bpf_tcp_sock, snd_una):
7155 BPF_TCP_SOCK_GET_COMMON(snd_una);
7156 break;
7157 case offsetof(struct bpf_tcp_sock, mss_cache):
7158 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7159 break;
7160 case offsetof(struct bpf_tcp_sock, ecn_flags):
7161 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7162 break;
7163 case offsetof(struct bpf_tcp_sock, rate_delivered):
7164 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7165 break;
7166 case offsetof(struct bpf_tcp_sock, rate_interval_us):
7167 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7168 break;
7169 case offsetof(struct bpf_tcp_sock, packets_out):
7170 BPF_TCP_SOCK_GET_COMMON(packets_out);
7171 break;
7172 case offsetof(struct bpf_tcp_sock, retrans_out):
7173 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7174 break;
7175 case offsetof(struct bpf_tcp_sock, total_retrans):
7176 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7177 break;
7178 case offsetof(struct bpf_tcp_sock, segs_in):
7179 BPF_TCP_SOCK_GET_COMMON(segs_in);
7180 break;
7181 case offsetof(struct bpf_tcp_sock, data_segs_in):
7182 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7183 break;
7184 case offsetof(struct bpf_tcp_sock, segs_out):
7185 BPF_TCP_SOCK_GET_COMMON(segs_out);
7186 break;
7187 case offsetof(struct bpf_tcp_sock, data_segs_out):
7188 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7189 break;
7190 case offsetof(struct bpf_tcp_sock, lost_out):
7191 BPF_TCP_SOCK_GET_COMMON(lost_out);
7192 break;
7193 case offsetof(struct bpf_tcp_sock, sacked_out):
7194 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7195 break;
7196 case offsetof(struct bpf_tcp_sock, bytes_received):
7197 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7198 break;
7199 case offsetof(struct bpf_tcp_sock, bytes_acked):
7200 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7201 break;
7202 case offsetof(struct bpf_tcp_sock, dsack_dups):
7203 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7204 break;
7205 case offsetof(struct bpf_tcp_sock, delivered):
7206 BPF_TCP_SOCK_GET_COMMON(delivered);
7207 break;
7208 case offsetof(struct bpf_tcp_sock, delivered_ce):
7209 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7210 break;
7211 case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7212 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7213 break;
7214 }
7215
7216 return insn - insn_buf;
7217 }
7218
BPF_CALL_1(bpf_tcp_sock,struct sock *,sk)7219 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7220 {
7221 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7222 return (unsigned long)sk;
7223
7224 return (unsigned long)NULL;
7225 }
7226
7227 const struct bpf_func_proto bpf_tcp_sock_proto = {
7228 .func = bpf_tcp_sock,
7229 .gpl_only = false,
7230 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL,
7231 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7232 };
7233
BPF_CALL_1(bpf_get_listener_sock,struct sock *,sk)7234 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7235 {
7236 sk = sk_to_full_sk(sk);
7237
7238 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7239 return (unsigned long)sk;
7240
7241 return (unsigned long)NULL;
7242 }
7243
7244 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7245 .func = bpf_get_listener_sock,
7246 .gpl_only = false,
7247 .ret_type = RET_PTR_TO_SOCKET_OR_NULL,
7248 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
7249 };
7250
BPF_CALL_1(bpf_skb_ecn_set_ce,struct sk_buff *,skb)7251 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7252 {
7253 unsigned int iphdr_len;
7254
7255 switch (skb_protocol(skb, true)) {
7256 case cpu_to_be16(ETH_P_IP):
7257 iphdr_len = sizeof(struct iphdr);
7258 break;
7259 case cpu_to_be16(ETH_P_IPV6):
7260 iphdr_len = sizeof(struct ipv6hdr);
7261 break;
7262 default:
7263 return 0;
7264 }
7265
7266 if (skb_headlen(skb) < iphdr_len)
7267 return 0;
7268
7269 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7270 return 0;
7271
7272 return INET_ECN_set_ce(skb);
7273 }
7274
bpf_xdp_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)7275 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7276 struct bpf_insn_access_aux *info)
7277 {
7278 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7279 return false;
7280
7281 if (off % size != 0)
7282 return false;
7283
7284 switch (off) {
7285 default:
7286 return size == sizeof(__u32);
7287 }
7288 }
7289
bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)7290 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7291 const struct bpf_insn *si,
7292 struct bpf_insn *insn_buf,
7293 struct bpf_prog *prog, u32 *target_size)
7294 {
7295 struct bpf_insn *insn = insn_buf;
7296
7297 #define BPF_XDP_SOCK_GET(FIELD) \
7298 do { \
7299 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \
7300 sizeof_field(struct bpf_xdp_sock, FIELD)); \
7301 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7302 si->dst_reg, si->src_reg, \
7303 offsetof(struct xdp_sock, FIELD)); \
7304 } while (0)
7305
7306 switch (si->off) {
7307 case offsetof(struct bpf_xdp_sock, queue_id):
7308 BPF_XDP_SOCK_GET(queue_id);
7309 break;
7310 }
7311
7312 return insn - insn_buf;
7313 }
7314
7315 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7316 .func = bpf_skb_ecn_set_ce,
7317 .gpl_only = false,
7318 .ret_type = RET_INTEGER,
7319 .arg1_type = ARG_PTR_TO_CTX,
7320 };
7321
BPF_CALL_5(bpf_tcp_check_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7322 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7323 struct tcphdr *, th, u32, th_len)
7324 {
7325 #ifdef CONFIG_SYN_COOKIES
7326 int ret;
7327
7328 if (unlikely(!sk || th_len < sizeof(*th)))
7329 return -EINVAL;
7330
7331 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7332 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7333 return -EINVAL;
7334
7335 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7336 return -EINVAL;
7337
7338 if (!th->ack || th->rst || th->syn)
7339 return -ENOENT;
7340
7341 if (unlikely(iph_len < sizeof(struct iphdr)))
7342 return -EINVAL;
7343
7344 if (tcp_synq_no_recent_overflow(sk))
7345 return -ENOENT;
7346
7347 /* Both struct iphdr and struct ipv6hdr have the version field at the
7348 * same offset so we can cast to the shorter header (struct iphdr).
7349 */
7350 switch (((struct iphdr *)iph)->version) {
7351 case 4:
7352 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7353 return -EINVAL;
7354
7355 ret = __cookie_v4_check((struct iphdr *)iph, th);
7356 break;
7357
7358 #if IS_BUILTIN(CONFIG_IPV6)
7359 case 6:
7360 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7361 return -EINVAL;
7362
7363 if (sk->sk_family != AF_INET6)
7364 return -EINVAL;
7365
7366 ret = __cookie_v6_check((struct ipv6hdr *)iph, th);
7367 break;
7368 #endif /* CONFIG_IPV6 */
7369
7370 default:
7371 return -EPROTONOSUPPORT;
7372 }
7373
7374 if (ret > 0)
7375 return 0;
7376
7377 return -ENOENT;
7378 #else
7379 return -ENOTSUPP;
7380 #endif
7381 }
7382
7383 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7384 .func = bpf_tcp_check_syncookie,
7385 .gpl_only = true,
7386 .pkt_access = true,
7387 .ret_type = RET_INTEGER,
7388 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7389 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7390 .arg3_type = ARG_CONST_SIZE,
7391 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7392 .arg5_type = ARG_CONST_SIZE,
7393 };
7394
BPF_CALL_5(bpf_tcp_gen_syncookie,struct sock *,sk,void *,iph,u32,iph_len,struct tcphdr *,th,u32,th_len)7395 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7396 struct tcphdr *, th, u32, th_len)
7397 {
7398 #ifdef CONFIG_SYN_COOKIES
7399 u32 cookie;
7400 u16 mss;
7401
7402 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7403 return -EINVAL;
7404
7405 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7406 return -EINVAL;
7407
7408 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7409 return -ENOENT;
7410
7411 if (!th->syn || th->ack || th->fin || th->rst)
7412 return -EINVAL;
7413
7414 if (unlikely(iph_len < sizeof(struct iphdr)))
7415 return -EINVAL;
7416
7417 /* Both struct iphdr and struct ipv6hdr have the version field at the
7418 * same offset so we can cast to the shorter header (struct iphdr).
7419 */
7420 switch (((struct iphdr *)iph)->version) {
7421 case 4:
7422 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7423 return -EINVAL;
7424
7425 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7426 break;
7427
7428 #if IS_BUILTIN(CONFIG_IPV6)
7429 case 6:
7430 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7431 return -EINVAL;
7432
7433 if (sk->sk_family != AF_INET6)
7434 return -EINVAL;
7435
7436 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7437 break;
7438 #endif /* CONFIG_IPV6 */
7439
7440 default:
7441 return -EPROTONOSUPPORT;
7442 }
7443 if (mss == 0)
7444 return -ENOENT;
7445
7446 return cookie | ((u64)mss << 32);
7447 #else
7448 return -EOPNOTSUPP;
7449 #endif /* CONFIG_SYN_COOKIES */
7450 }
7451
7452 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7453 .func = bpf_tcp_gen_syncookie,
7454 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */
7455 .pkt_access = true,
7456 .ret_type = RET_INTEGER,
7457 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7458 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7459 .arg3_type = ARG_CONST_SIZE,
7460 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7461 .arg5_type = ARG_CONST_SIZE,
7462 };
7463
BPF_CALL_3(bpf_sk_assign,struct sk_buff *,skb,struct sock *,sk,u64,flags)7464 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7465 {
7466 if (!sk || flags != 0)
7467 return -EINVAL;
7468 if (!skb_at_tc_ingress(skb))
7469 return -EOPNOTSUPP;
7470 if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7471 return -ENETUNREACH;
7472 if (sk_unhashed(sk))
7473 return -EOPNOTSUPP;
7474 if (sk_is_refcounted(sk) &&
7475 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7476 return -ENOENT;
7477
7478 skb_orphan(skb);
7479 skb->sk = sk;
7480 skb->destructor = sock_pfree;
7481
7482 return 0;
7483 }
7484
7485 static const struct bpf_func_proto bpf_sk_assign_proto = {
7486 .func = bpf_sk_assign,
7487 .gpl_only = false,
7488 .ret_type = RET_INTEGER,
7489 .arg1_type = ARG_PTR_TO_CTX,
7490 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7491 .arg3_type = ARG_ANYTHING,
7492 };
7493
bpf_search_tcp_opt(const u8 * op,const u8 * opend,u8 search_kind,const u8 * magic,u8 magic_len,bool * eol)7494 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7495 u8 search_kind, const u8 *magic,
7496 u8 magic_len, bool *eol)
7497 {
7498 u8 kind, kind_len;
7499
7500 *eol = false;
7501
7502 while (op < opend) {
7503 kind = op[0];
7504
7505 if (kind == TCPOPT_EOL) {
7506 *eol = true;
7507 return ERR_PTR(-ENOMSG);
7508 } else if (kind == TCPOPT_NOP) {
7509 op++;
7510 continue;
7511 }
7512
7513 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7514 /* Something is wrong in the received header.
7515 * Follow the TCP stack's tcp_parse_options()
7516 * and just bail here.
7517 */
7518 return ERR_PTR(-EFAULT);
7519
7520 kind_len = op[1];
7521 if (search_kind == kind) {
7522 if (!magic_len)
7523 return op;
7524
7525 if (magic_len > kind_len - 2)
7526 return ERR_PTR(-ENOMSG);
7527
7528 if (!memcmp(&op[2], magic, magic_len))
7529 return op;
7530 }
7531
7532 op += kind_len;
7533 }
7534
7535 return ERR_PTR(-ENOMSG);
7536 }
7537
BPF_CALL_4(bpf_sock_ops_load_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,void *,search_res,u32,len,u64,flags)7538 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7539 void *, search_res, u32, len, u64, flags)
7540 {
7541 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7542 const u8 *op, *opend, *magic, *search = search_res;
7543 u8 search_kind, search_len, copy_len, magic_len;
7544 int ret;
7545
7546 /* 2 byte is the minimal option len except TCPOPT_NOP and
7547 * TCPOPT_EOL which are useless for the bpf prog to learn
7548 * and this helper disallow loading them also.
7549 */
7550 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7551 return -EINVAL;
7552
7553 search_kind = search[0];
7554 search_len = search[1];
7555
7556 if (search_len > len || search_kind == TCPOPT_NOP ||
7557 search_kind == TCPOPT_EOL)
7558 return -EINVAL;
7559
7560 if (search_kind == TCPOPT_EXP || search_kind == 253) {
7561 /* 16 or 32 bit magic. +2 for kind and kind length */
7562 if (search_len != 4 && search_len != 6)
7563 return -EINVAL;
7564 magic = &search[2];
7565 magic_len = search_len - 2;
7566 } else {
7567 if (search_len)
7568 return -EINVAL;
7569 magic = NULL;
7570 magic_len = 0;
7571 }
7572
7573 if (load_syn) {
7574 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7575 if (ret < 0)
7576 return ret;
7577
7578 opend = op + ret;
7579 op += sizeof(struct tcphdr);
7580 } else {
7581 if (!bpf_sock->skb ||
7582 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7583 /* This bpf_sock->op cannot call this helper */
7584 return -EPERM;
7585
7586 opend = bpf_sock->skb_data_end;
7587 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7588 }
7589
7590 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7591 &eol);
7592 if (IS_ERR(op))
7593 return PTR_ERR(op);
7594
7595 copy_len = op[1];
7596 ret = copy_len;
7597 if (copy_len > len) {
7598 ret = -ENOSPC;
7599 copy_len = len;
7600 }
7601
7602 memcpy(search_res, op, copy_len);
7603 return ret;
7604 }
7605
7606 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7607 .func = bpf_sock_ops_load_hdr_opt,
7608 .gpl_only = false,
7609 .ret_type = RET_INTEGER,
7610 .arg1_type = ARG_PTR_TO_CTX,
7611 .arg2_type = ARG_PTR_TO_MEM,
7612 .arg3_type = ARG_CONST_SIZE,
7613 .arg4_type = ARG_ANYTHING,
7614 };
7615
BPF_CALL_4(bpf_sock_ops_store_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,const void *,from,u32,len,u64,flags)7616 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7617 const void *, from, u32, len, u64, flags)
7618 {
7619 u8 new_kind, new_kind_len, magic_len = 0, *opend;
7620 const u8 *op, *new_op, *magic = NULL;
7621 struct sk_buff *skb;
7622 bool eol;
7623
7624 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7625 return -EPERM;
7626
7627 if (len < 2 || flags)
7628 return -EINVAL;
7629
7630 new_op = from;
7631 new_kind = new_op[0];
7632 new_kind_len = new_op[1];
7633
7634 if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7635 new_kind == TCPOPT_EOL)
7636 return -EINVAL;
7637
7638 if (new_kind_len > bpf_sock->remaining_opt_len)
7639 return -ENOSPC;
7640
7641 /* 253 is another experimental kind */
7642 if (new_kind == TCPOPT_EXP || new_kind == 253) {
7643 if (new_kind_len < 4)
7644 return -EINVAL;
7645 /* Match for the 2 byte magic also.
7646 * RFC 6994: the magic could be 2 or 4 bytes.
7647 * Hence, matching by 2 byte only is on the
7648 * conservative side but it is the right
7649 * thing to do for the 'search-for-duplication'
7650 * purpose.
7651 */
7652 magic = &new_op[2];
7653 magic_len = 2;
7654 }
7655
7656 /* Check for duplication */
7657 skb = bpf_sock->skb;
7658 op = skb->data + sizeof(struct tcphdr);
7659 opend = bpf_sock->skb_data_end;
7660
7661 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7662 &eol);
7663 if (!IS_ERR(op))
7664 return -EEXIST;
7665
7666 if (PTR_ERR(op) != -ENOMSG)
7667 return PTR_ERR(op);
7668
7669 if (eol)
7670 /* The option has been ended. Treat it as no more
7671 * header option can be written.
7672 */
7673 return -ENOSPC;
7674
7675 /* No duplication found. Store the header option. */
7676 memcpy(opend, from, new_kind_len);
7677
7678 bpf_sock->remaining_opt_len -= new_kind_len;
7679 bpf_sock->skb_data_end += new_kind_len;
7680
7681 return 0;
7682 }
7683
7684 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7685 .func = bpf_sock_ops_store_hdr_opt,
7686 .gpl_only = false,
7687 .ret_type = RET_INTEGER,
7688 .arg1_type = ARG_PTR_TO_CTX,
7689 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
7690 .arg3_type = ARG_CONST_SIZE,
7691 .arg4_type = ARG_ANYTHING,
7692 };
7693
BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt,struct bpf_sock_ops_kern *,bpf_sock,u32,len,u64,flags)7694 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7695 u32, len, u64, flags)
7696 {
7697 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7698 return -EPERM;
7699
7700 if (flags || len < 2)
7701 return -EINVAL;
7702
7703 if (len > bpf_sock->remaining_opt_len)
7704 return -ENOSPC;
7705
7706 bpf_sock->remaining_opt_len -= len;
7707
7708 return 0;
7709 }
7710
7711 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7712 .func = bpf_sock_ops_reserve_hdr_opt,
7713 .gpl_only = false,
7714 .ret_type = RET_INTEGER,
7715 .arg1_type = ARG_PTR_TO_CTX,
7716 .arg2_type = ARG_ANYTHING,
7717 .arg3_type = ARG_ANYTHING,
7718 };
7719
BPF_CALL_3(bpf_skb_set_tstamp,struct sk_buff *,skb,u64,tstamp,u32,tstamp_type)7720 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7721 u64, tstamp, u32, tstamp_type)
7722 {
7723 /* skb_clear_delivery_time() is done for inet protocol */
7724 if (skb->protocol != htons(ETH_P_IP) &&
7725 skb->protocol != htons(ETH_P_IPV6))
7726 return -EOPNOTSUPP;
7727
7728 switch (tstamp_type) {
7729 case BPF_SKB_TSTAMP_DELIVERY_MONO:
7730 if (!tstamp)
7731 return -EINVAL;
7732 skb->tstamp = tstamp;
7733 skb->mono_delivery_time = 1;
7734 break;
7735 case BPF_SKB_TSTAMP_UNSPEC:
7736 if (tstamp)
7737 return -EINVAL;
7738 skb->tstamp = 0;
7739 skb->mono_delivery_time = 0;
7740 break;
7741 default:
7742 return -EINVAL;
7743 }
7744
7745 return 0;
7746 }
7747
7748 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7749 .func = bpf_skb_set_tstamp,
7750 .gpl_only = false,
7751 .ret_type = RET_INTEGER,
7752 .arg1_type = ARG_PTR_TO_CTX,
7753 .arg2_type = ARG_ANYTHING,
7754 .arg3_type = ARG_ANYTHING,
7755 };
7756
7757 #ifdef CONFIG_SYN_COOKIES
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th,u32,th_len)7758 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7759 struct tcphdr *, th, u32, th_len)
7760 {
7761 u32 cookie;
7762 u16 mss;
7763
7764 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7765 return -EINVAL;
7766
7767 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7768 cookie = __cookie_v4_init_sequence(iph, th, &mss);
7769
7770 return cookie | ((u64)mss << 32);
7771 }
7772
7773 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7774 .func = bpf_tcp_raw_gen_syncookie_ipv4,
7775 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */
7776 .pkt_access = true,
7777 .ret_type = RET_INTEGER,
7778 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7779 .arg1_size = sizeof(struct iphdr),
7780 .arg2_type = ARG_PTR_TO_MEM,
7781 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7782 };
7783
BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th,u32,th_len)7784 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7785 struct tcphdr *, th, u32, th_len)
7786 {
7787 #if IS_BUILTIN(CONFIG_IPV6)
7788 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7789 sizeof(struct ipv6hdr);
7790 u32 cookie;
7791 u16 mss;
7792
7793 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7794 return -EINVAL;
7795
7796 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7797 cookie = __cookie_v6_init_sequence(iph, th, &mss);
7798
7799 return cookie | ((u64)mss << 32);
7800 #else
7801 return -EPROTONOSUPPORT;
7802 #endif
7803 }
7804
7805 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7806 .func = bpf_tcp_raw_gen_syncookie_ipv6,
7807 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */
7808 .pkt_access = true,
7809 .ret_type = RET_INTEGER,
7810 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7811 .arg1_size = sizeof(struct ipv6hdr),
7812 .arg2_type = ARG_PTR_TO_MEM,
7813 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
7814 };
7815
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4,struct iphdr *,iph,struct tcphdr *,th)7816 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7817 struct tcphdr *, th)
7818 {
7819 if (__cookie_v4_check(iph, th) > 0)
7820 return 0;
7821
7822 return -EACCES;
7823 }
7824
7825 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7826 .func = bpf_tcp_raw_check_syncookie_ipv4,
7827 .gpl_only = true, /* __cookie_v4_check is GPL */
7828 .pkt_access = true,
7829 .ret_type = RET_INTEGER,
7830 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7831 .arg1_size = sizeof(struct iphdr),
7832 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7833 .arg2_size = sizeof(struct tcphdr),
7834 };
7835
BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6,struct ipv6hdr *,iph,struct tcphdr *,th)7836 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7837 struct tcphdr *, th)
7838 {
7839 #if IS_BUILTIN(CONFIG_IPV6)
7840 if (__cookie_v6_check(iph, th) > 0)
7841 return 0;
7842
7843 return -EACCES;
7844 #else
7845 return -EPROTONOSUPPORT;
7846 #endif
7847 }
7848
7849 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7850 .func = bpf_tcp_raw_check_syncookie_ipv6,
7851 .gpl_only = true, /* __cookie_v6_check is GPL */
7852 .pkt_access = true,
7853 .ret_type = RET_INTEGER,
7854 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7855 .arg1_size = sizeof(struct ipv6hdr),
7856 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM,
7857 .arg2_size = sizeof(struct tcphdr),
7858 };
7859 #endif /* CONFIG_SYN_COOKIES */
7860
7861 #endif /* CONFIG_INET */
7862
bpf_helper_changes_pkt_data(void * func)7863 bool bpf_helper_changes_pkt_data(void *func)
7864 {
7865 if (func == bpf_skb_vlan_push ||
7866 func == bpf_skb_vlan_pop ||
7867 func == bpf_skb_store_bytes ||
7868 func == bpf_skb_change_proto ||
7869 func == bpf_skb_change_head ||
7870 func == sk_skb_change_head ||
7871 func == bpf_skb_change_tail ||
7872 func == sk_skb_change_tail ||
7873 func == bpf_skb_adjust_room ||
7874 func == sk_skb_adjust_room ||
7875 func == bpf_skb_pull_data ||
7876 func == sk_skb_pull_data ||
7877 func == bpf_clone_redirect ||
7878 func == bpf_l3_csum_replace ||
7879 func == bpf_l4_csum_replace ||
7880 func == bpf_xdp_adjust_head ||
7881 func == bpf_xdp_adjust_meta ||
7882 func == bpf_msg_pull_data ||
7883 func == bpf_msg_push_data ||
7884 func == bpf_msg_pop_data ||
7885 func == bpf_xdp_adjust_tail ||
7886 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7887 func == bpf_lwt_seg6_store_bytes ||
7888 func == bpf_lwt_seg6_adjust_srh ||
7889 func == bpf_lwt_seg6_action ||
7890 #endif
7891 #ifdef CONFIG_INET
7892 func == bpf_sock_ops_store_hdr_opt ||
7893 #endif
7894 func == bpf_lwt_in_push_encap ||
7895 func == bpf_lwt_xmit_push_encap)
7896 return true;
7897
7898 return false;
7899 }
7900
7901 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7902 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7903
7904 static const struct bpf_func_proto *
sock_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7905 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7906 {
7907 const struct bpf_func_proto *func_proto;
7908
7909 func_proto = cgroup_common_func_proto(func_id, prog);
7910 if (func_proto)
7911 return func_proto;
7912
7913 func_proto = cgroup_current_func_proto(func_id, prog);
7914 if (func_proto)
7915 return func_proto;
7916
7917 switch (func_id) {
7918 case BPF_FUNC_get_socket_cookie:
7919 return &bpf_get_socket_cookie_sock_proto;
7920 case BPF_FUNC_get_netns_cookie:
7921 return &bpf_get_netns_cookie_sock_proto;
7922 case BPF_FUNC_perf_event_output:
7923 return &bpf_event_output_data_proto;
7924 case BPF_FUNC_sk_storage_get:
7925 return &bpf_sk_storage_get_cg_sock_proto;
7926 case BPF_FUNC_ktime_get_coarse_ns:
7927 return &bpf_ktime_get_coarse_ns_proto;
7928 default:
7929 return bpf_base_func_proto(func_id, prog);
7930 }
7931 }
7932
7933 static const struct bpf_func_proto *
sock_addr_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)7934 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7935 {
7936 const struct bpf_func_proto *func_proto;
7937
7938 func_proto = cgroup_common_func_proto(func_id, prog);
7939 if (func_proto)
7940 return func_proto;
7941
7942 func_proto = cgroup_current_func_proto(func_id, prog);
7943 if (func_proto)
7944 return func_proto;
7945
7946 switch (func_id) {
7947 case BPF_FUNC_bind:
7948 switch (prog->expected_attach_type) {
7949 case BPF_CGROUP_INET4_CONNECT:
7950 case BPF_CGROUP_INET6_CONNECT:
7951 return &bpf_bind_proto;
7952 default:
7953 return NULL;
7954 }
7955 case BPF_FUNC_get_socket_cookie:
7956 return &bpf_get_socket_cookie_sock_addr_proto;
7957 case BPF_FUNC_get_netns_cookie:
7958 return &bpf_get_netns_cookie_sock_addr_proto;
7959 case BPF_FUNC_perf_event_output:
7960 return &bpf_event_output_data_proto;
7961 #ifdef CONFIG_INET
7962 case BPF_FUNC_sk_lookup_tcp:
7963 return &bpf_sock_addr_sk_lookup_tcp_proto;
7964 case BPF_FUNC_sk_lookup_udp:
7965 return &bpf_sock_addr_sk_lookup_udp_proto;
7966 case BPF_FUNC_sk_release:
7967 return &bpf_sk_release_proto;
7968 case BPF_FUNC_skc_lookup_tcp:
7969 return &bpf_sock_addr_skc_lookup_tcp_proto;
7970 #endif /* CONFIG_INET */
7971 case BPF_FUNC_sk_storage_get:
7972 return &bpf_sk_storage_get_proto;
7973 case BPF_FUNC_sk_storage_delete:
7974 return &bpf_sk_storage_delete_proto;
7975 case BPF_FUNC_setsockopt:
7976 switch (prog->expected_attach_type) {
7977 case BPF_CGROUP_INET4_BIND:
7978 case BPF_CGROUP_INET6_BIND:
7979 case BPF_CGROUP_INET4_CONNECT:
7980 case BPF_CGROUP_INET6_CONNECT:
7981 case BPF_CGROUP_UNIX_CONNECT:
7982 case BPF_CGROUP_UDP4_RECVMSG:
7983 case BPF_CGROUP_UDP6_RECVMSG:
7984 case BPF_CGROUP_UNIX_RECVMSG:
7985 case BPF_CGROUP_UDP4_SENDMSG:
7986 case BPF_CGROUP_UDP6_SENDMSG:
7987 case BPF_CGROUP_UNIX_SENDMSG:
7988 case BPF_CGROUP_INET4_GETPEERNAME:
7989 case BPF_CGROUP_INET6_GETPEERNAME:
7990 case BPF_CGROUP_UNIX_GETPEERNAME:
7991 case BPF_CGROUP_INET4_GETSOCKNAME:
7992 case BPF_CGROUP_INET6_GETSOCKNAME:
7993 case BPF_CGROUP_UNIX_GETSOCKNAME:
7994 return &bpf_sock_addr_setsockopt_proto;
7995 default:
7996 return NULL;
7997 }
7998 case BPF_FUNC_getsockopt:
7999 switch (prog->expected_attach_type) {
8000 case BPF_CGROUP_INET4_BIND:
8001 case BPF_CGROUP_INET6_BIND:
8002 case BPF_CGROUP_INET4_CONNECT:
8003 case BPF_CGROUP_INET6_CONNECT:
8004 case BPF_CGROUP_UNIX_CONNECT:
8005 case BPF_CGROUP_UDP4_RECVMSG:
8006 case BPF_CGROUP_UDP6_RECVMSG:
8007 case BPF_CGROUP_UNIX_RECVMSG:
8008 case BPF_CGROUP_UDP4_SENDMSG:
8009 case BPF_CGROUP_UDP6_SENDMSG:
8010 case BPF_CGROUP_UNIX_SENDMSG:
8011 case BPF_CGROUP_INET4_GETPEERNAME:
8012 case BPF_CGROUP_INET6_GETPEERNAME:
8013 case BPF_CGROUP_UNIX_GETPEERNAME:
8014 case BPF_CGROUP_INET4_GETSOCKNAME:
8015 case BPF_CGROUP_INET6_GETSOCKNAME:
8016 case BPF_CGROUP_UNIX_GETSOCKNAME:
8017 return &bpf_sock_addr_getsockopt_proto;
8018 default:
8019 return NULL;
8020 }
8021 default:
8022 return bpf_sk_base_func_proto(func_id, prog);
8023 }
8024 }
8025
8026 static const struct bpf_func_proto *
sk_filter_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8027 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8028 {
8029 switch (func_id) {
8030 case BPF_FUNC_skb_load_bytes:
8031 return &bpf_skb_load_bytes_proto;
8032 case BPF_FUNC_skb_load_bytes_relative:
8033 return &bpf_skb_load_bytes_relative_proto;
8034 case BPF_FUNC_get_socket_cookie:
8035 return &bpf_get_socket_cookie_proto;
8036 case BPF_FUNC_get_socket_uid:
8037 return &bpf_get_socket_uid_proto;
8038 case BPF_FUNC_perf_event_output:
8039 return &bpf_skb_event_output_proto;
8040 default:
8041 return bpf_sk_base_func_proto(func_id, prog);
8042 }
8043 }
8044
8045 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
8046 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
8047
8048 static const struct bpf_func_proto *
cg_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8049 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8050 {
8051 const struct bpf_func_proto *func_proto;
8052
8053 func_proto = cgroup_common_func_proto(func_id, prog);
8054 if (func_proto)
8055 return func_proto;
8056
8057 switch (func_id) {
8058 case BPF_FUNC_sk_fullsock:
8059 return &bpf_sk_fullsock_proto;
8060 case BPF_FUNC_sk_storage_get:
8061 return &bpf_sk_storage_get_proto;
8062 case BPF_FUNC_sk_storage_delete:
8063 return &bpf_sk_storage_delete_proto;
8064 case BPF_FUNC_perf_event_output:
8065 return &bpf_skb_event_output_proto;
8066 #ifdef CONFIG_SOCK_CGROUP_DATA
8067 case BPF_FUNC_skb_cgroup_id:
8068 return &bpf_skb_cgroup_id_proto;
8069 case BPF_FUNC_skb_ancestor_cgroup_id:
8070 return &bpf_skb_ancestor_cgroup_id_proto;
8071 case BPF_FUNC_sk_cgroup_id:
8072 return &bpf_sk_cgroup_id_proto;
8073 case BPF_FUNC_sk_ancestor_cgroup_id:
8074 return &bpf_sk_ancestor_cgroup_id_proto;
8075 #endif
8076 #ifdef CONFIG_INET
8077 case BPF_FUNC_sk_lookup_tcp:
8078 return &bpf_sk_lookup_tcp_proto;
8079 case BPF_FUNC_sk_lookup_udp:
8080 return &bpf_sk_lookup_udp_proto;
8081 case BPF_FUNC_sk_release:
8082 return &bpf_sk_release_proto;
8083 case BPF_FUNC_skc_lookup_tcp:
8084 return &bpf_skc_lookup_tcp_proto;
8085 case BPF_FUNC_tcp_sock:
8086 return &bpf_tcp_sock_proto;
8087 case BPF_FUNC_get_listener_sock:
8088 return &bpf_get_listener_sock_proto;
8089 case BPF_FUNC_skb_ecn_set_ce:
8090 return &bpf_skb_ecn_set_ce_proto;
8091 #endif
8092 default:
8093 return sk_filter_func_proto(func_id, prog);
8094 }
8095 }
8096
8097 static const struct bpf_func_proto *
tc_cls_act_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8098 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8099 {
8100 switch (func_id) {
8101 case BPF_FUNC_skb_store_bytes:
8102 return &bpf_skb_store_bytes_proto;
8103 case BPF_FUNC_skb_load_bytes:
8104 return &bpf_skb_load_bytes_proto;
8105 case BPF_FUNC_skb_load_bytes_relative:
8106 return &bpf_skb_load_bytes_relative_proto;
8107 case BPF_FUNC_skb_pull_data:
8108 return &bpf_skb_pull_data_proto;
8109 case BPF_FUNC_csum_diff:
8110 return &bpf_csum_diff_proto;
8111 case BPF_FUNC_csum_update:
8112 return &bpf_csum_update_proto;
8113 case BPF_FUNC_csum_level:
8114 return &bpf_csum_level_proto;
8115 case BPF_FUNC_l3_csum_replace:
8116 return &bpf_l3_csum_replace_proto;
8117 case BPF_FUNC_l4_csum_replace:
8118 return &bpf_l4_csum_replace_proto;
8119 case BPF_FUNC_clone_redirect:
8120 return &bpf_clone_redirect_proto;
8121 case BPF_FUNC_get_cgroup_classid:
8122 return &bpf_get_cgroup_classid_proto;
8123 case BPF_FUNC_skb_vlan_push:
8124 return &bpf_skb_vlan_push_proto;
8125 case BPF_FUNC_skb_vlan_pop:
8126 return &bpf_skb_vlan_pop_proto;
8127 case BPF_FUNC_skb_change_proto:
8128 return &bpf_skb_change_proto_proto;
8129 case BPF_FUNC_skb_change_type:
8130 return &bpf_skb_change_type_proto;
8131 case BPF_FUNC_skb_adjust_room:
8132 return &bpf_skb_adjust_room_proto;
8133 case BPF_FUNC_skb_change_tail:
8134 return &bpf_skb_change_tail_proto;
8135 case BPF_FUNC_skb_change_head:
8136 return &bpf_skb_change_head_proto;
8137 case BPF_FUNC_skb_get_tunnel_key:
8138 return &bpf_skb_get_tunnel_key_proto;
8139 case BPF_FUNC_skb_set_tunnel_key:
8140 return bpf_get_skb_set_tunnel_proto(func_id);
8141 case BPF_FUNC_skb_get_tunnel_opt:
8142 return &bpf_skb_get_tunnel_opt_proto;
8143 case BPF_FUNC_skb_set_tunnel_opt:
8144 return bpf_get_skb_set_tunnel_proto(func_id);
8145 case BPF_FUNC_redirect:
8146 return &bpf_redirect_proto;
8147 case BPF_FUNC_redirect_neigh:
8148 return &bpf_redirect_neigh_proto;
8149 case BPF_FUNC_redirect_peer:
8150 return &bpf_redirect_peer_proto;
8151 case BPF_FUNC_get_route_realm:
8152 return &bpf_get_route_realm_proto;
8153 case BPF_FUNC_get_hash_recalc:
8154 return &bpf_get_hash_recalc_proto;
8155 case BPF_FUNC_set_hash_invalid:
8156 return &bpf_set_hash_invalid_proto;
8157 case BPF_FUNC_set_hash:
8158 return &bpf_set_hash_proto;
8159 case BPF_FUNC_perf_event_output:
8160 return &bpf_skb_event_output_proto;
8161 case BPF_FUNC_get_smp_processor_id:
8162 return &bpf_get_smp_processor_id_proto;
8163 case BPF_FUNC_skb_under_cgroup:
8164 return &bpf_skb_under_cgroup_proto;
8165 case BPF_FUNC_get_socket_cookie:
8166 return &bpf_get_socket_cookie_proto;
8167 case BPF_FUNC_get_socket_uid:
8168 return &bpf_get_socket_uid_proto;
8169 case BPF_FUNC_fib_lookup:
8170 return &bpf_skb_fib_lookup_proto;
8171 case BPF_FUNC_check_mtu:
8172 return &bpf_skb_check_mtu_proto;
8173 case BPF_FUNC_sk_fullsock:
8174 return &bpf_sk_fullsock_proto;
8175 case BPF_FUNC_sk_storage_get:
8176 return &bpf_sk_storage_get_proto;
8177 case BPF_FUNC_sk_storage_delete:
8178 return &bpf_sk_storage_delete_proto;
8179 #ifdef CONFIG_XFRM
8180 case BPF_FUNC_skb_get_xfrm_state:
8181 return &bpf_skb_get_xfrm_state_proto;
8182 #endif
8183 #ifdef CONFIG_CGROUP_NET_CLASSID
8184 case BPF_FUNC_skb_cgroup_classid:
8185 return &bpf_skb_cgroup_classid_proto;
8186 #endif
8187 #ifdef CONFIG_SOCK_CGROUP_DATA
8188 case BPF_FUNC_skb_cgroup_id:
8189 return &bpf_skb_cgroup_id_proto;
8190 case BPF_FUNC_skb_ancestor_cgroup_id:
8191 return &bpf_skb_ancestor_cgroup_id_proto;
8192 #endif
8193 #ifdef CONFIG_INET
8194 case BPF_FUNC_sk_lookup_tcp:
8195 return &bpf_tc_sk_lookup_tcp_proto;
8196 case BPF_FUNC_sk_lookup_udp:
8197 return &bpf_tc_sk_lookup_udp_proto;
8198 case BPF_FUNC_sk_release:
8199 return &bpf_sk_release_proto;
8200 case BPF_FUNC_tcp_sock:
8201 return &bpf_tcp_sock_proto;
8202 case BPF_FUNC_get_listener_sock:
8203 return &bpf_get_listener_sock_proto;
8204 case BPF_FUNC_skc_lookup_tcp:
8205 return &bpf_tc_skc_lookup_tcp_proto;
8206 case BPF_FUNC_tcp_check_syncookie:
8207 return &bpf_tcp_check_syncookie_proto;
8208 case BPF_FUNC_skb_ecn_set_ce:
8209 return &bpf_skb_ecn_set_ce_proto;
8210 case BPF_FUNC_tcp_gen_syncookie:
8211 return &bpf_tcp_gen_syncookie_proto;
8212 case BPF_FUNC_sk_assign:
8213 return &bpf_sk_assign_proto;
8214 case BPF_FUNC_skb_set_tstamp:
8215 return &bpf_skb_set_tstamp_proto;
8216 #ifdef CONFIG_SYN_COOKIES
8217 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8218 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8219 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8220 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8221 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8222 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8223 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8224 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8225 #endif
8226 #endif
8227 default:
8228 return bpf_sk_base_func_proto(func_id, prog);
8229 }
8230 }
8231
8232 static const struct bpf_func_proto *
xdp_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8233 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8234 {
8235 switch (func_id) {
8236 case BPF_FUNC_perf_event_output:
8237 return &bpf_xdp_event_output_proto;
8238 case BPF_FUNC_get_smp_processor_id:
8239 return &bpf_get_smp_processor_id_proto;
8240 case BPF_FUNC_csum_diff:
8241 return &bpf_csum_diff_proto;
8242 case BPF_FUNC_xdp_adjust_head:
8243 return &bpf_xdp_adjust_head_proto;
8244 case BPF_FUNC_xdp_adjust_meta:
8245 return &bpf_xdp_adjust_meta_proto;
8246 case BPF_FUNC_redirect:
8247 return &bpf_xdp_redirect_proto;
8248 case BPF_FUNC_redirect_map:
8249 return &bpf_xdp_redirect_map_proto;
8250 case BPF_FUNC_xdp_adjust_tail:
8251 return &bpf_xdp_adjust_tail_proto;
8252 case BPF_FUNC_xdp_get_buff_len:
8253 return &bpf_xdp_get_buff_len_proto;
8254 case BPF_FUNC_xdp_load_bytes:
8255 return &bpf_xdp_load_bytes_proto;
8256 case BPF_FUNC_xdp_store_bytes:
8257 return &bpf_xdp_store_bytes_proto;
8258 case BPF_FUNC_fib_lookup:
8259 return &bpf_xdp_fib_lookup_proto;
8260 case BPF_FUNC_check_mtu:
8261 return &bpf_xdp_check_mtu_proto;
8262 #ifdef CONFIG_INET
8263 case BPF_FUNC_sk_lookup_udp:
8264 return &bpf_xdp_sk_lookup_udp_proto;
8265 case BPF_FUNC_sk_lookup_tcp:
8266 return &bpf_xdp_sk_lookup_tcp_proto;
8267 case BPF_FUNC_sk_release:
8268 return &bpf_sk_release_proto;
8269 case BPF_FUNC_skc_lookup_tcp:
8270 return &bpf_xdp_skc_lookup_tcp_proto;
8271 case BPF_FUNC_tcp_check_syncookie:
8272 return &bpf_tcp_check_syncookie_proto;
8273 case BPF_FUNC_tcp_gen_syncookie:
8274 return &bpf_tcp_gen_syncookie_proto;
8275 #ifdef CONFIG_SYN_COOKIES
8276 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8277 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8278 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8279 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8280 case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8281 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8282 case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8283 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8284 #endif
8285 #endif
8286 default:
8287 return bpf_sk_base_func_proto(func_id, prog);
8288 }
8289
8290 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8291 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8292 * kfuncs are defined in two different modules, and we want to be able
8293 * to use them interchangeably with the same BTF type ID. Because modules
8294 * can't de-duplicate BTF IDs between each other, we need the type to be
8295 * referenced in the vmlinux BTF or the verifier will get confused about
8296 * the different types. So we add this dummy type reference which will
8297 * be included in vmlinux BTF, allowing both modules to refer to the
8298 * same type ID.
8299 */
8300 BTF_TYPE_EMIT(struct nf_conn___init);
8301 #endif
8302 }
8303
8304 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8305 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8306
8307 static const struct bpf_func_proto *
sock_ops_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8308 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8309 {
8310 const struct bpf_func_proto *func_proto;
8311
8312 func_proto = cgroup_common_func_proto(func_id, prog);
8313 if (func_proto)
8314 return func_proto;
8315
8316 switch (func_id) {
8317 case BPF_FUNC_setsockopt:
8318 return &bpf_sock_ops_setsockopt_proto;
8319 case BPF_FUNC_getsockopt:
8320 return &bpf_sock_ops_getsockopt_proto;
8321 case BPF_FUNC_sock_ops_cb_flags_set:
8322 return &bpf_sock_ops_cb_flags_set_proto;
8323 case BPF_FUNC_sock_map_update:
8324 return &bpf_sock_map_update_proto;
8325 case BPF_FUNC_sock_hash_update:
8326 return &bpf_sock_hash_update_proto;
8327 case BPF_FUNC_get_socket_cookie:
8328 return &bpf_get_socket_cookie_sock_ops_proto;
8329 case BPF_FUNC_perf_event_output:
8330 return &bpf_event_output_data_proto;
8331 case BPF_FUNC_sk_storage_get:
8332 return &bpf_sk_storage_get_proto;
8333 case BPF_FUNC_sk_storage_delete:
8334 return &bpf_sk_storage_delete_proto;
8335 case BPF_FUNC_get_netns_cookie:
8336 return &bpf_get_netns_cookie_sock_ops_proto;
8337 #ifdef CONFIG_INET
8338 case BPF_FUNC_load_hdr_opt:
8339 return &bpf_sock_ops_load_hdr_opt_proto;
8340 case BPF_FUNC_store_hdr_opt:
8341 return &bpf_sock_ops_store_hdr_opt_proto;
8342 case BPF_FUNC_reserve_hdr_opt:
8343 return &bpf_sock_ops_reserve_hdr_opt_proto;
8344 case BPF_FUNC_tcp_sock:
8345 return &bpf_tcp_sock_proto;
8346 #endif /* CONFIG_INET */
8347 default:
8348 return bpf_sk_base_func_proto(func_id, prog);
8349 }
8350 }
8351
8352 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8353 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8354
8355 static const struct bpf_func_proto *
sk_msg_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8356 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8357 {
8358 switch (func_id) {
8359 case BPF_FUNC_msg_redirect_map:
8360 return &bpf_msg_redirect_map_proto;
8361 case BPF_FUNC_msg_redirect_hash:
8362 return &bpf_msg_redirect_hash_proto;
8363 case BPF_FUNC_msg_apply_bytes:
8364 return &bpf_msg_apply_bytes_proto;
8365 case BPF_FUNC_msg_cork_bytes:
8366 return &bpf_msg_cork_bytes_proto;
8367 case BPF_FUNC_msg_pull_data:
8368 return &bpf_msg_pull_data_proto;
8369 case BPF_FUNC_msg_push_data:
8370 return &bpf_msg_push_data_proto;
8371 case BPF_FUNC_msg_pop_data:
8372 return &bpf_msg_pop_data_proto;
8373 case BPF_FUNC_perf_event_output:
8374 return &bpf_event_output_data_proto;
8375 case BPF_FUNC_get_current_uid_gid:
8376 return &bpf_get_current_uid_gid_proto;
8377 case BPF_FUNC_sk_storage_get:
8378 return &bpf_sk_storage_get_proto;
8379 case BPF_FUNC_sk_storage_delete:
8380 return &bpf_sk_storage_delete_proto;
8381 case BPF_FUNC_get_netns_cookie:
8382 return &bpf_get_netns_cookie_sk_msg_proto;
8383 #ifdef CONFIG_CGROUP_NET_CLASSID
8384 case BPF_FUNC_get_cgroup_classid:
8385 return &bpf_get_cgroup_classid_curr_proto;
8386 #endif
8387 default:
8388 return bpf_sk_base_func_proto(func_id, prog);
8389 }
8390 }
8391
8392 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8393 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8394
8395 static const struct bpf_func_proto *
sk_skb_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8396 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8397 {
8398 switch (func_id) {
8399 case BPF_FUNC_skb_store_bytes:
8400 return &bpf_skb_store_bytes_proto;
8401 case BPF_FUNC_skb_load_bytes:
8402 return &bpf_skb_load_bytes_proto;
8403 case BPF_FUNC_skb_pull_data:
8404 return &sk_skb_pull_data_proto;
8405 case BPF_FUNC_skb_change_tail:
8406 return &sk_skb_change_tail_proto;
8407 case BPF_FUNC_skb_change_head:
8408 return &sk_skb_change_head_proto;
8409 case BPF_FUNC_skb_adjust_room:
8410 return &sk_skb_adjust_room_proto;
8411 case BPF_FUNC_get_socket_cookie:
8412 return &bpf_get_socket_cookie_proto;
8413 case BPF_FUNC_get_socket_uid:
8414 return &bpf_get_socket_uid_proto;
8415 case BPF_FUNC_sk_redirect_map:
8416 return &bpf_sk_redirect_map_proto;
8417 case BPF_FUNC_sk_redirect_hash:
8418 return &bpf_sk_redirect_hash_proto;
8419 case BPF_FUNC_perf_event_output:
8420 return &bpf_skb_event_output_proto;
8421 #ifdef CONFIG_INET
8422 case BPF_FUNC_sk_lookup_tcp:
8423 return &bpf_sk_lookup_tcp_proto;
8424 case BPF_FUNC_sk_lookup_udp:
8425 return &bpf_sk_lookup_udp_proto;
8426 case BPF_FUNC_sk_release:
8427 return &bpf_sk_release_proto;
8428 case BPF_FUNC_skc_lookup_tcp:
8429 return &bpf_skc_lookup_tcp_proto;
8430 #endif
8431 default:
8432 return bpf_sk_base_func_proto(func_id, prog);
8433 }
8434 }
8435
8436 static const struct bpf_func_proto *
flow_dissector_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8437 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8438 {
8439 switch (func_id) {
8440 case BPF_FUNC_skb_load_bytes:
8441 return &bpf_flow_dissector_load_bytes_proto;
8442 default:
8443 return bpf_sk_base_func_proto(func_id, prog);
8444 }
8445 }
8446
8447 static const struct bpf_func_proto *
lwt_out_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8448 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8449 {
8450 switch (func_id) {
8451 case BPF_FUNC_skb_load_bytes:
8452 return &bpf_skb_load_bytes_proto;
8453 case BPF_FUNC_skb_pull_data:
8454 return &bpf_skb_pull_data_proto;
8455 case BPF_FUNC_csum_diff:
8456 return &bpf_csum_diff_proto;
8457 case BPF_FUNC_get_cgroup_classid:
8458 return &bpf_get_cgroup_classid_proto;
8459 case BPF_FUNC_get_route_realm:
8460 return &bpf_get_route_realm_proto;
8461 case BPF_FUNC_get_hash_recalc:
8462 return &bpf_get_hash_recalc_proto;
8463 case BPF_FUNC_perf_event_output:
8464 return &bpf_skb_event_output_proto;
8465 case BPF_FUNC_get_smp_processor_id:
8466 return &bpf_get_smp_processor_id_proto;
8467 case BPF_FUNC_skb_under_cgroup:
8468 return &bpf_skb_under_cgroup_proto;
8469 default:
8470 return bpf_sk_base_func_proto(func_id, prog);
8471 }
8472 }
8473
8474 static const struct bpf_func_proto *
lwt_in_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8475 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8476 {
8477 switch (func_id) {
8478 case BPF_FUNC_lwt_push_encap:
8479 return &bpf_lwt_in_push_encap_proto;
8480 default:
8481 return lwt_out_func_proto(func_id, prog);
8482 }
8483 }
8484
8485 static const struct bpf_func_proto *
lwt_xmit_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8486 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8487 {
8488 switch (func_id) {
8489 case BPF_FUNC_skb_get_tunnel_key:
8490 return &bpf_skb_get_tunnel_key_proto;
8491 case BPF_FUNC_skb_set_tunnel_key:
8492 return bpf_get_skb_set_tunnel_proto(func_id);
8493 case BPF_FUNC_skb_get_tunnel_opt:
8494 return &bpf_skb_get_tunnel_opt_proto;
8495 case BPF_FUNC_skb_set_tunnel_opt:
8496 return bpf_get_skb_set_tunnel_proto(func_id);
8497 case BPF_FUNC_redirect:
8498 return &bpf_redirect_proto;
8499 case BPF_FUNC_clone_redirect:
8500 return &bpf_clone_redirect_proto;
8501 case BPF_FUNC_skb_change_tail:
8502 return &bpf_skb_change_tail_proto;
8503 case BPF_FUNC_skb_change_head:
8504 return &bpf_skb_change_head_proto;
8505 case BPF_FUNC_skb_store_bytes:
8506 return &bpf_skb_store_bytes_proto;
8507 case BPF_FUNC_csum_update:
8508 return &bpf_csum_update_proto;
8509 case BPF_FUNC_csum_level:
8510 return &bpf_csum_level_proto;
8511 case BPF_FUNC_l3_csum_replace:
8512 return &bpf_l3_csum_replace_proto;
8513 case BPF_FUNC_l4_csum_replace:
8514 return &bpf_l4_csum_replace_proto;
8515 case BPF_FUNC_set_hash_invalid:
8516 return &bpf_set_hash_invalid_proto;
8517 case BPF_FUNC_lwt_push_encap:
8518 return &bpf_lwt_xmit_push_encap_proto;
8519 default:
8520 return lwt_out_func_proto(func_id, prog);
8521 }
8522 }
8523
8524 static const struct bpf_func_proto *
lwt_seg6local_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)8525 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8526 {
8527 switch (func_id) {
8528 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8529 case BPF_FUNC_lwt_seg6_store_bytes:
8530 return &bpf_lwt_seg6_store_bytes_proto;
8531 case BPF_FUNC_lwt_seg6_action:
8532 return &bpf_lwt_seg6_action_proto;
8533 case BPF_FUNC_lwt_seg6_adjust_srh:
8534 return &bpf_lwt_seg6_adjust_srh_proto;
8535 #endif
8536 default:
8537 return lwt_out_func_proto(func_id, prog);
8538 }
8539 }
8540
bpf_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8541 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8542 const struct bpf_prog *prog,
8543 struct bpf_insn_access_aux *info)
8544 {
8545 const int size_default = sizeof(__u32);
8546
8547 if (off < 0 || off >= sizeof(struct __sk_buff))
8548 return false;
8549
8550 /* The verifier guarantees that size > 0. */
8551 if (off % size != 0)
8552 return false;
8553
8554 switch (off) {
8555 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8556 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8557 return false;
8558 break;
8559 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8560 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8561 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8562 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8563 case bpf_ctx_range(struct __sk_buff, data):
8564 case bpf_ctx_range(struct __sk_buff, data_meta):
8565 case bpf_ctx_range(struct __sk_buff, data_end):
8566 if (size != size_default)
8567 return false;
8568 break;
8569 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8570 return false;
8571 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8572 if (type == BPF_WRITE || size != sizeof(__u64))
8573 return false;
8574 break;
8575 case bpf_ctx_range(struct __sk_buff, tstamp):
8576 if (size != sizeof(__u64))
8577 return false;
8578 break;
8579 case offsetof(struct __sk_buff, sk):
8580 if (type == BPF_WRITE || size != sizeof(__u64))
8581 return false;
8582 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8583 break;
8584 case offsetof(struct __sk_buff, tstamp_type):
8585 return false;
8586 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8587 /* Explicitly prohibit access to padding in __sk_buff. */
8588 return false;
8589 default:
8590 /* Only narrow read access allowed for now. */
8591 if (type == BPF_WRITE) {
8592 if (size != size_default)
8593 return false;
8594 } else {
8595 bpf_ctx_record_field_size(info, size_default);
8596 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8597 return false;
8598 }
8599 }
8600
8601 return true;
8602 }
8603
sk_filter_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8604 static bool sk_filter_is_valid_access(int off, int size,
8605 enum bpf_access_type type,
8606 const struct bpf_prog *prog,
8607 struct bpf_insn_access_aux *info)
8608 {
8609 switch (off) {
8610 case bpf_ctx_range(struct __sk_buff, tc_classid):
8611 case bpf_ctx_range(struct __sk_buff, data):
8612 case bpf_ctx_range(struct __sk_buff, data_meta):
8613 case bpf_ctx_range(struct __sk_buff, data_end):
8614 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8615 case bpf_ctx_range(struct __sk_buff, tstamp):
8616 case bpf_ctx_range(struct __sk_buff, wire_len):
8617 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8618 return false;
8619 }
8620
8621 if (type == BPF_WRITE) {
8622 switch (off) {
8623 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8624 break;
8625 default:
8626 return false;
8627 }
8628 }
8629
8630 return bpf_skb_is_valid_access(off, size, type, prog, info);
8631 }
8632
cg_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8633 static bool cg_skb_is_valid_access(int off, int size,
8634 enum bpf_access_type type,
8635 const struct bpf_prog *prog,
8636 struct bpf_insn_access_aux *info)
8637 {
8638 switch (off) {
8639 case bpf_ctx_range(struct __sk_buff, tc_classid):
8640 case bpf_ctx_range(struct __sk_buff, data_meta):
8641 case bpf_ctx_range(struct __sk_buff, wire_len):
8642 return false;
8643 case bpf_ctx_range(struct __sk_buff, data):
8644 case bpf_ctx_range(struct __sk_buff, data_end):
8645 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
8646 return false;
8647 break;
8648 }
8649
8650 if (type == BPF_WRITE) {
8651 switch (off) {
8652 case bpf_ctx_range(struct __sk_buff, mark):
8653 case bpf_ctx_range(struct __sk_buff, priority):
8654 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8655 break;
8656 case bpf_ctx_range(struct __sk_buff, tstamp):
8657 if (!bpf_token_capable(prog->aux->token, CAP_BPF))
8658 return false;
8659 break;
8660 default:
8661 return false;
8662 }
8663 }
8664
8665 switch (off) {
8666 case bpf_ctx_range(struct __sk_buff, data):
8667 info->reg_type = PTR_TO_PACKET;
8668 break;
8669 case bpf_ctx_range(struct __sk_buff, data_end):
8670 info->reg_type = PTR_TO_PACKET_END;
8671 break;
8672 }
8673
8674 return bpf_skb_is_valid_access(off, size, type, prog, info);
8675 }
8676
lwt_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8677 static bool lwt_is_valid_access(int off, int size,
8678 enum bpf_access_type type,
8679 const struct bpf_prog *prog,
8680 struct bpf_insn_access_aux *info)
8681 {
8682 switch (off) {
8683 case bpf_ctx_range(struct __sk_buff, tc_classid):
8684 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8685 case bpf_ctx_range(struct __sk_buff, data_meta):
8686 case bpf_ctx_range(struct __sk_buff, tstamp):
8687 case bpf_ctx_range(struct __sk_buff, wire_len):
8688 case bpf_ctx_range(struct __sk_buff, hwtstamp):
8689 return false;
8690 }
8691
8692 if (type == BPF_WRITE) {
8693 switch (off) {
8694 case bpf_ctx_range(struct __sk_buff, mark):
8695 case bpf_ctx_range(struct __sk_buff, priority):
8696 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8697 break;
8698 default:
8699 return false;
8700 }
8701 }
8702
8703 switch (off) {
8704 case bpf_ctx_range(struct __sk_buff, data):
8705 info->reg_type = PTR_TO_PACKET;
8706 break;
8707 case bpf_ctx_range(struct __sk_buff, data_end):
8708 info->reg_type = PTR_TO_PACKET_END;
8709 break;
8710 }
8711
8712 return bpf_skb_is_valid_access(off, size, type, prog, info);
8713 }
8714
8715 /* Attach type specific accesses */
__sock_filter_check_attach_type(int off,enum bpf_access_type access_type,enum bpf_attach_type attach_type)8716 static bool __sock_filter_check_attach_type(int off,
8717 enum bpf_access_type access_type,
8718 enum bpf_attach_type attach_type)
8719 {
8720 switch (off) {
8721 case offsetof(struct bpf_sock, bound_dev_if):
8722 case offsetof(struct bpf_sock, mark):
8723 case offsetof(struct bpf_sock, priority):
8724 switch (attach_type) {
8725 case BPF_CGROUP_INET_SOCK_CREATE:
8726 case BPF_CGROUP_INET_SOCK_RELEASE:
8727 goto full_access;
8728 default:
8729 return false;
8730 }
8731 case bpf_ctx_range(struct bpf_sock, src_ip4):
8732 switch (attach_type) {
8733 case BPF_CGROUP_INET4_POST_BIND:
8734 goto read_only;
8735 default:
8736 return false;
8737 }
8738 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8739 switch (attach_type) {
8740 case BPF_CGROUP_INET6_POST_BIND:
8741 goto read_only;
8742 default:
8743 return false;
8744 }
8745 case bpf_ctx_range(struct bpf_sock, src_port):
8746 switch (attach_type) {
8747 case BPF_CGROUP_INET4_POST_BIND:
8748 case BPF_CGROUP_INET6_POST_BIND:
8749 goto read_only;
8750 default:
8751 return false;
8752 }
8753 }
8754 read_only:
8755 return access_type == BPF_READ;
8756 full_access:
8757 return true;
8758 }
8759
bpf_sock_common_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8760 bool bpf_sock_common_is_valid_access(int off, int size,
8761 enum bpf_access_type type,
8762 struct bpf_insn_access_aux *info)
8763 {
8764 switch (off) {
8765 case bpf_ctx_range_till(struct bpf_sock, type, priority):
8766 return false;
8767 default:
8768 return bpf_sock_is_valid_access(off, size, type, info);
8769 }
8770 }
8771
bpf_sock_is_valid_access(int off,int size,enum bpf_access_type type,struct bpf_insn_access_aux * info)8772 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8773 struct bpf_insn_access_aux *info)
8774 {
8775 const int size_default = sizeof(__u32);
8776 int field_size;
8777
8778 if (off < 0 || off >= sizeof(struct bpf_sock))
8779 return false;
8780 if (off % size != 0)
8781 return false;
8782
8783 switch (off) {
8784 case offsetof(struct bpf_sock, state):
8785 case offsetof(struct bpf_sock, family):
8786 case offsetof(struct bpf_sock, type):
8787 case offsetof(struct bpf_sock, protocol):
8788 case offsetof(struct bpf_sock, src_port):
8789 case offsetof(struct bpf_sock, rx_queue_mapping):
8790 case bpf_ctx_range(struct bpf_sock, src_ip4):
8791 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8792 case bpf_ctx_range(struct bpf_sock, dst_ip4):
8793 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8794 bpf_ctx_record_field_size(info, size_default);
8795 return bpf_ctx_narrow_access_ok(off, size, size_default);
8796 case bpf_ctx_range(struct bpf_sock, dst_port):
8797 field_size = size == size_default ?
8798 size_default : sizeof_field(struct bpf_sock, dst_port);
8799 bpf_ctx_record_field_size(info, field_size);
8800 return bpf_ctx_narrow_access_ok(off, size, field_size);
8801 case offsetofend(struct bpf_sock, dst_port) ...
8802 offsetof(struct bpf_sock, dst_ip4) - 1:
8803 return false;
8804 }
8805
8806 return size == size_default;
8807 }
8808
sock_filter_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8809 static bool sock_filter_is_valid_access(int off, int size,
8810 enum bpf_access_type type,
8811 const struct bpf_prog *prog,
8812 struct bpf_insn_access_aux *info)
8813 {
8814 if (!bpf_sock_is_valid_access(off, size, type, info))
8815 return false;
8816 return __sock_filter_check_attach_type(off, type,
8817 prog->expected_attach_type);
8818 }
8819
bpf_noop_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8820 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8821 const struct bpf_prog *prog)
8822 {
8823 /* Neither direct read nor direct write requires any preliminary
8824 * action.
8825 */
8826 return 0;
8827 }
8828
bpf_unclone_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog,int drop_verdict)8829 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8830 const struct bpf_prog *prog, int drop_verdict)
8831 {
8832 struct bpf_insn *insn = insn_buf;
8833
8834 if (!direct_write)
8835 return 0;
8836
8837 /* if (!skb->cloned)
8838 * goto start;
8839 *
8840 * (Fast-path, otherwise approximation that we might be
8841 * a clone, do the rest in helper.)
8842 */
8843 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8844 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8845 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8846
8847 /* ret = bpf_skb_pull_data(skb, 0); */
8848 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8849 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8850 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8851 BPF_FUNC_skb_pull_data);
8852 /* if (!ret)
8853 * goto restore;
8854 * return TC_ACT_SHOT;
8855 */
8856 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8857 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8858 *insn++ = BPF_EXIT_INSN();
8859
8860 /* restore: */
8861 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8862 /* start: */
8863 *insn++ = prog->insnsi[0];
8864
8865 return insn - insn_buf;
8866 }
8867
bpf_gen_ld_abs(const struct bpf_insn * orig,struct bpf_insn * insn_buf)8868 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8869 struct bpf_insn *insn_buf)
8870 {
8871 bool indirect = BPF_MODE(orig->code) == BPF_IND;
8872 struct bpf_insn *insn = insn_buf;
8873
8874 if (!indirect) {
8875 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8876 } else {
8877 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8878 if (orig->imm)
8879 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8880 }
8881 /* We're guaranteed here that CTX is in R6. */
8882 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8883
8884 switch (BPF_SIZE(orig->code)) {
8885 case BPF_B:
8886 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8887 break;
8888 case BPF_H:
8889 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8890 break;
8891 case BPF_W:
8892 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8893 break;
8894 }
8895
8896 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8897 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8898 *insn++ = BPF_EXIT_INSN();
8899
8900 return insn - insn_buf;
8901 }
8902
tc_cls_act_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)8903 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8904 const struct bpf_prog *prog)
8905 {
8906 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8907 }
8908
tc_cls_act_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8909 static bool tc_cls_act_is_valid_access(int off, int size,
8910 enum bpf_access_type type,
8911 const struct bpf_prog *prog,
8912 struct bpf_insn_access_aux *info)
8913 {
8914 if (type == BPF_WRITE) {
8915 switch (off) {
8916 case bpf_ctx_range(struct __sk_buff, mark):
8917 case bpf_ctx_range(struct __sk_buff, tc_index):
8918 case bpf_ctx_range(struct __sk_buff, priority):
8919 case bpf_ctx_range(struct __sk_buff, tc_classid):
8920 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8921 case bpf_ctx_range(struct __sk_buff, tstamp):
8922 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8923 break;
8924 default:
8925 return false;
8926 }
8927 }
8928
8929 switch (off) {
8930 case bpf_ctx_range(struct __sk_buff, data):
8931 info->reg_type = PTR_TO_PACKET;
8932 break;
8933 case bpf_ctx_range(struct __sk_buff, data_meta):
8934 info->reg_type = PTR_TO_PACKET_META;
8935 break;
8936 case bpf_ctx_range(struct __sk_buff, data_end):
8937 info->reg_type = PTR_TO_PACKET_END;
8938 break;
8939 case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8940 return false;
8941 case offsetof(struct __sk_buff, tstamp_type):
8942 /* The convert_ctx_access() on reading and writing
8943 * __sk_buff->tstamp depends on whether the bpf prog
8944 * has used __sk_buff->tstamp_type or not.
8945 * Thus, we need to set prog->tstamp_type_access
8946 * earlier during is_valid_access() here.
8947 */
8948 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8949 return size == sizeof(__u8);
8950 }
8951
8952 return bpf_skb_is_valid_access(off, size, type, prog, info);
8953 }
8954
8955 DEFINE_MUTEX(nf_conn_btf_access_lock);
8956 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8957
8958 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8959 const struct bpf_reg_state *reg,
8960 int off, int size);
8961 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8962
tc_cls_act_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)8963 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8964 const struct bpf_reg_state *reg,
8965 int off, int size)
8966 {
8967 int ret = -EACCES;
8968
8969 mutex_lock(&nf_conn_btf_access_lock);
8970 if (nfct_btf_struct_access)
8971 ret = nfct_btf_struct_access(log, reg, off, size);
8972 mutex_unlock(&nf_conn_btf_access_lock);
8973
8974 return ret;
8975 }
8976
__is_valid_xdp_access(int off,int size)8977 static bool __is_valid_xdp_access(int off, int size)
8978 {
8979 if (off < 0 || off >= sizeof(struct xdp_md))
8980 return false;
8981 if (off % size != 0)
8982 return false;
8983 if (size != sizeof(__u32))
8984 return false;
8985
8986 return true;
8987 }
8988
xdp_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)8989 static bool xdp_is_valid_access(int off, int size,
8990 enum bpf_access_type type,
8991 const struct bpf_prog *prog,
8992 struct bpf_insn_access_aux *info)
8993 {
8994 if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8995 switch (off) {
8996 case offsetof(struct xdp_md, egress_ifindex):
8997 return false;
8998 }
8999 }
9000
9001 if (type == BPF_WRITE) {
9002 if (bpf_prog_is_offloaded(prog->aux)) {
9003 switch (off) {
9004 case offsetof(struct xdp_md, rx_queue_index):
9005 return __is_valid_xdp_access(off, size);
9006 }
9007 }
9008 return false;
9009 }
9010
9011 switch (off) {
9012 case offsetof(struct xdp_md, data):
9013 info->reg_type = PTR_TO_PACKET;
9014 break;
9015 case offsetof(struct xdp_md, data_meta):
9016 info->reg_type = PTR_TO_PACKET_META;
9017 break;
9018 case offsetof(struct xdp_md, data_end):
9019 info->reg_type = PTR_TO_PACKET_END;
9020 break;
9021 }
9022
9023 return __is_valid_xdp_access(off, size);
9024 }
9025
bpf_warn_invalid_xdp_action(struct net_device * dev,struct bpf_prog * prog,u32 act)9026 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
9027 {
9028 const u32 act_max = XDP_REDIRECT;
9029
9030 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
9031 act > act_max ? "Illegal" : "Driver unsupported",
9032 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
9033 }
9034 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
9035
xdp_btf_struct_access(struct bpf_verifier_log * log,const struct bpf_reg_state * reg,int off,int size)9036 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
9037 const struct bpf_reg_state *reg,
9038 int off, int size)
9039 {
9040 int ret = -EACCES;
9041
9042 mutex_lock(&nf_conn_btf_access_lock);
9043 if (nfct_btf_struct_access)
9044 ret = nfct_btf_struct_access(log, reg, off, size);
9045 mutex_unlock(&nf_conn_btf_access_lock);
9046
9047 return ret;
9048 }
9049
sock_addr_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9050 static bool sock_addr_is_valid_access(int off, int size,
9051 enum bpf_access_type type,
9052 const struct bpf_prog *prog,
9053 struct bpf_insn_access_aux *info)
9054 {
9055 const int size_default = sizeof(__u32);
9056
9057 if (off < 0 || off >= sizeof(struct bpf_sock_addr))
9058 return false;
9059 if (off % size != 0)
9060 return false;
9061
9062 /* Disallow access to fields not belonging to the attach type's address
9063 * family.
9064 */
9065 switch (off) {
9066 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9067 switch (prog->expected_attach_type) {
9068 case BPF_CGROUP_INET4_BIND:
9069 case BPF_CGROUP_INET4_CONNECT:
9070 case BPF_CGROUP_INET4_GETPEERNAME:
9071 case BPF_CGROUP_INET4_GETSOCKNAME:
9072 case BPF_CGROUP_UDP4_SENDMSG:
9073 case BPF_CGROUP_UDP4_RECVMSG:
9074 break;
9075 default:
9076 return false;
9077 }
9078 break;
9079 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9080 switch (prog->expected_attach_type) {
9081 case BPF_CGROUP_INET6_BIND:
9082 case BPF_CGROUP_INET6_CONNECT:
9083 case BPF_CGROUP_INET6_GETPEERNAME:
9084 case BPF_CGROUP_INET6_GETSOCKNAME:
9085 case BPF_CGROUP_UDP6_SENDMSG:
9086 case BPF_CGROUP_UDP6_RECVMSG:
9087 break;
9088 default:
9089 return false;
9090 }
9091 break;
9092 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9093 switch (prog->expected_attach_type) {
9094 case BPF_CGROUP_UDP4_SENDMSG:
9095 break;
9096 default:
9097 return false;
9098 }
9099 break;
9100 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9101 msg_src_ip6[3]):
9102 switch (prog->expected_attach_type) {
9103 case BPF_CGROUP_UDP6_SENDMSG:
9104 break;
9105 default:
9106 return false;
9107 }
9108 break;
9109 }
9110
9111 switch (off) {
9112 case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9113 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9114 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9115 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9116 msg_src_ip6[3]):
9117 case bpf_ctx_range(struct bpf_sock_addr, user_port):
9118 if (type == BPF_READ) {
9119 bpf_ctx_record_field_size(info, size_default);
9120
9121 if (bpf_ctx_wide_access_ok(off, size,
9122 struct bpf_sock_addr,
9123 user_ip6))
9124 return true;
9125
9126 if (bpf_ctx_wide_access_ok(off, size,
9127 struct bpf_sock_addr,
9128 msg_src_ip6))
9129 return true;
9130
9131 if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9132 return false;
9133 } else {
9134 if (bpf_ctx_wide_access_ok(off, size,
9135 struct bpf_sock_addr,
9136 user_ip6))
9137 return true;
9138
9139 if (bpf_ctx_wide_access_ok(off, size,
9140 struct bpf_sock_addr,
9141 msg_src_ip6))
9142 return true;
9143
9144 if (size != size_default)
9145 return false;
9146 }
9147 break;
9148 case offsetof(struct bpf_sock_addr, sk):
9149 if (type != BPF_READ)
9150 return false;
9151 if (size != sizeof(__u64))
9152 return false;
9153 info->reg_type = PTR_TO_SOCKET;
9154 break;
9155 default:
9156 if (type == BPF_READ) {
9157 if (size != size_default)
9158 return false;
9159 } else {
9160 return false;
9161 }
9162 }
9163
9164 return true;
9165 }
9166
sock_ops_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9167 static bool sock_ops_is_valid_access(int off, int size,
9168 enum bpf_access_type type,
9169 const struct bpf_prog *prog,
9170 struct bpf_insn_access_aux *info)
9171 {
9172 const int size_default = sizeof(__u32);
9173
9174 if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9175 return false;
9176
9177 /* The verifier guarantees that size > 0. */
9178 if (off % size != 0)
9179 return false;
9180
9181 if (type == BPF_WRITE) {
9182 switch (off) {
9183 case offsetof(struct bpf_sock_ops, reply):
9184 case offsetof(struct bpf_sock_ops, sk_txhash):
9185 if (size != size_default)
9186 return false;
9187 break;
9188 default:
9189 return false;
9190 }
9191 } else {
9192 switch (off) {
9193 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9194 bytes_acked):
9195 if (size != sizeof(__u64))
9196 return false;
9197 break;
9198 case offsetof(struct bpf_sock_ops, sk):
9199 if (size != sizeof(__u64))
9200 return false;
9201 info->reg_type = PTR_TO_SOCKET_OR_NULL;
9202 break;
9203 case offsetof(struct bpf_sock_ops, skb_data):
9204 if (size != sizeof(__u64))
9205 return false;
9206 info->reg_type = PTR_TO_PACKET;
9207 break;
9208 case offsetof(struct bpf_sock_ops, skb_data_end):
9209 if (size != sizeof(__u64))
9210 return false;
9211 info->reg_type = PTR_TO_PACKET_END;
9212 break;
9213 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9214 bpf_ctx_record_field_size(info, size_default);
9215 return bpf_ctx_narrow_access_ok(off, size,
9216 size_default);
9217 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9218 if (size != sizeof(__u64))
9219 return false;
9220 break;
9221 default:
9222 if (size != size_default)
9223 return false;
9224 break;
9225 }
9226 }
9227
9228 return true;
9229 }
9230
sk_skb_prologue(struct bpf_insn * insn_buf,bool direct_write,const struct bpf_prog * prog)9231 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9232 const struct bpf_prog *prog)
9233 {
9234 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9235 }
9236
sk_skb_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9237 static bool sk_skb_is_valid_access(int off, int size,
9238 enum bpf_access_type type,
9239 const struct bpf_prog *prog,
9240 struct bpf_insn_access_aux *info)
9241 {
9242 switch (off) {
9243 case bpf_ctx_range(struct __sk_buff, tc_classid):
9244 case bpf_ctx_range(struct __sk_buff, data_meta):
9245 case bpf_ctx_range(struct __sk_buff, tstamp):
9246 case bpf_ctx_range(struct __sk_buff, wire_len):
9247 case bpf_ctx_range(struct __sk_buff, hwtstamp):
9248 return false;
9249 }
9250
9251 if (type == BPF_WRITE) {
9252 switch (off) {
9253 case bpf_ctx_range(struct __sk_buff, tc_index):
9254 case bpf_ctx_range(struct __sk_buff, priority):
9255 break;
9256 default:
9257 return false;
9258 }
9259 }
9260
9261 switch (off) {
9262 case bpf_ctx_range(struct __sk_buff, mark):
9263 return false;
9264 case bpf_ctx_range(struct __sk_buff, data):
9265 info->reg_type = PTR_TO_PACKET;
9266 break;
9267 case bpf_ctx_range(struct __sk_buff, data_end):
9268 info->reg_type = PTR_TO_PACKET_END;
9269 break;
9270 }
9271
9272 return bpf_skb_is_valid_access(off, size, type, prog, info);
9273 }
9274
sk_msg_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9275 static bool sk_msg_is_valid_access(int off, int size,
9276 enum bpf_access_type type,
9277 const struct bpf_prog *prog,
9278 struct bpf_insn_access_aux *info)
9279 {
9280 if (type == BPF_WRITE)
9281 return false;
9282
9283 if (off % size != 0)
9284 return false;
9285
9286 switch (off) {
9287 case offsetof(struct sk_msg_md, data):
9288 info->reg_type = PTR_TO_PACKET;
9289 if (size != sizeof(__u64))
9290 return false;
9291 break;
9292 case offsetof(struct sk_msg_md, data_end):
9293 info->reg_type = PTR_TO_PACKET_END;
9294 if (size != sizeof(__u64))
9295 return false;
9296 break;
9297 case offsetof(struct sk_msg_md, sk):
9298 if (size != sizeof(__u64))
9299 return false;
9300 info->reg_type = PTR_TO_SOCKET;
9301 break;
9302 case bpf_ctx_range(struct sk_msg_md, family):
9303 case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9304 case bpf_ctx_range(struct sk_msg_md, local_ip4):
9305 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9306 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9307 case bpf_ctx_range(struct sk_msg_md, remote_port):
9308 case bpf_ctx_range(struct sk_msg_md, local_port):
9309 case bpf_ctx_range(struct sk_msg_md, size):
9310 if (size != sizeof(__u32))
9311 return false;
9312 break;
9313 default:
9314 return false;
9315 }
9316 return true;
9317 }
9318
flow_dissector_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)9319 static bool flow_dissector_is_valid_access(int off, int size,
9320 enum bpf_access_type type,
9321 const struct bpf_prog *prog,
9322 struct bpf_insn_access_aux *info)
9323 {
9324 const int size_default = sizeof(__u32);
9325
9326 if (off < 0 || off >= sizeof(struct __sk_buff))
9327 return false;
9328
9329 if (type == BPF_WRITE)
9330 return false;
9331
9332 switch (off) {
9333 case bpf_ctx_range(struct __sk_buff, data):
9334 if (size != size_default)
9335 return false;
9336 info->reg_type = PTR_TO_PACKET;
9337 return true;
9338 case bpf_ctx_range(struct __sk_buff, data_end):
9339 if (size != size_default)
9340 return false;
9341 info->reg_type = PTR_TO_PACKET_END;
9342 return true;
9343 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9344 if (size != sizeof(__u64))
9345 return false;
9346 info->reg_type = PTR_TO_FLOW_KEYS;
9347 return true;
9348 default:
9349 return false;
9350 }
9351 }
9352
flow_dissector_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9353 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9354 const struct bpf_insn *si,
9355 struct bpf_insn *insn_buf,
9356 struct bpf_prog *prog,
9357 u32 *target_size)
9358
9359 {
9360 struct bpf_insn *insn = insn_buf;
9361
9362 switch (si->off) {
9363 case offsetof(struct __sk_buff, data):
9364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9365 si->dst_reg, si->src_reg,
9366 offsetof(struct bpf_flow_dissector, data));
9367 break;
9368
9369 case offsetof(struct __sk_buff, data_end):
9370 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9371 si->dst_reg, si->src_reg,
9372 offsetof(struct bpf_flow_dissector, data_end));
9373 break;
9374
9375 case offsetof(struct __sk_buff, flow_keys):
9376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9377 si->dst_reg, si->src_reg,
9378 offsetof(struct bpf_flow_dissector, flow_keys));
9379 break;
9380 }
9381
9382 return insn - insn_buf;
9383 }
9384
bpf_convert_tstamp_type_read(const struct bpf_insn * si,struct bpf_insn * insn)9385 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9386 struct bpf_insn *insn)
9387 {
9388 __u8 value_reg = si->dst_reg;
9389 __u8 skb_reg = si->src_reg;
9390 /* AX is needed because src_reg and dst_reg could be the same */
9391 __u8 tmp_reg = BPF_REG_AX;
9392
9393 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9394 SKB_BF_MONO_TC_OFFSET);
9395 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9396 SKB_MONO_DELIVERY_TIME_MASK, 2);
9397 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9398 *insn++ = BPF_JMP_A(1);
9399 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9400
9401 return insn;
9402 }
9403
bpf_convert_shinfo_access(__u8 dst_reg,__u8 skb_reg,struct bpf_insn * insn)9404 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9405 struct bpf_insn *insn)
9406 {
9407 /* si->dst_reg = skb_shinfo(SKB); */
9408 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9409 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9410 BPF_REG_AX, skb_reg,
9411 offsetof(struct sk_buff, end));
9412 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9413 dst_reg, skb_reg,
9414 offsetof(struct sk_buff, head));
9415 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9416 #else
9417 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9418 dst_reg, skb_reg,
9419 offsetof(struct sk_buff, end));
9420 #endif
9421
9422 return insn;
9423 }
9424
bpf_convert_tstamp_read(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9425 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9426 const struct bpf_insn *si,
9427 struct bpf_insn *insn)
9428 {
9429 __u8 value_reg = si->dst_reg;
9430 __u8 skb_reg = si->src_reg;
9431
9432 #ifdef CONFIG_NET_XGRESS
9433 /* If the tstamp_type is read,
9434 * the bpf prog is aware the tstamp could have delivery time.
9435 * Thus, read skb->tstamp as is if tstamp_type_access is true.
9436 */
9437 if (!prog->tstamp_type_access) {
9438 /* AX is needed because src_reg and dst_reg could be the same */
9439 __u8 tmp_reg = BPF_REG_AX;
9440
9441 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9442 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9443 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9444 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9445 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9446 /* skb->tc_at_ingress && skb->mono_delivery_time,
9447 * read 0 as the (rcv) timestamp.
9448 */
9449 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9450 *insn++ = BPF_JMP_A(1);
9451 }
9452 #endif
9453
9454 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9455 offsetof(struct sk_buff, tstamp));
9456 return insn;
9457 }
9458
bpf_convert_tstamp_write(const struct bpf_prog * prog,const struct bpf_insn * si,struct bpf_insn * insn)9459 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9460 const struct bpf_insn *si,
9461 struct bpf_insn *insn)
9462 {
9463 __u8 value_reg = si->src_reg;
9464 __u8 skb_reg = si->dst_reg;
9465
9466 #ifdef CONFIG_NET_XGRESS
9467 /* If the tstamp_type is read,
9468 * the bpf prog is aware the tstamp could have delivery time.
9469 * Thus, write skb->tstamp as is if tstamp_type_access is true.
9470 * Otherwise, writing at ingress will have to clear the
9471 * mono_delivery_time bit also.
9472 */
9473 if (!prog->tstamp_type_access) {
9474 __u8 tmp_reg = BPF_REG_AX;
9475
9476 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9477 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9478 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9479 /* goto <store> */
9480 *insn++ = BPF_JMP_A(2);
9481 /* <clear>: mono_delivery_time */
9482 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9483 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9484 }
9485 #endif
9486
9487 /* <store>: skb->tstamp = tstamp */
9488 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9489 skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9490 return insn;
9491 }
9492
9493 #define BPF_EMIT_STORE(size, si, off) \
9494 BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM, \
9495 (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9496
bpf_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9497 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9498 const struct bpf_insn *si,
9499 struct bpf_insn *insn_buf,
9500 struct bpf_prog *prog, u32 *target_size)
9501 {
9502 struct bpf_insn *insn = insn_buf;
9503 int off;
9504
9505 switch (si->off) {
9506 case offsetof(struct __sk_buff, len):
9507 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9508 bpf_target_off(struct sk_buff, len, 4,
9509 target_size));
9510 break;
9511
9512 case offsetof(struct __sk_buff, protocol):
9513 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9514 bpf_target_off(struct sk_buff, protocol, 2,
9515 target_size));
9516 break;
9517
9518 case offsetof(struct __sk_buff, vlan_proto):
9519 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9520 bpf_target_off(struct sk_buff, vlan_proto, 2,
9521 target_size));
9522 break;
9523
9524 case offsetof(struct __sk_buff, priority):
9525 if (type == BPF_WRITE)
9526 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9527 bpf_target_off(struct sk_buff, priority, 4,
9528 target_size));
9529 else
9530 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9531 bpf_target_off(struct sk_buff, priority, 4,
9532 target_size));
9533 break;
9534
9535 case offsetof(struct __sk_buff, ingress_ifindex):
9536 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9537 bpf_target_off(struct sk_buff, skb_iif, 4,
9538 target_size));
9539 break;
9540
9541 case offsetof(struct __sk_buff, ifindex):
9542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9543 si->dst_reg, si->src_reg,
9544 offsetof(struct sk_buff, dev));
9545 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9546 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9547 bpf_target_off(struct net_device, ifindex, 4,
9548 target_size));
9549 break;
9550
9551 case offsetof(struct __sk_buff, hash):
9552 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9553 bpf_target_off(struct sk_buff, hash, 4,
9554 target_size));
9555 break;
9556
9557 case offsetof(struct __sk_buff, mark):
9558 if (type == BPF_WRITE)
9559 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9560 bpf_target_off(struct sk_buff, mark, 4,
9561 target_size));
9562 else
9563 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9564 bpf_target_off(struct sk_buff, mark, 4,
9565 target_size));
9566 break;
9567
9568 case offsetof(struct __sk_buff, pkt_type):
9569 *target_size = 1;
9570 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9571 PKT_TYPE_OFFSET);
9572 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9573 #ifdef __BIG_ENDIAN_BITFIELD
9574 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9575 #endif
9576 break;
9577
9578 case offsetof(struct __sk_buff, queue_mapping):
9579 if (type == BPF_WRITE) {
9580 u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9581
9582 if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9583 *insn++ = BPF_JMP_A(0); /* noop */
9584 break;
9585 }
9586
9587 if (BPF_CLASS(si->code) == BPF_STX)
9588 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9589 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9590 } else {
9591 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9592 bpf_target_off(struct sk_buff,
9593 queue_mapping,
9594 2, target_size));
9595 }
9596 break;
9597
9598 case offsetof(struct __sk_buff, vlan_present):
9599 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9600 bpf_target_off(struct sk_buff,
9601 vlan_all, 4, target_size));
9602 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9603 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9604 break;
9605
9606 case offsetof(struct __sk_buff, vlan_tci):
9607 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9608 bpf_target_off(struct sk_buff, vlan_tci, 2,
9609 target_size));
9610 break;
9611
9612 case offsetof(struct __sk_buff, cb[0]) ...
9613 offsetofend(struct __sk_buff, cb[4]) - 1:
9614 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9615 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9616 offsetof(struct qdisc_skb_cb, data)) %
9617 sizeof(__u64));
9618
9619 prog->cb_access = 1;
9620 off = si->off;
9621 off -= offsetof(struct __sk_buff, cb[0]);
9622 off += offsetof(struct sk_buff, cb);
9623 off += offsetof(struct qdisc_skb_cb, data);
9624 if (type == BPF_WRITE)
9625 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9626 else
9627 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9628 si->src_reg, off);
9629 break;
9630
9631 case offsetof(struct __sk_buff, tc_classid):
9632 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9633
9634 off = si->off;
9635 off -= offsetof(struct __sk_buff, tc_classid);
9636 off += offsetof(struct sk_buff, cb);
9637 off += offsetof(struct qdisc_skb_cb, tc_classid);
9638 *target_size = 2;
9639 if (type == BPF_WRITE)
9640 *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9641 else
9642 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9643 si->src_reg, off);
9644 break;
9645
9646 case offsetof(struct __sk_buff, data):
9647 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9648 si->dst_reg, si->src_reg,
9649 offsetof(struct sk_buff, data));
9650 break;
9651
9652 case offsetof(struct __sk_buff, data_meta):
9653 off = si->off;
9654 off -= offsetof(struct __sk_buff, data_meta);
9655 off += offsetof(struct sk_buff, cb);
9656 off += offsetof(struct bpf_skb_data_end, data_meta);
9657 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9658 si->src_reg, off);
9659 break;
9660
9661 case offsetof(struct __sk_buff, data_end):
9662 off = si->off;
9663 off -= offsetof(struct __sk_buff, data_end);
9664 off += offsetof(struct sk_buff, cb);
9665 off += offsetof(struct bpf_skb_data_end, data_end);
9666 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9667 si->src_reg, off);
9668 break;
9669
9670 case offsetof(struct __sk_buff, tc_index):
9671 #ifdef CONFIG_NET_SCHED
9672 if (type == BPF_WRITE)
9673 *insn++ = BPF_EMIT_STORE(BPF_H, si,
9674 bpf_target_off(struct sk_buff, tc_index, 2,
9675 target_size));
9676 else
9677 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9678 bpf_target_off(struct sk_buff, tc_index, 2,
9679 target_size));
9680 #else
9681 *target_size = 2;
9682 if (type == BPF_WRITE)
9683 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9684 else
9685 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9686 #endif
9687 break;
9688
9689 case offsetof(struct __sk_buff, napi_id):
9690 #if defined(CONFIG_NET_RX_BUSY_POLL)
9691 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9692 bpf_target_off(struct sk_buff, napi_id, 4,
9693 target_size));
9694 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9695 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9696 #else
9697 *target_size = 4;
9698 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9699 #endif
9700 break;
9701 case offsetof(struct __sk_buff, family):
9702 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9703
9704 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9705 si->dst_reg, si->src_reg,
9706 offsetof(struct sk_buff, sk));
9707 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9708 bpf_target_off(struct sock_common,
9709 skc_family,
9710 2, target_size));
9711 break;
9712 case offsetof(struct __sk_buff, remote_ip4):
9713 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9714
9715 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9716 si->dst_reg, si->src_reg,
9717 offsetof(struct sk_buff, sk));
9718 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9719 bpf_target_off(struct sock_common,
9720 skc_daddr,
9721 4, target_size));
9722 break;
9723 case offsetof(struct __sk_buff, local_ip4):
9724 BUILD_BUG_ON(sizeof_field(struct sock_common,
9725 skc_rcv_saddr) != 4);
9726
9727 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9728 si->dst_reg, si->src_reg,
9729 offsetof(struct sk_buff, sk));
9730 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9731 bpf_target_off(struct sock_common,
9732 skc_rcv_saddr,
9733 4, target_size));
9734 break;
9735 case offsetof(struct __sk_buff, remote_ip6[0]) ...
9736 offsetof(struct __sk_buff, remote_ip6[3]):
9737 #if IS_ENABLED(CONFIG_IPV6)
9738 BUILD_BUG_ON(sizeof_field(struct sock_common,
9739 skc_v6_daddr.s6_addr32[0]) != 4);
9740
9741 off = si->off;
9742 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9743
9744 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9745 si->dst_reg, si->src_reg,
9746 offsetof(struct sk_buff, sk));
9747 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9748 offsetof(struct sock_common,
9749 skc_v6_daddr.s6_addr32[0]) +
9750 off);
9751 #else
9752 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9753 #endif
9754 break;
9755 case offsetof(struct __sk_buff, local_ip6[0]) ...
9756 offsetof(struct __sk_buff, local_ip6[3]):
9757 #if IS_ENABLED(CONFIG_IPV6)
9758 BUILD_BUG_ON(sizeof_field(struct sock_common,
9759 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9760
9761 off = si->off;
9762 off -= offsetof(struct __sk_buff, local_ip6[0]);
9763
9764 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9765 si->dst_reg, si->src_reg,
9766 offsetof(struct sk_buff, sk));
9767 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9768 offsetof(struct sock_common,
9769 skc_v6_rcv_saddr.s6_addr32[0]) +
9770 off);
9771 #else
9772 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9773 #endif
9774 break;
9775
9776 case offsetof(struct __sk_buff, remote_port):
9777 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9778
9779 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9780 si->dst_reg, si->src_reg,
9781 offsetof(struct sk_buff, sk));
9782 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9783 bpf_target_off(struct sock_common,
9784 skc_dport,
9785 2, target_size));
9786 #ifndef __BIG_ENDIAN_BITFIELD
9787 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9788 #endif
9789 break;
9790
9791 case offsetof(struct __sk_buff, local_port):
9792 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9793
9794 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9795 si->dst_reg, si->src_reg,
9796 offsetof(struct sk_buff, sk));
9797 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9798 bpf_target_off(struct sock_common,
9799 skc_num, 2, target_size));
9800 break;
9801
9802 case offsetof(struct __sk_buff, tstamp):
9803 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9804
9805 if (type == BPF_WRITE)
9806 insn = bpf_convert_tstamp_write(prog, si, insn);
9807 else
9808 insn = bpf_convert_tstamp_read(prog, si, insn);
9809 break;
9810
9811 case offsetof(struct __sk_buff, tstamp_type):
9812 insn = bpf_convert_tstamp_type_read(si, insn);
9813 break;
9814
9815 case offsetof(struct __sk_buff, gso_segs):
9816 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9817 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9818 si->dst_reg, si->dst_reg,
9819 bpf_target_off(struct skb_shared_info,
9820 gso_segs, 2,
9821 target_size));
9822 break;
9823 case offsetof(struct __sk_buff, gso_size):
9824 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9825 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9826 si->dst_reg, si->dst_reg,
9827 bpf_target_off(struct skb_shared_info,
9828 gso_size, 2,
9829 target_size));
9830 break;
9831 case offsetof(struct __sk_buff, wire_len):
9832 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9833
9834 off = si->off;
9835 off -= offsetof(struct __sk_buff, wire_len);
9836 off += offsetof(struct sk_buff, cb);
9837 off += offsetof(struct qdisc_skb_cb, pkt_len);
9838 *target_size = 4;
9839 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9840 break;
9841
9842 case offsetof(struct __sk_buff, sk):
9843 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9844 si->dst_reg, si->src_reg,
9845 offsetof(struct sk_buff, sk));
9846 break;
9847 case offsetof(struct __sk_buff, hwtstamp):
9848 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9849 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9850
9851 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9852 *insn++ = BPF_LDX_MEM(BPF_DW,
9853 si->dst_reg, si->dst_reg,
9854 bpf_target_off(struct skb_shared_info,
9855 hwtstamps, 8,
9856 target_size));
9857 break;
9858 }
9859
9860 return insn - insn_buf;
9861 }
9862
bpf_sock_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)9863 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9864 const struct bpf_insn *si,
9865 struct bpf_insn *insn_buf,
9866 struct bpf_prog *prog, u32 *target_size)
9867 {
9868 struct bpf_insn *insn = insn_buf;
9869 int off;
9870
9871 switch (si->off) {
9872 case offsetof(struct bpf_sock, bound_dev_if):
9873 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9874
9875 if (type == BPF_WRITE)
9876 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9877 offsetof(struct sock, sk_bound_dev_if));
9878 else
9879 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9880 offsetof(struct sock, sk_bound_dev_if));
9881 break;
9882
9883 case offsetof(struct bpf_sock, mark):
9884 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9885
9886 if (type == BPF_WRITE)
9887 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9888 offsetof(struct sock, sk_mark));
9889 else
9890 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9891 offsetof(struct sock, sk_mark));
9892 break;
9893
9894 case offsetof(struct bpf_sock, priority):
9895 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9896
9897 if (type == BPF_WRITE)
9898 *insn++ = BPF_EMIT_STORE(BPF_W, si,
9899 offsetof(struct sock, sk_priority));
9900 else
9901 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9902 offsetof(struct sock, sk_priority));
9903 break;
9904
9905 case offsetof(struct bpf_sock, family):
9906 *insn++ = BPF_LDX_MEM(
9907 BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9908 si->dst_reg, si->src_reg,
9909 bpf_target_off(struct sock_common,
9910 skc_family,
9911 sizeof_field(struct sock_common,
9912 skc_family),
9913 target_size));
9914 break;
9915
9916 case offsetof(struct bpf_sock, type):
9917 *insn++ = BPF_LDX_MEM(
9918 BPF_FIELD_SIZEOF(struct sock, sk_type),
9919 si->dst_reg, si->src_reg,
9920 bpf_target_off(struct sock, sk_type,
9921 sizeof_field(struct sock, sk_type),
9922 target_size));
9923 break;
9924
9925 case offsetof(struct bpf_sock, protocol):
9926 *insn++ = BPF_LDX_MEM(
9927 BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9928 si->dst_reg, si->src_reg,
9929 bpf_target_off(struct sock, sk_protocol,
9930 sizeof_field(struct sock, sk_protocol),
9931 target_size));
9932 break;
9933
9934 case offsetof(struct bpf_sock, src_ip4):
9935 *insn++ = BPF_LDX_MEM(
9936 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9937 bpf_target_off(struct sock_common, skc_rcv_saddr,
9938 sizeof_field(struct sock_common,
9939 skc_rcv_saddr),
9940 target_size));
9941 break;
9942
9943 case offsetof(struct bpf_sock, dst_ip4):
9944 *insn++ = BPF_LDX_MEM(
9945 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9946 bpf_target_off(struct sock_common, skc_daddr,
9947 sizeof_field(struct sock_common,
9948 skc_daddr),
9949 target_size));
9950 break;
9951
9952 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9953 #if IS_ENABLED(CONFIG_IPV6)
9954 off = si->off;
9955 off -= offsetof(struct bpf_sock, src_ip6[0]);
9956 *insn++ = BPF_LDX_MEM(
9957 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9958 bpf_target_off(
9959 struct sock_common,
9960 skc_v6_rcv_saddr.s6_addr32[0],
9961 sizeof_field(struct sock_common,
9962 skc_v6_rcv_saddr.s6_addr32[0]),
9963 target_size) + off);
9964 #else
9965 (void)off;
9966 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9967 #endif
9968 break;
9969
9970 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9971 #if IS_ENABLED(CONFIG_IPV6)
9972 off = si->off;
9973 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9974 *insn++ = BPF_LDX_MEM(
9975 BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9976 bpf_target_off(struct sock_common,
9977 skc_v6_daddr.s6_addr32[0],
9978 sizeof_field(struct sock_common,
9979 skc_v6_daddr.s6_addr32[0]),
9980 target_size) + off);
9981 #else
9982 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9983 *target_size = 4;
9984 #endif
9985 break;
9986
9987 case offsetof(struct bpf_sock, src_port):
9988 *insn++ = BPF_LDX_MEM(
9989 BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9990 si->dst_reg, si->src_reg,
9991 bpf_target_off(struct sock_common, skc_num,
9992 sizeof_field(struct sock_common,
9993 skc_num),
9994 target_size));
9995 break;
9996
9997 case offsetof(struct bpf_sock, dst_port):
9998 *insn++ = BPF_LDX_MEM(
9999 BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
10000 si->dst_reg, si->src_reg,
10001 bpf_target_off(struct sock_common, skc_dport,
10002 sizeof_field(struct sock_common,
10003 skc_dport),
10004 target_size));
10005 break;
10006
10007 case offsetof(struct bpf_sock, state):
10008 *insn++ = BPF_LDX_MEM(
10009 BPF_FIELD_SIZEOF(struct sock_common, skc_state),
10010 si->dst_reg, si->src_reg,
10011 bpf_target_off(struct sock_common, skc_state,
10012 sizeof_field(struct sock_common,
10013 skc_state),
10014 target_size));
10015 break;
10016 case offsetof(struct bpf_sock, rx_queue_mapping):
10017 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
10018 *insn++ = BPF_LDX_MEM(
10019 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
10020 si->dst_reg, si->src_reg,
10021 bpf_target_off(struct sock, sk_rx_queue_mapping,
10022 sizeof_field(struct sock,
10023 sk_rx_queue_mapping),
10024 target_size));
10025 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
10026 1);
10027 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10028 #else
10029 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
10030 *target_size = 2;
10031 #endif
10032 break;
10033 }
10034
10035 return insn - insn_buf;
10036 }
10037
tc_cls_act_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10038 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
10039 const struct bpf_insn *si,
10040 struct bpf_insn *insn_buf,
10041 struct bpf_prog *prog, u32 *target_size)
10042 {
10043 struct bpf_insn *insn = insn_buf;
10044
10045 switch (si->off) {
10046 case offsetof(struct __sk_buff, ifindex):
10047 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
10048 si->dst_reg, si->src_reg,
10049 offsetof(struct sk_buff, dev));
10050 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10051 bpf_target_off(struct net_device, ifindex, 4,
10052 target_size));
10053 break;
10054 default:
10055 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10056 target_size);
10057 }
10058
10059 return insn - insn_buf;
10060 }
10061
xdp_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10062 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
10063 const struct bpf_insn *si,
10064 struct bpf_insn *insn_buf,
10065 struct bpf_prog *prog, u32 *target_size)
10066 {
10067 struct bpf_insn *insn = insn_buf;
10068
10069 switch (si->off) {
10070 case offsetof(struct xdp_md, data):
10071 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
10072 si->dst_reg, si->src_reg,
10073 offsetof(struct xdp_buff, data));
10074 break;
10075 case offsetof(struct xdp_md, data_meta):
10076 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
10077 si->dst_reg, si->src_reg,
10078 offsetof(struct xdp_buff, data_meta));
10079 break;
10080 case offsetof(struct xdp_md, data_end):
10081 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
10082 si->dst_reg, si->src_reg,
10083 offsetof(struct xdp_buff, data_end));
10084 break;
10085 case offsetof(struct xdp_md, ingress_ifindex):
10086 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10087 si->dst_reg, si->src_reg,
10088 offsetof(struct xdp_buff, rxq));
10089 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
10090 si->dst_reg, si->dst_reg,
10091 offsetof(struct xdp_rxq_info, dev));
10092 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10093 offsetof(struct net_device, ifindex));
10094 break;
10095 case offsetof(struct xdp_md, rx_queue_index):
10096 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
10097 si->dst_reg, si->src_reg,
10098 offsetof(struct xdp_buff, rxq));
10099 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10100 offsetof(struct xdp_rxq_info,
10101 queue_index));
10102 break;
10103 case offsetof(struct xdp_md, egress_ifindex):
10104 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
10105 si->dst_reg, si->src_reg,
10106 offsetof(struct xdp_buff, txq));
10107 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
10108 si->dst_reg, si->dst_reg,
10109 offsetof(struct xdp_txq_info, dev));
10110 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10111 offsetof(struct net_device, ifindex));
10112 break;
10113 }
10114
10115 return insn - insn_buf;
10116 }
10117
10118 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
10119 * context Structure, F is Field in context structure that contains a pointer
10120 * to Nested Structure of type NS that has the field NF.
10121 *
10122 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10123 * sure that SIZE is not greater than actual size of S.F.NF.
10124 *
10125 * If offset OFF is provided, the load happens from that offset relative to
10126 * offset of NF.
10127 */
10128 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \
10129 do { \
10130 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \
10131 si->src_reg, offsetof(S, F)); \
10132 *insn++ = BPF_LDX_MEM( \
10133 SIZE, si->dst_reg, si->dst_reg, \
10134 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10135 target_size) \
10136 + OFF); \
10137 } while (0)
10138
10139 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \
10140 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \
10141 BPF_FIELD_SIZEOF(NS, NF), 0)
10142
10143 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10144 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10145 *
10146 * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10147 * "register" since two registers available in convert_ctx_access are not
10148 * enough: we can't override neither SRC, since it contains value to store, nor
10149 * DST since it contains pointer to context that may be used by later
10150 * instructions. But we need a temporary place to save pointer to nested
10151 * structure whose field we want to store to.
10152 */
10153 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \
10154 do { \
10155 int tmp_reg = BPF_REG_9; \
10156 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10157 --tmp_reg; \
10158 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \
10159 --tmp_reg; \
10160 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \
10161 offsetof(S, TF)); \
10162 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \
10163 si->dst_reg, offsetof(S, F)); \
10164 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code), \
10165 tmp_reg, si->src_reg, \
10166 bpf_target_off(NS, NF, sizeof_field(NS, NF), \
10167 target_size) \
10168 + OFF, \
10169 si->imm); \
10170 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \
10171 offsetof(S, TF)); \
10172 } while (0)
10173
10174 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10175 TF) \
10176 do { \
10177 if (type == BPF_WRITE) { \
10178 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \
10179 OFF, TF); \
10180 } else { \
10181 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \
10182 S, NS, F, NF, SIZE, OFF); \
10183 } \
10184 } while (0)
10185
10186 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \
10187 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \
10188 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10189
sock_addr_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10190 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10191 const struct bpf_insn *si,
10192 struct bpf_insn *insn_buf,
10193 struct bpf_prog *prog, u32 *target_size)
10194 {
10195 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10196 struct bpf_insn *insn = insn_buf;
10197
10198 switch (si->off) {
10199 case offsetof(struct bpf_sock_addr, user_family):
10200 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10201 struct sockaddr, uaddr, sa_family);
10202 break;
10203
10204 case offsetof(struct bpf_sock_addr, user_ip4):
10205 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10206 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10207 sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10208 break;
10209
10210 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10211 off = si->off;
10212 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10213 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10214 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10215 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10216 tmp_reg);
10217 break;
10218
10219 case offsetof(struct bpf_sock_addr, user_port):
10220 /* To get port we need to know sa_family first and then treat
10221 * sockaddr as either sockaddr_in or sockaddr_in6.
10222 * Though we can simplify since port field has same offset and
10223 * size in both structures.
10224 * Here we check this invariant and use just one of the
10225 * structures if it's true.
10226 */
10227 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10228 offsetof(struct sockaddr_in6, sin6_port));
10229 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10230 sizeof_field(struct sockaddr_in6, sin6_port));
10231 /* Account for sin6_port being smaller than user_port. */
10232 port_size = min(port_size, BPF_LDST_BYTES(si));
10233 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10234 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10235 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10236 break;
10237
10238 case offsetof(struct bpf_sock_addr, family):
10239 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10240 struct sock, sk, sk_family);
10241 break;
10242
10243 case offsetof(struct bpf_sock_addr, type):
10244 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10245 struct sock, sk, sk_type);
10246 break;
10247
10248 case offsetof(struct bpf_sock_addr, protocol):
10249 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10250 struct sock, sk, sk_protocol);
10251 break;
10252
10253 case offsetof(struct bpf_sock_addr, msg_src_ip4):
10254 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10255 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10256 struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10257 s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10258 break;
10259
10260 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10261 msg_src_ip6[3]):
10262 off = si->off;
10263 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10264 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10265 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10266 struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10267 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10268 break;
10269 case offsetof(struct bpf_sock_addr, sk):
10270 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10271 si->dst_reg, si->src_reg,
10272 offsetof(struct bpf_sock_addr_kern, sk));
10273 break;
10274 }
10275
10276 return insn - insn_buf;
10277 }
10278
sock_ops_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10279 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10280 const struct bpf_insn *si,
10281 struct bpf_insn *insn_buf,
10282 struct bpf_prog *prog,
10283 u32 *target_size)
10284 {
10285 struct bpf_insn *insn = insn_buf;
10286 int off;
10287
10288 /* Helper macro for adding read access to tcp_sock or sock fields. */
10289 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10290 do { \
10291 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \
10292 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10293 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10294 if (si->dst_reg == reg || si->src_reg == reg) \
10295 reg--; \
10296 if (si->dst_reg == reg || si->src_reg == reg) \
10297 reg--; \
10298 if (si->dst_reg == si->src_reg) { \
10299 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10300 offsetof(struct bpf_sock_ops_kern, \
10301 temp)); \
10302 fullsock_reg = reg; \
10303 jmp += 2; \
10304 } \
10305 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10306 struct bpf_sock_ops_kern, \
10307 is_fullsock), \
10308 fullsock_reg, si->src_reg, \
10309 offsetof(struct bpf_sock_ops_kern, \
10310 is_fullsock)); \
10311 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10312 if (si->dst_reg == si->src_reg) \
10313 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10314 offsetof(struct bpf_sock_ops_kern, \
10315 temp)); \
10316 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10317 struct bpf_sock_ops_kern, sk),\
10318 si->dst_reg, si->src_reg, \
10319 offsetof(struct bpf_sock_ops_kern, sk));\
10320 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \
10321 OBJ_FIELD), \
10322 si->dst_reg, si->dst_reg, \
10323 offsetof(OBJ, OBJ_FIELD)); \
10324 if (si->dst_reg == si->src_reg) { \
10325 *insn++ = BPF_JMP_A(1); \
10326 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10327 offsetof(struct bpf_sock_ops_kern, \
10328 temp)); \
10329 } \
10330 } while (0)
10331
10332 #define SOCK_OPS_GET_SK() \
10333 do { \
10334 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \
10335 if (si->dst_reg == reg || si->src_reg == reg) \
10336 reg--; \
10337 if (si->dst_reg == reg || si->src_reg == reg) \
10338 reg--; \
10339 if (si->dst_reg == si->src_reg) { \
10340 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \
10341 offsetof(struct bpf_sock_ops_kern, \
10342 temp)); \
10343 fullsock_reg = reg; \
10344 jmp += 2; \
10345 } \
10346 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10347 struct bpf_sock_ops_kern, \
10348 is_fullsock), \
10349 fullsock_reg, si->src_reg, \
10350 offsetof(struct bpf_sock_ops_kern, \
10351 is_fullsock)); \
10352 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \
10353 if (si->dst_reg == si->src_reg) \
10354 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10355 offsetof(struct bpf_sock_ops_kern, \
10356 temp)); \
10357 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10358 struct bpf_sock_ops_kern, sk),\
10359 si->dst_reg, si->src_reg, \
10360 offsetof(struct bpf_sock_ops_kern, sk));\
10361 if (si->dst_reg == si->src_reg) { \
10362 *insn++ = BPF_JMP_A(1); \
10363 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \
10364 offsetof(struct bpf_sock_ops_kern, \
10365 temp)); \
10366 } \
10367 } while (0)
10368
10369 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10370 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10371
10372 /* Helper macro for adding write access to tcp_sock or sock fields.
10373 * The macro is called with two registers, dst_reg which contains a pointer
10374 * to ctx (context) and src_reg which contains the value that should be
10375 * stored. However, we need an additional register since we cannot overwrite
10376 * dst_reg because it may be used later in the program.
10377 * Instead we "borrow" one of the other register. We first save its value
10378 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10379 * it at the end of the macro.
10380 */
10381 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \
10382 do { \
10383 int reg = BPF_REG_9; \
10384 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \
10385 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \
10386 if (si->dst_reg == reg || si->src_reg == reg) \
10387 reg--; \
10388 if (si->dst_reg == reg || si->src_reg == reg) \
10389 reg--; \
10390 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \
10391 offsetof(struct bpf_sock_ops_kern, \
10392 temp)); \
10393 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10394 struct bpf_sock_ops_kern, \
10395 is_fullsock), \
10396 reg, si->dst_reg, \
10397 offsetof(struct bpf_sock_ops_kern, \
10398 is_fullsock)); \
10399 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \
10400 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \
10401 struct bpf_sock_ops_kern, sk),\
10402 reg, si->dst_reg, \
10403 offsetof(struct bpf_sock_ops_kern, sk));\
10404 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) | \
10405 BPF_MEM | BPF_CLASS(si->code), \
10406 reg, si->src_reg, \
10407 offsetof(OBJ, OBJ_FIELD), \
10408 si->imm); \
10409 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \
10410 offsetof(struct bpf_sock_ops_kern, \
10411 temp)); \
10412 } while (0)
10413
10414 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \
10415 do { \
10416 if (TYPE == BPF_WRITE) \
10417 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10418 else \
10419 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \
10420 } while (0)
10421
10422 switch (si->off) {
10423 case offsetof(struct bpf_sock_ops, op):
10424 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10425 op),
10426 si->dst_reg, si->src_reg,
10427 offsetof(struct bpf_sock_ops_kern, op));
10428 break;
10429
10430 case offsetof(struct bpf_sock_ops, replylong[0]) ...
10431 offsetof(struct bpf_sock_ops, replylong[3]):
10432 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10433 sizeof_field(struct bpf_sock_ops_kern, reply));
10434 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10435 sizeof_field(struct bpf_sock_ops_kern, replylong));
10436 off = si->off;
10437 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10438 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10439 if (type == BPF_WRITE)
10440 *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10441 else
10442 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10443 off);
10444 break;
10445
10446 case offsetof(struct bpf_sock_ops, family):
10447 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10448
10449 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10450 struct bpf_sock_ops_kern, sk),
10451 si->dst_reg, si->src_reg,
10452 offsetof(struct bpf_sock_ops_kern, sk));
10453 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10454 offsetof(struct sock_common, skc_family));
10455 break;
10456
10457 case offsetof(struct bpf_sock_ops, remote_ip4):
10458 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10459
10460 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10461 struct bpf_sock_ops_kern, sk),
10462 si->dst_reg, si->src_reg,
10463 offsetof(struct bpf_sock_ops_kern, sk));
10464 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10465 offsetof(struct sock_common, skc_daddr));
10466 break;
10467
10468 case offsetof(struct bpf_sock_ops, local_ip4):
10469 BUILD_BUG_ON(sizeof_field(struct sock_common,
10470 skc_rcv_saddr) != 4);
10471
10472 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10473 struct bpf_sock_ops_kern, sk),
10474 si->dst_reg, si->src_reg,
10475 offsetof(struct bpf_sock_ops_kern, sk));
10476 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10477 offsetof(struct sock_common,
10478 skc_rcv_saddr));
10479 break;
10480
10481 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10482 offsetof(struct bpf_sock_ops, remote_ip6[3]):
10483 #if IS_ENABLED(CONFIG_IPV6)
10484 BUILD_BUG_ON(sizeof_field(struct sock_common,
10485 skc_v6_daddr.s6_addr32[0]) != 4);
10486
10487 off = si->off;
10488 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10489 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10490 struct bpf_sock_ops_kern, sk),
10491 si->dst_reg, si->src_reg,
10492 offsetof(struct bpf_sock_ops_kern, sk));
10493 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10494 offsetof(struct sock_common,
10495 skc_v6_daddr.s6_addr32[0]) +
10496 off);
10497 #else
10498 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10499 #endif
10500 break;
10501
10502 case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10503 offsetof(struct bpf_sock_ops, local_ip6[3]):
10504 #if IS_ENABLED(CONFIG_IPV6)
10505 BUILD_BUG_ON(sizeof_field(struct sock_common,
10506 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10507
10508 off = si->off;
10509 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10510 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10511 struct bpf_sock_ops_kern, sk),
10512 si->dst_reg, si->src_reg,
10513 offsetof(struct bpf_sock_ops_kern, sk));
10514 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10515 offsetof(struct sock_common,
10516 skc_v6_rcv_saddr.s6_addr32[0]) +
10517 off);
10518 #else
10519 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10520 #endif
10521 break;
10522
10523 case offsetof(struct bpf_sock_ops, remote_port):
10524 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10525
10526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10527 struct bpf_sock_ops_kern, sk),
10528 si->dst_reg, si->src_reg,
10529 offsetof(struct bpf_sock_ops_kern, sk));
10530 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10531 offsetof(struct sock_common, skc_dport));
10532 #ifndef __BIG_ENDIAN_BITFIELD
10533 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10534 #endif
10535 break;
10536
10537 case offsetof(struct bpf_sock_ops, local_port):
10538 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10539
10540 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10541 struct bpf_sock_ops_kern, sk),
10542 si->dst_reg, si->src_reg,
10543 offsetof(struct bpf_sock_ops_kern, sk));
10544 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10545 offsetof(struct sock_common, skc_num));
10546 break;
10547
10548 case offsetof(struct bpf_sock_ops, is_fullsock):
10549 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10550 struct bpf_sock_ops_kern,
10551 is_fullsock),
10552 si->dst_reg, si->src_reg,
10553 offsetof(struct bpf_sock_ops_kern,
10554 is_fullsock));
10555 break;
10556
10557 case offsetof(struct bpf_sock_ops, state):
10558 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10559
10560 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10561 struct bpf_sock_ops_kern, sk),
10562 si->dst_reg, si->src_reg,
10563 offsetof(struct bpf_sock_ops_kern, sk));
10564 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10565 offsetof(struct sock_common, skc_state));
10566 break;
10567
10568 case offsetof(struct bpf_sock_ops, rtt_min):
10569 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10570 sizeof(struct minmax));
10571 BUILD_BUG_ON(sizeof(struct minmax) <
10572 sizeof(struct minmax_sample));
10573
10574 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10575 struct bpf_sock_ops_kern, sk),
10576 si->dst_reg, si->src_reg,
10577 offsetof(struct bpf_sock_ops_kern, sk));
10578 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10579 offsetof(struct tcp_sock, rtt_min) +
10580 sizeof_field(struct minmax_sample, t));
10581 break;
10582
10583 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10584 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10585 struct tcp_sock);
10586 break;
10587
10588 case offsetof(struct bpf_sock_ops, sk_txhash):
10589 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10590 struct sock, type);
10591 break;
10592 case offsetof(struct bpf_sock_ops, snd_cwnd):
10593 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10594 break;
10595 case offsetof(struct bpf_sock_ops, srtt_us):
10596 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10597 break;
10598 case offsetof(struct bpf_sock_ops, snd_ssthresh):
10599 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10600 break;
10601 case offsetof(struct bpf_sock_ops, rcv_nxt):
10602 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10603 break;
10604 case offsetof(struct bpf_sock_ops, snd_nxt):
10605 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10606 break;
10607 case offsetof(struct bpf_sock_ops, snd_una):
10608 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10609 break;
10610 case offsetof(struct bpf_sock_ops, mss_cache):
10611 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10612 break;
10613 case offsetof(struct bpf_sock_ops, ecn_flags):
10614 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10615 break;
10616 case offsetof(struct bpf_sock_ops, rate_delivered):
10617 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10618 break;
10619 case offsetof(struct bpf_sock_ops, rate_interval_us):
10620 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10621 break;
10622 case offsetof(struct bpf_sock_ops, packets_out):
10623 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10624 break;
10625 case offsetof(struct bpf_sock_ops, retrans_out):
10626 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10627 break;
10628 case offsetof(struct bpf_sock_ops, total_retrans):
10629 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10630 break;
10631 case offsetof(struct bpf_sock_ops, segs_in):
10632 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10633 break;
10634 case offsetof(struct bpf_sock_ops, data_segs_in):
10635 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10636 break;
10637 case offsetof(struct bpf_sock_ops, segs_out):
10638 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10639 break;
10640 case offsetof(struct bpf_sock_ops, data_segs_out):
10641 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10642 break;
10643 case offsetof(struct bpf_sock_ops, lost_out):
10644 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10645 break;
10646 case offsetof(struct bpf_sock_ops, sacked_out):
10647 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10648 break;
10649 case offsetof(struct bpf_sock_ops, bytes_received):
10650 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10651 break;
10652 case offsetof(struct bpf_sock_ops, bytes_acked):
10653 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10654 break;
10655 case offsetof(struct bpf_sock_ops, sk):
10656 SOCK_OPS_GET_SK();
10657 break;
10658 case offsetof(struct bpf_sock_ops, skb_data_end):
10659 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10660 skb_data_end),
10661 si->dst_reg, si->src_reg,
10662 offsetof(struct bpf_sock_ops_kern,
10663 skb_data_end));
10664 break;
10665 case offsetof(struct bpf_sock_ops, skb_data):
10666 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10667 skb),
10668 si->dst_reg, si->src_reg,
10669 offsetof(struct bpf_sock_ops_kern,
10670 skb));
10671 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10672 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10673 si->dst_reg, si->dst_reg,
10674 offsetof(struct sk_buff, data));
10675 break;
10676 case offsetof(struct bpf_sock_ops, skb_len):
10677 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10678 skb),
10679 si->dst_reg, si->src_reg,
10680 offsetof(struct bpf_sock_ops_kern,
10681 skb));
10682 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10683 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10684 si->dst_reg, si->dst_reg,
10685 offsetof(struct sk_buff, len));
10686 break;
10687 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10688 off = offsetof(struct sk_buff, cb);
10689 off += offsetof(struct tcp_skb_cb, tcp_flags);
10690 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10691 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10692 skb),
10693 si->dst_reg, si->src_reg,
10694 offsetof(struct bpf_sock_ops_kern,
10695 skb));
10696 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10697 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10698 tcp_flags),
10699 si->dst_reg, si->dst_reg, off);
10700 break;
10701 case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10702 struct bpf_insn *jmp_on_null_skb;
10703
10704 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10705 skb),
10706 si->dst_reg, si->src_reg,
10707 offsetof(struct bpf_sock_ops_kern,
10708 skb));
10709 /* Reserve one insn to test skb == NULL */
10710 jmp_on_null_skb = insn++;
10711 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10712 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10713 bpf_target_off(struct skb_shared_info,
10714 hwtstamps, 8,
10715 target_size));
10716 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10717 insn - jmp_on_null_skb - 1);
10718 break;
10719 }
10720 }
10721 return insn - insn_buf;
10722 }
10723
10724 /* data_end = skb->data + skb_headlen() */
bpf_convert_data_end_access(const struct bpf_insn * si,struct bpf_insn * insn)10725 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10726 struct bpf_insn *insn)
10727 {
10728 int reg;
10729 int temp_reg_off = offsetof(struct sk_buff, cb) +
10730 offsetof(struct sk_skb_cb, temp_reg);
10731
10732 if (si->src_reg == si->dst_reg) {
10733 /* We need an extra register, choose and save a register. */
10734 reg = BPF_REG_9;
10735 if (si->src_reg == reg || si->dst_reg == reg)
10736 reg--;
10737 if (si->src_reg == reg || si->dst_reg == reg)
10738 reg--;
10739 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10740 } else {
10741 reg = si->dst_reg;
10742 }
10743
10744 /* reg = skb->data */
10745 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10746 reg, si->src_reg,
10747 offsetof(struct sk_buff, data));
10748 /* AX = skb->len */
10749 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10750 BPF_REG_AX, si->src_reg,
10751 offsetof(struct sk_buff, len));
10752 /* reg = skb->data + skb->len */
10753 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10754 /* AX = skb->data_len */
10755 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10756 BPF_REG_AX, si->src_reg,
10757 offsetof(struct sk_buff, data_len));
10758
10759 /* reg = skb->data + skb->len - skb->data_len */
10760 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10761
10762 if (si->src_reg == si->dst_reg) {
10763 /* Restore the saved register */
10764 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10765 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10766 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10767 }
10768
10769 return insn;
10770 }
10771
sk_skb_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10772 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10773 const struct bpf_insn *si,
10774 struct bpf_insn *insn_buf,
10775 struct bpf_prog *prog, u32 *target_size)
10776 {
10777 struct bpf_insn *insn = insn_buf;
10778 int off;
10779
10780 switch (si->off) {
10781 case offsetof(struct __sk_buff, data_end):
10782 insn = bpf_convert_data_end_access(si, insn);
10783 break;
10784 case offsetof(struct __sk_buff, cb[0]) ...
10785 offsetofend(struct __sk_buff, cb[4]) - 1:
10786 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10787 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10788 offsetof(struct sk_skb_cb, data)) %
10789 sizeof(__u64));
10790
10791 prog->cb_access = 1;
10792 off = si->off;
10793 off -= offsetof(struct __sk_buff, cb[0]);
10794 off += offsetof(struct sk_buff, cb);
10795 off += offsetof(struct sk_skb_cb, data);
10796 if (type == BPF_WRITE)
10797 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10798 else
10799 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10800 si->src_reg, off);
10801 break;
10802
10803
10804 default:
10805 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10806 target_size);
10807 }
10808
10809 return insn - insn_buf;
10810 }
10811
sk_msg_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)10812 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10813 const struct bpf_insn *si,
10814 struct bpf_insn *insn_buf,
10815 struct bpf_prog *prog, u32 *target_size)
10816 {
10817 struct bpf_insn *insn = insn_buf;
10818 #if IS_ENABLED(CONFIG_IPV6)
10819 int off;
10820 #endif
10821
10822 /* convert ctx uses the fact sg element is first in struct */
10823 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10824
10825 switch (si->off) {
10826 case offsetof(struct sk_msg_md, data):
10827 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10828 si->dst_reg, si->src_reg,
10829 offsetof(struct sk_msg, data));
10830 break;
10831 case offsetof(struct sk_msg_md, data_end):
10832 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10833 si->dst_reg, si->src_reg,
10834 offsetof(struct sk_msg, data_end));
10835 break;
10836 case offsetof(struct sk_msg_md, family):
10837 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10838
10839 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10840 struct sk_msg, sk),
10841 si->dst_reg, si->src_reg,
10842 offsetof(struct sk_msg, sk));
10843 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10844 offsetof(struct sock_common, skc_family));
10845 break;
10846
10847 case offsetof(struct sk_msg_md, remote_ip4):
10848 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10849
10850 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10851 struct sk_msg, sk),
10852 si->dst_reg, si->src_reg,
10853 offsetof(struct sk_msg, sk));
10854 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10855 offsetof(struct sock_common, skc_daddr));
10856 break;
10857
10858 case offsetof(struct sk_msg_md, local_ip4):
10859 BUILD_BUG_ON(sizeof_field(struct sock_common,
10860 skc_rcv_saddr) != 4);
10861
10862 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10863 struct sk_msg, sk),
10864 si->dst_reg, si->src_reg,
10865 offsetof(struct sk_msg, sk));
10866 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10867 offsetof(struct sock_common,
10868 skc_rcv_saddr));
10869 break;
10870
10871 case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10872 offsetof(struct sk_msg_md, remote_ip6[3]):
10873 #if IS_ENABLED(CONFIG_IPV6)
10874 BUILD_BUG_ON(sizeof_field(struct sock_common,
10875 skc_v6_daddr.s6_addr32[0]) != 4);
10876
10877 off = si->off;
10878 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10879 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10880 struct sk_msg, sk),
10881 si->dst_reg, si->src_reg,
10882 offsetof(struct sk_msg, sk));
10883 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10884 offsetof(struct sock_common,
10885 skc_v6_daddr.s6_addr32[0]) +
10886 off);
10887 #else
10888 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10889 #endif
10890 break;
10891
10892 case offsetof(struct sk_msg_md, local_ip6[0]) ...
10893 offsetof(struct sk_msg_md, local_ip6[3]):
10894 #if IS_ENABLED(CONFIG_IPV6)
10895 BUILD_BUG_ON(sizeof_field(struct sock_common,
10896 skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10897
10898 off = si->off;
10899 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10900 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10901 struct sk_msg, sk),
10902 si->dst_reg, si->src_reg,
10903 offsetof(struct sk_msg, sk));
10904 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10905 offsetof(struct sock_common,
10906 skc_v6_rcv_saddr.s6_addr32[0]) +
10907 off);
10908 #else
10909 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10910 #endif
10911 break;
10912
10913 case offsetof(struct sk_msg_md, remote_port):
10914 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10915
10916 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10917 struct sk_msg, sk),
10918 si->dst_reg, si->src_reg,
10919 offsetof(struct sk_msg, sk));
10920 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10921 offsetof(struct sock_common, skc_dport));
10922 #ifndef __BIG_ENDIAN_BITFIELD
10923 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10924 #endif
10925 break;
10926
10927 case offsetof(struct sk_msg_md, local_port):
10928 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10929
10930 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10931 struct sk_msg, sk),
10932 si->dst_reg, si->src_reg,
10933 offsetof(struct sk_msg, sk));
10934 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10935 offsetof(struct sock_common, skc_num));
10936 break;
10937
10938 case offsetof(struct sk_msg_md, size):
10939 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10940 si->dst_reg, si->src_reg,
10941 offsetof(struct sk_msg_sg, size));
10942 break;
10943
10944 case offsetof(struct sk_msg_md, sk):
10945 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10946 si->dst_reg, si->src_reg,
10947 offsetof(struct sk_msg, sk));
10948 break;
10949 }
10950
10951 return insn - insn_buf;
10952 }
10953
10954 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10955 .get_func_proto = sk_filter_func_proto,
10956 .is_valid_access = sk_filter_is_valid_access,
10957 .convert_ctx_access = bpf_convert_ctx_access,
10958 .gen_ld_abs = bpf_gen_ld_abs,
10959 };
10960
10961 const struct bpf_prog_ops sk_filter_prog_ops = {
10962 .test_run = bpf_prog_test_run_skb,
10963 };
10964
10965 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10966 .get_func_proto = tc_cls_act_func_proto,
10967 .is_valid_access = tc_cls_act_is_valid_access,
10968 .convert_ctx_access = tc_cls_act_convert_ctx_access,
10969 .gen_prologue = tc_cls_act_prologue,
10970 .gen_ld_abs = bpf_gen_ld_abs,
10971 .btf_struct_access = tc_cls_act_btf_struct_access,
10972 };
10973
10974 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10975 .test_run = bpf_prog_test_run_skb,
10976 };
10977
10978 const struct bpf_verifier_ops xdp_verifier_ops = {
10979 .get_func_proto = xdp_func_proto,
10980 .is_valid_access = xdp_is_valid_access,
10981 .convert_ctx_access = xdp_convert_ctx_access,
10982 .gen_prologue = bpf_noop_prologue,
10983 .btf_struct_access = xdp_btf_struct_access,
10984 };
10985
10986 const struct bpf_prog_ops xdp_prog_ops = {
10987 .test_run = bpf_prog_test_run_xdp,
10988 };
10989
10990 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10991 .get_func_proto = cg_skb_func_proto,
10992 .is_valid_access = cg_skb_is_valid_access,
10993 .convert_ctx_access = bpf_convert_ctx_access,
10994 };
10995
10996 const struct bpf_prog_ops cg_skb_prog_ops = {
10997 .test_run = bpf_prog_test_run_skb,
10998 };
10999
11000 const struct bpf_verifier_ops lwt_in_verifier_ops = {
11001 .get_func_proto = lwt_in_func_proto,
11002 .is_valid_access = lwt_is_valid_access,
11003 .convert_ctx_access = bpf_convert_ctx_access,
11004 };
11005
11006 const struct bpf_prog_ops lwt_in_prog_ops = {
11007 .test_run = bpf_prog_test_run_skb,
11008 };
11009
11010 const struct bpf_verifier_ops lwt_out_verifier_ops = {
11011 .get_func_proto = lwt_out_func_proto,
11012 .is_valid_access = lwt_is_valid_access,
11013 .convert_ctx_access = bpf_convert_ctx_access,
11014 };
11015
11016 const struct bpf_prog_ops lwt_out_prog_ops = {
11017 .test_run = bpf_prog_test_run_skb,
11018 };
11019
11020 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
11021 .get_func_proto = lwt_xmit_func_proto,
11022 .is_valid_access = lwt_is_valid_access,
11023 .convert_ctx_access = bpf_convert_ctx_access,
11024 .gen_prologue = tc_cls_act_prologue,
11025 };
11026
11027 const struct bpf_prog_ops lwt_xmit_prog_ops = {
11028 .test_run = bpf_prog_test_run_skb,
11029 };
11030
11031 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
11032 .get_func_proto = lwt_seg6local_func_proto,
11033 .is_valid_access = lwt_is_valid_access,
11034 .convert_ctx_access = bpf_convert_ctx_access,
11035 };
11036
11037 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
11038 .test_run = bpf_prog_test_run_skb,
11039 };
11040
11041 const struct bpf_verifier_ops cg_sock_verifier_ops = {
11042 .get_func_proto = sock_filter_func_proto,
11043 .is_valid_access = sock_filter_is_valid_access,
11044 .convert_ctx_access = bpf_sock_convert_ctx_access,
11045 };
11046
11047 const struct bpf_prog_ops cg_sock_prog_ops = {
11048 };
11049
11050 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
11051 .get_func_proto = sock_addr_func_proto,
11052 .is_valid_access = sock_addr_is_valid_access,
11053 .convert_ctx_access = sock_addr_convert_ctx_access,
11054 };
11055
11056 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
11057 };
11058
11059 const struct bpf_verifier_ops sock_ops_verifier_ops = {
11060 .get_func_proto = sock_ops_func_proto,
11061 .is_valid_access = sock_ops_is_valid_access,
11062 .convert_ctx_access = sock_ops_convert_ctx_access,
11063 };
11064
11065 const struct bpf_prog_ops sock_ops_prog_ops = {
11066 };
11067
11068 const struct bpf_verifier_ops sk_skb_verifier_ops = {
11069 .get_func_proto = sk_skb_func_proto,
11070 .is_valid_access = sk_skb_is_valid_access,
11071 .convert_ctx_access = sk_skb_convert_ctx_access,
11072 .gen_prologue = sk_skb_prologue,
11073 };
11074
11075 const struct bpf_prog_ops sk_skb_prog_ops = {
11076 };
11077
11078 const struct bpf_verifier_ops sk_msg_verifier_ops = {
11079 .get_func_proto = sk_msg_func_proto,
11080 .is_valid_access = sk_msg_is_valid_access,
11081 .convert_ctx_access = sk_msg_convert_ctx_access,
11082 .gen_prologue = bpf_noop_prologue,
11083 };
11084
11085 const struct bpf_prog_ops sk_msg_prog_ops = {
11086 };
11087
11088 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
11089 .get_func_proto = flow_dissector_func_proto,
11090 .is_valid_access = flow_dissector_is_valid_access,
11091 .convert_ctx_access = flow_dissector_convert_ctx_access,
11092 };
11093
11094 const struct bpf_prog_ops flow_dissector_prog_ops = {
11095 .test_run = bpf_prog_test_run_flow_dissector,
11096 };
11097
sk_detach_filter(struct sock * sk)11098 int sk_detach_filter(struct sock *sk)
11099 {
11100 int ret = -ENOENT;
11101 struct sk_filter *filter;
11102
11103 if (sock_flag(sk, SOCK_FILTER_LOCKED))
11104 return -EPERM;
11105
11106 filter = rcu_dereference_protected(sk->sk_filter,
11107 lockdep_sock_is_held(sk));
11108 if (filter) {
11109 RCU_INIT_POINTER(sk->sk_filter, NULL);
11110 sk_filter_uncharge(sk, filter);
11111 ret = 0;
11112 }
11113
11114 return ret;
11115 }
11116 EXPORT_SYMBOL_GPL(sk_detach_filter);
11117
sk_get_filter(struct sock * sk,sockptr_t optval,unsigned int len)11118 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
11119 {
11120 struct sock_fprog_kern *fprog;
11121 struct sk_filter *filter;
11122 int ret = 0;
11123
11124 sockopt_lock_sock(sk);
11125 filter = rcu_dereference_protected(sk->sk_filter,
11126 lockdep_sock_is_held(sk));
11127 if (!filter)
11128 goto out;
11129
11130 /* We're copying the filter that has been originally attached,
11131 * so no conversion/decode needed anymore. eBPF programs that
11132 * have no original program cannot be dumped through this.
11133 */
11134 ret = -EACCES;
11135 fprog = filter->prog->orig_prog;
11136 if (!fprog)
11137 goto out;
11138
11139 ret = fprog->len;
11140 if (!len)
11141 /* User space only enquires number of filter blocks. */
11142 goto out;
11143
11144 ret = -EINVAL;
11145 if (len < fprog->len)
11146 goto out;
11147
11148 ret = -EFAULT;
11149 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11150 goto out;
11151
11152 /* Instead of bytes, the API requests to return the number
11153 * of filter blocks.
11154 */
11155 ret = fprog->len;
11156 out:
11157 sockopt_release_sock(sk);
11158 return ret;
11159 }
11160
11161 #ifdef CONFIG_INET
bpf_init_reuseport_kern(struct sk_reuseport_kern * reuse_kern,struct sock_reuseport * reuse,struct sock * sk,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)11162 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11163 struct sock_reuseport *reuse,
11164 struct sock *sk, struct sk_buff *skb,
11165 struct sock *migrating_sk,
11166 u32 hash)
11167 {
11168 reuse_kern->skb = skb;
11169 reuse_kern->sk = sk;
11170 reuse_kern->selected_sk = NULL;
11171 reuse_kern->migrating_sk = migrating_sk;
11172 reuse_kern->data_end = skb->data + skb_headlen(skb);
11173 reuse_kern->hash = hash;
11174 reuse_kern->reuseport_id = reuse->reuseport_id;
11175 reuse_kern->bind_inany = reuse->bind_inany;
11176 }
11177
bpf_run_sk_reuseport(struct sock_reuseport * reuse,struct sock * sk,struct bpf_prog * prog,struct sk_buff * skb,struct sock * migrating_sk,u32 hash)11178 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11179 struct bpf_prog *prog, struct sk_buff *skb,
11180 struct sock *migrating_sk,
11181 u32 hash)
11182 {
11183 struct sk_reuseport_kern reuse_kern;
11184 enum sk_action action;
11185
11186 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11187 action = bpf_prog_run(prog, &reuse_kern);
11188
11189 if (action == SK_PASS)
11190 return reuse_kern.selected_sk;
11191 else
11192 return ERR_PTR(-ECONNREFUSED);
11193 }
11194
BPF_CALL_4(sk_select_reuseport,struct sk_reuseport_kern *,reuse_kern,struct bpf_map *,map,void *,key,u32,flags)11195 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11196 struct bpf_map *, map, void *, key, u32, flags)
11197 {
11198 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11199 struct sock_reuseport *reuse;
11200 struct sock *selected_sk;
11201
11202 selected_sk = map->ops->map_lookup_elem(map, key);
11203 if (!selected_sk)
11204 return -ENOENT;
11205
11206 reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11207 if (!reuse) {
11208 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11209 if (sk_is_refcounted(selected_sk))
11210 sock_put(selected_sk);
11211
11212 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11213 * The only (!reuse) case here is - the sk has already been
11214 * unhashed (e.g. by close()), so treat it as -ENOENT.
11215 *
11216 * Other maps (e.g. sock_map) do not provide this guarantee and
11217 * the sk may never be in the reuseport group to begin with.
11218 */
11219 return is_sockarray ? -ENOENT : -EINVAL;
11220 }
11221
11222 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11223 struct sock *sk = reuse_kern->sk;
11224
11225 if (sk->sk_protocol != selected_sk->sk_protocol)
11226 return -EPROTOTYPE;
11227 else if (sk->sk_family != selected_sk->sk_family)
11228 return -EAFNOSUPPORT;
11229
11230 /* Catch all. Likely bound to a different sockaddr. */
11231 return -EBADFD;
11232 }
11233
11234 reuse_kern->selected_sk = selected_sk;
11235
11236 return 0;
11237 }
11238
11239 static const struct bpf_func_proto sk_select_reuseport_proto = {
11240 .func = sk_select_reuseport,
11241 .gpl_only = false,
11242 .ret_type = RET_INTEGER,
11243 .arg1_type = ARG_PTR_TO_CTX,
11244 .arg2_type = ARG_CONST_MAP_PTR,
11245 .arg3_type = ARG_PTR_TO_MAP_KEY,
11246 .arg4_type = ARG_ANYTHING,
11247 };
11248
BPF_CALL_4(sk_reuseport_load_bytes,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len)11249 BPF_CALL_4(sk_reuseport_load_bytes,
11250 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11251 void *, to, u32, len)
11252 {
11253 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11254 }
11255
11256 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11257 .func = sk_reuseport_load_bytes,
11258 .gpl_only = false,
11259 .ret_type = RET_INTEGER,
11260 .arg1_type = ARG_PTR_TO_CTX,
11261 .arg2_type = ARG_ANYTHING,
11262 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11263 .arg4_type = ARG_CONST_SIZE,
11264 };
11265
BPF_CALL_5(sk_reuseport_load_bytes_relative,const struct sk_reuseport_kern *,reuse_kern,u32,offset,void *,to,u32,len,u32,start_header)11266 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11267 const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11268 void *, to, u32, len, u32, start_header)
11269 {
11270 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11271 len, start_header);
11272 }
11273
11274 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11275 .func = sk_reuseport_load_bytes_relative,
11276 .gpl_only = false,
11277 .ret_type = RET_INTEGER,
11278 .arg1_type = ARG_PTR_TO_CTX,
11279 .arg2_type = ARG_ANYTHING,
11280 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
11281 .arg4_type = ARG_CONST_SIZE,
11282 .arg5_type = ARG_ANYTHING,
11283 };
11284
11285 static const struct bpf_func_proto *
sk_reuseport_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11286 sk_reuseport_func_proto(enum bpf_func_id func_id,
11287 const struct bpf_prog *prog)
11288 {
11289 switch (func_id) {
11290 case BPF_FUNC_sk_select_reuseport:
11291 return &sk_select_reuseport_proto;
11292 case BPF_FUNC_skb_load_bytes:
11293 return &sk_reuseport_load_bytes_proto;
11294 case BPF_FUNC_skb_load_bytes_relative:
11295 return &sk_reuseport_load_bytes_relative_proto;
11296 case BPF_FUNC_get_socket_cookie:
11297 return &bpf_get_socket_ptr_cookie_proto;
11298 case BPF_FUNC_ktime_get_coarse_ns:
11299 return &bpf_ktime_get_coarse_ns_proto;
11300 default:
11301 return bpf_base_func_proto(func_id, prog);
11302 }
11303 }
11304
11305 static bool
sk_reuseport_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)11306 sk_reuseport_is_valid_access(int off, int size,
11307 enum bpf_access_type type,
11308 const struct bpf_prog *prog,
11309 struct bpf_insn_access_aux *info)
11310 {
11311 const u32 size_default = sizeof(__u32);
11312
11313 if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11314 off % size || type != BPF_READ)
11315 return false;
11316
11317 switch (off) {
11318 case offsetof(struct sk_reuseport_md, data):
11319 info->reg_type = PTR_TO_PACKET;
11320 return size == sizeof(__u64);
11321
11322 case offsetof(struct sk_reuseport_md, data_end):
11323 info->reg_type = PTR_TO_PACKET_END;
11324 return size == sizeof(__u64);
11325
11326 case offsetof(struct sk_reuseport_md, hash):
11327 return size == size_default;
11328
11329 case offsetof(struct sk_reuseport_md, sk):
11330 info->reg_type = PTR_TO_SOCKET;
11331 return size == sizeof(__u64);
11332
11333 case offsetof(struct sk_reuseport_md, migrating_sk):
11334 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11335 return size == sizeof(__u64);
11336
11337 /* Fields that allow narrowing */
11338 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11339 if (size < sizeof_field(struct sk_buff, protocol))
11340 return false;
11341 fallthrough;
11342 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11343 case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11344 case bpf_ctx_range(struct sk_reuseport_md, len):
11345 bpf_ctx_record_field_size(info, size_default);
11346 return bpf_ctx_narrow_access_ok(off, size, size_default);
11347
11348 default:
11349 return false;
11350 }
11351 }
11352
11353 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \
11354 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11355 si->dst_reg, si->src_reg, \
11356 bpf_target_off(struct sk_reuseport_kern, F, \
11357 sizeof_field(struct sk_reuseport_kern, F), \
11358 target_size)); \
11359 })
11360
11361 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \
11362 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11363 struct sk_buff, \
11364 skb, \
11365 SKB_FIELD)
11366
11367 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \
11368 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \
11369 struct sock, \
11370 sk, \
11371 SK_FIELD)
11372
sk_reuseport_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)11373 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11374 const struct bpf_insn *si,
11375 struct bpf_insn *insn_buf,
11376 struct bpf_prog *prog,
11377 u32 *target_size)
11378 {
11379 struct bpf_insn *insn = insn_buf;
11380
11381 switch (si->off) {
11382 case offsetof(struct sk_reuseport_md, data):
11383 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11384 break;
11385
11386 case offsetof(struct sk_reuseport_md, len):
11387 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11388 break;
11389
11390 case offsetof(struct sk_reuseport_md, eth_protocol):
11391 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11392 break;
11393
11394 case offsetof(struct sk_reuseport_md, ip_protocol):
11395 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11396 break;
11397
11398 case offsetof(struct sk_reuseport_md, data_end):
11399 SK_REUSEPORT_LOAD_FIELD(data_end);
11400 break;
11401
11402 case offsetof(struct sk_reuseport_md, hash):
11403 SK_REUSEPORT_LOAD_FIELD(hash);
11404 break;
11405
11406 case offsetof(struct sk_reuseport_md, bind_inany):
11407 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11408 break;
11409
11410 case offsetof(struct sk_reuseport_md, sk):
11411 SK_REUSEPORT_LOAD_FIELD(sk);
11412 break;
11413
11414 case offsetof(struct sk_reuseport_md, migrating_sk):
11415 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11416 break;
11417 }
11418
11419 return insn - insn_buf;
11420 }
11421
11422 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11423 .get_func_proto = sk_reuseport_func_proto,
11424 .is_valid_access = sk_reuseport_is_valid_access,
11425 .convert_ctx_access = sk_reuseport_convert_ctx_access,
11426 };
11427
11428 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11429 };
11430
11431 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11432 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11433
BPF_CALL_3(bpf_sk_lookup_assign,struct bpf_sk_lookup_kern *,ctx,struct sock *,sk,u64,flags)11434 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11435 struct sock *, sk, u64, flags)
11436 {
11437 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11438 BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11439 return -EINVAL;
11440 if (unlikely(sk && sk_is_refcounted(sk)))
11441 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11442 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11443 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11444 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11445 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11446
11447 /* Check if socket is suitable for packet L3/L4 protocol */
11448 if (sk && sk->sk_protocol != ctx->protocol)
11449 return -EPROTOTYPE;
11450 if (sk && sk->sk_family != ctx->family &&
11451 (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11452 return -EAFNOSUPPORT;
11453
11454 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11455 return -EEXIST;
11456
11457 /* Select socket as lookup result */
11458 ctx->selected_sk = sk;
11459 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11460 return 0;
11461 }
11462
11463 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11464 .func = bpf_sk_lookup_assign,
11465 .gpl_only = false,
11466 .ret_type = RET_INTEGER,
11467 .arg1_type = ARG_PTR_TO_CTX,
11468 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL,
11469 .arg3_type = ARG_ANYTHING,
11470 };
11471
11472 static const struct bpf_func_proto *
sk_lookup_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11473 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11474 {
11475 switch (func_id) {
11476 case BPF_FUNC_perf_event_output:
11477 return &bpf_event_output_data_proto;
11478 case BPF_FUNC_sk_assign:
11479 return &bpf_sk_lookup_assign_proto;
11480 case BPF_FUNC_sk_release:
11481 return &bpf_sk_release_proto;
11482 default:
11483 return bpf_sk_base_func_proto(func_id, prog);
11484 }
11485 }
11486
sk_lookup_is_valid_access(int off,int size,enum bpf_access_type type,const struct bpf_prog * prog,struct bpf_insn_access_aux * info)11487 static bool sk_lookup_is_valid_access(int off, int size,
11488 enum bpf_access_type type,
11489 const struct bpf_prog *prog,
11490 struct bpf_insn_access_aux *info)
11491 {
11492 if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11493 return false;
11494 if (off % size != 0)
11495 return false;
11496 if (type != BPF_READ)
11497 return false;
11498
11499 switch (off) {
11500 case offsetof(struct bpf_sk_lookup, sk):
11501 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11502 return size == sizeof(__u64);
11503
11504 case bpf_ctx_range(struct bpf_sk_lookup, family):
11505 case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11506 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11507 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11508 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11509 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11510 case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11511 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11512 bpf_ctx_record_field_size(info, sizeof(__u32));
11513 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11514
11515 case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11516 /* Allow 4-byte access to 2-byte field for backward compatibility */
11517 if (size == sizeof(__u32))
11518 return true;
11519 bpf_ctx_record_field_size(info, sizeof(__be16));
11520 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11521
11522 case offsetofend(struct bpf_sk_lookup, remote_port) ...
11523 offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11524 /* Allow access to zero padding for backward compatibility */
11525 bpf_ctx_record_field_size(info, sizeof(__u16));
11526 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11527
11528 default:
11529 return false;
11530 }
11531 }
11532
sk_lookup_convert_ctx_access(enum bpf_access_type type,const struct bpf_insn * si,struct bpf_insn * insn_buf,struct bpf_prog * prog,u32 * target_size)11533 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11534 const struct bpf_insn *si,
11535 struct bpf_insn *insn_buf,
11536 struct bpf_prog *prog,
11537 u32 *target_size)
11538 {
11539 struct bpf_insn *insn = insn_buf;
11540
11541 switch (si->off) {
11542 case offsetof(struct bpf_sk_lookup, sk):
11543 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11544 offsetof(struct bpf_sk_lookup_kern, selected_sk));
11545 break;
11546
11547 case offsetof(struct bpf_sk_lookup, family):
11548 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11549 bpf_target_off(struct bpf_sk_lookup_kern,
11550 family, 2, target_size));
11551 break;
11552
11553 case offsetof(struct bpf_sk_lookup, protocol):
11554 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11555 bpf_target_off(struct bpf_sk_lookup_kern,
11556 protocol, 2, target_size));
11557 break;
11558
11559 case offsetof(struct bpf_sk_lookup, remote_ip4):
11560 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11561 bpf_target_off(struct bpf_sk_lookup_kern,
11562 v4.saddr, 4, target_size));
11563 break;
11564
11565 case offsetof(struct bpf_sk_lookup, local_ip4):
11566 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11567 bpf_target_off(struct bpf_sk_lookup_kern,
11568 v4.daddr, 4, target_size));
11569 break;
11570
11571 case bpf_ctx_range_till(struct bpf_sk_lookup,
11572 remote_ip6[0], remote_ip6[3]): {
11573 #if IS_ENABLED(CONFIG_IPV6)
11574 int off = si->off;
11575
11576 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11577 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11578 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11579 offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11580 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11581 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11582 #else
11583 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11584 #endif
11585 break;
11586 }
11587 case bpf_ctx_range_till(struct bpf_sk_lookup,
11588 local_ip6[0], local_ip6[3]): {
11589 #if IS_ENABLED(CONFIG_IPV6)
11590 int off = si->off;
11591
11592 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11593 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11594 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11595 offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11596 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11597 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11598 #else
11599 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11600 #endif
11601 break;
11602 }
11603 case offsetof(struct bpf_sk_lookup, remote_port):
11604 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11605 bpf_target_off(struct bpf_sk_lookup_kern,
11606 sport, 2, target_size));
11607 break;
11608
11609 case offsetofend(struct bpf_sk_lookup, remote_port):
11610 *target_size = 2;
11611 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11612 break;
11613
11614 case offsetof(struct bpf_sk_lookup, local_port):
11615 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11616 bpf_target_off(struct bpf_sk_lookup_kern,
11617 dport, 2, target_size));
11618 break;
11619
11620 case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11621 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11622 bpf_target_off(struct bpf_sk_lookup_kern,
11623 ingress_ifindex, 4, target_size));
11624 break;
11625 }
11626
11627 return insn - insn_buf;
11628 }
11629
11630 const struct bpf_prog_ops sk_lookup_prog_ops = {
11631 .test_run = bpf_prog_test_run_sk_lookup,
11632 };
11633
11634 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11635 .get_func_proto = sk_lookup_func_proto,
11636 .is_valid_access = sk_lookup_is_valid_access,
11637 .convert_ctx_access = sk_lookup_convert_ctx_access,
11638 };
11639
11640 #endif /* CONFIG_INET */
11641
DEFINE_BPF_DISPATCHER(xdp)11642 DEFINE_BPF_DISPATCHER(xdp)
11643
11644 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11645 {
11646 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11647 }
11648
BTF_ID_LIST_GLOBAL(btf_sock_ids,MAX_BTF_SOCK_TYPE)11649 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11650 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11651 BTF_SOCK_TYPE_xxx
11652 #undef BTF_SOCK_TYPE
11653
11654 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11655 {
11656 /* tcp6_sock type is not generated in dwarf and hence btf,
11657 * trigger an explicit type generation here.
11658 */
11659 BTF_TYPE_EMIT(struct tcp6_sock);
11660 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11661 sk->sk_family == AF_INET6)
11662 return (unsigned long)sk;
11663
11664 return (unsigned long)NULL;
11665 }
11666
11667 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11668 .func = bpf_skc_to_tcp6_sock,
11669 .gpl_only = false,
11670 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11671 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11672 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11673 };
11674
BPF_CALL_1(bpf_skc_to_tcp_sock,struct sock *,sk)11675 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11676 {
11677 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11678 return (unsigned long)sk;
11679
11680 return (unsigned long)NULL;
11681 }
11682
11683 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11684 .func = bpf_skc_to_tcp_sock,
11685 .gpl_only = false,
11686 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11687 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11688 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11689 };
11690
BPF_CALL_1(bpf_skc_to_tcp_timewait_sock,struct sock *,sk)11691 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11692 {
11693 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11694 * generated if CONFIG_INET=n. Trigger an explicit generation here.
11695 */
11696 BTF_TYPE_EMIT(struct inet_timewait_sock);
11697 BTF_TYPE_EMIT(struct tcp_timewait_sock);
11698
11699 #ifdef CONFIG_INET
11700 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11701 return (unsigned long)sk;
11702 #endif
11703
11704 #if IS_BUILTIN(CONFIG_IPV6)
11705 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11706 return (unsigned long)sk;
11707 #endif
11708
11709 return (unsigned long)NULL;
11710 }
11711
11712 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11713 .func = bpf_skc_to_tcp_timewait_sock,
11714 .gpl_only = false,
11715 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11716 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11717 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11718 };
11719
BPF_CALL_1(bpf_skc_to_tcp_request_sock,struct sock *,sk)11720 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11721 {
11722 #ifdef CONFIG_INET
11723 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11724 return (unsigned long)sk;
11725 #endif
11726
11727 #if IS_BUILTIN(CONFIG_IPV6)
11728 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11729 return (unsigned long)sk;
11730 #endif
11731
11732 return (unsigned long)NULL;
11733 }
11734
11735 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11736 .func = bpf_skc_to_tcp_request_sock,
11737 .gpl_only = false,
11738 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11739 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11740 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11741 };
11742
BPF_CALL_1(bpf_skc_to_udp6_sock,struct sock *,sk)11743 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11744 {
11745 /* udp6_sock type is not generated in dwarf and hence btf,
11746 * trigger an explicit type generation here.
11747 */
11748 BTF_TYPE_EMIT(struct udp6_sock);
11749 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11750 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11751 return (unsigned long)sk;
11752
11753 return (unsigned long)NULL;
11754 }
11755
11756 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11757 .func = bpf_skc_to_udp6_sock,
11758 .gpl_only = false,
11759 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11760 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11761 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11762 };
11763
BPF_CALL_1(bpf_skc_to_unix_sock,struct sock *,sk)11764 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11765 {
11766 /* unix_sock type is not generated in dwarf and hence btf,
11767 * trigger an explicit type generation here.
11768 */
11769 BTF_TYPE_EMIT(struct unix_sock);
11770 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11771 return (unsigned long)sk;
11772
11773 return (unsigned long)NULL;
11774 }
11775
11776 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11777 .func = bpf_skc_to_unix_sock,
11778 .gpl_only = false,
11779 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11780 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11781 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11782 };
11783
BPF_CALL_1(bpf_skc_to_mptcp_sock,struct sock *,sk)11784 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11785 {
11786 BTF_TYPE_EMIT(struct mptcp_sock);
11787 return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11788 }
11789
11790 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11791 .func = bpf_skc_to_mptcp_sock,
11792 .gpl_only = false,
11793 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11794 .arg1_type = ARG_PTR_TO_SOCK_COMMON,
11795 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11796 };
11797
BPF_CALL_1(bpf_sock_from_file,struct file *,file)11798 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11799 {
11800 return (unsigned long)sock_from_file(file);
11801 }
11802
11803 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11804 BTF_ID(struct, socket)
11805 BTF_ID(struct, file)
11806
11807 const struct bpf_func_proto bpf_sock_from_file_proto = {
11808 .func = bpf_sock_from_file,
11809 .gpl_only = false,
11810 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL,
11811 .ret_btf_id = &bpf_sock_from_file_btf_ids[0],
11812 .arg1_type = ARG_PTR_TO_BTF_ID,
11813 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1],
11814 };
11815
11816 static const struct bpf_func_proto *
bpf_sk_base_func_proto(enum bpf_func_id func_id,const struct bpf_prog * prog)11817 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11818 {
11819 const struct bpf_func_proto *func;
11820
11821 switch (func_id) {
11822 case BPF_FUNC_skc_to_tcp6_sock:
11823 func = &bpf_skc_to_tcp6_sock_proto;
11824 break;
11825 case BPF_FUNC_skc_to_tcp_sock:
11826 func = &bpf_skc_to_tcp_sock_proto;
11827 break;
11828 case BPF_FUNC_skc_to_tcp_timewait_sock:
11829 func = &bpf_skc_to_tcp_timewait_sock_proto;
11830 break;
11831 case BPF_FUNC_skc_to_tcp_request_sock:
11832 func = &bpf_skc_to_tcp_request_sock_proto;
11833 break;
11834 case BPF_FUNC_skc_to_udp6_sock:
11835 func = &bpf_skc_to_udp6_sock_proto;
11836 break;
11837 case BPF_FUNC_skc_to_unix_sock:
11838 func = &bpf_skc_to_unix_sock_proto;
11839 break;
11840 case BPF_FUNC_skc_to_mptcp_sock:
11841 func = &bpf_skc_to_mptcp_sock_proto;
11842 break;
11843 case BPF_FUNC_ktime_get_coarse_ns:
11844 return &bpf_ktime_get_coarse_ns_proto;
11845 default:
11846 return bpf_base_func_proto(func_id, prog);
11847 }
11848
11849 if (!bpf_token_capable(prog->aux->token, CAP_PERFMON))
11850 return NULL;
11851
11852 return func;
11853 }
11854
11855 __bpf_kfunc_start_defs();
bpf_dynptr_from_skb(struct sk_buff * skb,u64 flags,struct bpf_dynptr_kern * ptr__uninit)11856 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags,
11857 struct bpf_dynptr_kern *ptr__uninit)
11858 {
11859 if (flags) {
11860 bpf_dynptr_set_null(ptr__uninit);
11861 return -EINVAL;
11862 }
11863
11864 bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11865
11866 return 0;
11867 }
11868
bpf_dynptr_from_xdp(struct xdp_buff * xdp,u64 flags,struct bpf_dynptr_kern * ptr__uninit)11869 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags,
11870 struct bpf_dynptr_kern *ptr__uninit)
11871 {
11872 if (flags) {
11873 bpf_dynptr_set_null(ptr__uninit);
11874 return -EINVAL;
11875 }
11876
11877 bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11878
11879 return 0;
11880 }
11881
bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern * sa_kern,const u8 * sun_path,u32 sun_path__sz)11882 __bpf_kfunc int bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern *sa_kern,
11883 const u8 *sun_path, u32 sun_path__sz)
11884 {
11885 struct sockaddr_un *un;
11886
11887 if (sa_kern->sk->sk_family != AF_UNIX)
11888 return -EINVAL;
11889
11890 /* We do not allow changing the address to unnamed or larger than the
11891 * maximum allowed address size for a unix sockaddr.
11892 */
11893 if (sun_path__sz == 0 || sun_path__sz > UNIX_PATH_MAX)
11894 return -EINVAL;
11895
11896 un = (struct sockaddr_un *)sa_kern->uaddr;
11897 memcpy(un->sun_path, sun_path, sun_path__sz);
11898 sa_kern->uaddrlen = offsetof(struct sockaddr_un, sun_path) + sun_path__sz;
11899
11900 return 0;
11901 }
11902
bpf_sk_assign_tcp_reqsk(struct sk_buff * skb,struct sock * sk,struct bpf_tcp_req_attrs * attrs,int attrs__sz)11903 __bpf_kfunc int bpf_sk_assign_tcp_reqsk(struct sk_buff *skb, struct sock *sk,
11904 struct bpf_tcp_req_attrs *attrs, int attrs__sz)
11905 {
11906 #if IS_ENABLED(CONFIG_SYN_COOKIES)
11907 const struct request_sock_ops *ops;
11908 struct inet_request_sock *ireq;
11909 struct tcp_request_sock *treq;
11910 struct request_sock *req;
11911 struct net *net;
11912 __u16 min_mss;
11913 u32 tsoff = 0;
11914
11915 if (attrs__sz != sizeof(*attrs) ||
11916 attrs->reserved[0] || attrs->reserved[1] || attrs->reserved[2])
11917 return -EINVAL;
11918
11919 if (!skb_at_tc_ingress(skb))
11920 return -EINVAL;
11921
11922 net = dev_net(skb->dev);
11923 if (net != sock_net(sk))
11924 return -ENETUNREACH;
11925
11926 switch (skb->protocol) {
11927 case htons(ETH_P_IP):
11928 ops = &tcp_request_sock_ops;
11929 min_mss = 536;
11930 break;
11931 #if IS_BUILTIN(CONFIG_IPV6)
11932 case htons(ETH_P_IPV6):
11933 ops = &tcp6_request_sock_ops;
11934 min_mss = IPV6_MIN_MTU - 60;
11935 break;
11936 #endif
11937 default:
11938 return -EINVAL;
11939 }
11940
11941 if (sk->sk_type != SOCK_STREAM || sk->sk_state != TCP_LISTEN ||
11942 sk_is_mptcp(sk))
11943 return -EINVAL;
11944
11945 if (attrs->mss < min_mss)
11946 return -EINVAL;
11947
11948 if (attrs->wscale_ok) {
11949 if (!READ_ONCE(net->ipv4.sysctl_tcp_window_scaling))
11950 return -EINVAL;
11951
11952 if (attrs->snd_wscale > TCP_MAX_WSCALE ||
11953 attrs->rcv_wscale > TCP_MAX_WSCALE)
11954 return -EINVAL;
11955 }
11956
11957 if (attrs->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
11958 return -EINVAL;
11959
11960 if (attrs->tstamp_ok) {
11961 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
11962 return -EINVAL;
11963
11964 tsoff = attrs->rcv_tsecr - tcp_ns_to_ts(attrs->usec_ts_ok, tcp_clock_ns());
11965 }
11966
11967 req = inet_reqsk_alloc(ops, sk, false);
11968 if (!req)
11969 return -ENOMEM;
11970
11971 ireq = inet_rsk(req);
11972 treq = tcp_rsk(req);
11973
11974 req->rsk_listener = sk;
11975 req->syncookie = 1;
11976 req->mss = attrs->mss;
11977 req->ts_recent = attrs->rcv_tsval;
11978
11979 ireq->snd_wscale = attrs->snd_wscale;
11980 ireq->rcv_wscale = attrs->rcv_wscale;
11981 ireq->tstamp_ok = !!attrs->tstamp_ok;
11982 ireq->sack_ok = !!attrs->sack_ok;
11983 ireq->wscale_ok = !!attrs->wscale_ok;
11984 ireq->ecn_ok = !!attrs->ecn_ok;
11985
11986 treq->req_usec_ts = !!attrs->usec_ts_ok;
11987 treq->ts_off = tsoff;
11988
11989 skb_orphan(skb);
11990 skb->sk = req_to_sk(req);
11991 skb->destructor = sock_pfree;
11992
11993 return 0;
11994 #else
11995 return -EOPNOTSUPP;
11996 #endif
11997 }
11998
11999 __bpf_kfunc_end_defs();
12000
bpf_dynptr_from_skb_rdonly(struct sk_buff * skb,u64 flags,struct bpf_dynptr_kern * ptr__uninit)12001 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
12002 struct bpf_dynptr_kern *ptr__uninit)
12003 {
12004 int err;
12005
12006 err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
12007 if (err)
12008 return err;
12009
12010 bpf_dynptr_set_rdonly(ptr__uninit);
12011
12012 return 0;
12013 }
12014
12015 BTF_KFUNCS_START(bpf_kfunc_check_set_skb)
12016 BTF_ID_FLAGS(func, bpf_dynptr_from_skb)
12017 BTF_KFUNCS_END(bpf_kfunc_check_set_skb)
12018
12019 BTF_KFUNCS_START(bpf_kfunc_check_set_xdp)
12020 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
12021 BTF_KFUNCS_END(bpf_kfunc_check_set_xdp)
12022
12023 BTF_KFUNCS_START(bpf_kfunc_check_set_sock_addr)
12024 BTF_ID_FLAGS(func, bpf_sock_addr_set_sun_path)
12025 BTF_KFUNCS_END(bpf_kfunc_check_set_sock_addr)
12026
12027 BTF_KFUNCS_START(bpf_kfunc_check_set_tcp_reqsk)
12028 BTF_ID_FLAGS(func, bpf_sk_assign_tcp_reqsk, KF_TRUSTED_ARGS)
12029 BTF_KFUNCS_END(bpf_kfunc_check_set_tcp_reqsk)
12030
12031 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
12032 .owner = THIS_MODULE,
12033 .set = &bpf_kfunc_check_set_skb,
12034 };
12035
12036 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
12037 .owner = THIS_MODULE,
12038 .set = &bpf_kfunc_check_set_xdp,
12039 };
12040
12041 static const struct btf_kfunc_id_set bpf_kfunc_set_sock_addr = {
12042 .owner = THIS_MODULE,
12043 .set = &bpf_kfunc_check_set_sock_addr,
12044 };
12045
12046 static const struct btf_kfunc_id_set bpf_kfunc_set_tcp_reqsk = {
12047 .owner = THIS_MODULE,
12048 .set = &bpf_kfunc_check_set_tcp_reqsk,
12049 };
12050
bpf_kfunc_init(void)12051 static int __init bpf_kfunc_init(void)
12052 {
12053 int ret;
12054
12055 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
12056 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
12057 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
12058 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
12059 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
12060 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
12061 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
12062 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
12063 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
12064 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
12065 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
12066 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
12067 &bpf_kfunc_set_sock_addr);
12068 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_tcp_reqsk);
12069 }
12070 late_initcall(bpf_kfunc_init);
12071
12072 __bpf_kfunc_start_defs();
12073
12074 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
12075 *
12076 * The function expects a non-NULL pointer to a socket, and invokes the
12077 * protocol specific socket destroy handlers.
12078 *
12079 * The helper can only be called from BPF contexts that have acquired the socket
12080 * locks.
12081 *
12082 * Parameters:
12083 * @sock: Pointer to socket to be destroyed
12084 *
12085 * Return:
12086 * On error, may return EPROTONOSUPPORT, EINVAL.
12087 * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
12088 * 0 otherwise
12089 */
bpf_sock_destroy(struct sock_common * sock)12090 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
12091 {
12092 struct sock *sk = (struct sock *)sock;
12093
12094 /* The locking semantics that allow for synchronous execution of the
12095 * destroy handlers are only supported for TCP and UDP.
12096 * Supporting protocols will need to acquire sock lock in the BPF context
12097 * prior to invoking this kfunc.
12098 */
12099 if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
12100 sk->sk_protocol != IPPROTO_UDP))
12101 return -EOPNOTSUPP;
12102
12103 return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
12104 }
12105
12106 __bpf_kfunc_end_defs();
12107
12108 BTF_KFUNCS_START(bpf_sk_iter_kfunc_ids)
BTF_ID_FLAGS(func,bpf_sock_destroy,KF_TRUSTED_ARGS)12109 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
12110 BTF_KFUNCS_END(bpf_sk_iter_kfunc_ids)
12111
12112 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
12113 {
12114 if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
12115 prog->expected_attach_type != BPF_TRACE_ITER)
12116 return -EACCES;
12117 return 0;
12118 }
12119
12120 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
12121 .owner = THIS_MODULE,
12122 .set = &bpf_sk_iter_kfunc_ids,
12123 .filter = tracing_iter_filter,
12124 };
12125
init_subsystem(void)12126 static int init_subsystem(void)
12127 {
12128 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
12129 }
12130 late_initcall(init_subsystem);
12131