1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/net/sunrpc/svcsock.c 4 * 5 * These are the RPC server socket internals. 6 * 7 * The server scheduling algorithm does not always distribute the load 8 * evenly when servicing a single client. May need to modify the 9 * svc_xprt_enqueue procedure... 10 * 11 * TCP support is largely untested and may be a little slow. The problem 12 * is that we currently do two separate recvfrom's, one for the 4-byte 13 * record length, and the second for the actual record. This could possibly 14 * be improved by always reading a minimum size of around 100 bytes and 15 * tucking any superfluous bytes away in a temporary store. Still, that 16 * leaves write requests out in the rain. An alternative may be to peek at 17 * the first skb in the queue, and if it matches the next TCP sequence 18 * number, to extract the record marker. Yuck. 19 * 20 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> 21 */ 22 23 #include <linux/kernel.h> 24 #include <linux/sched.h> 25 #include <linux/module.h> 26 #include <linux/errno.h> 27 #include <linux/fcntl.h> 28 #include <linux/net.h> 29 #include <linux/in.h> 30 #include <linux/inet.h> 31 #include <linux/udp.h> 32 #include <linux/tcp.h> 33 #include <linux/unistd.h> 34 #include <linux/slab.h> 35 #include <linux/netdevice.h> 36 #include <linux/skbuff.h> 37 #include <linux/file.h> 38 #include <linux/freezer.h> 39 #include <net/sock.h> 40 #include <net/checksum.h> 41 #include <net/ip.h> 42 #include <net/ipv6.h> 43 #include <net/udp.h> 44 #include <net/tcp.h> 45 #include <net/tcp_states.h> 46 #include <linux/uaccess.h> 47 #include <asm/ioctls.h> 48 #include <trace/events/skb.h> 49 50 #include <linux/sunrpc/types.h> 51 #include <linux/sunrpc/clnt.h> 52 #include <linux/sunrpc/xdr.h> 53 #include <linux/sunrpc/msg_prot.h> 54 #include <linux/sunrpc/svcsock.h> 55 #include <linux/sunrpc/stats.h> 56 #include <linux/sunrpc/xprt.h> 57 58 #include "socklib.h" 59 #include "sunrpc.h" 60 61 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 62 63 64 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, 65 int flags); 66 static int svc_udp_recvfrom(struct svc_rqst *); 67 static int svc_udp_sendto(struct svc_rqst *); 68 static void svc_sock_detach(struct svc_xprt *); 69 static void svc_tcp_sock_detach(struct svc_xprt *); 70 static void svc_sock_free(struct svc_xprt *); 71 72 static struct svc_xprt *svc_create_socket(struct svc_serv *, int, 73 struct net *, struct sockaddr *, 74 int, int); 75 #ifdef CONFIG_DEBUG_LOCK_ALLOC 76 static struct lock_class_key svc_key[2]; 77 static struct lock_class_key svc_slock_key[2]; 78 79 static void svc_reclassify_socket(struct socket *sock) 80 { 81 struct sock *sk = sock->sk; 82 83 if (WARN_ON_ONCE(!sock_allow_reclassification(sk))) 84 return; 85 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 static void svc_release_udp_skb(struct svc_rqst *rqstp) 129 { 130 struct sk_buff *skb = rqstp->rq_xprt_ctxt; 131 132 if (skb) { 133 rqstp->rq_xprt_ctxt = NULL; 134 135 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); 136 consume_skb(skb); 137 } 138 } 139 140 union svc_pktinfo_u { 141 struct in_pktinfo pkti; 142 struct in6_pktinfo pkti6; 143 }; 144 #define SVC_PKTINFO_SPACE \ 145 CMSG_SPACE(sizeof(union svc_pktinfo_u)) 146 147 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) 148 { 149 struct svc_sock *svsk = 150 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 151 switch (svsk->sk_sk->sk_family) { 152 case AF_INET: { 153 struct in_pktinfo *pki = CMSG_DATA(cmh); 154 155 cmh->cmsg_level = SOL_IP; 156 cmh->cmsg_type = IP_PKTINFO; 157 pki->ipi_ifindex = 0; 158 pki->ipi_spec_dst.s_addr = 159 svc_daddr_in(rqstp)->sin_addr.s_addr; 160 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 161 } 162 break; 163 164 case AF_INET6: { 165 struct in6_pktinfo *pki = CMSG_DATA(cmh); 166 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 167 168 cmh->cmsg_level = SOL_IPV6; 169 cmh->cmsg_type = IPV6_PKTINFO; 170 pki->ipi6_ifindex = daddr->sin6_scope_id; 171 pki->ipi6_addr = daddr->sin6_addr; 172 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 173 } 174 break; 175 } 176 } 177 178 static int svc_sock_read_payload(struct svc_rqst *rqstp, unsigned int offset, 179 unsigned int length) 180 { 181 return 0; 182 } 183 184 /* 185 * Report socket names for nfsdfs 186 */ 187 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) 188 { 189 const struct sock *sk = svsk->sk_sk; 190 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? 191 "udp" : "tcp"; 192 int len; 193 194 switch (sk->sk_family) { 195 case PF_INET: 196 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", 197 proto_name, 198 &inet_sk(sk)->inet_rcv_saddr, 199 inet_sk(sk)->inet_num); 200 break; 201 #if IS_ENABLED(CONFIG_IPV6) 202 case PF_INET6: 203 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", 204 proto_name, 205 &sk->sk_v6_rcv_saddr, 206 inet_sk(sk)->inet_num); 207 break; 208 #endif 209 default: 210 len = snprintf(buf, remaining, "*unknown-%d*\n", 211 sk->sk_family); 212 } 213 214 if (len >= remaining) { 215 *buf = '\0'; 216 return -ENAMETOOLONG; 217 } 218 return len; 219 } 220 221 /* 222 * Generic recvfrom routine. 223 */ 224 static ssize_t svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, 225 unsigned int nr, size_t buflen, unsigned int base) 226 { 227 struct svc_sock *svsk = 228 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 229 struct msghdr msg = { NULL }; 230 ssize_t len; 231 232 rqstp->rq_xprt_hlen = 0; 233 234 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 235 iov_iter_kvec(&msg.msg_iter, READ, iov, nr, buflen); 236 if (base != 0) { 237 iov_iter_advance(&msg.msg_iter, base); 238 buflen -= base; 239 } 240 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT); 241 /* If we read a full record, then assume there may be more 242 * data to read (stream based sockets only!) 243 */ 244 if (len == buflen) 245 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 246 247 dprintk("svc: socket %p recvfrom(%p, %zu) = %zd\n", 248 svsk, iov[0].iov_base, iov[0].iov_len, len); 249 return len; 250 } 251 252 /* 253 * Set socket snd and rcv buffer lengths 254 */ 255 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs) 256 { 257 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg; 258 struct socket *sock = svsk->sk_sock; 259 260 nreqs = min(nreqs, INT_MAX / 2 / max_mesg); 261 262 lock_sock(sock->sk); 263 sock->sk->sk_sndbuf = nreqs * max_mesg * 2; 264 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2; 265 sock->sk->sk_write_space(sock->sk); 266 release_sock(sock->sk); 267 } 268 269 static void svc_sock_secure_port(struct svc_rqst *rqstp) 270 { 271 if (svc_port_is_privileged(svc_addr(rqstp))) 272 set_bit(RQ_SECURE, &rqstp->rq_flags); 273 else 274 clear_bit(RQ_SECURE, &rqstp->rq_flags); 275 } 276 277 /* 278 * INET callback when data has been received on the socket. 279 */ 280 static void svc_data_ready(struct sock *sk) 281 { 282 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 283 284 if (svsk) { 285 dprintk("svc: socket %p(inet %p), busy=%d\n", 286 svsk, sk, 287 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 288 289 /* Refer to svc_setup_socket() for details. */ 290 rmb(); 291 svsk->sk_odata(sk); 292 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags)) 293 svc_xprt_enqueue(&svsk->sk_xprt); 294 } 295 } 296 297 /* 298 * INET callback when space is newly available on the socket. 299 */ 300 static void svc_write_space(struct sock *sk) 301 { 302 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 303 304 if (svsk) { 305 dprintk("svc: socket %p(inet %p), write_space busy=%d\n", 306 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 307 308 /* Refer to svc_setup_socket() for details. */ 309 rmb(); 310 svsk->sk_owspace(sk); 311 svc_xprt_enqueue(&svsk->sk_xprt); 312 } 313 } 314 315 static int svc_tcp_has_wspace(struct svc_xprt *xprt) 316 { 317 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 318 319 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) 320 return 1; 321 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 322 } 323 324 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt) 325 { 326 struct svc_sock *svsk; 327 struct socket *sock; 328 struct linger no_linger = { 329 .l_onoff = 1, 330 .l_linger = 0, 331 }; 332 333 svsk = container_of(xprt, struct svc_sock, sk_xprt); 334 sock = svsk->sk_sock; 335 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER, 336 (char *)&no_linger, sizeof(no_linger)); 337 } 338 339 /* 340 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo 341 */ 342 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, 343 struct cmsghdr *cmh) 344 { 345 struct in_pktinfo *pki = CMSG_DATA(cmh); 346 struct sockaddr_in *daddr = svc_daddr_in(rqstp); 347 348 if (cmh->cmsg_type != IP_PKTINFO) 349 return 0; 350 351 daddr->sin_family = AF_INET; 352 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; 353 return 1; 354 } 355 356 /* 357 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl 358 */ 359 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, 360 struct cmsghdr *cmh) 361 { 362 struct in6_pktinfo *pki = CMSG_DATA(cmh); 363 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 364 365 if (cmh->cmsg_type != IPV6_PKTINFO) 366 return 0; 367 368 daddr->sin6_family = AF_INET6; 369 daddr->sin6_addr = pki->ipi6_addr; 370 daddr->sin6_scope_id = pki->ipi6_ifindex; 371 return 1; 372 } 373 374 /* 375 * Copy the UDP datagram's destination address to the rqstp structure. 376 * The 'destination' address in this case is the address to which the 377 * peer sent the datagram, i.e. our local address. For multihomed 378 * hosts, this can change from msg to msg. Note that only the IP 379 * address changes, the port number should remain the same. 380 */ 381 static int svc_udp_get_dest_address(struct svc_rqst *rqstp, 382 struct cmsghdr *cmh) 383 { 384 switch (cmh->cmsg_level) { 385 case SOL_IP: 386 return svc_udp_get_dest_address4(rqstp, cmh); 387 case SOL_IPV6: 388 return svc_udp_get_dest_address6(rqstp, cmh); 389 } 390 391 return 0; 392 } 393 394 /* 395 * Receive a datagram from a UDP socket. 396 */ 397 static int svc_udp_recvfrom(struct svc_rqst *rqstp) 398 { 399 struct svc_sock *svsk = 400 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 401 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 402 struct sk_buff *skb; 403 union { 404 struct cmsghdr hdr; 405 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 406 } buffer; 407 struct cmsghdr *cmh = &buffer.hdr; 408 struct msghdr msg = { 409 .msg_name = svc_addr(rqstp), 410 .msg_control = cmh, 411 .msg_controllen = sizeof(buffer), 412 .msg_flags = MSG_DONTWAIT, 413 }; 414 size_t len; 415 int err; 416 417 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 418 /* udp sockets need large rcvbuf as all pending 419 * requests are still in that buffer. sndbuf must 420 * also be large enough that there is enough space 421 * for one reply per thread. We count all threads 422 * rather than threads in a particular pool, which 423 * provides an upper bound on the number of threads 424 * which will access the socket. 425 */ 426 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3); 427 428 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 429 skb = NULL; 430 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 431 0, 0, MSG_PEEK | MSG_DONTWAIT); 432 if (err >= 0) 433 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err); 434 435 if (skb == NULL) { 436 if (err != -EAGAIN) { 437 /* possibly an icmp error */ 438 dprintk("svc: recvfrom returned error %d\n", -err); 439 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 440 } 441 return 0; 442 } 443 len = svc_addr_len(svc_addr(rqstp)); 444 rqstp->rq_addrlen = len; 445 if (skb->tstamp == 0) { 446 skb->tstamp = ktime_get_real(); 447 /* Don't enable netstamp, sunrpc doesn't 448 need that much accuracy */ 449 } 450 sock_write_timestamp(svsk->sk_sk, skb->tstamp); 451 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ 452 453 len = skb->len; 454 rqstp->rq_arg.len = len; 455 456 rqstp->rq_prot = IPPROTO_UDP; 457 458 if (!svc_udp_get_dest_address(rqstp, cmh)) { 459 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", 460 cmh->cmsg_level, cmh->cmsg_type); 461 goto out_free; 462 } 463 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); 464 465 if (skb_is_nonlinear(skb)) { 466 /* we have to copy */ 467 local_bh_disable(); 468 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { 469 local_bh_enable(); 470 /* checksum error */ 471 goto out_free; 472 } 473 local_bh_enable(); 474 consume_skb(skb); 475 } else { 476 /* we can use it in-place */ 477 rqstp->rq_arg.head[0].iov_base = skb->data; 478 rqstp->rq_arg.head[0].iov_len = len; 479 if (skb_checksum_complete(skb)) 480 goto out_free; 481 rqstp->rq_xprt_ctxt = skb; 482 } 483 484 rqstp->rq_arg.page_base = 0; 485 if (len <= rqstp->rq_arg.head[0].iov_len) { 486 rqstp->rq_arg.head[0].iov_len = len; 487 rqstp->rq_arg.page_len = 0; 488 rqstp->rq_respages = rqstp->rq_pages+1; 489 } else { 490 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 491 rqstp->rq_respages = rqstp->rq_pages + 1 + 492 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); 493 } 494 rqstp->rq_next_page = rqstp->rq_respages+1; 495 496 if (serv->sv_stats) 497 serv->sv_stats->netudpcnt++; 498 499 return len; 500 out_free: 501 kfree_skb(skb); 502 return 0; 503 } 504 505 /** 506 * svc_udp_sendto - Send out a reply on a UDP socket 507 * @rqstp: completed svc_rqst 508 * 509 * Returns the number of bytes sent, or a negative errno. 510 */ 511 static int svc_udp_sendto(struct svc_rqst *rqstp) 512 { 513 struct svc_xprt *xprt = rqstp->rq_xprt; 514 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 515 struct xdr_buf *xdr = &rqstp->rq_res; 516 union { 517 struct cmsghdr hdr; 518 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 519 } buffer; 520 struct cmsghdr *cmh = &buffer.hdr; 521 struct msghdr msg = { 522 .msg_name = &rqstp->rq_addr, 523 .msg_namelen = rqstp->rq_addrlen, 524 .msg_control = cmh, 525 .msg_controllen = sizeof(buffer), 526 }; 527 unsigned int uninitialized_var(sent); 528 int err; 529 530 svc_release_udp_skb(rqstp); 531 532 svc_set_cmsg_data(rqstp, cmh); 533 534 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent); 535 xdr_free_bvec(xdr); 536 if (err == -ECONNREFUSED) { 537 /* ICMP error on earlier request. */ 538 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent); 539 xdr_free_bvec(xdr); 540 } 541 if (err < 0) 542 return err; 543 return sent; 544 } 545 546 static int svc_udp_has_wspace(struct svc_xprt *xprt) 547 { 548 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 549 struct svc_serv *serv = xprt->xpt_server; 550 unsigned long required; 551 552 /* 553 * Set the SOCK_NOSPACE flag before checking the available 554 * sock space. 555 */ 556 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 557 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; 558 if (required*2 > sock_wspace(svsk->sk_sk)) 559 return 0; 560 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 561 return 1; 562 } 563 564 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) 565 { 566 BUG(); 567 return NULL; 568 } 569 570 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt) 571 { 572 } 573 574 static struct svc_xprt *svc_udp_create(struct svc_serv *serv, 575 struct net *net, 576 struct sockaddr *sa, int salen, 577 int flags) 578 { 579 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); 580 } 581 582 static const struct svc_xprt_ops svc_udp_ops = { 583 .xpo_create = svc_udp_create, 584 .xpo_recvfrom = svc_udp_recvfrom, 585 .xpo_sendto = svc_udp_sendto, 586 .xpo_read_payload = svc_sock_read_payload, 587 .xpo_release_rqst = svc_release_udp_skb, 588 .xpo_detach = svc_sock_detach, 589 .xpo_free = svc_sock_free, 590 .xpo_has_wspace = svc_udp_has_wspace, 591 .xpo_accept = svc_udp_accept, 592 .xpo_secure_port = svc_sock_secure_port, 593 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt, 594 }; 595 596 static struct svc_xprt_class svc_udp_class = { 597 .xcl_name = "udp", 598 .xcl_owner = THIS_MODULE, 599 .xcl_ops = &svc_udp_ops, 600 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, 601 .xcl_ident = XPRT_TRANSPORT_UDP, 602 }; 603 604 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) 605 { 606 int err, level, optname, one = 1; 607 608 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, 609 &svsk->sk_xprt, serv); 610 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 611 svsk->sk_sk->sk_data_ready = svc_data_ready; 612 svsk->sk_sk->sk_write_space = svc_write_space; 613 614 /* initialise setting must have enough space to 615 * receive and respond to one request. 616 * svc_udp_recvfrom will re-adjust if necessary 617 */ 618 svc_sock_setbufsize(svsk, 3); 619 620 /* data might have come in before data_ready set up */ 621 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 622 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 623 624 /* make sure we get destination address info */ 625 switch (svsk->sk_sk->sk_family) { 626 case AF_INET: 627 level = SOL_IP; 628 optname = IP_PKTINFO; 629 break; 630 case AF_INET6: 631 level = SOL_IPV6; 632 optname = IPV6_RECVPKTINFO; 633 break; 634 default: 635 BUG(); 636 } 637 err = kernel_setsockopt(svsk->sk_sock, level, optname, 638 (char *)&one, sizeof(one)); 639 dprintk("svc: kernel_setsockopt returned %d\n", err); 640 } 641 642 /* 643 * A data_ready event on a listening socket means there's a connection 644 * pending. Do not use state_change as a substitute for it. 645 */ 646 static void svc_tcp_listen_data_ready(struct sock *sk) 647 { 648 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 649 650 dprintk("svc: socket %p TCP (listen) state change %d\n", 651 sk, sk->sk_state); 652 653 if (svsk) { 654 /* Refer to svc_setup_socket() for details. */ 655 rmb(); 656 svsk->sk_odata(sk); 657 } 658 659 /* 660 * This callback may called twice when a new connection 661 * is established as a child socket inherits everything 662 * from a parent LISTEN socket. 663 * 1) data_ready method of the parent socket will be called 664 * when one of child sockets become ESTABLISHED. 665 * 2) data_ready method of the child socket may be called 666 * when it receives data before the socket is accepted. 667 * In case of 2, we should ignore it silently. 668 */ 669 if (sk->sk_state == TCP_LISTEN) { 670 if (svsk) { 671 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 672 svc_xprt_enqueue(&svsk->sk_xprt); 673 } else 674 printk("svc: socket %p: no user data\n", sk); 675 } 676 } 677 678 /* 679 * A state change on a connected socket means it's dying or dead. 680 */ 681 static void svc_tcp_state_change(struct sock *sk) 682 { 683 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 684 685 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", 686 sk, sk->sk_state, sk->sk_user_data); 687 688 if (!svsk) 689 printk("svc: socket %p: no user data\n", sk); 690 else { 691 /* Refer to svc_setup_socket() for details. */ 692 rmb(); 693 svsk->sk_ostate(sk); 694 if (sk->sk_state != TCP_ESTABLISHED) { 695 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 696 svc_xprt_enqueue(&svsk->sk_xprt); 697 } 698 } 699 } 700 701 /* 702 * Accept a TCP connection 703 */ 704 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) 705 { 706 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 707 struct sockaddr_storage addr; 708 struct sockaddr *sin = (struct sockaddr *) &addr; 709 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 710 struct socket *sock = svsk->sk_sock; 711 struct socket *newsock; 712 struct svc_sock *newsvsk; 713 int err, slen; 714 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 715 716 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); 717 if (!sock) 718 return NULL; 719 720 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 721 err = kernel_accept(sock, &newsock, O_NONBLOCK); 722 if (err < 0) { 723 if (err == -ENOMEM) 724 printk(KERN_WARNING "%s: no more sockets!\n", 725 serv->sv_name); 726 else if (err != -EAGAIN) 727 net_warn_ratelimited("%s: accept failed (err %d)!\n", 728 serv->sv_name, -err); 729 return NULL; 730 } 731 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 732 733 err = kernel_getpeername(newsock, sin); 734 if (err < 0) { 735 net_warn_ratelimited("%s: peername failed (err %d)!\n", 736 serv->sv_name, -err); 737 goto failed; /* aborted connection or whatever */ 738 } 739 slen = err; 740 741 /* Ideally, we would want to reject connections from unauthorized 742 * hosts here, but when we get encryption, the IP of the host won't 743 * tell us anything. For now just warn about unpriv connections. 744 */ 745 if (!svc_port_is_privileged(sin)) { 746 dprintk("%s: connect from unprivileged port: %s\n", 747 serv->sv_name, 748 __svc_print_addr(sin, buf, sizeof(buf))); 749 } 750 dprintk("%s: connect from %s\n", serv->sv_name, 751 __svc_print_addr(sin, buf, sizeof(buf))); 752 753 /* Reset the inherited callbacks before calling svc_setup_socket */ 754 newsock->sk->sk_state_change = svsk->sk_ostate; 755 newsock->sk->sk_data_ready = svsk->sk_odata; 756 newsock->sk->sk_write_space = svsk->sk_owspace; 757 758 /* make sure that a write doesn't block forever when 759 * low on memory 760 */ 761 newsock->sk->sk_sndtimeo = HZ*30; 762 763 newsvsk = svc_setup_socket(serv, newsock, 764 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); 765 if (IS_ERR(newsvsk)) 766 goto failed; 767 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); 768 err = kernel_getsockname(newsock, sin); 769 slen = err; 770 if (unlikely(err < 0)) { 771 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); 772 slen = offsetof(struct sockaddr, sa_data); 773 } 774 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); 775 776 if (sock_is_loopback(newsock->sk)) 777 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 778 else 779 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 780 if (serv->sv_stats) 781 serv->sv_stats->nettcpconn++; 782 783 return &newsvsk->sk_xprt; 784 785 failed: 786 sock_release(newsock); 787 return NULL; 788 } 789 790 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 791 { 792 unsigned int i, len, npages; 793 794 if (svsk->sk_datalen == 0) 795 return 0; 796 len = svsk->sk_datalen; 797 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 798 for (i = 0; i < npages; i++) { 799 if (rqstp->rq_pages[i] != NULL) 800 put_page(rqstp->rq_pages[i]); 801 BUG_ON(svsk->sk_pages[i] == NULL); 802 rqstp->rq_pages[i] = svsk->sk_pages[i]; 803 svsk->sk_pages[i] = NULL; 804 } 805 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); 806 return len; 807 } 808 809 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 810 { 811 unsigned int i, len, npages; 812 813 if (svsk->sk_datalen == 0) 814 return; 815 len = svsk->sk_datalen; 816 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 817 for (i = 0; i < npages; i++) { 818 svsk->sk_pages[i] = rqstp->rq_pages[i]; 819 rqstp->rq_pages[i] = NULL; 820 } 821 } 822 823 static void svc_tcp_clear_pages(struct svc_sock *svsk) 824 { 825 unsigned int i, len, npages; 826 827 if (svsk->sk_datalen == 0) 828 goto out; 829 len = svsk->sk_datalen; 830 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 831 for (i = 0; i < npages; i++) { 832 if (svsk->sk_pages[i] == NULL) { 833 WARN_ON_ONCE(1); 834 continue; 835 } 836 put_page(svsk->sk_pages[i]); 837 svsk->sk_pages[i] = NULL; 838 } 839 out: 840 svsk->sk_tcplen = 0; 841 svsk->sk_datalen = 0; 842 } 843 844 /* 845 * Receive fragment record header. 846 * If we haven't gotten the record length yet, get the next four bytes. 847 */ 848 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) 849 { 850 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 851 unsigned int want; 852 int len; 853 854 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { 855 struct kvec iov; 856 857 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; 858 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; 859 iov.iov_len = want; 860 len = svc_recvfrom(rqstp, &iov, 1, want, 0); 861 if (len < 0) 862 goto error; 863 svsk->sk_tcplen += len; 864 865 if (len < want) { 866 dprintk("svc: short recvfrom while reading record " 867 "length (%d of %d)\n", len, want); 868 return -EAGAIN; 869 } 870 871 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk)); 872 if (svc_sock_reclen(svsk) + svsk->sk_datalen > 873 serv->sv_max_mesg) { 874 net_notice_ratelimited("RPC: fragment too large: %d\n", 875 svc_sock_reclen(svsk)); 876 goto err_delete; 877 } 878 } 879 880 return svc_sock_reclen(svsk); 881 error: 882 dprintk("RPC: TCP recv_record got %d\n", len); 883 return len; 884 err_delete: 885 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 886 return -EAGAIN; 887 } 888 889 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) 890 { 891 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; 892 struct rpc_rqst *req = NULL; 893 struct kvec *src, *dst; 894 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 895 __be32 xid; 896 __be32 calldir; 897 898 xid = *p++; 899 calldir = *p; 900 901 if (!bc_xprt) 902 return -EAGAIN; 903 spin_lock(&bc_xprt->queue_lock); 904 req = xprt_lookup_rqst(bc_xprt, xid); 905 if (!req) 906 goto unlock_notfound; 907 908 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); 909 /* 910 * XXX!: cheating for now! Only copying HEAD. 911 * But we know this is good enough for now (in fact, for any 912 * callback reply in the forseeable future). 913 */ 914 dst = &req->rq_private_buf.head[0]; 915 src = &rqstp->rq_arg.head[0]; 916 if (dst->iov_len < src->iov_len) 917 goto unlock_eagain; /* whatever; just giving up. */ 918 memcpy(dst->iov_base, src->iov_base, src->iov_len); 919 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); 920 rqstp->rq_arg.len = 0; 921 spin_unlock(&bc_xprt->queue_lock); 922 return 0; 923 unlock_notfound: 924 printk(KERN_NOTICE 925 "%s: Got unrecognized reply: " 926 "calldir 0x%x xpt_bc_xprt %p xid %08x\n", 927 __func__, ntohl(calldir), 928 bc_xprt, ntohl(xid)); 929 unlock_eagain: 930 spin_unlock(&bc_xprt->queue_lock); 931 return -EAGAIN; 932 } 933 934 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len) 935 { 936 int i = 0; 937 int t = 0; 938 939 while (t < len) { 940 vec[i].iov_base = page_address(pages[i]); 941 vec[i].iov_len = PAGE_SIZE; 942 i++; 943 t += PAGE_SIZE; 944 } 945 return i; 946 } 947 948 static void svc_tcp_fragment_received(struct svc_sock *svsk) 949 { 950 /* If we have more data, signal svc_xprt_enqueue() to try again */ 951 dprintk("svc: TCP %s record (%d bytes)\n", 952 svc_sock_final_rec(svsk) ? "final" : "nonfinal", 953 svc_sock_reclen(svsk)); 954 svsk->sk_tcplen = 0; 955 svsk->sk_reclen = 0; 956 } 957 958 /* 959 * Receive data from a TCP socket. 960 */ 961 static int svc_tcp_recvfrom(struct svc_rqst *rqstp) 962 { 963 struct svc_sock *svsk = 964 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 965 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 966 int len; 967 struct kvec *vec; 968 unsigned int want, base; 969 __be32 *p; 970 __be32 calldir; 971 int pnum; 972 973 dprintk("svc: tcp_recv %p data %d conn %d close %d\n", 974 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), 975 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), 976 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 977 978 len = svc_tcp_recv_record(svsk, rqstp); 979 if (len < 0) 980 goto error; 981 982 base = svc_tcp_restore_pages(svsk, rqstp); 983 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); 984 985 vec = rqstp->rq_vec; 986 987 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], base + want); 988 989 rqstp->rq_respages = &rqstp->rq_pages[pnum]; 990 rqstp->rq_next_page = rqstp->rq_respages + 1; 991 992 /* Now receive data */ 993 len = svc_recvfrom(rqstp, vec, pnum, base + want, base); 994 if (len >= 0) { 995 svsk->sk_tcplen += len; 996 svsk->sk_datalen += len; 997 } 998 if (len != want || !svc_sock_final_rec(svsk)) { 999 svc_tcp_save_pages(svsk, rqstp); 1000 if (len < 0 && len != -EAGAIN) 1001 goto err_delete; 1002 if (len == want) 1003 svc_tcp_fragment_received(svsk); 1004 else 1005 dprintk("svc: incomplete TCP record (%d of %d)\n", 1006 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)), 1007 svc_sock_reclen(svsk)); 1008 goto err_noclose; 1009 } 1010 1011 if (svsk->sk_datalen < 8) { 1012 svsk->sk_datalen = 0; 1013 goto err_delete; /* client is nuts. */ 1014 } 1015 1016 rqstp->rq_arg.len = svsk->sk_datalen; 1017 rqstp->rq_arg.page_base = 0; 1018 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { 1019 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; 1020 rqstp->rq_arg.page_len = 0; 1021 } else 1022 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1023 1024 rqstp->rq_xprt_ctxt = NULL; 1025 rqstp->rq_prot = IPPROTO_TCP; 1026 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags)) 1027 set_bit(RQ_LOCAL, &rqstp->rq_flags); 1028 else 1029 clear_bit(RQ_LOCAL, &rqstp->rq_flags); 1030 1031 p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1032 calldir = p[1]; 1033 if (calldir) 1034 len = receive_cb_reply(svsk, rqstp); 1035 1036 /* Reset TCP read info */ 1037 svsk->sk_datalen = 0; 1038 svc_tcp_fragment_received(svsk); 1039 1040 if (len < 0) 1041 goto error; 1042 1043 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); 1044 if (serv->sv_stats) 1045 serv->sv_stats->nettcpcnt++; 1046 1047 return rqstp->rq_arg.len; 1048 1049 error: 1050 if (len != -EAGAIN) 1051 goto err_delete; 1052 dprintk("RPC: TCP recvfrom got EAGAIN\n"); 1053 return 0; 1054 err_delete: 1055 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", 1056 svsk->sk_xprt.xpt_server->sv_name, -len); 1057 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1058 err_noclose: 1059 return 0; /* record not complete */ 1060 } 1061 1062 /** 1063 * svc_tcp_sendto - Send out a reply on a TCP socket 1064 * @rqstp: completed svc_rqst 1065 * 1066 * Returns the number of bytes sent, or a negative errno. 1067 */ 1068 static int svc_tcp_sendto(struct svc_rqst *rqstp) 1069 { 1070 struct svc_xprt *xprt = rqstp->rq_xprt; 1071 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1072 struct xdr_buf *xdr = &rqstp->rq_res; 1073 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 1074 (u32)xdr->len); 1075 struct msghdr msg = { 1076 .msg_flags = 0, 1077 }; 1078 unsigned int uninitialized_var(sent); 1079 int err; 1080 1081 svc_release_skb(rqstp); 1082 1083 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, marker, &sent); 1084 xdr_free_bvec(xdr); 1085 if (err < 0 || sent != (xdr->len + sizeof(marker))) 1086 goto out_close; 1087 return sent; 1088 1089 out_close: 1090 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", 1091 xprt->xpt_server->sv_name, 1092 (err < 0) ? "got error" : "sent", 1093 (err < 0) ? err : sent, xdr->len); 1094 set_bit(XPT_CLOSE, &xprt->xpt_flags); 1095 svc_xprt_enqueue(xprt); 1096 return -EAGAIN; 1097 } 1098 1099 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, 1100 struct net *net, 1101 struct sockaddr *sa, int salen, 1102 int flags) 1103 { 1104 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1105 } 1106 1107 static const struct svc_xprt_ops svc_tcp_ops = { 1108 .xpo_create = svc_tcp_create, 1109 .xpo_recvfrom = svc_tcp_recvfrom, 1110 .xpo_sendto = svc_tcp_sendto, 1111 .xpo_read_payload = svc_sock_read_payload, 1112 .xpo_release_rqst = svc_release_skb, 1113 .xpo_detach = svc_tcp_sock_detach, 1114 .xpo_free = svc_sock_free, 1115 .xpo_has_wspace = svc_tcp_has_wspace, 1116 .xpo_accept = svc_tcp_accept, 1117 .xpo_secure_port = svc_sock_secure_port, 1118 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt, 1119 }; 1120 1121 static struct svc_xprt_class svc_tcp_class = { 1122 .xcl_name = "tcp", 1123 .xcl_owner = THIS_MODULE, 1124 .xcl_ops = &svc_tcp_ops, 1125 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1126 .xcl_ident = XPRT_TRANSPORT_TCP, 1127 }; 1128 1129 void svc_init_xprt_sock(void) 1130 { 1131 svc_reg_xprt_class(&svc_tcp_class); 1132 svc_reg_xprt_class(&svc_udp_class); 1133 } 1134 1135 void svc_cleanup_xprt_sock(void) 1136 { 1137 svc_unreg_xprt_class(&svc_tcp_class); 1138 svc_unreg_xprt_class(&svc_udp_class); 1139 } 1140 1141 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) 1142 { 1143 struct sock *sk = svsk->sk_sk; 1144 1145 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, 1146 &svsk->sk_xprt, serv); 1147 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 1148 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags); 1149 if (sk->sk_state == TCP_LISTEN) { 1150 dprintk("setting up TCP socket for listening\n"); 1151 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener"); 1152 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); 1153 sk->sk_data_ready = svc_tcp_listen_data_ready; 1154 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 1155 } else { 1156 dprintk("setting up TCP socket for reading\n"); 1157 sk->sk_state_change = svc_tcp_state_change; 1158 sk->sk_data_ready = svc_data_ready; 1159 sk->sk_write_space = svc_write_space; 1160 1161 svsk->sk_reclen = 0; 1162 svsk->sk_tcplen = 0; 1163 svsk->sk_datalen = 0; 1164 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); 1165 1166 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1167 1168 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1169 switch (sk->sk_state) { 1170 case TCP_SYN_RECV: 1171 case TCP_ESTABLISHED: 1172 break; 1173 default: 1174 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1175 } 1176 } 1177 } 1178 1179 void svc_sock_update_bufs(struct svc_serv *serv) 1180 { 1181 /* 1182 * The number of server threads has changed. Update 1183 * rcvbuf and sndbuf accordingly on all sockets 1184 */ 1185 struct svc_sock *svsk; 1186 1187 spin_lock_bh(&serv->sv_lock); 1188 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) 1189 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1190 spin_unlock_bh(&serv->sv_lock); 1191 } 1192 EXPORT_SYMBOL_GPL(svc_sock_update_bufs); 1193 1194 /* 1195 * Initialize socket for RPC use and create svc_sock struct 1196 */ 1197 static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1198 struct socket *sock, 1199 int flags) 1200 { 1201 struct svc_sock *svsk; 1202 struct sock *inet; 1203 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1204 int err = 0; 1205 1206 dprintk("svc: svc_setup_socket %p\n", sock); 1207 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1208 if (!svsk) 1209 return ERR_PTR(-ENOMEM); 1210 1211 inet = sock->sk; 1212 1213 /* Register socket with portmapper */ 1214 if (pmap_register) 1215 err = svc_register(serv, sock_net(sock->sk), inet->sk_family, 1216 inet->sk_protocol, 1217 ntohs(inet_sk(inet)->inet_sport)); 1218 1219 if (err < 0) { 1220 kfree(svsk); 1221 return ERR_PTR(err); 1222 } 1223 1224 svsk->sk_sock = sock; 1225 svsk->sk_sk = inet; 1226 svsk->sk_ostate = inet->sk_state_change; 1227 svsk->sk_odata = inet->sk_data_ready; 1228 svsk->sk_owspace = inet->sk_write_space; 1229 /* 1230 * This barrier is necessary in order to prevent race condition 1231 * with svc_data_ready(), svc_listen_data_ready() and others 1232 * when calling callbacks above. 1233 */ 1234 wmb(); 1235 inet->sk_user_data = svsk; 1236 1237 /* Initialize the socket */ 1238 if (sock->type == SOCK_DGRAM) 1239 svc_udp_init(svsk, serv); 1240 else 1241 svc_tcp_init(svsk, serv); 1242 1243 dprintk("svc: svc_setup_socket created %p (inet %p), " 1244 "listen %d close %d\n", 1245 svsk, svsk->sk_sk, 1246 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags), 1247 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1248 1249 return svsk; 1250 } 1251 1252 bool svc_alien_sock(struct net *net, int fd) 1253 { 1254 int err; 1255 struct socket *sock = sockfd_lookup(fd, &err); 1256 bool ret = false; 1257 1258 if (!sock) 1259 goto out; 1260 if (sock_net(sock->sk) != net) 1261 ret = true; 1262 sockfd_put(sock); 1263 out: 1264 return ret; 1265 } 1266 EXPORT_SYMBOL_GPL(svc_alien_sock); 1267 1268 /** 1269 * svc_addsock - add a listener socket to an RPC service 1270 * @serv: pointer to RPC service to which to add a new listener 1271 * @fd: file descriptor of the new listener 1272 * @name_return: pointer to buffer to fill in with name of listener 1273 * @len: size of the buffer 1274 * @cred: credential 1275 * 1276 * Fills in socket name and returns positive length of name if successful. 1277 * Name is terminated with '\n'. On error, returns a negative errno 1278 * value. 1279 */ 1280 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, 1281 const size_t len, const struct cred *cred) 1282 { 1283 int err = 0; 1284 struct socket *so = sockfd_lookup(fd, &err); 1285 struct svc_sock *svsk = NULL; 1286 struct sockaddr_storage addr; 1287 struct sockaddr *sin = (struct sockaddr *)&addr; 1288 int salen; 1289 1290 if (!so) 1291 return err; 1292 err = -EAFNOSUPPORT; 1293 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) 1294 goto out; 1295 err = -EPROTONOSUPPORT; 1296 if (so->sk->sk_protocol != IPPROTO_TCP && 1297 so->sk->sk_protocol != IPPROTO_UDP) 1298 goto out; 1299 err = -EISCONN; 1300 if (so->state > SS_UNCONNECTED) 1301 goto out; 1302 err = -ENOENT; 1303 if (!try_module_get(THIS_MODULE)) 1304 goto out; 1305 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); 1306 if (IS_ERR(svsk)) { 1307 module_put(THIS_MODULE); 1308 err = PTR_ERR(svsk); 1309 goto out; 1310 } 1311 salen = kernel_getsockname(svsk->sk_sock, sin); 1312 if (salen >= 0) 1313 svc_xprt_set_local(&svsk->sk_xprt, sin, salen); 1314 svsk->sk_xprt.xpt_cred = get_cred(cred); 1315 svc_add_new_perm_xprt(serv, &svsk->sk_xprt); 1316 return svc_one_sock_name(svsk, name_return, len); 1317 out: 1318 sockfd_put(so); 1319 return err; 1320 } 1321 EXPORT_SYMBOL_GPL(svc_addsock); 1322 1323 /* 1324 * Create socket for RPC service. 1325 */ 1326 static struct svc_xprt *svc_create_socket(struct svc_serv *serv, 1327 int protocol, 1328 struct net *net, 1329 struct sockaddr *sin, int len, 1330 int flags) 1331 { 1332 struct svc_sock *svsk; 1333 struct socket *sock; 1334 int error; 1335 int type; 1336 struct sockaddr_storage addr; 1337 struct sockaddr *newsin = (struct sockaddr *)&addr; 1338 int newlen; 1339 int family; 1340 int val; 1341 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 1342 1343 dprintk("svc: svc_create_socket(%s, %d, %s)\n", 1344 serv->sv_program->pg_name, protocol, 1345 __svc_print_addr(sin, buf, sizeof(buf))); 1346 1347 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1348 printk(KERN_WARNING "svc: only UDP and TCP " 1349 "sockets supported\n"); 1350 return ERR_PTR(-EINVAL); 1351 } 1352 1353 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1354 switch (sin->sa_family) { 1355 case AF_INET6: 1356 family = PF_INET6; 1357 break; 1358 case AF_INET: 1359 family = PF_INET; 1360 break; 1361 default: 1362 return ERR_PTR(-EINVAL); 1363 } 1364 1365 error = __sock_create(net, family, type, protocol, &sock, 1); 1366 if (error < 0) 1367 return ERR_PTR(error); 1368 1369 svc_reclassify_socket(sock); 1370 1371 /* 1372 * If this is an PF_INET6 listener, we want to avoid 1373 * getting requests from IPv4 remotes. Those should 1374 * be shunted to a PF_INET listener via rpcbind. 1375 */ 1376 val = 1; 1377 if (family == PF_INET6) 1378 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, 1379 (char *)&val, sizeof(val)); 1380 1381 if (type == SOCK_STREAM) 1382 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ 1383 error = kernel_bind(sock, sin, len); 1384 if (error < 0) 1385 goto bummer; 1386 1387 error = kernel_getsockname(sock, newsin); 1388 if (error < 0) 1389 goto bummer; 1390 newlen = error; 1391 1392 if (protocol == IPPROTO_TCP) { 1393 if ((error = kernel_listen(sock, 64)) < 0) 1394 goto bummer; 1395 } 1396 1397 svsk = svc_setup_socket(serv, sock, flags); 1398 if (IS_ERR(svsk)) { 1399 error = PTR_ERR(svsk); 1400 goto bummer; 1401 } 1402 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); 1403 return (struct svc_xprt *)svsk; 1404 bummer: 1405 dprintk("svc: svc_create_socket error = %d\n", -error); 1406 sock_release(sock); 1407 return ERR_PTR(error); 1408 } 1409 1410 /* 1411 * Detach the svc_sock from the socket so that no 1412 * more callbacks occur. 1413 */ 1414 static void svc_sock_detach(struct svc_xprt *xprt) 1415 { 1416 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1417 struct sock *sk = svsk->sk_sk; 1418 1419 dprintk("svc: svc_sock_detach(%p)\n", svsk); 1420 1421 /* put back the old socket callbacks */ 1422 lock_sock(sk); 1423 sk->sk_state_change = svsk->sk_ostate; 1424 sk->sk_data_ready = svsk->sk_odata; 1425 sk->sk_write_space = svsk->sk_owspace; 1426 sk->sk_user_data = NULL; 1427 release_sock(sk); 1428 } 1429 1430 /* 1431 * Disconnect the socket, and reset the callbacks 1432 */ 1433 static void svc_tcp_sock_detach(struct svc_xprt *xprt) 1434 { 1435 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1436 1437 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); 1438 1439 svc_sock_detach(xprt); 1440 1441 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { 1442 svc_tcp_clear_pages(svsk); 1443 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); 1444 } 1445 } 1446 1447 /* 1448 * Free the svc_sock's socket resources and the svc_sock itself. 1449 */ 1450 static void svc_sock_free(struct svc_xprt *xprt) 1451 { 1452 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1453 dprintk("svc: svc_sock_free(%p)\n", svsk); 1454 1455 if (svsk->sk_sock->file) 1456 sockfd_put(svsk->sk_sock); 1457 else 1458 sock_release(svsk->sk_sock); 1459 kfree(svsk); 1460 } 1461