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 void svc_udp_data_ready(struct sock *, int); 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 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 74 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, 75 struct net *, struct sockaddr *, 76 int, int); 77 static void svc_bc_sock_free(struct svc_xprt *xprt); 78 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 79 80 #ifdef CONFIG_DEBUG_LOCK_ALLOC 81 static struct lock_class_key svc_key[2]; 82 static struct lock_class_key svc_slock_key[2]; 83 84 static void svc_reclassify_socket(struct socket *sock) 85 { 86 struct sock *sk = sock->sk; 87 88 WARN_ON_ONCE(sock_owned_by_user(sk)); 89 if (sock_owned_by_user(sk)) 90 return; 91 92 switch (sk->sk_family) { 93 case AF_INET: 94 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", 95 &svc_slock_key[0], 96 "sk_xprt.xpt_lock-AF_INET-NFSD", 97 &svc_key[0]); 98 break; 99 100 case AF_INET6: 101 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", 102 &svc_slock_key[1], 103 "sk_xprt.xpt_lock-AF_INET6-NFSD", 104 &svc_key[1]); 105 break; 106 107 default: 108 BUG(); 109 } 110 } 111 #else 112 static void svc_reclassify_socket(struct socket *sock) 113 { 114 } 115 #endif 116 117 /* 118 * Release an skbuff after use 119 */ 120 static void svc_release_skb(struct svc_rqst *rqstp) 121 { 122 struct sk_buff *skb = rqstp->rq_xprt_ctxt; 123 124 if (skb) { 125 struct svc_sock *svsk = 126 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 127 rqstp->rq_xprt_ctxt = NULL; 128 129 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); 130 skb_free_datagram_locked(svsk->sk_sk, skb); 131 } 132 } 133 134 union svc_pktinfo_u { 135 struct in_pktinfo pkti; 136 struct in6_pktinfo pkti6; 137 }; 138 #define SVC_PKTINFO_SPACE \ 139 CMSG_SPACE(sizeof(union svc_pktinfo_u)) 140 141 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) 142 { 143 struct svc_sock *svsk = 144 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 145 switch (svsk->sk_sk->sk_family) { 146 case AF_INET: { 147 struct in_pktinfo *pki = CMSG_DATA(cmh); 148 149 cmh->cmsg_level = SOL_IP; 150 cmh->cmsg_type = IP_PKTINFO; 151 pki->ipi_ifindex = 0; 152 pki->ipi_spec_dst.s_addr = 153 svc_daddr_in(rqstp)->sin_addr.s_addr; 154 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 155 } 156 break; 157 158 case AF_INET6: { 159 struct in6_pktinfo *pki = CMSG_DATA(cmh); 160 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 161 162 cmh->cmsg_level = SOL_IPV6; 163 cmh->cmsg_type = IPV6_PKTINFO; 164 pki->ipi6_ifindex = daddr->sin6_scope_id; 165 pki->ipi6_addr = daddr->sin6_addr; 166 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 167 } 168 break; 169 } 170 } 171 172 /* 173 * send routine intended to be shared by the fore- and back-channel 174 */ 175 int svc_send_common(struct socket *sock, struct xdr_buf *xdr, 176 struct page *headpage, unsigned long headoffset, 177 struct page *tailpage, unsigned long tailoffset) 178 { 179 int result; 180 int size; 181 struct page **ppage = xdr->pages; 182 size_t base = xdr->page_base; 183 unsigned int pglen = xdr->page_len; 184 unsigned int flags = MSG_MORE; 185 int slen; 186 int len = 0; 187 188 slen = xdr->len; 189 190 /* send head */ 191 if (slen == xdr->head[0].iov_len) 192 flags = 0; 193 len = kernel_sendpage(sock, headpage, headoffset, 194 xdr->head[0].iov_len, flags); 195 if (len != xdr->head[0].iov_len) 196 goto out; 197 slen -= xdr->head[0].iov_len; 198 if (slen == 0) 199 goto out; 200 201 /* send page data */ 202 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; 203 while (pglen > 0) { 204 if (slen == size) 205 flags = 0; 206 result = kernel_sendpage(sock, *ppage, base, size, flags); 207 if (result > 0) 208 len += result; 209 if (result != size) 210 goto out; 211 slen -= size; 212 pglen -= size; 213 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; 214 base = 0; 215 ppage++; 216 } 217 218 /* send tail */ 219 if (xdr->tail[0].iov_len) { 220 result = kernel_sendpage(sock, tailpage, tailoffset, 221 xdr->tail[0].iov_len, 0); 222 if (result > 0) 223 len += result; 224 } 225 226 out: 227 return len; 228 } 229 230 231 /* 232 * Generic sendto routine 233 */ 234 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) 235 { 236 struct svc_sock *svsk = 237 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 238 struct socket *sock = svsk->sk_sock; 239 union { 240 struct cmsghdr hdr; 241 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 242 } buffer; 243 struct cmsghdr *cmh = &buffer.hdr; 244 int len = 0; 245 unsigned long tailoff; 246 unsigned long headoff; 247 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 248 249 if (rqstp->rq_prot == IPPROTO_UDP) { 250 struct msghdr msg = { 251 .msg_name = &rqstp->rq_addr, 252 .msg_namelen = rqstp->rq_addrlen, 253 .msg_control = cmh, 254 .msg_controllen = sizeof(buffer), 255 .msg_flags = MSG_MORE, 256 }; 257 258 svc_set_cmsg_data(rqstp, cmh); 259 260 if (sock_sendmsg(sock, &msg, 0) < 0) 261 goto out; 262 } 263 264 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); 265 headoff = 0; 266 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, 267 rqstp->rq_respages[0], tailoff); 268 269 out: 270 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", 271 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, 272 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); 273 274 return len; 275 } 276 277 /* 278 * Report socket names for nfsdfs 279 */ 280 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) 281 { 282 const struct sock *sk = svsk->sk_sk; 283 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? 284 "udp" : "tcp"; 285 int len; 286 287 switch (sk->sk_family) { 288 case PF_INET: 289 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", 290 proto_name, 291 &inet_sk(sk)->inet_rcv_saddr, 292 inet_sk(sk)->inet_num); 293 break; 294 case PF_INET6: 295 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", 296 proto_name, 297 &inet6_sk(sk)->rcv_saddr, 298 inet_sk(sk)->inet_num); 299 break; 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 * Check input queue length 314 */ 315 static int svc_recv_available(struct svc_sock *svsk) 316 { 317 struct socket *sock = svsk->sk_sock; 318 int avail, err; 319 320 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); 321 322 return (err >= 0)? avail : err; 323 } 324 325 /* 326 * Generic recvfrom routine. 327 */ 328 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, 329 int buflen) 330 { 331 struct svc_sock *svsk = 332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 333 struct msghdr msg = { 334 .msg_flags = MSG_DONTWAIT, 335 }; 336 int len; 337 338 rqstp->rq_xprt_hlen = 0; 339 340 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, 341 msg.msg_flags); 342 343 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", 344 svsk, iov[0].iov_base, iov[0].iov_len, len); 345 return len; 346 } 347 348 static int svc_partial_recvfrom(struct svc_rqst *rqstp, 349 struct kvec *iov, int nr, 350 int buflen, unsigned int base) 351 { 352 size_t save_iovlen; 353 void *save_iovbase; 354 unsigned int i; 355 int ret; 356 357 if (base == 0) 358 return svc_recvfrom(rqstp, iov, nr, buflen); 359 360 for (i = 0; i < nr; i++) { 361 if (iov[i].iov_len > base) 362 break; 363 base -= iov[i].iov_len; 364 } 365 save_iovlen = iov[i].iov_len; 366 save_iovbase = iov[i].iov_base; 367 iov[i].iov_len -= base; 368 iov[i].iov_base += base; 369 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen); 370 iov[i].iov_len = save_iovlen; 371 iov[i].iov_base = save_iovbase; 372 return ret; 373 } 374 375 /* 376 * Set socket snd and rcv buffer lengths 377 */ 378 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, 379 unsigned int rcv) 380 { 381 #if 0 382 mm_segment_t oldfs; 383 oldfs = get_fs(); set_fs(KERNEL_DS); 384 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, 385 (char*)&snd, sizeof(snd)); 386 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, 387 (char*)&rcv, sizeof(rcv)); 388 #else 389 /* sock_setsockopt limits use to sysctl_?mem_max, 390 * which isn't acceptable. Until that is made conditional 391 * on not having CAP_SYS_RESOURCE or similar, we go direct... 392 * DaveM said I could! 393 */ 394 lock_sock(sock->sk); 395 sock->sk->sk_sndbuf = snd * 2; 396 sock->sk->sk_rcvbuf = rcv * 2; 397 sock->sk->sk_write_space(sock->sk); 398 release_sock(sock->sk); 399 #endif 400 } 401 /* 402 * INET callback when data has been received on the socket. 403 */ 404 static void svc_udp_data_ready(struct sock *sk, int count) 405 { 406 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 407 wait_queue_head_t *wq = sk_sleep(sk); 408 409 if (svsk) { 410 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", 411 svsk, sk, count, 412 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 413 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 414 svc_xprt_enqueue(&svsk->sk_xprt); 415 } 416 if (wq && waitqueue_active(wq)) 417 wake_up_interruptible(wq); 418 } 419 420 /* 421 * INET callback when space is newly available on the socket. 422 */ 423 static void svc_write_space(struct sock *sk) 424 { 425 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 426 wait_queue_head_t *wq = sk_sleep(sk); 427 428 if (svsk) { 429 dprintk("svc: socket %p(inet %p), write_space busy=%d\n", 430 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 431 svc_xprt_enqueue(&svsk->sk_xprt); 432 } 433 434 if (wq && waitqueue_active(wq)) { 435 dprintk("RPC svc_write_space: someone sleeping on %p\n", 436 svsk); 437 wake_up_interruptible(wq); 438 } 439 } 440 441 static void svc_tcp_write_space(struct sock *sk) 442 { 443 struct socket *sock = sk->sk_socket; 444 445 if (sk_stream_is_writeable(sk) && sock) 446 clear_bit(SOCK_NOSPACE, &sock->flags); 447 svc_write_space(sk); 448 } 449 450 /* 451 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo 452 */ 453 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, 454 struct cmsghdr *cmh) 455 { 456 struct in_pktinfo *pki = CMSG_DATA(cmh); 457 struct sockaddr_in *daddr = svc_daddr_in(rqstp); 458 459 if (cmh->cmsg_type != IP_PKTINFO) 460 return 0; 461 462 daddr->sin_family = AF_INET; 463 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; 464 return 1; 465 } 466 467 /* 468 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl 469 */ 470 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, 471 struct cmsghdr *cmh) 472 { 473 struct in6_pktinfo *pki = CMSG_DATA(cmh); 474 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 475 476 if (cmh->cmsg_type != IPV6_PKTINFO) 477 return 0; 478 479 daddr->sin6_family = AF_INET6; 480 daddr->sin6_addr = pki->ipi6_addr; 481 daddr->sin6_scope_id = pki->ipi6_ifindex; 482 return 1; 483 } 484 485 /* 486 * Copy the UDP datagram's destination address to the rqstp structure. 487 * The 'destination' address in this case is the address to which the 488 * peer sent the datagram, i.e. our local address. For multihomed 489 * hosts, this can change from msg to msg. Note that only the IP 490 * address changes, the port number should remain the same. 491 */ 492 static int svc_udp_get_dest_address(struct svc_rqst *rqstp, 493 struct cmsghdr *cmh) 494 { 495 switch (cmh->cmsg_level) { 496 case SOL_IP: 497 return svc_udp_get_dest_address4(rqstp, cmh); 498 case SOL_IPV6: 499 return svc_udp_get_dest_address6(rqstp, cmh); 500 } 501 502 return 0; 503 } 504 505 /* 506 * Receive a datagram from a UDP socket. 507 */ 508 static int svc_udp_recvfrom(struct svc_rqst *rqstp) 509 { 510 struct svc_sock *svsk = 511 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 512 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 513 struct sk_buff *skb; 514 union { 515 struct cmsghdr hdr; 516 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 517 } buffer; 518 struct cmsghdr *cmh = &buffer.hdr; 519 struct msghdr msg = { 520 .msg_name = svc_addr(rqstp), 521 .msg_control = cmh, 522 .msg_controllen = sizeof(buffer), 523 .msg_flags = MSG_DONTWAIT, 524 }; 525 size_t len; 526 int err; 527 528 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 529 /* udp sockets need large rcvbuf as all pending 530 * requests are still in that buffer. sndbuf must 531 * also be large enough that there is enough space 532 * for one reply per thread. We count all threads 533 * rather than threads in a particular pool, which 534 * provides an upper bound on the number of threads 535 * which will access the socket. 536 */ 537 svc_sock_setbufsize(svsk->sk_sock, 538 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 539 (serv->sv_nrthreads+3) * serv->sv_max_mesg); 540 541 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 542 skb = NULL; 543 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 544 0, 0, MSG_PEEK | MSG_DONTWAIT); 545 if (err >= 0) 546 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); 547 548 if (skb == NULL) { 549 if (err != -EAGAIN) { 550 /* possibly an icmp error */ 551 dprintk("svc: recvfrom returned error %d\n", -err); 552 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 553 } 554 return 0; 555 } 556 len = svc_addr_len(svc_addr(rqstp)); 557 rqstp->rq_addrlen = len; 558 if (skb->tstamp.tv64 == 0) { 559 skb->tstamp = ktime_get_real(); 560 /* Don't enable netstamp, sunrpc doesn't 561 need that much accuracy */ 562 } 563 svsk->sk_sk->sk_stamp = skb->tstamp; 564 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ 565 566 len = skb->len - sizeof(struct udphdr); 567 rqstp->rq_arg.len = len; 568 569 rqstp->rq_prot = IPPROTO_UDP; 570 571 if (!svc_udp_get_dest_address(rqstp, cmh)) { 572 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", 573 cmh->cmsg_level, cmh->cmsg_type); 574 goto out_free; 575 } 576 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); 577 578 if (skb_is_nonlinear(skb)) { 579 /* we have to copy */ 580 local_bh_disable(); 581 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { 582 local_bh_enable(); 583 /* checksum error */ 584 goto out_free; 585 } 586 local_bh_enable(); 587 skb_free_datagram_locked(svsk->sk_sk, skb); 588 } else { 589 /* we can use it in-place */ 590 rqstp->rq_arg.head[0].iov_base = skb->data + 591 sizeof(struct udphdr); 592 rqstp->rq_arg.head[0].iov_len = len; 593 if (skb_checksum_complete(skb)) 594 goto out_free; 595 rqstp->rq_xprt_ctxt = skb; 596 } 597 598 rqstp->rq_arg.page_base = 0; 599 if (len <= rqstp->rq_arg.head[0].iov_len) { 600 rqstp->rq_arg.head[0].iov_len = len; 601 rqstp->rq_arg.page_len = 0; 602 rqstp->rq_respages = rqstp->rq_pages+1; 603 } else { 604 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 605 rqstp->rq_respages = rqstp->rq_pages + 1 + 606 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); 607 } 608 rqstp->rq_next_page = rqstp->rq_respages+1; 609 610 if (serv->sv_stats) 611 serv->sv_stats->netudpcnt++; 612 613 return len; 614 out_free: 615 trace_kfree_skb(skb, svc_udp_recvfrom); 616 skb_free_datagram_locked(svsk->sk_sk, skb); 617 return 0; 618 } 619 620 static int 621 svc_udp_sendto(struct svc_rqst *rqstp) 622 { 623 int error; 624 625 error = svc_sendto(rqstp, &rqstp->rq_res); 626 if (error == -ECONNREFUSED) 627 /* ICMP error on earlier request. */ 628 error = svc_sendto(rqstp, &rqstp->rq_res); 629 630 return error; 631 } 632 633 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) 634 { 635 } 636 637 static int svc_udp_has_wspace(struct svc_xprt *xprt) 638 { 639 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 640 struct svc_serv *serv = xprt->xpt_server; 641 unsigned long required; 642 643 /* 644 * Set the SOCK_NOSPACE flag before checking the available 645 * sock space. 646 */ 647 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 648 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; 649 if (required*2 > sock_wspace(svsk->sk_sk)) 650 return 0; 651 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 652 return 1; 653 } 654 655 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) 656 { 657 BUG(); 658 return NULL; 659 } 660 661 static struct svc_xprt *svc_udp_create(struct svc_serv *serv, 662 struct net *net, 663 struct sockaddr *sa, int salen, 664 int flags) 665 { 666 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); 667 } 668 669 static struct svc_xprt_ops svc_udp_ops = { 670 .xpo_create = svc_udp_create, 671 .xpo_recvfrom = svc_udp_recvfrom, 672 .xpo_sendto = svc_udp_sendto, 673 .xpo_release_rqst = svc_release_skb, 674 .xpo_detach = svc_sock_detach, 675 .xpo_free = svc_sock_free, 676 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, 677 .xpo_has_wspace = svc_udp_has_wspace, 678 .xpo_accept = svc_udp_accept, 679 }; 680 681 static struct svc_xprt_class svc_udp_class = { 682 .xcl_name = "udp", 683 .xcl_owner = THIS_MODULE, 684 .xcl_ops = &svc_udp_ops, 685 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, 686 }; 687 688 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) 689 { 690 int err, level, optname, one = 1; 691 692 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, 693 &svsk->sk_xprt, serv); 694 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 695 svsk->sk_sk->sk_data_ready = svc_udp_data_ready; 696 svsk->sk_sk->sk_write_space = svc_write_space; 697 698 /* initialise setting must have enough space to 699 * receive and respond to one request. 700 * svc_udp_recvfrom will re-adjust if necessary 701 */ 702 svc_sock_setbufsize(svsk->sk_sock, 703 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, 704 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); 705 706 /* data might have come in before data_ready set up */ 707 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 708 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 709 710 /* make sure we get destination address info */ 711 switch (svsk->sk_sk->sk_family) { 712 case AF_INET: 713 level = SOL_IP; 714 optname = IP_PKTINFO; 715 break; 716 case AF_INET6: 717 level = SOL_IPV6; 718 optname = IPV6_RECVPKTINFO; 719 break; 720 default: 721 BUG(); 722 } 723 err = kernel_setsockopt(svsk->sk_sock, level, optname, 724 (char *)&one, sizeof(one)); 725 dprintk("svc: kernel_setsockopt returned %d\n", err); 726 } 727 728 /* 729 * A data_ready event on a listening socket means there's a connection 730 * pending. Do not use state_change as a substitute for it. 731 */ 732 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) 733 { 734 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 735 wait_queue_head_t *wq; 736 737 dprintk("svc: socket %p TCP (listen) state change %d\n", 738 sk, sk->sk_state); 739 740 /* 741 * This callback may called twice when a new connection 742 * is established as a child socket inherits everything 743 * from a parent LISTEN socket. 744 * 1) data_ready method of the parent socket will be called 745 * when one of child sockets become ESTABLISHED. 746 * 2) data_ready method of the child socket may be called 747 * when it receives data before the socket is accepted. 748 * In case of 2, we should ignore it silently. 749 */ 750 if (sk->sk_state == TCP_LISTEN) { 751 if (svsk) { 752 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 753 svc_xprt_enqueue(&svsk->sk_xprt); 754 } else 755 printk("svc: socket %p: no user data\n", sk); 756 } 757 758 wq = sk_sleep(sk); 759 if (wq && waitqueue_active(wq)) 760 wake_up_interruptible_all(wq); 761 } 762 763 /* 764 * A state change on a connected socket means it's dying or dead. 765 */ 766 static void svc_tcp_state_change(struct sock *sk) 767 { 768 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 769 wait_queue_head_t *wq = sk_sleep(sk); 770 771 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", 772 sk, sk->sk_state, sk->sk_user_data); 773 774 if (!svsk) 775 printk("svc: socket %p: no user data\n", sk); 776 else { 777 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 778 svc_xprt_enqueue(&svsk->sk_xprt); 779 } 780 if (wq && waitqueue_active(wq)) 781 wake_up_interruptible_all(wq); 782 } 783 784 static void svc_tcp_data_ready(struct sock *sk, int count) 785 { 786 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 787 wait_queue_head_t *wq = sk_sleep(sk); 788 789 dprintk("svc: socket %p TCP data ready (svsk %p)\n", 790 sk, sk->sk_user_data); 791 if (svsk) { 792 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 793 svc_xprt_enqueue(&svsk->sk_xprt); 794 } 795 if (wq && waitqueue_active(wq)) 796 wake_up_interruptible(wq); 797 } 798 799 /* 800 * Accept a TCP connection 801 */ 802 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) 803 { 804 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 805 struct sockaddr_storage addr; 806 struct sockaddr *sin = (struct sockaddr *) &addr; 807 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 808 struct socket *sock = svsk->sk_sock; 809 struct socket *newsock; 810 struct svc_sock *newsvsk; 811 int err, slen; 812 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 813 814 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); 815 if (!sock) 816 return NULL; 817 818 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 819 err = kernel_accept(sock, &newsock, O_NONBLOCK); 820 if (err < 0) { 821 if (err == -ENOMEM) 822 printk(KERN_WARNING "%s: no more sockets!\n", 823 serv->sv_name); 824 else if (err != -EAGAIN) 825 net_warn_ratelimited("%s: accept failed (err %d)!\n", 826 serv->sv_name, -err); 827 return NULL; 828 } 829 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 830 831 err = kernel_getpeername(newsock, sin, &slen); 832 if (err < 0) { 833 net_warn_ratelimited("%s: peername failed (err %d)!\n", 834 serv->sv_name, -err); 835 goto failed; /* aborted connection or whatever */ 836 } 837 838 /* Ideally, we would want to reject connections from unauthorized 839 * hosts here, but when we get encryption, the IP of the host won't 840 * tell us anything. For now just warn about unpriv connections. 841 */ 842 if (!svc_port_is_privileged(sin)) { 843 dprintk(KERN_WARNING 844 "%s: connect from unprivileged port: %s\n", 845 serv->sv_name, 846 __svc_print_addr(sin, buf, sizeof(buf))); 847 } 848 dprintk("%s: connect from %s\n", serv->sv_name, 849 __svc_print_addr(sin, buf, sizeof(buf))); 850 851 /* make sure that a write doesn't block forever when 852 * low on memory 853 */ 854 newsock->sk->sk_sndtimeo = HZ*30; 855 856 newsvsk = svc_setup_socket(serv, newsock, 857 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); 858 if (IS_ERR(newsvsk)) 859 goto failed; 860 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); 861 err = kernel_getsockname(newsock, sin, &slen); 862 if (unlikely(err < 0)) { 863 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); 864 slen = offsetof(struct sockaddr, sa_data); 865 } 866 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); 867 868 if (serv->sv_stats) 869 serv->sv_stats->nettcpconn++; 870 871 return &newsvsk->sk_xprt; 872 873 failed: 874 sock_release(newsock); 875 return NULL; 876 } 877 878 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 879 { 880 unsigned int i, len, npages; 881 882 if (svsk->sk_datalen == 0) 883 return 0; 884 len = svsk->sk_datalen; 885 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 886 for (i = 0; i < npages; i++) { 887 if (rqstp->rq_pages[i] != NULL) 888 put_page(rqstp->rq_pages[i]); 889 BUG_ON(svsk->sk_pages[i] == NULL); 890 rqstp->rq_pages[i] = svsk->sk_pages[i]; 891 svsk->sk_pages[i] = NULL; 892 } 893 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); 894 return len; 895 } 896 897 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 898 { 899 unsigned int i, len, npages; 900 901 if (svsk->sk_datalen == 0) 902 return; 903 len = svsk->sk_datalen; 904 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 905 for (i = 0; i < npages; i++) { 906 svsk->sk_pages[i] = rqstp->rq_pages[i]; 907 rqstp->rq_pages[i] = NULL; 908 } 909 } 910 911 static void svc_tcp_clear_pages(struct svc_sock *svsk) 912 { 913 unsigned int i, len, npages; 914 915 if (svsk->sk_datalen == 0) 916 goto out; 917 len = svsk->sk_datalen; 918 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 919 for (i = 0; i < npages; i++) { 920 if (svsk->sk_pages[i] == NULL) { 921 WARN_ON_ONCE(1); 922 continue; 923 } 924 put_page(svsk->sk_pages[i]); 925 svsk->sk_pages[i] = NULL; 926 } 927 out: 928 svsk->sk_tcplen = 0; 929 svsk->sk_datalen = 0; 930 } 931 932 /* 933 * Receive fragment record header. 934 * If we haven't gotten the record length yet, get the next four bytes. 935 */ 936 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) 937 { 938 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 939 unsigned int want; 940 int len; 941 942 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 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 req = xprt_lookup_rqst(bc_xprt, xid); 992 993 if (!req) { 994 printk(KERN_NOTICE 995 "%s: Got unrecognized reply: " 996 "calldir 0x%x xpt_bc_xprt %p xid %08x\n", 997 __func__, ntohl(calldir), 998 bc_xprt, xid); 999 return -EAGAIN; 1000 } 1001 1002 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); 1003 /* 1004 * XXX!: cheating for now! Only copying HEAD. 1005 * But we know this is good enough for now (in fact, for any 1006 * callback reply in the forseeable future). 1007 */ 1008 dst = &req->rq_private_buf.head[0]; 1009 src = &rqstp->rq_arg.head[0]; 1010 if (dst->iov_len < src->iov_len) 1011 return -EAGAIN; /* whatever; just giving up. */ 1012 memcpy(dst->iov_base, src->iov_base, src->iov_len); 1013 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); 1014 rqstp->rq_arg.len = 0; 1015 return 0; 1016 } 1017 1018 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len) 1019 { 1020 int i = 0; 1021 int t = 0; 1022 1023 while (t < len) { 1024 vec[i].iov_base = page_address(pages[i]); 1025 vec[i].iov_len = PAGE_SIZE; 1026 i++; 1027 t += PAGE_SIZE; 1028 } 1029 return i; 1030 } 1031 1032 static void svc_tcp_fragment_received(struct svc_sock *svsk) 1033 { 1034 /* If we have more data, signal svc_xprt_enqueue() to try again */ 1035 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr)) 1036 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1037 dprintk("svc: TCP %s record (%d bytes)\n", 1038 svc_sock_final_rec(svsk) ? "final" : "nonfinal", 1039 svc_sock_reclen(svsk)); 1040 svsk->sk_tcplen = 0; 1041 svsk->sk_reclen = 0; 1042 } 1043 1044 /* 1045 * Receive data from a TCP socket. 1046 */ 1047 static int svc_tcp_recvfrom(struct svc_rqst *rqstp) 1048 { 1049 struct svc_sock *svsk = 1050 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 1051 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1052 int len; 1053 struct kvec *vec; 1054 unsigned int want, base; 1055 __be32 *p; 1056 __be32 calldir; 1057 int pnum; 1058 1059 dprintk("svc: tcp_recv %p data %d conn %d close %d\n", 1060 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), 1061 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), 1062 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1063 1064 len = svc_tcp_recv_record(svsk, rqstp); 1065 if (len < 0) 1066 goto error; 1067 1068 base = svc_tcp_restore_pages(svsk, rqstp); 1069 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); 1070 1071 vec = rqstp->rq_vec; 1072 1073 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], 1074 svsk->sk_datalen + want); 1075 1076 rqstp->rq_respages = &rqstp->rq_pages[pnum]; 1077 rqstp->rq_next_page = rqstp->rq_respages + 1; 1078 1079 /* Now receive data */ 1080 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base); 1081 if (len >= 0) { 1082 svsk->sk_tcplen += len; 1083 svsk->sk_datalen += len; 1084 } 1085 if (len != want || !svc_sock_final_rec(svsk)) { 1086 svc_tcp_save_pages(svsk, rqstp); 1087 if (len < 0 && len != -EAGAIN) 1088 goto err_delete; 1089 if (len == want) 1090 svc_tcp_fragment_received(svsk); 1091 else 1092 dprintk("svc: incomplete TCP record (%d of %d)\n", 1093 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)), 1094 svc_sock_reclen(svsk)); 1095 goto err_noclose; 1096 } 1097 1098 if (svsk->sk_datalen < 8) { 1099 svsk->sk_datalen = 0; 1100 goto err_delete; /* client is nuts. */ 1101 } 1102 1103 rqstp->rq_arg.len = svsk->sk_datalen; 1104 rqstp->rq_arg.page_base = 0; 1105 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { 1106 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; 1107 rqstp->rq_arg.page_len = 0; 1108 } else 1109 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1110 1111 rqstp->rq_xprt_ctxt = NULL; 1112 rqstp->rq_prot = IPPROTO_TCP; 1113 1114 p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1115 calldir = p[1]; 1116 if (calldir) 1117 len = receive_cb_reply(svsk, rqstp); 1118 1119 /* Reset TCP read info */ 1120 svsk->sk_datalen = 0; 1121 svc_tcp_fragment_received(svsk); 1122 1123 if (len < 0) 1124 goto error; 1125 1126 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); 1127 if (serv->sv_stats) 1128 serv->sv_stats->nettcpcnt++; 1129 1130 return rqstp->rq_arg.len; 1131 1132 error: 1133 if (len != -EAGAIN) 1134 goto err_delete; 1135 dprintk("RPC: TCP recvfrom got EAGAIN\n"); 1136 return 0; 1137 err_delete: 1138 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", 1139 svsk->sk_xprt.xpt_server->sv_name, -len); 1140 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1141 err_noclose: 1142 return 0; /* record not complete */ 1143 } 1144 1145 /* 1146 * Send out data on TCP socket. 1147 */ 1148 static int svc_tcp_sendto(struct svc_rqst *rqstp) 1149 { 1150 struct xdr_buf *xbufp = &rqstp->rq_res; 1151 int sent; 1152 __be32 reclen; 1153 1154 /* Set up the first element of the reply kvec. 1155 * Any other kvecs that may be in use have been taken 1156 * care of by the server implementation itself. 1157 */ 1158 reclen = htonl(0x80000000|((xbufp->len ) - 4)); 1159 memcpy(xbufp->head[0].iov_base, &reclen, 4); 1160 1161 sent = svc_sendto(rqstp, &rqstp->rq_res); 1162 if (sent != xbufp->len) { 1163 printk(KERN_NOTICE 1164 "rpc-srv/tcp: %s: %s %d when sending %d bytes " 1165 "- shutting down socket\n", 1166 rqstp->rq_xprt->xpt_server->sv_name, 1167 (sent<0)?"got error":"sent only", 1168 sent, xbufp->len); 1169 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); 1170 svc_xprt_enqueue(rqstp->rq_xprt); 1171 sent = -EAGAIN; 1172 } 1173 return sent; 1174 } 1175 1176 /* 1177 * Setup response header. TCP has a 4B record length field. 1178 */ 1179 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) 1180 { 1181 struct kvec *resv = &rqstp->rq_res.head[0]; 1182 1183 /* tcp needs a space for the record length... */ 1184 svc_putnl(resv, 0); 1185 } 1186 1187 static int svc_tcp_has_wspace(struct svc_xprt *xprt) 1188 { 1189 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1190 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1191 int required; 1192 1193 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) 1194 return 1; 1195 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; 1196 if (sk_stream_wspace(svsk->sk_sk) >= required || 1197 (sk_stream_min_wspace(svsk->sk_sk) == 0 && 1198 atomic_read(&xprt->xpt_reserved) == 0)) 1199 return 1; 1200 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 1201 return 0; 1202 } 1203 1204 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, 1205 struct net *net, 1206 struct sockaddr *sa, int salen, 1207 int flags) 1208 { 1209 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1210 } 1211 1212 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1213 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, 1214 struct net *, struct sockaddr *, 1215 int, int); 1216 static void svc_bc_sock_free(struct svc_xprt *xprt); 1217 1218 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv, 1219 struct net *net, 1220 struct sockaddr *sa, int salen, 1221 int flags) 1222 { 1223 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1224 } 1225 1226 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt) 1227 { 1228 } 1229 1230 static struct svc_xprt_ops svc_tcp_bc_ops = { 1231 .xpo_create = svc_bc_tcp_create, 1232 .xpo_detach = svc_bc_tcp_sock_detach, 1233 .xpo_free = svc_bc_sock_free, 1234 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1235 }; 1236 1237 static struct svc_xprt_class svc_tcp_bc_class = { 1238 .xcl_name = "tcp-bc", 1239 .xcl_owner = THIS_MODULE, 1240 .xcl_ops = &svc_tcp_bc_ops, 1241 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1242 }; 1243 1244 static void svc_init_bc_xprt_sock(void) 1245 { 1246 svc_reg_xprt_class(&svc_tcp_bc_class); 1247 } 1248 1249 static void svc_cleanup_bc_xprt_sock(void) 1250 { 1251 svc_unreg_xprt_class(&svc_tcp_bc_class); 1252 } 1253 #else /* CONFIG_SUNRPC_BACKCHANNEL */ 1254 static void svc_init_bc_xprt_sock(void) 1255 { 1256 } 1257 1258 static void svc_cleanup_bc_xprt_sock(void) 1259 { 1260 } 1261 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1262 1263 static struct svc_xprt_ops svc_tcp_ops = { 1264 .xpo_create = svc_tcp_create, 1265 .xpo_recvfrom = svc_tcp_recvfrom, 1266 .xpo_sendto = svc_tcp_sendto, 1267 .xpo_release_rqst = svc_release_skb, 1268 .xpo_detach = svc_tcp_sock_detach, 1269 .xpo_free = svc_sock_free, 1270 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1271 .xpo_has_wspace = svc_tcp_has_wspace, 1272 .xpo_accept = svc_tcp_accept, 1273 }; 1274 1275 static struct svc_xprt_class svc_tcp_class = { 1276 .xcl_name = "tcp", 1277 .xcl_owner = THIS_MODULE, 1278 .xcl_ops = &svc_tcp_ops, 1279 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1280 }; 1281 1282 void svc_init_xprt_sock(void) 1283 { 1284 svc_reg_xprt_class(&svc_tcp_class); 1285 svc_reg_xprt_class(&svc_udp_class); 1286 svc_init_bc_xprt_sock(); 1287 } 1288 1289 void svc_cleanup_xprt_sock(void) 1290 { 1291 svc_unreg_xprt_class(&svc_tcp_class); 1292 svc_unreg_xprt_class(&svc_udp_class); 1293 svc_cleanup_bc_xprt_sock(); 1294 } 1295 1296 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) 1297 { 1298 struct sock *sk = svsk->sk_sk; 1299 1300 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, 1301 &svsk->sk_xprt, serv); 1302 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 1303 if (sk->sk_state == TCP_LISTEN) { 1304 dprintk("setting up TCP socket for listening\n"); 1305 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); 1306 sk->sk_data_ready = svc_tcp_listen_data_ready; 1307 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 1308 } else { 1309 dprintk("setting up TCP socket for reading\n"); 1310 sk->sk_state_change = svc_tcp_state_change; 1311 sk->sk_data_ready = svc_tcp_data_ready; 1312 sk->sk_write_space = svc_tcp_write_space; 1313 1314 svsk->sk_reclen = 0; 1315 svsk->sk_tcplen = 0; 1316 svsk->sk_datalen = 0; 1317 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); 1318 1319 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1320 1321 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1322 if (sk->sk_state != TCP_ESTABLISHED) 1323 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1324 } 1325 } 1326 1327 void svc_sock_update_bufs(struct svc_serv *serv) 1328 { 1329 /* 1330 * The number of server threads has changed. Update 1331 * rcvbuf and sndbuf accordingly on all sockets 1332 */ 1333 struct svc_sock *svsk; 1334 1335 spin_lock_bh(&serv->sv_lock); 1336 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) 1337 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1338 spin_unlock_bh(&serv->sv_lock); 1339 } 1340 EXPORT_SYMBOL_GPL(svc_sock_update_bufs); 1341 1342 /* 1343 * Initialize socket for RPC use and create svc_sock struct 1344 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. 1345 */ 1346 static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1347 struct socket *sock, 1348 int flags) 1349 { 1350 struct svc_sock *svsk; 1351 struct sock *inet; 1352 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1353 int err = 0; 1354 1355 dprintk("svc: svc_setup_socket %p\n", sock); 1356 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1357 if (!svsk) 1358 return ERR_PTR(-ENOMEM); 1359 1360 inet = sock->sk; 1361 1362 /* Register socket with portmapper */ 1363 if (pmap_register) 1364 err = svc_register(serv, sock_net(sock->sk), inet->sk_family, 1365 inet->sk_protocol, 1366 ntohs(inet_sk(inet)->inet_sport)); 1367 1368 if (err < 0) { 1369 kfree(svsk); 1370 return ERR_PTR(err); 1371 } 1372 1373 inet->sk_user_data = svsk; 1374 svsk->sk_sock = sock; 1375 svsk->sk_sk = inet; 1376 svsk->sk_ostate = inet->sk_state_change; 1377 svsk->sk_odata = inet->sk_data_ready; 1378 svsk->sk_owspace = inet->sk_write_space; 1379 1380 /* Initialize the socket */ 1381 if (sock->type == SOCK_DGRAM) 1382 svc_udp_init(svsk, serv); 1383 else { 1384 /* initialise setting must have enough space to 1385 * receive and respond to one request. 1386 */ 1387 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg, 1388 4 * serv->sv_max_mesg); 1389 svc_tcp_init(svsk, serv); 1390 } 1391 1392 dprintk("svc: svc_setup_socket created %p (inet %p)\n", 1393 svsk, svsk->sk_sk); 1394 1395 return svsk; 1396 } 1397 1398 /** 1399 * svc_addsock - add a listener socket to an RPC service 1400 * @serv: pointer to RPC service to which to add a new listener 1401 * @fd: file descriptor of the new listener 1402 * @name_return: pointer to buffer to fill in with name of listener 1403 * @len: size of the buffer 1404 * 1405 * Fills in socket name and returns positive length of name if successful. 1406 * Name is terminated with '\n'. On error, returns a negative errno 1407 * value. 1408 */ 1409 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, 1410 const size_t len) 1411 { 1412 int err = 0; 1413 struct socket *so = sockfd_lookup(fd, &err); 1414 struct svc_sock *svsk = NULL; 1415 struct sockaddr_storage addr; 1416 struct sockaddr *sin = (struct sockaddr *)&addr; 1417 int salen; 1418 1419 if (!so) 1420 return err; 1421 err = -EAFNOSUPPORT; 1422 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) 1423 goto out; 1424 err = -EPROTONOSUPPORT; 1425 if (so->sk->sk_protocol != IPPROTO_TCP && 1426 so->sk->sk_protocol != IPPROTO_UDP) 1427 goto out; 1428 err = -EISCONN; 1429 if (so->state > SS_UNCONNECTED) 1430 goto out; 1431 err = -ENOENT; 1432 if (!try_module_get(THIS_MODULE)) 1433 goto out; 1434 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); 1435 if (IS_ERR(svsk)) { 1436 module_put(THIS_MODULE); 1437 err = PTR_ERR(svsk); 1438 goto out; 1439 } 1440 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) 1441 svc_xprt_set_local(&svsk->sk_xprt, sin, salen); 1442 svc_add_new_perm_xprt(serv, &svsk->sk_xprt); 1443 return svc_one_sock_name(svsk, name_return, len); 1444 out: 1445 sockfd_put(so); 1446 return err; 1447 } 1448 EXPORT_SYMBOL_GPL(svc_addsock); 1449 1450 /* 1451 * Create socket for RPC service. 1452 */ 1453 static struct svc_xprt *svc_create_socket(struct svc_serv *serv, 1454 int protocol, 1455 struct net *net, 1456 struct sockaddr *sin, int len, 1457 int flags) 1458 { 1459 struct svc_sock *svsk; 1460 struct socket *sock; 1461 int error; 1462 int type; 1463 struct sockaddr_storage addr; 1464 struct sockaddr *newsin = (struct sockaddr *)&addr; 1465 int newlen; 1466 int family; 1467 int val; 1468 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 1469 1470 dprintk("svc: svc_create_socket(%s, %d, %s)\n", 1471 serv->sv_program->pg_name, protocol, 1472 __svc_print_addr(sin, buf, sizeof(buf))); 1473 1474 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1475 printk(KERN_WARNING "svc: only UDP and TCP " 1476 "sockets supported\n"); 1477 return ERR_PTR(-EINVAL); 1478 } 1479 1480 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1481 switch (sin->sa_family) { 1482 case AF_INET6: 1483 family = PF_INET6; 1484 break; 1485 case AF_INET: 1486 family = PF_INET; 1487 break; 1488 default: 1489 return ERR_PTR(-EINVAL); 1490 } 1491 1492 error = __sock_create(net, family, type, protocol, &sock, 1); 1493 if (error < 0) 1494 return ERR_PTR(error); 1495 1496 svc_reclassify_socket(sock); 1497 1498 /* 1499 * If this is an PF_INET6 listener, we want to avoid 1500 * getting requests from IPv4 remotes. Those should 1501 * be shunted to a PF_INET listener via rpcbind. 1502 */ 1503 val = 1; 1504 if (family == PF_INET6) 1505 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, 1506 (char *)&val, sizeof(val)); 1507 1508 if (type == SOCK_STREAM) 1509 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ 1510 error = kernel_bind(sock, sin, len); 1511 if (error < 0) 1512 goto bummer; 1513 1514 newlen = len; 1515 error = kernel_getsockname(sock, newsin, &newlen); 1516 if (error < 0) 1517 goto bummer; 1518 1519 if (protocol == IPPROTO_TCP) { 1520 if ((error = kernel_listen(sock, 64)) < 0) 1521 goto bummer; 1522 } 1523 1524 svsk = svc_setup_socket(serv, sock, flags); 1525 if (IS_ERR(svsk)) { 1526 error = PTR_ERR(svsk); 1527 goto bummer; 1528 } 1529 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); 1530 return (struct svc_xprt *)svsk; 1531 bummer: 1532 dprintk("svc: svc_create_socket error = %d\n", -error); 1533 sock_release(sock); 1534 return ERR_PTR(error); 1535 } 1536 1537 /* 1538 * Detach the svc_sock from the socket so that no 1539 * more callbacks occur. 1540 */ 1541 static void svc_sock_detach(struct svc_xprt *xprt) 1542 { 1543 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1544 struct sock *sk = svsk->sk_sk; 1545 wait_queue_head_t *wq; 1546 1547 dprintk("svc: svc_sock_detach(%p)\n", svsk); 1548 1549 /* put back the old socket callbacks */ 1550 sk->sk_state_change = svsk->sk_ostate; 1551 sk->sk_data_ready = svsk->sk_odata; 1552 sk->sk_write_space = svsk->sk_owspace; 1553 1554 wq = sk_sleep(sk); 1555 if (wq && waitqueue_active(wq)) 1556 wake_up_interruptible(wq); 1557 } 1558 1559 /* 1560 * Disconnect the socket, and reset the callbacks 1561 */ 1562 static void svc_tcp_sock_detach(struct svc_xprt *xprt) 1563 { 1564 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1565 1566 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); 1567 1568 svc_sock_detach(xprt); 1569 1570 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { 1571 svc_tcp_clear_pages(svsk); 1572 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); 1573 } 1574 } 1575 1576 /* 1577 * Free the svc_sock's socket resources and the svc_sock itself. 1578 */ 1579 static void svc_sock_free(struct svc_xprt *xprt) 1580 { 1581 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1582 dprintk("svc: svc_sock_free(%p)\n", svsk); 1583 1584 if (svsk->sk_sock->file) 1585 sockfd_put(svsk->sk_sock); 1586 else 1587 sock_release(svsk->sk_sock); 1588 kfree(svsk); 1589 } 1590 1591 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1592 /* 1593 * Create a back channel svc_xprt which shares the fore channel socket. 1594 */ 1595 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv, 1596 int protocol, 1597 struct net *net, 1598 struct sockaddr *sin, int len, 1599 int flags) 1600 { 1601 struct svc_sock *svsk; 1602 struct svc_xprt *xprt; 1603 1604 if (protocol != IPPROTO_TCP) { 1605 printk(KERN_WARNING "svc: only TCP sockets" 1606 " supported on shared back channel\n"); 1607 return ERR_PTR(-EINVAL); 1608 } 1609 1610 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1611 if (!svsk) 1612 return ERR_PTR(-ENOMEM); 1613 1614 xprt = &svsk->sk_xprt; 1615 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv); 1616 1617 serv->sv_bc_xprt = xprt; 1618 1619 return xprt; 1620 } 1621 1622 /* 1623 * Free a back channel svc_sock. 1624 */ 1625 static void svc_bc_sock_free(struct svc_xprt *xprt) 1626 { 1627 if (xprt) 1628 kfree(container_of(xprt, struct svc_sock, sk_xprt)); 1629 } 1630 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1631