1 /* 2 * linux/net/sunrpc/svcsock.c 3 * 4 * These are the RPC server socket internals. 5 * 6 * The server scheduling algorithm does not always distribute the load 7 * evenly when servicing a single client. May need to modify the 8 * svc_xprt_enqueue procedure... 9 * 10 * TCP support is largely untested and may be a little slow. The problem 11 * is that we currently do two separate recvfrom's, one for the 4-byte 12 * record length, and the second for the actual record. This could possibly 13 * be improved by always reading a minimum size of around 100 bytes and 14 * tucking any superfluous bytes away in a temporary store. Still, that 15 * leaves write requests out in the rain. An alternative may be to peek at 16 * the first skb in the queue, and if it matches the next TCP sequence 17 * number, to extract the record marker. Yuck. 18 * 19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> 20 */ 21 22 #include <linux/kernel.h> 23 #include <linux/sched.h> 24 #include <linux/errno.h> 25 #include <linux/fcntl.h> 26 #include <linux/net.h> 27 #include <linux/in.h> 28 #include <linux/inet.h> 29 #include <linux/udp.h> 30 #include <linux/tcp.h> 31 #include <linux/unistd.h> 32 #include <linux/slab.h> 33 #include <linux/netdevice.h> 34 #include <linux/skbuff.h> 35 #include <linux/file.h> 36 #include <linux/freezer.h> 37 #include <net/sock.h> 38 #include <net/checksum.h> 39 #include <net/ip.h> 40 #include <net/ipv6.h> 41 #include <net/tcp.h> 42 #include <net/tcp_states.h> 43 #include <asm/uaccess.h> 44 #include <asm/ioctls.h> 45 46 #include <linux/sunrpc/types.h> 47 #include <linux/sunrpc/clnt.h> 48 #include <linux/sunrpc/xdr.h> 49 #include <linux/sunrpc/msg_prot.h> 50 #include <linux/sunrpc/svcsock.h> 51 #include <linux/sunrpc/stats.h> 52 #include <linux/sunrpc/xprt.h> 53 54 #include "sunrpc.h" 55 56 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 57 58 59 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, 60 int *errp, int flags); 61 static void svc_udp_data_ready(struct sock *, int); 62 static int svc_udp_recvfrom(struct svc_rqst *); 63 static int svc_udp_sendto(struct svc_rqst *); 64 static void svc_sock_detach(struct svc_xprt *); 65 static void svc_tcp_sock_detach(struct svc_xprt *); 66 static void svc_sock_free(struct svc_xprt *); 67 68 static struct svc_xprt *svc_create_socket(struct svc_serv *, int, 69 struct net *, struct sockaddr *, 70 int, int); 71 #if defined(CONFIG_NFS_V4_1) 72 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, 73 struct net *, struct sockaddr *, 74 int, int); 75 static void svc_bc_sock_free(struct svc_xprt *xprt); 76 #endif /* CONFIG_NFS_V4_1 */ 77 78 #ifdef CONFIG_DEBUG_LOCK_ALLOC 79 static struct lock_class_key svc_key[2]; 80 static struct lock_class_key svc_slock_key[2]; 81 82 static void svc_reclassify_socket(struct socket *sock) 83 { 84 struct sock *sk = sock->sk; 85 BUG_ON(sock_owned_by_user(sk)); 86 switch (sk->sk_family) { 87 case AF_INET: 88 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", 89 &svc_slock_key[0], 90 "sk_xprt.xpt_lock-AF_INET-NFSD", 91 &svc_key[0]); 92 break; 93 94 case AF_INET6: 95 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", 96 &svc_slock_key[1], 97 "sk_xprt.xpt_lock-AF_INET6-NFSD", 98 &svc_key[1]); 99 break; 100 101 default: 102 BUG(); 103 } 104 } 105 #else 106 static void svc_reclassify_socket(struct socket *sock) 107 { 108 } 109 #endif 110 111 /* 112 * Release an skbuff after use 113 */ 114 static void svc_release_skb(struct svc_rqst *rqstp) 115 { 116 struct sk_buff *skb = rqstp->rq_xprt_ctxt; 117 118 if (skb) { 119 struct svc_sock *svsk = 120 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 121 rqstp->rq_xprt_ctxt = NULL; 122 123 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); 124 skb_free_datagram_locked(svsk->sk_sk, skb); 125 } 126 } 127 128 union svc_pktinfo_u { 129 struct in_pktinfo pkti; 130 struct in6_pktinfo pkti6; 131 }; 132 #define SVC_PKTINFO_SPACE \ 133 CMSG_SPACE(sizeof(union svc_pktinfo_u)) 134 135 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) 136 { 137 struct svc_sock *svsk = 138 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 139 switch (svsk->sk_sk->sk_family) { 140 case AF_INET: { 141 struct in_pktinfo *pki = CMSG_DATA(cmh); 142 143 cmh->cmsg_level = SOL_IP; 144 cmh->cmsg_type = IP_PKTINFO; 145 pki->ipi_ifindex = 0; 146 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; 147 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 148 } 149 break; 150 151 case AF_INET6: { 152 struct in6_pktinfo *pki = CMSG_DATA(cmh); 153 154 cmh->cmsg_level = SOL_IPV6; 155 cmh->cmsg_type = IPV6_PKTINFO; 156 pki->ipi6_ifindex = 0; 157 ipv6_addr_copy(&pki->ipi6_addr, 158 &rqstp->rq_daddr.addr6); 159 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 160 } 161 break; 162 } 163 } 164 165 /* 166 * send routine intended to be shared by the fore- and back-channel 167 */ 168 int svc_send_common(struct socket *sock, struct xdr_buf *xdr, 169 struct page *headpage, unsigned long headoffset, 170 struct page *tailpage, unsigned long tailoffset) 171 { 172 int result; 173 int size; 174 struct page **ppage = xdr->pages; 175 size_t base = xdr->page_base; 176 unsigned int pglen = xdr->page_len; 177 unsigned int flags = MSG_MORE; 178 int slen; 179 int len = 0; 180 181 slen = xdr->len; 182 183 /* send head */ 184 if (slen == xdr->head[0].iov_len) 185 flags = 0; 186 len = kernel_sendpage(sock, headpage, headoffset, 187 xdr->head[0].iov_len, flags); 188 if (len != xdr->head[0].iov_len) 189 goto out; 190 slen -= xdr->head[0].iov_len; 191 if (slen == 0) 192 goto out; 193 194 /* send page data */ 195 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; 196 while (pglen > 0) { 197 if (slen == size) 198 flags = 0; 199 result = kernel_sendpage(sock, *ppage, base, size, flags); 200 if (result > 0) 201 len += result; 202 if (result != size) 203 goto out; 204 slen -= size; 205 pglen -= size; 206 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; 207 base = 0; 208 ppage++; 209 } 210 211 /* send tail */ 212 if (xdr->tail[0].iov_len) { 213 result = kernel_sendpage(sock, tailpage, tailoffset, 214 xdr->tail[0].iov_len, 0); 215 if (result > 0) 216 len += result; 217 } 218 219 out: 220 return len; 221 } 222 223 224 /* 225 * Generic sendto routine 226 */ 227 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) 228 { 229 struct svc_sock *svsk = 230 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 231 struct socket *sock = svsk->sk_sock; 232 union { 233 struct cmsghdr hdr; 234 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 235 } buffer; 236 struct cmsghdr *cmh = &buffer.hdr; 237 int len = 0; 238 unsigned long tailoff; 239 unsigned long headoff; 240 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 241 242 if (rqstp->rq_prot == IPPROTO_UDP) { 243 struct msghdr msg = { 244 .msg_name = &rqstp->rq_addr, 245 .msg_namelen = rqstp->rq_addrlen, 246 .msg_control = cmh, 247 .msg_controllen = sizeof(buffer), 248 .msg_flags = MSG_MORE, 249 }; 250 251 svc_set_cmsg_data(rqstp, cmh); 252 253 if (sock_sendmsg(sock, &msg, 0) < 0) 254 goto out; 255 } 256 257 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); 258 headoff = 0; 259 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, 260 rqstp->rq_respages[0], tailoff); 261 262 out: 263 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", 264 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, 265 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); 266 267 return len; 268 } 269 270 /* 271 * Report socket names for nfsdfs 272 */ 273 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) 274 { 275 const struct sock *sk = svsk->sk_sk; 276 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? 277 "udp" : "tcp"; 278 int len; 279 280 switch (sk->sk_family) { 281 case PF_INET: 282 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", 283 proto_name, 284 &inet_sk(sk)->inet_rcv_saddr, 285 inet_sk(sk)->inet_num); 286 break; 287 case PF_INET6: 288 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", 289 proto_name, 290 &inet6_sk(sk)->rcv_saddr, 291 inet_sk(sk)->inet_num); 292 break; 293 default: 294 len = snprintf(buf, remaining, "*unknown-%d*\n", 295 sk->sk_family); 296 } 297 298 if (len >= remaining) { 299 *buf = '\0'; 300 return -ENAMETOOLONG; 301 } 302 return len; 303 } 304 305 /** 306 * svc_sock_names - construct a list of listener names in a string 307 * @serv: pointer to RPC service 308 * @buf: pointer to a buffer to fill in with socket names 309 * @buflen: size of the buffer to be filled 310 * @toclose: pointer to '\0'-terminated C string containing the name 311 * of a listener to be closed 312 * 313 * Fills in @buf with a '\n'-separated list of names of listener 314 * sockets. If @toclose is not NULL, the socket named by @toclose 315 * is closed, and is not included in the output list. 316 * 317 * Returns positive length of the socket name string, or a negative 318 * errno value on error. 319 */ 320 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, 321 const char *toclose) 322 { 323 struct svc_sock *svsk, *closesk = NULL; 324 int len = 0; 325 326 if (!serv) 327 return 0; 328 329 spin_lock_bh(&serv->sv_lock); 330 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { 331 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); 332 if (onelen < 0) { 333 len = onelen; 334 break; 335 } 336 if (toclose && strcmp(toclose, buf + len) == 0) { 337 closesk = svsk; 338 svc_xprt_get(&closesk->sk_xprt); 339 } else 340 len += onelen; 341 } 342 spin_unlock_bh(&serv->sv_lock); 343 344 if (closesk) { 345 /* Should unregister with portmap, but you cannot 346 * unregister just one protocol... 347 */ 348 svc_close_xprt(&closesk->sk_xprt); 349 svc_xprt_put(&closesk->sk_xprt); 350 } else if (toclose) 351 return -ENOENT; 352 return len; 353 } 354 EXPORT_SYMBOL_GPL(svc_sock_names); 355 356 /* 357 * Check input queue length 358 */ 359 static int svc_recv_available(struct svc_sock *svsk) 360 { 361 struct socket *sock = svsk->sk_sock; 362 int avail, err; 363 364 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); 365 366 return (err >= 0)? avail : err; 367 } 368 369 /* 370 * Generic recvfrom routine. 371 */ 372 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, 373 int buflen) 374 { 375 struct svc_sock *svsk = 376 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 377 struct msghdr msg = { 378 .msg_flags = MSG_DONTWAIT, 379 }; 380 int len; 381 382 rqstp->rq_xprt_hlen = 0; 383 384 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, 385 msg.msg_flags); 386 387 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", 388 svsk, iov[0].iov_base, iov[0].iov_len, len); 389 return len; 390 } 391 392 static int svc_partial_recvfrom(struct svc_rqst *rqstp, 393 struct kvec *iov, int nr, 394 int buflen, unsigned int base) 395 { 396 size_t save_iovlen; 397 void __user *save_iovbase; 398 unsigned int i; 399 int ret; 400 401 if (base == 0) 402 return svc_recvfrom(rqstp, iov, nr, buflen); 403 404 for (i = 0; i < nr; i++) { 405 if (iov[i].iov_len > base) 406 break; 407 base -= iov[i].iov_len; 408 } 409 save_iovlen = iov[i].iov_len; 410 save_iovbase = iov[i].iov_base; 411 iov[i].iov_len -= base; 412 iov[i].iov_base += base; 413 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen); 414 iov[i].iov_len = save_iovlen; 415 iov[i].iov_base = save_iovbase; 416 return ret; 417 } 418 419 /* 420 * Set socket snd and rcv buffer lengths 421 */ 422 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, 423 unsigned int rcv) 424 { 425 #if 0 426 mm_segment_t oldfs; 427 oldfs = get_fs(); set_fs(KERNEL_DS); 428 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, 429 (char*)&snd, sizeof(snd)); 430 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, 431 (char*)&rcv, sizeof(rcv)); 432 #else 433 /* sock_setsockopt limits use to sysctl_?mem_max, 434 * which isn't acceptable. Until that is made conditional 435 * on not having CAP_SYS_RESOURCE or similar, we go direct... 436 * DaveM said I could! 437 */ 438 lock_sock(sock->sk); 439 sock->sk->sk_sndbuf = snd * 2; 440 sock->sk->sk_rcvbuf = rcv * 2; 441 sock->sk->sk_write_space(sock->sk); 442 release_sock(sock->sk); 443 #endif 444 } 445 /* 446 * INET callback when data has been received on the socket. 447 */ 448 static void svc_udp_data_ready(struct sock *sk, int count) 449 { 450 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 451 wait_queue_head_t *wq = sk_sleep(sk); 452 453 if (svsk) { 454 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", 455 svsk, sk, count, 456 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 457 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 458 svc_xprt_enqueue(&svsk->sk_xprt); 459 } 460 if (wq && waitqueue_active(wq)) 461 wake_up_interruptible(wq); 462 } 463 464 /* 465 * INET callback when space is newly available on the socket. 466 */ 467 static void svc_write_space(struct sock *sk) 468 { 469 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 470 wait_queue_head_t *wq = sk_sleep(sk); 471 472 if (svsk) { 473 dprintk("svc: socket %p(inet %p), write_space busy=%d\n", 474 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 475 svc_xprt_enqueue(&svsk->sk_xprt); 476 } 477 478 if (wq && waitqueue_active(wq)) { 479 dprintk("RPC svc_write_space: someone sleeping on %p\n", 480 svsk); 481 wake_up_interruptible(wq); 482 } 483 } 484 485 static void svc_tcp_write_space(struct sock *sk) 486 { 487 struct socket *sock = sk->sk_socket; 488 489 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock) 490 clear_bit(SOCK_NOSPACE, &sock->flags); 491 svc_write_space(sk); 492 } 493 494 /* 495 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo 496 */ 497 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, 498 struct cmsghdr *cmh) 499 { 500 struct in_pktinfo *pki = CMSG_DATA(cmh); 501 if (cmh->cmsg_type != IP_PKTINFO) 502 return 0; 503 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; 504 return 1; 505 } 506 507 /* 508 * See net/ipv6/datagram.c : datagram_recv_ctl 509 */ 510 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, 511 struct cmsghdr *cmh) 512 { 513 struct in6_pktinfo *pki = CMSG_DATA(cmh); 514 if (cmh->cmsg_type != IPV6_PKTINFO) 515 return 0; 516 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); 517 return 1; 518 } 519 520 /* 521 * Copy the UDP datagram's destination address to the rqstp structure. 522 * The 'destination' address in this case is the address to which the 523 * peer sent the datagram, i.e. our local address. For multihomed 524 * hosts, this can change from msg to msg. Note that only the IP 525 * address changes, the port number should remain the same. 526 */ 527 static int svc_udp_get_dest_address(struct svc_rqst *rqstp, 528 struct cmsghdr *cmh) 529 { 530 switch (cmh->cmsg_level) { 531 case SOL_IP: 532 return svc_udp_get_dest_address4(rqstp, cmh); 533 case SOL_IPV6: 534 return svc_udp_get_dest_address6(rqstp, cmh); 535 } 536 537 return 0; 538 } 539 540 /* 541 * Receive a datagram from a UDP socket. 542 */ 543 static int svc_udp_recvfrom(struct svc_rqst *rqstp) 544 { 545 struct svc_sock *svsk = 546 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 547 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 548 struct sk_buff *skb; 549 union { 550 struct cmsghdr hdr; 551 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 552 } buffer; 553 struct cmsghdr *cmh = &buffer.hdr; 554 struct msghdr msg = { 555 .msg_name = svc_addr(rqstp), 556 .msg_control = cmh, 557 .msg_controllen = sizeof(buffer), 558 .msg_flags = MSG_DONTWAIT, 559 }; 560 size_t len; 561 int err; 562 563 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 564 /* udp sockets need large rcvbuf as all pending 565 * requests are still in that buffer. sndbuf must 566 * also be large enough that there is enough space 567 * for one reply per thread. We count all threads 568 * rather than threads in a particular pool, which 569 * provides an upper bound on the number of threads 570 * which will access the socket. 571 */ 572 svc_sock_setbufsize(svsk->sk_sock, 573 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 574 (serv->sv_nrthreads+3) * serv->sv_max_mesg); 575 576 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 577 skb = NULL; 578 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 579 0, 0, MSG_PEEK | MSG_DONTWAIT); 580 if (err >= 0) 581 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); 582 583 if (skb == NULL) { 584 if (err != -EAGAIN) { 585 /* possibly an icmp error */ 586 dprintk("svc: recvfrom returned error %d\n", -err); 587 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 588 } 589 return -EAGAIN; 590 } 591 len = svc_addr_len(svc_addr(rqstp)); 592 if (len == 0) 593 return -EAFNOSUPPORT; 594 rqstp->rq_addrlen = len; 595 if (skb->tstamp.tv64 == 0) { 596 skb->tstamp = ktime_get_real(); 597 /* Don't enable netstamp, sunrpc doesn't 598 need that much accuracy */ 599 } 600 svsk->sk_sk->sk_stamp = skb->tstamp; 601 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ 602 603 len = skb->len - sizeof(struct udphdr); 604 rqstp->rq_arg.len = len; 605 606 rqstp->rq_prot = IPPROTO_UDP; 607 608 if (!svc_udp_get_dest_address(rqstp, cmh)) { 609 if (net_ratelimit()) 610 printk(KERN_WARNING 611 "svc: received unknown control message %d/%d; " 612 "dropping RPC reply datagram\n", 613 cmh->cmsg_level, cmh->cmsg_type); 614 skb_free_datagram_locked(svsk->sk_sk, skb); 615 return 0; 616 } 617 618 if (skb_is_nonlinear(skb)) { 619 /* we have to copy */ 620 local_bh_disable(); 621 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { 622 local_bh_enable(); 623 /* checksum error */ 624 skb_free_datagram_locked(svsk->sk_sk, skb); 625 return 0; 626 } 627 local_bh_enable(); 628 skb_free_datagram_locked(svsk->sk_sk, skb); 629 } else { 630 /* we can use it in-place */ 631 rqstp->rq_arg.head[0].iov_base = skb->data + 632 sizeof(struct udphdr); 633 rqstp->rq_arg.head[0].iov_len = len; 634 if (skb_checksum_complete(skb)) { 635 skb_free_datagram_locked(svsk->sk_sk, skb); 636 return 0; 637 } 638 rqstp->rq_xprt_ctxt = skb; 639 } 640 641 rqstp->rq_arg.page_base = 0; 642 if (len <= rqstp->rq_arg.head[0].iov_len) { 643 rqstp->rq_arg.head[0].iov_len = len; 644 rqstp->rq_arg.page_len = 0; 645 rqstp->rq_respages = rqstp->rq_pages+1; 646 } else { 647 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 648 rqstp->rq_respages = rqstp->rq_pages + 1 + 649 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); 650 } 651 652 if (serv->sv_stats) 653 serv->sv_stats->netudpcnt++; 654 655 return len; 656 } 657 658 static int 659 svc_udp_sendto(struct svc_rqst *rqstp) 660 { 661 int error; 662 663 error = svc_sendto(rqstp, &rqstp->rq_res); 664 if (error == -ECONNREFUSED) 665 /* ICMP error on earlier request. */ 666 error = svc_sendto(rqstp, &rqstp->rq_res); 667 668 return error; 669 } 670 671 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) 672 { 673 } 674 675 static int svc_udp_has_wspace(struct svc_xprt *xprt) 676 { 677 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 678 struct svc_serv *serv = xprt->xpt_server; 679 unsigned long required; 680 681 /* 682 * Set the SOCK_NOSPACE flag before checking the available 683 * sock space. 684 */ 685 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 686 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; 687 if (required*2 > sock_wspace(svsk->sk_sk)) 688 return 0; 689 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 690 return 1; 691 } 692 693 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) 694 { 695 BUG(); 696 return NULL; 697 } 698 699 static struct svc_xprt *svc_udp_create(struct svc_serv *serv, 700 struct net *net, 701 struct sockaddr *sa, int salen, 702 int flags) 703 { 704 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); 705 } 706 707 static struct svc_xprt_ops svc_udp_ops = { 708 .xpo_create = svc_udp_create, 709 .xpo_recvfrom = svc_udp_recvfrom, 710 .xpo_sendto = svc_udp_sendto, 711 .xpo_release_rqst = svc_release_skb, 712 .xpo_detach = svc_sock_detach, 713 .xpo_free = svc_sock_free, 714 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, 715 .xpo_has_wspace = svc_udp_has_wspace, 716 .xpo_accept = svc_udp_accept, 717 }; 718 719 static struct svc_xprt_class svc_udp_class = { 720 .xcl_name = "udp", 721 .xcl_owner = THIS_MODULE, 722 .xcl_ops = &svc_udp_ops, 723 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, 724 }; 725 726 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) 727 { 728 int err, level, optname, one = 1; 729 730 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); 731 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 732 svsk->sk_sk->sk_data_ready = svc_udp_data_ready; 733 svsk->sk_sk->sk_write_space = svc_write_space; 734 735 /* initialise setting must have enough space to 736 * receive and respond to one request. 737 * svc_udp_recvfrom will re-adjust if necessary 738 */ 739 svc_sock_setbufsize(svsk->sk_sock, 740 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, 741 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); 742 743 /* data might have come in before data_ready set up */ 744 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 745 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 746 747 /* make sure we get destination address info */ 748 switch (svsk->sk_sk->sk_family) { 749 case AF_INET: 750 level = SOL_IP; 751 optname = IP_PKTINFO; 752 break; 753 case AF_INET6: 754 level = SOL_IPV6; 755 optname = IPV6_RECVPKTINFO; 756 break; 757 default: 758 BUG(); 759 } 760 err = kernel_setsockopt(svsk->sk_sock, level, optname, 761 (char *)&one, sizeof(one)); 762 dprintk("svc: kernel_setsockopt returned %d\n", err); 763 } 764 765 /* 766 * A data_ready event on a listening socket means there's a connection 767 * pending. Do not use state_change as a substitute for it. 768 */ 769 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) 770 { 771 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 772 wait_queue_head_t *wq; 773 774 dprintk("svc: socket %p TCP (listen) state change %d\n", 775 sk, sk->sk_state); 776 777 /* 778 * This callback may called twice when a new connection 779 * is established as a child socket inherits everything 780 * from a parent LISTEN socket. 781 * 1) data_ready method of the parent socket will be called 782 * when one of child sockets become ESTABLISHED. 783 * 2) data_ready method of the child socket may be called 784 * when it receives data before the socket is accepted. 785 * In case of 2, we should ignore it silently. 786 */ 787 if (sk->sk_state == TCP_LISTEN) { 788 if (svsk) { 789 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 790 svc_xprt_enqueue(&svsk->sk_xprt); 791 } else 792 printk("svc: socket %p: no user data\n", sk); 793 } 794 795 wq = sk_sleep(sk); 796 if (wq && waitqueue_active(wq)) 797 wake_up_interruptible_all(wq); 798 } 799 800 /* 801 * A state change on a connected socket means it's dying or dead. 802 */ 803 static void svc_tcp_state_change(struct sock *sk) 804 { 805 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 806 wait_queue_head_t *wq = sk_sleep(sk); 807 808 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", 809 sk, sk->sk_state, sk->sk_user_data); 810 811 if (!svsk) 812 printk("svc: socket %p: no user data\n", sk); 813 else { 814 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 815 svc_xprt_enqueue(&svsk->sk_xprt); 816 } 817 if (wq && waitqueue_active(wq)) 818 wake_up_interruptible_all(wq); 819 } 820 821 static void svc_tcp_data_ready(struct sock *sk, int count) 822 { 823 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 824 wait_queue_head_t *wq = sk_sleep(sk); 825 826 dprintk("svc: socket %p TCP data ready (svsk %p)\n", 827 sk, sk->sk_user_data); 828 if (svsk) { 829 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 830 svc_xprt_enqueue(&svsk->sk_xprt); 831 } 832 if (wq && waitqueue_active(wq)) 833 wake_up_interruptible(wq); 834 } 835 836 /* 837 * Accept a TCP connection 838 */ 839 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) 840 { 841 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 842 struct sockaddr_storage addr; 843 struct sockaddr *sin = (struct sockaddr *) &addr; 844 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 845 struct socket *sock = svsk->sk_sock; 846 struct socket *newsock; 847 struct svc_sock *newsvsk; 848 int err, slen; 849 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 850 851 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); 852 if (!sock) 853 return NULL; 854 855 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 856 err = kernel_accept(sock, &newsock, O_NONBLOCK); 857 if (err < 0) { 858 if (err == -ENOMEM) 859 printk(KERN_WARNING "%s: no more sockets!\n", 860 serv->sv_name); 861 else if (err != -EAGAIN && net_ratelimit()) 862 printk(KERN_WARNING "%s: accept failed (err %d)!\n", 863 serv->sv_name, -err); 864 return NULL; 865 } 866 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 867 868 err = kernel_getpeername(newsock, sin, &slen); 869 if (err < 0) { 870 if (net_ratelimit()) 871 printk(KERN_WARNING "%s: peername failed (err %d)!\n", 872 serv->sv_name, -err); 873 goto failed; /* aborted connection or whatever */ 874 } 875 876 /* Ideally, we would want to reject connections from unauthorized 877 * hosts here, but when we get encryption, the IP of the host won't 878 * tell us anything. For now just warn about unpriv connections. 879 */ 880 if (!svc_port_is_privileged(sin)) { 881 dprintk(KERN_WARNING 882 "%s: connect from unprivileged port: %s\n", 883 serv->sv_name, 884 __svc_print_addr(sin, buf, sizeof(buf))); 885 } 886 dprintk("%s: connect from %s\n", serv->sv_name, 887 __svc_print_addr(sin, buf, sizeof(buf))); 888 889 /* make sure that a write doesn't block forever when 890 * low on memory 891 */ 892 newsock->sk->sk_sndtimeo = HZ*30; 893 894 if (!(newsvsk = svc_setup_socket(serv, newsock, &err, 895 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) 896 goto failed; 897 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); 898 err = kernel_getsockname(newsock, sin, &slen); 899 if (unlikely(err < 0)) { 900 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); 901 slen = offsetof(struct sockaddr, sa_data); 902 } 903 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); 904 905 if (serv->sv_stats) 906 serv->sv_stats->nettcpconn++; 907 908 return &newsvsk->sk_xprt; 909 910 failed: 911 sock_release(newsock); 912 return NULL; 913 } 914 915 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 916 { 917 unsigned int i, len, npages; 918 919 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr)) 920 return 0; 921 len = svsk->sk_tcplen - sizeof(rpc_fraghdr); 922 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 923 for (i = 0; i < npages; i++) { 924 if (rqstp->rq_pages[i] != NULL) 925 put_page(rqstp->rq_pages[i]); 926 BUG_ON(svsk->sk_pages[i] == NULL); 927 rqstp->rq_pages[i] = svsk->sk_pages[i]; 928 svsk->sk_pages[i] = NULL; 929 } 930 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); 931 return len; 932 } 933 934 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 935 { 936 unsigned int i, len, npages; 937 938 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr)) 939 return; 940 len = svsk->sk_tcplen - sizeof(rpc_fraghdr); 941 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 942 for (i = 0; i < npages; i++) { 943 svsk->sk_pages[i] = rqstp->rq_pages[i]; 944 rqstp->rq_pages[i] = NULL; 945 } 946 } 947 948 static void svc_tcp_clear_pages(struct svc_sock *svsk) 949 { 950 unsigned int i, len, npages; 951 952 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr)) 953 goto out; 954 len = svsk->sk_tcplen - sizeof(rpc_fraghdr); 955 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 956 for (i = 0; i < npages; i++) { 957 BUG_ON(svsk->sk_pages[i] == NULL); 958 put_page(svsk->sk_pages[i]); 959 svsk->sk_pages[i] = NULL; 960 } 961 out: 962 svsk->sk_tcplen = 0; 963 } 964 965 /* 966 * Receive data. 967 * If we haven't gotten the record length yet, get the next four bytes. 968 * Otherwise try to gobble up as much as possible up to the complete 969 * record length. 970 */ 971 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) 972 { 973 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 974 unsigned int want; 975 int len; 976 977 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 978 979 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { 980 struct kvec iov; 981 982 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; 983 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; 984 iov.iov_len = want; 985 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) 986 goto error; 987 svsk->sk_tcplen += len; 988 989 if (len < want) { 990 dprintk("svc: short recvfrom while reading record " 991 "length (%d of %d)\n", len, want); 992 return -EAGAIN; 993 } 994 995 svsk->sk_reclen = ntohl(svsk->sk_reclen); 996 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { 997 /* FIXME: technically, a record can be fragmented, 998 * and non-terminal fragments will not have the top 999 * bit set in the fragment length header. 1000 * But apparently no known nfs clients send fragmented 1001 * records. */ 1002 if (net_ratelimit()) 1003 printk(KERN_NOTICE "RPC: multiple fragments " 1004 "per record not supported\n"); 1005 goto err_delete; 1006 } 1007 1008 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; 1009 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); 1010 if (svsk->sk_reclen > serv->sv_max_mesg) { 1011 if (net_ratelimit()) 1012 printk(KERN_NOTICE "RPC: " 1013 "fragment too large: 0x%08lx\n", 1014 (unsigned long)svsk->sk_reclen); 1015 goto err_delete; 1016 } 1017 } 1018 1019 if (svsk->sk_reclen < 8) 1020 goto err_delete; /* client is nuts. */ 1021 1022 len = svsk->sk_reclen; 1023 1024 return len; 1025 error: 1026 dprintk("RPC: TCP recv_record got %d\n", len); 1027 return len; 1028 err_delete: 1029 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1030 return -EAGAIN; 1031 } 1032 1033 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) 1034 { 1035 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; 1036 struct rpc_rqst *req = NULL; 1037 struct kvec *src, *dst; 1038 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1039 __be32 xid; 1040 __be32 calldir; 1041 1042 xid = *p++; 1043 calldir = *p; 1044 1045 if (bc_xprt) 1046 req = xprt_lookup_rqst(bc_xprt, xid); 1047 1048 if (!req) { 1049 printk(KERN_NOTICE 1050 "%s: Got unrecognized reply: " 1051 "calldir 0x%x xpt_bc_xprt %p xid %08x\n", 1052 __func__, ntohl(calldir), 1053 bc_xprt, xid); 1054 return -EAGAIN; 1055 } 1056 1057 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); 1058 /* 1059 * XXX!: cheating for now! Only copying HEAD. 1060 * But we know this is good enough for now (in fact, for any 1061 * callback reply in the forseeable future). 1062 */ 1063 dst = &req->rq_private_buf.head[0]; 1064 src = &rqstp->rq_arg.head[0]; 1065 if (dst->iov_len < src->iov_len) 1066 return -EAGAIN; /* whatever; just giving up. */ 1067 memcpy(dst->iov_base, src->iov_base, src->iov_len); 1068 xprt_complete_rqst(req->rq_task, svsk->sk_reclen); 1069 rqstp->rq_arg.len = 0; 1070 return 0; 1071 } 1072 1073 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len) 1074 { 1075 int i = 0; 1076 int t = 0; 1077 1078 while (t < len) { 1079 vec[i].iov_base = page_address(pages[i]); 1080 vec[i].iov_len = PAGE_SIZE; 1081 i++; 1082 t += PAGE_SIZE; 1083 } 1084 return i; 1085 } 1086 1087 1088 /* 1089 * Receive data from a TCP socket. 1090 */ 1091 static int svc_tcp_recvfrom(struct svc_rqst *rqstp) 1092 { 1093 struct svc_sock *svsk = 1094 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 1095 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1096 int len; 1097 struct kvec *vec; 1098 unsigned int want, base; 1099 __be32 *p; 1100 __be32 calldir; 1101 int pnum; 1102 1103 dprintk("svc: tcp_recv %p data %d conn %d close %d\n", 1104 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), 1105 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), 1106 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1107 1108 len = svc_tcp_recv_record(svsk, rqstp); 1109 if (len < 0) 1110 goto error; 1111 1112 base = svc_tcp_restore_pages(svsk, rqstp); 1113 want = svsk->sk_reclen - base; 1114 1115 vec = rqstp->rq_vec; 1116 1117 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], 1118 svsk->sk_reclen); 1119 1120 rqstp->rq_respages = &rqstp->rq_pages[pnum]; 1121 1122 /* Now receive data */ 1123 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base); 1124 if (len >= 0) 1125 svsk->sk_tcplen += len; 1126 if (len != want) { 1127 if (len < 0 && len != -EAGAIN) 1128 goto err_other; 1129 svc_tcp_save_pages(svsk, rqstp); 1130 dprintk("svc: incomplete TCP record (%d of %d)\n", 1131 svsk->sk_tcplen, svsk->sk_reclen); 1132 goto err_noclose; 1133 } 1134 1135 rqstp->rq_arg.len = svsk->sk_reclen; 1136 rqstp->rq_arg.page_base = 0; 1137 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { 1138 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; 1139 rqstp->rq_arg.page_len = 0; 1140 } else 1141 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1142 1143 rqstp->rq_xprt_ctxt = NULL; 1144 rqstp->rq_prot = IPPROTO_TCP; 1145 1146 p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1147 calldir = p[1]; 1148 if (calldir) 1149 len = receive_cb_reply(svsk, rqstp); 1150 1151 /* Reset TCP read info */ 1152 svsk->sk_reclen = 0; 1153 svsk->sk_tcplen = 0; 1154 /* If we have more data, signal svc_xprt_enqueue() to try again */ 1155 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr)) 1156 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1157 1158 if (len < 0) 1159 goto error; 1160 1161 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); 1162 if (serv->sv_stats) 1163 serv->sv_stats->nettcpcnt++; 1164 1165 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len); 1166 return rqstp->rq_arg.len; 1167 1168 error: 1169 if (len != -EAGAIN) 1170 goto err_other; 1171 dprintk("RPC: TCP recvfrom got EAGAIN\n"); 1172 return -EAGAIN; 1173 err_other: 1174 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", 1175 svsk->sk_xprt.xpt_server->sv_name, -len); 1176 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1177 err_noclose: 1178 return -EAGAIN; /* record not complete */ 1179 } 1180 1181 /* 1182 * Send out data on TCP socket. 1183 */ 1184 static int svc_tcp_sendto(struct svc_rqst *rqstp) 1185 { 1186 struct xdr_buf *xbufp = &rqstp->rq_res; 1187 int sent; 1188 __be32 reclen; 1189 1190 /* Set up the first element of the reply kvec. 1191 * Any other kvecs that may be in use have been taken 1192 * care of by the server implementation itself. 1193 */ 1194 reclen = htonl(0x80000000|((xbufp->len ) - 4)); 1195 memcpy(xbufp->head[0].iov_base, &reclen, 4); 1196 1197 sent = svc_sendto(rqstp, &rqstp->rq_res); 1198 if (sent != xbufp->len) { 1199 printk(KERN_NOTICE 1200 "rpc-srv/tcp: %s: %s %d when sending %d bytes " 1201 "- shutting down socket\n", 1202 rqstp->rq_xprt->xpt_server->sv_name, 1203 (sent<0)?"got error":"sent only", 1204 sent, xbufp->len); 1205 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); 1206 svc_xprt_enqueue(rqstp->rq_xprt); 1207 sent = -EAGAIN; 1208 } 1209 return sent; 1210 } 1211 1212 /* 1213 * Setup response header. TCP has a 4B record length field. 1214 */ 1215 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) 1216 { 1217 struct kvec *resv = &rqstp->rq_res.head[0]; 1218 1219 /* tcp needs a space for the record length... */ 1220 svc_putnl(resv, 0); 1221 } 1222 1223 static int svc_tcp_has_wspace(struct svc_xprt *xprt) 1224 { 1225 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1226 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1227 int required; 1228 1229 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) 1230 return 1; 1231 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; 1232 if (sk_stream_wspace(svsk->sk_sk) >= required) 1233 return 1; 1234 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 1235 return 0; 1236 } 1237 1238 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, 1239 struct net *net, 1240 struct sockaddr *sa, int salen, 1241 int flags) 1242 { 1243 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1244 } 1245 1246 #if defined(CONFIG_NFS_V4_1) 1247 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, 1248 struct net *, struct sockaddr *, 1249 int, int); 1250 static void svc_bc_sock_free(struct svc_xprt *xprt); 1251 1252 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv, 1253 struct net *net, 1254 struct sockaddr *sa, int salen, 1255 int flags) 1256 { 1257 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1258 } 1259 1260 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt) 1261 { 1262 } 1263 1264 static struct svc_xprt_ops svc_tcp_bc_ops = { 1265 .xpo_create = svc_bc_tcp_create, 1266 .xpo_detach = svc_bc_tcp_sock_detach, 1267 .xpo_free = svc_bc_sock_free, 1268 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1269 }; 1270 1271 static struct svc_xprt_class svc_tcp_bc_class = { 1272 .xcl_name = "tcp-bc", 1273 .xcl_owner = THIS_MODULE, 1274 .xcl_ops = &svc_tcp_bc_ops, 1275 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1276 }; 1277 1278 static void svc_init_bc_xprt_sock(void) 1279 { 1280 svc_reg_xprt_class(&svc_tcp_bc_class); 1281 } 1282 1283 static void svc_cleanup_bc_xprt_sock(void) 1284 { 1285 svc_unreg_xprt_class(&svc_tcp_bc_class); 1286 } 1287 #else /* CONFIG_NFS_V4_1 */ 1288 static void svc_init_bc_xprt_sock(void) 1289 { 1290 } 1291 1292 static void svc_cleanup_bc_xprt_sock(void) 1293 { 1294 } 1295 #endif /* CONFIG_NFS_V4_1 */ 1296 1297 static struct svc_xprt_ops svc_tcp_ops = { 1298 .xpo_create = svc_tcp_create, 1299 .xpo_recvfrom = svc_tcp_recvfrom, 1300 .xpo_sendto = svc_tcp_sendto, 1301 .xpo_release_rqst = svc_release_skb, 1302 .xpo_detach = svc_tcp_sock_detach, 1303 .xpo_free = svc_sock_free, 1304 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1305 .xpo_has_wspace = svc_tcp_has_wspace, 1306 .xpo_accept = svc_tcp_accept, 1307 }; 1308 1309 static struct svc_xprt_class svc_tcp_class = { 1310 .xcl_name = "tcp", 1311 .xcl_owner = THIS_MODULE, 1312 .xcl_ops = &svc_tcp_ops, 1313 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1314 }; 1315 1316 void svc_init_xprt_sock(void) 1317 { 1318 svc_reg_xprt_class(&svc_tcp_class); 1319 svc_reg_xprt_class(&svc_udp_class); 1320 svc_init_bc_xprt_sock(); 1321 } 1322 1323 void svc_cleanup_xprt_sock(void) 1324 { 1325 svc_unreg_xprt_class(&svc_tcp_class); 1326 svc_unreg_xprt_class(&svc_udp_class); 1327 svc_cleanup_bc_xprt_sock(); 1328 } 1329 1330 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) 1331 { 1332 struct sock *sk = svsk->sk_sk; 1333 1334 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); 1335 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 1336 if (sk->sk_state == TCP_LISTEN) { 1337 dprintk("setting up TCP socket for listening\n"); 1338 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); 1339 sk->sk_data_ready = svc_tcp_listen_data_ready; 1340 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 1341 } else { 1342 dprintk("setting up TCP socket for reading\n"); 1343 sk->sk_state_change = svc_tcp_state_change; 1344 sk->sk_data_ready = svc_tcp_data_ready; 1345 sk->sk_write_space = svc_tcp_write_space; 1346 1347 svsk->sk_reclen = 0; 1348 svsk->sk_tcplen = 0; 1349 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); 1350 1351 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1352 1353 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1354 if (sk->sk_state != TCP_ESTABLISHED) 1355 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1356 } 1357 } 1358 1359 void svc_sock_update_bufs(struct svc_serv *serv) 1360 { 1361 /* 1362 * The number of server threads has changed. Update 1363 * rcvbuf and sndbuf accordingly on all sockets 1364 */ 1365 struct svc_sock *svsk; 1366 1367 spin_lock_bh(&serv->sv_lock); 1368 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) 1369 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1370 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list) 1371 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1372 spin_unlock_bh(&serv->sv_lock); 1373 } 1374 EXPORT_SYMBOL_GPL(svc_sock_update_bufs); 1375 1376 /* 1377 * Initialize socket for RPC use and create svc_sock struct 1378 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. 1379 */ 1380 static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1381 struct socket *sock, 1382 int *errp, int flags) 1383 { 1384 struct svc_sock *svsk; 1385 struct sock *inet; 1386 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1387 1388 dprintk("svc: svc_setup_socket %p\n", sock); 1389 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { 1390 *errp = -ENOMEM; 1391 return NULL; 1392 } 1393 1394 inet = sock->sk; 1395 1396 /* Register socket with portmapper */ 1397 if (*errp >= 0 && pmap_register) 1398 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, 1399 ntohs(inet_sk(inet)->inet_sport)); 1400 1401 if (*errp < 0) { 1402 kfree(svsk); 1403 return NULL; 1404 } 1405 1406 inet->sk_user_data = svsk; 1407 svsk->sk_sock = sock; 1408 svsk->sk_sk = inet; 1409 svsk->sk_ostate = inet->sk_state_change; 1410 svsk->sk_odata = inet->sk_data_ready; 1411 svsk->sk_owspace = inet->sk_write_space; 1412 1413 /* Initialize the socket */ 1414 if (sock->type == SOCK_DGRAM) 1415 svc_udp_init(svsk, serv); 1416 else { 1417 /* initialise setting must have enough space to 1418 * receive and respond to one request. 1419 */ 1420 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg, 1421 4 * serv->sv_max_mesg); 1422 svc_tcp_init(svsk, serv); 1423 } 1424 1425 dprintk("svc: svc_setup_socket created %p (inet %p)\n", 1426 svsk, svsk->sk_sk); 1427 1428 return svsk; 1429 } 1430 1431 /** 1432 * svc_addsock - add a listener socket to an RPC service 1433 * @serv: pointer to RPC service to which to add a new listener 1434 * @fd: file descriptor of the new listener 1435 * @name_return: pointer to buffer to fill in with name of listener 1436 * @len: size of the buffer 1437 * 1438 * Fills in socket name and returns positive length of name if successful. 1439 * Name is terminated with '\n'. On error, returns a negative errno 1440 * value. 1441 */ 1442 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, 1443 const size_t len) 1444 { 1445 int err = 0; 1446 struct socket *so = sockfd_lookup(fd, &err); 1447 struct svc_sock *svsk = NULL; 1448 1449 if (!so) 1450 return err; 1451 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) 1452 err = -EAFNOSUPPORT; 1453 else if (so->sk->sk_protocol != IPPROTO_TCP && 1454 so->sk->sk_protocol != IPPROTO_UDP) 1455 err = -EPROTONOSUPPORT; 1456 else if (so->state > SS_UNCONNECTED) 1457 err = -EISCONN; 1458 else { 1459 if (!try_module_get(THIS_MODULE)) 1460 err = -ENOENT; 1461 else 1462 svsk = svc_setup_socket(serv, so, &err, 1463 SVC_SOCK_DEFAULTS); 1464 if (svsk) { 1465 struct sockaddr_storage addr; 1466 struct sockaddr *sin = (struct sockaddr *)&addr; 1467 int salen; 1468 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) 1469 svc_xprt_set_local(&svsk->sk_xprt, sin, salen); 1470 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); 1471 spin_lock_bh(&serv->sv_lock); 1472 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); 1473 spin_unlock_bh(&serv->sv_lock); 1474 svc_xprt_received(&svsk->sk_xprt); 1475 err = 0; 1476 } else 1477 module_put(THIS_MODULE); 1478 } 1479 if (err) { 1480 sockfd_put(so); 1481 return err; 1482 } 1483 return svc_one_sock_name(svsk, name_return, len); 1484 } 1485 EXPORT_SYMBOL_GPL(svc_addsock); 1486 1487 /* 1488 * Create socket for RPC service. 1489 */ 1490 static struct svc_xprt *svc_create_socket(struct svc_serv *serv, 1491 int protocol, 1492 struct net *net, 1493 struct sockaddr *sin, int len, 1494 int flags) 1495 { 1496 struct svc_sock *svsk; 1497 struct socket *sock; 1498 int error; 1499 int type; 1500 struct sockaddr_storage addr; 1501 struct sockaddr *newsin = (struct sockaddr *)&addr; 1502 int newlen; 1503 int family; 1504 int val; 1505 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 1506 1507 dprintk("svc: svc_create_socket(%s, %d, %s)\n", 1508 serv->sv_program->pg_name, protocol, 1509 __svc_print_addr(sin, buf, sizeof(buf))); 1510 1511 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1512 printk(KERN_WARNING "svc: only UDP and TCP " 1513 "sockets supported\n"); 1514 return ERR_PTR(-EINVAL); 1515 } 1516 1517 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1518 switch (sin->sa_family) { 1519 case AF_INET6: 1520 family = PF_INET6; 1521 break; 1522 case AF_INET: 1523 family = PF_INET; 1524 break; 1525 default: 1526 return ERR_PTR(-EINVAL); 1527 } 1528 1529 error = __sock_create(net, family, type, protocol, &sock, 1); 1530 if (error < 0) 1531 return ERR_PTR(error); 1532 1533 svc_reclassify_socket(sock); 1534 1535 /* 1536 * If this is an PF_INET6 listener, we want to avoid 1537 * getting requests from IPv4 remotes. Those should 1538 * be shunted to a PF_INET listener via rpcbind. 1539 */ 1540 val = 1; 1541 if (family == PF_INET6) 1542 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, 1543 (char *)&val, sizeof(val)); 1544 1545 if (type == SOCK_STREAM) 1546 sock->sk->sk_reuse = 1; /* allow address reuse */ 1547 error = kernel_bind(sock, sin, len); 1548 if (error < 0) 1549 goto bummer; 1550 1551 newlen = len; 1552 error = kernel_getsockname(sock, newsin, &newlen); 1553 if (error < 0) 1554 goto bummer; 1555 1556 if (protocol == IPPROTO_TCP) { 1557 if ((error = kernel_listen(sock, 64)) < 0) 1558 goto bummer; 1559 } 1560 1561 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { 1562 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); 1563 return (struct svc_xprt *)svsk; 1564 } 1565 1566 bummer: 1567 dprintk("svc: svc_create_socket error = %d\n", -error); 1568 sock_release(sock); 1569 return ERR_PTR(error); 1570 } 1571 1572 /* 1573 * Detach the svc_sock from the socket so that no 1574 * more callbacks occur. 1575 */ 1576 static void svc_sock_detach(struct svc_xprt *xprt) 1577 { 1578 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1579 struct sock *sk = svsk->sk_sk; 1580 wait_queue_head_t *wq; 1581 1582 dprintk("svc: svc_sock_detach(%p)\n", svsk); 1583 1584 /* put back the old socket callbacks */ 1585 sk->sk_state_change = svsk->sk_ostate; 1586 sk->sk_data_ready = svsk->sk_odata; 1587 sk->sk_write_space = svsk->sk_owspace; 1588 1589 wq = sk_sleep(sk); 1590 if (wq && waitqueue_active(wq)) 1591 wake_up_interruptible(wq); 1592 } 1593 1594 /* 1595 * Disconnect the socket, and reset the callbacks 1596 */ 1597 static void svc_tcp_sock_detach(struct svc_xprt *xprt) 1598 { 1599 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1600 1601 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); 1602 1603 svc_sock_detach(xprt); 1604 1605 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { 1606 svc_tcp_clear_pages(svsk); 1607 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); 1608 } 1609 } 1610 1611 /* 1612 * Free the svc_sock's socket resources and the svc_sock itself. 1613 */ 1614 static void svc_sock_free(struct svc_xprt *xprt) 1615 { 1616 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1617 dprintk("svc: svc_sock_free(%p)\n", svsk); 1618 1619 if (svsk->sk_sock->file) 1620 sockfd_put(svsk->sk_sock); 1621 else 1622 sock_release(svsk->sk_sock); 1623 kfree(svsk); 1624 } 1625 1626 #if defined(CONFIG_NFS_V4_1) 1627 /* 1628 * Create a back channel svc_xprt which shares the fore channel socket. 1629 */ 1630 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv, 1631 int protocol, 1632 struct net *net, 1633 struct sockaddr *sin, int len, 1634 int flags) 1635 { 1636 struct svc_sock *svsk; 1637 struct svc_xprt *xprt; 1638 1639 if (protocol != IPPROTO_TCP) { 1640 printk(KERN_WARNING "svc: only TCP sockets" 1641 " supported on shared back channel\n"); 1642 return ERR_PTR(-EINVAL); 1643 } 1644 1645 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1646 if (!svsk) 1647 return ERR_PTR(-ENOMEM); 1648 1649 xprt = &svsk->sk_xprt; 1650 svc_xprt_init(&svc_tcp_bc_class, xprt, serv); 1651 1652 serv->sv_bc_xprt = xprt; 1653 1654 return xprt; 1655 } 1656 1657 /* 1658 * Free a back channel svc_sock. 1659 */ 1660 static void svc_bc_sock_free(struct svc_xprt *xprt) 1661 { 1662 if (xprt) 1663 kfree(container_of(xprt, struct svc_sock, sk_xprt)); 1664 } 1665 #endif /* CONFIG_NFS_V4_1 */ 1666