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 <net/tls.h> 47 #include <net/tls_prot.h> 48 #include <net/handshake.h> 49 #include <linux/uaccess.h> 50 #include <linux/highmem.h> 51 #include <asm/ioctls.h> 52 #include <linux/key.h> 53 54 #include <linux/sunrpc/types.h> 55 #include <linux/sunrpc/clnt.h> 56 #include <linux/sunrpc/xdr.h> 57 #include <linux/sunrpc/msg_prot.h> 58 #include <linux/sunrpc/svcsock.h> 59 #include <linux/sunrpc/stats.h> 60 #include <linux/sunrpc/xprt.h> 61 62 #include <trace/events/sock.h> 63 #include <trace/events/sunrpc.h> 64 65 #include "socklib.h" 66 #include "sunrpc.h" 67 68 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 69 70 /* To-do: to avoid tying up an nfsd thread while waiting for a 71 * handshake request, the request could instead be deferred. 72 */ 73 enum { 74 SVC_HANDSHAKE_TO = 5U * HZ 75 }; 76 77 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, 78 int flags); 79 static int svc_udp_recvfrom(struct svc_rqst *); 80 static int svc_udp_sendto(struct svc_rqst *); 81 static void svc_sock_detach(struct svc_xprt *); 82 static void svc_tcp_sock_detach(struct svc_xprt *); 83 static void svc_sock_free(struct svc_xprt *); 84 85 static struct svc_xprt *svc_create_socket(struct svc_serv *, int, 86 struct net *, struct sockaddr *, 87 int, int); 88 #ifdef CONFIG_DEBUG_LOCK_ALLOC 89 static struct lock_class_key svc_key[2]; 90 static struct lock_class_key svc_slock_key[2]; 91 92 static void svc_reclassify_socket(struct socket *sock) 93 { 94 struct sock *sk = sock->sk; 95 96 if (WARN_ON_ONCE(!sock_allow_reclassification(sk))) 97 return; 98 99 switch (sk->sk_family) { 100 case AF_INET: 101 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", 102 &svc_slock_key[0], 103 "sk_xprt.xpt_lock-AF_INET-NFSD", 104 &svc_key[0]); 105 break; 106 107 case AF_INET6: 108 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", 109 &svc_slock_key[1], 110 "sk_xprt.xpt_lock-AF_INET6-NFSD", 111 &svc_key[1]); 112 break; 113 114 default: 115 BUG(); 116 } 117 } 118 #else 119 static void svc_reclassify_socket(struct socket *sock) 120 { 121 } 122 #endif 123 124 /** 125 * svc_tcp_release_ctxt - Release transport-related resources 126 * @xprt: the transport which owned the context 127 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt 128 * 129 */ 130 static void svc_tcp_release_ctxt(struct svc_xprt *xprt, void *ctxt) 131 { 132 } 133 134 /** 135 * svc_udp_release_ctxt - Release transport-related resources 136 * @xprt: the transport which owned the context 137 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt 138 * 139 */ 140 static void svc_udp_release_ctxt(struct svc_xprt *xprt, void *ctxt) 141 { 142 struct sk_buff *skb = ctxt; 143 144 if (skb) 145 consume_skb(skb); 146 } 147 148 union svc_pktinfo_u { 149 struct in_pktinfo pkti; 150 struct in6_pktinfo pkti6; 151 }; 152 #define SVC_PKTINFO_SPACE \ 153 CMSG_SPACE(sizeof(union svc_pktinfo_u)) 154 155 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) 156 { 157 struct svc_sock *svsk = 158 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 159 switch (svsk->sk_sk->sk_family) { 160 case AF_INET: { 161 struct in_pktinfo *pki = CMSG_DATA(cmh); 162 163 cmh->cmsg_level = SOL_IP; 164 cmh->cmsg_type = IP_PKTINFO; 165 pki->ipi_ifindex = 0; 166 pki->ipi_spec_dst.s_addr = 167 svc_daddr_in(rqstp)->sin_addr.s_addr; 168 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 169 } 170 break; 171 172 case AF_INET6: { 173 struct in6_pktinfo *pki = CMSG_DATA(cmh); 174 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 175 176 cmh->cmsg_level = SOL_IPV6; 177 cmh->cmsg_type = IPV6_PKTINFO; 178 pki->ipi6_ifindex = daddr->sin6_scope_id; 179 pki->ipi6_addr = daddr->sin6_addr; 180 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 181 } 182 break; 183 } 184 } 185 186 static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset, 187 unsigned int length) 188 { 189 return 0; 190 } 191 192 /* 193 * Report socket names for nfsdfs 194 */ 195 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) 196 { 197 const struct sock *sk = svsk->sk_sk; 198 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? 199 "udp" : "tcp"; 200 int len; 201 202 switch (sk->sk_family) { 203 case PF_INET: 204 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", 205 proto_name, 206 &inet_sk(sk)->inet_rcv_saddr, 207 inet_sk(sk)->inet_num); 208 break; 209 #if IS_ENABLED(CONFIG_IPV6) 210 case PF_INET6: 211 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", 212 proto_name, 213 &sk->sk_v6_rcv_saddr, 214 inet_sk(sk)->inet_num); 215 break; 216 #endif 217 default: 218 len = snprintf(buf, remaining, "*unknown-%d*\n", 219 sk->sk_family); 220 } 221 222 if (len >= remaining) { 223 *buf = '\0'; 224 return -ENAMETOOLONG; 225 } 226 return len; 227 } 228 229 static int 230 svc_tcp_sock_process_cmsg(struct svc_sock *svsk, struct msghdr *msg, 231 struct cmsghdr *cmsg, int ret) 232 { 233 if (cmsg->cmsg_level == SOL_TLS && 234 cmsg->cmsg_type == TLS_GET_RECORD_TYPE) { 235 u8 content_type = *((u8 *)CMSG_DATA(cmsg)); 236 237 switch (content_type) { 238 case TLS_RECORD_TYPE_DATA: 239 /* TLS sets EOR at the end of each application data 240 * record, even though there might be more frames 241 * waiting to be decrypted. 242 */ 243 msg->msg_flags &= ~MSG_EOR; 244 break; 245 case TLS_RECORD_TYPE_ALERT: 246 ret = -ENOTCONN; 247 break; 248 default: 249 ret = -EAGAIN; 250 } 251 } 252 return ret; 253 } 254 255 static int 256 svc_tcp_sock_recv_cmsg(struct svc_sock *svsk, struct msghdr *msg) 257 { 258 union { 259 struct cmsghdr cmsg; 260 u8 buf[CMSG_SPACE(sizeof(u8))]; 261 } u; 262 int ret; 263 264 msg->msg_control = &u; 265 msg->msg_controllen = sizeof(u); 266 ret = sock_recvmsg(svsk->sk_sock, msg, MSG_DONTWAIT); 267 if (unlikely(msg->msg_controllen != sizeof(u))) 268 ret = svc_tcp_sock_process_cmsg(svsk, msg, &u.cmsg, ret); 269 return ret; 270 } 271 272 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 273 static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek) 274 { 275 struct bvec_iter bi = { 276 .bi_size = size + seek, 277 }; 278 struct bio_vec bv; 279 280 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK); 281 for_each_bvec(bv, bvec, bi, bi) 282 flush_dcache_page(bv.bv_page); 283 } 284 #else 285 static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size, 286 size_t seek) 287 { 288 } 289 #endif 290 291 /* 292 * Read from @rqstp's transport socket. The incoming message fills whole 293 * pages in @rqstp's rq_pages array until the last page of the message 294 * has been received into a partial page. 295 */ 296 static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen, 297 size_t seek) 298 { 299 struct svc_sock *svsk = 300 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 301 struct bio_vec *bvec = rqstp->rq_bvec; 302 struct msghdr msg = { NULL }; 303 unsigned int i; 304 ssize_t len; 305 size_t t; 306 307 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 308 309 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE) 310 bvec_set_page(&bvec[i], rqstp->rq_pages[i], PAGE_SIZE, 0); 311 rqstp->rq_respages = &rqstp->rq_pages[i]; 312 rqstp->rq_next_page = rqstp->rq_respages + 1; 313 314 iov_iter_bvec(&msg.msg_iter, ITER_DEST, bvec, i, buflen); 315 if (seek) { 316 iov_iter_advance(&msg.msg_iter, seek); 317 buflen -= seek; 318 } 319 len = svc_tcp_sock_recv_cmsg(svsk, &msg); 320 if (len > 0) 321 svc_flush_bvec(bvec, len, seek); 322 323 /* If we read a full record, then assume there may be more 324 * data to read (stream based sockets only!) 325 */ 326 if (len == buflen) 327 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 328 329 return len; 330 } 331 332 /* 333 * Set socket snd and rcv buffer lengths 334 */ 335 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs) 336 { 337 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg; 338 struct socket *sock = svsk->sk_sock; 339 340 nreqs = min(nreqs, INT_MAX / 2 / max_mesg); 341 342 lock_sock(sock->sk); 343 sock->sk->sk_sndbuf = nreqs * max_mesg * 2; 344 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2; 345 sock->sk->sk_write_space(sock->sk); 346 release_sock(sock->sk); 347 } 348 349 static void svc_sock_secure_port(struct svc_rqst *rqstp) 350 { 351 if (svc_port_is_privileged(svc_addr(rqstp))) 352 set_bit(RQ_SECURE, &rqstp->rq_flags); 353 else 354 clear_bit(RQ_SECURE, &rqstp->rq_flags); 355 } 356 357 /* 358 * INET callback when data has been received on the socket. 359 */ 360 static void svc_data_ready(struct sock *sk) 361 { 362 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 363 364 trace_sk_data_ready(sk); 365 366 if (svsk) { 367 /* Refer to svc_setup_socket() for details. */ 368 rmb(); 369 svsk->sk_odata(sk); 370 trace_svcsock_data_ready(&svsk->sk_xprt, 0); 371 if (test_bit(XPT_HANDSHAKE, &svsk->sk_xprt.xpt_flags)) 372 return; 373 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags)) 374 svc_xprt_enqueue(&svsk->sk_xprt); 375 } 376 } 377 378 /* 379 * INET callback when space is newly available on the socket. 380 */ 381 static void svc_write_space(struct sock *sk) 382 { 383 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 384 385 if (svsk) { 386 /* Refer to svc_setup_socket() for details. */ 387 rmb(); 388 trace_svcsock_write_space(&svsk->sk_xprt, 0); 389 svsk->sk_owspace(sk); 390 svc_xprt_enqueue(&svsk->sk_xprt); 391 } 392 } 393 394 static int svc_tcp_has_wspace(struct svc_xprt *xprt) 395 { 396 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 397 398 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) 399 return 1; 400 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 401 } 402 403 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt) 404 { 405 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 406 407 sock_no_linger(svsk->sk_sock->sk); 408 } 409 410 /** 411 * svc_tcp_handshake_done - Handshake completion handler 412 * @data: address of xprt to wake 413 * @status: status of handshake 414 * @peerid: serial number of key containing the remote peer's identity 415 * 416 * If a security policy is specified as an export option, we don't 417 * have a specific export here to check. So we set a "TLS session 418 * is present" flag on the xprt and let an upper layer enforce local 419 * security policy. 420 */ 421 static void svc_tcp_handshake_done(void *data, int status, key_serial_t peerid) 422 { 423 struct svc_xprt *xprt = data; 424 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 425 426 if (!status) { 427 if (peerid != TLS_NO_PEERID) 428 set_bit(XPT_PEER_AUTH, &xprt->xpt_flags); 429 set_bit(XPT_TLS_SESSION, &xprt->xpt_flags); 430 } 431 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags); 432 complete_all(&svsk->sk_handshake_done); 433 } 434 435 /** 436 * svc_tcp_handshake - Perform a transport-layer security handshake 437 * @xprt: connected transport endpoint 438 * 439 */ 440 static void svc_tcp_handshake(struct svc_xprt *xprt) 441 { 442 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 443 struct sock *sk = svsk->sk_sock->sk; 444 struct tls_handshake_args args = { 445 .ta_sock = svsk->sk_sock, 446 .ta_done = svc_tcp_handshake_done, 447 .ta_data = xprt, 448 }; 449 int ret; 450 451 trace_svc_tls_upcall(xprt); 452 453 clear_bit(XPT_TLS_SESSION, &xprt->xpt_flags); 454 init_completion(&svsk->sk_handshake_done); 455 456 ret = tls_server_hello_x509(&args, GFP_KERNEL); 457 if (ret) { 458 trace_svc_tls_not_started(xprt); 459 goto out_failed; 460 } 461 462 ret = wait_for_completion_interruptible_timeout(&svsk->sk_handshake_done, 463 SVC_HANDSHAKE_TO); 464 if (ret <= 0) { 465 if (tls_handshake_cancel(sk)) { 466 trace_svc_tls_timed_out(xprt); 467 goto out_close; 468 } 469 } 470 471 if (!test_bit(XPT_TLS_SESSION, &xprt->xpt_flags)) { 472 trace_svc_tls_unavailable(xprt); 473 goto out_close; 474 } 475 476 /* Mark the transport ready in case the remote sent RPC 477 * traffic before the kernel received the handshake 478 * completion downcall. 479 */ 480 set_bit(XPT_DATA, &xprt->xpt_flags); 481 svc_xprt_enqueue(xprt); 482 return; 483 484 out_close: 485 set_bit(XPT_CLOSE, &xprt->xpt_flags); 486 out_failed: 487 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags); 488 set_bit(XPT_DATA, &xprt->xpt_flags); 489 svc_xprt_enqueue(xprt); 490 } 491 492 /* 493 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo 494 */ 495 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, 496 struct cmsghdr *cmh) 497 { 498 struct in_pktinfo *pki = CMSG_DATA(cmh); 499 struct sockaddr_in *daddr = svc_daddr_in(rqstp); 500 501 if (cmh->cmsg_type != IP_PKTINFO) 502 return 0; 503 504 daddr->sin_family = AF_INET; 505 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; 506 return 1; 507 } 508 509 /* 510 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl 511 */ 512 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, 513 struct cmsghdr *cmh) 514 { 515 struct in6_pktinfo *pki = CMSG_DATA(cmh); 516 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 517 518 if (cmh->cmsg_type != IPV6_PKTINFO) 519 return 0; 520 521 daddr->sin6_family = AF_INET6; 522 daddr->sin6_addr = pki->ipi6_addr; 523 daddr->sin6_scope_id = pki->ipi6_ifindex; 524 return 1; 525 } 526 527 /* 528 * Copy the UDP datagram's destination address to the rqstp structure. 529 * The 'destination' address in this case is the address to which the 530 * peer sent the datagram, i.e. our local address. For multihomed 531 * hosts, this can change from msg to msg. Note that only the IP 532 * address changes, the port number should remain the same. 533 */ 534 static int svc_udp_get_dest_address(struct svc_rqst *rqstp, 535 struct cmsghdr *cmh) 536 { 537 switch (cmh->cmsg_level) { 538 case SOL_IP: 539 return svc_udp_get_dest_address4(rqstp, cmh); 540 case SOL_IPV6: 541 return svc_udp_get_dest_address6(rqstp, cmh); 542 } 543 544 return 0; 545 } 546 547 /** 548 * svc_udp_recvfrom - Receive a datagram from a UDP socket. 549 * @rqstp: request structure into which to receive an RPC Call 550 * 551 * Called in a loop when XPT_DATA has been set. 552 * 553 * Returns: 554 * On success, the number of bytes in a received RPC Call, or 555 * %0 if a complete RPC Call message was not ready to return 556 */ 557 static int svc_udp_recvfrom(struct svc_rqst *rqstp) 558 { 559 struct svc_sock *svsk = 560 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 561 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 562 struct sk_buff *skb; 563 union { 564 struct cmsghdr hdr; 565 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 566 } buffer; 567 struct cmsghdr *cmh = &buffer.hdr; 568 struct msghdr msg = { 569 .msg_name = svc_addr(rqstp), 570 .msg_control = cmh, 571 .msg_controllen = sizeof(buffer), 572 .msg_flags = MSG_DONTWAIT, 573 }; 574 size_t len; 575 int err; 576 577 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 578 /* udp sockets need large rcvbuf as all pending 579 * requests are still in that buffer. sndbuf must 580 * also be large enough that there is enough space 581 * for one reply per thread. We count all threads 582 * rather than threads in a particular pool, which 583 * provides an upper bound on the number of threads 584 * which will access the socket. 585 */ 586 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3); 587 588 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 589 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 590 0, 0, MSG_PEEK | MSG_DONTWAIT); 591 if (err < 0) 592 goto out_recv_err; 593 skb = skb_recv_udp(svsk->sk_sk, MSG_DONTWAIT, &err); 594 if (!skb) 595 goto out_recv_err; 596 597 len = svc_addr_len(svc_addr(rqstp)); 598 rqstp->rq_addrlen = len; 599 if (skb->tstamp == 0) { 600 skb->tstamp = ktime_get_real(); 601 /* Don't enable netstamp, sunrpc doesn't 602 need that much accuracy */ 603 } 604 sock_write_timestamp(svsk->sk_sk, skb->tstamp); 605 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ 606 607 len = skb->len; 608 rqstp->rq_arg.len = len; 609 trace_svcsock_udp_recv(&svsk->sk_xprt, len); 610 611 rqstp->rq_prot = IPPROTO_UDP; 612 613 if (!svc_udp_get_dest_address(rqstp, cmh)) 614 goto out_cmsg_err; 615 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); 616 617 if (skb_is_nonlinear(skb)) { 618 /* we have to copy */ 619 local_bh_disable(); 620 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) 621 goto out_bh_enable; 622 local_bh_enable(); 623 consume_skb(skb); 624 } else { 625 /* we can use it in-place */ 626 rqstp->rq_arg.head[0].iov_base = skb->data; 627 rqstp->rq_arg.head[0].iov_len = len; 628 if (skb_checksum_complete(skb)) 629 goto out_free; 630 rqstp->rq_xprt_ctxt = skb; 631 } 632 633 rqstp->rq_arg.page_base = 0; 634 if (len <= rqstp->rq_arg.head[0].iov_len) { 635 rqstp->rq_arg.head[0].iov_len = len; 636 rqstp->rq_arg.page_len = 0; 637 rqstp->rq_respages = rqstp->rq_pages+1; 638 } else { 639 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 640 rqstp->rq_respages = rqstp->rq_pages + 1 + 641 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); 642 } 643 rqstp->rq_next_page = rqstp->rq_respages+1; 644 645 if (serv->sv_stats) 646 serv->sv_stats->netudpcnt++; 647 648 svc_sock_secure_port(rqstp); 649 svc_xprt_received(rqstp->rq_xprt); 650 return len; 651 652 out_recv_err: 653 if (err != -EAGAIN) { 654 /* possibly an icmp error */ 655 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 656 } 657 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err); 658 goto out_clear_busy; 659 out_cmsg_err: 660 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", 661 cmh->cmsg_level, cmh->cmsg_type); 662 goto out_free; 663 out_bh_enable: 664 local_bh_enable(); 665 out_free: 666 kfree_skb(skb); 667 out_clear_busy: 668 svc_xprt_received(rqstp->rq_xprt); 669 return 0; 670 } 671 672 /** 673 * svc_udp_sendto - Send out a reply on a UDP socket 674 * @rqstp: completed svc_rqst 675 * 676 * xpt_mutex ensures @rqstp's whole message is written to the socket 677 * without interruption. 678 * 679 * Returns the number of bytes sent, or a negative errno. 680 */ 681 static int svc_udp_sendto(struct svc_rqst *rqstp) 682 { 683 struct svc_xprt *xprt = rqstp->rq_xprt; 684 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 685 struct xdr_buf *xdr = &rqstp->rq_res; 686 union { 687 struct cmsghdr hdr; 688 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 689 } buffer; 690 struct cmsghdr *cmh = &buffer.hdr; 691 struct msghdr msg = { 692 .msg_name = &rqstp->rq_addr, 693 .msg_namelen = rqstp->rq_addrlen, 694 .msg_control = cmh, 695 .msg_controllen = sizeof(buffer), 696 }; 697 unsigned int sent; 698 int err; 699 700 svc_udp_release_ctxt(xprt, rqstp->rq_xprt_ctxt); 701 rqstp->rq_xprt_ctxt = NULL; 702 703 svc_set_cmsg_data(rqstp, cmh); 704 705 mutex_lock(&xprt->xpt_mutex); 706 707 if (svc_xprt_is_dead(xprt)) 708 goto out_notconn; 709 710 err = xdr_alloc_bvec(xdr, GFP_KERNEL); 711 if (err < 0) 712 goto out_unlock; 713 714 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent); 715 if (err == -ECONNREFUSED) { 716 /* ICMP error on earlier request. */ 717 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent); 718 } 719 xdr_free_bvec(xdr); 720 trace_svcsock_udp_send(xprt, err); 721 out_unlock: 722 mutex_unlock(&xprt->xpt_mutex); 723 if (err < 0) 724 return err; 725 return sent; 726 727 out_notconn: 728 mutex_unlock(&xprt->xpt_mutex); 729 return -ENOTCONN; 730 } 731 732 static int svc_udp_has_wspace(struct svc_xprt *xprt) 733 { 734 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 735 struct svc_serv *serv = xprt->xpt_server; 736 unsigned long required; 737 738 /* 739 * Set the SOCK_NOSPACE flag before checking the available 740 * sock space. 741 */ 742 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 743 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; 744 if (required*2 > sock_wspace(svsk->sk_sk)) 745 return 0; 746 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 747 return 1; 748 } 749 750 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) 751 { 752 BUG(); 753 return NULL; 754 } 755 756 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt) 757 { 758 } 759 760 static struct svc_xprt *svc_udp_create(struct svc_serv *serv, 761 struct net *net, 762 struct sockaddr *sa, int salen, 763 int flags) 764 { 765 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); 766 } 767 768 static const struct svc_xprt_ops svc_udp_ops = { 769 .xpo_create = svc_udp_create, 770 .xpo_recvfrom = svc_udp_recvfrom, 771 .xpo_sendto = svc_udp_sendto, 772 .xpo_result_payload = svc_sock_result_payload, 773 .xpo_release_ctxt = svc_udp_release_ctxt, 774 .xpo_detach = svc_sock_detach, 775 .xpo_free = svc_sock_free, 776 .xpo_has_wspace = svc_udp_has_wspace, 777 .xpo_accept = svc_udp_accept, 778 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt, 779 }; 780 781 static struct svc_xprt_class svc_udp_class = { 782 .xcl_name = "udp", 783 .xcl_owner = THIS_MODULE, 784 .xcl_ops = &svc_udp_ops, 785 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, 786 .xcl_ident = XPRT_TRANSPORT_UDP, 787 }; 788 789 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) 790 { 791 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, 792 &svsk->sk_xprt, serv); 793 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 794 svsk->sk_sk->sk_data_ready = svc_data_ready; 795 svsk->sk_sk->sk_write_space = svc_write_space; 796 797 /* initialise setting must have enough space to 798 * receive and respond to one request. 799 * svc_udp_recvfrom will re-adjust if necessary 800 */ 801 svc_sock_setbufsize(svsk, 3); 802 803 /* data might have come in before data_ready set up */ 804 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 805 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 806 807 /* make sure we get destination address info */ 808 switch (svsk->sk_sk->sk_family) { 809 case AF_INET: 810 ip_sock_set_pktinfo(svsk->sk_sock->sk); 811 break; 812 case AF_INET6: 813 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk); 814 break; 815 default: 816 BUG(); 817 } 818 } 819 820 /* 821 * A data_ready event on a listening socket means there's a connection 822 * pending. Do not use state_change as a substitute for it. 823 */ 824 static void svc_tcp_listen_data_ready(struct sock *sk) 825 { 826 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 827 828 trace_sk_data_ready(sk); 829 830 /* 831 * This callback may called twice when a new connection 832 * is established as a child socket inherits everything 833 * from a parent LISTEN socket. 834 * 1) data_ready method of the parent socket will be called 835 * when one of child sockets become ESTABLISHED. 836 * 2) data_ready method of the child socket may be called 837 * when it receives data before the socket is accepted. 838 * In case of 2, we should ignore it silently and DO NOT 839 * dereference svsk. 840 */ 841 if (sk->sk_state != TCP_LISTEN) 842 return; 843 844 if (svsk) { 845 /* Refer to svc_setup_socket() for details. */ 846 rmb(); 847 svsk->sk_odata(sk); 848 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 849 svc_xprt_enqueue(&svsk->sk_xprt); 850 } 851 } 852 853 /* 854 * A state change on a connected socket means it's dying or dead. 855 */ 856 static void svc_tcp_state_change(struct sock *sk) 857 { 858 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 859 860 if (svsk) { 861 /* Refer to svc_setup_socket() for details. */ 862 rmb(); 863 svsk->sk_ostate(sk); 864 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock); 865 if (sk->sk_state != TCP_ESTABLISHED) 866 svc_xprt_deferred_close(&svsk->sk_xprt); 867 } 868 } 869 870 /* 871 * Accept a TCP connection 872 */ 873 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) 874 { 875 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 876 struct sockaddr_storage addr; 877 struct sockaddr *sin = (struct sockaddr *) &addr; 878 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 879 struct socket *sock = svsk->sk_sock; 880 struct socket *newsock; 881 struct svc_sock *newsvsk; 882 int err, slen; 883 884 if (!sock) 885 return NULL; 886 887 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 888 err = kernel_accept(sock, &newsock, O_NONBLOCK); 889 if (err < 0) { 890 if (err != -EAGAIN) 891 trace_svcsock_accept_err(xprt, serv->sv_name, err); 892 return NULL; 893 } 894 if (IS_ERR(sock_alloc_file(newsock, O_NONBLOCK, NULL))) 895 return NULL; 896 897 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 898 899 err = kernel_getpeername(newsock, sin); 900 if (err < 0) { 901 trace_svcsock_getpeername_err(xprt, serv->sv_name, err); 902 goto failed; /* aborted connection or whatever */ 903 } 904 slen = err; 905 906 /* Reset the inherited callbacks before calling svc_setup_socket */ 907 newsock->sk->sk_state_change = svsk->sk_ostate; 908 newsock->sk->sk_data_ready = svsk->sk_odata; 909 newsock->sk->sk_write_space = svsk->sk_owspace; 910 911 /* make sure that a write doesn't block forever when 912 * low on memory 913 */ 914 newsock->sk->sk_sndtimeo = HZ*30; 915 916 newsvsk = svc_setup_socket(serv, newsock, 917 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); 918 if (IS_ERR(newsvsk)) 919 goto failed; 920 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); 921 err = kernel_getsockname(newsock, sin); 922 slen = err; 923 if (unlikely(err < 0)) 924 slen = offsetof(struct sockaddr, sa_data); 925 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); 926 927 if (sock_is_loopback(newsock->sk)) 928 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 929 else 930 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 931 if (serv->sv_stats) 932 serv->sv_stats->nettcpconn++; 933 934 return &newsvsk->sk_xprt; 935 936 failed: 937 sockfd_put(newsock); 938 return NULL; 939 } 940 941 static size_t svc_tcp_restore_pages(struct svc_sock *svsk, 942 struct svc_rqst *rqstp) 943 { 944 size_t len = svsk->sk_datalen; 945 unsigned int i, npages; 946 947 if (!len) 948 return 0; 949 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 950 for (i = 0; i < npages; i++) { 951 if (rqstp->rq_pages[i] != NULL) 952 put_page(rqstp->rq_pages[i]); 953 BUG_ON(svsk->sk_pages[i] == NULL); 954 rqstp->rq_pages[i] = svsk->sk_pages[i]; 955 svsk->sk_pages[i] = NULL; 956 } 957 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); 958 return len; 959 } 960 961 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 962 { 963 unsigned int i, len, npages; 964 965 if (svsk->sk_datalen == 0) 966 return; 967 len = svsk->sk_datalen; 968 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 969 for (i = 0; i < npages; i++) { 970 svsk->sk_pages[i] = rqstp->rq_pages[i]; 971 rqstp->rq_pages[i] = NULL; 972 } 973 } 974 975 static void svc_tcp_clear_pages(struct svc_sock *svsk) 976 { 977 unsigned int i, len, npages; 978 979 if (svsk->sk_datalen == 0) 980 goto out; 981 len = svsk->sk_datalen; 982 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 983 for (i = 0; i < npages; i++) { 984 if (svsk->sk_pages[i] == NULL) { 985 WARN_ON_ONCE(1); 986 continue; 987 } 988 put_page(svsk->sk_pages[i]); 989 svsk->sk_pages[i] = NULL; 990 } 991 out: 992 svsk->sk_tcplen = 0; 993 svsk->sk_datalen = 0; 994 } 995 996 /* 997 * Receive fragment record header into sk_marker. 998 */ 999 static ssize_t svc_tcp_read_marker(struct svc_sock *svsk, 1000 struct svc_rqst *rqstp) 1001 { 1002 ssize_t want, len; 1003 1004 /* If we haven't gotten the record length yet, 1005 * get the next four bytes. 1006 */ 1007 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { 1008 struct msghdr msg = { NULL }; 1009 struct kvec iov; 1010 1011 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; 1012 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen; 1013 iov.iov_len = want; 1014 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want); 1015 len = svc_tcp_sock_recv_cmsg(svsk, &msg); 1016 if (len < 0) 1017 return len; 1018 svsk->sk_tcplen += len; 1019 if (len < want) { 1020 /* call again to read the remaining bytes */ 1021 goto err_short; 1022 } 1023 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker); 1024 if (svc_sock_reclen(svsk) + svsk->sk_datalen > 1025 svsk->sk_xprt.xpt_server->sv_max_mesg) 1026 goto err_too_large; 1027 } 1028 return svc_sock_reclen(svsk); 1029 1030 err_too_large: 1031 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n", 1032 __func__, svsk->sk_xprt.xpt_server->sv_name, 1033 svc_sock_reclen(svsk)); 1034 svc_xprt_deferred_close(&svsk->sk_xprt); 1035 err_short: 1036 return -EAGAIN; 1037 } 1038 1039 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) 1040 { 1041 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; 1042 struct rpc_rqst *req = NULL; 1043 struct kvec *src, *dst; 1044 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1045 __be32 xid; 1046 __be32 calldir; 1047 1048 xid = *p++; 1049 calldir = *p; 1050 1051 if (!bc_xprt) 1052 return -EAGAIN; 1053 spin_lock(&bc_xprt->queue_lock); 1054 req = xprt_lookup_rqst(bc_xprt, xid); 1055 if (!req) 1056 goto unlock_notfound; 1057 1058 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); 1059 /* 1060 * XXX!: cheating for now! Only copying HEAD. 1061 * But we know this is good enough for now (in fact, for any 1062 * callback reply in the forseeable future). 1063 */ 1064 dst = &req->rq_private_buf.head[0]; 1065 src = &rqstp->rq_arg.head[0]; 1066 if (dst->iov_len < src->iov_len) 1067 goto unlock_eagain; /* whatever; just giving up. */ 1068 memcpy(dst->iov_base, src->iov_base, src->iov_len); 1069 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); 1070 rqstp->rq_arg.len = 0; 1071 spin_unlock(&bc_xprt->queue_lock); 1072 return 0; 1073 unlock_notfound: 1074 printk(KERN_NOTICE 1075 "%s: Got unrecognized reply: " 1076 "calldir 0x%x xpt_bc_xprt %p xid %08x\n", 1077 __func__, ntohl(calldir), 1078 bc_xprt, ntohl(xid)); 1079 unlock_eagain: 1080 spin_unlock(&bc_xprt->queue_lock); 1081 return -EAGAIN; 1082 } 1083 1084 static void svc_tcp_fragment_received(struct svc_sock *svsk) 1085 { 1086 /* If we have more data, signal svc_xprt_enqueue() to try again */ 1087 svsk->sk_tcplen = 0; 1088 svsk->sk_marker = xdr_zero; 1089 } 1090 1091 /** 1092 * svc_tcp_recvfrom - Receive data from a TCP socket 1093 * @rqstp: request structure into which to receive an RPC Call 1094 * 1095 * Called in a loop when XPT_DATA has been set. 1096 * 1097 * Read the 4-byte stream record marker, then use the record length 1098 * in that marker to set up exactly the resources needed to receive 1099 * the next RPC message into @rqstp. 1100 * 1101 * Returns: 1102 * On success, the number of bytes in a received RPC Call, or 1103 * %0 if a complete RPC Call message was not ready to return 1104 * 1105 * The zero return case handles partial receives and callback Replies. 1106 * The state of a partial receive is preserved in the svc_sock for 1107 * the next call to svc_tcp_recvfrom. 1108 */ 1109 static int svc_tcp_recvfrom(struct svc_rqst *rqstp) 1110 { 1111 struct svc_sock *svsk = 1112 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 1113 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1114 size_t want, base; 1115 ssize_t len; 1116 __be32 *p; 1117 __be32 calldir; 1118 1119 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1120 len = svc_tcp_read_marker(svsk, rqstp); 1121 if (len < 0) 1122 goto error; 1123 1124 base = svc_tcp_restore_pages(svsk, rqstp); 1125 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); 1126 len = svc_tcp_read_msg(rqstp, base + want, base); 1127 if (len >= 0) { 1128 trace_svcsock_tcp_recv(&svsk->sk_xprt, len); 1129 svsk->sk_tcplen += len; 1130 svsk->sk_datalen += len; 1131 } 1132 if (len != want || !svc_sock_final_rec(svsk)) 1133 goto err_incomplete; 1134 if (svsk->sk_datalen < 8) 1135 goto err_nuts; 1136 1137 rqstp->rq_arg.len = svsk->sk_datalen; 1138 rqstp->rq_arg.page_base = 0; 1139 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { 1140 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; 1141 rqstp->rq_arg.page_len = 0; 1142 } else 1143 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1144 1145 rqstp->rq_xprt_ctxt = NULL; 1146 rqstp->rq_prot = IPPROTO_TCP; 1147 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags)) 1148 set_bit(RQ_LOCAL, &rqstp->rq_flags); 1149 else 1150 clear_bit(RQ_LOCAL, &rqstp->rq_flags); 1151 1152 p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1153 calldir = p[1]; 1154 if (calldir) 1155 len = receive_cb_reply(svsk, rqstp); 1156 1157 /* Reset TCP read info */ 1158 svsk->sk_datalen = 0; 1159 svc_tcp_fragment_received(svsk); 1160 1161 if (len < 0) 1162 goto error; 1163 1164 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); 1165 if (serv->sv_stats) 1166 serv->sv_stats->nettcpcnt++; 1167 1168 svc_sock_secure_port(rqstp); 1169 svc_xprt_received(rqstp->rq_xprt); 1170 return rqstp->rq_arg.len; 1171 1172 err_incomplete: 1173 svc_tcp_save_pages(svsk, rqstp); 1174 if (len < 0 && len != -EAGAIN) 1175 goto err_delete; 1176 if (len == want) 1177 svc_tcp_fragment_received(svsk); 1178 else 1179 trace_svcsock_tcp_recv_short(&svsk->sk_xprt, 1180 svc_sock_reclen(svsk), 1181 svsk->sk_tcplen - sizeof(rpc_fraghdr)); 1182 goto err_noclose; 1183 error: 1184 if (len != -EAGAIN) 1185 goto err_delete; 1186 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0); 1187 goto err_noclose; 1188 err_nuts: 1189 svsk->sk_datalen = 0; 1190 err_delete: 1191 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len); 1192 svc_xprt_deferred_close(&svsk->sk_xprt); 1193 err_noclose: 1194 svc_xprt_received(rqstp->rq_xprt); 1195 return 0; /* record not complete */ 1196 } 1197 1198 static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec, 1199 int flags) 1200 { 1201 struct msghdr msg = { .msg_flags = MSG_SPLICE_PAGES | flags, }; 1202 1203 iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, vec, 1, vec->iov_len); 1204 return sock_sendmsg(sock, &msg); 1205 } 1206 1207 /* 1208 * MSG_SPLICE_PAGES is used exclusively to reduce the number of 1209 * copy operations in this path. Therefore the caller must ensure 1210 * that the pages backing @xdr are unchanging. 1211 * 1212 * In addition, the logic assumes that * .bv_len is never larger 1213 * than PAGE_SIZE. 1214 */ 1215 static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr, 1216 rpc_fraghdr marker, unsigned int *sentp) 1217 { 1218 const struct kvec *head = xdr->head; 1219 const struct kvec *tail = xdr->tail; 1220 struct kvec rm = { 1221 .iov_base = &marker, 1222 .iov_len = sizeof(marker), 1223 }; 1224 struct msghdr msg = { 1225 .msg_flags = 0, 1226 }; 1227 int ret; 1228 1229 *sentp = 0; 1230 ret = xdr_alloc_bvec(xdr, GFP_KERNEL); 1231 if (ret < 0) 1232 return ret; 1233 1234 ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len); 1235 if (ret < 0) 1236 return ret; 1237 *sentp += ret; 1238 if (ret != rm.iov_len) 1239 return -EAGAIN; 1240 1241 ret = svc_tcp_send_kvec(sock, head, 0); 1242 if (ret < 0) 1243 return ret; 1244 *sentp += ret; 1245 if (ret != head->iov_len) 1246 goto out; 1247 1248 msg.msg_flags = MSG_SPLICE_PAGES; 1249 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, xdr->bvec, 1250 xdr_buf_pagecount(xdr), xdr->page_len); 1251 ret = sock_sendmsg(sock, &msg); 1252 if (ret < 0) 1253 return ret; 1254 *sentp += ret; 1255 1256 if (tail->iov_len) { 1257 ret = svc_tcp_send_kvec(sock, tail, 0); 1258 if (ret < 0) 1259 return ret; 1260 *sentp += ret; 1261 } 1262 1263 out: 1264 return 0; 1265 } 1266 1267 /** 1268 * svc_tcp_sendto - Send out a reply on a TCP socket 1269 * @rqstp: completed svc_rqst 1270 * 1271 * xpt_mutex ensures @rqstp's whole message is written to the socket 1272 * without interruption. 1273 * 1274 * Returns the number of bytes sent, or a negative errno. 1275 */ 1276 static int svc_tcp_sendto(struct svc_rqst *rqstp) 1277 { 1278 struct svc_xprt *xprt = rqstp->rq_xprt; 1279 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1280 struct xdr_buf *xdr = &rqstp->rq_res; 1281 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | 1282 (u32)xdr->len); 1283 unsigned int sent; 1284 int err; 1285 1286 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt); 1287 rqstp->rq_xprt_ctxt = NULL; 1288 1289 atomic_inc(&svsk->sk_sendqlen); 1290 mutex_lock(&xprt->xpt_mutex); 1291 if (svc_xprt_is_dead(xprt)) 1292 goto out_notconn; 1293 tcp_sock_set_cork(svsk->sk_sk, true); 1294 err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent); 1295 xdr_free_bvec(xdr); 1296 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent); 1297 if (err < 0 || sent != (xdr->len + sizeof(marker))) 1298 goto out_close; 1299 if (atomic_dec_and_test(&svsk->sk_sendqlen)) 1300 tcp_sock_set_cork(svsk->sk_sk, false); 1301 mutex_unlock(&xprt->xpt_mutex); 1302 return sent; 1303 1304 out_notconn: 1305 atomic_dec(&svsk->sk_sendqlen); 1306 mutex_unlock(&xprt->xpt_mutex); 1307 return -ENOTCONN; 1308 out_close: 1309 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", 1310 xprt->xpt_server->sv_name, 1311 (err < 0) ? "got error" : "sent", 1312 (err < 0) ? err : sent, xdr->len); 1313 svc_xprt_deferred_close(xprt); 1314 atomic_dec(&svsk->sk_sendqlen); 1315 mutex_unlock(&xprt->xpt_mutex); 1316 return -EAGAIN; 1317 } 1318 1319 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, 1320 struct net *net, 1321 struct sockaddr *sa, int salen, 1322 int flags) 1323 { 1324 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1325 } 1326 1327 static const struct svc_xprt_ops svc_tcp_ops = { 1328 .xpo_create = svc_tcp_create, 1329 .xpo_recvfrom = svc_tcp_recvfrom, 1330 .xpo_sendto = svc_tcp_sendto, 1331 .xpo_result_payload = svc_sock_result_payload, 1332 .xpo_release_ctxt = svc_tcp_release_ctxt, 1333 .xpo_detach = svc_tcp_sock_detach, 1334 .xpo_free = svc_sock_free, 1335 .xpo_has_wspace = svc_tcp_has_wspace, 1336 .xpo_accept = svc_tcp_accept, 1337 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt, 1338 .xpo_handshake = svc_tcp_handshake, 1339 }; 1340 1341 static struct svc_xprt_class svc_tcp_class = { 1342 .xcl_name = "tcp", 1343 .xcl_owner = THIS_MODULE, 1344 .xcl_ops = &svc_tcp_ops, 1345 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1346 .xcl_ident = XPRT_TRANSPORT_TCP, 1347 }; 1348 1349 void svc_init_xprt_sock(void) 1350 { 1351 svc_reg_xprt_class(&svc_tcp_class); 1352 svc_reg_xprt_class(&svc_udp_class); 1353 } 1354 1355 void svc_cleanup_xprt_sock(void) 1356 { 1357 svc_unreg_xprt_class(&svc_tcp_class); 1358 svc_unreg_xprt_class(&svc_udp_class); 1359 } 1360 1361 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) 1362 { 1363 struct sock *sk = svsk->sk_sk; 1364 1365 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, 1366 &svsk->sk_xprt, serv); 1367 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 1368 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags); 1369 if (sk->sk_state == TCP_LISTEN) { 1370 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener"); 1371 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); 1372 sk->sk_data_ready = svc_tcp_listen_data_ready; 1373 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 1374 } else { 1375 sk->sk_state_change = svc_tcp_state_change; 1376 sk->sk_data_ready = svc_data_ready; 1377 sk->sk_write_space = svc_write_space; 1378 1379 svsk->sk_marker = xdr_zero; 1380 svsk->sk_tcplen = 0; 1381 svsk->sk_datalen = 0; 1382 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); 1383 1384 tcp_sock_set_nodelay(sk); 1385 1386 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1387 switch (sk->sk_state) { 1388 case TCP_SYN_RECV: 1389 case TCP_ESTABLISHED: 1390 break; 1391 default: 1392 svc_xprt_deferred_close(&svsk->sk_xprt); 1393 } 1394 } 1395 } 1396 1397 void svc_sock_update_bufs(struct svc_serv *serv) 1398 { 1399 /* 1400 * The number of server threads has changed. Update 1401 * rcvbuf and sndbuf accordingly on all sockets 1402 */ 1403 struct svc_sock *svsk; 1404 1405 spin_lock_bh(&serv->sv_lock); 1406 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) 1407 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1408 spin_unlock_bh(&serv->sv_lock); 1409 } 1410 EXPORT_SYMBOL_GPL(svc_sock_update_bufs); 1411 1412 /* 1413 * Initialize socket for RPC use and create svc_sock struct 1414 */ 1415 static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1416 struct socket *sock, 1417 int flags) 1418 { 1419 struct svc_sock *svsk; 1420 struct sock *inet; 1421 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1422 1423 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1424 if (!svsk) 1425 return ERR_PTR(-ENOMEM); 1426 1427 inet = sock->sk; 1428 1429 if (pmap_register) { 1430 int err; 1431 1432 err = svc_register(serv, sock_net(sock->sk), inet->sk_family, 1433 inet->sk_protocol, 1434 ntohs(inet_sk(inet)->inet_sport)); 1435 if (err < 0) { 1436 kfree(svsk); 1437 return ERR_PTR(err); 1438 } 1439 } 1440 1441 svsk->sk_sock = sock; 1442 svsk->sk_sk = inet; 1443 svsk->sk_ostate = inet->sk_state_change; 1444 svsk->sk_odata = inet->sk_data_ready; 1445 svsk->sk_owspace = inet->sk_write_space; 1446 /* 1447 * This barrier is necessary in order to prevent race condition 1448 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others 1449 * when calling callbacks above. 1450 */ 1451 wmb(); 1452 inet->sk_user_data = svsk; 1453 1454 /* Initialize the socket */ 1455 if (sock->type == SOCK_DGRAM) 1456 svc_udp_init(svsk, serv); 1457 else 1458 svc_tcp_init(svsk, serv); 1459 1460 trace_svcsock_new(svsk, sock); 1461 return svsk; 1462 } 1463 1464 /** 1465 * svc_addsock - add a listener socket to an RPC service 1466 * @serv: pointer to RPC service to which to add a new listener 1467 * @net: caller's network namespace 1468 * @fd: file descriptor of the new listener 1469 * @name_return: pointer to buffer to fill in with name of listener 1470 * @len: size of the buffer 1471 * @cred: credential 1472 * 1473 * Fills in socket name and returns positive length of name if successful. 1474 * Name is terminated with '\n'. On error, returns a negative errno 1475 * value. 1476 */ 1477 int svc_addsock(struct svc_serv *serv, struct net *net, const int fd, 1478 char *name_return, const size_t len, const struct cred *cred) 1479 { 1480 int err = 0; 1481 struct socket *so = sockfd_lookup(fd, &err); 1482 struct svc_sock *svsk = NULL; 1483 struct sockaddr_storage addr; 1484 struct sockaddr *sin = (struct sockaddr *)&addr; 1485 int salen; 1486 1487 if (!so) 1488 return err; 1489 err = -EINVAL; 1490 if (sock_net(so->sk) != net) 1491 goto out; 1492 err = -EAFNOSUPPORT; 1493 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) 1494 goto out; 1495 err = -EPROTONOSUPPORT; 1496 if (so->sk->sk_protocol != IPPROTO_TCP && 1497 so->sk->sk_protocol != IPPROTO_UDP) 1498 goto out; 1499 err = -EISCONN; 1500 if (so->state > SS_UNCONNECTED) 1501 goto out; 1502 err = -ENOENT; 1503 if (!try_module_get(THIS_MODULE)) 1504 goto out; 1505 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); 1506 if (IS_ERR(svsk)) { 1507 module_put(THIS_MODULE); 1508 err = PTR_ERR(svsk); 1509 goto out; 1510 } 1511 salen = kernel_getsockname(svsk->sk_sock, sin); 1512 if (salen >= 0) 1513 svc_xprt_set_local(&svsk->sk_xprt, sin, salen); 1514 svsk->sk_xprt.xpt_cred = get_cred(cred); 1515 svc_add_new_perm_xprt(serv, &svsk->sk_xprt); 1516 return svc_one_sock_name(svsk, name_return, len); 1517 out: 1518 sockfd_put(so); 1519 return err; 1520 } 1521 EXPORT_SYMBOL_GPL(svc_addsock); 1522 1523 /* 1524 * Create socket for RPC service. 1525 */ 1526 static struct svc_xprt *svc_create_socket(struct svc_serv *serv, 1527 int protocol, 1528 struct net *net, 1529 struct sockaddr *sin, int len, 1530 int flags) 1531 { 1532 struct svc_sock *svsk; 1533 struct socket *sock; 1534 int error; 1535 int type; 1536 struct sockaddr_storage addr; 1537 struct sockaddr *newsin = (struct sockaddr *)&addr; 1538 int newlen; 1539 int family; 1540 1541 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1542 printk(KERN_WARNING "svc: only UDP and TCP " 1543 "sockets supported\n"); 1544 return ERR_PTR(-EINVAL); 1545 } 1546 1547 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1548 switch (sin->sa_family) { 1549 case AF_INET6: 1550 family = PF_INET6; 1551 break; 1552 case AF_INET: 1553 family = PF_INET; 1554 break; 1555 default: 1556 return ERR_PTR(-EINVAL); 1557 } 1558 1559 error = __sock_create(net, family, type, protocol, &sock, 1); 1560 if (error < 0) 1561 return ERR_PTR(error); 1562 1563 svc_reclassify_socket(sock); 1564 1565 /* 1566 * If this is an PF_INET6 listener, we want to avoid 1567 * getting requests from IPv4 remotes. Those should 1568 * be shunted to a PF_INET listener via rpcbind. 1569 */ 1570 if (family == PF_INET6) 1571 ip6_sock_set_v6only(sock->sk); 1572 if (type == SOCK_STREAM) 1573 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ 1574 error = kernel_bind(sock, sin, len); 1575 if (error < 0) 1576 goto bummer; 1577 1578 error = kernel_getsockname(sock, newsin); 1579 if (error < 0) 1580 goto bummer; 1581 newlen = error; 1582 1583 if (protocol == IPPROTO_TCP) { 1584 if ((error = kernel_listen(sock, 64)) < 0) 1585 goto bummer; 1586 } 1587 1588 svsk = svc_setup_socket(serv, sock, flags); 1589 if (IS_ERR(svsk)) { 1590 error = PTR_ERR(svsk); 1591 goto bummer; 1592 } 1593 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); 1594 return (struct svc_xprt *)svsk; 1595 bummer: 1596 sock_release(sock); 1597 return ERR_PTR(error); 1598 } 1599 1600 /* 1601 * Detach the svc_sock from the socket so that no 1602 * more callbacks occur. 1603 */ 1604 static void svc_sock_detach(struct svc_xprt *xprt) 1605 { 1606 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1607 struct sock *sk = svsk->sk_sk; 1608 1609 /* put back the old socket callbacks */ 1610 lock_sock(sk); 1611 sk->sk_state_change = svsk->sk_ostate; 1612 sk->sk_data_ready = svsk->sk_odata; 1613 sk->sk_write_space = svsk->sk_owspace; 1614 sk->sk_user_data = NULL; 1615 release_sock(sk); 1616 } 1617 1618 /* 1619 * Disconnect the socket, and reset the callbacks 1620 */ 1621 static void svc_tcp_sock_detach(struct svc_xprt *xprt) 1622 { 1623 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1624 1625 svc_sock_detach(xprt); 1626 1627 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { 1628 svc_tcp_clear_pages(svsk); 1629 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); 1630 } 1631 } 1632 1633 /* 1634 * Free the svc_sock's socket resources and the svc_sock itself. 1635 */ 1636 static void svc_sock_free(struct svc_xprt *xprt) 1637 { 1638 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1639 struct socket *sock = svsk->sk_sock; 1640 1641 trace_svcsock_free(svsk, sock); 1642 1643 tls_handshake_cancel(sock->sk); 1644 if (sock->file) 1645 sockfd_put(sock); 1646 else 1647 sock_release(sock); 1648 kfree(svsk); 1649 } 1650