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 61static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, 62 int flags); 63static void svc_udp_data_ready(struct sock *); 64static int svc_udp_recvfrom(struct svc_rqst *); 65static int svc_udp_sendto(struct svc_rqst *); 66static void svc_sock_detach(struct svc_xprt *); 67static void svc_tcp_sock_detach(struct svc_xprt *); 68static void svc_sock_free(struct svc_xprt *); 69 70static struct svc_xprt *svc_create_socket(struct svc_serv *, int, 71 struct net *, struct sockaddr *, 72 int, int); 73#if defined(CONFIG_SUNRPC_BACKCHANNEL) 74static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, 75 struct net *, struct sockaddr *, 76 int, int); 77static void svc_bc_sock_free(struct svc_xprt *xprt); 78#endif /* CONFIG_SUNRPC_BACKCHANNEL */ 79 80#ifdef CONFIG_DEBUG_LOCK_ALLOC 81static struct lock_class_key svc_key[2]; 82static struct lock_class_key svc_slock_key[2]; 83 84static 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 112static void svc_reclassify_socket(struct socket *sock) 113{ 114} 115#endif 116 117/* 118 * Release an skbuff after use 119 */ 120static 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 134union 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 141static 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 */ 175int 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 226out: 227 return len; 228} 229 230 231/* 232 * Generic sendto routine 233 */ 234static 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) 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 269out: 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 */ 280static 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#if IS_ENABLED(CONFIG_IPV6) 295 case PF_INET6: 296 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", 297 proto_name, 298 &sk->sk_v6_rcv_saddr, 299 inet_sk(sk)->inet_num); 300 break; 301#endif 302 default: 303 len = snprintf(buf, remaining, "*unknown-%d*\n", 304 sk->sk_family); 305 } 306 307 if (len >= remaining) { 308 *buf = '\0'; 309 return -ENAMETOOLONG; 310 } 311 return len; 312} 313 314/* 315 * Generic recvfrom routine. 316 */ 317static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, 318 int buflen) 319{ 320 struct svc_sock *svsk = 321 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 322 struct msghdr msg = { 323 .msg_flags = MSG_DONTWAIT, 324 }; 325 int len; 326 327 rqstp->rq_xprt_hlen = 0; 328 329 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 330 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, 331 msg.msg_flags); 332 /* If we read a full record, then assume there may be more 333 * data to read (stream based sockets only!) 334 */ 335 if (len == buflen) 336 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 337 338 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", 339 svsk, iov[0].iov_base, iov[0].iov_len, len); 340 return len; 341} 342 343static int svc_partial_recvfrom(struct svc_rqst *rqstp, 344 struct kvec *iov, int nr, 345 int buflen, unsigned int base) 346{ 347 size_t save_iovlen; 348 void *save_iovbase; 349 unsigned int i; 350 int ret; 351 352 if (base == 0) 353 return svc_recvfrom(rqstp, iov, nr, buflen); 354 355 for (i = 0; i < nr; i++) { 356 if (iov[i].iov_len > base) 357 break; 358 base -= iov[i].iov_len; 359 } 360 save_iovlen = iov[i].iov_len; 361 save_iovbase = iov[i].iov_base; 362 iov[i].iov_len -= base; 363 iov[i].iov_base += base; 364 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen); 365 iov[i].iov_len = save_iovlen; 366 iov[i].iov_base = save_iovbase; 367 return ret; 368} 369 370/* 371 * Set socket snd and rcv buffer lengths 372 */ 373static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, 374 unsigned int rcv) 375{ 376#if 0 377 mm_segment_t oldfs; 378 oldfs = get_fs(); set_fs(KERNEL_DS); 379 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, 380 (char*)&snd, sizeof(snd)); 381 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, 382 (char*)&rcv, sizeof(rcv)); 383#else 384 /* sock_setsockopt limits use to sysctl_?mem_max, 385 * which isn't acceptable. Until that is made conditional 386 * on not having CAP_SYS_RESOURCE or similar, we go direct... 387 * DaveM said I could! 388 */ 389 lock_sock(sock->sk); 390 sock->sk->sk_sndbuf = snd * 2; 391 sock->sk->sk_rcvbuf = rcv * 2; 392 sock->sk->sk_write_space(sock->sk); 393 release_sock(sock->sk); 394#endif 395} 396 397static int svc_sock_secure_port(struct svc_rqst *rqstp) 398{ 399 return svc_port_is_privileged(svc_addr(rqstp)); 400} 401 402/* 403 * INET callback when data has been received on the socket. 404 */ 405static void svc_udp_data_ready(struct sock *sk) 406{ 407 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 408 wait_queue_head_t *wq = sk_sleep(sk); 409 410 if (svsk) { 411 dprintk("svc: socket %p(inet %p), busy=%d\n", 412 svsk, sk, 413 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 414 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 415 svc_xprt_enqueue(&svsk->sk_xprt); 416 } 417 if (wq && waitqueue_active(wq)) 418 wake_up_interruptible(wq); 419} 420 421/* 422 * INET callback when space is newly available on the socket. 423 */ 424static void svc_write_space(struct sock *sk) 425{ 426 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 427 wait_queue_head_t *wq = sk_sleep(sk); 428 429 if (svsk) { 430 dprintk("svc: socket %p(inet %p), write_space busy=%d\n", 431 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 432 svc_xprt_enqueue(&svsk->sk_xprt); 433 } 434 435 if (wq && waitqueue_active(wq)) { 436 dprintk("RPC svc_write_space: someone sleeping on %p\n", 437 svsk); 438 wake_up_interruptible(wq); 439 } 440} 441 442static int svc_tcp_has_wspace(struct svc_xprt *xprt) 443{ 444 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 445 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 446 int required; 447 448 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) 449 return 1; 450 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; 451 if (sk_stream_wspace(svsk->sk_sk) >= required || 452 (sk_stream_min_wspace(svsk->sk_sk) == 0 && 453 atomic_read(&xprt->xpt_reserved) == 0)) 454 return 1; 455 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 456 return 0; 457} 458 459static void svc_tcp_write_space(struct sock *sk) 460{ 461 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 462 struct socket *sock = sk->sk_socket; 463 464 if (!sk_stream_is_writeable(sk) || !sock) 465 return; 466 if (!svsk || svc_tcp_has_wspace(&svsk->sk_xprt)) 467 clear_bit(SOCK_NOSPACE, &sock->flags); 468 svc_write_space(sk); 469} 470 471static void svc_tcp_adjust_wspace(struct svc_xprt *xprt) 472{ 473 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 474 475 if (svc_tcp_has_wspace(xprt)) 476 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 477} 478 479/* 480 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo 481 */ 482static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, 483 struct cmsghdr *cmh) 484{ 485 struct in_pktinfo *pki = CMSG_DATA(cmh); 486 struct sockaddr_in *daddr = svc_daddr_in(rqstp); 487 488 if (cmh->cmsg_type != IP_PKTINFO) 489 return 0; 490 491 daddr->sin_family = AF_INET; 492 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr; 493 return 1; 494} 495 496/* 497 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl 498 */ 499static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, 500 struct cmsghdr *cmh) 501{ 502 struct in6_pktinfo *pki = CMSG_DATA(cmh); 503 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp); 504 505 if (cmh->cmsg_type != IPV6_PKTINFO) 506 return 0; 507 508 daddr->sin6_family = AF_INET6; 509 daddr->sin6_addr = pki->ipi6_addr; 510 daddr->sin6_scope_id = pki->ipi6_ifindex; 511 return 1; 512} 513 514/* 515 * Copy the UDP datagram's destination address to the rqstp structure. 516 * The 'destination' address in this case is the address to which the 517 * peer sent the datagram, i.e. our local address. For multihomed 518 * hosts, this can change from msg to msg. Note that only the IP 519 * address changes, the port number should remain the same. 520 */ 521static int svc_udp_get_dest_address(struct svc_rqst *rqstp, 522 struct cmsghdr *cmh) 523{ 524 switch (cmh->cmsg_level) { 525 case SOL_IP: 526 return svc_udp_get_dest_address4(rqstp, cmh); 527 case SOL_IPV6: 528 return svc_udp_get_dest_address6(rqstp, cmh); 529 } 530 531 return 0; 532} 533 534/* 535 * Receive a datagram from a UDP socket. 536 */ 537static int svc_udp_recvfrom(struct svc_rqst *rqstp) 538{ 539 struct svc_sock *svsk = 540 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 541 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 542 struct sk_buff *skb; 543 union { 544 struct cmsghdr hdr; 545 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 546 } buffer; 547 struct cmsghdr *cmh = &buffer.hdr; 548 struct msghdr msg = { 549 .msg_name = svc_addr(rqstp), 550 .msg_control = cmh, 551 .msg_controllen = sizeof(buffer), 552 .msg_flags = MSG_DONTWAIT, 553 }; 554 size_t len; 555 int err; 556 557 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 558 /* udp sockets need large rcvbuf as all pending 559 * requests are still in that buffer. sndbuf must 560 * also be large enough that there is enough space 561 * for one reply per thread. We count all threads 562 * rather than threads in a particular pool, which 563 * provides an upper bound on the number of threads 564 * which will access the socket. 565 */ 566 svc_sock_setbufsize(svsk->sk_sock, 567 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 568 (serv->sv_nrthreads+3) * serv->sv_max_mesg); 569 570 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 571 skb = NULL; 572 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 573 0, 0, MSG_PEEK | MSG_DONTWAIT); 574 if (err >= 0) 575 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); 576 577 if (skb == NULL) { 578 if (err != -EAGAIN) { 579 /* possibly an icmp error */ 580 dprintk("svc: recvfrom returned error %d\n", -err); 581 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 582 } 583 return 0; 584 } 585 len = svc_addr_len(svc_addr(rqstp)); 586 rqstp->rq_addrlen = len; 587 if (skb->tstamp.tv64 == 0) { 588 skb->tstamp = ktime_get_real(); 589 /* Don't enable netstamp, sunrpc doesn't 590 need that much accuracy */ 591 } 592 svsk->sk_sk->sk_stamp = skb->tstamp; 593 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ 594 595 len = skb->len - sizeof(struct udphdr); 596 rqstp->rq_arg.len = len; 597 598 rqstp->rq_prot = IPPROTO_UDP; 599 600 if (!svc_udp_get_dest_address(rqstp, cmh)) { 601 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n", 602 cmh->cmsg_level, cmh->cmsg_type); 603 goto out_free; 604 } 605 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp)); 606 607 if (skb_is_nonlinear(skb)) { 608 /* we have to copy */ 609 local_bh_disable(); 610 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { 611 local_bh_enable(); 612 /* checksum error */ 613 goto out_free; 614 } 615 local_bh_enable(); 616 skb_free_datagram_locked(svsk->sk_sk, skb); 617 } else { 618 /* we can use it in-place */ 619 rqstp->rq_arg.head[0].iov_base = skb->data + 620 sizeof(struct udphdr); 621 rqstp->rq_arg.head[0].iov_len = len; 622 if (skb_checksum_complete(skb)) 623 goto out_free; 624 rqstp->rq_xprt_ctxt = skb; 625 } 626 627 rqstp->rq_arg.page_base = 0; 628 if (len <= rqstp->rq_arg.head[0].iov_len) { 629 rqstp->rq_arg.head[0].iov_len = len; 630 rqstp->rq_arg.page_len = 0; 631 rqstp->rq_respages = rqstp->rq_pages+1; 632 } else { 633 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 634 rqstp->rq_respages = rqstp->rq_pages + 1 + 635 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); 636 } 637 rqstp->rq_next_page = rqstp->rq_respages+1; 638 639 if (serv->sv_stats) 640 serv->sv_stats->netudpcnt++; 641 642 return len; 643out_free: 644 trace_kfree_skb(skb, svc_udp_recvfrom); 645 skb_free_datagram_locked(svsk->sk_sk, skb); 646 return 0; 647} 648 649static int 650svc_udp_sendto(struct svc_rqst *rqstp) 651{ 652 int error; 653 654 error = svc_sendto(rqstp, &rqstp->rq_res); 655 if (error == -ECONNREFUSED) 656 /* ICMP error on earlier request. */ 657 error = svc_sendto(rqstp, &rqstp->rq_res); 658 659 return error; 660} 661 662static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) 663{ 664} 665 666static int svc_udp_has_wspace(struct svc_xprt *xprt) 667{ 668 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 669 struct svc_serv *serv = xprt->xpt_server; 670 unsigned long required; 671 672 /* 673 * Set the SOCK_NOSPACE flag before checking the available 674 * sock space. 675 */ 676 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 677 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; 678 if (required*2 > sock_wspace(svsk->sk_sk)) 679 return 0; 680 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 681 return 1; 682} 683 684static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) 685{ 686 BUG(); 687 return NULL; 688} 689 690static struct svc_xprt *svc_udp_create(struct svc_serv *serv, 691 struct net *net, 692 struct sockaddr *sa, int salen, 693 int flags) 694{ 695 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); 696} 697 698static struct svc_xprt_ops svc_udp_ops = { 699 .xpo_create = svc_udp_create, 700 .xpo_recvfrom = svc_udp_recvfrom, 701 .xpo_sendto = svc_udp_sendto, 702 .xpo_release_rqst = svc_release_skb, 703 .xpo_detach = svc_sock_detach, 704 .xpo_free = svc_sock_free, 705 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, 706 .xpo_has_wspace = svc_udp_has_wspace, 707 .xpo_accept = svc_udp_accept, 708 .xpo_secure_port = svc_sock_secure_port, 709}; 710 711static struct svc_xprt_class svc_udp_class = { 712 .xcl_name = "udp", 713 .xcl_owner = THIS_MODULE, 714 .xcl_ops = &svc_udp_ops, 715 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, 716 .xcl_ident = XPRT_TRANSPORT_UDP, 717}; 718 719static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) 720{ 721 int err, level, optname, one = 1; 722 723 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class, 724 &svsk->sk_xprt, serv); 725 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 726 svsk->sk_sk->sk_data_ready = svc_udp_data_ready; 727 svsk->sk_sk->sk_write_space = svc_write_space; 728 729 /* initialise setting must have enough space to 730 * receive and respond to one request. 731 * svc_udp_recvfrom will re-adjust if necessary 732 */ 733 svc_sock_setbufsize(svsk->sk_sock, 734 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, 735 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); 736 737 /* data might have come in before data_ready set up */ 738 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 739 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 740 741 /* make sure we get destination address info */ 742 switch (svsk->sk_sk->sk_family) { 743 case AF_INET: 744 level = SOL_IP; 745 optname = IP_PKTINFO; 746 break; 747 case AF_INET6: 748 level = SOL_IPV6; 749 optname = IPV6_RECVPKTINFO; 750 break; 751 default: 752 BUG(); 753 } 754 err = kernel_setsockopt(svsk->sk_sock, level, optname, 755 (char *)&one, sizeof(one)); 756 dprintk("svc: kernel_setsockopt returned %d\n", err); 757} 758 759/* 760 * A data_ready event on a listening socket means there's a connection 761 * pending. Do not use state_change as a substitute for it. 762 */ 763static void svc_tcp_listen_data_ready(struct sock *sk) 764{ 765 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 766 wait_queue_head_t *wq; 767 768 dprintk("svc: socket %p TCP (listen) state change %d\n", 769 sk, sk->sk_state); 770 771 /* 772 * This callback may called twice when a new connection 773 * is established as a child socket inherits everything 774 * from a parent LISTEN socket. 775 * 1) data_ready method of the parent socket will be called 776 * when one of child sockets become ESTABLISHED. 777 * 2) data_ready method of the child socket may be called 778 * when it receives data before the socket is accepted. 779 * In case of 2, we should ignore it silently. 780 */ 781 if (sk->sk_state == TCP_LISTEN) { 782 if (svsk) { 783 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 784 svc_xprt_enqueue(&svsk->sk_xprt); 785 } else 786 printk("svc: socket %p: no user data\n", sk); 787 } 788 789 wq = sk_sleep(sk); 790 if (wq && waitqueue_active(wq)) 791 wake_up_interruptible_all(wq); 792} 793 794/* 795 * A state change on a connected socket means it's dying or dead. 796 */ 797static void svc_tcp_state_change(struct sock *sk) 798{ 799 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 800 wait_queue_head_t *wq = sk_sleep(sk); 801 802 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", 803 sk, sk->sk_state, sk->sk_user_data); 804 805 if (!svsk) 806 printk("svc: socket %p: no user data\n", sk); 807 else { 808 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 809 svc_xprt_enqueue(&svsk->sk_xprt); 810 } 811 if (wq && waitqueue_active(wq)) 812 wake_up_interruptible_all(wq); 813} 814 815static void svc_tcp_data_ready(struct sock *sk) 816{ 817 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 818 wait_queue_head_t *wq = sk_sleep(sk); 819 820 dprintk("svc: socket %p TCP data ready (svsk %p)\n", 821 sk, sk->sk_user_data); 822 if (svsk) { 823 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 824 svc_xprt_enqueue(&svsk->sk_xprt); 825 } 826 if (wq && waitqueue_active(wq)) 827 wake_up_interruptible(wq); 828} 829 830/* 831 * Accept a TCP connection 832 */ 833static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) 834{ 835 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 836 struct sockaddr_storage addr; 837 struct sockaddr *sin = (struct sockaddr *) &addr; 838 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 839 struct socket *sock = svsk->sk_sock; 840 struct socket *newsock; 841 struct svc_sock *newsvsk; 842 int err, slen; 843 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 844 845 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); 846 if (!sock) 847 return NULL; 848 849 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 850 err = kernel_accept(sock, &newsock, O_NONBLOCK); 851 if (err < 0) { 852 if (err == -ENOMEM) 853 printk(KERN_WARNING "%s: no more sockets!\n", 854 serv->sv_name); 855 else if (err != -EAGAIN) 856 net_warn_ratelimited("%s: accept failed (err %d)!\n", 857 serv->sv_name, -err); 858 return NULL; 859 } 860 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 861 862 err = kernel_getpeername(newsock, sin, &slen); 863 if (err < 0) { 864 net_warn_ratelimited("%s: peername failed (err %d)!\n", 865 serv->sv_name, -err); 866 goto failed; /* aborted connection or whatever */ 867 } 868 869 /* Ideally, we would want to reject connections from unauthorized 870 * hosts here, but when we get encryption, the IP of the host won't 871 * tell us anything. For now just warn about unpriv connections. 872 */ 873 if (!svc_port_is_privileged(sin)) { 874 dprintk("%s: connect from unprivileged port: %s\n", 875 serv->sv_name, 876 __svc_print_addr(sin, buf, sizeof(buf))); 877 } 878 dprintk("%s: connect from %s\n", serv->sv_name, 879 __svc_print_addr(sin, buf, sizeof(buf))); 880 881 /* make sure that a write doesn't block forever when 882 * low on memory 883 */ 884 newsock->sk->sk_sndtimeo = HZ*30; 885 886 newsvsk = svc_setup_socket(serv, newsock, 887 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)); 888 if (IS_ERR(newsvsk)) 889 goto failed; 890 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); 891 err = kernel_getsockname(newsock, sin, &slen); 892 if (unlikely(err < 0)) { 893 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); 894 slen = offsetof(struct sockaddr, sa_data); 895 } 896 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); 897 898 if (sock_is_loopback(newsock->sk)) 899 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 900 else 901 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags); 902 if (serv->sv_stats) 903 serv->sv_stats->nettcpconn++; 904 905 return &newsvsk->sk_xprt; 906 907failed: 908 sock_release(newsock); 909 return NULL; 910} 911 912static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 913{ 914 unsigned int i, len, npages; 915 916 if (svsk->sk_datalen == 0) 917 return 0; 918 len = svsk->sk_datalen; 919 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 920 for (i = 0; i < npages; i++) { 921 if (rqstp->rq_pages[i] != NULL) 922 put_page(rqstp->rq_pages[i]); 923 BUG_ON(svsk->sk_pages[i] == NULL); 924 rqstp->rq_pages[i] = svsk->sk_pages[i]; 925 svsk->sk_pages[i] = NULL; 926 } 927 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]); 928 return len; 929} 930 931static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp) 932{ 933 unsigned int i, len, npages; 934 935 if (svsk->sk_datalen == 0) 936 return; 937 len = svsk->sk_datalen; 938 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 939 for (i = 0; i < npages; i++) { 940 svsk->sk_pages[i] = rqstp->rq_pages[i]; 941 rqstp->rq_pages[i] = NULL; 942 } 943} 944 945static void svc_tcp_clear_pages(struct svc_sock *svsk) 946{ 947 unsigned int i, len, npages; 948 949 if (svsk->sk_datalen == 0) 950 goto out; 951 len = svsk->sk_datalen; 952 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 953 for (i = 0; i < npages; i++) { 954 if (svsk->sk_pages[i] == NULL) { 955 WARN_ON_ONCE(1); 956 continue; 957 } 958 put_page(svsk->sk_pages[i]); 959 svsk->sk_pages[i] = NULL; 960 } 961out: 962 svsk->sk_tcplen = 0; 963 svsk->sk_datalen = 0; 964} 965 966/* 967 * Receive fragment record header. 968 * If we haven't gotten the record length yet, get the next four bytes. 969 */ 970static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) 971{ 972 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 973 unsigned int want; 974 int len; 975 976 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { 977 struct kvec iov; 978 979 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; 980 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; 981 iov.iov_len = want; 982 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) 983 goto error; 984 svsk->sk_tcplen += len; 985 986 if (len < want) { 987 dprintk("svc: short recvfrom while reading record " 988 "length (%d of %d)\n", len, want); 989 return -EAGAIN; 990 } 991 992 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk)); 993 if (svc_sock_reclen(svsk) + svsk->sk_datalen > 994 serv->sv_max_mesg) { 995 net_notice_ratelimited("RPC: fragment too large: %d\n", 996 svc_sock_reclen(svsk)); 997 goto err_delete; 998 } 999 } 1000 1001 return svc_sock_reclen(svsk); 1002error: 1003 dprintk("RPC: TCP recv_record got %d\n", len); 1004 return len; 1005err_delete: 1006 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1007 return -EAGAIN; 1008} 1009 1010static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp) 1011{ 1012 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt; 1013 struct rpc_rqst *req = NULL; 1014 struct kvec *src, *dst; 1015 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1016 __be32 xid; 1017 __be32 calldir; 1018 1019 xid = *p++; 1020 calldir = *p; 1021 1022 if (!bc_xprt) 1023 return -EAGAIN; 1024 spin_lock_bh(&bc_xprt->transport_lock); 1025 req = xprt_lookup_rqst(bc_xprt, xid); 1026 if (!req) 1027 goto unlock_notfound; 1028 1029 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); 1030 /* 1031 * XXX!: cheating for now! Only copying HEAD. 1032 * But we know this is good enough for now (in fact, for any 1033 * callback reply in the forseeable future). 1034 */ 1035 dst = &req->rq_private_buf.head[0]; 1036 src = &rqstp->rq_arg.head[0]; 1037 if (dst->iov_len < src->iov_len) 1038 goto unlock_eagain; /* whatever; just giving up. */ 1039 memcpy(dst->iov_base, src->iov_base, src->iov_len); 1040 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len); 1041 rqstp->rq_arg.len = 0; 1042 spin_unlock_bh(&bc_xprt->transport_lock); 1043 return 0; 1044unlock_notfound: 1045 printk(KERN_NOTICE 1046 "%s: Got unrecognized reply: " 1047 "calldir 0x%x xpt_bc_xprt %p xid %08x\n", 1048 __func__, ntohl(calldir), 1049 bc_xprt, ntohl(xid)); 1050unlock_eagain: 1051 spin_unlock_bh(&bc_xprt->transport_lock); 1052 return -EAGAIN; 1053} 1054 1055static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len) 1056{ 1057 int i = 0; 1058 int t = 0; 1059 1060 while (t < len) { 1061 vec[i].iov_base = page_address(pages[i]); 1062 vec[i].iov_len = PAGE_SIZE; 1063 i++; 1064 t += PAGE_SIZE; 1065 } 1066 return i; 1067} 1068 1069static void svc_tcp_fragment_received(struct svc_sock *svsk) 1070{ 1071 /* If we have more data, signal svc_xprt_enqueue() to try again */ 1072 dprintk("svc: TCP %s record (%d bytes)\n", 1073 svc_sock_final_rec(svsk) ? "final" : "nonfinal", 1074 svc_sock_reclen(svsk)); 1075 svsk->sk_tcplen = 0; 1076 svsk->sk_reclen = 0; 1077} 1078 1079/* 1080 * Receive data from a TCP socket. 1081 */ 1082static int svc_tcp_recvfrom(struct svc_rqst *rqstp) 1083{ 1084 struct svc_sock *svsk = 1085 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 1086 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1087 int len; 1088 struct kvec *vec; 1089 unsigned int want, base; 1090 __be32 *p; 1091 __be32 calldir; 1092 int pnum; 1093 1094 dprintk("svc: tcp_recv %p data %d conn %d close %d\n", 1095 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), 1096 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), 1097 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1098 1099 len = svc_tcp_recv_record(svsk, rqstp); 1100 if (len < 0) 1101 goto error; 1102 1103 base = svc_tcp_restore_pages(svsk, rqstp); 1104 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr)); 1105 1106 vec = rqstp->rq_vec; 1107 1108 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], 1109 svsk->sk_datalen + want); 1110 1111 rqstp->rq_respages = &rqstp->rq_pages[pnum]; 1112 rqstp->rq_next_page = rqstp->rq_respages + 1; 1113 1114 /* Now receive data */ 1115 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base); 1116 if (len >= 0) { 1117 svsk->sk_tcplen += len; 1118 svsk->sk_datalen += len; 1119 } 1120 if (len != want || !svc_sock_final_rec(svsk)) { 1121 svc_tcp_save_pages(svsk, rqstp); 1122 if (len < 0 && len != -EAGAIN) 1123 goto err_delete; 1124 if (len == want) 1125 svc_tcp_fragment_received(svsk); 1126 else 1127 dprintk("svc: incomplete TCP record (%d of %d)\n", 1128 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)), 1129 svc_sock_reclen(svsk)); 1130 goto err_noclose; 1131 } 1132 1133 if (svsk->sk_datalen < 8) { 1134 svsk->sk_datalen = 0; 1135 goto err_delete; /* client is nuts. */ 1136 } 1137 1138 rqstp->rq_arg.len = svsk->sk_datalen; 1139 rqstp->rq_arg.page_base = 0; 1140 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { 1141 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; 1142 rqstp->rq_arg.page_len = 0; 1143 } else 1144 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1145 1146 rqstp->rq_xprt_ctxt = NULL; 1147 rqstp->rq_prot = IPPROTO_TCP; 1148 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags)) 1149 set_bit(RQ_LOCAL, &rqstp->rq_flags); 1150 else 1151 clear_bit(RQ_LOCAL, &rqstp->rq_flags); 1152 1153 p = (__be32 *)rqstp->rq_arg.head[0].iov_base; 1154 calldir = p[1]; 1155 if (calldir) 1156 len = receive_cb_reply(svsk, rqstp); 1157 1158 /* Reset TCP read info */ 1159 svsk->sk_datalen = 0; 1160 svc_tcp_fragment_received(svsk); 1161 1162 if (len < 0) 1163 goto error; 1164 1165 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); 1166 if (serv->sv_stats) 1167 serv->sv_stats->nettcpcnt++; 1168 1169 return rqstp->rq_arg.len; 1170 1171error: 1172 if (len != -EAGAIN) 1173 goto err_delete; 1174 dprintk("RPC: TCP recvfrom got EAGAIN\n"); 1175 return 0; 1176err_delete: 1177 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", 1178 svsk->sk_xprt.xpt_server->sv_name, -len); 1179 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1180err_noclose: 1181 return 0; /* record not complete */ 1182} 1183 1184/* 1185 * Send out data on TCP socket. 1186 */ 1187static int svc_tcp_sendto(struct svc_rqst *rqstp) 1188{ 1189 struct xdr_buf *xbufp = &rqstp->rq_res; 1190 int sent; 1191 __be32 reclen; 1192 1193 /* Set up the first element of the reply kvec. 1194 * Any other kvecs that may be in use have been taken 1195 * care of by the server implementation itself. 1196 */ 1197 reclen = htonl(0x80000000|((xbufp->len ) - 4)); 1198 memcpy(xbufp->head[0].iov_base, &reclen, 4); 1199 1200 sent = svc_sendto(rqstp, &rqstp->rq_res); 1201 if (sent != xbufp->len) { 1202 printk(KERN_NOTICE 1203 "rpc-srv/tcp: %s: %s %d when sending %d bytes " 1204 "- shutting down socket\n", 1205 rqstp->rq_xprt->xpt_server->sv_name, 1206 (sent<0)?"got error":"sent only", 1207 sent, xbufp->len); 1208 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); 1209 svc_xprt_enqueue(rqstp->rq_xprt); 1210 sent = -EAGAIN; 1211 } 1212 return sent; 1213} 1214 1215/* 1216 * Setup response header. TCP has a 4B record length field. 1217 */ 1218static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) 1219{ 1220 struct kvec *resv = &rqstp->rq_res.head[0]; 1221 1222 /* tcp needs a space for the record length... */ 1223 svc_putnl(resv, 0); 1224} 1225 1226static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, 1227 struct net *net, 1228 struct sockaddr *sa, int salen, 1229 int flags) 1230{ 1231 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1232} 1233 1234#if defined(CONFIG_SUNRPC_BACKCHANNEL) 1235static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int, 1236 struct net *, struct sockaddr *, 1237 int, int); 1238static void svc_bc_sock_free(struct svc_xprt *xprt); 1239 1240static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv, 1241 struct net *net, 1242 struct sockaddr *sa, int salen, 1243 int flags) 1244{ 1245 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1246} 1247 1248static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt) 1249{ 1250} 1251 1252static struct svc_xprt_ops svc_tcp_bc_ops = { 1253 .xpo_create = svc_bc_tcp_create, 1254 .xpo_detach = svc_bc_tcp_sock_detach, 1255 .xpo_free = svc_bc_sock_free, 1256 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1257 .xpo_secure_port = svc_sock_secure_port, 1258}; 1259 1260static struct svc_xprt_class svc_tcp_bc_class = { 1261 .xcl_name = "tcp-bc", 1262 .xcl_owner = THIS_MODULE, 1263 .xcl_ops = &svc_tcp_bc_ops, 1264 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1265}; 1266 1267static void svc_init_bc_xprt_sock(void) 1268{ 1269 svc_reg_xprt_class(&svc_tcp_bc_class); 1270} 1271 1272static void svc_cleanup_bc_xprt_sock(void) 1273{ 1274 svc_unreg_xprt_class(&svc_tcp_bc_class); 1275} 1276#else /* CONFIG_SUNRPC_BACKCHANNEL */ 1277static void svc_init_bc_xprt_sock(void) 1278{ 1279} 1280 1281static void svc_cleanup_bc_xprt_sock(void) 1282{ 1283} 1284#endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1285 1286static struct svc_xprt_ops svc_tcp_ops = { 1287 .xpo_create = svc_tcp_create, 1288 .xpo_recvfrom = svc_tcp_recvfrom, 1289 .xpo_sendto = svc_tcp_sendto, 1290 .xpo_release_rqst = svc_release_skb, 1291 .xpo_detach = svc_tcp_sock_detach, 1292 .xpo_free = svc_sock_free, 1293 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1294 .xpo_has_wspace = svc_tcp_has_wspace, 1295 .xpo_accept = svc_tcp_accept, 1296 .xpo_secure_port = svc_sock_secure_port, 1297 .xpo_adjust_wspace = svc_tcp_adjust_wspace, 1298}; 1299 1300static struct svc_xprt_class svc_tcp_class = { 1301 .xcl_name = "tcp", 1302 .xcl_owner = THIS_MODULE, 1303 .xcl_ops = &svc_tcp_ops, 1304 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1305 .xcl_ident = XPRT_TRANSPORT_TCP, 1306}; 1307 1308void svc_init_xprt_sock(void) 1309{ 1310 svc_reg_xprt_class(&svc_tcp_class); 1311 svc_reg_xprt_class(&svc_udp_class); 1312 svc_init_bc_xprt_sock(); 1313} 1314 1315void svc_cleanup_xprt_sock(void) 1316{ 1317 svc_unreg_xprt_class(&svc_tcp_class); 1318 svc_unreg_xprt_class(&svc_udp_class); 1319 svc_cleanup_bc_xprt_sock(); 1320} 1321 1322static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) 1323{ 1324 struct sock *sk = svsk->sk_sk; 1325 1326 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class, 1327 &svsk->sk_xprt, serv); 1328 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 1329 if (sk->sk_state == TCP_LISTEN) { 1330 dprintk("setting up TCP socket for listening\n"); 1331 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); 1332 sk->sk_data_ready = svc_tcp_listen_data_ready; 1333 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 1334 } else { 1335 dprintk("setting up TCP socket for reading\n"); 1336 sk->sk_state_change = svc_tcp_state_change; 1337 sk->sk_data_ready = svc_tcp_data_ready; 1338 sk->sk_write_space = svc_tcp_write_space; 1339 1340 svsk->sk_reclen = 0; 1341 svsk->sk_tcplen = 0; 1342 svsk->sk_datalen = 0; 1343 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages)); 1344 1345 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1346 1347 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1348 if (sk->sk_state != TCP_ESTABLISHED) 1349 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1350 } 1351} 1352 1353void svc_sock_update_bufs(struct svc_serv *serv) 1354{ 1355 /* 1356 * The number of server threads has changed. Update 1357 * rcvbuf and sndbuf accordingly on all sockets 1358 */ 1359 struct svc_sock *svsk; 1360 1361 spin_lock_bh(&serv->sv_lock); 1362 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) 1363 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1364 spin_unlock_bh(&serv->sv_lock); 1365} 1366EXPORT_SYMBOL_GPL(svc_sock_update_bufs); 1367 1368/* 1369 * Initialize socket for RPC use and create svc_sock struct 1370 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. 1371 */ 1372static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1373 struct socket *sock, 1374 int flags) 1375{ 1376 struct svc_sock *svsk; 1377 struct sock *inet; 1378 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1379 int err = 0; 1380 1381 dprintk("svc: svc_setup_socket %p\n", sock); 1382 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1383 if (!svsk) 1384 return ERR_PTR(-ENOMEM); 1385 1386 inet = sock->sk; 1387 1388 /* Register socket with portmapper */ 1389 if (pmap_register) 1390 err = svc_register(serv, sock_net(sock->sk), inet->sk_family, 1391 inet->sk_protocol, 1392 ntohs(inet_sk(inet)->inet_sport)); 1393 1394 if (err < 0) { 1395 kfree(svsk); 1396 return ERR_PTR(err); 1397 } 1398 1399 inet->sk_user_data = svsk; 1400 svsk->sk_sock = sock; 1401 svsk->sk_sk = inet; 1402 svsk->sk_ostate = inet->sk_state_change; 1403 svsk->sk_odata = inet->sk_data_ready; 1404 svsk->sk_owspace = inet->sk_write_space; 1405 1406 /* Initialize the socket */ 1407 if (sock->type == SOCK_DGRAM) 1408 svc_udp_init(svsk, serv); 1409 else { 1410 /* initialise setting must have enough space to 1411 * receive and respond to one request. 1412 */ 1413 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg, 1414 4 * serv->sv_max_mesg); 1415 svc_tcp_init(svsk, serv); 1416 } 1417 1418 dprintk("svc: svc_setup_socket created %p (inet %p)\n", 1419 svsk, svsk->sk_sk); 1420 1421 return svsk; 1422} 1423 1424bool svc_alien_sock(struct net *net, int fd) 1425{ 1426 int err; 1427 struct socket *sock = sockfd_lookup(fd, &err); 1428 bool ret = false; 1429 1430 if (!sock) 1431 goto out; 1432 if (sock_net(sock->sk) != net) 1433 ret = true; 1434 sockfd_put(sock); 1435out: 1436 return ret; 1437} 1438EXPORT_SYMBOL_GPL(svc_alien_sock); 1439 1440/** 1441 * svc_addsock - add a listener socket to an RPC service 1442 * @serv: pointer to RPC service to which to add a new listener 1443 * @fd: file descriptor of the new listener 1444 * @name_return: pointer to buffer to fill in with name of listener 1445 * @len: size of the buffer 1446 * 1447 * Fills in socket name and returns positive length of name if successful. 1448 * Name is terminated with '\n'. On error, returns a negative errno 1449 * value. 1450 */ 1451int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, 1452 const size_t len) 1453{ 1454 int err = 0; 1455 struct socket *so = sockfd_lookup(fd, &err); 1456 struct svc_sock *svsk = NULL; 1457 struct sockaddr_storage addr; 1458 struct sockaddr *sin = (struct sockaddr *)&addr; 1459 int salen; 1460 1461 if (!so) 1462 return err; 1463 err = -EAFNOSUPPORT; 1464 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) 1465 goto out; 1466 err = -EPROTONOSUPPORT; 1467 if (so->sk->sk_protocol != IPPROTO_TCP && 1468 so->sk->sk_protocol != IPPROTO_UDP) 1469 goto out; 1470 err = -EISCONN; 1471 if (so->state > SS_UNCONNECTED) 1472 goto out; 1473 err = -ENOENT; 1474 if (!try_module_get(THIS_MODULE)) 1475 goto out; 1476 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS); 1477 if (IS_ERR(svsk)) { 1478 module_put(THIS_MODULE); 1479 err = PTR_ERR(svsk); 1480 goto out; 1481 } 1482 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) 1483 svc_xprt_set_local(&svsk->sk_xprt, sin, salen); 1484 svc_add_new_perm_xprt(serv, &svsk->sk_xprt); 1485 return svc_one_sock_name(svsk, name_return, len); 1486out: 1487 sockfd_put(so); 1488 return err; 1489} 1490EXPORT_SYMBOL_GPL(svc_addsock); 1491 1492/* 1493 * Create socket for RPC service. 1494 */ 1495static struct svc_xprt *svc_create_socket(struct svc_serv *serv, 1496 int protocol, 1497 struct net *net, 1498 struct sockaddr *sin, int len, 1499 int flags) 1500{ 1501 struct svc_sock *svsk; 1502 struct socket *sock; 1503 int error; 1504 int type; 1505 struct sockaddr_storage addr; 1506 struct sockaddr *newsin = (struct sockaddr *)&addr; 1507 int newlen; 1508 int family; 1509 int val; 1510 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 1511 1512 dprintk("svc: svc_create_socket(%s, %d, %s)\n", 1513 serv->sv_program->pg_name, protocol, 1514 __svc_print_addr(sin, buf, sizeof(buf))); 1515 1516 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1517 printk(KERN_WARNING "svc: only UDP and TCP " 1518 "sockets supported\n"); 1519 return ERR_PTR(-EINVAL); 1520 } 1521 1522 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1523 switch (sin->sa_family) { 1524 case AF_INET6: 1525 family = PF_INET6; 1526 break; 1527 case AF_INET: 1528 family = PF_INET; 1529 break; 1530 default: 1531 return ERR_PTR(-EINVAL); 1532 } 1533 1534 error = __sock_create(net, family, type, protocol, &sock, 1); 1535 if (error < 0) 1536 return ERR_PTR(error); 1537 1538 svc_reclassify_socket(sock); 1539 1540 /* 1541 * If this is an PF_INET6 listener, we want to avoid 1542 * getting requests from IPv4 remotes. Those should 1543 * be shunted to a PF_INET listener via rpcbind. 1544 */ 1545 val = 1; 1546 if (family == PF_INET6) 1547 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, 1548 (char *)&val, sizeof(val)); 1549 1550 if (type == SOCK_STREAM) 1551 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */ 1552 error = kernel_bind(sock, sin, len); 1553 if (error < 0) 1554 goto bummer; 1555 1556 newlen = len; 1557 error = kernel_getsockname(sock, newsin, &newlen); 1558 if (error < 0) 1559 goto bummer; 1560 1561 if (protocol == IPPROTO_TCP) { 1562 if ((error = kernel_listen(sock, 64)) < 0) 1563 goto bummer; 1564 } 1565 1566 svsk = svc_setup_socket(serv, sock, flags); 1567 if (IS_ERR(svsk)) { 1568 error = PTR_ERR(svsk); 1569 goto bummer; 1570 } 1571 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); 1572 return (struct svc_xprt *)svsk; 1573bummer: 1574 dprintk("svc: svc_create_socket error = %d\n", -error); 1575 sock_release(sock); 1576 return ERR_PTR(error); 1577} 1578 1579/* 1580 * Detach the svc_sock from the socket so that no 1581 * more callbacks occur. 1582 */ 1583static void svc_sock_detach(struct svc_xprt *xprt) 1584{ 1585 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1586 struct sock *sk = svsk->sk_sk; 1587 wait_queue_head_t *wq; 1588 1589 dprintk("svc: svc_sock_detach(%p)\n", svsk); 1590 1591 /* put back the old socket callbacks */ 1592 sk->sk_state_change = svsk->sk_ostate; 1593 sk->sk_data_ready = svsk->sk_odata; 1594 sk->sk_write_space = svsk->sk_owspace; 1595 1596 wq = sk_sleep(sk); 1597 if (wq && waitqueue_active(wq)) 1598 wake_up_interruptible(wq); 1599} 1600 1601/* 1602 * Disconnect the socket, and reset the callbacks 1603 */ 1604static void svc_tcp_sock_detach(struct svc_xprt *xprt) 1605{ 1606 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1607 1608 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); 1609 1610 svc_sock_detach(xprt); 1611 1612 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) { 1613 svc_tcp_clear_pages(svsk); 1614 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); 1615 } 1616} 1617 1618/* 1619 * Free the svc_sock's socket resources and the svc_sock itself. 1620 */ 1621static void svc_sock_free(struct svc_xprt *xprt) 1622{ 1623 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1624 dprintk("svc: svc_sock_free(%p)\n", svsk); 1625 1626 if (svsk->sk_sock->file) 1627 sockfd_put(svsk->sk_sock); 1628 else 1629 sock_release(svsk->sk_sock); 1630 kfree(svsk); 1631} 1632 1633#if defined(CONFIG_SUNRPC_BACKCHANNEL) 1634/* 1635 * Create a back channel svc_xprt which shares the fore channel socket. 1636 */ 1637static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv, 1638 int protocol, 1639 struct net *net, 1640 struct sockaddr *sin, int len, 1641 int flags) 1642{ 1643 struct svc_sock *svsk; 1644 struct svc_xprt *xprt; 1645 1646 if (protocol != IPPROTO_TCP) { 1647 printk(KERN_WARNING "svc: only TCP sockets" 1648 " supported on shared back channel\n"); 1649 return ERR_PTR(-EINVAL); 1650 } 1651 1652 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1653 if (!svsk) 1654 return ERR_PTR(-ENOMEM); 1655 1656 xprt = &svsk->sk_xprt; 1657 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv); 1658 1659 serv->sv_bc_xprt = xprt; 1660 1661 return xprt; 1662} 1663 1664/* 1665 * Free a back channel svc_sock. 1666 */ 1667static void svc_bc_sock_free(struct svc_xprt *xprt) 1668{ 1669 if (xprt) 1670 kfree(container_of(xprt, struct svc_sock, sk_xprt)); 1671} 1672#endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1673