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/tcp.h> 43 #include <net/tcp_states.h> 44 #include <asm/uaccess.h> 45 #include <asm/ioctls.h> 46 #include <trace/events/skb.h> 47 48 #include <linux/sunrpc/types.h> 49 #include <linux/sunrpc/clnt.h> 50 #include <linux/sunrpc/xdr.h> 51 #include <linux/sunrpc/msg_prot.h> 52 #include <linux/sunrpc/svcsock.h> 53 #include <linux/sunrpc/stats.h> 54 #include <linux/sunrpc/xprt.h> 55 56 #include "sunrpc.h" 57 58 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 59 60 61 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, 62 int flags); 63 static int svc_udp_recvfrom(struct svc_rqst *); 64 static int svc_udp_sendto(struct svc_rqst *); 65 static void svc_sock_detach(struct svc_xprt *); 66 static void svc_tcp_sock_detach(struct svc_xprt *); 67 static void svc_sock_free(struct svc_xprt *); 68 69 static struct svc_xprt *svc_create_socket(struct svc_serv *, int, 70 struct net *, struct sockaddr *, 71 int, int); 72 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 73 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, 74 struct net *, struct sockaddr *, 75 int, int); 76 static void svc_bc_sock_free(struct svc_xprt *xprt); 77 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 78 79 #ifdef CONFIG_DEBUG_LOCK_ALLOC 80 static struct lock_class_key svc_key[2]; 81 static struct lock_class_key svc_slock_key[2]; 82 83 static void svc_reclassify_socket(struct socket *sock) 84 { 85 struct sock *sk = sock->sk; 86 87 if (WARN_ON_ONCE(!sock_allow_reclassification(sk))) 88 return; 89 90 switch (sk->sk_family) { 91 case AF_INET: 92 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", 93 &svc_slock_key[0], 94 "sk_xprt.xpt_lock-AF_INET-NFSD", 95 &svc_key[0]); 96 break; 97 98 case AF_INET6: 99 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", 100 &svc_slock_key[1], 101 "sk_xprt.xpt_lock-AF_INET6-NFSD", 102 &svc_key[1]); 103 break; 104 105 default: 106 BUG(); 107 } 108 } 109 #else 110 static void svc_reclassify_socket(struct socket *sock) 111 { 112 } 113 #endif 114 115 /* 116 * Release an skbuff after use 117 */ 118 static void svc_release_skb(struct svc_rqst *rqstp) 119 { 120 struct sk_buff *skb = rqstp->rq_xprt_ctxt; 121 122 if (skb) { 123 struct svc_sock *svsk = 124 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 125 rqstp->rq_xprt_ctxt = NULL; 126 127 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); 128 skb_free_datagram_locked(svsk->sk_sk, skb); 129 } 130 } 131 132 union svc_pktinfo_u { 133 struct in_pktinfo pkti; 134 struct in6_pktinfo pkti6; 135 }; 136 #define SVC_PKTINFO_SPACE \ 137 CMSG_SPACE(sizeof(union svc_pktinfo_u)) 138 139 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) 140 { 141 struct svc_sock *svsk = 142 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 143 switch (svsk->sk_sk->sk_family) { 144 case AF_INET: { 145 struct in_pktinfo *pki = CMSG_DATA(cmh); 146 147 cmh->cmsg_level = SOL_IP; 148 cmh->cmsg_type = IP_PKTINFO; 149 pki->ipi_ifindex = 0; 150 pki->ipi_spec_dst.s_addr = 151 svc_daddr_in(rqstp)->sin_addr.s_addr; 152 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 153 } 154 break; 155 156 case AF_INET6: { 157 struct in6_pktinfo *pki = CMSG_DATA(cmh); 158 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 159 160 cmh->cmsg_level = SOL_IPV6; 161 cmh->cmsg_type = IPV6_PKTINFO; 162 pki->ipi6_ifindex = daddr->sin6_scope_id; 163 pki->ipi6_addr = daddr->sin6_addr; 164 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 165 } 166 break; 167 } 168 } 169 170 /* 171 * send routine intended to be shared by the fore- and back-channel 172 */ 173 int svc_send_common(struct socket *sock, struct xdr_buf *xdr, 174 struct page *headpage, unsigned long headoffset, 175 struct page *tailpage, unsigned long tailoffset) 176 { 177 int result; 178 int size; 179 struct page **ppage = xdr->pages; 180 size_t base = xdr->page_base; 181 unsigned int pglen = xdr->page_len; 182 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST; 183 int slen; 184 int len = 0; 185 186 slen = xdr->len; 187 188 /* send head */ 189 if (slen == xdr->head[0].iov_len) 190 flags = 0; 191 len = kernel_sendpage(sock, headpage, headoffset, 192 xdr->head[0].iov_len, flags); 193 if (len != xdr->head[0].iov_len) 194 goto out; 195 slen -= xdr->head[0].iov_len; 196 if (slen == 0) 197 goto out; 198 199 /* send page data */ 200 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; 201 while (pglen > 0) { 202 if (slen == size) 203 flags = 0; 204 result = kernel_sendpage(sock, *ppage, base, size, flags); 205 if (result > 0) 206 len += result; 207 if (result != size) 208 goto out; 209 slen -= size; 210 pglen -= size; 211 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; 212 base = 0; 213 ppage++; 214 } 215 216 /* send tail */ 217 if (xdr->tail[0].iov_len) { 218 result = kernel_sendpage(sock, tailpage, tailoffset, 219 xdr->tail[0].iov_len, 0); 220 if (result > 0) 221 len += result; 222 } 223 224 out: 225 return len; 226 } 227 228 229 /* 230 * Generic sendto routine 231 */ 232 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) 233 { 234 struct svc_sock *svsk = 235 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 236 struct socket *sock = svsk->sk_sock; 237 union { 238 struct cmsghdr hdr; 239 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 240 } buffer; 241 struct cmsghdr *cmh = &buffer.hdr; 242 int len = 0; 243 unsigned long tailoff; 244 unsigned long headoff; 245 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 246 247 if (rqstp->rq_prot == IPPROTO_UDP) { 248 struct msghdr msg = { 249 .msg_name = &rqstp->rq_addr, 250 .msg_namelen = rqstp->rq_addrlen, 251 .msg_control = cmh, 252 .msg_controllen = sizeof(buffer), 253 .msg_flags = MSG_MORE, 254 }; 255 256 svc_set_cmsg_data(rqstp, cmh); 257 258 if (sock_sendmsg(sock, &msg) < 0) 259 goto out; 260 } 261 262 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); 263 headoff = 0; 264 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, 265 rqstp->rq_respages[0], tailoff); 266 267 out: 268 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", 269 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, 270 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); 271 272 return len; 273 } 274 275 /* 276 * Report socket names for nfsdfs 277 */ 278 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) 279 { 280 const struct sock *sk = svsk->sk_sk; 281 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? 282 "udp" : "tcp"; 283 int len; 284 285 switch (sk->sk_family) { 286 case PF_INET: 287 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", 288 proto_name, 289 &inet_sk(sk)->inet_rcv_saddr, 290 inet_sk(sk)->inet_num); 291 break; 292 #if IS_ENABLED(CONFIG_IPV6) 293 case PF_INET6: 294 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", 295 proto_name, 296 &sk->sk_v6_rcv_saddr, 297 inet_sk(sk)->inet_num); 298 break; 299 #endif 300 default: 301 len = snprintf(buf, remaining, "*unknown-%d*\n", 302 sk->sk_family); 303 } 304 305 if (len >= remaining) { 306 *buf = '\0'; 307 return -ENAMETOOLONG; 308 } 309 return len; 310 } 311 312 /* 313 * Generic recvfrom routine. 314 */ 315 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, 316 int buflen) 317 { 318 struct svc_sock *svsk = 319 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 320 struct msghdr msg = { 321 .msg_flags = MSG_DONTWAIT, 322 }; 323 int len; 324 325 rqstp->rq_xprt_hlen = 0; 326 327 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 328 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, 329 msg.msg_flags); 330 /* If we read a full record, then assume there may be more 331 * data to read (stream based sockets only!) 332 */ 333 if (len == buflen) 334 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 335 336 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", 337 svsk, iov[0].iov_base, iov[0].iov_len, len); 338 return len; 339 } 340 341 static int svc_partial_recvfrom(struct svc_rqst *rqstp, 342 struct kvec *iov, int nr, 343 int buflen, unsigned int base) 344 { 345 size_t save_iovlen; 346 void *save_iovbase; 347 unsigned int i; 348 int ret; 349 350 if (base == 0) 351 return svc_recvfrom(rqstp, iov, nr, buflen); 352 353 for (i = 0; i < nr; i++) { 354 if (iov[i].iov_len > base) 355 break; 356 base -= iov[i].iov_len; 357 } 358 save_iovlen = iov[i].iov_len; 359 save_iovbase = iov[i].iov_base; 360 iov[i].iov_len -= base; 361 iov[i].iov_base += base; 362 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen); 363 iov[i].iov_len = save_iovlen; 364 iov[i].iov_base = save_iovbase; 365 return ret; 366 } 367 368 /* 369 * Set socket snd and rcv buffer lengths 370 */ 371 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, 372 unsigned int rcv) 373 { 374 #if 0 375 mm_segment_t oldfs; 376 oldfs = get_fs(); set_fs(KERNEL_DS); 377 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, 378 (char*)&snd, sizeof(snd)); 379 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, 380 (char*)&rcv, sizeof(rcv)); 381 #else 382 /* sock_setsockopt limits use to sysctl_?mem_max, 383 * which isn't acceptable. Until that is made conditional 384 * on not having CAP_SYS_RESOURCE or similar, we go direct... 385 * DaveM said I could! 386 */ 387 lock_sock(sock->sk); 388 sock->sk->sk_sndbuf = snd * 2; 389 sock->sk->sk_rcvbuf = rcv * 2; 390 sock->sk->sk_write_space(sock->sk); 391 release_sock(sock->sk); 392 #endif 393 } 394 395 static int svc_sock_secure_port(struct svc_rqst *rqstp) 396 { 397 return svc_port_is_privileged(svc_addr(rqstp)); 398 } 399 400 /* 401 * INET callback when data has been received on the socket. 402 */ 403 static void svc_data_ready(struct sock *sk) 404 { 405 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 406 407 if (svsk) { 408 dprintk("svc: socket %p(inet %p), busy=%d\n", 409 svsk, sk, 410 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 411 svsk->sk_odata(sk); 412 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags)) 413 svc_xprt_enqueue(&svsk->sk_xprt); 414 } 415 } 416 417 /* 418 * INET callback when space is newly available on the socket. 419 */ 420 static void svc_write_space(struct sock *sk) 421 { 422 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 423 424 if (svsk) { 425 dprintk("svc: socket %p(inet %p), write_space busy=%d\n", 426 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 427 svsk->sk_owspace(sk); 428 svc_xprt_enqueue(&svsk->sk_xprt); 429 } 430 } 431 432 static int svc_tcp_has_wspace(struct svc_xprt *xprt) 433 { 434 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 435 436 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) 437 return 1; 438 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 439 } 440 441 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt) 442 { 443 struct svc_sock *svsk; 444 struct socket *sock; 445 struct linger no_linger = { 446 .l_onoff = 1, 447 .l_linger = 0, 448 }; 449 450 svsk = container_of(xprt, struct svc_sock, sk_xprt); 451 sock = svsk->sk_sock; 452 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER, 453 (char *)&no_linger, sizeof(no_linger)); 454 } 455 456 /* 457 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo 458 */ 459 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, 460 struct cmsghdr *cmh) 461 { 462 struct in_pktinfo *pki = CMSG_DATA(cmh); 463 struct sockaddr_in *daddr = svc_daddr_in(rqstp); 464 465 if (cmh->cmsg_type != IP_PKTINFO) 466 return 0; 467 468 daddr->sin_family = AF_INET; 469 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; 470 return 1; 471 } 472 473 /* 474 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl 475 */ 476 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, 477 struct cmsghdr *cmh) 478 { 479 struct in6_pktinfo *pki = CMSG_DATA(cmh); 480 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 481 482 if (cmh->cmsg_type != IPV6_PKTINFO) 483 return 0; 484 485 daddr->sin6_family = AF_INET6; 486 daddr->sin6_addr = pki->ipi6_addr; 487 daddr->sin6_scope_id = pki->ipi6_ifindex; 488 return 1; 489 } 490 491 /* 492 * Copy the UDP datagram's destination address to the rqstp structure. 493 * The 'destination' address in this case is the address to which the 494 * peer sent the datagram, i.e. our local address. For multihomed 495 * hosts, this can change from msg to msg. Note that only the IP 496 * address changes, the port number should remain the same. 497 */ 498 static int svc_udp_get_dest_address(struct svc_rqst *rqstp, 499 struct cmsghdr *cmh) 500 { 501 switch (cmh->cmsg_level) { 502 case SOL_IP: 503 return svc_udp_get_dest_address4(rqstp, cmh); 504 case SOL_IPV6: 505 return svc_udp_get_dest_address6(rqstp, cmh); 506 } 507 508 return 0; 509 } 510 511 /* 512 * Receive a datagram from a UDP socket. 513 */ 514 static int svc_udp_recvfrom(struct svc_rqst *rqstp) 515 { 516 struct svc_sock *svsk = 517 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 518 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 519 struct sk_buff *skb; 520 union { 521 struct cmsghdr hdr; 522 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 523 } buffer; 524 struct cmsghdr *cmh = &buffer.hdr; 525 struct msghdr msg = { 526 .msg_name = svc_addr(rqstp), 527 .msg_control = cmh, 528 .msg_controllen = sizeof(buffer), 529 .msg_flags = MSG_DONTWAIT, 530 }; 531 size_t len; 532 int err; 533 534 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 535 /* udp sockets need large rcvbuf as all pending 536 * requests are still in that buffer. sndbuf must 537 * also be large enough that there is enough space 538 * for one reply per thread. We count all threads 539 * rather than threads in a particular pool, which 540 * provides an upper bound on the number of threads 541 * which will access the socket. 542 */ 543 svc_sock_setbufsize(svsk->sk_sock, 544 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 545 (serv->sv_nrthreads+3) * serv->sv_max_mesg); 546 547 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 548 skb = NULL; 549 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 550 0, 0, MSG_PEEK | MSG_DONTWAIT); 551 if (err >= 0) 552 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); 553 554 if (skb == NULL) { 555 if (err != -EAGAIN) { 556 /* possibly an icmp error */ 557 dprintk("svc: recvfrom returned error %d\n", -err); 558 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 559 } 560 return 0; 561 } 562 len = svc_addr_len(svc_addr(rqstp)); 563 rqstp->rq_addrlen = len; 564 if (skb->tstamp.tv64 == 0) { 565 skb->tstamp = ktime_get_real(); 566 /* Don't enable netstamp, sunrpc doesn't 567 need that much accuracy */ 568 } 569 svsk->sk_sk->sk_stamp = skb->tstamp; 570 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ 571 572 len = skb->len; 573 rqstp->rq_arg.len = len; 574 575 rqstp->rq_prot = IPPROTO_UDP; 576 577 if (!svc_udp_get_dest_address(rqstp, cmh)) { 578 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", 579 cmh->cmsg_level, cmh->cmsg_type); 580 goto out_free; 581 } 582 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); 583 584 if (skb_is_nonlinear(skb)) { 585 /* we have to copy */ 586 local_bh_disable(); 587 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { 588 local_bh_enable(); 589 /* checksum error */ 590 goto out_free; 591 } 592 local_bh_enable(); 593 skb_free_datagram_locked(svsk->sk_sk, skb); 594 } else { 595 /* we can use it in-place */ 596 rqstp->rq_arg.head[0].iov_base = skb->data; 597 rqstp->rq_arg.head[0].iov_len = len; 598 if (skb_checksum_complete(skb)) 599 goto out_free; 600 rqstp->rq_xprt_ctxt = skb; 601 } 602 603 rqstp->rq_arg.page_base = 0; 604 if (len <= rqstp->rq_arg.head[0].iov_len) { 605 rqstp->rq_arg.head[0].iov_len = len; 606 rqstp->rq_arg.page_len = 0; 607 rqstp->rq_respages = rqstp->rq_pages+1; 608 } else { 609 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 610 rqstp->rq_respages = rqstp->rq_pages + 1 + 611 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); 612 } 613 rqstp->rq_next_page = rqstp->rq_respages+1; 614 615 if (serv->sv_stats) 616 serv->sv_stats->netudpcnt++; 617 618 return len; 619 out_free: 620 trace_kfree_skb(skb, svc_udp_recvfrom); 621 skb_free_datagram_locked(svsk->sk_sk, skb); 622 return 0; 623 } 624 625 static int 626 svc_udp_sendto(struct svc_rqst *rqstp) 627 { 628 int error; 629 630 error = svc_sendto(rqstp, &rqstp->rq_res); 631 if (error == -ECONNREFUSED) 632 /* ICMP error on earlier request. */ 633 error = svc_sendto(rqstp, &rqstp->rq_res); 634 635 return error; 636 } 637 638 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) 639 { 640 } 641 642 static int svc_udp_has_wspace(struct svc_xprt *xprt) 643 { 644 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 645 struct svc_serv *serv = xprt->xpt_server; 646 unsigned long required; 647 648 /* 649 * Set the SOCK_NOSPACE flag before checking the available 650 * sock space. 651 */ 652 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 653 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; 654 if (required*2 > sock_wspace(svsk->sk_sk)) 655 return 0; 656 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 657 return 1; 658 } 659 660 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) 661 { 662 BUG(); 663 return NULL; 664 } 665 666 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt) 667 { 668 } 669 670 static struct svc_xprt *svc_udp_create(struct svc_serv *serv, 671 struct net *net, 672 struct sockaddr *sa, int salen, 673 int flags) 674 { 675 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); 676 } 677 678 static struct svc_xprt_ops svc_udp_ops = { 679 .xpo_create = svc_udp_create, 680 .xpo_recvfrom = svc_udp_recvfrom, 681 .xpo_sendto = svc_udp_sendto, 682 .xpo_release_rqst = svc_release_skb, 683 .xpo_detach = svc_sock_detach, 684 .xpo_free = svc_sock_free, 685 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, 686 .xpo_has_wspace = svc_udp_has_wspace, 687 .xpo_accept = svc_udp_accept, 688 .xpo_secure_port = svc_sock_secure_port, 689 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt, 690 }; 691 692 static struct svc_xprt_class svc_udp_class = { 693 .xcl_name = "udp", 694 .xcl_owner = THIS_MODULE, 695 .xcl_ops = &svc_udp_ops, 696 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, 697 .xcl_ident = XPRT_TRANSPORT_UDP, 698 }; 699 700 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) 701 { 702 int err, level, optname, one = 1; 703 704 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, 705 &svsk->sk_xprt, serv); 706 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 707 svsk->sk_sk->sk_data_ready = svc_data_ready; 708 svsk->sk_sk->sk_write_space = svc_write_space; 709 710 /* initialise setting must have enough space to 711 * receive and respond to one request. 712 * svc_udp_recvfrom will re-adjust if necessary 713 */ 714 svc_sock_setbufsize(svsk->sk_sock, 715 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, 716 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); 717 718 /* data might have come in before data_ready set up */ 719 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 720 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 721 722 /* make sure we get destination address info */ 723 switch (svsk->sk_sk->sk_family) { 724 case AF_INET: 725 level = SOL_IP; 726 optname = IP_PKTINFO; 727 break; 728 case AF_INET6: 729 level = SOL_IPV6; 730 optname = IPV6_RECVPKTINFO; 731 break; 732 default: 733 BUG(); 734 } 735 err = kernel_setsockopt(svsk->sk_sock, level, optname, 736 (char *)&one, sizeof(one)); 737 dprintk("svc: kernel_setsockopt returned %d\n", err); 738 } 739 740 /* 741 * A data_ready event on a listening socket means there's a connection 742 * pending. Do not use state_change as a substitute for it. 743 */ 744 static void svc_tcp_listen_data_ready(struct sock *sk) 745 { 746 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 747 748 dprintk("svc: socket %p TCP (listen) state change %d\n", 749 sk, sk->sk_state); 750 751 if (svsk) 752 svsk->sk_odata(sk); 753 /* 754 * This callback may called twice when a new connection 755 * is established as a child socket inherits everything 756 * from a parent LISTEN socket. 757 * 1) data_ready method of the parent socket will be called 758 * when one of child sockets become ESTABLISHED. 759 * 2) data_ready method of the child socket may be called 760 * when it receives data before the socket is accepted. 761 * In case of 2, we should ignore it silently. 762 */ 763 if (sk->sk_state == TCP_LISTEN) { 764 if (svsk) { 765 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 766 svc_xprt_enqueue(&svsk->sk_xprt); 767 } else 768 printk("svc: socket %p: no user data\n", sk); 769 } 770 } 771 772 /* 773 * A state change on a connected socket means it's dying or dead. 774 */ 775 static void svc_tcp_state_change(struct sock *sk) 776 { 777 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 778 779 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", 780 sk, sk->sk_state, sk->sk_user_data); 781 782 if (!svsk) 783 printk("svc: socket %p: no user data\n", sk); 784 else { 785 svsk->sk_ostate(sk); 786 if (sk->sk_state != TCP_ESTABLISHED) { 787 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 788 svc_xprt_enqueue(&svsk->sk_xprt); 789 } 790 } 791 } 792 793 /* 794 * Accept a TCP connection 795 */ 796 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) 797 { 798 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 799 struct sockaddr_storage addr; 800 struct sockaddr *sin = (struct sockaddr *) &addr; 801 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 802 struct socket *sock = svsk->sk_sock; 803 struct socket *newsock; 804 struct svc_sock *newsvsk; 805 int err, slen; 806 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 807 808 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); 809 if (!sock) 810 return NULL; 811 812 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 813 err = kernel_accept(sock, &newsock, O_NONBLOCK); 814 if (err < 0) { 815 if (err == -ENOMEM) 816 printk(KERN_WARNING "%s: no more sockets!\n", 817 serv->sv_name); 818 else if (err != -EAGAIN) 819 net_warn_ratelimited("%s: accept failed (err %d)!\n", 820 serv->sv_name, -err); 821 return NULL; 822 } 823 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 824 825 err = kernel_getpeername(newsock, sin, &slen); 826 if (err < 0) { 827 net_warn_ratelimited("%s: peername failed (err %d)!\n", 828 serv->sv_name, -err); 829 goto failed; /* aborted connection or whatever */ 830 } 831 832 /* Ideally, we would want to reject connections from unauthorized 833 * hosts here, but when we get encryption, the IP of the host won't 834 * tell us anything. For now just warn about unpriv connections. 835 */ 836 if (!svc_port_is_privileged(sin)) { 837 dprintk("%s: connect from unprivileged port: %s\n", 838 serv->sv_name, 839 __svc_print_addr(sin, buf, sizeof(buf))); 840 } 841 dprintk("%s: connect from %s\n", serv->sv_name, 842 __svc_print_addr(sin, buf, sizeof(buf))); 843 844 /* Reset the inherited callbacks before calling svc_setup_socket */ 845 newsock->sk->sk_state_change = svsk->sk_ostate; 846 newsock->sk->sk_data_ready = svsk->sk_odata; 847 newsock->sk->sk_write_space = svsk->sk_owspace; 848 849 /* make sure that a write doesn't block forever when 850 * low on memory 851 */ 852 newsock->sk->sk_sndtimeo = HZ*30; 853 854 newsvsk = svc_setup_socket(serv, newsock, 855 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); 856 if (IS_ERR(newsvsk)) 857 goto failed; 858 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); 859 err = kernel_getsockname(newsock, sin, &slen); 860 if (unlikely(err < 0)) { 861 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); 862 slen = offsetof(struct sockaddr, sa_data); 863 } 864 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); 865 866 if (sock_is_loopback(newsock->sk)) 867 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 868 else 869 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 870 if (serv->sv_stats) 871 serv->sv_stats->nettcpconn++; 872 873 return &newsvsk->sk_xprt; 874 875 failed: 876 sock_release(newsock); 877 return NULL; 878 } 879 880 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 881 { 882 unsigned int i, len, npages; 883 884 if (svsk->sk_datalen == 0) 885 return 0; 886 len = svsk->sk_datalen; 887 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 888 for (i = 0; i < npages; i++) { 889 if (rqstp->rq_pages[i] != NULL) 890 put_page(rqstp->rq_pages[i]); 891 BUG_ON(svsk->sk_pages[i] == NULL); 892 rqstp->rq_pages[i] = svsk->sk_pages[i]; 893 svsk->sk_pages[i] = NULL; 894 } 895 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); 896 return len; 897 } 898 899 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 900 { 901 unsigned int i, len, npages; 902 903 if (svsk->sk_datalen == 0) 904 return; 905 len = svsk->sk_datalen; 906 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 907 for (i = 0; i < npages; i++) { 908 svsk->sk_pages[i] = rqstp->rq_pages[i]; 909 rqstp->rq_pages[i] = NULL; 910 } 911 } 912 913 static void svc_tcp_clear_pages(struct svc_sock *svsk) 914 { 915 unsigned int i, len, npages; 916 917 if (svsk->sk_datalen == 0) 918 goto out; 919 len = svsk->sk_datalen; 920 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 921 for (i = 0; i < npages; i++) { 922 if (svsk->sk_pages[i] == NULL) { 923 WARN_ON_ONCE(1); 924 continue; 925 } 926 put_page(svsk->sk_pages[i]); 927 svsk->sk_pages[i] = NULL; 928 } 929 out: 930 svsk->sk_tcplen = 0; 931 svsk->sk_datalen = 0; 932 } 933 934 /* 935 * Receive fragment record header. 936 * If we haven't gotten the record length yet, get the next four bytes. 937 */ 938 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) 939 { 940 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 941 unsigned int want; 942 int len; 943 944 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { 945 struct kvec iov; 946 947 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; 948 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; 949 iov.iov_len = want; 950 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) 951 goto error; 952 svsk->sk_tcplen += len; 953 954 if (len < want) { 955 dprintk("svc: short recvfrom while reading record " 956 "length (%d of %d)\n", len, want); 957 return -EAGAIN; 958 } 959 960 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk)); 961 if (svc_sock_reclen(svsk) + svsk->sk_datalen > 962 serv->sv_max_mesg) { 963 net_notice_ratelimited("RPC: fragment too large: %d\n", 964 svc_sock_reclen(svsk)); 965 goto err_delete; 966 } 967 } 968 969 return svc_sock_reclen(svsk); 970 error: 971 dprintk("RPC: TCP recv_record got %d\n", len); 972 return len; 973 err_delete: 974 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 975 return -EAGAIN; 976 } 977 978 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) 979 { 980 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; 981 struct rpc_rqst *req = NULL; 982 struct kvec *src, *dst; 983 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 984 __be32 xid; 985 __be32 calldir; 986 987 xid = *p++; 988 calldir = *p; 989 990 if (!bc_xprt) 991 return -EAGAIN; 992 spin_lock_bh(&bc_xprt->transport_lock); 993 req = xprt_lookup_rqst(bc_xprt, xid); 994 if (!req) 995 goto unlock_notfound; 996 997 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); 998 /* 999 * XXX!: cheating for now! Only copying HEAD. 1000 * But we know this is good enough for now (in fact, for any 1001 * callback reply in the forseeable future). 1002 */ 1003 dst = &req->rq_private_buf.head[0]; 1004 src = &rqstp->rq_arg.head[0]; 1005 if (dst->iov_len < src->iov_len) 1006 goto unlock_eagain; /* whatever; just giving up. */ 1007 memcpy(dst->iov_base, src->iov_base, src->iov_len); 1008 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); 1009 rqstp->rq_arg.len = 0; 1010 spin_unlock_bh(&bc_xprt->transport_lock); 1011 return 0; 1012 unlock_notfound: 1013 printk(KERN_NOTICE 1014 "%s: Got unrecognized reply: " 1015 "calldir 0x%x xpt_bc_xprt %p xid %08x\n", 1016 __func__, ntohl(calldir), 1017 bc_xprt, ntohl(xid)); 1018 unlock_eagain: 1019 spin_unlock_bh(&bc_xprt->transport_lock); 1020 return -EAGAIN; 1021 } 1022 1023 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len) 1024 { 1025 int i = 0; 1026 int t = 0; 1027 1028 while (t < len) { 1029 vec[i].iov_base = page_address(pages[i]); 1030 vec[i].iov_len = PAGE_SIZE; 1031 i++; 1032 t += PAGE_SIZE; 1033 } 1034 return i; 1035 } 1036 1037 static void svc_tcp_fragment_received(struct svc_sock *svsk) 1038 { 1039 /* If we have more data, signal svc_xprt_enqueue() to try again */ 1040 dprintk("svc: TCP %s record (%d bytes)\n", 1041 svc_sock_final_rec(svsk) ? "final" : "nonfinal", 1042 svc_sock_reclen(svsk)); 1043 svsk->sk_tcplen = 0; 1044 svsk->sk_reclen = 0; 1045 } 1046 1047 /* 1048 * Receive data from a TCP socket. 1049 */ 1050 static int svc_tcp_recvfrom(struct svc_rqst *rqstp) 1051 { 1052 struct svc_sock *svsk = 1053 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 1054 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1055 int len; 1056 struct kvec *vec; 1057 unsigned int want, base; 1058 __be32 *p; 1059 __be32 calldir; 1060 int pnum; 1061 1062 dprintk("svc: tcp_recv %p data %d conn %d close %d\n", 1063 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), 1064 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), 1065 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1066 1067 len = svc_tcp_recv_record(svsk, rqstp); 1068 if (len < 0) 1069 goto error; 1070 1071 base = svc_tcp_restore_pages(svsk, rqstp); 1072 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); 1073 1074 vec = rqstp->rq_vec; 1075 1076 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], 1077 svsk->sk_datalen + want); 1078 1079 rqstp->rq_respages = &rqstp->rq_pages[pnum]; 1080 rqstp->rq_next_page = rqstp->rq_respages + 1; 1081 1082 /* Now receive data */ 1083 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base); 1084 if (len >= 0) { 1085 svsk->sk_tcplen += len; 1086 svsk->sk_datalen += len; 1087 } 1088 if (len != want || !svc_sock_final_rec(svsk)) { 1089 svc_tcp_save_pages(svsk, rqstp); 1090 if (len < 0 && len != -EAGAIN) 1091 goto err_delete; 1092 if (len == want) 1093 svc_tcp_fragment_received(svsk); 1094 else 1095 dprintk("svc: incomplete TCP record (%d of %d)\n", 1096 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)), 1097 svc_sock_reclen(svsk)); 1098 goto err_noclose; 1099 } 1100 1101 if (svsk->sk_datalen < 8) { 1102 svsk->sk_datalen = 0; 1103 goto err_delete; /* client is nuts. */ 1104 } 1105 1106 rqstp->rq_arg.len = svsk->sk_datalen; 1107 rqstp->rq_arg.page_base = 0; 1108 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { 1109 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; 1110 rqstp->rq_arg.page_len = 0; 1111 } else 1112 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1113 1114 rqstp->rq_xprt_ctxt = NULL; 1115 rqstp->rq_prot = IPPROTO_TCP; 1116 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags)) 1117 set_bit(RQ_LOCAL, &rqstp->rq_flags); 1118 else 1119 clear_bit(RQ_LOCAL, &rqstp->rq_flags); 1120 1121 p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1122 calldir = p[1]; 1123 if (calldir) 1124 len = receive_cb_reply(svsk, rqstp); 1125 1126 /* Reset TCP read info */ 1127 svsk->sk_datalen = 0; 1128 svc_tcp_fragment_received(svsk); 1129 1130 if (len < 0) 1131 goto error; 1132 1133 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); 1134 if (serv->sv_stats) 1135 serv->sv_stats->nettcpcnt++; 1136 1137 return rqstp->rq_arg.len; 1138 1139 error: 1140 if (len != -EAGAIN) 1141 goto err_delete; 1142 dprintk("RPC: TCP recvfrom got EAGAIN\n"); 1143 return 0; 1144 err_delete: 1145 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", 1146 svsk->sk_xprt.xpt_server->sv_name, -len); 1147 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1148 err_noclose: 1149 return 0; /* record not complete */ 1150 } 1151 1152 /* 1153 * Send out data on TCP socket. 1154 */ 1155 static int svc_tcp_sendto(struct svc_rqst *rqstp) 1156 { 1157 struct xdr_buf *xbufp = &rqstp->rq_res; 1158 int sent; 1159 __be32 reclen; 1160 1161 /* Set up the first element of the reply kvec. 1162 * Any other kvecs that may be in use have been taken 1163 * care of by the server implementation itself. 1164 */ 1165 reclen = htonl(0x80000000|((xbufp->len ) - 4)); 1166 memcpy(xbufp->head[0].iov_base, &reclen, 4); 1167 1168 sent = svc_sendto(rqstp, &rqstp->rq_res); 1169 if (sent != xbufp->len) { 1170 printk(KERN_NOTICE 1171 "rpc-srv/tcp: %s: %s %d when sending %d bytes " 1172 "- shutting down socket\n", 1173 rqstp->rq_xprt->xpt_server->sv_name, 1174 (sent<0)?"got error":"sent only", 1175 sent, xbufp->len); 1176 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); 1177 svc_xprt_enqueue(rqstp->rq_xprt); 1178 sent = -EAGAIN; 1179 } 1180 return sent; 1181 } 1182 1183 /* 1184 * Setup response header. TCP has a 4B record length field. 1185 */ 1186 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) 1187 { 1188 struct kvec *resv = &rqstp->rq_res.head[0]; 1189 1190 /* tcp needs a space for the record length... */ 1191 svc_putnl(resv, 0); 1192 } 1193 1194 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, 1195 struct net *net, 1196 struct sockaddr *sa, int salen, 1197 int flags) 1198 { 1199 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1200 } 1201 1202 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1203 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, 1204 struct net *, struct sockaddr *, 1205 int, int); 1206 static void svc_bc_sock_free(struct svc_xprt *xprt); 1207 1208 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv, 1209 struct net *net, 1210 struct sockaddr *sa, int salen, 1211 int flags) 1212 { 1213 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1214 } 1215 1216 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt) 1217 { 1218 } 1219 1220 static struct svc_xprt_ops svc_tcp_bc_ops = { 1221 .xpo_create = svc_bc_tcp_create, 1222 .xpo_detach = svc_bc_tcp_sock_detach, 1223 .xpo_free = svc_bc_sock_free, 1224 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1225 .xpo_secure_port = svc_sock_secure_port, 1226 }; 1227 1228 static struct svc_xprt_class svc_tcp_bc_class = { 1229 .xcl_name = "tcp-bc", 1230 .xcl_owner = THIS_MODULE, 1231 .xcl_ops = &svc_tcp_bc_ops, 1232 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1233 }; 1234 1235 static void svc_init_bc_xprt_sock(void) 1236 { 1237 svc_reg_xprt_class(&svc_tcp_bc_class); 1238 } 1239 1240 static void svc_cleanup_bc_xprt_sock(void) 1241 { 1242 svc_unreg_xprt_class(&svc_tcp_bc_class); 1243 } 1244 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 1245 static void svc_init_bc_xprt_sock(void) 1246 { 1247 } 1248 1249 static void svc_cleanup_bc_xprt_sock(void) 1250 { 1251 } 1252 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1253 1254 static struct svc_xprt_ops svc_tcp_ops = { 1255 .xpo_create = svc_tcp_create, 1256 .xpo_recvfrom = svc_tcp_recvfrom, 1257 .xpo_sendto = svc_tcp_sendto, 1258 .xpo_release_rqst = svc_release_skb, 1259 .xpo_detach = svc_tcp_sock_detach, 1260 .xpo_free = svc_sock_free, 1261 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1262 .xpo_has_wspace = svc_tcp_has_wspace, 1263 .xpo_accept = svc_tcp_accept, 1264 .xpo_secure_port = svc_sock_secure_port, 1265 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt, 1266 }; 1267 1268 static struct svc_xprt_class svc_tcp_class = { 1269 .xcl_name = "tcp", 1270 .xcl_owner = THIS_MODULE, 1271 .xcl_ops = &svc_tcp_ops, 1272 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1273 .xcl_ident = XPRT_TRANSPORT_TCP, 1274 }; 1275 1276 void svc_init_xprt_sock(void) 1277 { 1278 svc_reg_xprt_class(&svc_tcp_class); 1279 svc_reg_xprt_class(&svc_udp_class); 1280 svc_init_bc_xprt_sock(); 1281 } 1282 1283 void svc_cleanup_xprt_sock(void) 1284 { 1285 svc_unreg_xprt_class(&svc_tcp_class); 1286 svc_unreg_xprt_class(&svc_udp_class); 1287 svc_cleanup_bc_xprt_sock(); 1288 } 1289 1290 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) 1291 { 1292 struct sock *sk = svsk->sk_sk; 1293 1294 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, 1295 &svsk->sk_xprt, serv); 1296 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 1297 if (sk->sk_state == TCP_LISTEN) { 1298 dprintk("setting up TCP socket for listening\n"); 1299 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); 1300 sk->sk_data_ready = svc_tcp_listen_data_ready; 1301 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 1302 } else { 1303 dprintk("setting up TCP socket for reading\n"); 1304 sk->sk_state_change = svc_tcp_state_change; 1305 sk->sk_data_ready = svc_data_ready; 1306 sk->sk_write_space = svc_write_space; 1307 1308 svsk->sk_reclen = 0; 1309 svsk->sk_tcplen = 0; 1310 svsk->sk_datalen = 0; 1311 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); 1312 1313 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1314 1315 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1316 switch (sk->sk_state) { 1317 case TCP_SYN_RECV: 1318 case TCP_ESTABLISHED: 1319 break; 1320 default: 1321 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1322 } 1323 } 1324 } 1325 1326 void svc_sock_update_bufs(struct svc_serv *serv) 1327 { 1328 /* 1329 * The number of server threads has changed. Update 1330 * rcvbuf and sndbuf accordingly on all sockets 1331 */ 1332 struct svc_sock *svsk; 1333 1334 spin_lock_bh(&serv->sv_lock); 1335 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) 1336 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1337 spin_unlock_bh(&serv->sv_lock); 1338 } 1339 EXPORT_SYMBOL_GPL(svc_sock_update_bufs); 1340 1341 /* 1342 * Initialize socket for RPC use and create svc_sock struct 1343 */ 1344 static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1345 struct socket *sock, 1346 int flags) 1347 { 1348 struct svc_sock *svsk; 1349 struct sock *inet; 1350 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1351 int err = 0; 1352 1353 dprintk("svc: svc_setup_socket %p\n", sock); 1354 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1355 if (!svsk) 1356 return ERR_PTR(-ENOMEM); 1357 1358 inet = sock->sk; 1359 1360 /* Register socket with portmapper */ 1361 if (pmap_register) 1362 err = svc_register(serv, sock_net(sock->sk), inet->sk_family, 1363 inet->sk_protocol, 1364 ntohs(inet_sk(inet)->inet_sport)); 1365 1366 if (err < 0) { 1367 kfree(svsk); 1368 return ERR_PTR(err); 1369 } 1370 1371 inet->sk_user_data = svsk; 1372 svsk->sk_sock = sock; 1373 svsk->sk_sk = inet; 1374 svsk->sk_ostate = inet->sk_state_change; 1375 svsk->sk_odata = inet->sk_data_ready; 1376 svsk->sk_owspace = inet->sk_write_space; 1377 1378 /* Initialize the socket */ 1379 if (sock->type == SOCK_DGRAM) 1380 svc_udp_init(svsk, serv); 1381 else 1382 svc_tcp_init(svsk, serv); 1383 1384 dprintk("svc: svc_setup_socket created %p (inet %p), " 1385 "listen %d close %d\n", 1386 svsk, svsk->sk_sk, 1387 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags), 1388 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1389 1390 return svsk; 1391 } 1392 1393 bool svc_alien_sock(struct net *net, int fd) 1394 { 1395 int err; 1396 struct socket *sock = sockfd_lookup(fd, &err); 1397 bool ret = false; 1398 1399 if (!sock) 1400 goto out; 1401 if (sock_net(sock->sk) != net) 1402 ret = true; 1403 sockfd_put(sock); 1404 out: 1405 return ret; 1406 } 1407 EXPORT_SYMBOL_GPL(svc_alien_sock); 1408 1409 /** 1410 * svc_addsock - add a listener socket to an RPC service 1411 * @serv: pointer to RPC service to which to add a new listener 1412 * @fd: file descriptor of the new listener 1413 * @name_return: pointer to buffer to fill in with name of listener 1414 * @len: size of the buffer 1415 * 1416 * Fills in socket name and returns positive length of name if successful. 1417 * Name is terminated with '\n'. On error, returns a negative errno 1418 * value. 1419 */ 1420 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, 1421 const size_t len) 1422 { 1423 int err = 0; 1424 struct socket *so = sockfd_lookup(fd, &err); 1425 struct svc_sock *svsk = NULL; 1426 struct sockaddr_storage addr; 1427 struct sockaddr *sin = (struct sockaddr *)&addr; 1428 int salen; 1429 1430 if (!so) 1431 return err; 1432 err = -EAFNOSUPPORT; 1433 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) 1434 goto out; 1435 err = -EPROTONOSUPPORT; 1436 if (so->sk->sk_protocol != IPPROTO_TCP && 1437 so->sk->sk_protocol != IPPROTO_UDP) 1438 goto out; 1439 err = -EISCONN; 1440 if (so->state > SS_UNCONNECTED) 1441 goto out; 1442 err = -ENOENT; 1443 if (!try_module_get(THIS_MODULE)) 1444 goto out; 1445 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); 1446 if (IS_ERR(svsk)) { 1447 module_put(THIS_MODULE); 1448 err = PTR_ERR(svsk); 1449 goto out; 1450 } 1451 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) 1452 svc_xprt_set_local(&svsk->sk_xprt, sin, salen); 1453 svc_add_new_perm_xprt(serv, &svsk->sk_xprt); 1454 return svc_one_sock_name(svsk, name_return, len); 1455 out: 1456 sockfd_put(so); 1457 return err; 1458 } 1459 EXPORT_SYMBOL_GPL(svc_addsock); 1460 1461 /* 1462 * Create socket for RPC service. 1463 */ 1464 static struct svc_xprt *svc_create_socket(struct svc_serv *serv, 1465 int protocol, 1466 struct net *net, 1467 struct sockaddr *sin, int len, 1468 int flags) 1469 { 1470 struct svc_sock *svsk; 1471 struct socket *sock; 1472 int error; 1473 int type; 1474 struct sockaddr_storage addr; 1475 struct sockaddr *newsin = (struct sockaddr *)&addr; 1476 int newlen; 1477 int family; 1478 int val; 1479 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 1480 1481 dprintk("svc: svc_create_socket(%s, %d, %s)\n", 1482 serv->sv_program->pg_name, protocol, 1483 __svc_print_addr(sin, buf, sizeof(buf))); 1484 1485 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1486 printk(KERN_WARNING "svc: only UDP and TCP " 1487 "sockets supported\n"); 1488 return ERR_PTR(-EINVAL); 1489 } 1490 1491 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1492 switch (sin->sa_family) { 1493 case AF_INET6: 1494 family = PF_INET6; 1495 break; 1496 case AF_INET: 1497 family = PF_INET; 1498 break; 1499 default: 1500 return ERR_PTR(-EINVAL); 1501 } 1502 1503 error = __sock_create(net, family, type, protocol, &sock, 1); 1504 if (error < 0) 1505 return ERR_PTR(error); 1506 1507 svc_reclassify_socket(sock); 1508 1509 /* 1510 * If this is an PF_INET6 listener, we want to avoid 1511 * getting requests from IPv4 remotes. Those should 1512 * be shunted to a PF_INET listener via rpcbind. 1513 */ 1514 val = 1; 1515 if (family == PF_INET6) 1516 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, 1517 (char *)&val, sizeof(val)); 1518 1519 if (type == SOCK_STREAM) 1520 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ 1521 error = kernel_bind(sock, sin, len); 1522 if (error < 0) 1523 goto bummer; 1524 1525 newlen = len; 1526 error = kernel_getsockname(sock, newsin, &newlen); 1527 if (error < 0) 1528 goto bummer; 1529 1530 if (protocol == IPPROTO_TCP) { 1531 if ((error = kernel_listen(sock, 64)) < 0) 1532 goto bummer; 1533 } 1534 1535 svsk = svc_setup_socket(serv, sock, flags); 1536 if (IS_ERR(svsk)) { 1537 error = PTR_ERR(svsk); 1538 goto bummer; 1539 } 1540 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); 1541 return (struct svc_xprt *)svsk; 1542 bummer: 1543 dprintk("svc: svc_create_socket error = %d\n", -error); 1544 sock_release(sock); 1545 return ERR_PTR(error); 1546 } 1547 1548 /* 1549 * Detach the svc_sock from the socket so that no 1550 * more callbacks occur. 1551 */ 1552 static void svc_sock_detach(struct svc_xprt *xprt) 1553 { 1554 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1555 struct sock *sk = svsk->sk_sk; 1556 1557 dprintk("svc: svc_sock_detach(%p)\n", svsk); 1558 1559 /* put back the old socket callbacks */ 1560 lock_sock(sk); 1561 sk->sk_state_change = svsk->sk_ostate; 1562 sk->sk_data_ready = svsk->sk_odata; 1563 sk->sk_write_space = svsk->sk_owspace; 1564 sk->sk_user_data = NULL; 1565 release_sock(sk); 1566 } 1567 1568 /* 1569 * Disconnect the socket, and reset the callbacks 1570 */ 1571 static void svc_tcp_sock_detach(struct svc_xprt *xprt) 1572 { 1573 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1574 1575 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); 1576 1577 svc_sock_detach(xprt); 1578 1579 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { 1580 svc_tcp_clear_pages(svsk); 1581 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); 1582 } 1583 } 1584 1585 /* 1586 * Free the svc_sock's socket resources and the svc_sock itself. 1587 */ 1588 static void svc_sock_free(struct svc_xprt *xprt) 1589 { 1590 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1591 dprintk("svc: svc_sock_free(%p)\n", svsk); 1592 1593 if (svsk->sk_sock->file) 1594 sockfd_put(svsk->sk_sock); 1595 else 1596 sock_release(svsk->sk_sock); 1597 kfree(svsk); 1598 } 1599 1600 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1601 /* 1602 * Create a back channel svc_xprt which shares the fore channel socket. 1603 */ 1604 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv, 1605 int protocol, 1606 struct net *net, 1607 struct sockaddr *sin, int len, 1608 int flags) 1609 { 1610 struct svc_sock *svsk; 1611 struct svc_xprt *xprt; 1612 1613 if (protocol != IPPROTO_TCP) { 1614 printk(KERN_WARNING "svc: only TCP sockets" 1615 " supported on shared back channel\n"); 1616 return ERR_PTR(-EINVAL); 1617 } 1618 1619 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1620 if (!svsk) 1621 return ERR_PTR(-ENOMEM); 1622 1623 xprt = &svsk->sk_xprt; 1624 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv); 1625 1626 serv->sv_bc_xprt = xprt; 1627 1628 return xprt; 1629 } 1630 1631 /* 1632 * Free a back channel svc_sock. 1633 */ 1634 static void svc_bc_sock_free(struct svc_xprt *xprt) 1635 { 1636 if (xprt) 1637 kfree(container_of(xprt, struct svc_sock, sk_xprt)); 1638 } 1639 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1640