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/errno.h> 25 #include <linux/fcntl.h> 26 #include <linux/net.h> 27 #include <linux/in.h> 28 #include <linux/inet.h> 29 #include <linux/udp.h> 30 #include <linux/tcp.h> 31 #include <linux/unistd.h> 32 #include <linux/slab.h> 33 #include <linux/netdevice.h> 34 #include <linux/skbuff.h> 35 #include <linux/file.h> 36 #include <linux/freezer.h> 37 #include <net/sock.h> 38 #include <net/checksum.h> 39 #include <net/ip.h> 40 #include <net/ipv6.h> 41 #include <net/tcp.h> 42 #include <net/tcp_states.h> 43 #include <asm/uaccess.h> 44 #include <asm/ioctls.h> 45 46 #include <linux/sunrpc/types.h> 47 #include <linux/sunrpc/clnt.h> 48 #include <linux/sunrpc/xdr.h> 49 #include <linux/sunrpc/msg_prot.h> 50 #include <linux/sunrpc/svcsock.h> 51 #include <linux/sunrpc/stats.h> 52 #include <linux/sunrpc/xprt.h> 53 54 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 55 56 57 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, 58 int *errp, int flags); 59 static void svc_udp_data_ready(struct sock *, int); 60 static int svc_udp_recvfrom(struct svc_rqst *); 61 static int svc_udp_sendto(struct svc_rqst *); 62 static void svc_sock_detach(struct svc_xprt *); 63 static void svc_tcp_sock_detach(struct svc_xprt *); 64 static void svc_sock_free(struct svc_xprt *); 65 66 static struct svc_xprt *svc_create_socket(struct svc_serv *, int, 67 struct net *, struct sockaddr *, 68 int, int); 69 #ifdef CONFIG_DEBUG_LOCK_ALLOC 70 static struct lock_class_key svc_key[2]; 71 static struct lock_class_key svc_slock_key[2]; 72 73 static void svc_reclassify_socket(struct socket *sock) 74 { 75 struct sock *sk = sock->sk; 76 BUG_ON(sock_owned_by_user(sk)); 77 switch (sk->sk_family) { 78 case AF_INET: 79 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", 80 &svc_slock_key[0], 81 "sk_xprt.xpt_lock-AF_INET-NFSD", 82 &svc_key[0]); 83 break; 84 85 case AF_INET6: 86 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", 87 &svc_slock_key[1], 88 "sk_xprt.xpt_lock-AF_INET6-NFSD", 89 &svc_key[1]); 90 break; 91 92 default: 93 BUG(); 94 } 95 } 96 #else 97 static void svc_reclassify_socket(struct socket *sock) 98 { 99 } 100 #endif 101 102 /* 103 * Release an skbuff after use 104 */ 105 static void svc_release_skb(struct svc_rqst *rqstp) 106 { 107 struct sk_buff *skb = rqstp->rq_xprt_ctxt; 108 109 if (skb) { 110 struct svc_sock *svsk = 111 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 112 rqstp->rq_xprt_ctxt = NULL; 113 114 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); 115 skb_free_datagram_locked(svsk->sk_sk, skb); 116 } 117 } 118 119 union svc_pktinfo_u { 120 struct in_pktinfo pkti; 121 struct in6_pktinfo pkti6; 122 }; 123 #define SVC_PKTINFO_SPACE \ 124 CMSG_SPACE(sizeof(union svc_pktinfo_u)) 125 126 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) 127 { 128 struct svc_sock *svsk = 129 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 130 switch (svsk->sk_sk->sk_family) { 131 case AF_INET: { 132 struct in_pktinfo *pki = CMSG_DATA(cmh); 133 134 cmh->cmsg_level = SOL_IP; 135 cmh->cmsg_type = IP_PKTINFO; 136 pki->ipi_ifindex = 0; 137 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; 138 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 139 } 140 break; 141 142 case AF_INET6: { 143 struct in6_pktinfo *pki = CMSG_DATA(cmh); 144 145 cmh->cmsg_level = SOL_IPV6; 146 cmh->cmsg_type = IPV6_PKTINFO; 147 pki->ipi6_ifindex = 0; 148 ipv6_addr_copy(&pki->ipi6_addr, 149 &rqstp->rq_daddr.addr6); 150 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 151 } 152 break; 153 } 154 } 155 156 /* 157 * send routine intended to be shared by the fore- and back-channel 158 */ 159 int svc_send_common(struct socket *sock, struct xdr_buf *xdr, 160 struct page *headpage, unsigned long headoffset, 161 struct page *tailpage, unsigned long tailoffset) 162 { 163 int result; 164 int size; 165 struct page **ppage = xdr->pages; 166 size_t base = xdr->page_base; 167 unsigned int pglen = xdr->page_len; 168 unsigned int flags = MSG_MORE; 169 int slen; 170 int len = 0; 171 172 slen = xdr->len; 173 174 /* send head */ 175 if (slen == xdr->head[0].iov_len) 176 flags = 0; 177 len = kernel_sendpage(sock, headpage, headoffset, 178 xdr->head[0].iov_len, flags); 179 if (len != xdr->head[0].iov_len) 180 goto out; 181 slen -= xdr->head[0].iov_len; 182 if (slen == 0) 183 goto out; 184 185 /* send page data */ 186 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; 187 while (pglen > 0) { 188 if (slen == size) 189 flags = 0; 190 result = kernel_sendpage(sock, *ppage, base, size, flags); 191 if (result > 0) 192 len += result; 193 if (result != size) 194 goto out; 195 slen -= size; 196 pglen -= size; 197 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; 198 base = 0; 199 ppage++; 200 } 201 202 /* send tail */ 203 if (xdr->tail[0].iov_len) { 204 result = kernel_sendpage(sock, tailpage, tailoffset, 205 xdr->tail[0].iov_len, 0); 206 if (result > 0) 207 len += result; 208 } 209 210 out: 211 return len; 212 } 213 214 215 /* 216 * Generic sendto routine 217 */ 218 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) 219 { 220 struct svc_sock *svsk = 221 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 222 struct socket *sock = svsk->sk_sock; 223 union { 224 struct cmsghdr hdr; 225 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 226 } buffer; 227 struct cmsghdr *cmh = &buffer.hdr; 228 int len = 0; 229 unsigned long tailoff; 230 unsigned long headoff; 231 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 232 233 if (rqstp->rq_prot == IPPROTO_UDP) { 234 struct msghdr msg = { 235 .msg_name = &rqstp->rq_addr, 236 .msg_namelen = rqstp->rq_addrlen, 237 .msg_control = cmh, 238 .msg_controllen = sizeof(buffer), 239 .msg_flags = MSG_MORE, 240 }; 241 242 svc_set_cmsg_data(rqstp, cmh); 243 244 if (sock_sendmsg(sock, &msg, 0) < 0) 245 goto out; 246 } 247 248 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1); 249 headoff = 0; 250 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff, 251 rqstp->rq_respages[0], tailoff); 252 253 out: 254 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", 255 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len, 256 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); 257 258 return len; 259 } 260 261 /* 262 * Report socket names for nfsdfs 263 */ 264 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining) 265 { 266 const struct sock *sk = svsk->sk_sk; 267 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ? 268 "udp" : "tcp"; 269 int len; 270 271 switch (sk->sk_family) { 272 case PF_INET: 273 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n", 274 proto_name, 275 &inet_sk(sk)->inet_rcv_saddr, 276 inet_sk(sk)->inet_num); 277 break; 278 case PF_INET6: 279 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n", 280 proto_name, 281 &inet6_sk(sk)->rcv_saddr, 282 inet_sk(sk)->inet_num); 283 break; 284 default: 285 len = snprintf(buf, remaining, "*unknown-%d*\n", 286 sk->sk_family); 287 } 288 289 if (len >= remaining) { 290 *buf = '\0'; 291 return -ENAMETOOLONG; 292 } 293 return len; 294 } 295 296 /** 297 * svc_sock_names - construct a list of listener names in a string 298 * @serv: pointer to RPC service 299 * @buf: pointer to a buffer to fill in with socket names 300 * @buflen: size of the buffer to be filled 301 * @toclose: pointer to '\0'-terminated C string containing the name 302 * of a listener to be closed 303 * 304 * Fills in @buf with a '\n'-separated list of names of listener 305 * sockets. If @toclose is not NULL, the socket named by @toclose 306 * is closed, and is not included in the output list. 307 * 308 * Returns positive length of the socket name string, or a negative 309 * errno value on error. 310 */ 311 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen, 312 const char *toclose) 313 { 314 struct svc_sock *svsk, *closesk = NULL; 315 int len = 0; 316 317 if (!serv) 318 return 0; 319 320 spin_lock_bh(&serv->sv_lock); 321 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) { 322 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len); 323 if (onelen < 0) { 324 len = onelen; 325 break; 326 } 327 if (toclose && strcmp(toclose, buf + len) == 0) 328 closesk = svsk; 329 else 330 len += onelen; 331 } 332 spin_unlock_bh(&serv->sv_lock); 333 334 if (closesk) 335 /* Should unregister with portmap, but you cannot 336 * unregister just one protocol... 337 */ 338 svc_close_xprt(&closesk->sk_xprt); 339 else if (toclose) 340 return -ENOENT; 341 return len; 342 } 343 EXPORT_SYMBOL_GPL(svc_sock_names); 344 345 /* 346 * Check input queue length 347 */ 348 static int svc_recv_available(struct svc_sock *svsk) 349 { 350 struct socket *sock = svsk->sk_sock; 351 int avail, err; 352 353 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); 354 355 return (err >= 0)? avail : err; 356 } 357 358 /* 359 * Generic recvfrom routine. 360 */ 361 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, 362 int buflen) 363 { 364 struct svc_sock *svsk = 365 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 366 struct msghdr msg = { 367 .msg_flags = MSG_DONTWAIT, 368 }; 369 int len; 370 371 rqstp->rq_xprt_hlen = 0; 372 373 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, 374 msg.msg_flags); 375 376 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", 377 svsk, iov[0].iov_base, iov[0].iov_len, len); 378 return len; 379 } 380 381 /* 382 * Set socket snd and rcv buffer lengths 383 */ 384 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, 385 unsigned int rcv) 386 { 387 #if 0 388 mm_segment_t oldfs; 389 oldfs = get_fs(); set_fs(KERNEL_DS); 390 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, 391 (char*)&snd, sizeof(snd)); 392 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, 393 (char*)&rcv, sizeof(rcv)); 394 #else 395 /* sock_setsockopt limits use to sysctl_?mem_max, 396 * which isn't acceptable. Until that is made conditional 397 * on not having CAP_SYS_RESOURCE or similar, we go direct... 398 * DaveM said I could! 399 */ 400 lock_sock(sock->sk); 401 sock->sk->sk_sndbuf = snd * 2; 402 sock->sk->sk_rcvbuf = rcv * 2; 403 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; 404 sock->sk->sk_write_space(sock->sk); 405 release_sock(sock->sk); 406 #endif 407 } 408 /* 409 * INET callback when data has been received on the socket. 410 */ 411 static void svc_udp_data_ready(struct sock *sk, int count) 412 { 413 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 414 415 if (svsk) { 416 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", 417 svsk, sk, count, 418 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 419 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 420 svc_xprt_enqueue(&svsk->sk_xprt); 421 } 422 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) 423 wake_up_interruptible(sk_sleep(sk)); 424 } 425 426 /* 427 * INET callback when space is newly available on the socket. 428 */ 429 static void svc_write_space(struct sock *sk) 430 { 431 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 432 433 if (svsk) { 434 dprintk("svc: socket %p(inet %p), write_space busy=%d\n", 435 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags)); 436 svc_xprt_enqueue(&svsk->sk_xprt); 437 } 438 439 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) { 440 dprintk("RPC svc_write_space: someone sleeping on %p\n", 441 svsk); 442 wake_up_interruptible(sk_sleep(sk)); 443 } 444 } 445 446 static void svc_tcp_write_space(struct sock *sk) 447 { 448 struct socket *sock = sk->sk_socket; 449 450 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock) 451 clear_bit(SOCK_NOSPACE, &sock->flags); 452 svc_write_space(sk); 453 } 454 455 /* 456 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo 457 */ 458 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp, 459 struct cmsghdr *cmh) 460 { 461 struct in_pktinfo *pki = CMSG_DATA(cmh); 462 if (cmh->cmsg_type != IP_PKTINFO) 463 return 0; 464 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; 465 return 1; 466 } 467 468 /* 469 * See net/ipv6/datagram.c : datagram_recv_ctl 470 */ 471 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp, 472 struct cmsghdr *cmh) 473 { 474 struct in6_pktinfo *pki = CMSG_DATA(cmh); 475 if (cmh->cmsg_type != IPV6_PKTINFO) 476 return 0; 477 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); 478 return 1; 479 } 480 481 /* 482 * Copy the UDP datagram's destination address to the rqstp structure. 483 * The 'destination' address in this case is the address to which the 484 * peer sent the datagram, i.e. our local address. For multihomed 485 * hosts, this can change from msg to msg. Note that only the IP 486 * address changes, the port number should remain the same. 487 */ 488 static int svc_udp_get_dest_address(struct svc_rqst *rqstp, 489 struct cmsghdr *cmh) 490 { 491 switch (cmh->cmsg_level) { 492 case SOL_IP: 493 return svc_udp_get_dest_address4(rqstp, cmh); 494 case SOL_IPV6: 495 return svc_udp_get_dest_address6(rqstp, cmh); 496 } 497 498 return 0; 499 } 500 501 /* 502 * Receive a datagram from a UDP socket. 503 */ 504 static int svc_udp_recvfrom(struct svc_rqst *rqstp) 505 { 506 struct svc_sock *svsk = 507 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 508 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 509 struct sk_buff *skb; 510 union { 511 struct cmsghdr hdr; 512 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 513 } buffer; 514 struct cmsghdr *cmh = &buffer.hdr; 515 struct msghdr msg = { 516 .msg_name = svc_addr(rqstp), 517 .msg_control = cmh, 518 .msg_controllen = sizeof(buffer), 519 .msg_flags = MSG_DONTWAIT, 520 }; 521 size_t len; 522 int err; 523 524 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 525 /* udp sockets need large rcvbuf as all pending 526 * requests are still in that buffer. sndbuf must 527 * also be large enough that there is enough space 528 * for one reply per thread. We count all threads 529 * rather than threads in a particular pool, which 530 * provides an upper bound on the number of threads 531 * which will access the socket. 532 */ 533 svc_sock_setbufsize(svsk->sk_sock, 534 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 535 (serv->sv_nrthreads+3) * serv->sv_max_mesg); 536 537 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 538 skb = NULL; 539 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 540 0, 0, MSG_PEEK | MSG_DONTWAIT); 541 if (err >= 0) 542 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); 543 544 if (skb == NULL) { 545 if (err != -EAGAIN) { 546 /* possibly an icmp error */ 547 dprintk("svc: recvfrom returned error %d\n", -err); 548 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 549 } 550 return -EAGAIN; 551 } 552 len = svc_addr_len(svc_addr(rqstp)); 553 if (len == 0) 554 return -EAFNOSUPPORT; 555 rqstp->rq_addrlen = len; 556 if (skb->tstamp.tv64 == 0) { 557 skb->tstamp = ktime_get_real(); 558 /* Don't enable netstamp, sunrpc doesn't 559 need that much accuracy */ 560 } 561 svsk->sk_sk->sk_stamp = skb->tstamp; 562 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */ 563 564 len = skb->len - sizeof(struct udphdr); 565 rqstp->rq_arg.len = len; 566 567 rqstp->rq_prot = IPPROTO_UDP; 568 569 if (!svc_udp_get_dest_address(rqstp, cmh)) { 570 if (net_ratelimit()) 571 printk(KERN_WARNING 572 "svc: received unknown control message %d/%d; " 573 "dropping RPC reply datagram\n", 574 cmh->cmsg_level, cmh->cmsg_type); 575 skb_free_datagram_locked(svsk->sk_sk, skb); 576 return 0; 577 } 578 579 if (skb_is_nonlinear(skb)) { 580 /* we have to copy */ 581 local_bh_disable(); 582 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { 583 local_bh_enable(); 584 /* checksum error */ 585 skb_free_datagram_locked(svsk->sk_sk, skb); 586 return 0; 587 } 588 local_bh_enable(); 589 skb_free_datagram_locked(svsk->sk_sk, skb); 590 } else { 591 /* we can use it in-place */ 592 rqstp->rq_arg.head[0].iov_base = skb->data + 593 sizeof(struct udphdr); 594 rqstp->rq_arg.head[0].iov_len = len; 595 if (skb_checksum_complete(skb)) { 596 skb_free_datagram_locked(svsk->sk_sk, skb); 597 return 0; 598 } 599 rqstp->rq_xprt_ctxt = skb; 600 } 601 602 rqstp->rq_arg.page_base = 0; 603 if (len <= rqstp->rq_arg.head[0].iov_len) { 604 rqstp->rq_arg.head[0].iov_len = len; 605 rqstp->rq_arg.page_len = 0; 606 rqstp->rq_respages = rqstp->rq_pages+1; 607 } else { 608 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 609 rqstp->rq_respages = rqstp->rq_pages + 1 + 610 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE); 611 } 612 613 if (serv->sv_stats) 614 serv->sv_stats->netudpcnt++; 615 616 return len; 617 } 618 619 static int 620 svc_udp_sendto(struct svc_rqst *rqstp) 621 { 622 int error; 623 624 error = svc_sendto(rqstp, &rqstp->rq_res); 625 if (error == -ECONNREFUSED) 626 /* ICMP error on earlier request. */ 627 error = svc_sendto(rqstp, &rqstp->rq_res); 628 629 return error; 630 } 631 632 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp) 633 { 634 } 635 636 static int svc_udp_has_wspace(struct svc_xprt *xprt) 637 { 638 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 639 struct svc_serv *serv = xprt->xpt_server; 640 unsigned long required; 641 642 /* 643 * Set the SOCK_NOSPACE flag before checking the available 644 * sock space. 645 */ 646 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 647 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg; 648 if (required*2 > sock_wspace(svsk->sk_sk)) 649 return 0; 650 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 651 return 1; 652 } 653 654 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt) 655 { 656 BUG(); 657 return NULL; 658 } 659 660 static struct svc_xprt *svc_udp_create(struct svc_serv *serv, 661 struct net *net, 662 struct sockaddr *sa, int salen, 663 int flags) 664 { 665 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags); 666 } 667 668 static struct svc_xprt_ops svc_udp_ops = { 669 .xpo_create = svc_udp_create, 670 .xpo_recvfrom = svc_udp_recvfrom, 671 .xpo_sendto = svc_udp_sendto, 672 .xpo_release_rqst = svc_release_skb, 673 .xpo_detach = svc_sock_detach, 674 .xpo_free = svc_sock_free, 675 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr, 676 .xpo_has_wspace = svc_udp_has_wspace, 677 .xpo_accept = svc_udp_accept, 678 }; 679 680 static struct svc_xprt_class svc_udp_class = { 681 .xcl_name = "udp", 682 .xcl_owner = THIS_MODULE, 683 .xcl_ops = &svc_udp_ops, 684 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP, 685 }; 686 687 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv) 688 { 689 int err, level, optname, one = 1; 690 691 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv); 692 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 693 svsk->sk_sk->sk_data_ready = svc_udp_data_ready; 694 svsk->sk_sk->sk_write_space = svc_write_space; 695 696 /* initialise setting must have enough space to 697 * receive and respond to one request. 698 * svc_udp_recvfrom will re-adjust if necessary 699 */ 700 svc_sock_setbufsize(svsk->sk_sock, 701 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, 702 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); 703 704 /* data might have come in before data_ready set up */ 705 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 706 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 707 708 /* make sure we get destination address info */ 709 switch (svsk->sk_sk->sk_family) { 710 case AF_INET: 711 level = SOL_IP; 712 optname = IP_PKTINFO; 713 break; 714 case AF_INET6: 715 level = SOL_IPV6; 716 optname = IPV6_RECVPKTINFO; 717 break; 718 default: 719 BUG(); 720 } 721 err = kernel_setsockopt(svsk->sk_sock, level, optname, 722 (char *)&one, sizeof(one)); 723 dprintk("svc: kernel_setsockopt returned %d\n", err); 724 } 725 726 /* 727 * A data_ready event on a listening socket means there's a connection 728 * pending. Do not use state_change as a substitute for it. 729 */ 730 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused) 731 { 732 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 733 734 dprintk("svc: socket %p TCP (listen) state change %d\n", 735 sk, sk->sk_state); 736 737 /* 738 * This callback may called twice when a new connection 739 * is established as a child socket inherits everything 740 * from a parent LISTEN socket. 741 * 1) data_ready method of the parent socket will be called 742 * when one of child sockets become ESTABLISHED. 743 * 2) data_ready method of the child socket may be called 744 * when it receives data before the socket is accepted. 745 * In case of 2, we should ignore it silently. 746 */ 747 if (sk->sk_state == TCP_LISTEN) { 748 if (svsk) { 749 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 750 svc_xprt_enqueue(&svsk->sk_xprt); 751 } else 752 printk("svc: socket %p: no user data\n", sk); 753 } 754 755 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) 756 wake_up_interruptible_all(sk_sleep(sk)); 757 } 758 759 /* 760 * A state change on a connected socket means it's dying or dead. 761 */ 762 static void svc_tcp_state_change(struct sock *sk) 763 { 764 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 765 766 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", 767 sk, sk->sk_state, sk->sk_user_data); 768 769 if (!svsk) 770 printk("svc: socket %p: no user data\n", sk); 771 else { 772 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 773 svc_xprt_enqueue(&svsk->sk_xprt); 774 } 775 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) 776 wake_up_interruptible_all(sk_sleep(sk)); 777 } 778 779 static void svc_tcp_data_ready(struct sock *sk, int count) 780 { 781 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 782 783 dprintk("svc: socket %p TCP data ready (svsk %p)\n", 784 sk, sk->sk_user_data); 785 if (svsk) { 786 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 787 svc_xprt_enqueue(&svsk->sk_xprt); 788 } 789 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) 790 wake_up_interruptible(sk_sleep(sk)); 791 } 792 793 /* 794 * Accept a TCP connection 795 */ 796 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt) 797 { 798 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 799 struct sockaddr_storage addr; 800 struct sockaddr *sin = (struct sockaddr *) &addr; 801 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 802 struct socket *sock = svsk->sk_sock; 803 struct socket *newsock; 804 struct svc_sock *newsvsk; 805 int err, slen; 806 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 807 808 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); 809 if (!sock) 810 return NULL; 811 812 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 813 err = kernel_accept(sock, &newsock, O_NONBLOCK); 814 if (err < 0) { 815 if (err == -ENOMEM) 816 printk(KERN_WARNING "%s: no more sockets!\n", 817 serv->sv_name); 818 else if (err != -EAGAIN && net_ratelimit()) 819 printk(KERN_WARNING "%s: accept failed (err %d)!\n", 820 serv->sv_name, -err); 821 return NULL; 822 } 823 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 824 825 err = kernel_getpeername(newsock, sin, &slen); 826 if (err < 0) { 827 if (net_ratelimit()) 828 printk(KERN_WARNING "%s: peername failed (err %d)!\n", 829 serv->sv_name, -err); 830 goto failed; /* aborted connection or whatever */ 831 } 832 833 /* Ideally, we would want to reject connections from unauthorized 834 * hosts here, but when we get encryption, the IP of the host won't 835 * tell us anything. For now just warn about unpriv connections. 836 */ 837 if (!svc_port_is_privileged(sin)) { 838 dprintk(KERN_WARNING 839 "%s: connect from unprivileged port: %s\n", 840 serv->sv_name, 841 __svc_print_addr(sin, buf, sizeof(buf))); 842 } 843 dprintk("%s: connect from %s\n", serv->sv_name, 844 __svc_print_addr(sin, buf, sizeof(buf))); 845 846 /* make sure that a write doesn't block forever when 847 * low on memory 848 */ 849 newsock->sk->sk_sndtimeo = HZ*30; 850 851 if (!(newsvsk = svc_setup_socket(serv, newsock, &err, 852 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) 853 goto failed; 854 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen); 855 err = kernel_getsockname(newsock, sin, &slen); 856 if (unlikely(err < 0)) { 857 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); 858 slen = offsetof(struct sockaddr, sa_data); 859 } 860 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen); 861 862 if (serv->sv_stats) 863 serv->sv_stats->nettcpconn++; 864 865 return &newsvsk->sk_xprt; 866 867 failed: 868 sock_release(newsock); 869 return NULL; 870 } 871 872 /* 873 * Receive data. 874 * If we haven't gotten the record length yet, get the next four bytes. 875 * Otherwise try to gobble up as much as possible up to the complete 876 * record length. 877 */ 878 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp) 879 { 880 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 881 int len; 882 883 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) 884 /* sndbuf needs to have room for one request 885 * per thread, otherwise we can stall even when the 886 * network isn't a bottleneck. 887 * 888 * We count all threads rather than threads in a 889 * particular pool, which provides an upper bound 890 * on the number of threads which will access the socket. 891 * 892 * rcvbuf just needs to be able to hold a few requests. 893 * Normally they will be removed from the queue 894 * as soon a a complete request arrives. 895 */ 896 svc_sock_setbufsize(svsk->sk_sock, 897 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 898 3 * serv->sv_max_mesg); 899 900 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 901 902 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) { 903 int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen; 904 struct kvec iov; 905 906 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; 907 iov.iov_len = want; 908 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) 909 goto error; 910 svsk->sk_tcplen += len; 911 912 if (len < want) { 913 dprintk("svc: short recvfrom while reading record " 914 "length (%d of %d)\n", len, want); 915 goto err_again; /* record header not complete */ 916 } 917 918 svsk->sk_reclen = ntohl(svsk->sk_reclen); 919 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) { 920 /* FIXME: technically, a record can be fragmented, 921 * and non-terminal fragments will not have the top 922 * bit set in the fragment length header. 923 * But apparently no known nfs clients send fragmented 924 * records. */ 925 if (net_ratelimit()) 926 printk(KERN_NOTICE "RPC: multiple fragments " 927 "per record not supported\n"); 928 goto err_delete; 929 } 930 931 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK; 932 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); 933 if (svsk->sk_reclen > serv->sv_max_mesg) { 934 if (net_ratelimit()) 935 printk(KERN_NOTICE "RPC: " 936 "fragment too large: 0x%08lx\n", 937 (unsigned long)svsk->sk_reclen); 938 goto err_delete; 939 } 940 } 941 942 /* Check whether enough data is available */ 943 len = svc_recv_available(svsk); 944 if (len < 0) 945 goto error; 946 947 if (len < svsk->sk_reclen) { 948 dprintk("svc: incomplete TCP record (%d of %d)\n", 949 len, svsk->sk_reclen); 950 goto err_again; /* record not complete */ 951 } 952 len = svsk->sk_reclen; 953 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 954 955 return len; 956 error: 957 if (len == -EAGAIN) 958 dprintk("RPC: TCP recv_record got EAGAIN\n"); 959 return len; 960 err_delete: 961 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 962 err_again: 963 return -EAGAIN; 964 } 965 966 static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp, 967 struct rpc_rqst **reqpp, struct kvec *vec) 968 { 969 struct rpc_rqst *req = NULL; 970 u32 *p; 971 u32 xid; 972 u32 calldir; 973 int len; 974 975 len = svc_recvfrom(rqstp, vec, 1, 8); 976 if (len < 0) 977 goto error; 978 979 p = (u32 *)rqstp->rq_arg.head[0].iov_base; 980 xid = *p++; 981 calldir = *p; 982 983 if (calldir == 0) { 984 /* REQUEST is the most common case */ 985 vec[0] = rqstp->rq_arg.head[0]; 986 } else { 987 /* REPLY */ 988 if (svsk->sk_bc_xprt) 989 req = xprt_lookup_rqst(svsk->sk_bc_xprt, xid); 990 991 if (!req) { 992 printk(KERN_NOTICE 993 "%s: Got unrecognized reply: " 994 "calldir 0x%x sk_bc_xprt %p xid %08x\n", 995 __func__, ntohl(calldir), 996 svsk->sk_bc_xprt, xid); 997 vec[0] = rqstp->rq_arg.head[0]; 998 goto out; 999 } 1000 1001 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, 1002 sizeof(struct xdr_buf)); 1003 /* copy the xid and call direction */ 1004 memcpy(req->rq_private_buf.head[0].iov_base, 1005 rqstp->rq_arg.head[0].iov_base, 8); 1006 vec[0] = req->rq_private_buf.head[0]; 1007 } 1008 out: 1009 vec[0].iov_base += 8; 1010 vec[0].iov_len -= 8; 1011 len = svsk->sk_reclen - 8; 1012 error: 1013 *reqpp = req; 1014 return len; 1015 } 1016 1017 /* 1018 * Receive data from a TCP socket. 1019 */ 1020 static int svc_tcp_recvfrom(struct svc_rqst *rqstp) 1021 { 1022 struct svc_sock *svsk = 1023 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt); 1024 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1025 int len; 1026 struct kvec *vec; 1027 int pnum, vlen; 1028 struct rpc_rqst *req = NULL; 1029 1030 dprintk("svc: tcp_recv %p data %d conn %d close %d\n", 1031 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags), 1032 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), 1033 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 1034 1035 len = svc_tcp_recv_record(svsk, rqstp); 1036 if (len < 0) 1037 goto error; 1038 1039 vec = rqstp->rq_vec; 1040 vec[0] = rqstp->rq_arg.head[0]; 1041 vlen = PAGE_SIZE; 1042 1043 /* 1044 * We have enough data for the whole tcp record. Let's try and read the 1045 * first 8 bytes to get the xid and the call direction. We can use this 1046 * to figure out if this is a call or a reply to a callback. If 1047 * sk_reclen is < 8 (xid and calldir), then this is a malformed packet. 1048 * In that case, don't bother with the calldir and just read the data. 1049 * It will be rejected in svc_process. 1050 */ 1051 if (len >= 8) { 1052 len = svc_process_calldir(svsk, rqstp, &req, vec); 1053 if (len < 0) 1054 goto err_again; 1055 vlen -= 8; 1056 } 1057 1058 pnum = 1; 1059 while (vlen < len) { 1060 vec[pnum].iov_base = (req) ? 1061 page_address(req->rq_private_buf.pages[pnum - 1]) : 1062 page_address(rqstp->rq_pages[pnum]); 1063 vec[pnum].iov_len = PAGE_SIZE; 1064 pnum++; 1065 vlen += PAGE_SIZE; 1066 } 1067 rqstp->rq_respages = &rqstp->rq_pages[pnum]; 1068 1069 /* Now receive data */ 1070 len = svc_recvfrom(rqstp, vec, pnum, len); 1071 if (len < 0) 1072 goto err_again; 1073 1074 /* 1075 * Account for the 8 bytes we read earlier 1076 */ 1077 len += 8; 1078 1079 if (req) { 1080 xprt_complete_rqst(req->rq_task, len); 1081 len = 0; 1082 goto out; 1083 } 1084 dprintk("svc: TCP complete record (%d bytes)\n", len); 1085 rqstp->rq_arg.len = len; 1086 rqstp->rq_arg.page_base = 0; 1087 if (len <= rqstp->rq_arg.head[0].iov_len) { 1088 rqstp->rq_arg.head[0].iov_len = len; 1089 rqstp->rq_arg.page_len = 0; 1090 } else { 1091 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 1092 } 1093 1094 rqstp->rq_xprt_ctxt = NULL; 1095 rqstp->rq_prot = IPPROTO_TCP; 1096 1097 out: 1098 /* Reset TCP read info */ 1099 svsk->sk_reclen = 0; 1100 svsk->sk_tcplen = 0; 1101 1102 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt); 1103 if (serv->sv_stats) 1104 serv->sv_stats->nettcpcnt++; 1105 1106 return len; 1107 1108 err_again: 1109 if (len == -EAGAIN) { 1110 dprintk("RPC: TCP recvfrom got EAGAIN\n"); 1111 return len; 1112 } 1113 error: 1114 if (len != -EAGAIN) { 1115 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", 1116 svsk->sk_xprt.xpt_server->sv_name, -len); 1117 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1118 } 1119 return -EAGAIN; 1120 } 1121 1122 /* 1123 * Send out data on TCP socket. 1124 */ 1125 static int svc_tcp_sendto(struct svc_rqst *rqstp) 1126 { 1127 struct xdr_buf *xbufp = &rqstp->rq_res; 1128 int sent; 1129 __be32 reclen; 1130 1131 /* Set up the first element of the reply kvec. 1132 * Any other kvecs that may be in use have been taken 1133 * care of by the server implementation itself. 1134 */ 1135 reclen = htonl(0x80000000|((xbufp->len ) - 4)); 1136 memcpy(xbufp->head[0].iov_base, &reclen, 4); 1137 1138 sent = svc_sendto(rqstp, &rqstp->rq_res); 1139 if (sent != xbufp->len) { 1140 printk(KERN_NOTICE 1141 "rpc-srv/tcp: %s: %s %d when sending %d bytes " 1142 "- shutting down socket\n", 1143 rqstp->rq_xprt->xpt_server->sv_name, 1144 (sent<0)?"got error":"sent only", 1145 sent, xbufp->len); 1146 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags); 1147 svc_xprt_enqueue(rqstp->rq_xprt); 1148 sent = -EAGAIN; 1149 } 1150 return sent; 1151 } 1152 1153 /* 1154 * Setup response header. TCP has a 4B record length field. 1155 */ 1156 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp) 1157 { 1158 struct kvec *resv = &rqstp->rq_res.head[0]; 1159 1160 /* tcp needs a space for the record length... */ 1161 svc_putnl(resv, 0); 1162 } 1163 1164 static int svc_tcp_has_wspace(struct svc_xprt *xprt) 1165 { 1166 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1167 struct svc_serv *serv = svsk->sk_xprt.xpt_server; 1168 int required; 1169 1170 if (test_bit(XPT_LISTENER, &xprt->xpt_flags)) 1171 return 1; 1172 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg; 1173 if (sk_stream_wspace(svsk->sk_sk) >= required) 1174 return 1; 1175 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 1176 return 0; 1177 } 1178 1179 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv, 1180 struct net *net, 1181 struct sockaddr *sa, int salen, 1182 int flags) 1183 { 1184 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags); 1185 } 1186 1187 static struct svc_xprt_ops svc_tcp_ops = { 1188 .xpo_create = svc_tcp_create, 1189 .xpo_recvfrom = svc_tcp_recvfrom, 1190 .xpo_sendto = svc_tcp_sendto, 1191 .xpo_release_rqst = svc_release_skb, 1192 .xpo_detach = svc_tcp_sock_detach, 1193 .xpo_free = svc_sock_free, 1194 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr, 1195 .xpo_has_wspace = svc_tcp_has_wspace, 1196 .xpo_accept = svc_tcp_accept, 1197 }; 1198 1199 static struct svc_xprt_class svc_tcp_class = { 1200 .xcl_name = "tcp", 1201 .xcl_owner = THIS_MODULE, 1202 .xcl_ops = &svc_tcp_ops, 1203 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP, 1204 }; 1205 1206 void svc_init_xprt_sock(void) 1207 { 1208 svc_reg_xprt_class(&svc_tcp_class); 1209 svc_reg_xprt_class(&svc_udp_class); 1210 } 1211 1212 void svc_cleanup_xprt_sock(void) 1213 { 1214 svc_unreg_xprt_class(&svc_tcp_class); 1215 svc_unreg_xprt_class(&svc_udp_class); 1216 } 1217 1218 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) 1219 { 1220 struct sock *sk = svsk->sk_sk; 1221 1222 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv); 1223 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags); 1224 if (sk->sk_state == TCP_LISTEN) { 1225 dprintk("setting up TCP socket for listening\n"); 1226 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags); 1227 sk->sk_data_ready = svc_tcp_listen_data_ready; 1228 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags); 1229 } else { 1230 dprintk("setting up TCP socket for reading\n"); 1231 sk->sk_state_change = svc_tcp_state_change; 1232 sk->sk_data_ready = svc_tcp_data_ready; 1233 sk->sk_write_space = svc_tcp_write_space; 1234 1235 svsk->sk_reclen = 0; 1236 svsk->sk_tcplen = 0; 1237 1238 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1239 1240 /* initialise setting must have enough space to 1241 * receive and respond to one request. 1242 * svc_tcp_recvfrom will re-adjust if necessary 1243 */ 1244 svc_sock_setbufsize(svsk->sk_sock, 1245 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, 1246 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); 1247 1248 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1249 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1250 if (sk->sk_state != TCP_ESTABLISHED) 1251 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1252 } 1253 } 1254 1255 void svc_sock_update_bufs(struct svc_serv *serv) 1256 { 1257 /* 1258 * The number of server threads has changed. Update 1259 * rcvbuf and sndbuf accordingly on all sockets 1260 */ 1261 struct svc_sock *svsk; 1262 1263 spin_lock_bh(&serv->sv_lock); 1264 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) 1265 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1266 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list) 1267 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); 1268 spin_unlock_bh(&serv->sv_lock); 1269 } 1270 EXPORT_SYMBOL_GPL(svc_sock_update_bufs); 1271 1272 /* 1273 * Initialize socket for RPC use and create svc_sock struct 1274 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. 1275 */ 1276 static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1277 struct socket *sock, 1278 int *errp, int flags) 1279 { 1280 struct svc_sock *svsk; 1281 struct sock *inet; 1282 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1283 1284 dprintk("svc: svc_setup_socket %p\n", sock); 1285 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { 1286 *errp = -ENOMEM; 1287 return NULL; 1288 } 1289 1290 inet = sock->sk; 1291 1292 /* Register socket with portmapper */ 1293 if (*errp >= 0 && pmap_register) 1294 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol, 1295 ntohs(inet_sk(inet)->inet_sport)); 1296 1297 if (*errp < 0) { 1298 kfree(svsk); 1299 return NULL; 1300 } 1301 1302 inet->sk_user_data = svsk; 1303 svsk->sk_sock = sock; 1304 svsk->sk_sk = inet; 1305 svsk->sk_ostate = inet->sk_state_change; 1306 svsk->sk_odata = inet->sk_data_ready; 1307 svsk->sk_owspace = inet->sk_write_space; 1308 1309 /* Initialize the socket */ 1310 if (sock->type == SOCK_DGRAM) 1311 svc_udp_init(svsk, serv); 1312 else 1313 svc_tcp_init(svsk, serv); 1314 1315 dprintk("svc: svc_setup_socket created %p (inet %p)\n", 1316 svsk, svsk->sk_sk); 1317 1318 return svsk; 1319 } 1320 1321 /** 1322 * svc_addsock - add a listener socket to an RPC service 1323 * @serv: pointer to RPC service to which to add a new listener 1324 * @fd: file descriptor of the new listener 1325 * @name_return: pointer to buffer to fill in with name of listener 1326 * @len: size of the buffer 1327 * 1328 * Fills in socket name and returns positive length of name if successful. 1329 * Name is terminated with '\n'. On error, returns a negative errno 1330 * value. 1331 */ 1332 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return, 1333 const size_t len) 1334 { 1335 int err = 0; 1336 struct socket *so = sockfd_lookup(fd, &err); 1337 struct svc_sock *svsk = NULL; 1338 1339 if (!so) 1340 return err; 1341 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6)) 1342 err = -EAFNOSUPPORT; 1343 else if (so->sk->sk_protocol != IPPROTO_TCP && 1344 so->sk->sk_protocol != IPPROTO_UDP) 1345 err = -EPROTONOSUPPORT; 1346 else if (so->state > SS_UNCONNECTED) 1347 err = -EISCONN; 1348 else { 1349 if (!try_module_get(THIS_MODULE)) 1350 err = -ENOENT; 1351 else 1352 svsk = svc_setup_socket(serv, so, &err, 1353 SVC_SOCK_DEFAULTS); 1354 if (svsk) { 1355 struct sockaddr_storage addr; 1356 struct sockaddr *sin = (struct sockaddr *)&addr; 1357 int salen; 1358 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0) 1359 svc_xprt_set_local(&svsk->sk_xprt, sin, salen); 1360 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags); 1361 spin_lock_bh(&serv->sv_lock); 1362 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks); 1363 spin_unlock_bh(&serv->sv_lock); 1364 svc_xprt_received(&svsk->sk_xprt); 1365 err = 0; 1366 } else 1367 module_put(THIS_MODULE); 1368 } 1369 if (err) { 1370 sockfd_put(so); 1371 return err; 1372 } 1373 return svc_one_sock_name(svsk, name_return, len); 1374 } 1375 EXPORT_SYMBOL_GPL(svc_addsock); 1376 1377 /* 1378 * Create socket for RPC service. 1379 */ 1380 static struct svc_xprt *svc_create_socket(struct svc_serv *serv, 1381 int protocol, 1382 struct net *net, 1383 struct sockaddr *sin, int len, 1384 int flags) 1385 { 1386 struct svc_sock *svsk; 1387 struct socket *sock; 1388 int error; 1389 int type; 1390 struct sockaddr_storage addr; 1391 struct sockaddr *newsin = (struct sockaddr *)&addr; 1392 int newlen; 1393 int family; 1394 int val; 1395 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 1396 1397 dprintk("svc: svc_create_socket(%s, %d, %s)\n", 1398 serv->sv_program->pg_name, protocol, 1399 __svc_print_addr(sin, buf, sizeof(buf))); 1400 1401 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1402 printk(KERN_WARNING "svc: only UDP and TCP " 1403 "sockets supported\n"); 1404 return ERR_PTR(-EINVAL); 1405 } 1406 1407 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1408 switch (sin->sa_family) { 1409 case AF_INET6: 1410 family = PF_INET6; 1411 break; 1412 case AF_INET: 1413 family = PF_INET; 1414 break; 1415 default: 1416 return ERR_PTR(-EINVAL); 1417 } 1418 1419 error = __sock_create(net, family, type, protocol, &sock, 1); 1420 if (error < 0) 1421 return ERR_PTR(error); 1422 1423 svc_reclassify_socket(sock); 1424 1425 /* 1426 * If this is an PF_INET6 listener, we want to avoid 1427 * getting requests from IPv4 remotes. Those should 1428 * be shunted to a PF_INET listener via rpcbind. 1429 */ 1430 val = 1; 1431 if (family == PF_INET6) 1432 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY, 1433 (char *)&val, sizeof(val)); 1434 1435 if (type == SOCK_STREAM) 1436 sock->sk->sk_reuse = 1; /* allow address reuse */ 1437 error = kernel_bind(sock, sin, len); 1438 if (error < 0) 1439 goto bummer; 1440 1441 newlen = len; 1442 error = kernel_getsockname(sock, newsin, &newlen); 1443 if (error < 0) 1444 goto bummer; 1445 1446 if (protocol == IPPROTO_TCP) { 1447 if ((error = kernel_listen(sock, 64)) < 0) 1448 goto bummer; 1449 } 1450 1451 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { 1452 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen); 1453 return (struct svc_xprt *)svsk; 1454 } 1455 1456 bummer: 1457 dprintk("svc: svc_create_socket error = %d\n", -error); 1458 sock_release(sock); 1459 return ERR_PTR(error); 1460 } 1461 1462 /* 1463 * Detach the svc_sock from the socket so that no 1464 * more callbacks occur. 1465 */ 1466 static void svc_sock_detach(struct svc_xprt *xprt) 1467 { 1468 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1469 struct sock *sk = svsk->sk_sk; 1470 1471 dprintk("svc: svc_sock_detach(%p)\n", svsk); 1472 1473 /* put back the old socket callbacks */ 1474 sk->sk_state_change = svsk->sk_ostate; 1475 sk->sk_data_ready = svsk->sk_odata; 1476 sk->sk_write_space = svsk->sk_owspace; 1477 1478 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) 1479 wake_up_interruptible(sk_sleep(sk)); 1480 } 1481 1482 /* 1483 * Disconnect the socket, and reset the callbacks 1484 */ 1485 static void svc_tcp_sock_detach(struct svc_xprt *xprt) 1486 { 1487 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1488 1489 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk); 1490 1491 svc_sock_detach(xprt); 1492 1493 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) 1494 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR); 1495 } 1496 1497 /* 1498 * Free the svc_sock's socket resources and the svc_sock itself. 1499 */ 1500 static void svc_sock_free(struct svc_xprt *xprt) 1501 { 1502 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt); 1503 dprintk("svc: svc_sock_free(%p)\n", svsk); 1504 1505 if (svsk->sk_sock->file) 1506 sockfd_put(svsk->sk_sock); 1507 else 1508 sock_release(svsk->sk_sock); 1509 kfree(svsk); 1510 } 1511 1512 /* 1513 * Create a svc_xprt. 1514 * 1515 * For internal use only (e.g. nfsv4.1 backchannel). 1516 * Callers should typically use the xpo_create() method. 1517 */ 1518 struct svc_xprt *svc_sock_create(struct svc_serv *serv, int prot) 1519 { 1520 struct svc_sock *svsk; 1521 struct svc_xprt *xprt = NULL; 1522 1523 dprintk("svc: %s\n", __func__); 1524 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL); 1525 if (!svsk) 1526 goto out; 1527 1528 xprt = &svsk->sk_xprt; 1529 if (prot == IPPROTO_TCP) 1530 svc_xprt_init(&svc_tcp_class, xprt, serv); 1531 else if (prot == IPPROTO_UDP) 1532 svc_xprt_init(&svc_udp_class, xprt, serv); 1533 else 1534 BUG(); 1535 out: 1536 dprintk("svc: %s return %p\n", __func__, xprt); 1537 return xprt; 1538 } 1539 EXPORT_SYMBOL_GPL(svc_sock_create); 1540 1541 /* 1542 * Destroy a svc_sock. 1543 */ 1544 void svc_sock_destroy(struct svc_xprt *xprt) 1545 { 1546 if (xprt) 1547 kfree(container_of(xprt, struct svc_sock, sk_xprt)); 1548 } 1549 EXPORT_SYMBOL_GPL(svc_sock_destroy); 1550