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/module.h> 25 #include <linux/errno.h> 26 #include <linux/fcntl.h> 27 #include <linux/net.h> 28 #include <linux/in.h> 29 #include <linux/inet.h> 30 #include <linux/udp.h> 31 #include <linux/tcp.h> 32 #include <linux/unistd.h> 33 #include <linux/slab.h> 34 #include <linux/netdevice.h> 35 #include <linux/skbuff.h> 36 #include <linux/file.h> 37 #include <linux/freezer.h> 38 #include <net/sock.h> 39 #include <net/checksum.h> 40 #include <net/ip.h> 41 #include <net/ipv6.h> 42 #include <net/udp.h> 43 #include <net/tcp.h> 44 #include <net/tcp_states.h> 45 #include <linux/uaccess.h> 46 #include <asm/ioctls.h> 47 #include <trace/events/skb.h> 48 49 #include <linux/sunrpc/types.h> 50 #include <linux/sunrpc/clnt.h> 51 #include <linux/sunrpc/xdr.h> 52 #include <linux/sunrpc/msg_prot.h> 53 #include <linux/sunrpc/svcsock.h> 54 #include <linux/sunrpc/stats.h> 55 #include <linux/sunrpc/xprt.h> 56 57 #include "sunrpc.h" 58 59 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 60 61 62 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, 63 int flags); 64 static int svc_udp_recvfrom(struct svc_rqst *); 65 static int svc_udp_sendto(struct svc_rqst *); 66 static void svc_sock_detach(struct svc_xprt *); 67 static void svc_tcp_sock_detach(struct svc_xprt *); 68 static void svc_sock_free(struct svc_xprt *); 69 70 static struct svc_xprt *svc_create_socket(struct svc_serv *, int, 71 struct net *, struct sockaddr *, 72 int, int); 73 #ifdef CONFIG_DEBUG_LOCK_ALLOC 74 static struct lock_class_key svc_key[2]; 75 static struct lock_class_key svc_slock_key[2]; 76 77 static void svc_reclassify_socket(struct socket *sock) 78 { 79 struct sock *sk = sock->sk; 80 81 if (WARN_ON_ONCE(!sock_allow_reclassification(sk))) 82 return; 83 84 switch (sk->sk_family) { 85 case AF_INET: 86 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", 87 &svc_slock_key[0], 88 "sk_xprt.xpt_lock-AF_INET-NFSD", 89 &svc_key[0]); 90 break; 91 92 case AF_INET6: 93 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", 94 &svc_slock_key[1], 95 "sk_xprt.xpt_lock-AF_INET6-NFSD", 96 &svc_key[1]); 97 break; 98 99 default: 100 BUG(); 101 } 102 } 103 #else 104 static void svc_reclassify_socket(struct socket *sock) 105 { 106 } 107 #endif 108 109 /* 110 * Release an skbuff after use 111 */ 112 static void svc_release_skb(struct svc_rqst *rqstp) 113 { 114 struct sk_buff *skb = rqstp->rq_xprt_ctxt; 115 116 if (skb) { 117 struct svc_sock *svsk = 118 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 119 rqstp->rq_xprt_ctxt = NULL; 120 121 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); 122 skb_free_datagram_locked(svsk->sk_sk, skb); 123 } 124 } 125 126 static void svc_release_udp_skb(struct svc_rqst *rqstp) 127 { 128 struct sk_buff *skb = rqstp->rq_xprt_ctxt; 129 130 if (skb) { 131 rqstp->rq_xprt_ctxt = NULL; 132 133 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); 134 consume_skb(skb); 135 } 136 } 137 138 union svc_pktinfo_u { 139 struct in_pktinfo pkti; 140 struct in6_pktinfo pkti6; 141 }; 142 #define SVC_PKTINFO_SPACE \ 143 CMSG_SPACE(sizeof(union svc_pktinfo_u)) 144 145 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) 146 { 147 struct svc_sock *svsk = 148 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 149 switch (svsk->sk_sk->sk_family) { 150 case AF_INET: { 151 struct in_pktinfo *pki = CMSG_DATA(cmh); 152 153 cmh->cmsg_level = SOL_IP; 154 cmh->cmsg_type = IP_PKTINFO; 155 pki->ipi_ifindex = 0; 156 pki->ipi_spec_dst.s_addr = 157 svc_daddr_in(rqstp)->sin_addr.s_addr; 158 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 159 } 160 break; 161 162 case AF_INET6: { 163 struct in6_pktinfo *pki = CMSG_DATA(cmh); 164 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 165 166 cmh->cmsg_level = SOL_IPV6; 167 cmh->cmsg_type = IPV6_PKTINFO; 168 pki->ipi6_ifindex = daddr->sin6_scope_id; 169 pki->ipi6_addr = daddr->sin6_addr; 170 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 171 } 172 break; 173 } 174 } 175 176 /* 177 * send routine intended to be shared by the fore- and back-channel 178 */ 179 int svc_send_common(struct socket *sock, struct xdr_buf *xdr, 180 struct page *headpage, unsigned long headoffset, 181 struct page *tailpage, unsigned long tailoffset) 182 { 183 int result; 184 int size; 185 struct page **ppage = xdr->pages; 186 size_t base = xdr->page_base; 187 unsigned int pglen = xdr->page_len; 188 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST; 189 int slen; 190 int len = 0; 191 192 slen = xdr->len; 193 194 /* send head */ 195 if (slen == xdr->head[0].iov_len) 196 flags = 0; 197 len = kernel_sendpage(sock, headpage, headoffset, 198 xdr->head[0].iov_len, flags); 199 if (len != xdr->head[0].iov_len) 200 goto out; 201 slen -= xdr->head[0].iov_len; 202 if (slen == 0) 203 goto out; 204 205 /* send page data */ 206 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; 207 while (pglen > 0) { 208 if (slen == size) 209 flags = 0; 210 result = kernel_sendpage(sock, *ppage, base, size, flags); 211 if (result > 0) 212 len += result; 213 if (result != size) 214 goto out; 215 slen -= size; 216 pglen -= size; 217 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; 218 base = 0; 219 ppage++; 220 } 221 222 /* send tail */ 223 if (xdr->tail[0].iov_len) { 224 result = kernel_sendpage(sock, tailpage, tailoffset, 225 xdr->tail[0].iov_len, 0); 226 if (result > 0) 227 len += result; 228 } 229 230 out: 231 return len; 232 } 233 234 235 /* 236 * Generic sendto routine 237 */ 238 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) 239 { 240 struct svc_sock *svsk = 241 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 242 struct socket *sock = svsk->sk_sock; 243 union { 244 struct cmsghdr hdr; 245 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 246 } buffer; 247 struct cmsghdr *cmh = &buffer.hdr; 248 int len = 0; 249 unsigned long tailoff; 250 unsigned long headoff; 251 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 252 253 if (rqstp->rq_prot == IPPROTO_UDP) { 254 struct msghdr msg = { 255 .msg_name = &rqstp->rq_addr, 256 .msg_namelen = rqstp->rq_addrlen, 257 .msg_control = cmh, 258 .msg_controllen = sizeof(buffer), 259 .msg_flags = MSG_MORE, 260 }; 261 262 svc_set_cmsg_data(rqstp, cmh); 263 264 if (sock_sendmsg(sock, &msg) < 0) 265 goto out; 266 } 267 268 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); 269 headoff = 0; 270 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, 271 rqstp->rq_respages[0], tailoff); 272 273 out: 274 dprintk("svc: socket %p sendto([%p %zu... ], %d) = %d (addr %s)\n", 275 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, 276 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); 277 278 return len; 279 } 280 281 /* 282 * Report socket names for nfsdfs 283 */ 284 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) 285 { 286 const struct sock *sk = svsk->sk_sk; 287 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? 288 "udp" : "tcp"; 289 int len; 290 291 switch (sk->sk_family) { 292 case PF_INET: 293 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", 294 proto_name, 295 &inet_sk(sk)->inet_rcv_saddr, 296 inet_sk(sk)->inet_num); 297 break; 298 #if IS_ENABLED(CONFIG_IPV6) 299 case PF_INET6: 300 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", 301 proto_name, 302 &sk->sk_v6_rcv_saddr, 303 inet_sk(sk)->inet_num); 304 break; 305 #endif 306 default: 307 len = snprintf(buf, remaining, "*unknown-%d*\n", 308 sk->sk_family); 309 } 310 311 if (len >= remaining) { 312 *buf = '\0'; 313 return -ENAMETOOLONG; 314 } 315 return len; 316 } 317 318 /* 319 * Generic recvfrom routine. 320 */ 321 static ssize_t svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, 322 unsigned int nr, size_t buflen, unsigned int base) 323 { 324 struct svc_sock *svsk = 325 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 326 struct msghdr msg = { NULL }; 327 ssize_t len; 328 329 rqstp->rq_xprt_hlen = 0; 330 331 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 332 iov_iter_kvec(&msg.msg_iter, READ, iov, nr, buflen); 333 if (base != 0) { 334 iov_iter_advance(&msg.msg_iter, base); 335 buflen -= base; 336 } 337 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT); 338 /* If we read a full record, then assume there may be more 339 * data to read (stream based sockets only!) 340 */ 341 if (len == buflen) 342 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 343 344 dprintk("svc: socket %p recvfrom(%p, %zu) = %zd\n", 345 svsk, iov[0].iov_base, iov[0].iov_len, len); 346 return len; 347 } 348 349 /* 350 * Set socket snd and rcv buffer lengths 351 */ 352 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, 353 unsigned int rcv) 354 { 355 lock_sock(sock->sk); 356 sock->sk->sk_sndbuf = snd * 2; 357 sock->sk->sk_rcvbuf = rcv * 2; 358 sock->sk->sk_write_space(sock->sk); 359 release_sock(sock->sk); 360 } 361 362 static void svc_sock_secure_port(struct svc_rqst *rqstp) 363 { 364 if (svc_port_is_privileged(svc_addr(rqstp))) 365 set_bit(RQ_SECURE, &rqstp->rq_flags); 366 else 367 clear_bit(RQ_SECURE, &rqstp->rq_flags); 368 } 369 370 /* 371 * INET callback when data has been received on the socket. 372 */ 373 static void svc_data_ready(struct sock *sk) 374 { 375 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 376 377 if (svsk) { 378 dprintk("svc: socket %p(inet %p), busy=%d\n", 379 svsk, sk, 380 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 381 382 /* Refer to svc_setup_socket() for details. */ 383 rmb(); 384 svsk->sk_odata(sk); 385 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags)) 386 svc_xprt_enqueue(&svsk->sk_xprt); 387 } 388 } 389 390 /* 391 * INET callback when space is newly available on the socket. 392 */ 393 static void svc_write_space(struct sock *sk) 394 { 395 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 396 397 if (svsk) { 398 dprintk("svc: socket %p(inet %p), write_space busy=%d\n", 399 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 400 401 /* Refer to svc_setup_socket() for details. */ 402 rmb(); 403 svsk->sk_owspace(sk); 404 svc_xprt_enqueue(&svsk->sk_xprt); 405 } 406 } 407 408 static int svc_tcp_has_wspace(struct svc_xprt *xprt) 409 { 410 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 411 412 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) 413 return 1; 414 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 415 } 416 417 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt) 418 { 419 struct svc_sock *svsk; 420 struct socket *sock; 421 struct linger no_linger = { 422 .l_onoff = 1, 423 .l_linger = 0, 424 }; 425 426 svsk = container_of(xprt, struct svc_sock, sk_xprt); 427 sock = svsk->sk_sock; 428 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER, 429 (char *)&no_linger, sizeof(no_linger)); 430 } 431 432 /* 433 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo 434 */ 435 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, 436 struct cmsghdr *cmh) 437 { 438 struct in_pktinfo *pki = CMSG_DATA(cmh); 439 struct sockaddr_in *daddr = svc_daddr_in(rqstp); 440 441 if (cmh->cmsg_type != IP_PKTINFO) 442 return 0; 443 444 daddr->sin_family = AF_INET; 445 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; 446 return 1; 447 } 448 449 /* 450 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl 451 */ 452 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, 453 struct cmsghdr *cmh) 454 { 455 struct in6_pktinfo *pki = CMSG_DATA(cmh); 456 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 457 458 if (cmh->cmsg_type != IPV6_PKTINFO) 459 return 0; 460 461 daddr->sin6_family = AF_INET6; 462 daddr->sin6_addr = pki->ipi6_addr; 463 daddr->sin6_scope_id = pki->ipi6_ifindex; 464 return 1; 465 } 466 467 /* 468 * Copy the UDP datagram's destination address to the rqstp structure. 469 * The 'destination' address in this case is the address to which the 470 * peer sent the datagram, i.e. our local address. For multihomed 471 * hosts, this can change from msg to msg. Note that only the IP 472 * address changes, the port number should remain the same. 473 */ 474 static int svc_udp_get_dest_address(struct svc_rqst *rqstp, 475 struct cmsghdr *cmh) 476 { 477 switch (cmh->cmsg_level) { 478 case SOL_IP: 479 return svc_udp_get_dest_address4(rqstp, cmh); 480 case SOL_IPV6: 481 return svc_udp_get_dest_address6(rqstp, cmh); 482 } 483 484 return 0; 485 } 486 487 /* 488 * Receive a datagram from a UDP socket. 489 */ 490 static int svc_udp_recvfrom(struct svc_rqst *rqstp) 491 { 492 struct svc_sock *svsk = 493 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 494 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 495 struct sk_buff *skb; 496 union { 497 struct cmsghdr hdr; 498 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 499 } buffer; 500 struct cmsghdr *cmh = &buffer.hdr; 501 struct msghdr msg = { 502 .msg_name = svc_addr(rqstp), 503 .msg_control = cmh, 504 .msg_controllen = sizeof(buffer), 505 .msg_flags = MSG_DONTWAIT, 506 }; 507 size_t len; 508 int err; 509 510 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 511 /* udp sockets need large rcvbuf as all pending 512 * requests are still in that buffer. sndbuf must 513 * also be large enough that there is enough space 514 * for one reply per thread. We count all threads 515 * rather than threads in a particular pool, which 516 * provides an upper bound on the number of threads 517 * which will access the socket. 518 */ 519 svc_sock_setbufsize(svsk->sk_sock, 520 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 521 (serv->sv_nrthreads+3) * serv->sv_max_mesg); 522 523 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 524 skb = NULL; 525 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 526 0, 0, MSG_PEEK | MSG_DONTWAIT); 527 if (err >= 0) 528 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err); 529 530 if (skb == NULL) { 531 if (err != -EAGAIN) { 532 /* possibly an icmp error */ 533 dprintk("svc: recvfrom returned error %d\n", -err); 534 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 535 } 536 return 0; 537 } 538 len = svc_addr_len(svc_addr(rqstp)); 539 rqstp->rq_addrlen = len; 540 if (skb->tstamp == 0) { 541 skb->tstamp = ktime_get_real(); 542 /* Don't enable netstamp, sunrpc doesn't 543 need that much accuracy */ 544 } 545 svsk->sk_sk->sk_stamp = skb->tstamp; 546 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ 547 548 len = skb->len; 549 rqstp->rq_arg.len = len; 550 551 rqstp->rq_prot = IPPROTO_UDP; 552 553 if (!svc_udp_get_dest_address(rqstp, cmh)) { 554 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", 555 cmh->cmsg_level, cmh->cmsg_type); 556 goto out_free; 557 } 558 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); 559 560 if (skb_is_nonlinear(skb)) { 561 /* we have to copy */ 562 local_bh_disable(); 563 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { 564 local_bh_enable(); 565 /* checksum error */ 566 goto out_free; 567 } 568 local_bh_enable(); 569 consume_skb(skb); 570 } else { 571 /* we can use it in-place */ 572 rqstp->rq_arg.head[0].iov_base = skb->data; 573 rqstp->rq_arg.head[0].iov_len = len; 574 if (skb_checksum_complete(skb)) 575 goto out_free; 576 rqstp->rq_xprt_ctxt = skb; 577 } 578 579 rqstp->rq_arg.page_base = 0; 580 if (len <= rqstp->rq_arg.head[0].iov_len) { 581 rqstp->rq_arg.head[0].iov_len = len; 582 rqstp->rq_arg.page_len = 0; 583 rqstp->rq_respages = rqstp->rq_pages+1; 584 } else { 585 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 586 rqstp->rq_respages = rqstp->rq_pages + 1 + 587 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); 588 } 589 rqstp->rq_next_page = rqstp->rq_respages+1; 590 591 if (serv->sv_stats) 592 serv->sv_stats->netudpcnt++; 593 594 return len; 595 out_free: 596 kfree_skb(skb); 597 return 0; 598 } 599 600 static int 601 svc_udp_sendto(struct svc_rqst *rqstp) 602 { 603 int error; 604 605 error = svc_sendto(rqstp, &rqstp->rq_res); 606 if (error == -ECONNREFUSED) 607 /* ICMP error on earlier request. */ 608 error = svc_sendto(rqstp, &rqstp->rq_res); 609 610 return error; 611 } 612 613 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) 614 { 615 } 616 617 static int svc_udp_has_wspace(struct svc_xprt *xprt) 618 { 619 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 620 struct svc_serv *serv = xprt->xpt_server; 621 unsigned long required; 622 623 /* 624 * Set the SOCK_NOSPACE flag before checking the available 625 * sock space. 626 */ 627 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 628 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; 629 if (required*2 > sock_wspace(svsk->sk_sk)) 630 return 0; 631 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 632 return 1; 633 } 634 635 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) 636 { 637 BUG(); 638 return NULL; 639 } 640 641 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt) 642 { 643 } 644 645 static struct svc_xprt *svc_udp_create(struct svc_serv *serv, 646 struct net *net, 647 struct sockaddr *sa, int salen, 648 int flags) 649 { 650 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); 651 } 652 653 static const struct svc_xprt_ops svc_udp_ops = { 654 .xpo_create = svc_udp_create, 655 .xpo_recvfrom = svc_udp_recvfrom, 656 .xpo_sendto = svc_udp_sendto, 657 .xpo_release_rqst = svc_release_udp_skb, 658 .xpo_detach = svc_sock_detach, 659 .xpo_free = svc_sock_free, 660 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, 661 .xpo_has_wspace = svc_udp_has_wspace, 662 .xpo_accept = svc_udp_accept, 663 .xpo_secure_port = svc_sock_secure_port, 664 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt, 665 }; 666 667 static struct svc_xprt_class svc_udp_class = { 668 .xcl_name = "udp", 669 .xcl_owner = THIS_MODULE, 670 .xcl_ops = &svc_udp_ops, 671 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, 672 .xcl_ident = XPRT_TRANSPORT_UDP, 673 }; 674 675 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) 676 { 677 int err, level, optname, one = 1; 678 679 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, 680 &svsk->sk_xprt, serv); 681 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 682 svsk->sk_sk->sk_data_ready = svc_data_ready; 683 svsk->sk_sk->sk_write_space = svc_write_space; 684 685 /* initialise setting must have enough space to 686 * receive and respond to one request. 687 * svc_udp_recvfrom will re-adjust if necessary 688 */ 689 svc_sock_setbufsize(svsk->sk_sock, 690 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, 691 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); 692 693 /* data might have come in before data_ready set up */ 694 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 695 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 696 697 /* make sure we get destination address info */ 698 switch (svsk->sk_sk->sk_family) { 699 case AF_INET: 700 level = SOL_IP; 701 optname = IP_PKTINFO; 702 break; 703 case AF_INET6: 704 level = SOL_IPV6; 705 optname = IPV6_RECVPKTINFO; 706 break; 707 default: 708 BUG(); 709 } 710 err = kernel_setsockopt(svsk->sk_sock, level, optname, 711 (char *)&one, sizeof(one)); 712 dprintk("svc: kernel_setsockopt returned %d\n", err); 713 } 714 715 /* 716 * A data_ready event on a listening socket means there's a connection 717 * pending. Do not use state_change as a substitute for it. 718 */ 719 static void svc_tcp_listen_data_ready(struct sock *sk) 720 { 721 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 722 723 dprintk("svc: socket %p TCP (listen) state change %d\n", 724 sk, sk->sk_state); 725 726 if (svsk) { 727 /* Refer to svc_setup_socket() for details. */ 728 rmb(); 729 svsk->sk_odata(sk); 730 } 731 732 /* 733 * This callback may called twice when a new connection 734 * is established as a child socket inherits everything 735 * from a parent LISTEN socket. 736 * 1) data_ready method of the parent socket will be called 737 * when one of child sockets become ESTABLISHED. 738 * 2) data_ready method of the child socket may be called 739 * when it receives data before the socket is accepted. 740 * In case of 2, we should ignore it silently. 741 */ 742 if (sk->sk_state == TCP_LISTEN) { 743 if (svsk) { 744 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 745 svc_xprt_enqueue(&svsk->sk_xprt); 746 } else 747 printk("svc: socket %p: no user data\n", sk); 748 } 749 } 750 751 /* 752 * A state change on a connected socket means it's dying or dead. 753 */ 754 static void svc_tcp_state_change(struct sock *sk) 755 { 756 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 757 758 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", 759 sk, sk->sk_state, sk->sk_user_data); 760 761 if (!svsk) 762 printk("svc: socket %p: no user data\n", sk); 763 else { 764 /* Refer to svc_setup_socket() for details. */ 765 rmb(); 766 svsk->sk_ostate(sk); 767 if (sk->sk_state != TCP_ESTABLISHED) { 768 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 769 svc_xprt_enqueue(&svsk->sk_xprt); 770 } 771 } 772 } 773 774 /* 775 * Accept a TCP connection 776 */ 777 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) 778 { 779 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 780 struct sockaddr_storage addr; 781 struct sockaddr *sin = (struct sockaddr *) &addr; 782 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 783 struct socket *sock = svsk->sk_sock; 784 struct socket *newsock; 785 struct svc_sock *newsvsk; 786 int err, slen; 787 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 788 789 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); 790 if (!sock) 791 return NULL; 792 793 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 794 err = kernel_accept(sock, &newsock, O_NONBLOCK); 795 if (err < 0) { 796 if (err == -ENOMEM) 797 printk(KERN_WARNING "%s: no more sockets!\n", 798 serv->sv_name); 799 else if (err != -EAGAIN) 800 net_warn_ratelimited("%s: accept failed (err %d)!\n", 801 serv->sv_name, -err); 802 return NULL; 803 } 804 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 805 806 err = kernel_getpeername(newsock, sin); 807 if (err < 0) { 808 net_warn_ratelimited("%s: peername failed (err %d)!\n", 809 serv->sv_name, -err); 810 goto failed; /* aborted connection or whatever */ 811 } 812 slen = err; 813 814 /* Ideally, we would want to reject connections from unauthorized 815 * hosts here, but when we get encryption, the IP of the host won't 816 * tell us anything. For now just warn about unpriv connections. 817 */ 818 if (!svc_port_is_privileged(sin)) { 819 dprintk("%s: connect from unprivileged port: %s\n", 820 serv->sv_name, 821 __svc_print_addr(sin, buf, sizeof(buf))); 822 } 823 dprintk("%s: connect from %s\n", serv->sv_name, 824 __svc_print_addr(sin, buf, sizeof(buf))); 825 826 /* Reset the inherited callbacks before calling svc_setup_socket */ 827 newsock->sk->sk_state_change = svsk->sk_ostate; 828 newsock->sk->sk_data_ready = svsk->sk_odata; 829 newsock->sk->sk_write_space = svsk->sk_owspace; 830 831 /* make sure that a write doesn't block forever when 832 * low on memory 833 */ 834 newsock->sk->sk_sndtimeo = HZ*30; 835 836 newsvsk = svc_setup_socket(serv, newsock, 837 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); 838 if (IS_ERR(newsvsk)) 839 goto failed; 840 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); 841 err = kernel_getsockname(newsock, sin); 842 slen = err; 843 if (unlikely(err < 0)) { 844 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); 845 slen = offsetof(struct sockaddr, sa_data); 846 } 847 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); 848 849 if (sock_is_loopback(newsock->sk)) 850 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 851 else 852 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 853 if (serv->sv_stats) 854 serv->sv_stats->nettcpconn++; 855 856 return &newsvsk->sk_xprt; 857 858 failed: 859 sock_release(newsock); 860 return NULL; 861 } 862 863 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 864 { 865 unsigned int i, len, npages; 866 867 if (svsk->sk_datalen == 0) 868 return 0; 869 len = svsk->sk_datalen; 870 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 871 for (i = 0; i < npages; i++) { 872 if (rqstp->rq_pages[i] != NULL) 873 put_page(rqstp->rq_pages[i]); 874 BUG_ON(svsk->sk_pages[i] == NULL); 875 rqstp->rq_pages[i] = svsk->sk_pages[i]; 876 svsk->sk_pages[i] = NULL; 877 } 878 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); 879 return len; 880 } 881 882 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 883 { 884 unsigned int i, len, npages; 885 886 if (svsk->sk_datalen == 0) 887 return; 888 len = svsk->sk_datalen; 889 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 890 for (i = 0; i < npages; i++) { 891 svsk->sk_pages[i] = rqstp->rq_pages[i]; 892 rqstp->rq_pages[i] = NULL; 893 } 894 } 895 896 static void svc_tcp_clear_pages(struct svc_sock *svsk) 897 { 898 unsigned int i, len, npages; 899 900 if (svsk->sk_datalen == 0) 901 goto out; 902 len = svsk->sk_datalen; 903 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 904 for (i = 0; i < npages; i++) { 905 if (svsk->sk_pages[i] == NULL) { 906 WARN_ON_ONCE(1); 907 continue; 908 } 909 put_page(svsk->sk_pages[i]); 910 svsk->sk_pages[i] = NULL; 911 } 912 out: 913 svsk->sk_tcplen = 0; 914 svsk->sk_datalen = 0; 915 } 916 917 /* 918 * Receive fragment record header. 919 * If we haven't gotten the record length yet, get the next four bytes. 920 */ 921 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) 922 { 923 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 924 unsigned int want; 925 int len; 926 927 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { 928 struct kvec iov; 929 930 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; 931 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; 932 iov.iov_len = want; 933 len = svc_recvfrom(rqstp, &iov, 1, want, 0); 934 if (len < 0) 935 goto error; 936 svsk->sk_tcplen += len; 937 938 if (len < want) { 939 dprintk("svc: short recvfrom while reading record " 940 "length (%d of %d)\n", len, want); 941 return -EAGAIN; 942 } 943 944 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk)); 945 if (svc_sock_reclen(svsk) + svsk->sk_datalen > 946 serv->sv_max_mesg) { 947 net_notice_ratelimited("RPC: fragment too large: %d\n", 948 svc_sock_reclen(svsk)); 949 goto err_delete; 950 } 951 } 952 953 return svc_sock_reclen(svsk); 954 error: 955 dprintk("RPC: TCP recv_record got %d\n", len); 956 return len; 957 err_delete: 958 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 959 return -EAGAIN; 960 } 961 962 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) 963 { 964 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; 965 struct rpc_rqst *req = NULL; 966 struct kvec *src, *dst; 967 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 968 __be32 xid; 969 __be32 calldir; 970 971 xid = *p++; 972 calldir = *p; 973 974 if (!bc_xprt) 975 return -EAGAIN; 976 spin_lock(&bc_xprt->queue_lock); 977 req = xprt_lookup_rqst(bc_xprt, xid); 978 if (!req) 979 goto unlock_notfound; 980 981 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); 982 /* 983 * XXX!: cheating for now! Only copying HEAD. 984 * But we know this is good enough for now (in fact, for any 985 * callback reply in the forseeable future). 986 */ 987 dst = &req->rq_private_buf.head[0]; 988 src = &rqstp->rq_arg.head[0]; 989 if (dst->iov_len < src->iov_len) 990 goto unlock_eagain; /* whatever; just giving up. */ 991 memcpy(dst->iov_base, src->iov_base, src->iov_len); 992 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); 993 rqstp->rq_arg.len = 0; 994 spin_unlock(&bc_xprt->queue_lock); 995 return 0; 996 unlock_notfound: 997 printk(KERN_NOTICE 998 "%s: Got unrecognized reply: " 999 "calldir 0x%x xpt_bc_xprt %p xid %08x\n", 1000 __func__, ntohl(calldir), 1001 bc_xprt, ntohl(xid)); 1002 unlock_eagain: 1003 spin_unlock(&bc_xprt->queue_lock); 1004 return -EAGAIN; 1005 } 1006 1007 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len) 1008 { 1009 int i = 0; 1010 int t = 0; 1011 1012 while (t < len) { 1013 vec[i].iov_base = page_address(pages[i]); 1014 vec[i].iov_len = PAGE_SIZE; 1015 i++; 1016 t += PAGE_SIZE; 1017 } 1018 return i; 1019 } 1020 1021 static void svc_tcp_fragment_received(struct svc_sock *svsk) 1022 { 1023 /* If we have more data, signal svc_xprt_enqueue() to try again */ 1024 dprintk("svc: TCP %s record (%d bytes)\n", 1025 svc_sock_final_rec(svsk) ? "final" : "nonfinal", 1026 svc_sock_reclen(svsk)); 1027 svsk->sk_tcplen = 0; 1028 svsk->sk_reclen = 0; 1029 } 1030 1031 /* 1032 * Receive data from a TCP socket. 1033 */ 1034 static int svc_tcp_recvfrom(struct svc_rqst *rqstp) 1035 { 1036 struct svc_sock *svsk = 1037 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 1038 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1039 int len; 1040 struct kvec *vec; 1041 unsigned int want, base; 1042 __be32 *p; 1043 __be32 calldir; 1044 int pnum; 1045 1046 dprintk("svc: tcp_recv %p data %d conn %d close %d\n", 1047 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), 1048 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), 1049 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1050 1051 len = svc_tcp_recv_record(svsk, rqstp); 1052 if (len < 0) 1053 goto error; 1054 1055 base = svc_tcp_restore_pages(svsk, rqstp); 1056 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); 1057 1058 vec = rqstp->rq_vec; 1059 1060 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], base + want); 1061 1062 rqstp->rq_respages = &rqstp->rq_pages[pnum]; 1063 rqstp->rq_next_page = rqstp->rq_respages + 1; 1064 1065 /* Now receive data */ 1066 len = svc_recvfrom(rqstp, vec, pnum, base + want, base); 1067 if (len >= 0) { 1068 svsk->sk_tcplen += len; 1069 svsk->sk_datalen += len; 1070 } 1071 if (len != want || !svc_sock_final_rec(svsk)) { 1072 svc_tcp_save_pages(svsk, rqstp); 1073 if (len < 0 && len != -EAGAIN) 1074 goto err_delete; 1075 if (len == want) 1076 svc_tcp_fragment_received(svsk); 1077 else 1078 dprintk("svc: incomplete TCP record (%d of %d)\n", 1079 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)), 1080 svc_sock_reclen(svsk)); 1081 goto err_noclose; 1082 } 1083 1084 if (svsk->sk_datalen < 8) { 1085 svsk->sk_datalen = 0; 1086 goto err_delete; /* client is nuts. */ 1087 } 1088 1089 rqstp->rq_arg.len = svsk->sk_datalen; 1090 rqstp->rq_arg.page_base = 0; 1091 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { 1092 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; 1093 rqstp->rq_arg.page_len = 0; 1094 } else 1095 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1096 1097 rqstp->rq_xprt_ctxt = NULL; 1098 rqstp->rq_prot = IPPROTO_TCP; 1099 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags)) 1100 set_bit(RQ_LOCAL, &rqstp->rq_flags); 1101 else 1102 clear_bit(RQ_LOCAL, &rqstp->rq_flags); 1103 1104 p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1105 calldir = p[1]; 1106 if (calldir) 1107 len = receive_cb_reply(svsk, rqstp); 1108 1109 /* Reset TCP read info */ 1110 svsk->sk_datalen = 0; 1111 svc_tcp_fragment_received(svsk); 1112 1113 if (len < 0) 1114 goto error; 1115 1116 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); 1117 if (serv->sv_stats) 1118 serv->sv_stats->nettcpcnt++; 1119 1120 return rqstp->rq_arg.len; 1121 1122 error: 1123 if (len != -EAGAIN) 1124 goto err_delete; 1125 dprintk("RPC: TCP recvfrom got EAGAIN\n"); 1126 return 0; 1127 err_delete: 1128 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", 1129 svsk->sk_xprt.xpt_server->sv_name, -len); 1130 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1131 err_noclose: 1132 return 0; /* record not complete */ 1133 } 1134 1135 /* 1136 * Send out data on TCP socket. 1137 */ 1138 static int svc_tcp_sendto(struct svc_rqst *rqstp) 1139 { 1140 struct xdr_buf *xbufp = &rqstp->rq_res; 1141 int sent; 1142 __be32 reclen; 1143 1144 /* Set up the first element of the reply kvec. 1145 * Any other kvecs that may be in use have been taken 1146 * care of by the server implementation itself. 1147 */ 1148 reclen = htonl(0x80000000|((xbufp->len ) - 4)); 1149 memcpy(xbufp->head[0].iov_base, &reclen, 4); 1150 1151 sent = svc_sendto(rqstp, &rqstp->rq_res); 1152 if (sent != xbufp->len) { 1153 printk(KERN_NOTICE 1154 "rpc-srv/tcp: %s: %s %d when sending %d bytes " 1155 "- shutting down socket\n", 1156 rqstp->rq_xprt->xpt_server->sv_name, 1157 (sent<0)?"got error":"sent only", 1158 sent, xbufp->len); 1159 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); 1160 svc_xprt_enqueue(rqstp->rq_xprt); 1161 sent = -EAGAIN; 1162 } 1163 return sent; 1164 } 1165 1166 /* 1167 * Setup response header. TCP has a 4B record length field. 1168 */ 1169 void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) 1170 { 1171 struct kvec *resv = &rqstp->rq_res.head[0]; 1172 1173 /* tcp needs a space for the record length... */ 1174 svc_putnl(resv, 0); 1175 } 1176 1177 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, 1178 struct net *net, 1179 struct sockaddr *sa, int salen, 1180 int flags) 1181 { 1182 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1183 } 1184 1185 static const struct svc_xprt_ops svc_tcp_ops = { 1186 .xpo_create = svc_tcp_create, 1187 .xpo_recvfrom = svc_tcp_recvfrom, 1188 .xpo_sendto = svc_tcp_sendto, 1189 .xpo_release_rqst = svc_release_skb, 1190 .xpo_detach = svc_tcp_sock_detach, 1191 .xpo_free = svc_sock_free, 1192 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1193 .xpo_has_wspace = svc_tcp_has_wspace, 1194 .xpo_accept = svc_tcp_accept, 1195 .xpo_secure_port = svc_sock_secure_port, 1196 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt, 1197 }; 1198 1199 static struct svc_xprt_class svc_tcp_class = { 1200 .xcl_name = "tcp", 1201 .xcl_owner = THIS_MODULE, 1202 .xcl_ops = &svc_tcp_ops, 1203 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1204 .xcl_ident = XPRT_TRANSPORT_TCP, 1205 }; 1206 1207 void svc_init_xprt_sock(void) 1208 { 1209 svc_reg_xprt_class(&svc_tcp_class); 1210 svc_reg_xprt_class(&svc_udp_class); 1211 } 1212 1213 void svc_cleanup_xprt_sock(void) 1214 { 1215 svc_unreg_xprt_class(&svc_tcp_class); 1216 svc_unreg_xprt_class(&svc_udp_class); 1217 } 1218 1219 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) 1220 { 1221 struct sock *sk = svsk->sk_sk; 1222 1223 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, 1224 &svsk->sk_xprt, serv); 1225 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 1226 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags); 1227 if (sk->sk_state == TCP_LISTEN) { 1228 dprintk("setting up TCP socket for listening\n"); 1229 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener"); 1230 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); 1231 sk->sk_data_ready = svc_tcp_listen_data_ready; 1232 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 1233 } else { 1234 dprintk("setting up TCP socket for reading\n"); 1235 sk->sk_state_change = svc_tcp_state_change; 1236 sk->sk_data_ready = svc_data_ready; 1237 sk->sk_write_space = svc_write_space; 1238 1239 svsk->sk_reclen = 0; 1240 svsk->sk_tcplen = 0; 1241 svsk->sk_datalen = 0; 1242 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); 1243 1244 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1245 1246 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1247 switch (sk->sk_state) { 1248 case TCP_SYN_RECV: 1249 case TCP_ESTABLISHED: 1250 break; 1251 default: 1252 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1253 } 1254 } 1255 } 1256 1257 void svc_sock_update_bufs(struct svc_serv *serv) 1258 { 1259 /* 1260 * The number of server threads has changed. Update 1261 * rcvbuf and sndbuf accordingly on all sockets 1262 */ 1263 struct svc_sock *svsk; 1264 1265 spin_lock_bh(&serv->sv_lock); 1266 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) 1267 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1268 spin_unlock_bh(&serv->sv_lock); 1269 } 1270 EXPORT_SYMBOL_GPL(svc_sock_update_bufs); 1271 1272 /* 1273 * Initialize socket for RPC use and create svc_sock struct 1274 */ 1275 static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1276 struct socket *sock, 1277 int flags) 1278 { 1279 struct svc_sock *svsk; 1280 struct sock *inet; 1281 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1282 int err = 0; 1283 1284 dprintk("svc: svc_setup_socket %p\n", sock); 1285 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1286 if (!svsk) 1287 return ERR_PTR(-ENOMEM); 1288 1289 inet = sock->sk; 1290 1291 /* Register socket with portmapper */ 1292 if (pmap_register) 1293 err = svc_register(serv, sock_net(sock->sk), inet->sk_family, 1294 inet->sk_protocol, 1295 ntohs(inet_sk(inet)->inet_sport)); 1296 1297 if (err < 0) { 1298 kfree(svsk); 1299 return ERR_PTR(err); 1300 } 1301 1302 svsk->sk_sock = sock; 1303 svsk->sk_sk = inet; 1304 svsk->sk_ostate = inet->sk_state_change; 1305 svsk->sk_odata = inet->sk_data_ready; 1306 svsk->sk_owspace = inet->sk_write_space; 1307 /* 1308 * This barrier is necessary in order to prevent race condition 1309 * with svc_data_ready(), svc_listen_data_ready() and others 1310 * when calling callbacks above. 1311 */ 1312 wmb(); 1313 inet->sk_user_data = svsk; 1314 1315 /* Initialize the socket */ 1316 if (sock->type == SOCK_DGRAM) 1317 svc_udp_init(svsk, serv); 1318 else 1319 svc_tcp_init(svsk, serv); 1320 1321 dprintk("svc: svc_setup_socket created %p (inet %p), " 1322 "listen %d close %d\n", 1323 svsk, svsk->sk_sk, 1324 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags), 1325 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1326 1327 return svsk; 1328 } 1329 1330 bool svc_alien_sock(struct net *net, int fd) 1331 { 1332 int err; 1333 struct socket *sock = sockfd_lookup(fd, &err); 1334 bool ret = false; 1335 1336 if (!sock) 1337 goto out; 1338 if (sock_net(sock->sk) != net) 1339 ret = true; 1340 sockfd_put(sock); 1341 out: 1342 return ret; 1343 } 1344 EXPORT_SYMBOL_GPL(svc_alien_sock); 1345 1346 /** 1347 * svc_addsock - add a listener socket to an RPC service 1348 * @serv: pointer to RPC service to which to add a new listener 1349 * @fd: file descriptor of the new listener 1350 * @name_return: pointer to buffer to fill in with name of listener 1351 * @len: size of the buffer 1352 * 1353 * Fills in socket name and returns positive length of name if successful. 1354 * Name is terminated with '\n'. On error, returns a negative errno 1355 * value. 1356 */ 1357 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, 1358 const size_t len) 1359 { 1360 int err = 0; 1361 struct socket *so = sockfd_lookup(fd, &err); 1362 struct svc_sock *svsk = NULL; 1363 struct sockaddr_storage addr; 1364 struct sockaddr *sin = (struct sockaddr *)&addr; 1365 int salen; 1366 1367 if (!so) 1368 return err; 1369 err = -EAFNOSUPPORT; 1370 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) 1371 goto out; 1372 err = -EPROTONOSUPPORT; 1373 if (so->sk->sk_protocol != IPPROTO_TCP && 1374 so->sk->sk_protocol != IPPROTO_UDP) 1375 goto out; 1376 err = -EISCONN; 1377 if (so->state > SS_UNCONNECTED) 1378 goto out; 1379 err = -ENOENT; 1380 if (!try_module_get(THIS_MODULE)) 1381 goto out; 1382 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); 1383 if (IS_ERR(svsk)) { 1384 module_put(THIS_MODULE); 1385 err = PTR_ERR(svsk); 1386 goto out; 1387 } 1388 salen = kernel_getsockname(svsk->sk_sock, sin); 1389 if (salen >= 0) 1390 svc_xprt_set_local(&svsk->sk_xprt, sin, salen); 1391 svc_add_new_perm_xprt(serv, &svsk->sk_xprt); 1392 return svc_one_sock_name(svsk, name_return, len); 1393 out: 1394 sockfd_put(so); 1395 return err; 1396 } 1397 EXPORT_SYMBOL_GPL(svc_addsock); 1398 1399 /* 1400 * Create socket for RPC service. 1401 */ 1402 static struct svc_xprt *svc_create_socket(struct svc_serv *serv, 1403 int protocol, 1404 struct net *net, 1405 struct sockaddr *sin, int len, 1406 int flags) 1407 { 1408 struct svc_sock *svsk; 1409 struct socket *sock; 1410 int error; 1411 int type; 1412 struct sockaddr_storage addr; 1413 struct sockaddr *newsin = (struct sockaddr *)&addr; 1414 int newlen; 1415 int family; 1416 int val; 1417 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 1418 1419 dprintk("svc: svc_create_socket(%s, %d, %s)\n", 1420 serv->sv_program->pg_name, protocol, 1421 __svc_print_addr(sin, buf, sizeof(buf))); 1422 1423 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1424 printk(KERN_WARNING "svc: only UDP and TCP " 1425 "sockets supported\n"); 1426 return ERR_PTR(-EINVAL); 1427 } 1428 1429 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1430 switch (sin->sa_family) { 1431 case AF_INET6: 1432 family = PF_INET6; 1433 break; 1434 case AF_INET: 1435 family = PF_INET; 1436 break; 1437 default: 1438 return ERR_PTR(-EINVAL); 1439 } 1440 1441 error = __sock_create(net, family, type, protocol, &sock, 1); 1442 if (error < 0) 1443 return ERR_PTR(error); 1444 1445 svc_reclassify_socket(sock); 1446 1447 /* 1448 * If this is an PF_INET6 listener, we want to avoid 1449 * getting requests from IPv4 remotes. Those should 1450 * be shunted to a PF_INET listener via rpcbind. 1451 */ 1452 val = 1; 1453 if (family == PF_INET6) 1454 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, 1455 (char *)&val, sizeof(val)); 1456 1457 if (type == SOCK_STREAM) 1458 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ 1459 error = kernel_bind(sock, sin, len); 1460 if (error < 0) 1461 goto bummer; 1462 1463 error = kernel_getsockname(sock, newsin); 1464 if (error < 0) 1465 goto bummer; 1466 newlen = error; 1467 1468 if (protocol == IPPROTO_TCP) { 1469 if ((error = kernel_listen(sock, 64)) < 0) 1470 goto bummer; 1471 } 1472 1473 svsk = svc_setup_socket(serv, sock, flags); 1474 if (IS_ERR(svsk)) { 1475 error = PTR_ERR(svsk); 1476 goto bummer; 1477 } 1478 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); 1479 return (struct svc_xprt *)svsk; 1480 bummer: 1481 dprintk("svc: svc_create_socket error = %d\n", -error); 1482 sock_release(sock); 1483 return ERR_PTR(error); 1484 } 1485 1486 /* 1487 * Detach the svc_sock from the socket so that no 1488 * more callbacks occur. 1489 */ 1490 static void svc_sock_detach(struct svc_xprt *xprt) 1491 { 1492 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1493 struct sock *sk = svsk->sk_sk; 1494 1495 dprintk("svc: svc_sock_detach(%p)\n", svsk); 1496 1497 /* put back the old socket callbacks */ 1498 lock_sock(sk); 1499 sk->sk_state_change = svsk->sk_ostate; 1500 sk->sk_data_ready = svsk->sk_odata; 1501 sk->sk_write_space = svsk->sk_owspace; 1502 sk->sk_user_data = NULL; 1503 release_sock(sk); 1504 } 1505 1506 /* 1507 * Disconnect the socket, and reset the callbacks 1508 */ 1509 static void svc_tcp_sock_detach(struct svc_xprt *xprt) 1510 { 1511 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1512 1513 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); 1514 1515 svc_sock_detach(xprt); 1516 1517 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { 1518 svc_tcp_clear_pages(svsk); 1519 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); 1520 } 1521 } 1522 1523 /* 1524 * Free the svc_sock's socket resources and the svc_sock itself. 1525 */ 1526 static void svc_sock_free(struct svc_xprt *xprt) 1527 { 1528 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1529 dprintk("svc: svc_sock_free(%p)\n", svsk); 1530 1531 if (svsk->sk_sock->file) 1532 sockfd_put(svsk->sk_sock); 1533 else 1534 sock_release(svsk->sk_sock); 1535 kfree(svsk); 1536 } 1537