1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <gw4pts@gw4pts.ampr.org> 7 * Florian La Roche, <rzsfl@rz.uni-sb.de> 8 */ 9 10 #ifndef _LINUX_SKBUFF_H 11 #define _LINUX_SKBUFF_H 12 13 #include <linux/kernel.h> 14 #include <linux/compiler.h> 15 #include <linux/time.h> 16 #include <linux/bug.h> 17 #include <linux/bvec.h> 18 #include <linux/cache.h> 19 #include <linux/rbtree.h> 20 #include <linux/socket.h> 21 #include <linux/refcount.h> 22 23 #include <linux/atomic.h> 24 #include <asm/types.h> 25 #include <linux/spinlock.h> 26 #include <net/checksum.h> 27 #include <linux/rcupdate.h> 28 #include <linux/dma-mapping.h> 29 #include <linux/netdev_features.h> 30 #include <net/flow_dissector.h> 31 #include <linux/in6.h> 32 #include <linux/if_packet.h> 33 #include <linux/llist.h> 34 #include <net/flow.h> 35 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 36 #include <linux/netfilter/nf_conntrack_common.h> 37 #endif 38 #include <net/net_debug.h> 39 #include <net/dropreason-core.h> 40 41 /** 42 * DOC: skb checksums 43 * 44 * The interface for checksum offload between the stack and networking drivers 45 * is as follows... 46 * 47 * IP checksum related features 48 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 49 * 50 * Drivers advertise checksum offload capabilities in the features of a device. 51 * From the stack's point of view these are capabilities offered by the driver. 52 * A driver typically only advertises features that it is capable of offloading 53 * to its device. 54 * 55 * .. flat-table:: Checksum related device features 56 * :widths: 1 10 57 * 58 * * - %NETIF_F_HW_CSUM 59 * - The driver (or its device) is able to compute one 60 * IP (one's complement) checksum for any combination 61 * of protocols or protocol layering. The checksum is 62 * computed and set in a packet per the CHECKSUM_PARTIAL 63 * interface (see below). 64 * 65 * * - %NETIF_F_IP_CSUM 66 * - Driver (device) is only able to checksum plain 67 * TCP or UDP packets over IPv4. These are specifically 68 * unencapsulated packets of the form IPv4|TCP or 69 * IPv4|UDP where the Protocol field in the IPv4 header 70 * is TCP or UDP. The IPv4 header may contain IP options. 71 * This feature cannot be set in features for a device 72 * with NETIF_F_HW_CSUM also set. This feature is being 73 * DEPRECATED (see below). 74 * 75 * * - %NETIF_F_IPV6_CSUM 76 * - Driver (device) is only able to checksum plain 77 * TCP or UDP packets over IPv6. These are specifically 78 * unencapsulated packets of the form IPv6|TCP or 79 * IPv6|UDP where the Next Header field in the IPv6 80 * header is either TCP or UDP. IPv6 extension headers 81 * are not supported with this feature. This feature 82 * cannot be set in features for a device with 83 * NETIF_F_HW_CSUM also set. This feature is being 84 * DEPRECATED (see below). 85 * 86 * * - %NETIF_F_RXCSUM 87 * - Driver (device) performs receive checksum offload. 88 * This flag is only used to disable the RX checksum 89 * feature for a device. The stack will accept receive 90 * checksum indication in packets received on a device 91 * regardless of whether NETIF_F_RXCSUM is set. 92 * 93 * Checksumming of received packets by device 94 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 95 * 96 * Indication of checksum verification is set in &sk_buff.ip_summed. 97 * Possible values are: 98 * 99 * - %CHECKSUM_NONE 100 * 101 * Device did not checksum this packet e.g. due to lack of capabilities. 102 * The packet contains full (though not verified) checksum in packet but 103 * not in skb->csum. Thus, skb->csum is undefined in this case. 104 * 105 * - %CHECKSUM_UNNECESSARY 106 * 107 * The hardware you're dealing with doesn't calculate the full checksum 108 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 109 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 110 * if their checksums are okay. &sk_buff.csum is still undefined in this case 111 * though. A driver or device must never modify the checksum field in the 112 * packet even if checksum is verified. 113 * 114 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 115 * 116 * - TCP: IPv6 and IPv4. 117 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 118 * zero UDP checksum for either IPv4 or IPv6, the networking stack 119 * may perform further validation in this case. 120 * - GRE: only if the checksum is present in the header. 121 * - SCTP: indicates the CRC in SCTP header has been validated. 122 * - FCOE: indicates the CRC in FC frame has been validated. 123 * 124 * &sk_buff.csum_level indicates the number of consecutive checksums found in 125 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 126 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 127 * and a device is able to verify the checksums for UDP (possibly zero), 128 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 129 * two. If the device were only able to verify the UDP checksum and not 130 * GRE, either because it doesn't support GRE checksum or because GRE 131 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 132 * not considered in this case). 133 * 134 * - %CHECKSUM_COMPLETE 135 * 136 * This is the most generic way. The device supplied checksum of the _whole_ 137 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 138 * hardware doesn't need to parse L3/L4 headers to implement this. 139 * 140 * Notes: 141 * 142 * - Even if device supports only some protocols, but is able to produce 143 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 144 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 145 * 146 * - %CHECKSUM_PARTIAL 147 * 148 * A checksum is set up to be offloaded to a device as described in the 149 * output description for CHECKSUM_PARTIAL. This may occur on a packet 150 * received directly from another Linux OS, e.g., a virtualized Linux kernel 151 * on the same host, or it may be set in the input path in GRO or remote 152 * checksum offload. For the purposes of checksum verification, the checksum 153 * referred to by skb->csum_start + skb->csum_offset and any preceding 154 * checksums in the packet are considered verified. Any checksums in the 155 * packet that are after the checksum being offloaded are not considered to 156 * be verified. 157 * 158 * Checksumming on transmit for non-GSO 159 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 160 * 161 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 162 * Values are: 163 * 164 * - %CHECKSUM_PARTIAL 165 * 166 * The driver is required to checksum the packet as seen by hard_start_xmit() 167 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 168 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 169 * A driver may verify that the 170 * csum_start and csum_offset values are valid values given the length and 171 * offset of the packet, but it should not attempt to validate that the 172 * checksum refers to a legitimate transport layer checksum -- it is the 173 * purview of the stack to validate that csum_start and csum_offset are set 174 * correctly. 175 * 176 * When the stack requests checksum offload for a packet, the driver MUST 177 * ensure that the checksum is set correctly. A driver can either offload the 178 * checksum calculation to the device, or call skb_checksum_help (in the case 179 * that the device does not support offload for a particular checksum). 180 * 181 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 182 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 183 * checksum offload capability. 184 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 185 * on network device checksumming capabilities: if a packet does not match 186 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 187 * &sk_buff.csum_not_inet, see :ref:`crc`) 188 * is called to resolve the checksum. 189 * 190 * - %CHECKSUM_NONE 191 * 192 * The skb was already checksummed by the protocol, or a checksum is not 193 * required. 194 * 195 * - %CHECKSUM_UNNECESSARY 196 * 197 * This has the same meaning as CHECKSUM_NONE for checksum offload on 198 * output. 199 * 200 * - %CHECKSUM_COMPLETE 201 * 202 * Not used in checksum output. If a driver observes a packet with this value 203 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 204 * 205 * .. _crc: 206 * 207 * Non-IP checksum (CRC) offloads 208 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 209 * 210 * .. flat-table:: 211 * :widths: 1 10 212 * 213 * * - %NETIF_F_SCTP_CRC 214 * - This feature indicates that a device is capable of 215 * offloading the SCTP CRC in a packet. To perform this offload the stack 216 * will set csum_start and csum_offset accordingly, set ip_summed to 217 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 218 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 219 * A driver that supports both IP checksum offload and SCTP CRC32c offload 220 * must verify which offload is configured for a packet by testing the 221 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 222 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 223 * 224 * * - %NETIF_F_FCOE_CRC 225 * - This feature indicates that a device is capable of offloading the FCOE 226 * CRC in a packet. To perform this offload the stack will set ip_summed 227 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 228 * accordingly. Note that there is no indication in the skbuff that the 229 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 230 * both IP checksum offload and FCOE CRC offload must verify which offload 231 * is configured for a packet, presumably by inspecting packet headers. 232 * 233 * Checksumming on output with GSO 234 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 235 * 236 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 237 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 238 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 239 * part of the GSO operation is implied. If a checksum is being offloaded 240 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 241 * csum_offset are set to refer to the outermost checksum being offloaded 242 * (two offloaded checksums are possible with UDP encapsulation). 243 */ 244 245 /* Don't change this without changing skb_csum_unnecessary! */ 246 #define CHECKSUM_NONE 0 247 #define CHECKSUM_UNNECESSARY 1 248 #define CHECKSUM_COMPLETE 2 249 #define CHECKSUM_PARTIAL 3 250 251 /* Maximum value in skb->csum_level */ 252 #define SKB_MAX_CSUM_LEVEL 3 253 254 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 255 #define SKB_WITH_OVERHEAD(X) \ 256 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 257 258 /* For X bytes available in skb->head, what is the minimal 259 * allocation needed, knowing struct skb_shared_info needs 260 * to be aligned. 261 */ 262 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \ 263 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 264 265 #define SKB_MAX_ORDER(X, ORDER) \ 266 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 267 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 268 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 269 270 /* return minimum truesize of one skb containing X bytes of data */ 271 #define SKB_TRUESIZE(X) ((X) + \ 272 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 273 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 274 275 struct ahash_request; 276 struct net_device; 277 struct scatterlist; 278 struct pipe_inode_info; 279 struct iov_iter; 280 struct napi_struct; 281 struct bpf_prog; 282 union bpf_attr; 283 struct skb_ext; 284 struct ts_config; 285 286 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 287 struct nf_bridge_info { 288 enum { 289 BRNF_PROTO_UNCHANGED, 290 BRNF_PROTO_8021Q, 291 BRNF_PROTO_PPPOE 292 } orig_proto:8; 293 u8 pkt_otherhost:1; 294 u8 in_prerouting:1; 295 u8 bridged_dnat:1; 296 u8 sabotage_in_done:1; 297 __u16 frag_max_size; 298 int physinif; 299 300 /* always valid & non-NULL from FORWARD on, for physdev match */ 301 struct net_device *physoutdev; 302 union { 303 /* prerouting: detect dnat in orig/reply direction */ 304 __be32 ipv4_daddr; 305 struct in6_addr ipv6_daddr; 306 307 /* after prerouting + nat detected: store original source 308 * mac since neigh resolution overwrites it, only used while 309 * skb is out in neigh layer. 310 */ 311 char neigh_header[8]; 312 }; 313 }; 314 #endif 315 316 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 317 /* Chain in tc_skb_ext will be used to share the tc chain with 318 * ovs recirc_id. It will be set to the current chain by tc 319 * and read by ovs to recirc_id. 320 */ 321 struct tc_skb_ext { 322 union { 323 u64 act_miss_cookie; 324 __u32 chain; 325 }; 326 __u16 mru; 327 __u16 zone; 328 u8 post_ct:1; 329 u8 post_ct_snat:1; 330 u8 post_ct_dnat:1; 331 u8 act_miss:1; /* Set if act_miss_cookie is used */ 332 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */ 333 }; 334 #endif 335 336 struct sk_buff_head { 337 /* These two members must be first to match sk_buff. */ 338 struct_group_tagged(sk_buff_list, list, 339 struct sk_buff *next; 340 struct sk_buff *prev; 341 ); 342 343 __u32 qlen; 344 spinlock_t lock; 345 }; 346 347 struct sk_buff; 348 349 #ifndef CONFIG_MAX_SKB_FRAGS 350 # define CONFIG_MAX_SKB_FRAGS 17 351 #endif 352 353 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS 354 355 extern int sysctl_max_skb_frags; 356 357 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 358 * segment using its current segmentation instead. 359 */ 360 #define GSO_BY_FRAGS 0xFFFF 361 362 typedef struct bio_vec skb_frag_t; 363 364 /** 365 * skb_frag_size() - Returns the size of a skb fragment 366 * @frag: skb fragment 367 */ 368 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 369 { 370 return frag->bv_len; 371 } 372 373 /** 374 * skb_frag_size_set() - Sets the size of a skb fragment 375 * @frag: skb fragment 376 * @size: size of fragment 377 */ 378 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 379 { 380 frag->bv_len = size; 381 } 382 383 /** 384 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 385 * @frag: skb fragment 386 * @delta: value to add 387 */ 388 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 389 { 390 frag->bv_len += delta; 391 } 392 393 /** 394 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 395 * @frag: skb fragment 396 * @delta: value to subtract 397 */ 398 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 399 { 400 frag->bv_len -= delta; 401 } 402 403 /** 404 * skb_frag_must_loop - Test if %p is a high memory page 405 * @p: fragment's page 406 */ 407 static inline bool skb_frag_must_loop(struct page *p) 408 { 409 #if defined(CONFIG_HIGHMEM) 410 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 411 return true; 412 #endif 413 return false; 414 } 415 416 /** 417 * skb_frag_foreach_page - loop over pages in a fragment 418 * 419 * @f: skb frag to operate on 420 * @f_off: offset from start of f->bv_page 421 * @f_len: length from f_off to loop over 422 * @p: (temp var) current page 423 * @p_off: (temp var) offset from start of current page, 424 * non-zero only on first page. 425 * @p_len: (temp var) length in current page, 426 * < PAGE_SIZE only on first and last page. 427 * @copied: (temp var) length so far, excluding current p_len. 428 * 429 * A fragment can hold a compound page, in which case per-page 430 * operations, notably kmap_atomic, must be called for each 431 * regular page. 432 */ 433 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 434 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 435 p_off = (f_off) & (PAGE_SIZE - 1), \ 436 p_len = skb_frag_must_loop(p) ? \ 437 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 438 copied = 0; \ 439 copied < f_len; \ 440 copied += p_len, p++, p_off = 0, \ 441 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 442 443 /** 444 * struct skb_shared_hwtstamps - hardware time stamps 445 * @hwtstamp: hardware time stamp transformed into duration 446 * since arbitrary point in time 447 * @netdev_data: address/cookie of network device driver used as 448 * reference to actual hardware time stamp 449 * 450 * Software time stamps generated by ktime_get_real() are stored in 451 * skb->tstamp. 452 * 453 * hwtstamps can only be compared against other hwtstamps from 454 * the same device. 455 * 456 * This structure is attached to packets as part of the 457 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 458 */ 459 struct skb_shared_hwtstamps { 460 union { 461 ktime_t hwtstamp; 462 void *netdev_data; 463 }; 464 }; 465 466 /* Definitions for tx_flags in struct skb_shared_info */ 467 enum { 468 /* generate hardware time stamp */ 469 SKBTX_HW_TSTAMP = 1 << 0, 470 471 /* generate software time stamp when queueing packet to NIC */ 472 SKBTX_SW_TSTAMP = 1 << 1, 473 474 /* device driver is going to provide hardware time stamp */ 475 SKBTX_IN_PROGRESS = 1 << 2, 476 477 /* generate hardware time stamp based on cycles if supported */ 478 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3, 479 480 /* generate wifi status information (where possible) */ 481 SKBTX_WIFI_STATUS = 1 << 4, 482 483 /* determine hardware time stamp based on time or cycles */ 484 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 485 486 /* generate software time stamp when entering packet scheduling */ 487 SKBTX_SCHED_TSTAMP = 1 << 6, 488 }; 489 490 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 491 SKBTX_SCHED_TSTAMP) 492 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 493 SKBTX_HW_TSTAMP_USE_CYCLES | \ 494 SKBTX_ANY_SW_TSTAMP) 495 496 /* Definitions for flags in struct skb_shared_info */ 497 enum { 498 /* use zcopy routines */ 499 SKBFL_ZEROCOPY_ENABLE = BIT(0), 500 501 /* This indicates at least one fragment might be overwritten 502 * (as in vmsplice(), sendfile() ...) 503 * If we need to compute a TX checksum, we'll need to copy 504 * all frags to avoid possible bad checksum 505 */ 506 SKBFL_SHARED_FRAG = BIT(1), 507 508 /* segment contains only zerocopy data and should not be 509 * charged to the kernel memory. 510 */ 511 SKBFL_PURE_ZEROCOPY = BIT(2), 512 513 SKBFL_DONT_ORPHAN = BIT(3), 514 515 /* page references are managed by the ubuf_info, so it's safe to 516 * use frags only up until ubuf_info is released 517 */ 518 SKBFL_MANAGED_FRAG_REFS = BIT(4), 519 }; 520 521 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 522 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 523 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 524 525 /* 526 * The callback notifies userspace to release buffers when skb DMA is done in 527 * lower device, the skb last reference should be 0 when calling this. 528 * The zerocopy_success argument is true if zero copy transmit occurred, 529 * false on data copy or out of memory error caused by data copy attempt. 530 * The ctx field is used to track device context. 531 * The desc field is used to track userspace buffer index. 532 */ 533 struct ubuf_info { 534 void (*callback)(struct sk_buff *, struct ubuf_info *, 535 bool zerocopy_success); 536 refcount_t refcnt; 537 u8 flags; 538 }; 539 540 struct ubuf_info_msgzc { 541 struct ubuf_info ubuf; 542 543 union { 544 struct { 545 unsigned long desc; 546 void *ctx; 547 }; 548 struct { 549 u32 id; 550 u16 len; 551 u16 zerocopy:1; 552 u32 bytelen; 553 }; 554 }; 555 556 struct mmpin { 557 struct user_struct *user; 558 unsigned int num_pg; 559 } mmp; 560 }; 561 562 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 563 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 564 ubuf) 565 566 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 567 void mm_unaccount_pinned_pages(struct mmpin *mmp); 568 569 /* Preserve some data across TX submission and completion. 570 * 571 * Note, this state is stored in the driver. Extending the layout 572 * might need some special care. 573 */ 574 struct xsk_tx_metadata_compl { 575 __u64 *tx_timestamp; 576 }; 577 578 /* This data is invariant across clones and lives at 579 * the end of the header data, ie. at skb->end. 580 */ 581 struct skb_shared_info { 582 __u8 flags; 583 __u8 meta_len; 584 __u8 nr_frags; 585 __u8 tx_flags; 586 unsigned short gso_size; 587 /* Warning: this field is not always filled in (UFO)! */ 588 unsigned short gso_segs; 589 struct sk_buff *frag_list; 590 union { 591 struct skb_shared_hwtstamps hwtstamps; 592 struct xsk_tx_metadata_compl xsk_meta; 593 }; 594 unsigned int gso_type; 595 u32 tskey; 596 597 /* 598 * Warning : all fields before dataref are cleared in __alloc_skb() 599 */ 600 atomic_t dataref; 601 unsigned int xdp_frags_size; 602 603 /* Intermediate layers must ensure that destructor_arg 604 * remains valid until skb destructor */ 605 void * destructor_arg; 606 607 /* must be last field, see pskb_expand_head() */ 608 skb_frag_t frags[MAX_SKB_FRAGS]; 609 }; 610 611 /** 612 * DOC: dataref and headerless skbs 613 * 614 * Transport layers send out clones of payload skbs they hold for 615 * retransmissions. To allow lower layers of the stack to prepend their headers 616 * we split &skb_shared_info.dataref into two halves. 617 * The lower 16 bits count the overall number of references. 618 * The higher 16 bits indicate how many of the references are payload-only. 619 * skb_header_cloned() checks if skb is allowed to add / write the headers. 620 * 621 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 622 * (via __skb_header_release()). Any clone created from marked skb will get 623 * &sk_buff.hdr_len populated with the available headroom. 624 * If there's the only clone in existence it's able to modify the headroom 625 * at will. The sequence of calls inside the transport layer is:: 626 * 627 * <alloc skb> 628 * skb_reserve() 629 * __skb_header_release() 630 * skb_clone() 631 * // send the clone down the stack 632 * 633 * This is not a very generic construct and it depends on the transport layers 634 * doing the right thing. In practice there's usually only one payload-only skb. 635 * Having multiple payload-only skbs with different lengths of hdr_len is not 636 * possible. The payload-only skbs should never leave their owner. 637 */ 638 #define SKB_DATAREF_SHIFT 16 639 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 640 641 642 enum { 643 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 644 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 645 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 646 }; 647 648 enum { 649 SKB_GSO_TCPV4 = 1 << 0, 650 651 /* This indicates the skb is from an untrusted source. */ 652 SKB_GSO_DODGY = 1 << 1, 653 654 /* This indicates the tcp segment has CWR set. */ 655 SKB_GSO_TCP_ECN = 1 << 2, 656 657 SKB_GSO_TCP_FIXEDID = 1 << 3, 658 659 SKB_GSO_TCPV6 = 1 << 4, 660 661 SKB_GSO_FCOE = 1 << 5, 662 663 SKB_GSO_GRE = 1 << 6, 664 665 SKB_GSO_GRE_CSUM = 1 << 7, 666 667 SKB_GSO_IPXIP4 = 1 << 8, 668 669 SKB_GSO_IPXIP6 = 1 << 9, 670 671 SKB_GSO_UDP_TUNNEL = 1 << 10, 672 673 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 674 675 SKB_GSO_PARTIAL = 1 << 12, 676 677 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 678 679 SKB_GSO_SCTP = 1 << 14, 680 681 SKB_GSO_ESP = 1 << 15, 682 683 SKB_GSO_UDP = 1 << 16, 684 685 SKB_GSO_UDP_L4 = 1 << 17, 686 687 SKB_GSO_FRAGLIST = 1 << 18, 688 }; 689 690 #if BITS_PER_LONG > 32 691 #define NET_SKBUFF_DATA_USES_OFFSET 1 692 #endif 693 694 #ifdef NET_SKBUFF_DATA_USES_OFFSET 695 typedef unsigned int sk_buff_data_t; 696 #else 697 typedef unsigned char *sk_buff_data_t; 698 #endif 699 700 /** 701 * DOC: Basic sk_buff geometry 702 * 703 * struct sk_buff itself is a metadata structure and does not hold any packet 704 * data. All the data is held in associated buffers. 705 * 706 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 707 * into two parts: 708 * 709 * - data buffer, containing headers and sometimes payload; 710 * this is the part of the skb operated on by the common helpers 711 * such as skb_put() or skb_pull(); 712 * - shared info (struct skb_shared_info) which holds an array of pointers 713 * to read-only data in the (page, offset, length) format. 714 * 715 * Optionally &skb_shared_info.frag_list may point to another skb. 716 * 717 * Basic diagram may look like this:: 718 * 719 * --------------- 720 * | sk_buff | 721 * --------------- 722 * ,--------------------------- + head 723 * / ,----------------- + data 724 * / / ,----------- + tail 725 * | | | , + end 726 * | | | | 727 * v v v v 728 * ----------------------------------------------- 729 * | headroom | data | tailroom | skb_shared_info | 730 * ----------------------------------------------- 731 * + [page frag] 732 * + [page frag] 733 * + [page frag] 734 * + [page frag] --------- 735 * + frag_list --> | sk_buff | 736 * --------- 737 * 738 */ 739 740 /** 741 * struct sk_buff - socket buffer 742 * @next: Next buffer in list 743 * @prev: Previous buffer in list 744 * @tstamp: Time we arrived/left 745 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 746 * for retransmit timer 747 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 748 * @list: queue head 749 * @ll_node: anchor in an llist (eg socket defer_list) 750 * @sk: Socket we are owned by 751 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in 752 * fragmentation management 753 * @dev: Device we arrived on/are leaving by 754 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 755 * @cb: Control buffer. Free for use by every layer. Put private vars here 756 * @_skb_refdst: destination entry (with norefcount bit) 757 * @len: Length of actual data 758 * @data_len: Data length 759 * @mac_len: Length of link layer header 760 * @hdr_len: writable header length of cloned skb 761 * @csum: Checksum (must include start/offset pair) 762 * @csum_start: Offset from skb->head where checksumming should start 763 * @csum_offset: Offset from csum_start where checksum should be stored 764 * @priority: Packet queueing priority 765 * @ignore_df: allow local fragmentation 766 * @cloned: Head may be cloned (check refcnt to be sure) 767 * @ip_summed: Driver fed us an IP checksum 768 * @nohdr: Payload reference only, must not modify header 769 * @pkt_type: Packet class 770 * @fclone: skbuff clone status 771 * @ipvs_property: skbuff is owned by ipvs 772 * @inner_protocol_type: whether the inner protocol is 773 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 774 * @remcsum_offload: remote checksum offload is enabled 775 * @offload_fwd_mark: Packet was L2-forwarded in hardware 776 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 777 * @tc_skip_classify: do not classify packet. set by IFB device 778 * @tc_at_ingress: used within tc_classify to distinguish in/egress 779 * @redirected: packet was redirected by packet classifier 780 * @from_ingress: packet was redirected from the ingress path 781 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 782 * @peeked: this packet has been seen already, so stats have been 783 * done for it, don't do them again 784 * @nf_trace: netfilter packet trace flag 785 * @protocol: Packet protocol from driver 786 * @destructor: Destruct function 787 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 788 * @_sk_redir: socket redirection information for skmsg 789 * @_nfct: Associated connection, if any (with nfctinfo bits) 790 * @skb_iif: ifindex of device we arrived on 791 * @tc_index: Traffic control index 792 * @hash: the packet hash 793 * @queue_mapping: Queue mapping for multiqueue devices 794 * @head_frag: skb was allocated from page fragments, 795 * not allocated by kmalloc() or vmalloc(). 796 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 797 * @pp_recycle: mark the packet for recycling instead of freeing (implies 798 * page_pool support on driver) 799 * @active_extensions: active extensions (skb_ext_id types) 800 * @ndisc_nodetype: router type (from link layer) 801 * @ooo_okay: allow the mapping of a socket to a queue to be changed 802 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 803 * ports. 804 * @sw_hash: indicates hash was computed in software stack 805 * @wifi_acked_valid: wifi_acked was set 806 * @wifi_acked: whether frame was acked on wifi or not 807 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 808 * @encapsulation: indicates the inner headers in the skbuff are valid 809 * @encap_hdr_csum: software checksum is needed 810 * @csum_valid: checksum is already valid 811 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 812 * @csum_complete_sw: checksum was completed by software 813 * @csum_level: indicates the number of consecutive checksums found in 814 * the packet minus one that have been verified as 815 * CHECKSUM_UNNECESSARY (max 3) 816 * @dst_pending_confirm: need to confirm neighbour 817 * @decrypted: Decrypted SKB 818 * @slow_gro: state present at GRO time, slower prepare step required 819 * @mono_delivery_time: When set, skb->tstamp has the 820 * delivery_time in mono clock base (i.e. EDT). Otherwise, the 821 * skb->tstamp has the (rcv) timestamp at ingress and 822 * delivery_time at egress. 823 * @napi_id: id of the NAPI struct this skb came from 824 * @sender_cpu: (aka @napi_id) source CPU in XPS 825 * @alloc_cpu: CPU which did the skb allocation. 826 * @secmark: security marking 827 * @mark: Generic packet mark 828 * @reserved_tailroom: (aka @mark) number of bytes of free space available 829 * at the tail of an sk_buff 830 * @vlan_all: vlan fields (proto & tci) 831 * @vlan_proto: vlan encapsulation protocol 832 * @vlan_tci: vlan tag control information 833 * @inner_protocol: Protocol (encapsulation) 834 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 835 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 836 * @inner_transport_header: Inner transport layer header (encapsulation) 837 * @inner_network_header: Network layer header (encapsulation) 838 * @inner_mac_header: Link layer header (encapsulation) 839 * @transport_header: Transport layer header 840 * @network_header: Network layer header 841 * @mac_header: Link layer header 842 * @kcov_handle: KCOV remote handle for remote coverage collection 843 * @tail: Tail pointer 844 * @end: End pointer 845 * @head: Head of buffer 846 * @data: Data head pointer 847 * @truesize: Buffer size 848 * @users: User count - see {datagram,tcp}.c 849 * @extensions: allocated extensions, valid if active_extensions is nonzero 850 */ 851 852 struct sk_buff { 853 union { 854 struct { 855 /* These two members must be first to match sk_buff_head. */ 856 struct sk_buff *next; 857 struct sk_buff *prev; 858 859 union { 860 struct net_device *dev; 861 /* Some protocols might use this space to store information, 862 * while device pointer would be NULL. 863 * UDP receive path is one user. 864 */ 865 unsigned long dev_scratch; 866 }; 867 }; 868 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 869 struct list_head list; 870 struct llist_node ll_node; 871 }; 872 873 union { 874 struct sock *sk; 875 int ip_defrag_offset; 876 }; 877 878 union { 879 ktime_t tstamp; 880 u64 skb_mstamp_ns; /* earliest departure time */ 881 }; 882 /* 883 * This is the control buffer. It is free to use for every 884 * layer. Please put your private variables there. If you 885 * want to keep them across layers you have to do a skb_clone() 886 * first. This is owned by whoever has the skb queued ATM. 887 */ 888 char cb[48] __aligned(8); 889 890 union { 891 struct { 892 unsigned long _skb_refdst; 893 void (*destructor)(struct sk_buff *skb); 894 }; 895 struct list_head tcp_tsorted_anchor; 896 #ifdef CONFIG_NET_SOCK_MSG 897 unsigned long _sk_redir; 898 #endif 899 }; 900 901 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 902 unsigned long _nfct; 903 #endif 904 unsigned int len, 905 data_len; 906 __u16 mac_len, 907 hdr_len; 908 909 /* Following fields are _not_ copied in __copy_skb_header() 910 * Note that queue_mapping is here mostly to fill a hole. 911 */ 912 __u16 queue_mapping; 913 914 /* if you move cloned around you also must adapt those constants */ 915 #ifdef __BIG_ENDIAN_BITFIELD 916 #define CLONED_MASK (1 << 7) 917 #else 918 #define CLONED_MASK 1 919 #endif 920 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 921 922 /* private: */ 923 __u8 __cloned_offset[0]; 924 /* public: */ 925 __u8 cloned:1, 926 nohdr:1, 927 fclone:2, 928 peeked:1, 929 head_frag:1, 930 pfmemalloc:1, 931 pp_recycle:1; /* page_pool recycle indicator */ 932 #ifdef CONFIG_SKB_EXTENSIONS 933 __u8 active_extensions; 934 #endif 935 936 /* Fields enclosed in headers group are copied 937 * using a single memcpy() in __copy_skb_header() 938 */ 939 struct_group(headers, 940 941 /* private: */ 942 __u8 __pkt_type_offset[0]; 943 /* public: */ 944 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 945 __u8 ignore_df:1; 946 __u8 dst_pending_confirm:1; 947 __u8 ip_summed:2; 948 __u8 ooo_okay:1; 949 950 /* private: */ 951 __u8 __mono_tc_offset[0]; 952 /* public: */ 953 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */ 954 #ifdef CONFIG_NET_XGRESS 955 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 956 __u8 tc_skip_classify:1; 957 #endif 958 __u8 remcsum_offload:1; 959 __u8 csum_complete_sw:1; 960 __u8 csum_level:2; 961 __u8 inner_protocol_type:1; 962 963 __u8 l4_hash:1; 964 __u8 sw_hash:1; 965 #ifdef CONFIG_WIRELESS 966 __u8 wifi_acked_valid:1; 967 __u8 wifi_acked:1; 968 #endif 969 __u8 no_fcs:1; 970 /* Indicates the inner headers are valid in the skbuff. */ 971 __u8 encapsulation:1; 972 __u8 encap_hdr_csum:1; 973 __u8 csum_valid:1; 974 #ifdef CONFIG_IPV6_NDISC_NODETYPE 975 __u8 ndisc_nodetype:2; 976 #endif 977 978 #if IS_ENABLED(CONFIG_IP_VS) 979 __u8 ipvs_property:1; 980 #endif 981 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 982 __u8 nf_trace:1; 983 #endif 984 #ifdef CONFIG_NET_SWITCHDEV 985 __u8 offload_fwd_mark:1; 986 __u8 offload_l3_fwd_mark:1; 987 #endif 988 __u8 redirected:1; 989 #ifdef CONFIG_NET_REDIRECT 990 __u8 from_ingress:1; 991 #endif 992 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 993 __u8 nf_skip_egress:1; 994 #endif 995 #ifdef CONFIG_TLS_DEVICE 996 __u8 decrypted:1; 997 #endif 998 __u8 slow_gro:1; 999 #if IS_ENABLED(CONFIG_IP_SCTP) 1000 __u8 csum_not_inet:1; 1001 #endif 1002 1003 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1004 __u16 tc_index; /* traffic control index */ 1005 #endif 1006 1007 u16 alloc_cpu; 1008 1009 union { 1010 __wsum csum; 1011 struct { 1012 __u16 csum_start; 1013 __u16 csum_offset; 1014 }; 1015 }; 1016 __u32 priority; 1017 int skb_iif; 1018 __u32 hash; 1019 union { 1020 u32 vlan_all; 1021 struct { 1022 __be16 vlan_proto; 1023 __u16 vlan_tci; 1024 }; 1025 }; 1026 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1027 union { 1028 unsigned int napi_id; 1029 unsigned int sender_cpu; 1030 }; 1031 #endif 1032 #ifdef CONFIG_NETWORK_SECMARK 1033 __u32 secmark; 1034 #endif 1035 1036 union { 1037 __u32 mark; 1038 __u32 reserved_tailroom; 1039 }; 1040 1041 union { 1042 __be16 inner_protocol; 1043 __u8 inner_ipproto; 1044 }; 1045 1046 __u16 inner_transport_header; 1047 __u16 inner_network_header; 1048 __u16 inner_mac_header; 1049 1050 __be16 protocol; 1051 __u16 transport_header; 1052 __u16 network_header; 1053 __u16 mac_header; 1054 1055 #ifdef CONFIG_KCOV 1056 u64 kcov_handle; 1057 #endif 1058 1059 ); /* end headers group */ 1060 1061 /* These elements must be at the end, see alloc_skb() for details. */ 1062 sk_buff_data_t tail; 1063 sk_buff_data_t end; 1064 unsigned char *head, 1065 *data; 1066 unsigned int truesize; 1067 refcount_t users; 1068 1069 #ifdef CONFIG_SKB_EXTENSIONS 1070 /* only usable after checking ->active_extensions != 0 */ 1071 struct skb_ext *extensions; 1072 #endif 1073 }; 1074 1075 /* if you move pkt_type around you also must adapt those constants */ 1076 #ifdef __BIG_ENDIAN_BITFIELD 1077 #define PKT_TYPE_MAX (7 << 5) 1078 #else 1079 #define PKT_TYPE_MAX 7 1080 #endif 1081 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1082 1083 /* if you move tc_at_ingress or mono_delivery_time 1084 * around, you also must adapt these constants. 1085 */ 1086 #ifdef __BIG_ENDIAN_BITFIELD 1087 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 7) 1088 #define TC_AT_INGRESS_MASK (1 << 6) 1089 #else 1090 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 0) 1091 #define TC_AT_INGRESS_MASK (1 << 1) 1092 #endif 1093 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1094 1095 #ifdef __KERNEL__ 1096 /* 1097 * Handling routines are only of interest to the kernel 1098 */ 1099 1100 #define SKB_ALLOC_FCLONE 0x01 1101 #define SKB_ALLOC_RX 0x02 1102 #define SKB_ALLOC_NAPI 0x04 1103 1104 /** 1105 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1106 * @skb: buffer 1107 */ 1108 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1109 { 1110 return unlikely(skb->pfmemalloc); 1111 } 1112 1113 /* 1114 * skb might have a dst pointer attached, refcounted or not. 1115 * _skb_refdst low order bit is set if refcount was _not_ taken 1116 */ 1117 #define SKB_DST_NOREF 1UL 1118 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1119 1120 /** 1121 * skb_dst - returns skb dst_entry 1122 * @skb: buffer 1123 * 1124 * Returns skb dst_entry, regardless of reference taken or not. 1125 */ 1126 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1127 { 1128 /* If refdst was not refcounted, check we still are in a 1129 * rcu_read_lock section 1130 */ 1131 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1132 !rcu_read_lock_held() && 1133 !rcu_read_lock_bh_held()); 1134 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1135 } 1136 1137 /** 1138 * skb_dst_set - sets skb dst 1139 * @skb: buffer 1140 * @dst: dst entry 1141 * 1142 * Sets skb dst, assuming a reference was taken on dst and should 1143 * be released by skb_dst_drop() 1144 */ 1145 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1146 { 1147 skb->slow_gro |= !!dst; 1148 skb->_skb_refdst = (unsigned long)dst; 1149 } 1150 1151 /** 1152 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1153 * @skb: buffer 1154 * @dst: dst entry 1155 * 1156 * Sets skb dst, assuming a reference was not taken on dst. 1157 * If dst entry is cached, we do not take reference and dst_release 1158 * will be avoided by refdst_drop. If dst entry is not cached, we take 1159 * reference, so that last dst_release can destroy the dst immediately. 1160 */ 1161 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1162 { 1163 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1164 skb->slow_gro |= !!dst; 1165 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1166 } 1167 1168 /** 1169 * skb_dst_is_noref - Test if skb dst isn't refcounted 1170 * @skb: buffer 1171 */ 1172 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1173 { 1174 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1175 } 1176 1177 /** 1178 * skb_rtable - Returns the skb &rtable 1179 * @skb: buffer 1180 */ 1181 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 1182 { 1183 return (struct rtable *)skb_dst(skb); 1184 } 1185 1186 /* For mangling skb->pkt_type from user space side from applications 1187 * such as nft, tc, etc, we only allow a conservative subset of 1188 * possible pkt_types to be set. 1189 */ 1190 static inline bool skb_pkt_type_ok(u32 ptype) 1191 { 1192 return ptype <= PACKET_OTHERHOST; 1193 } 1194 1195 /** 1196 * skb_napi_id - Returns the skb's NAPI id 1197 * @skb: buffer 1198 */ 1199 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1200 { 1201 #ifdef CONFIG_NET_RX_BUSY_POLL 1202 return skb->napi_id; 1203 #else 1204 return 0; 1205 #endif 1206 } 1207 1208 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1209 { 1210 #ifdef CONFIG_WIRELESS 1211 return skb->wifi_acked_valid; 1212 #else 1213 return 0; 1214 #endif 1215 } 1216 1217 /** 1218 * skb_unref - decrement the skb's reference count 1219 * @skb: buffer 1220 * 1221 * Returns true if we can free the skb. 1222 */ 1223 static inline bool skb_unref(struct sk_buff *skb) 1224 { 1225 if (unlikely(!skb)) 1226 return false; 1227 if (likely(refcount_read(&skb->users) == 1)) 1228 smp_rmb(); 1229 else if (likely(!refcount_dec_and_test(&skb->users))) 1230 return false; 1231 1232 return true; 1233 } 1234 1235 void __fix_address 1236 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason); 1237 1238 /** 1239 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1240 * @skb: buffer to free 1241 */ 1242 static inline void kfree_skb(struct sk_buff *skb) 1243 { 1244 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1245 } 1246 1247 void skb_release_head_state(struct sk_buff *skb); 1248 void kfree_skb_list_reason(struct sk_buff *segs, 1249 enum skb_drop_reason reason); 1250 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1251 void skb_tx_error(struct sk_buff *skb); 1252 1253 static inline void kfree_skb_list(struct sk_buff *segs) 1254 { 1255 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1256 } 1257 1258 #ifdef CONFIG_TRACEPOINTS 1259 void consume_skb(struct sk_buff *skb); 1260 #else 1261 static inline void consume_skb(struct sk_buff *skb) 1262 { 1263 return kfree_skb(skb); 1264 } 1265 #endif 1266 1267 void __consume_stateless_skb(struct sk_buff *skb); 1268 void __kfree_skb(struct sk_buff *skb); 1269 extern struct kmem_cache *skbuff_cache; 1270 1271 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1272 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1273 bool *fragstolen, int *delta_truesize); 1274 1275 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1276 int node); 1277 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1278 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1279 struct sk_buff *build_skb_around(struct sk_buff *skb, 1280 void *data, unsigned int frag_size); 1281 void skb_attempt_defer_free(struct sk_buff *skb); 1282 1283 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1284 struct sk_buff *slab_build_skb(void *data); 1285 1286 /** 1287 * alloc_skb - allocate a network buffer 1288 * @size: size to allocate 1289 * @priority: allocation mask 1290 * 1291 * This function is a convenient wrapper around __alloc_skb(). 1292 */ 1293 static inline struct sk_buff *alloc_skb(unsigned int size, 1294 gfp_t priority) 1295 { 1296 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1297 } 1298 1299 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1300 unsigned long data_len, 1301 int max_page_order, 1302 int *errcode, 1303 gfp_t gfp_mask); 1304 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1305 1306 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1307 struct sk_buff_fclones { 1308 struct sk_buff skb1; 1309 1310 struct sk_buff skb2; 1311 1312 refcount_t fclone_ref; 1313 }; 1314 1315 /** 1316 * skb_fclone_busy - check if fclone is busy 1317 * @sk: socket 1318 * @skb: buffer 1319 * 1320 * Returns true if skb is a fast clone, and its clone is not freed. 1321 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1322 * so we also check that didn't happen. 1323 */ 1324 static inline bool skb_fclone_busy(const struct sock *sk, 1325 const struct sk_buff *skb) 1326 { 1327 const struct sk_buff_fclones *fclones; 1328 1329 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1330 1331 return skb->fclone == SKB_FCLONE_ORIG && 1332 refcount_read(&fclones->fclone_ref) > 1 && 1333 READ_ONCE(fclones->skb2.sk) == sk; 1334 } 1335 1336 /** 1337 * alloc_skb_fclone - allocate a network buffer from fclone cache 1338 * @size: size to allocate 1339 * @priority: allocation mask 1340 * 1341 * This function is a convenient wrapper around __alloc_skb(). 1342 */ 1343 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1344 gfp_t priority) 1345 { 1346 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1347 } 1348 1349 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1350 void skb_headers_offset_update(struct sk_buff *skb, int off); 1351 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1352 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1353 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1354 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1355 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1356 gfp_t gfp_mask, bool fclone); 1357 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1358 gfp_t gfp_mask) 1359 { 1360 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1361 } 1362 1363 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1364 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1365 unsigned int headroom); 1366 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1367 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1368 int newtailroom, gfp_t priority); 1369 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1370 int offset, int len); 1371 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1372 int offset, int len); 1373 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1374 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1375 1376 /** 1377 * skb_pad - zero pad the tail of an skb 1378 * @skb: buffer to pad 1379 * @pad: space to pad 1380 * 1381 * Ensure that a buffer is followed by a padding area that is zero 1382 * filled. Used by network drivers which may DMA or transfer data 1383 * beyond the buffer end onto the wire. 1384 * 1385 * May return error in out of memory cases. The skb is freed on error. 1386 */ 1387 static inline int skb_pad(struct sk_buff *skb, int pad) 1388 { 1389 return __skb_pad(skb, pad, true); 1390 } 1391 #define dev_kfree_skb(a) consume_skb(a) 1392 1393 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1394 int offset, size_t size, size_t max_frags); 1395 1396 struct skb_seq_state { 1397 __u32 lower_offset; 1398 __u32 upper_offset; 1399 __u32 frag_idx; 1400 __u32 stepped_offset; 1401 struct sk_buff *root_skb; 1402 struct sk_buff *cur_skb; 1403 __u8 *frag_data; 1404 __u32 frag_off; 1405 }; 1406 1407 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1408 unsigned int to, struct skb_seq_state *st); 1409 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1410 struct skb_seq_state *st); 1411 void skb_abort_seq_read(struct skb_seq_state *st); 1412 1413 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1414 unsigned int to, struct ts_config *config); 1415 1416 /* 1417 * Packet hash types specify the type of hash in skb_set_hash. 1418 * 1419 * Hash types refer to the protocol layer addresses which are used to 1420 * construct a packet's hash. The hashes are used to differentiate or identify 1421 * flows of the protocol layer for the hash type. Hash types are either 1422 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1423 * 1424 * Properties of hashes: 1425 * 1426 * 1) Two packets in different flows have different hash values 1427 * 2) Two packets in the same flow should have the same hash value 1428 * 1429 * A hash at a higher layer is considered to be more specific. A driver should 1430 * set the most specific hash possible. 1431 * 1432 * A driver cannot indicate a more specific hash than the layer at which a hash 1433 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1434 * 1435 * A driver may indicate a hash level which is less specific than the 1436 * actual layer the hash was computed on. For instance, a hash computed 1437 * at L4 may be considered an L3 hash. This should only be done if the 1438 * driver can't unambiguously determine that the HW computed the hash at 1439 * the higher layer. Note that the "should" in the second property above 1440 * permits this. 1441 */ 1442 enum pkt_hash_types { 1443 PKT_HASH_TYPE_NONE, /* Undefined type */ 1444 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1445 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1446 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1447 }; 1448 1449 static inline void skb_clear_hash(struct sk_buff *skb) 1450 { 1451 skb->hash = 0; 1452 skb->sw_hash = 0; 1453 skb->l4_hash = 0; 1454 } 1455 1456 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1457 { 1458 if (!skb->l4_hash) 1459 skb_clear_hash(skb); 1460 } 1461 1462 static inline void 1463 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1464 { 1465 skb->l4_hash = is_l4; 1466 skb->sw_hash = is_sw; 1467 skb->hash = hash; 1468 } 1469 1470 static inline void 1471 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1472 { 1473 /* Used by drivers to set hash from HW */ 1474 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1475 } 1476 1477 static inline void 1478 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1479 { 1480 __skb_set_hash(skb, hash, true, is_l4); 1481 } 1482 1483 void __skb_get_hash(struct sk_buff *skb); 1484 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1485 u32 skb_get_poff(const struct sk_buff *skb); 1486 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1487 const struct flow_keys_basic *keys, int hlen); 1488 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1489 const void *data, int hlen_proto); 1490 1491 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1492 int thoff, u8 ip_proto) 1493 { 1494 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1495 } 1496 1497 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1498 const struct flow_dissector_key *key, 1499 unsigned int key_count); 1500 1501 struct bpf_flow_dissector; 1502 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1503 __be16 proto, int nhoff, int hlen, unsigned int flags); 1504 1505 bool __skb_flow_dissect(const struct net *net, 1506 const struct sk_buff *skb, 1507 struct flow_dissector *flow_dissector, 1508 void *target_container, const void *data, 1509 __be16 proto, int nhoff, int hlen, unsigned int flags); 1510 1511 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1512 struct flow_dissector *flow_dissector, 1513 void *target_container, unsigned int flags) 1514 { 1515 return __skb_flow_dissect(NULL, skb, flow_dissector, 1516 target_container, NULL, 0, 0, 0, flags); 1517 } 1518 1519 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1520 struct flow_keys *flow, 1521 unsigned int flags) 1522 { 1523 memset(flow, 0, sizeof(*flow)); 1524 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1525 flow, NULL, 0, 0, 0, flags); 1526 } 1527 1528 static inline bool 1529 skb_flow_dissect_flow_keys_basic(const struct net *net, 1530 const struct sk_buff *skb, 1531 struct flow_keys_basic *flow, 1532 const void *data, __be16 proto, 1533 int nhoff, int hlen, unsigned int flags) 1534 { 1535 memset(flow, 0, sizeof(*flow)); 1536 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1537 data, proto, nhoff, hlen, flags); 1538 } 1539 1540 void skb_flow_dissect_meta(const struct sk_buff *skb, 1541 struct flow_dissector *flow_dissector, 1542 void *target_container); 1543 1544 /* Gets a skb connection tracking info, ctinfo map should be a 1545 * map of mapsize to translate enum ip_conntrack_info states 1546 * to user states. 1547 */ 1548 void 1549 skb_flow_dissect_ct(const struct sk_buff *skb, 1550 struct flow_dissector *flow_dissector, 1551 void *target_container, 1552 u16 *ctinfo_map, size_t mapsize, 1553 bool post_ct, u16 zone); 1554 void 1555 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1556 struct flow_dissector *flow_dissector, 1557 void *target_container); 1558 1559 void skb_flow_dissect_hash(const struct sk_buff *skb, 1560 struct flow_dissector *flow_dissector, 1561 void *target_container); 1562 1563 static inline __u32 skb_get_hash(struct sk_buff *skb) 1564 { 1565 if (!skb->l4_hash && !skb->sw_hash) 1566 __skb_get_hash(skb); 1567 1568 return skb->hash; 1569 } 1570 1571 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1572 { 1573 if (!skb->l4_hash && !skb->sw_hash) { 1574 struct flow_keys keys; 1575 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1576 1577 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1578 } 1579 1580 return skb->hash; 1581 } 1582 1583 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1584 const siphash_key_t *perturb); 1585 1586 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1587 { 1588 return skb->hash; 1589 } 1590 1591 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1592 { 1593 to->hash = from->hash; 1594 to->sw_hash = from->sw_hash; 1595 to->l4_hash = from->l4_hash; 1596 }; 1597 1598 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1599 const struct sk_buff *skb2) 1600 { 1601 #ifdef CONFIG_TLS_DEVICE 1602 return skb2->decrypted - skb1->decrypted; 1603 #else 1604 return 0; 1605 #endif 1606 } 1607 1608 static inline void skb_copy_decrypted(struct sk_buff *to, 1609 const struct sk_buff *from) 1610 { 1611 #ifdef CONFIG_TLS_DEVICE 1612 to->decrypted = from->decrypted; 1613 #endif 1614 } 1615 1616 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1617 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1618 { 1619 return skb->head + skb->end; 1620 } 1621 1622 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1623 { 1624 return skb->end; 1625 } 1626 1627 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1628 { 1629 skb->end = offset; 1630 } 1631 #else 1632 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1633 { 1634 return skb->end; 1635 } 1636 1637 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1638 { 1639 return skb->end - skb->head; 1640 } 1641 1642 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1643 { 1644 skb->end = skb->head + offset; 1645 } 1646 #endif 1647 1648 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1649 struct ubuf_info *uarg); 1650 1651 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1652 1653 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 1654 bool success); 1655 1656 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1657 struct sk_buff *skb, struct iov_iter *from, 1658 size_t length); 1659 1660 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1661 struct msghdr *msg, int len) 1662 { 1663 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1664 } 1665 1666 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1667 struct msghdr *msg, int len, 1668 struct ubuf_info *uarg); 1669 1670 /* Internal */ 1671 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1672 1673 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1674 { 1675 return &skb_shinfo(skb)->hwtstamps; 1676 } 1677 1678 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1679 { 1680 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1681 1682 return is_zcopy ? skb_uarg(skb) : NULL; 1683 } 1684 1685 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1686 { 1687 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1688 } 1689 1690 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1691 { 1692 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1693 } 1694 1695 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1696 const struct sk_buff *skb2) 1697 { 1698 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1699 } 1700 1701 static inline void net_zcopy_get(struct ubuf_info *uarg) 1702 { 1703 refcount_inc(&uarg->refcnt); 1704 } 1705 1706 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1707 { 1708 skb_shinfo(skb)->destructor_arg = uarg; 1709 skb_shinfo(skb)->flags |= uarg->flags; 1710 } 1711 1712 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1713 bool *have_ref) 1714 { 1715 if (skb && uarg && !skb_zcopy(skb)) { 1716 if (unlikely(have_ref && *have_ref)) 1717 *have_ref = false; 1718 else 1719 net_zcopy_get(uarg); 1720 skb_zcopy_init(skb, uarg); 1721 } 1722 } 1723 1724 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1725 { 1726 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1727 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1728 } 1729 1730 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1731 { 1732 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1733 } 1734 1735 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1736 { 1737 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1738 } 1739 1740 static inline void net_zcopy_put(struct ubuf_info *uarg) 1741 { 1742 if (uarg) 1743 uarg->callback(NULL, uarg, true); 1744 } 1745 1746 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1747 { 1748 if (uarg) { 1749 if (uarg->callback == msg_zerocopy_callback) 1750 msg_zerocopy_put_abort(uarg, have_uref); 1751 else if (have_uref) 1752 net_zcopy_put(uarg); 1753 } 1754 } 1755 1756 /* Release a reference on a zerocopy structure */ 1757 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1758 { 1759 struct ubuf_info *uarg = skb_zcopy(skb); 1760 1761 if (uarg) { 1762 if (!skb_zcopy_is_nouarg(skb)) 1763 uarg->callback(skb, uarg, zerocopy_success); 1764 1765 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1766 } 1767 } 1768 1769 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1770 1771 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1772 { 1773 if (unlikely(skb_zcopy_managed(skb))) 1774 __skb_zcopy_downgrade_managed(skb); 1775 } 1776 1777 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1778 { 1779 skb->next = NULL; 1780 } 1781 1782 static inline void skb_poison_list(struct sk_buff *skb) 1783 { 1784 #ifdef CONFIG_DEBUG_NET 1785 skb->next = SKB_LIST_POISON_NEXT; 1786 #endif 1787 } 1788 1789 /* Iterate through singly-linked GSO fragments of an skb. */ 1790 #define skb_list_walk_safe(first, skb, next_skb) \ 1791 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1792 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1793 1794 static inline void skb_list_del_init(struct sk_buff *skb) 1795 { 1796 __list_del_entry(&skb->list); 1797 skb_mark_not_on_list(skb); 1798 } 1799 1800 /** 1801 * skb_queue_empty - check if a queue is empty 1802 * @list: queue head 1803 * 1804 * Returns true if the queue is empty, false otherwise. 1805 */ 1806 static inline int skb_queue_empty(const struct sk_buff_head *list) 1807 { 1808 return list->next == (const struct sk_buff *) list; 1809 } 1810 1811 /** 1812 * skb_queue_empty_lockless - check if a queue is empty 1813 * @list: queue head 1814 * 1815 * Returns true if the queue is empty, false otherwise. 1816 * This variant can be used in lockless contexts. 1817 */ 1818 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1819 { 1820 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1821 } 1822 1823 1824 /** 1825 * skb_queue_is_last - check if skb is the last entry in the queue 1826 * @list: queue head 1827 * @skb: buffer 1828 * 1829 * Returns true if @skb is the last buffer on the list. 1830 */ 1831 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1832 const struct sk_buff *skb) 1833 { 1834 return skb->next == (const struct sk_buff *) list; 1835 } 1836 1837 /** 1838 * skb_queue_is_first - check if skb is the first entry in the queue 1839 * @list: queue head 1840 * @skb: buffer 1841 * 1842 * Returns true if @skb is the first buffer on the list. 1843 */ 1844 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1845 const struct sk_buff *skb) 1846 { 1847 return skb->prev == (const struct sk_buff *) list; 1848 } 1849 1850 /** 1851 * skb_queue_next - return the next packet in the queue 1852 * @list: queue head 1853 * @skb: current buffer 1854 * 1855 * Return the next packet in @list after @skb. It is only valid to 1856 * call this if skb_queue_is_last() evaluates to false. 1857 */ 1858 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1859 const struct sk_buff *skb) 1860 { 1861 /* This BUG_ON may seem severe, but if we just return then we 1862 * are going to dereference garbage. 1863 */ 1864 BUG_ON(skb_queue_is_last(list, skb)); 1865 return skb->next; 1866 } 1867 1868 /** 1869 * skb_queue_prev - return the prev packet in the queue 1870 * @list: queue head 1871 * @skb: current buffer 1872 * 1873 * Return the prev packet in @list before @skb. It is only valid to 1874 * call this if skb_queue_is_first() evaluates to false. 1875 */ 1876 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1877 const struct sk_buff *skb) 1878 { 1879 /* This BUG_ON may seem severe, but if we just return then we 1880 * are going to dereference garbage. 1881 */ 1882 BUG_ON(skb_queue_is_first(list, skb)); 1883 return skb->prev; 1884 } 1885 1886 /** 1887 * skb_get - reference buffer 1888 * @skb: buffer to reference 1889 * 1890 * Makes another reference to a socket buffer and returns a pointer 1891 * to the buffer. 1892 */ 1893 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1894 { 1895 refcount_inc(&skb->users); 1896 return skb; 1897 } 1898 1899 /* 1900 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1901 */ 1902 1903 /** 1904 * skb_cloned - is the buffer a clone 1905 * @skb: buffer to check 1906 * 1907 * Returns true if the buffer was generated with skb_clone() and is 1908 * one of multiple shared copies of the buffer. Cloned buffers are 1909 * shared data so must not be written to under normal circumstances. 1910 */ 1911 static inline int skb_cloned(const struct sk_buff *skb) 1912 { 1913 return skb->cloned && 1914 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1915 } 1916 1917 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1918 { 1919 might_sleep_if(gfpflags_allow_blocking(pri)); 1920 1921 if (skb_cloned(skb)) 1922 return pskb_expand_head(skb, 0, 0, pri); 1923 1924 return 0; 1925 } 1926 1927 /* This variant of skb_unclone() makes sure skb->truesize 1928 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1929 * 1930 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 1931 * when various debugging features are in place. 1932 */ 1933 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 1934 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 1935 { 1936 might_sleep_if(gfpflags_allow_blocking(pri)); 1937 1938 if (skb_cloned(skb)) 1939 return __skb_unclone_keeptruesize(skb, pri); 1940 return 0; 1941 } 1942 1943 /** 1944 * skb_header_cloned - is the header a clone 1945 * @skb: buffer to check 1946 * 1947 * Returns true if modifying the header part of the buffer requires 1948 * the data to be copied. 1949 */ 1950 static inline int skb_header_cloned(const struct sk_buff *skb) 1951 { 1952 int dataref; 1953 1954 if (!skb->cloned) 1955 return 0; 1956 1957 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1958 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1959 return dataref != 1; 1960 } 1961 1962 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1963 { 1964 might_sleep_if(gfpflags_allow_blocking(pri)); 1965 1966 if (skb_header_cloned(skb)) 1967 return pskb_expand_head(skb, 0, 0, pri); 1968 1969 return 0; 1970 } 1971 1972 /** 1973 * __skb_header_release() - allow clones to use the headroom 1974 * @skb: buffer to operate on 1975 * 1976 * See "DOC: dataref and headerless skbs". 1977 */ 1978 static inline void __skb_header_release(struct sk_buff *skb) 1979 { 1980 skb->nohdr = 1; 1981 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1982 } 1983 1984 1985 /** 1986 * skb_shared - is the buffer shared 1987 * @skb: buffer to check 1988 * 1989 * Returns true if more than one person has a reference to this 1990 * buffer. 1991 */ 1992 static inline int skb_shared(const struct sk_buff *skb) 1993 { 1994 return refcount_read(&skb->users) != 1; 1995 } 1996 1997 /** 1998 * skb_share_check - check if buffer is shared and if so clone it 1999 * @skb: buffer to check 2000 * @pri: priority for memory allocation 2001 * 2002 * If the buffer is shared the buffer is cloned and the old copy 2003 * drops a reference. A new clone with a single reference is returned. 2004 * If the buffer is not shared the original buffer is returned. When 2005 * being called from interrupt status or with spinlocks held pri must 2006 * be GFP_ATOMIC. 2007 * 2008 * NULL is returned on a memory allocation failure. 2009 */ 2010 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2011 { 2012 might_sleep_if(gfpflags_allow_blocking(pri)); 2013 if (skb_shared(skb)) { 2014 struct sk_buff *nskb = skb_clone(skb, pri); 2015 2016 if (likely(nskb)) 2017 consume_skb(skb); 2018 else 2019 kfree_skb(skb); 2020 skb = nskb; 2021 } 2022 return skb; 2023 } 2024 2025 /* 2026 * Copy shared buffers into a new sk_buff. We effectively do COW on 2027 * packets to handle cases where we have a local reader and forward 2028 * and a couple of other messy ones. The normal one is tcpdumping 2029 * a packet that's being forwarded. 2030 */ 2031 2032 /** 2033 * skb_unshare - make a copy of a shared buffer 2034 * @skb: buffer to check 2035 * @pri: priority for memory allocation 2036 * 2037 * If the socket buffer is a clone then this function creates a new 2038 * copy of the data, drops a reference count on the old copy and returns 2039 * the new copy with the reference count at 1. If the buffer is not a clone 2040 * the original buffer is returned. When called with a spinlock held or 2041 * from interrupt state @pri must be %GFP_ATOMIC 2042 * 2043 * %NULL is returned on a memory allocation failure. 2044 */ 2045 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2046 gfp_t pri) 2047 { 2048 might_sleep_if(gfpflags_allow_blocking(pri)); 2049 if (skb_cloned(skb)) { 2050 struct sk_buff *nskb = skb_copy(skb, pri); 2051 2052 /* Free our shared copy */ 2053 if (likely(nskb)) 2054 consume_skb(skb); 2055 else 2056 kfree_skb(skb); 2057 skb = nskb; 2058 } 2059 return skb; 2060 } 2061 2062 /** 2063 * skb_peek - peek at the head of an &sk_buff_head 2064 * @list_: list to peek at 2065 * 2066 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2067 * be careful with this one. A peek leaves the buffer on the 2068 * list and someone else may run off with it. You must hold 2069 * the appropriate locks or have a private queue to do this. 2070 * 2071 * Returns %NULL for an empty list or a pointer to the head element. 2072 * The reference count is not incremented and the reference is therefore 2073 * volatile. Use with caution. 2074 */ 2075 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2076 { 2077 struct sk_buff *skb = list_->next; 2078 2079 if (skb == (struct sk_buff *)list_) 2080 skb = NULL; 2081 return skb; 2082 } 2083 2084 /** 2085 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2086 * @list_: list to peek at 2087 * 2088 * Like skb_peek(), but the caller knows that the list is not empty. 2089 */ 2090 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2091 { 2092 return list_->next; 2093 } 2094 2095 /** 2096 * skb_peek_next - peek skb following the given one from a queue 2097 * @skb: skb to start from 2098 * @list_: list to peek at 2099 * 2100 * Returns %NULL when the end of the list is met or a pointer to the 2101 * next element. The reference count is not incremented and the 2102 * reference is therefore volatile. Use with caution. 2103 */ 2104 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2105 const struct sk_buff_head *list_) 2106 { 2107 struct sk_buff *next = skb->next; 2108 2109 if (next == (struct sk_buff *)list_) 2110 next = NULL; 2111 return next; 2112 } 2113 2114 /** 2115 * skb_peek_tail - peek at the tail of an &sk_buff_head 2116 * @list_: list to peek at 2117 * 2118 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2119 * be careful with this one. A peek leaves the buffer on the 2120 * list and someone else may run off with it. You must hold 2121 * the appropriate locks or have a private queue to do this. 2122 * 2123 * Returns %NULL for an empty list or a pointer to the tail element. 2124 * The reference count is not incremented and the reference is therefore 2125 * volatile. Use with caution. 2126 */ 2127 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2128 { 2129 struct sk_buff *skb = READ_ONCE(list_->prev); 2130 2131 if (skb == (struct sk_buff *)list_) 2132 skb = NULL; 2133 return skb; 2134 2135 } 2136 2137 /** 2138 * skb_queue_len - get queue length 2139 * @list_: list to measure 2140 * 2141 * Return the length of an &sk_buff queue. 2142 */ 2143 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2144 { 2145 return list_->qlen; 2146 } 2147 2148 /** 2149 * skb_queue_len_lockless - get queue length 2150 * @list_: list to measure 2151 * 2152 * Return the length of an &sk_buff queue. 2153 * This variant can be used in lockless contexts. 2154 */ 2155 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2156 { 2157 return READ_ONCE(list_->qlen); 2158 } 2159 2160 /** 2161 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2162 * @list: queue to initialize 2163 * 2164 * This initializes only the list and queue length aspects of 2165 * an sk_buff_head object. This allows to initialize the list 2166 * aspects of an sk_buff_head without reinitializing things like 2167 * the spinlock. It can also be used for on-stack sk_buff_head 2168 * objects where the spinlock is known to not be used. 2169 */ 2170 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2171 { 2172 list->prev = list->next = (struct sk_buff *)list; 2173 list->qlen = 0; 2174 } 2175 2176 /* 2177 * This function creates a split out lock class for each invocation; 2178 * this is needed for now since a whole lot of users of the skb-queue 2179 * infrastructure in drivers have different locking usage (in hardirq) 2180 * than the networking core (in softirq only). In the long run either the 2181 * network layer or drivers should need annotation to consolidate the 2182 * main types of usage into 3 classes. 2183 */ 2184 static inline void skb_queue_head_init(struct sk_buff_head *list) 2185 { 2186 spin_lock_init(&list->lock); 2187 __skb_queue_head_init(list); 2188 } 2189 2190 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2191 struct lock_class_key *class) 2192 { 2193 skb_queue_head_init(list); 2194 lockdep_set_class(&list->lock, class); 2195 } 2196 2197 /* 2198 * Insert an sk_buff on a list. 2199 * 2200 * The "__skb_xxxx()" functions are the non-atomic ones that 2201 * can only be called with interrupts disabled. 2202 */ 2203 static inline void __skb_insert(struct sk_buff *newsk, 2204 struct sk_buff *prev, struct sk_buff *next, 2205 struct sk_buff_head *list) 2206 { 2207 /* See skb_queue_empty_lockless() and skb_peek_tail() 2208 * for the opposite READ_ONCE() 2209 */ 2210 WRITE_ONCE(newsk->next, next); 2211 WRITE_ONCE(newsk->prev, prev); 2212 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2213 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2214 WRITE_ONCE(list->qlen, list->qlen + 1); 2215 } 2216 2217 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2218 struct sk_buff *prev, 2219 struct sk_buff *next) 2220 { 2221 struct sk_buff *first = list->next; 2222 struct sk_buff *last = list->prev; 2223 2224 WRITE_ONCE(first->prev, prev); 2225 WRITE_ONCE(prev->next, first); 2226 2227 WRITE_ONCE(last->next, next); 2228 WRITE_ONCE(next->prev, last); 2229 } 2230 2231 /** 2232 * skb_queue_splice - join two skb lists, this is designed for stacks 2233 * @list: the new list to add 2234 * @head: the place to add it in the first list 2235 */ 2236 static inline void skb_queue_splice(const struct sk_buff_head *list, 2237 struct sk_buff_head *head) 2238 { 2239 if (!skb_queue_empty(list)) { 2240 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2241 head->qlen += list->qlen; 2242 } 2243 } 2244 2245 /** 2246 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2247 * @list: the new list to add 2248 * @head: the place to add it in the first list 2249 * 2250 * The list at @list is reinitialised 2251 */ 2252 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2253 struct sk_buff_head *head) 2254 { 2255 if (!skb_queue_empty(list)) { 2256 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2257 head->qlen += list->qlen; 2258 __skb_queue_head_init(list); 2259 } 2260 } 2261 2262 /** 2263 * skb_queue_splice_tail - join two skb lists, each list being a queue 2264 * @list: the new list to add 2265 * @head: the place to add it in the first list 2266 */ 2267 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2268 struct sk_buff_head *head) 2269 { 2270 if (!skb_queue_empty(list)) { 2271 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2272 head->qlen += list->qlen; 2273 } 2274 } 2275 2276 /** 2277 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2278 * @list: the new list to add 2279 * @head: the place to add it in the first list 2280 * 2281 * Each of the lists is a queue. 2282 * The list at @list is reinitialised 2283 */ 2284 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2285 struct sk_buff_head *head) 2286 { 2287 if (!skb_queue_empty(list)) { 2288 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2289 head->qlen += list->qlen; 2290 __skb_queue_head_init(list); 2291 } 2292 } 2293 2294 /** 2295 * __skb_queue_after - queue a buffer at the list head 2296 * @list: list to use 2297 * @prev: place after this buffer 2298 * @newsk: buffer to queue 2299 * 2300 * Queue a buffer int the middle of a list. This function takes no locks 2301 * and you must therefore hold required locks before calling it. 2302 * 2303 * A buffer cannot be placed on two lists at the same time. 2304 */ 2305 static inline void __skb_queue_after(struct sk_buff_head *list, 2306 struct sk_buff *prev, 2307 struct sk_buff *newsk) 2308 { 2309 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2310 } 2311 2312 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2313 struct sk_buff_head *list); 2314 2315 static inline void __skb_queue_before(struct sk_buff_head *list, 2316 struct sk_buff *next, 2317 struct sk_buff *newsk) 2318 { 2319 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2320 } 2321 2322 /** 2323 * __skb_queue_head - queue a buffer at the list head 2324 * @list: list to use 2325 * @newsk: buffer to queue 2326 * 2327 * Queue a buffer at the start of a list. This function takes no locks 2328 * and you must therefore hold required locks before calling it. 2329 * 2330 * A buffer cannot be placed on two lists at the same time. 2331 */ 2332 static inline void __skb_queue_head(struct sk_buff_head *list, 2333 struct sk_buff *newsk) 2334 { 2335 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2336 } 2337 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2338 2339 /** 2340 * __skb_queue_tail - queue a buffer at the list tail 2341 * @list: list to use 2342 * @newsk: buffer to queue 2343 * 2344 * Queue a buffer at the end of a list. This function takes no locks 2345 * and you must therefore hold required locks before calling it. 2346 * 2347 * A buffer cannot be placed on two lists at the same time. 2348 */ 2349 static inline void __skb_queue_tail(struct sk_buff_head *list, 2350 struct sk_buff *newsk) 2351 { 2352 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2353 } 2354 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2355 2356 /* 2357 * remove sk_buff from list. _Must_ be called atomically, and with 2358 * the list known.. 2359 */ 2360 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2361 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2362 { 2363 struct sk_buff *next, *prev; 2364 2365 WRITE_ONCE(list->qlen, list->qlen - 1); 2366 next = skb->next; 2367 prev = skb->prev; 2368 skb->next = skb->prev = NULL; 2369 WRITE_ONCE(next->prev, prev); 2370 WRITE_ONCE(prev->next, next); 2371 } 2372 2373 /** 2374 * __skb_dequeue - remove from the head of the queue 2375 * @list: list to dequeue from 2376 * 2377 * Remove the head of the list. This function does not take any locks 2378 * so must be used with appropriate locks held only. The head item is 2379 * returned or %NULL if the list is empty. 2380 */ 2381 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2382 { 2383 struct sk_buff *skb = skb_peek(list); 2384 if (skb) 2385 __skb_unlink(skb, list); 2386 return skb; 2387 } 2388 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2389 2390 /** 2391 * __skb_dequeue_tail - remove from the tail of the queue 2392 * @list: list to dequeue from 2393 * 2394 * Remove the tail of the list. This function does not take any locks 2395 * so must be used with appropriate locks held only. The tail item is 2396 * returned or %NULL if the list is empty. 2397 */ 2398 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2399 { 2400 struct sk_buff *skb = skb_peek_tail(list); 2401 if (skb) 2402 __skb_unlink(skb, list); 2403 return skb; 2404 } 2405 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2406 2407 2408 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2409 { 2410 return skb->data_len; 2411 } 2412 2413 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2414 { 2415 return skb->len - skb->data_len; 2416 } 2417 2418 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2419 { 2420 unsigned int i, len = 0; 2421 2422 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2423 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2424 return len; 2425 } 2426 2427 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2428 { 2429 return skb_headlen(skb) + __skb_pagelen(skb); 2430 } 2431 2432 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2433 struct page *page, 2434 int off, int size) 2435 { 2436 frag->bv_page = page; 2437 frag->bv_offset = off; 2438 skb_frag_size_set(frag, size); 2439 } 2440 2441 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2442 int i, struct page *page, 2443 int off, int size) 2444 { 2445 skb_frag_t *frag = &shinfo->frags[i]; 2446 2447 skb_frag_fill_page_desc(frag, page, off, size); 2448 } 2449 2450 /** 2451 * skb_len_add - adds a number to len fields of skb 2452 * @skb: buffer to add len to 2453 * @delta: number of bytes to add 2454 */ 2455 static inline void skb_len_add(struct sk_buff *skb, int delta) 2456 { 2457 skb->len += delta; 2458 skb->data_len += delta; 2459 skb->truesize += delta; 2460 } 2461 2462 /** 2463 * __skb_fill_page_desc - initialise a paged fragment in an skb 2464 * @skb: buffer containing fragment to be initialised 2465 * @i: paged fragment index to initialise 2466 * @page: the page to use for this fragment 2467 * @off: the offset to the data with @page 2468 * @size: the length of the data 2469 * 2470 * Initialises the @i'th fragment of @skb to point to &size bytes at 2471 * offset @off within @page. 2472 * 2473 * Does not take any additional reference on the fragment. 2474 */ 2475 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2476 struct page *page, int off, int size) 2477 { 2478 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size); 2479 2480 /* Propagate page pfmemalloc to the skb if we can. The problem is 2481 * that not all callers have unique ownership of the page but rely 2482 * on page_is_pfmemalloc doing the right thing(tm). 2483 */ 2484 page = compound_head(page); 2485 if (page_is_pfmemalloc(page)) 2486 skb->pfmemalloc = true; 2487 } 2488 2489 /** 2490 * skb_fill_page_desc - initialise a paged fragment in an skb 2491 * @skb: buffer containing fragment to be initialised 2492 * @i: paged fragment index to initialise 2493 * @page: the page to use for this fragment 2494 * @off: the offset to the data with @page 2495 * @size: the length of the data 2496 * 2497 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2498 * @skb to point to @size bytes at offset @off within @page. In 2499 * addition updates @skb such that @i is the last fragment. 2500 * 2501 * Does not take any additional reference on the fragment. 2502 */ 2503 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2504 struct page *page, int off, int size) 2505 { 2506 __skb_fill_page_desc(skb, i, page, off, size); 2507 skb_shinfo(skb)->nr_frags = i + 1; 2508 } 2509 2510 /** 2511 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2512 * @skb: buffer containing fragment to be initialised 2513 * @i: paged fragment index to initialise 2514 * @page: the page to use for this fragment 2515 * @off: the offset to the data with @page 2516 * @size: the length of the data 2517 * 2518 * Variant of skb_fill_page_desc() which does not deal with 2519 * pfmemalloc, if page is not owned by us. 2520 */ 2521 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2522 struct page *page, int off, 2523 int size) 2524 { 2525 struct skb_shared_info *shinfo = skb_shinfo(skb); 2526 2527 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2528 shinfo->nr_frags = i + 1; 2529 } 2530 2531 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2532 int size, unsigned int truesize); 2533 2534 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2535 unsigned int truesize); 2536 2537 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2538 2539 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2540 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2541 { 2542 return skb->head + skb->tail; 2543 } 2544 2545 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2546 { 2547 skb->tail = skb->data - skb->head; 2548 } 2549 2550 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2551 { 2552 skb_reset_tail_pointer(skb); 2553 skb->tail += offset; 2554 } 2555 2556 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2557 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2558 { 2559 return skb->tail; 2560 } 2561 2562 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2563 { 2564 skb->tail = skb->data; 2565 } 2566 2567 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2568 { 2569 skb->tail = skb->data + offset; 2570 } 2571 2572 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2573 2574 static inline void skb_assert_len(struct sk_buff *skb) 2575 { 2576 #ifdef CONFIG_DEBUG_NET 2577 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2578 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2579 #endif /* CONFIG_DEBUG_NET */ 2580 } 2581 2582 /* 2583 * Add data to an sk_buff 2584 */ 2585 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2586 void *skb_put(struct sk_buff *skb, unsigned int len); 2587 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2588 { 2589 void *tmp = skb_tail_pointer(skb); 2590 SKB_LINEAR_ASSERT(skb); 2591 skb->tail += len; 2592 skb->len += len; 2593 return tmp; 2594 } 2595 2596 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2597 { 2598 void *tmp = __skb_put(skb, len); 2599 2600 memset(tmp, 0, len); 2601 return tmp; 2602 } 2603 2604 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2605 unsigned int len) 2606 { 2607 void *tmp = __skb_put(skb, len); 2608 2609 memcpy(tmp, data, len); 2610 return tmp; 2611 } 2612 2613 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2614 { 2615 *(u8 *)__skb_put(skb, 1) = val; 2616 } 2617 2618 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2619 { 2620 void *tmp = skb_put(skb, len); 2621 2622 memset(tmp, 0, len); 2623 2624 return tmp; 2625 } 2626 2627 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2628 unsigned int len) 2629 { 2630 void *tmp = skb_put(skb, len); 2631 2632 memcpy(tmp, data, len); 2633 2634 return tmp; 2635 } 2636 2637 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2638 { 2639 *(u8 *)skb_put(skb, 1) = val; 2640 } 2641 2642 void *skb_push(struct sk_buff *skb, unsigned int len); 2643 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2644 { 2645 skb->data -= len; 2646 skb->len += len; 2647 return skb->data; 2648 } 2649 2650 void *skb_pull(struct sk_buff *skb, unsigned int len); 2651 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2652 { 2653 skb->len -= len; 2654 if (unlikely(skb->len < skb->data_len)) { 2655 #if defined(CONFIG_DEBUG_NET) 2656 skb->len += len; 2657 pr_err("__skb_pull(len=%u)\n", len); 2658 skb_dump(KERN_ERR, skb, false); 2659 #endif 2660 BUG(); 2661 } 2662 return skb->data += len; 2663 } 2664 2665 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2666 { 2667 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2668 } 2669 2670 void *skb_pull_data(struct sk_buff *skb, size_t len); 2671 2672 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2673 2674 static inline enum skb_drop_reason 2675 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2676 { 2677 if (likely(len <= skb_headlen(skb))) 2678 return SKB_NOT_DROPPED_YET; 2679 2680 if (unlikely(len > skb->len)) 2681 return SKB_DROP_REASON_PKT_TOO_SMALL; 2682 2683 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2684 return SKB_DROP_REASON_NOMEM; 2685 2686 return SKB_NOT_DROPPED_YET; 2687 } 2688 2689 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2690 { 2691 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2692 } 2693 2694 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2695 { 2696 if (!pskb_may_pull(skb, len)) 2697 return NULL; 2698 2699 skb->len -= len; 2700 return skb->data += len; 2701 } 2702 2703 void skb_condense(struct sk_buff *skb); 2704 2705 /** 2706 * skb_headroom - bytes at buffer head 2707 * @skb: buffer to check 2708 * 2709 * Return the number of bytes of free space at the head of an &sk_buff. 2710 */ 2711 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2712 { 2713 return skb->data - skb->head; 2714 } 2715 2716 /** 2717 * skb_tailroom - bytes at buffer end 2718 * @skb: buffer to check 2719 * 2720 * Return the number of bytes of free space at the tail of an sk_buff 2721 */ 2722 static inline int skb_tailroom(const struct sk_buff *skb) 2723 { 2724 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2725 } 2726 2727 /** 2728 * skb_availroom - bytes at buffer end 2729 * @skb: buffer to check 2730 * 2731 * Return the number of bytes of free space at the tail of an sk_buff 2732 * allocated by sk_stream_alloc() 2733 */ 2734 static inline int skb_availroom(const struct sk_buff *skb) 2735 { 2736 if (skb_is_nonlinear(skb)) 2737 return 0; 2738 2739 return skb->end - skb->tail - skb->reserved_tailroom; 2740 } 2741 2742 /** 2743 * skb_reserve - adjust headroom 2744 * @skb: buffer to alter 2745 * @len: bytes to move 2746 * 2747 * Increase the headroom of an empty &sk_buff by reducing the tail 2748 * room. This is only allowed for an empty buffer. 2749 */ 2750 static inline void skb_reserve(struct sk_buff *skb, int len) 2751 { 2752 skb->data += len; 2753 skb->tail += len; 2754 } 2755 2756 /** 2757 * skb_tailroom_reserve - adjust reserved_tailroom 2758 * @skb: buffer to alter 2759 * @mtu: maximum amount of headlen permitted 2760 * @needed_tailroom: minimum amount of reserved_tailroom 2761 * 2762 * Set reserved_tailroom so that headlen can be as large as possible but 2763 * not larger than mtu and tailroom cannot be smaller than 2764 * needed_tailroom. 2765 * The required headroom should already have been reserved before using 2766 * this function. 2767 */ 2768 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2769 unsigned int needed_tailroom) 2770 { 2771 SKB_LINEAR_ASSERT(skb); 2772 if (mtu < skb_tailroom(skb) - needed_tailroom) 2773 /* use at most mtu */ 2774 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2775 else 2776 /* use up to all available space */ 2777 skb->reserved_tailroom = needed_tailroom; 2778 } 2779 2780 #define ENCAP_TYPE_ETHER 0 2781 #define ENCAP_TYPE_IPPROTO 1 2782 2783 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2784 __be16 protocol) 2785 { 2786 skb->inner_protocol = protocol; 2787 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2788 } 2789 2790 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2791 __u8 ipproto) 2792 { 2793 skb->inner_ipproto = ipproto; 2794 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2795 } 2796 2797 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2798 { 2799 skb->inner_mac_header = skb->mac_header; 2800 skb->inner_network_header = skb->network_header; 2801 skb->inner_transport_header = skb->transport_header; 2802 } 2803 2804 static inline void skb_reset_mac_len(struct sk_buff *skb) 2805 { 2806 skb->mac_len = skb->network_header - skb->mac_header; 2807 } 2808 2809 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2810 *skb) 2811 { 2812 return skb->head + skb->inner_transport_header; 2813 } 2814 2815 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2816 { 2817 return skb_inner_transport_header(skb) - skb->data; 2818 } 2819 2820 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2821 { 2822 skb->inner_transport_header = skb->data - skb->head; 2823 } 2824 2825 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2826 const int offset) 2827 { 2828 skb_reset_inner_transport_header(skb); 2829 skb->inner_transport_header += offset; 2830 } 2831 2832 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2833 { 2834 return skb->head + skb->inner_network_header; 2835 } 2836 2837 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2838 { 2839 skb->inner_network_header = skb->data - skb->head; 2840 } 2841 2842 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2843 const int offset) 2844 { 2845 skb_reset_inner_network_header(skb); 2846 skb->inner_network_header += offset; 2847 } 2848 2849 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2850 { 2851 return skb->head + skb->inner_mac_header; 2852 } 2853 2854 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2855 { 2856 skb->inner_mac_header = skb->data - skb->head; 2857 } 2858 2859 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2860 const int offset) 2861 { 2862 skb_reset_inner_mac_header(skb); 2863 skb->inner_mac_header += offset; 2864 } 2865 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2866 { 2867 return skb->transport_header != (typeof(skb->transport_header))~0U; 2868 } 2869 2870 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2871 { 2872 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 2873 return skb->head + skb->transport_header; 2874 } 2875 2876 static inline void skb_reset_transport_header(struct sk_buff *skb) 2877 { 2878 skb->transport_header = skb->data - skb->head; 2879 } 2880 2881 static inline void skb_set_transport_header(struct sk_buff *skb, 2882 const int offset) 2883 { 2884 skb_reset_transport_header(skb); 2885 skb->transport_header += offset; 2886 } 2887 2888 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2889 { 2890 return skb->head + skb->network_header; 2891 } 2892 2893 static inline void skb_reset_network_header(struct sk_buff *skb) 2894 { 2895 skb->network_header = skb->data - skb->head; 2896 } 2897 2898 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2899 { 2900 skb_reset_network_header(skb); 2901 skb->network_header += offset; 2902 } 2903 2904 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2905 { 2906 return skb->mac_header != (typeof(skb->mac_header))~0U; 2907 } 2908 2909 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2910 { 2911 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2912 return skb->head + skb->mac_header; 2913 } 2914 2915 static inline int skb_mac_offset(const struct sk_buff *skb) 2916 { 2917 return skb_mac_header(skb) - skb->data; 2918 } 2919 2920 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2921 { 2922 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2923 return skb->network_header - skb->mac_header; 2924 } 2925 2926 static inline void skb_unset_mac_header(struct sk_buff *skb) 2927 { 2928 skb->mac_header = (typeof(skb->mac_header))~0U; 2929 } 2930 2931 static inline void skb_reset_mac_header(struct sk_buff *skb) 2932 { 2933 skb->mac_header = skb->data - skb->head; 2934 } 2935 2936 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2937 { 2938 skb_reset_mac_header(skb); 2939 skb->mac_header += offset; 2940 } 2941 2942 static inline void skb_pop_mac_header(struct sk_buff *skb) 2943 { 2944 skb->mac_header = skb->network_header; 2945 } 2946 2947 static inline void skb_probe_transport_header(struct sk_buff *skb) 2948 { 2949 struct flow_keys_basic keys; 2950 2951 if (skb_transport_header_was_set(skb)) 2952 return; 2953 2954 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2955 NULL, 0, 0, 0, 0)) 2956 skb_set_transport_header(skb, keys.control.thoff); 2957 } 2958 2959 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2960 { 2961 if (skb_mac_header_was_set(skb)) { 2962 const unsigned char *old_mac = skb_mac_header(skb); 2963 2964 skb_set_mac_header(skb, -skb->mac_len); 2965 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2966 } 2967 } 2968 2969 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2970 { 2971 return skb->csum_start - skb_headroom(skb); 2972 } 2973 2974 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2975 { 2976 return skb->head + skb->csum_start; 2977 } 2978 2979 static inline int skb_transport_offset(const struct sk_buff *skb) 2980 { 2981 return skb_transport_header(skb) - skb->data; 2982 } 2983 2984 static inline u32 skb_network_header_len(const struct sk_buff *skb) 2985 { 2986 return skb->transport_header - skb->network_header; 2987 } 2988 2989 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2990 { 2991 return skb->inner_transport_header - skb->inner_network_header; 2992 } 2993 2994 static inline int skb_network_offset(const struct sk_buff *skb) 2995 { 2996 return skb_network_header(skb) - skb->data; 2997 } 2998 2999 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3000 { 3001 return skb_inner_network_header(skb) - skb->data; 3002 } 3003 3004 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3005 { 3006 return pskb_may_pull(skb, skb_network_offset(skb) + len); 3007 } 3008 3009 /* 3010 * CPUs often take a performance hit when accessing unaligned memory 3011 * locations. The actual performance hit varies, it can be small if the 3012 * hardware handles it or large if we have to take an exception and fix it 3013 * in software. 3014 * 3015 * Since an ethernet header is 14 bytes network drivers often end up with 3016 * the IP header at an unaligned offset. The IP header can be aligned by 3017 * shifting the start of the packet by 2 bytes. Drivers should do this 3018 * with: 3019 * 3020 * skb_reserve(skb, NET_IP_ALIGN); 3021 * 3022 * The downside to this alignment of the IP header is that the DMA is now 3023 * unaligned. On some architectures the cost of an unaligned DMA is high 3024 * and this cost outweighs the gains made by aligning the IP header. 3025 * 3026 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3027 * to be overridden. 3028 */ 3029 #ifndef NET_IP_ALIGN 3030 #define NET_IP_ALIGN 2 3031 #endif 3032 3033 /* 3034 * The networking layer reserves some headroom in skb data (via 3035 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3036 * the header has to grow. In the default case, if the header has to grow 3037 * 32 bytes or less we avoid the reallocation. 3038 * 3039 * Unfortunately this headroom changes the DMA alignment of the resulting 3040 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3041 * on some architectures. An architecture can override this value, 3042 * perhaps setting it to a cacheline in size (since that will maintain 3043 * cacheline alignment of the DMA). It must be a power of 2. 3044 * 3045 * Various parts of the networking layer expect at least 32 bytes of 3046 * headroom, you should not reduce this. 3047 * 3048 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3049 * to reduce average number of cache lines per packet. 3050 * get_rps_cpu() for example only access one 64 bytes aligned block : 3051 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3052 */ 3053 #ifndef NET_SKB_PAD 3054 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3055 #endif 3056 3057 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3058 3059 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3060 { 3061 if (WARN_ON(skb_is_nonlinear(skb))) 3062 return; 3063 skb->len = len; 3064 skb_set_tail_pointer(skb, len); 3065 } 3066 3067 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3068 { 3069 __skb_set_length(skb, len); 3070 } 3071 3072 void skb_trim(struct sk_buff *skb, unsigned int len); 3073 3074 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3075 { 3076 if (skb->data_len) 3077 return ___pskb_trim(skb, len); 3078 __skb_trim(skb, len); 3079 return 0; 3080 } 3081 3082 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3083 { 3084 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3085 } 3086 3087 /** 3088 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3089 * @skb: buffer to alter 3090 * @len: new length 3091 * 3092 * This is identical to pskb_trim except that the caller knows that 3093 * the skb is not cloned so we should never get an error due to out- 3094 * of-memory. 3095 */ 3096 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3097 { 3098 int err = pskb_trim(skb, len); 3099 BUG_ON(err); 3100 } 3101 3102 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3103 { 3104 unsigned int diff = len - skb->len; 3105 3106 if (skb_tailroom(skb) < diff) { 3107 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3108 GFP_ATOMIC); 3109 if (ret) 3110 return ret; 3111 } 3112 __skb_set_length(skb, len); 3113 return 0; 3114 } 3115 3116 /** 3117 * skb_orphan - orphan a buffer 3118 * @skb: buffer to orphan 3119 * 3120 * If a buffer currently has an owner then we call the owner's 3121 * destructor function and make the @skb unowned. The buffer continues 3122 * to exist but is no longer charged to its former owner. 3123 */ 3124 static inline void skb_orphan(struct sk_buff *skb) 3125 { 3126 if (skb->destructor) { 3127 skb->destructor(skb); 3128 skb->destructor = NULL; 3129 skb->sk = NULL; 3130 } else { 3131 BUG_ON(skb->sk); 3132 } 3133 } 3134 3135 /** 3136 * skb_orphan_frags - orphan the frags contained in a buffer 3137 * @skb: buffer to orphan frags from 3138 * @gfp_mask: allocation mask for replacement pages 3139 * 3140 * For each frag in the SKB which needs a destructor (i.e. has an 3141 * owner) create a copy of that frag and release the original 3142 * page by calling the destructor. 3143 */ 3144 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3145 { 3146 if (likely(!skb_zcopy(skb))) 3147 return 0; 3148 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3149 return 0; 3150 return skb_copy_ubufs(skb, gfp_mask); 3151 } 3152 3153 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3154 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3155 { 3156 if (likely(!skb_zcopy(skb))) 3157 return 0; 3158 return skb_copy_ubufs(skb, gfp_mask); 3159 } 3160 3161 /** 3162 * __skb_queue_purge_reason - empty a list 3163 * @list: list to empty 3164 * @reason: drop reason 3165 * 3166 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3167 * the list and one reference dropped. This function does not take the 3168 * list lock and the caller must hold the relevant locks to use it. 3169 */ 3170 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3171 enum skb_drop_reason reason) 3172 { 3173 struct sk_buff *skb; 3174 3175 while ((skb = __skb_dequeue(list)) != NULL) 3176 kfree_skb_reason(skb, reason); 3177 } 3178 3179 static inline void __skb_queue_purge(struct sk_buff_head *list) 3180 { 3181 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3182 } 3183 3184 void skb_queue_purge_reason(struct sk_buff_head *list, 3185 enum skb_drop_reason reason); 3186 3187 static inline void skb_queue_purge(struct sk_buff_head *list) 3188 { 3189 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3190 } 3191 3192 unsigned int skb_rbtree_purge(struct rb_root *root); 3193 void skb_errqueue_purge(struct sk_buff_head *list); 3194 3195 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3196 3197 /** 3198 * netdev_alloc_frag - allocate a page fragment 3199 * @fragsz: fragment size 3200 * 3201 * Allocates a frag from a page for receive buffer. 3202 * Uses GFP_ATOMIC allocations. 3203 */ 3204 static inline void *netdev_alloc_frag(unsigned int fragsz) 3205 { 3206 return __netdev_alloc_frag_align(fragsz, ~0u); 3207 } 3208 3209 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3210 unsigned int align) 3211 { 3212 WARN_ON_ONCE(!is_power_of_2(align)); 3213 return __netdev_alloc_frag_align(fragsz, -align); 3214 } 3215 3216 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3217 gfp_t gfp_mask); 3218 3219 /** 3220 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3221 * @dev: network device to receive on 3222 * @length: length to allocate 3223 * 3224 * Allocate a new &sk_buff and assign it a usage count of one. The 3225 * buffer has unspecified headroom built in. Users should allocate 3226 * the headroom they think they need without accounting for the 3227 * built in space. The built in space is used for optimisations. 3228 * 3229 * %NULL is returned if there is no free memory. Although this function 3230 * allocates memory it can be called from an interrupt. 3231 */ 3232 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3233 unsigned int length) 3234 { 3235 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3236 } 3237 3238 /* legacy helper around __netdev_alloc_skb() */ 3239 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3240 gfp_t gfp_mask) 3241 { 3242 return __netdev_alloc_skb(NULL, length, gfp_mask); 3243 } 3244 3245 /* legacy helper around netdev_alloc_skb() */ 3246 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3247 { 3248 return netdev_alloc_skb(NULL, length); 3249 } 3250 3251 3252 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3253 unsigned int length, gfp_t gfp) 3254 { 3255 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3256 3257 if (NET_IP_ALIGN && skb) 3258 skb_reserve(skb, NET_IP_ALIGN); 3259 return skb; 3260 } 3261 3262 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3263 unsigned int length) 3264 { 3265 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3266 } 3267 3268 static inline void skb_free_frag(void *addr) 3269 { 3270 page_frag_free(addr); 3271 } 3272 3273 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3274 3275 static inline void *napi_alloc_frag(unsigned int fragsz) 3276 { 3277 return __napi_alloc_frag_align(fragsz, ~0u); 3278 } 3279 3280 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3281 unsigned int align) 3282 { 3283 WARN_ON_ONCE(!is_power_of_2(align)); 3284 return __napi_alloc_frag_align(fragsz, -align); 3285 } 3286 3287 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 3288 unsigned int length, gfp_t gfp_mask); 3289 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 3290 unsigned int length) 3291 { 3292 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 3293 } 3294 void napi_consume_skb(struct sk_buff *skb, int budget); 3295 3296 void napi_skb_free_stolen_head(struct sk_buff *skb); 3297 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3298 3299 /** 3300 * __dev_alloc_pages - allocate page for network Rx 3301 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3302 * @order: size of the allocation 3303 * 3304 * Allocate a new page. 3305 * 3306 * %NULL is returned if there is no free memory. 3307 */ 3308 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 3309 unsigned int order) 3310 { 3311 /* This piece of code contains several assumptions. 3312 * 1. This is for device Rx, therefore a cold page is preferred. 3313 * 2. The expectation is the user wants a compound page. 3314 * 3. If requesting a order 0 page it will not be compound 3315 * due to the check to see if order has a value in prep_new_page 3316 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3317 * code in gfp_to_alloc_flags that should be enforcing this. 3318 */ 3319 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3320 3321 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 3322 } 3323 3324 static inline struct page *dev_alloc_pages(unsigned int order) 3325 { 3326 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 3327 } 3328 3329 /** 3330 * __dev_alloc_page - allocate a page for network Rx 3331 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3332 * 3333 * Allocate a new page. 3334 * 3335 * %NULL is returned if there is no free memory. 3336 */ 3337 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 3338 { 3339 return __dev_alloc_pages(gfp_mask, 0); 3340 } 3341 3342 static inline struct page *dev_alloc_page(void) 3343 { 3344 return dev_alloc_pages(0); 3345 } 3346 3347 /** 3348 * dev_page_is_reusable - check whether a page can be reused for network Rx 3349 * @page: the page to test 3350 * 3351 * A page shouldn't be considered for reusing/recycling if it was allocated 3352 * under memory pressure or at a distant memory node. 3353 * 3354 * Returns false if this page should be returned to page allocator, true 3355 * otherwise. 3356 */ 3357 static inline bool dev_page_is_reusable(const struct page *page) 3358 { 3359 return likely(page_to_nid(page) == numa_mem_id() && 3360 !page_is_pfmemalloc(page)); 3361 } 3362 3363 /** 3364 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3365 * @page: The page that was allocated from skb_alloc_page 3366 * @skb: The skb that may need pfmemalloc set 3367 */ 3368 static inline void skb_propagate_pfmemalloc(const struct page *page, 3369 struct sk_buff *skb) 3370 { 3371 if (page_is_pfmemalloc(page)) 3372 skb->pfmemalloc = true; 3373 } 3374 3375 /** 3376 * skb_frag_off() - Returns the offset of a skb fragment 3377 * @frag: the paged fragment 3378 */ 3379 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3380 { 3381 return frag->bv_offset; 3382 } 3383 3384 /** 3385 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3386 * @frag: skb fragment 3387 * @delta: value to add 3388 */ 3389 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3390 { 3391 frag->bv_offset += delta; 3392 } 3393 3394 /** 3395 * skb_frag_off_set() - Sets the offset of a skb fragment 3396 * @frag: skb fragment 3397 * @offset: offset of fragment 3398 */ 3399 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3400 { 3401 frag->bv_offset = offset; 3402 } 3403 3404 /** 3405 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3406 * @fragto: skb fragment where offset is set 3407 * @fragfrom: skb fragment offset is copied from 3408 */ 3409 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3410 const skb_frag_t *fragfrom) 3411 { 3412 fragto->bv_offset = fragfrom->bv_offset; 3413 } 3414 3415 /** 3416 * skb_frag_page - retrieve the page referred to by a paged fragment 3417 * @frag: the paged fragment 3418 * 3419 * Returns the &struct page associated with @frag. 3420 */ 3421 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3422 { 3423 return frag->bv_page; 3424 } 3425 3426 /** 3427 * __skb_frag_ref - take an addition reference on a paged fragment. 3428 * @frag: the paged fragment 3429 * 3430 * Takes an additional reference on the paged fragment @frag. 3431 */ 3432 static inline void __skb_frag_ref(skb_frag_t *frag) 3433 { 3434 get_page(skb_frag_page(frag)); 3435 } 3436 3437 /** 3438 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 3439 * @skb: the buffer 3440 * @f: the fragment offset. 3441 * 3442 * Takes an additional reference on the @f'th paged fragment of @skb. 3443 */ 3444 static inline void skb_frag_ref(struct sk_buff *skb, int f) 3445 { 3446 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 3447 } 3448 3449 bool napi_pp_put_page(struct page *page, bool napi_safe); 3450 3451 static inline void 3452 napi_frag_unref(skb_frag_t *frag, bool recycle, bool napi_safe) 3453 { 3454 struct page *page = skb_frag_page(frag); 3455 3456 #ifdef CONFIG_PAGE_POOL 3457 if (recycle && napi_pp_put_page(page, napi_safe)) 3458 return; 3459 #endif 3460 put_page(page); 3461 } 3462 3463 /** 3464 * __skb_frag_unref - release a reference on a paged fragment. 3465 * @frag: the paged fragment 3466 * @recycle: recycle the page if allocated via page_pool 3467 * 3468 * Releases a reference on the paged fragment @frag 3469 * or recycles the page via the page_pool API. 3470 */ 3471 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle) 3472 { 3473 napi_frag_unref(frag, recycle, false); 3474 } 3475 3476 /** 3477 * skb_frag_unref - release a reference on a paged fragment of an skb. 3478 * @skb: the buffer 3479 * @f: the fragment offset 3480 * 3481 * Releases a reference on the @f'th paged fragment of @skb. 3482 */ 3483 static inline void skb_frag_unref(struct sk_buff *skb, int f) 3484 { 3485 struct skb_shared_info *shinfo = skb_shinfo(skb); 3486 3487 if (!skb_zcopy_managed(skb)) 3488 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle); 3489 } 3490 3491 /** 3492 * skb_frag_address - gets the address of the data contained in a paged fragment 3493 * @frag: the paged fragment buffer 3494 * 3495 * Returns the address of the data within @frag. The page must already 3496 * be mapped. 3497 */ 3498 static inline void *skb_frag_address(const skb_frag_t *frag) 3499 { 3500 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3501 } 3502 3503 /** 3504 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3505 * @frag: the paged fragment buffer 3506 * 3507 * Returns the address of the data within @frag. Checks that the page 3508 * is mapped and returns %NULL otherwise. 3509 */ 3510 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3511 { 3512 void *ptr = page_address(skb_frag_page(frag)); 3513 if (unlikely(!ptr)) 3514 return NULL; 3515 3516 return ptr + skb_frag_off(frag); 3517 } 3518 3519 /** 3520 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3521 * @fragto: skb fragment where page is set 3522 * @fragfrom: skb fragment page is copied from 3523 */ 3524 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3525 const skb_frag_t *fragfrom) 3526 { 3527 fragto->bv_page = fragfrom->bv_page; 3528 } 3529 3530 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3531 3532 /** 3533 * skb_frag_dma_map - maps a paged fragment via the DMA API 3534 * @dev: the device to map the fragment to 3535 * @frag: the paged fragment to map 3536 * @offset: the offset within the fragment (starting at the 3537 * fragment's own offset) 3538 * @size: the number of bytes to map 3539 * @dir: the direction of the mapping (``PCI_DMA_*``) 3540 * 3541 * Maps the page associated with @frag to @device. 3542 */ 3543 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3544 const skb_frag_t *frag, 3545 size_t offset, size_t size, 3546 enum dma_data_direction dir) 3547 { 3548 return dma_map_page(dev, skb_frag_page(frag), 3549 skb_frag_off(frag) + offset, size, dir); 3550 } 3551 3552 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3553 gfp_t gfp_mask) 3554 { 3555 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3556 } 3557 3558 3559 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3560 gfp_t gfp_mask) 3561 { 3562 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3563 } 3564 3565 3566 /** 3567 * skb_clone_writable - is the header of a clone writable 3568 * @skb: buffer to check 3569 * @len: length up to which to write 3570 * 3571 * Returns true if modifying the header part of the cloned buffer 3572 * does not requires the data to be copied. 3573 */ 3574 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3575 { 3576 return !skb_header_cloned(skb) && 3577 skb_headroom(skb) + len <= skb->hdr_len; 3578 } 3579 3580 static inline int skb_try_make_writable(struct sk_buff *skb, 3581 unsigned int write_len) 3582 { 3583 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3584 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3585 } 3586 3587 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3588 int cloned) 3589 { 3590 int delta = 0; 3591 3592 if (headroom > skb_headroom(skb)) 3593 delta = headroom - skb_headroom(skb); 3594 3595 if (delta || cloned) 3596 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3597 GFP_ATOMIC); 3598 return 0; 3599 } 3600 3601 /** 3602 * skb_cow - copy header of skb when it is required 3603 * @skb: buffer to cow 3604 * @headroom: needed headroom 3605 * 3606 * If the skb passed lacks sufficient headroom or its data part 3607 * is shared, data is reallocated. If reallocation fails, an error 3608 * is returned and original skb is not changed. 3609 * 3610 * The result is skb with writable area skb->head...skb->tail 3611 * and at least @headroom of space at head. 3612 */ 3613 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3614 { 3615 return __skb_cow(skb, headroom, skb_cloned(skb)); 3616 } 3617 3618 /** 3619 * skb_cow_head - skb_cow but only making the head writable 3620 * @skb: buffer to cow 3621 * @headroom: needed headroom 3622 * 3623 * This function is identical to skb_cow except that we replace the 3624 * skb_cloned check by skb_header_cloned. It should be used when 3625 * you only need to push on some header and do not need to modify 3626 * the data. 3627 */ 3628 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3629 { 3630 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3631 } 3632 3633 /** 3634 * skb_padto - pad an skbuff up to a minimal size 3635 * @skb: buffer to pad 3636 * @len: minimal length 3637 * 3638 * Pads up a buffer to ensure the trailing bytes exist and are 3639 * blanked. If the buffer already contains sufficient data it 3640 * is untouched. Otherwise it is extended. Returns zero on 3641 * success. The skb is freed on error. 3642 */ 3643 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3644 { 3645 unsigned int size = skb->len; 3646 if (likely(size >= len)) 3647 return 0; 3648 return skb_pad(skb, len - size); 3649 } 3650 3651 /** 3652 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3653 * @skb: buffer to pad 3654 * @len: minimal length 3655 * @free_on_error: free buffer on error 3656 * 3657 * Pads up a buffer to ensure the trailing bytes exist and are 3658 * blanked. If the buffer already contains sufficient data it 3659 * is untouched. Otherwise it is extended. Returns zero on 3660 * success. The skb is freed on error if @free_on_error is true. 3661 */ 3662 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3663 unsigned int len, 3664 bool free_on_error) 3665 { 3666 unsigned int size = skb->len; 3667 3668 if (unlikely(size < len)) { 3669 len -= size; 3670 if (__skb_pad(skb, len, free_on_error)) 3671 return -ENOMEM; 3672 __skb_put(skb, len); 3673 } 3674 return 0; 3675 } 3676 3677 /** 3678 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3679 * @skb: buffer to pad 3680 * @len: minimal length 3681 * 3682 * Pads up a buffer to ensure the trailing bytes exist and are 3683 * blanked. If the buffer already contains sufficient data it 3684 * is untouched. Otherwise it is extended. Returns zero on 3685 * success. The skb is freed on error. 3686 */ 3687 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3688 { 3689 return __skb_put_padto(skb, len, true); 3690 } 3691 3692 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3693 __must_check; 3694 3695 static inline int skb_add_data(struct sk_buff *skb, 3696 struct iov_iter *from, int copy) 3697 { 3698 const int off = skb->len; 3699 3700 if (skb->ip_summed == CHECKSUM_NONE) { 3701 __wsum csum = 0; 3702 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3703 &csum, from)) { 3704 skb->csum = csum_block_add(skb->csum, csum, off); 3705 return 0; 3706 } 3707 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3708 return 0; 3709 3710 __skb_trim(skb, off); 3711 return -EFAULT; 3712 } 3713 3714 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3715 const struct page *page, int off) 3716 { 3717 if (skb_zcopy(skb)) 3718 return false; 3719 if (i) { 3720 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3721 3722 return page == skb_frag_page(frag) && 3723 off == skb_frag_off(frag) + skb_frag_size(frag); 3724 } 3725 return false; 3726 } 3727 3728 static inline int __skb_linearize(struct sk_buff *skb) 3729 { 3730 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3731 } 3732 3733 /** 3734 * skb_linearize - convert paged skb to linear one 3735 * @skb: buffer to linarize 3736 * 3737 * If there is no free memory -ENOMEM is returned, otherwise zero 3738 * is returned and the old skb data released. 3739 */ 3740 static inline int skb_linearize(struct sk_buff *skb) 3741 { 3742 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3743 } 3744 3745 /** 3746 * skb_has_shared_frag - can any frag be overwritten 3747 * @skb: buffer to test 3748 * 3749 * Return true if the skb has at least one frag that might be modified 3750 * by an external entity (as in vmsplice()/sendfile()) 3751 */ 3752 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3753 { 3754 return skb_is_nonlinear(skb) && 3755 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3756 } 3757 3758 /** 3759 * skb_linearize_cow - make sure skb is linear and writable 3760 * @skb: buffer to process 3761 * 3762 * If there is no free memory -ENOMEM is returned, otherwise zero 3763 * is returned and the old skb data released. 3764 */ 3765 static inline int skb_linearize_cow(struct sk_buff *skb) 3766 { 3767 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3768 __skb_linearize(skb) : 0; 3769 } 3770 3771 static __always_inline void 3772 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3773 unsigned int off) 3774 { 3775 if (skb->ip_summed == CHECKSUM_COMPLETE) 3776 skb->csum = csum_block_sub(skb->csum, 3777 csum_partial(start, len, 0), off); 3778 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3779 skb_checksum_start_offset(skb) < 0) 3780 skb->ip_summed = CHECKSUM_NONE; 3781 } 3782 3783 /** 3784 * skb_postpull_rcsum - update checksum for received skb after pull 3785 * @skb: buffer to update 3786 * @start: start of data before pull 3787 * @len: length of data pulled 3788 * 3789 * After doing a pull on a received packet, you need to call this to 3790 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3791 * CHECKSUM_NONE so that it can be recomputed from scratch. 3792 */ 3793 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3794 const void *start, unsigned int len) 3795 { 3796 if (skb->ip_summed == CHECKSUM_COMPLETE) 3797 skb->csum = wsum_negate(csum_partial(start, len, 3798 wsum_negate(skb->csum))); 3799 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3800 skb_checksum_start_offset(skb) < 0) 3801 skb->ip_summed = CHECKSUM_NONE; 3802 } 3803 3804 static __always_inline void 3805 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3806 unsigned int off) 3807 { 3808 if (skb->ip_summed == CHECKSUM_COMPLETE) 3809 skb->csum = csum_block_add(skb->csum, 3810 csum_partial(start, len, 0), off); 3811 } 3812 3813 /** 3814 * skb_postpush_rcsum - update checksum for received skb after push 3815 * @skb: buffer to update 3816 * @start: start of data after push 3817 * @len: length of data pushed 3818 * 3819 * After doing a push on a received packet, you need to call this to 3820 * update the CHECKSUM_COMPLETE checksum. 3821 */ 3822 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3823 const void *start, unsigned int len) 3824 { 3825 __skb_postpush_rcsum(skb, start, len, 0); 3826 } 3827 3828 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3829 3830 /** 3831 * skb_push_rcsum - push skb and update receive checksum 3832 * @skb: buffer to update 3833 * @len: length of data pulled 3834 * 3835 * This function performs an skb_push on the packet and updates 3836 * the CHECKSUM_COMPLETE checksum. It should be used on 3837 * receive path processing instead of skb_push unless you know 3838 * that the checksum difference is zero (e.g., a valid IP header) 3839 * or you are setting ip_summed to CHECKSUM_NONE. 3840 */ 3841 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3842 { 3843 skb_push(skb, len); 3844 skb_postpush_rcsum(skb, skb->data, len); 3845 return skb->data; 3846 } 3847 3848 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3849 /** 3850 * pskb_trim_rcsum - trim received skb and update checksum 3851 * @skb: buffer to trim 3852 * @len: new length 3853 * 3854 * This is exactly the same as pskb_trim except that it ensures the 3855 * checksum of received packets are still valid after the operation. 3856 * It can change skb pointers. 3857 */ 3858 3859 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3860 { 3861 if (likely(len >= skb->len)) 3862 return 0; 3863 return pskb_trim_rcsum_slow(skb, len); 3864 } 3865 3866 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3867 { 3868 if (skb->ip_summed == CHECKSUM_COMPLETE) 3869 skb->ip_summed = CHECKSUM_NONE; 3870 __skb_trim(skb, len); 3871 return 0; 3872 } 3873 3874 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3875 { 3876 if (skb->ip_summed == CHECKSUM_COMPLETE) 3877 skb->ip_summed = CHECKSUM_NONE; 3878 return __skb_grow(skb, len); 3879 } 3880 3881 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3882 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3883 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3884 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3885 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3886 3887 #define skb_queue_walk(queue, skb) \ 3888 for (skb = (queue)->next; \ 3889 skb != (struct sk_buff *)(queue); \ 3890 skb = skb->next) 3891 3892 #define skb_queue_walk_safe(queue, skb, tmp) \ 3893 for (skb = (queue)->next, tmp = skb->next; \ 3894 skb != (struct sk_buff *)(queue); \ 3895 skb = tmp, tmp = skb->next) 3896 3897 #define skb_queue_walk_from(queue, skb) \ 3898 for (; skb != (struct sk_buff *)(queue); \ 3899 skb = skb->next) 3900 3901 #define skb_rbtree_walk(skb, root) \ 3902 for (skb = skb_rb_first(root); skb != NULL; \ 3903 skb = skb_rb_next(skb)) 3904 3905 #define skb_rbtree_walk_from(skb) \ 3906 for (; skb != NULL; \ 3907 skb = skb_rb_next(skb)) 3908 3909 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3910 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3911 skb = tmp) 3912 3913 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3914 for (tmp = skb->next; \ 3915 skb != (struct sk_buff *)(queue); \ 3916 skb = tmp, tmp = skb->next) 3917 3918 #define skb_queue_reverse_walk(queue, skb) \ 3919 for (skb = (queue)->prev; \ 3920 skb != (struct sk_buff *)(queue); \ 3921 skb = skb->prev) 3922 3923 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3924 for (skb = (queue)->prev, tmp = skb->prev; \ 3925 skb != (struct sk_buff *)(queue); \ 3926 skb = tmp, tmp = skb->prev) 3927 3928 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3929 for (tmp = skb->prev; \ 3930 skb != (struct sk_buff *)(queue); \ 3931 skb = tmp, tmp = skb->prev) 3932 3933 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3934 { 3935 return skb_shinfo(skb)->frag_list != NULL; 3936 } 3937 3938 static inline void skb_frag_list_init(struct sk_buff *skb) 3939 { 3940 skb_shinfo(skb)->frag_list = NULL; 3941 } 3942 3943 #define skb_walk_frags(skb, iter) \ 3944 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3945 3946 3947 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3948 int *err, long *timeo_p, 3949 const struct sk_buff *skb); 3950 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3951 struct sk_buff_head *queue, 3952 unsigned int flags, 3953 int *off, int *err, 3954 struct sk_buff **last); 3955 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3956 struct sk_buff_head *queue, 3957 unsigned int flags, int *off, int *err, 3958 struct sk_buff **last); 3959 struct sk_buff *__skb_recv_datagram(struct sock *sk, 3960 struct sk_buff_head *sk_queue, 3961 unsigned int flags, int *off, int *err); 3962 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 3963 __poll_t datagram_poll(struct file *file, struct socket *sock, 3964 struct poll_table_struct *wait); 3965 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3966 struct iov_iter *to, int size); 3967 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3968 struct msghdr *msg, int size) 3969 { 3970 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3971 } 3972 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3973 struct msghdr *msg); 3974 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3975 struct iov_iter *to, int len, 3976 struct ahash_request *hash); 3977 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3978 struct iov_iter *from, int len); 3979 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3980 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3981 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3982 static inline void skb_free_datagram_locked(struct sock *sk, 3983 struct sk_buff *skb) 3984 { 3985 __skb_free_datagram_locked(sk, skb, 0); 3986 } 3987 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3988 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3989 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3990 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3991 int len); 3992 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3993 struct pipe_inode_info *pipe, unsigned int len, 3994 unsigned int flags); 3995 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3996 int len); 3997 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3998 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3999 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4000 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4001 int len, int hlen); 4002 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4003 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4004 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4005 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4006 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4007 unsigned int offset); 4008 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4009 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4010 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4011 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4012 int skb_vlan_pop(struct sk_buff *skb); 4013 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4014 int skb_eth_pop(struct sk_buff *skb); 4015 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4016 const unsigned char *src); 4017 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4018 int mac_len, bool ethernet); 4019 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4020 bool ethernet); 4021 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4022 int skb_mpls_dec_ttl(struct sk_buff *skb); 4023 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4024 gfp_t gfp); 4025 4026 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4027 { 4028 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4029 } 4030 4031 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4032 { 4033 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4034 } 4035 4036 struct skb_checksum_ops { 4037 __wsum (*update)(const void *mem, int len, __wsum wsum); 4038 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 4039 }; 4040 4041 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 4042 4043 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 4044 __wsum csum, const struct skb_checksum_ops *ops); 4045 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4046 __wsum csum); 4047 4048 static inline void * __must_check 4049 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4050 const void *data, int hlen, void *buffer) 4051 { 4052 if (likely(hlen - offset >= len)) 4053 return (void *)data + offset; 4054 4055 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4056 return NULL; 4057 4058 return buffer; 4059 } 4060 4061 static inline void * __must_check 4062 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4063 { 4064 return __skb_header_pointer(skb, offset, len, skb->data, 4065 skb_headlen(skb), buffer); 4066 } 4067 4068 static inline void * __must_check 4069 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4070 { 4071 if (likely(skb_headlen(skb) - offset >= len)) 4072 return skb->data + offset; 4073 return NULL; 4074 } 4075 4076 /** 4077 * skb_needs_linearize - check if we need to linearize a given skb 4078 * depending on the given device features. 4079 * @skb: socket buffer to check 4080 * @features: net device features 4081 * 4082 * Returns true if either: 4083 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4084 * 2. skb is fragmented and the device does not support SG. 4085 */ 4086 static inline bool skb_needs_linearize(struct sk_buff *skb, 4087 netdev_features_t features) 4088 { 4089 return skb_is_nonlinear(skb) && 4090 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4091 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4092 } 4093 4094 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4095 void *to, 4096 const unsigned int len) 4097 { 4098 memcpy(to, skb->data, len); 4099 } 4100 4101 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4102 const int offset, void *to, 4103 const unsigned int len) 4104 { 4105 memcpy(to, skb->data + offset, len); 4106 } 4107 4108 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4109 const void *from, 4110 const unsigned int len) 4111 { 4112 memcpy(skb->data, from, len); 4113 } 4114 4115 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4116 const int offset, 4117 const void *from, 4118 const unsigned int len) 4119 { 4120 memcpy(skb->data + offset, from, len); 4121 } 4122 4123 void skb_init(void); 4124 4125 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4126 { 4127 return skb->tstamp; 4128 } 4129 4130 /** 4131 * skb_get_timestamp - get timestamp from a skb 4132 * @skb: skb to get stamp from 4133 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4134 * 4135 * Timestamps are stored in the skb as offsets to a base timestamp. 4136 * This function converts the offset back to a struct timeval and stores 4137 * it in stamp. 4138 */ 4139 static inline void skb_get_timestamp(const struct sk_buff *skb, 4140 struct __kernel_old_timeval *stamp) 4141 { 4142 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4143 } 4144 4145 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4146 struct __kernel_sock_timeval *stamp) 4147 { 4148 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4149 4150 stamp->tv_sec = ts.tv_sec; 4151 stamp->tv_usec = ts.tv_nsec / 1000; 4152 } 4153 4154 static inline void skb_get_timestampns(const struct sk_buff *skb, 4155 struct __kernel_old_timespec *stamp) 4156 { 4157 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4158 4159 stamp->tv_sec = ts.tv_sec; 4160 stamp->tv_nsec = ts.tv_nsec; 4161 } 4162 4163 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4164 struct __kernel_timespec *stamp) 4165 { 4166 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4167 4168 stamp->tv_sec = ts.tv_sec; 4169 stamp->tv_nsec = ts.tv_nsec; 4170 } 4171 4172 static inline void __net_timestamp(struct sk_buff *skb) 4173 { 4174 skb->tstamp = ktime_get_real(); 4175 skb->mono_delivery_time = 0; 4176 } 4177 4178 static inline ktime_t net_timedelta(ktime_t t) 4179 { 4180 return ktime_sub(ktime_get_real(), t); 4181 } 4182 4183 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4184 bool mono) 4185 { 4186 skb->tstamp = kt; 4187 skb->mono_delivery_time = kt && mono; 4188 } 4189 4190 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4191 4192 /* It is used in the ingress path to clear the delivery_time. 4193 * If needed, set the skb->tstamp to the (rcv) timestamp. 4194 */ 4195 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4196 { 4197 if (skb->mono_delivery_time) { 4198 skb->mono_delivery_time = 0; 4199 if (static_branch_unlikely(&netstamp_needed_key)) 4200 skb->tstamp = ktime_get_real(); 4201 else 4202 skb->tstamp = 0; 4203 } 4204 } 4205 4206 static inline void skb_clear_tstamp(struct sk_buff *skb) 4207 { 4208 if (skb->mono_delivery_time) 4209 return; 4210 4211 skb->tstamp = 0; 4212 } 4213 4214 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4215 { 4216 if (skb->mono_delivery_time) 4217 return 0; 4218 4219 return skb->tstamp; 4220 } 4221 4222 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4223 { 4224 if (!skb->mono_delivery_time && skb->tstamp) 4225 return skb->tstamp; 4226 4227 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4228 return ktime_get_real(); 4229 4230 return 0; 4231 } 4232 4233 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4234 { 4235 return skb_shinfo(skb)->meta_len; 4236 } 4237 4238 static inline void *skb_metadata_end(const struct sk_buff *skb) 4239 { 4240 return skb_mac_header(skb); 4241 } 4242 4243 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4244 const struct sk_buff *skb_b, 4245 u8 meta_len) 4246 { 4247 const void *a = skb_metadata_end(skb_a); 4248 const void *b = skb_metadata_end(skb_b); 4249 u64 diffs = 0; 4250 4251 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4252 BITS_PER_LONG != 64) 4253 goto slow; 4254 4255 /* Using more efficient variant than plain call to memcmp(). */ 4256 switch (meta_len) { 4257 #define __it(x, op) (x -= sizeof(u##op)) 4258 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4259 case 32: diffs |= __it_diff(a, b, 64); 4260 fallthrough; 4261 case 24: diffs |= __it_diff(a, b, 64); 4262 fallthrough; 4263 case 16: diffs |= __it_diff(a, b, 64); 4264 fallthrough; 4265 case 8: diffs |= __it_diff(a, b, 64); 4266 break; 4267 case 28: diffs |= __it_diff(a, b, 64); 4268 fallthrough; 4269 case 20: diffs |= __it_diff(a, b, 64); 4270 fallthrough; 4271 case 12: diffs |= __it_diff(a, b, 64); 4272 fallthrough; 4273 case 4: diffs |= __it_diff(a, b, 32); 4274 break; 4275 default: 4276 slow: 4277 return memcmp(a - meta_len, b - meta_len, meta_len); 4278 } 4279 return diffs; 4280 } 4281 4282 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4283 const struct sk_buff *skb_b) 4284 { 4285 u8 len_a = skb_metadata_len(skb_a); 4286 u8 len_b = skb_metadata_len(skb_b); 4287 4288 if (!(len_a | len_b)) 4289 return false; 4290 4291 return len_a != len_b ? 4292 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4293 } 4294 4295 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4296 { 4297 skb_shinfo(skb)->meta_len = meta_len; 4298 } 4299 4300 static inline void skb_metadata_clear(struct sk_buff *skb) 4301 { 4302 skb_metadata_set(skb, 0); 4303 } 4304 4305 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4306 4307 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4308 4309 void skb_clone_tx_timestamp(struct sk_buff *skb); 4310 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4311 4312 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4313 4314 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4315 { 4316 } 4317 4318 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4319 { 4320 return false; 4321 } 4322 4323 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4324 4325 /** 4326 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4327 * 4328 * PHY drivers may accept clones of transmitted packets for 4329 * timestamping via their phy_driver.txtstamp method. These drivers 4330 * must call this function to return the skb back to the stack with a 4331 * timestamp. 4332 * 4333 * @skb: clone of the original outgoing packet 4334 * @hwtstamps: hardware time stamps 4335 * 4336 */ 4337 void skb_complete_tx_timestamp(struct sk_buff *skb, 4338 struct skb_shared_hwtstamps *hwtstamps); 4339 4340 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4341 struct skb_shared_hwtstamps *hwtstamps, 4342 struct sock *sk, int tstype); 4343 4344 /** 4345 * skb_tstamp_tx - queue clone of skb with send time stamps 4346 * @orig_skb: the original outgoing packet 4347 * @hwtstamps: hardware time stamps, may be NULL if not available 4348 * 4349 * If the skb has a socket associated, then this function clones the 4350 * skb (thus sharing the actual data and optional structures), stores 4351 * the optional hardware time stamping information (if non NULL) or 4352 * generates a software time stamp (otherwise), then queues the clone 4353 * to the error queue of the socket. Errors are silently ignored. 4354 */ 4355 void skb_tstamp_tx(struct sk_buff *orig_skb, 4356 struct skb_shared_hwtstamps *hwtstamps); 4357 4358 /** 4359 * skb_tx_timestamp() - Driver hook for transmit timestamping 4360 * 4361 * Ethernet MAC Drivers should call this function in their hard_xmit() 4362 * function immediately before giving the sk_buff to the MAC hardware. 4363 * 4364 * Specifically, one should make absolutely sure that this function is 4365 * called before TX completion of this packet can trigger. Otherwise 4366 * the packet could potentially already be freed. 4367 * 4368 * @skb: A socket buffer. 4369 */ 4370 static inline void skb_tx_timestamp(struct sk_buff *skb) 4371 { 4372 skb_clone_tx_timestamp(skb); 4373 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4374 skb_tstamp_tx(skb, NULL); 4375 } 4376 4377 /** 4378 * skb_complete_wifi_ack - deliver skb with wifi status 4379 * 4380 * @skb: the original outgoing packet 4381 * @acked: ack status 4382 * 4383 */ 4384 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4385 4386 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4387 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4388 4389 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4390 { 4391 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4392 skb->csum_valid || 4393 (skb->ip_summed == CHECKSUM_PARTIAL && 4394 skb_checksum_start_offset(skb) >= 0)); 4395 } 4396 4397 /** 4398 * skb_checksum_complete - Calculate checksum of an entire packet 4399 * @skb: packet to process 4400 * 4401 * This function calculates the checksum over the entire packet plus 4402 * the value of skb->csum. The latter can be used to supply the 4403 * checksum of a pseudo header as used by TCP/UDP. It returns the 4404 * checksum. 4405 * 4406 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4407 * this function can be used to verify that checksum on received 4408 * packets. In that case the function should return zero if the 4409 * checksum is correct. In particular, this function will return zero 4410 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4411 * hardware has already verified the correctness of the checksum. 4412 */ 4413 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4414 { 4415 return skb_csum_unnecessary(skb) ? 4416 0 : __skb_checksum_complete(skb); 4417 } 4418 4419 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4420 { 4421 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4422 if (skb->csum_level == 0) 4423 skb->ip_summed = CHECKSUM_NONE; 4424 else 4425 skb->csum_level--; 4426 } 4427 } 4428 4429 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4430 { 4431 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4432 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4433 skb->csum_level++; 4434 } else if (skb->ip_summed == CHECKSUM_NONE) { 4435 skb->ip_summed = CHECKSUM_UNNECESSARY; 4436 skb->csum_level = 0; 4437 } 4438 } 4439 4440 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4441 { 4442 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4443 skb->ip_summed = CHECKSUM_NONE; 4444 skb->csum_level = 0; 4445 } 4446 } 4447 4448 /* Check if we need to perform checksum complete validation. 4449 * 4450 * Returns true if checksum complete is needed, false otherwise 4451 * (either checksum is unnecessary or zero checksum is allowed). 4452 */ 4453 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4454 bool zero_okay, 4455 __sum16 check) 4456 { 4457 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4458 skb->csum_valid = 1; 4459 __skb_decr_checksum_unnecessary(skb); 4460 return false; 4461 } 4462 4463 return true; 4464 } 4465 4466 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4467 * in checksum_init. 4468 */ 4469 #define CHECKSUM_BREAK 76 4470 4471 /* Unset checksum-complete 4472 * 4473 * Unset checksum complete can be done when packet is being modified 4474 * (uncompressed for instance) and checksum-complete value is 4475 * invalidated. 4476 */ 4477 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4478 { 4479 if (skb->ip_summed == CHECKSUM_COMPLETE) 4480 skb->ip_summed = CHECKSUM_NONE; 4481 } 4482 4483 /* Validate (init) checksum based on checksum complete. 4484 * 4485 * Return values: 4486 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4487 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4488 * checksum is stored in skb->csum for use in __skb_checksum_complete 4489 * non-zero: value of invalid checksum 4490 * 4491 */ 4492 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4493 bool complete, 4494 __wsum psum) 4495 { 4496 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4497 if (!csum_fold(csum_add(psum, skb->csum))) { 4498 skb->csum_valid = 1; 4499 return 0; 4500 } 4501 } 4502 4503 skb->csum = psum; 4504 4505 if (complete || skb->len <= CHECKSUM_BREAK) { 4506 __sum16 csum; 4507 4508 csum = __skb_checksum_complete(skb); 4509 skb->csum_valid = !csum; 4510 return csum; 4511 } 4512 4513 return 0; 4514 } 4515 4516 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4517 { 4518 return 0; 4519 } 4520 4521 /* Perform checksum validate (init). Note that this is a macro since we only 4522 * want to calculate the pseudo header which is an input function if necessary. 4523 * First we try to validate without any computation (checksum unnecessary) and 4524 * then calculate based on checksum complete calling the function to compute 4525 * pseudo header. 4526 * 4527 * Return values: 4528 * 0: checksum is validated or try to in skb_checksum_complete 4529 * non-zero: value of invalid checksum 4530 */ 4531 #define __skb_checksum_validate(skb, proto, complete, \ 4532 zero_okay, check, compute_pseudo) \ 4533 ({ \ 4534 __sum16 __ret = 0; \ 4535 skb->csum_valid = 0; \ 4536 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4537 __ret = __skb_checksum_validate_complete(skb, \ 4538 complete, compute_pseudo(skb, proto)); \ 4539 __ret; \ 4540 }) 4541 4542 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4543 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4544 4545 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4546 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4547 4548 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4549 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4550 4551 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4552 compute_pseudo) \ 4553 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4554 4555 #define skb_checksum_simple_validate(skb) \ 4556 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4557 4558 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4559 { 4560 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4561 } 4562 4563 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4564 { 4565 skb->csum = ~pseudo; 4566 skb->ip_summed = CHECKSUM_COMPLETE; 4567 } 4568 4569 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4570 do { \ 4571 if (__skb_checksum_convert_check(skb)) \ 4572 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4573 } while (0) 4574 4575 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4576 u16 start, u16 offset) 4577 { 4578 skb->ip_summed = CHECKSUM_PARTIAL; 4579 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4580 skb->csum_offset = offset - start; 4581 } 4582 4583 /* Update skbuf and packet to reflect the remote checksum offload operation. 4584 * When called, ptr indicates the starting point for skb->csum when 4585 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4586 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4587 */ 4588 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4589 int start, int offset, bool nopartial) 4590 { 4591 __wsum delta; 4592 4593 if (!nopartial) { 4594 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4595 return; 4596 } 4597 4598 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4599 __skb_checksum_complete(skb); 4600 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4601 } 4602 4603 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4604 4605 /* Adjust skb->csum since we changed the packet */ 4606 skb->csum = csum_add(skb->csum, delta); 4607 } 4608 4609 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4610 { 4611 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4612 return (void *)(skb->_nfct & NFCT_PTRMASK); 4613 #else 4614 return NULL; 4615 #endif 4616 } 4617 4618 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4619 { 4620 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4621 return skb->_nfct; 4622 #else 4623 return 0UL; 4624 #endif 4625 } 4626 4627 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4628 { 4629 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4630 skb->slow_gro |= !!nfct; 4631 skb->_nfct = nfct; 4632 #endif 4633 } 4634 4635 #ifdef CONFIG_SKB_EXTENSIONS 4636 enum skb_ext_id { 4637 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4638 SKB_EXT_BRIDGE_NF, 4639 #endif 4640 #ifdef CONFIG_XFRM 4641 SKB_EXT_SEC_PATH, 4642 #endif 4643 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4644 TC_SKB_EXT, 4645 #endif 4646 #if IS_ENABLED(CONFIG_MPTCP) 4647 SKB_EXT_MPTCP, 4648 #endif 4649 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4650 SKB_EXT_MCTP, 4651 #endif 4652 SKB_EXT_NUM, /* must be last */ 4653 }; 4654 4655 /** 4656 * struct skb_ext - sk_buff extensions 4657 * @refcnt: 1 on allocation, deallocated on 0 4658 * @offset: offset to add to @data to obtain extension address 4659 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4660 * @data: start of extension data, variable sized 4661 * 4662 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4663 * to use 'u8' types while allowing up to 2kb worth of extension data. 4664 */ 4665 struct skb_ext { 4666 refcount_t refcnt; 4667 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4668 u8 chunks; /* same */ 4669 char data[] __aligned(8); 4670 }; 4671 4672 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4673 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4674 struct skb_ext *ext); 4675 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4676 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4677 void __skb_ext_put(struct skb_ext *ext); 4678 4679 static inline void skb_ext_put(struct sk_buff *skb) 4680 { 4681 if (skb->active_extensions) 4682 __skb_ext_put(skb->extensions); 4683 } 4684 4685 static inline void __skb_ext_copy(struct sk_buff *dst, 4686 const struct sk_buff *src) 4687 { 4688 dst->active_extensions = src->active_extensions; 4689 4690 if (src->active_extensions) { 4691 struct skb_ext *ext = src->extensions; 4692 4693 refcount_inc(&ext->refcnt); 4694 dst->extensions = ext; 4695 } 4696 } 4697 4698 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4699 { 4700 skb_ext_put(dst); 4701 __skb_ext_copy(dst, src); 4702 } 4703 4704 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4705 { 4706 return !!ext->offset[i]; 4707 } 4708 4709 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4710 { 4711 return skb->active_extensions & (1 << id); 4712 } 4713 4714 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4715 { 4716 if (skb_ext_exist(skb, id)) 4717 __skb_ext_del(skb, id); 4718 } 4719 4720 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4721 { 4722 if (skb_ext_exist(skb, id)) { 4723 struct skb_ext *ext = skb->extensions; 4724 4725 return (void *)ext + (ext->offset[id] << 3); 4726 } 4727 4728 return NULL; 4729 } 4730 4731 static inline void skb_ext_reset(struct sk_buff *skb) 4732 { 4733 if (unlikely(skb->active_extensions)) { 4734 __skb_ext_put(skb->extensions); 4735 skb->active_extensions = 0; 4736 } 4737 } 4738 4739 static inline bool skb_has_extensions(struct sk_buff *skb) 4740 { 4741 return unlikely(skb->active_extensions); 4742 } 4743 #else 4744 static inline void skb_ext_put(struct sk_buff *skb) {} 4745 static inline void skb_ext_reset(struct sk_buff *skb) {} 4746 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4747 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4748 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4749 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4750 #endif /* CONFIG_SKB_EXTENSIONS */ 4751 4752 static inline void nf_reset_ct(struct sk_buff *skb) 4753 { 4754 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4755 nf_conntrack_put(skb_nfct(skb)); 4756 skb->_nfct = 0; 4757 #endif 4758 } 4759 4760 static inline void nf_reset_trace(struct sk_buff *skb) 4761 { 4762 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4763 skb->nf_trace = 0; 4764 #endif 4765 } 4766 4767 static inline void ipvs_reset(struct sk_buff *skb) 4768 { 4769 #if IS_ENABLED(CONFIG_IP_VS) 4770 skb->ipvs_property = 0; 4771 #endif 4772 } 4773 4774 /* Note: This doesn't put any conntrack info in dst. */ 4775 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4776 bool copy) 4777 { 4778 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4779 dst->_nfct = src->_nfct; 4780 nf_conntrack_get(skb_nfct(src)); 4781 #endif 4782 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4783 if (copy) 4784 dst->nf_trace = src->nf_trace; 4785 #endif 4786 } 4787 4788 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4789 { 4790 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4791 nf_conntrack_put(skb_nfct(dst)); 4792 #endif 4793 dst->slow_gro = src->slow_gro; 4794 __nf_copy(dst, src, true); 4795 } 4796 4797 #ifdef CONFIG_NETWORK_SECMARK 4798 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4799 { 4800 to->secmark = from->secmark; 4801 } 4802 4803 static inline void skb_init_secmark(struct sk_buff *skb) 4804 { 4805 skb->secmark = 0; 4806 } 4807 #else 4808 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4809 { } 4810 4811 static inline void skb_init_secmark(struct sk_buff *skb) 4812 { } 4813 #endif 4814 4815 static inline int secpath_exists(const struct sk_buff *skb) 4816 { 4817 #ifdef CONFIG_XFRM 4818 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4819 #else 4820 return 0; 4821 #endif 4822 } 4823 4824 static inline bool skb_irq_freeable(const struct sk_buff *skb) 4825 { 4826 return !skb->destructor && 4827 !secpath_exists(skb) && 4828 !skb_nfct(skb) && 4829 !skb->_skb_refdst && 4830 !skb_has_frag_list(skb); 4831 } 4832 4833 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4834 { 4835 skb->queue_mapping = queue_mapping; 4836 } 4837 4838 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4839 { 4840 return skb->queue_mapping; 4841 } 4842 4843 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4844 { 4845 to->queue_mapping = from->queue_mapping; 4846 } 4847 4848 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4849 { 4850 skb->queue_mapping = rx_queue + 1; 4851 } 4852 4853 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4854 { 4855 return skb->queue_mapping - 1; 4856 } 4857 4858 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4859 { 4860 return skb->queue_mapping != 0; 4861 } 4862 4863 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4864 { 4865 skb->dst_pending_confirm = val; 4866 } 4867 4868 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4869 { 4870 return skb->dst_pending_confirm != 0; 4871 } 4872 4873 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4874 { 4875 #ifdef CONFIG_XFRM 4876 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4877 #else 4878 return NULL; 4879 #endif 4880 } 4881 4882 static inline bool skb_is_gso(const struct sk_buff *skb) 4883 { 4884 return skb_shinfo(skb)->gso_size; 4885 } 4886 4887 /* Note: Should be called only if skb_is_gso(skb) is true */ 4888 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4889 { 4890 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4891 } 4892 4893 /* Note: Should be called only if skb_is_gso(skb) is true */ 4894 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4895 { 4896 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4897 } 4898 4899 /* Note: Should be called only if skb_is_gso(skb) is true */ 4900 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4901 { 4902 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4903 } 4904 4905 static inline void skb_gso_reset(struct sk_buff *skb) 4906 { 4907 skb_shinfo(skb)->gso_size = 0; 4908 skb_shinfo(skb)->gso_segs = 0; 4909 skb_shinfo(skb)->gso_type = 0; 4910 } 4911 4912 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4913 u16 increment) 4914 { 4915 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4916 return; 4917 shinfo->gso_size += increment; 4918 } 4919 4920 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4921 u16 decrement) 4922 { 4923 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4924 return; 4925 shinfo->gso_size -= decrement; 4926 } 4927 4928 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4929 4930 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4931 { 4932 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4933 * wanted then gso_type will be set. */ 4934 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4935 4936 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4937 unlikely(shinfo->gso_type == 0)) { 4938 __skb_warn_lro_forwarding(skb); 4939 return true; 4940 } 4941 return false; 4942 } 4943 4944 static inline void skb_forward_csum(struct sk_buff *skb) 4945 { 4946 /* Unfortunately we don't support this one. Any brave souls? */ 4947 if (skb->ip_summed == CHECKSUM_COMPLETE) 4948 skb->ip_summed = CHECKSUM_NONE; 4949 } 4950 4951 /** 4952 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4953 * @skb: skb to check 4954 * 4955 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4956 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4957 * use this helper, to document places where we make this assertion. 4958 */ 4959 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4960 { 4961 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 4962 } 4963 4964 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4965 4966 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4967 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4968 unsigned int transport_len, 4969 __sum16(*skb_chkf)(struct sk_buff *skb)); 4970 4971 /** 4972 * skb_head_is_locked - Determine if the skb->head is locked down 4973 * @skb: skb to check 4974 * 4975 * The head on skbs build around a head frag can be removed if they are 4976 * not cloned. This function returns true if the skb head is locked down 4977 * due to either being allocated via kmalloc, or by being a clone with 4978 * multiple references to the head. 4979 */ 4980 static inline bool skb_head_is_locked(const struct sk_buff *skb) 4981 { 4982 return !skb->head_frag || skb_cloned(skb); 4983 } 4984 4985 /* Local Checksum Offload. 4986 * Compute outer checksum based on the assumption that the 4987 * inner checksum will be offloaded later. 4988 * See Documentation/networking/checksum-offloads.rst for 4989 * explanation of how this works. 4990 * Fill in outer checksum adjustment (e.g. with sum of outer 4991 * pseudo-header) before calling. 4992 * Also ensure that inner checksum is in linear data area. 4993 */ 4994 static inline __wsum lco_csum(struct sk_buff *skb) 4995 { 4996 unsigned char *csum_start = skb_checksum_start(skb); 4997 unsigned char *l4_hdr = skb_transport_header(skb); 4998 __wsum partial; 4999 5000 /* Start with complement of inner checksum adjustment */ 5001 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5002 skb->csum_offset)); 5003 5004 /* Add in checksum of our headers (incl. outer checksum 5005 * adjustment filled in by caller) and return result. 5006 */ 5007 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5008 } 5009 5010 static inline bool skb_is_redirected(const struct sk_buff *skb) 5011 { 5012 return skb->redirected; 5013 } 5014 5015 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5016 { 5017 skb->redirected = 1; 5018 #ifdef CONFIG_NET_REDIRECT 5019 skb->from_ingress = from_ingress; 5020 if (skb->from_ingress) 5021 skb_clear_tstamp(skb); 5022 #endif 5023 } 5024 5025 static inline void skb_reset_redirect(struct sk_buff *skb) 5026 { 5027 skb->redirected = 0; 5028 } 5029 5030 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5031 bool from_ingress) 5032 { 5033 skb->redirected = 1; 5034 #ifdef CONFIG_NET_REDIRECT 5035 skb->from_ingress = from_ingress; 5036 #endif 5037 } 5038 5039 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5040 { 5041 #if IS_ENABLED(CONFIG_IP_SCTP) 5042 return skb->csum_not_inet; 5043 #else 5044 return 0; 5045 #endif 5046 } 5047 5048 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5049 { 5050 skb->ip_summed = CHECKSUM_NONE; 5051 #if IS_ENABLED(CONFIG_IP_SCTP) 5052 skb->csum_not_inet = 0; 5053 #endif 5054 } 5055 5056 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5057 const u64 kcov_handle) 5058 { 5059 #ifdef CONFIG_KCOV 5060 skb->kcov_handle = kcov_handle; 5061 #endif 5062 } 5063 5064 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5065 { 5066 #ifdef CONFIG_KCOV 5067 return skb->kcov_handle; 5068 #else 5069 return 0; 5070 #endif 5071 } 5072 5073 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5074 { 5075 #ifdef CONFIG_PAGE_POOL 5076 skb->pp_recycle = 1; 5077 #endif 5078 } 5079 5080 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5081 ssize_t maxsize, gfp_t gfp); 5082 5083 #endif /* __KERNEL__ */ 5084 #endif /* _LINUX_SKBUFF_H */ 5085