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_sock_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/sched.h> 23 #include <linux/errno.h> 24 #include <linux/fcntl.h> 25 #include <linux/net.h> 26 #include <linux/in.h> 27 #include <linux/inet.h> 28 #include <linux/udp.h> 29 #include <linux/tcp.h> 30 #include <linux/unistd.h> 31 #include <linux/slab.h> 32 #include <linux/netdevice.h> 33 #include <linux/skbuff.h> 34 #include <linux/file.h> 35 #include <linux/freezer.h> 36 #include <net/sock.h> 37 #include <net/checksum.h> 38 #include <net/ip.h> 39 #include <net/ipv6.h> 40 #include <net/tcp_states.h> 41 #include <asm/uaccess.h> 42 #include <asm/ioctls.h> 43 44 #include <linux/sunrpc/types.h> 45 #include <linux/sunrpc/clnt.h> 46 #include <linux/sunrpc/xdr.h> 47 #include <linux/sunrpc/svcsock.h> 48 #include <linux/sunrpc/stats.h> 49 50 /* SMP locking strategy: 51 * 52 * svc_pool->sp_lock protects most of the fields of that pool. 53 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt. 54 * when both need to be taken (rare), svc_serv->sv_lock is first. 55 * BKL protects svc_serv->sv_nrthread. 56 * svc_sock->sk_lock protects the svc_sock->sk_deferred list 57 * and the ->sk_info_authunix cache. 58 * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply. 59 * 60 * Some flags can be set to certain values at any time 61 * providing that certain rules are followed: 62 * 63 * SK_CONN, SK_DATA, can be set or cleared at any time. 64 * after a set, svc_sock_enqueue must be called. 65 * after a clear, the socket must be read/accepted 66 * if this succeeds, it must be set again. 67 * SK_CLOSE can set at any time. It is never cleared. 68 * sk_inuse contains a bias of '1' until SK_DEAD is set. 69 * so when sk_inuse hits zero, we know the socket is dead 70 * and no-one is using it. 71 * SK_DEAD can only be set while SK_BUSY is held which ensures 72 * no other thread will be using the socket or will try to 73 * set SK_DEAD. 74 * 75 */ 76 77 #define RPCDBG_FACILITY RPCDBG_SVCSOCK 78 79 80 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *, 81 int *errp, int flags); 82 static void svc_delete_socket(struct svc_sock *svsk); 83 static void svc_udp_data_ready(struct sock *, int); 84 static int svc_udp_recvfrom(struct svc_rqst *); 85 static int svc_udp_sendto(struct svc_rqst *); 86 static void svc_close_socket(struct svc_sock *svsk); 87 88 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk); 89 static int svc_deferred_recv(struct svc_rqst *rqstp); 90 static struct cache_deferred_req *svc_defer(struct cache_req *req); 91 92 /* apparently the "standard" is that clients close 93 * idle connections after 5 minutes, servers after 94 * 6 minutes 95 * http://www.connectathon.org/talks96/nfstcp.pdf 96 */ 97 static int svc_conn_age_period = 6*60; 98 99 #ifdef CONFIG_DEBUG_LOCK_ALLOC 100 static struct lock_class_key svc_key[2]; 101 static struct lock_class_key svc_slock_key[2]; 102 103 static inline void svc_reclassify_socket(struct socket *sock) 104 { 105 struct sock *sk = sock->sk; 106 BUG_ON(sk->sk_lock.owner != NULL); 107 switch (sk->sk_family) { 108 case AF_INET: 109 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD", 110 &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]); 111 break; 112 113 case AF_INET6: 114 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD", 115 &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]); 116 break; 117 118 default: 119 BUG(); 120 } 121 } 122 #else 123 static inline void svc_reclassify_socket(struct socket *sock) 124 { 125 } 126 #endif 127 128 static char *__svc_print_addr(struct sockaddr *addr, char *buf, size_t len) 129 { 130 switch (addr->sa_family) { 131 case AF_INET: 132 snprintf(buf, len, "%u.%u.%u.%u, port=%u", 133 NIPQUAD(((struct sockaddr_in *) addr)->sin_addr), 134 ntohs(((struct sockaddr_in *) addr)->sin_port)); 135 break; 136 137 case AF_INET6: 138 snprintf(buf, len, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u", 139 NIP6(((struct sockaddr_in6 *) addr)->sin6_addr), 140 ntohs(((struct sockaddr_in6 *) addr)->sin6_port)); 141 break; 142 143 default: 144 snprintf(buf, len, "unknown address type: %d", addr->sa_family); 145 break; 146 } 147 return buf; 148 } 149 150 /** 151 * svc_print_addr - Format rq_addr field for printing 152 * @rqstp: svc_rqst struct containing address to print 153 * @buf: target buffer for formatted address 154 * @len: length of target buffer 155 * 156 */ 157 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len) 158 { 159 return __svc_print_addr(svc_addr(rqstp), buf, len); 160 } 161 EXPORT_SYMBOL_GPL(svc_print_addr); 162 163 /* 164 * Queue up an idle server thread. Must have pool->sp_lock held. 165 * Note: this is really a stack rather than a queue, so that we only 166 * use as many different threads as we need, and the rest don't pollute 167 * the cache. 168 */ 169 static inline void 170 svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp) 171 { 172 list_add(&rqstp->rq_list, &pool->sp_threads); 173 } 174 175 /* 176 * Dequeue an nfsd thread. Must have pool->sp_lock held. 177 */ 178 static inline void 179 svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp) 180 { 181 list_del(&rqstp->rq_list); 182 } 183 184 /* 185 * Release an skbuff after use 186 */ 187 static inline void 188 svc_release_skb(struct svc_rqst *rqstp) 189 { 190 struct sk_buff *skb = rqstp->rq_skbuff; 191 struct svc_deferred_req *dr = rqstp->rq_deferred; 192 193 if (skb) { 194 rqstp->rq_skbuff = NULL; 195 196 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb); 197 skb_free_datagram(rqstp->rq_sock->sk_sk, skb); 198 } 199 if (dr) { 200 rqstp->rq_deferred = NULL; 201 kfree(dr); 202 } 203 } 204 205 /* 206 * Any space to write? 207 */ 208 static inline unsigned long 209 svc_sock_wspace(struct svc_sock *svsk) 210 { 211 int wspace; 212 213 if (svsk->sk_sock->type == SOCK_STREAM) 214 wspace = sk_stream_wspace(svsk->sk_sk); 215 else 216 wspace = sock_wspace(svsk->sk_sk); 217 218 return wspace; 219 } 220 221 /* 222 * Queue up a socket with data pending. If there are idle nfsd 223 * processes, wake 'em up. 224 * 225 */ 226 static void 227 svc_sock_enqueue(struct svc_sock *svsk) 228 { 229 struct svc_serv *serv = svsk->sk_server; 230 struct svc_pool *pool; 231 struct svc_rqst *rqstp; 232 int cpu; 233 234 if (!(svsk->sk_flags & 235 ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) )) 236 return; 237 if (test_bit(SK_DEAD, &svsk->sk_flags)) 238 return; 239 240 cpu = get_cpu(); 241 pool = svc_pool_for_cpu(svsk->sk_server, cpu); 242 put_cpu(); 243 244 spin_lock_bh(&pool->sp_lock); 245 246 if (!list_empty(&pool->sp_threads) && 247 !list_empty(&pool->sp_sockets)) 248 printk(KERN_ERR 249 "svc_sock_enqueue: threads and sockets both waiting??\n"); 250 251 if (test_bit(SK_DEAD, &svsk->sk_flags)) { 252 /* Don't enqueue dead sockets */ 253 dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk); 254 goto out_unlock; 255 } 256 257 /* Mark socket as busy. It will remain in this state until the 258 * server has processed all pending data and put the socket back 259 * on the idle list. We update SK_BUSY atomically because 260 * it also guards against trying to enqueue the svc_sock twice. 261 */ 262 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) { 263 /* Don't enqueue socket while already enqueued */ 264 dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk); 265 goto out_unlock; 266 } 267 BUG_ON(svsk->sk_pool != NULL); 268 svsk->sk_pool = pool; 269 270 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 271 if (((atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg)*2 272 > svc_sock_wspace(svsk)) 273 && !test_bit(SK_CLOSE, &svsk->sk_flags) 274 && !test_bit(SK_CONN, &svsk->sk_flags)) { 275 /* Don't enqueue while not enough space for reply */ 276 dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n", 277 svsk->sk_sk, atomic_read(&svsk->sk_reserved)+serv->sv_max_mesg, 278 svc_sock_wspace(svsk)); 279 svsk->sk_pool = NULL; 280 clear_bit(SK_BUSY, &svsk->sk_flags); 281 goto out_unlock; 282 } 283 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags); 284 285 286 if (!list_empty(&pool->sp_threads)) { 287 rqstp = list_entry(pool->sp_threads.next, 288 struct svc_rqst, 289 rq_list); 290 dprintk("svc: socket %p served by daemon %p\n", 291 svsk->sk_sk, rqstp); 292 svc_thread_dequeue(pool, rqstp); 293 if (rqstp->rq_sock) 294 printk(KERN_ERR 295 "svc_sock_enqueue: server %p, rq_sock=%p!\n", 296 rqstp, rqstp->rq_sock); 297 rqstp->rq_sock = svsk; 298 atomic_inc(&svsk->sk_inuse); 299 rqstp->rq_reserved = serv->sv_max_mesg; 300 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); 301 BUG_ON(svsk->sk_pool != pool); 302 wake_up(&rqstp->rq_wait); 303 } else { 304 dprintk("svc: socket %p put into queue\n", svsk->sk_sk); 305 list_add_tail(&svsk->sk_ready, &pool->sp_sockets); 306 BUG_ON(svsk->sk_pool != pool); 307 } 308 309 out_unlock: 310 spin_unlock_bh(&pool->sp_lock); 311 } 312 313 /* 314 * Dequeue the first socket. Must be called with the pool->sp_lock held. 315 */ 316 static inline struct svc_sock * 317 svc_sock_dequeue(struct svc_pool *pool) 318 { 319 struct svc_sock *svsk; 320 321 if (list_empty(&pool->sp_sockets)) 322 return NULL; 323 324 svsk = list_entry(pool->sp_sockets.next, 325 struct svc_sock, sk_ready); 326 list_del_init(&svsk->sk_ready); 327 328 dprintk("svc: socket %p dequeued, inuse=%d\n", 329 svsk->sk_sk, atomic_read(&svsk->sk_inuse)); 330 331 return svsk; 332 } 333 334 /* 335 * Having read something from a socket, check whether it 336 * needs to be re-enqueued. 337 * Note: SK_DATA only gets cleared when a read-attempt finds 338 * no (or insufficient) data. 339 */ 340 static inline void 341 svc_sock_received(struct svc_sock *svsk) 342 { 343 svsk->sk_pool = NULL; 344 clear_bit(SK_BUSY, &svsk->sk_flags); 345 svc_sock_enqueue(svsk); 346 } 347 348 349 /** 350 * svc_reserve - change the space reserved for the reply to a request. 351 * @rqstp: The request in question 352 * @space: new max space to reserve 353 * 354 * Each request reserves some space on the output queue of the socket 355 * to make sure the reply fits. This function reduces that reserved 356 * space to be the amount of space used already, plus @space. 357 * 358 */ 359 void svc_reserve(struct svc_rqst *rqstp, int space) 360 { 361 space += rqstp->rq_res.head[0].iov_len; 362 363 if (space < rqstp->rq_reserved) { 364 struct svc_sock *svsk = rqstp->rq_sock; 365 atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved); 366 rqstp->rq_reserved = space; 367 368 svc_sock_enqueue(svsk); 369 } 370 } 371 372 /* 373 * Release a socket after use. 374 */ 375 static inline void 376 svc_sock_put(struct svc_sock *svsk) 377 { 378 if (atomic_dec_and_test(&svsk->sk_inuse)) { 379 BUG_ON(! test_bit(SK_DEAD, &svsk->sk_flags)); 380 381 dprintk("svc: releasing dead socket\n"); 382 if (svsk->sk_sock->file) 383 sockfd_put(svsk->sk_sock); 384 else 385 sock_release(svsk->sk_sock); 386 if (svsk->sk_info_authunix != NULL) 387 svcauth_unix_info_release(svsk->sk_info_authunix); 388 kfree(svsk); 389 } 390 } 391 392 static void 393 svc_sock_release(struct svc_rqst *rqstp) 394 { 395 struct svc_sock *svsk = rqstp->rq_sock; 396 397 svc_release_skb(rqstp); 398 399 svc_free_res_pages(rqstp); 400 rqstp->rq_res.page_len = 0; 401 rqstp->rq_res.page_base = 0; 402 403 404 /* Reset response buffer and release 405 * the reservation. 406 * But first, check that enough space was reserved 407 * for the reply, otherwise we have a bug! 408 */ 409 if ((rqstp->rq_res.len) > rqstp->rq_reserved) 410 printk(KERN_ERR "RPC request reserved %d but used %d\n", 411 rqstp->rq_reserved, 412 rqstp->rq_res.len); 413 414 rqstp->rq_res.head[0].iov_len = 0; 415 svc_reserve(rqstp, 0); 416 rqstp->rq_sock = NULL; 417 418 svc_sock_put(svsk); 419 } 420 421 /* 422 * External function to wake up a server waiting for data 423 * This really only makes sense for services like lockd 424 * which have exactly one thread anyway. 425 */ 426 void 427 svc_wake_up(struct svc_serv *serv) 428 { 429 struct svc_rqst *rqstp; 430 unsigned int i; 431 struct svc_pool *pool; 432 433 for (i = 0; i < serv->sv_nrpools; i++) { 434 pool = &serv->sv_pools[i]; 435 436 spin_lock_bh(&pool->sp_lock); 437 if (!list_empty(&pool->sp_threads)) { 438 rqstp = list_entry(pool->sp_threads.next, 439 struct svc_rqst, 440 rq_list); 441 dprintk("svc: daemon %p woken up.\n", rqstp); 442 /* 443 svc_thread_dequeue(pool, rqstp); 444 rqstp->rq_sock = NULL; 445 */ 446 wake_up(&rqstp->rq_wait); 447 } 448 spin_unlock_bh(&pool->sp_lock); 449 } 450 } 451 452 union svc_pktinfo_u { 453 struct in_pktinfo pkti; 454 struct in6_pktinfo pkti6; 455 }; 456 #define SVC_PKTINFO_SPACE \ 457 CMSG_SPACE(sizeof(union svc_pktinfo_u)) 458 459 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh) 460 { 461 switch (rqstp->rq_sock->sk_sk->sk_family) { 462 case AF_INET: { 463 struct in_pktinfo *pki = CMSG_DATA(cmh); 464 465 cmh->cmsg_level = SOL_IP; 466 cmh->cmsg_type = IP_PKTINFO; 467 pki->ipi_ifindex = 0; 468 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr; 469 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 470 } 471 break; 472 473 case AF_INET6: { 474 struct in6_pktinfo *pki = CMSG_DATA(cmh); 475 476 cmh->cmsg_level = SOL_IPV6; 477 cmh->cmsg_type = IPV6_PKTINFO; 478 pki->ipi6_ifindex = 0; 479 ipv6_addr_copy(&pki->ipi6_addr, 480 &rqstp->rq_daddr.addr6); 481 cmh->cmsg_len = CMSG_LEN(sizeof(*pki)); 482 } 483 break; 484 } 485 return; 486 } 487 488 /* 489 * Generic sendto routine 490 */ 491 static int 492 svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr) 493 { 494 struct svc_sock *svsk = rqstp->rq_sock; 495 struct socket *sock = svsk->sk_sock; 496 int slen; 497 union { 498 struct cmsghdr hdr; 499 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 500 } buffer; 501 struct cmsghdr *cmh = &buffer.hdr; 502 int len = 0; 503 int result; 504 int size; 505 struct page **ppage = xdr->pages; 506 size_t base = xdr->page_base; 507 unsigned int pglen = xdr->page_len; 508 unsigned int flags = MSG_MORE; 509 char buf[RPC_MAX_ADDRBUFLEN]; 510 511 slen = xdr->len; 512 513 if (rqstp->rq_prot == IPPROTO_UDP) { 514 struct msghdr msg = { 515 .msg_name = &rqstp->rq_addr, 516 .msg_namelen = rqstp->rq_addrlen, 517 .msg_control = cmh, 518 .msg_controllen = sizeof(buffer), 519 .msg_flags = MSG_MORE, 520 }; 521 522 svc_set_cmsg_data(rqstp, cmh); 523 524 if (sock_sendmsg(sock, &msg, 0) < 0) 525 goto out; 526 } 527 528 /* send head */ 529 if (slen == xdr->head[0].iov_len) 530 flags = 0; 531 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0, 532 xdr->head[0].iov_len, flags); 533 if (len != xdr->head[0].iov_len) 534 goto out; 535 slen -= xdr->head[0].iov_len; 536 if (slen == 0) 537 goto out; 538 539 /* send page data */ 540 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen; 541 while (pglen > 0) { 542 if (slen == size) 543 flags = 0; 544 result = kernel_sendpage(sock, *ppage, base, size, flags); 545 if (result > 0) 546 len += result; 547 if (result != size) 548 goto out; 549 slen -= size; 550 pglen -= size; 551 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen; 552 base = 0; 553 ppage++; 554 } 555 /* send tail */ 556 if (xdr->tail[0].iov_len) { 557 result = kernel_sendpage(sock, rqstp->rq_respages[0], 558 ((unsigned long)xdr->tail[0].iov_base) 559 & (PAGE_SIZE-1), 560 xdr->tail[0].iov_len, 0); 561 562 if (result > 0) 563 len += result; 564 } 565 out: 566 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n", 567 rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, 568 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf))); 569 570 return len; 571 } 572 573 /* 574 * Report socket names for nfsdfs 575 */ 576 static int one_sock_name(char *buf, struct svc_sock *svsk) 577 { 578 int len; 579 580 switch(svsk->sk_sk->sk_family) { 581 case AF_INET: 582 len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n", 583 svsk->sk_sk->sk_protocol==IPPROTO_UDP? 584 "udp" : "tcp", 585 NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr), 586 inet_sk(svsk->sk_sk)->num); 587 break; 588 default: 589 len = sprintf(buf, "*unknown-%d*\n", 590 svsk->sk_sk->sk_family); 591 } 592 return len; 593 } 594 595 int 596 svc_sock_names(char *buf, struct svc_serv *serv, char *toclose) 597 { 598 struct svc_sock *svsk, *closesk = NULL; 599 int len = 0; 600 601 if (!serv) 602 return 0; 603 spin_lock_bh(&serv->sv_lock); 604 list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) { 605 int onelen = one_sock_name(buf+len, svsk); 606 if (toclose && strcmp(toclose, buf+len) == 0) 607 closesk = svsk; 608 else 609 len += onelen; 610 } 611 spin_unlock_bh(&serv->sv_lock); 612 if (closesk) 613 /* Should unregister with portmap, but you cannot 614 * unregister just one protocol... 615 */ 616 svc_close_socket(closesk); 617 else if (toclose) 618 return -ENOENT; 619 return len; 620 } 621 EXPORT_SYMBOL(svc_sock_names); 622 623 /* 624 * Check input queue length 625 */ 626 static int 627 svc_recv_available(struct svc_sock *svsk) 628 { 629 struct socket *sock = svsk->sk_sock; 630 int avail, err; 631 632 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail); 633 634 return (err >= 0)? avail : err; 635 } 636 637 /* 638 * Generic recvfrom routine. 639 */ 640 static int 641 svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen) 642 { 643 struct svc_sock *svsk = rqstp->rq_sock; 644 struct msghdr msg = { 645 .msg_flags = MSG_DONTWAIT, 646 }; 647 struct sockaddr *sin; 648 int len; 649 650 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen, 651 msg.msg_flags); 652 653 /* sock_recvmsg doesn't fill in the name/namelen, so we must.. 654 */ 655 memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen); 656 rqstp->rq_addrlen = svsk->sk_remotelen; 657 658 /* Destination address in request is needed for binding the 659 * source address in RPC callbacks later. 660 */ 661 sin = (struct sockaddr *)&svsk->sk_local; 662 switch (sin->sa_family) { 663 case AF_INET: 664 rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr; 665 break; 666 case AF_INET6: 667 rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr; 668 break; 669 } 670 671 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n", 672 svsk, iov[0].iov_base, iov[0].iov_len, len); 673 674 return len; 675 } 676 677 /* 678 * Set socket snd and rcv buffer lengths 679 */ 680 static inline void 681 svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv) 682 { 683 #if 0 684 mm_segment_t oldfs; 685 oldfs = get_fs(); set_fs(KERNEL_DS); 686 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF, 687 (char*)&snd, sizeof(snd)); 688 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, 689 (char*)&rcv, sizeof(rcv)); 690 #else 691 /* sock_setsockopt limits use to sysctl_?mem_max, 692 * which isn't acceptable. Until that is made conditional 693 * on not having CAP_SYS_RESOURCE or similar, we go direct... 694 * DaveM said I could! 695 */ 696 lock_sock(sock->sk); 697 sock->sk->sk_sndbuf = snd * 2; 698 sock->sk->sk_rcvbuf = rcv * 2; 699 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; 700 release_sock(sock->sk); 701 #endif 702 } 703 /* 704 * INET callback when data has been received on the socket. 705 */ 706 static void 707 svc_udp_data_ready(struct sock *sk, int count) 708 { 709 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 710 711 if (svsk) { 712 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n", 713 svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags)); 714 set_bit(SK_DATA, &svsk->sk_flags); 715 svc_sock_enqueue(svsk); 716 } 717 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 718 wake_up_interruptible(sk->sk_sleep); 719 } 720 721 /* 722 * INET callback when space is newly available on the socket. 723 */ 724 static void 725 svc_write_space(struct sock *sk) 726 { 727 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data); 728 729 if (svsk) { 730 dprintk("svc: socket %p(inet %p), write_space busy=%d\n", 731 svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags)); 732 svc_sock_enqueue(svsk); 733 } 734 735 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) { 736 dprintk("RPC svc_write_space: someone sleeping on %p\n", 737 svsk); 738 wake_up_interruptible(sk->sk_sleep); 739 } 740 } 741 742 static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp, 743 struct cmsghdr *cmh) 744 { 745 switch (rqstp->rq_sock->sk_sk->sk_family) { 746 case AF_INET: { 747 struct in_pktinfo *pki = CMSG_DATA(cmh); 748 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr; 749 break; 750 } 751 case AF_INET6: { 752 struct in6_pktinfo *pki = CMSG_DATA(cmh); 753 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr); 754 break; 755 } 756 } 757 } 758 759 /* 760 * Receive a datagram from a UDP socket. 761 */ 762 static int 763 svc_udp_recvfrom(struct svc_rqst *rqstp) 764 { 765 struct svc_sock *svsk = rqstp->rq_sock; 766 struct svc_serv *serv = svsk->sk_server; 767 struct sk_buff *skb; 768 union { 769 struct cmsghdr hdr; 770 long all[SVC_PKTINFO_SPACE / sizeof(long)]; 771 } buffer; 772 struct cmsghdr *cmh = &buffer.hdr; 773 int err, len; 774 struct msghdr msg = { 775 .msg_name = svc_addr(rqstp), 776 .msg_control = cmh, 777 .msg_controllen = sizeof(buffer), 778 .msg_flags = MSG_DONTWAIT, 779 }; 780 781 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) 782 /* udp sockets need large rcvbuf as all pending 783 * requests are still in that buffer. sndbuf must 784 * also be large enough that there is enough space 785 * for one reply per thread. We count all threads 786 * rather than threads in a particular pool, which 787 * provides an upper bound on the number of threads 788 * which will access the socket. 789 */ 790 svc_sock_setbufsize(svsk->sk_sock, 791 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 792 (serv->sv_nrthreads+3) * serv->sv_max_mesg); 793 794 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { 795 svc_sock_received(svsk); 796 return svc_deferred_recv(rqstp); 797 } 798 799 if (test_bit(SK_CLOSE, &svsk->sk_flags)) { 800 svc_delete_socket(svsk); 801 return 0; 802 } 803 804 clear_bit(SK_DATA, &svsk->sk_flags); 805 skb = NULL; 806 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL, 807 0, 0, MSG_PEEK | MSG_DONTWAIT); 808 if (err >= 0) 809 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err); 810 811 if (skb == NULL) { 812 if (err != -EAGAIN) { 813 /* possibly an icmp error */ 814 dprintk("svc: recvfrom returned error %d\n", -err); 815 set_bit(SK_DATA, &svsk->sk_flags); 816 } 817 svc_sock_received(svsk); 818 return -EAGAIN; 819 } 820 rqstp->rq_addrlen = sizeof(rqstp->rq_addr); 821 if (skb->tstamp.tv64 == 0) { 822 skb->tstamp = ktime_get_real(); 823 /* Don't enable netstamp, sunrpc doesn't 824 need that much accuracy */ 825 } 826 svsk->sk_sk->sk_stamp = skb->tstamp; 827 set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */ 828 829 /* 830 * Maybe more packets - kick another thread ASAP. 831 */ 832 svc_sock_received(svsk); 833 834 len = skb->len - sizeof(struct udphdr); 835 rqstp->rq_arg.len = len; 836 837 rqstp->rq_prot = IPPROTO_UDP; 838 839 if (cmh->cmsg_level != IPPROTO_IP || 840 cmh->cmsg_type != IP_PKTINFO) { 841 if (net_ratelimit()) 842 printk("rpcsvc: received unknown control message:" 843 "%d/%d\n", 844 cmh->cmsg_level, cmh->cmsg_type); 845 skb_free_datagram(svsk->sk_sk, skb); 846 return 0; 847 } 848 svc_udp_get_dest_address(rqstp, cmh); 849 850 if (skb_is_nonlinear(skb)) { 851 /* we have to copy */ 852 local_bh_disable(); 853 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) { 854 local_bh_enable(); 855 /* checksum error */ 856 skb_free_datagram(svsk->sk_sk, skb); 857 return 0; 858 } 859 local_bh_enable(); 860 skb_free_datagram(svsk->sk_sk, skb); 861 } else { 862 /* we can use it in-place */ 863 rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr); 864 rqstp->rq_arg.head[0].iov_len = len; 865 if (skb_checksum_complete(skb)) { 866 skb_free_datagram(svsk->sk_sk, skb); 867 return 0; 868 } 869 rqstp->rq_skbuff = skb; 870 } 871 872 rqstp->rq_arg.page_base = 0; 873 if (len <= rqstp->rq_arg.head[0].iov_len) { 874 rqstp->rq_arg.head[0].iov_len = len; 875 rqstp->rq_arg.page_len = 0; 876 rqstp->rq_respages = rqstp->rq_pages+1; 877 } else { 878 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 879 rqstp->rq_respages = rqstp->rq_pages + 1 + 880 (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE; 881 } 882 883 if (serv->sv_stats) 884 serv->sv_stats->netudpcnt++; 885 886 return len; 887 } 888 889 static int 890 svc_udp_sendto(struct svc_rqst *rqstp) 891 { 892 int error; 893 894 error = svc_sendto(rqstp, &rqstp->rq_res); 895 if (error == -ECONNREFUSED) 896 /* ICMP error on earlier request. */ 897 error = svc_sendto(rqstp, &rqstp->rq_res); 898 899 return error; 900 } 901 902 static void 903 svc_udp_init(struct svc_sock *svsk) 904 { 905 int one = 1; 906 mm_segment_t oldfs; 907 908 svsk->sk_sk->sk_data_ready = svc_udp_data_ready; 909 svsk->sk_sk->sk_write_space = svc_write_space; 910 svsk->sk_recvfrom = svc_udp_recvfrom; 911 svsk->sk_sendto = svc_udp_sendto; 912 913 /* initialise setting must have enough space to 914 * receive and respond to one request. 915 * svc_udp_recvfrom will re-adjust if necessary 916 */ 917 svc_sock_setbufsize(svsk->sk_sock, 918 3 * svsk->sk_server->sv_max_mesg, 919 3 * svsk->sk_server->sv_max_mesg); 920 921 set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */ 922 set_bit(SK_CHNGBUF, &svsk->sk_flags); 923 924 oldfs = get_fs(); 925 set_fs(KERNEL_DS); 926 /* make sure we get destination address info */ 927 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO, 928 (char __user *)&one, sizeof(one)); 929 set_fs(oldfs); 930 } 931 932 /* 933 * A data_ready event on a listening socket means there's a connection 934 * pending. Do not use state_change as a substitute for it. 935 */ 936 static void 937 svc_tcp_listen_data_ready(struct sock *sk, int count_unused) 938 { 939 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 940 941 dprintk("svc: socket %p TCP (listen) state change %d\n", 942 sk, sk->sk_state); 943 944 /* 945 * This callback may called twice when a new connection 946 * is established as a child socket inherits everything 947 * from a parent LISTEN socket. 948 * 1) data_ready method of the parent socket will be called 949 * when one of child sockets become ESTABLISHED. 950 * 2) data_ready method of the child socket may be called 951 * when it receives data before the socket is accepted. 952 * In case of 2, we should ignore it silently. 953 */ 954 if (sk->sk_state == TCP_LISTEN) { 955 if (svsk) { 956 set_bit(SK_CONN, &svsk->sk_flags); 957 svc_sock_enqueue(svsk); 958 } else 959 printk("svc: socket %p: no user data\n", sk); 960 } 961 962 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 963 wake_up_interruptible_all(sk->sk_sleep); 964 } 965 966 /* 967 * A state change on a connected socket means it's dying or dead. 968 */ 969 static void 970 svc_tcp_state_change(struct sock *sk) 971 { 972 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 973 974 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n", 975 sk, sk->sk_state, sk->sk_user_data); 976 977 if (!svsk) 978 printk("svc: socket %p: no user data\n", sk); 979 else { 980 set_bit(SK_CLOSE, &svsk->sk_flags); 981 svc_sock_enqueue(svsk); 982 } 983 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 984 wake_up_interruptible_all(sk->sk_sleep); 985 } 986 987 static void 988 svc_tcp_data_ready(struct sock *sk, int count) 989 { 990 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data; 991 992 dprintk("svc: socket %p TCP data ready (svsk %p)\n", 993 sk, sk->sk_user_data); 994 if (svsk) { 995 set_bit(SK_DATA, &svsk->sk_flags); 996 svc_sock_enqueue(svsk); 997 } 998 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 999 wake_up_interruptible(sk->sk_sleep); 1000 } 1001 1002 static inline int svc_port_is_privileged(struct sockaddr *sin) 1003 { 1004 switch (sin->sa_family) { 1005 case AF_INET: 1006 return ntohs(((struct sockaddr_in *)sin)->sin_port) 1007 < PROT_SOCK; 1008 case AF_INET6: 1009 return ntohs(((struct sockaddr_in6 *)sin)->sin6_port) 1010 < PROT_SOCK; 1011 default: 1012 return 0; 1013 } 1014 } 1015 1016 /* 1017 * Accept a TCP connection 1018 */ 1019 static void 1020 svc_tcp_accept(struct svc_sock *svsk) 1021 { 1022 struct sockaddr_storage addr; 1023 struct sockaddr *sin = (struct sockaddr *) &addr; 1024 struct svc_serv *serv = svsk->sk_server; 1025 struct socket *sock = svsk->sk_sock; 1026 struct socket *newsock; 1027 struct svc_sock *newsvsk; 1028 int err, slen; 1029 char buf[RPC_MAX_ADDRBUFLEN]; 1030 1031 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock); 1032 if (!sock) 1033 return; 1034 1035 clear_bit(SK_CONN, &svsk->sk_flags); 1036 err = kernel_accept(sock, &newsock, O_NONBLOCK); 1037 if (err < 0) { 1038 if (err == -ENOMEM) 1039 printk(KERN_WARNING "%s: no more sockets!\n", 1040 serv->sv_name); 1041 else if (err != -EAGAIN && net_ratelimit()) 1042 printk(KERN_WARNING "%s: accept failed (err %d)!\n", 1043 serv->sv_name, -err); 1044 return; 1045 } 1046 1047 set_bit(SK_CONN, &svsk->sk_flags); 1048 svc_sock_enqueue(svsk); 1049 1050 err = kernel_getpeername(newsock, sin, &slen); 1051 if (err < 0) { 1052 if (net_ratelimit()) 1053 printk(KERN_WARNING "%s: peername failed (err %d)!\n", 1054 serv->sv_name, -err); 1055 goto failed; /* aborted connection or whatever */ 1056 } 1057 1058 /* Ideally, we would want to reject connections from unauthorized 1059 * hosts here, but when we get encryption, the IP of the host won't 1060 * tell us anything. For now just warn about unpriv connections. 1061 */ 1062 if (!svc_port_is_privileged(sin)) { 1063 dprintk(KERN_WARNING 1064 "%s: connect from unprivileged port: %s\n", 1065 serv->sv_name, 1066 __svc_print_addr(sin, buf, sizeof(buf))); 1067 } 1068 dprintk("%s: connect from %s\n", serv->sv_name, 1069 __svc_print_addr(sin, buf, sizeof(buf))); 1070 1071 /* make sure that a write doesn't block forever when 1072 * low on memory 1073 */ 1074 newsock->sk->sk_sndtimeo = HZ*30; 1075 1076 if (!(newsvsk = svc_setup_socket(serv, newsock, &err, 1077 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY)))) 1078 goto failed; 1079 memcpy(&newsvsk->sk_remote, sin, slen); 1080 newsvsk->sk_remotelen = slen; 1081 err = kernel_getsockname(newsock, sin, &slen); 1082 if (unlikely(err < 0)) { 1083 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err); 1084 slen = offsetof(struct sockaddr, sa_data); 1085 } 1086 memcpy(&newsvsk->sk_local, sin, slen); 1087 1088 svc_sock_received(newsvsk); 1089 1090 /* make sure that we don't have too many active connections. 1091 * If we have, something must be dropped. 1092 * 1093 * There's no point in trying to do random drop here for 1094 * DoS prevention. The NFS clients does 1 reconnect in 15 1095 * seconds. An attacker can easily beat that. 1096 * 1097 * The only somewhat efficient mechanism would be if drop 1098 * old connections from the same IP first. But right now 1099 * we don't even record the client IP in svc_sock. 1100 */ 1101 if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) { 1102 struct svc_sock *svsk = NULL; 1103 spin_lock_bh(&serv->sv_lock); 1104 if (!list_empty(&serv->sv_tempsocks)) { 1105 if (net_ratelimit()) { 1106 /* Try to help the admin */ 1107 printk(KERN_NOTICE "%s: too many open TCP " 1108 "sockets, consider increasing the " 1109 "number of nfsd threads\n", 1110 serv->sv_name); 1111 printk(KERN_NOTICE 1112 "%s: last TCP connect from %s\n", 1113 serv->sv_name, __svc_print_addr(sin, 1114 buf, sizeof(buf))); 1115 } 1116 /* 1117 * Always select the oldest socket. It's not fair, 1118 * but so is life 1119 */ 1120 svsk = list_entry(serv->sv_tempsocks.prev, 1121 struct svc_sock, 1122 sk_list); 1123 set_bit(SK_CLOSE, &svsk->sk_flags); 1124 atomic_inc(&svsk->sk_inuse); 1125 } 1126 spin_unlock_bh(&serv->sv_lock); 1127 1128 if (svsk) { 1129 svc_sock_enqueue(svsk); 1130 svc_sock_put(svsk); 1131 } 1132 1133 } 1134 1135 if (serv->sv_stats) 1136 serv->sv_stats->nettcpconn++; 1137 1138 return; 1139 1140 failed: 1141 sock_release(newsock); 1142 return; 1143 } 1144 1145 /* 1146 * Receive data from a TCP socket. 1147 */ 1148 static int 1149 svc_tcp_recvfrom(struct svc_rqst *rqstp) 1150 { 1151 struct svc_sock *svsk = rqstp->rq_sock; 1152 struct svc_serv *serv = svsk->sk_server; 1153 int len; 1154 struct kvec *vec; 1155 int pnum, vlen; 1156 1157 dprintk("svc: tcp_recv %p data %d conn %d close %d\n", 1158 svsk, test_bit(SK_DATA, &svsk->sk_flags), 1159 test_bit(SK_CONN, &svsk->sk_flags), 1160 test_bit(SK_CLOSE, &svsk->sk_flags)); 1161 1162 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) { 1163 svc_sock_received(svsk); 1164 return svc_deferred_recv(rqstp); 1165 } 1166 1167 if (test_bit(SK_CLOSE, &svsk->sk_flags)) { 1168 svc_delete_socket(svsk); 1169 return 0; 1170 } 1171 1172 if (svsk->sk_sk->sk_state == TCP_LISTEN) { 1173 svc_tcp_accept(svsk); 1174 svc_sock_received(svsk); 1175 return 0; 1176 } 1177 1178 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags)) 1179 /* sndbuf needs to have room for one request 1180 * per thread, otherwise we can stall even when the 1181 * network isn't a bottleneck. 1182 * 1183 * We count all threads rather than threads in a 1184 * particular pool, which provides an upper bound 1185 * on the number of threads which will access the socket. 1186 * 1187 * rcvbuf just needs to be able to hold a few requests. 1188 * Normally they will be removed from the queue 1189 * as soon a a complete request arrives. 1190 */ 1191 svc_sock_setbufsize(svsk->sk_sock, 1192 (serv->sv_nrthreads+3) * serv->sv_max_mesg, 1193 3 * serv->sv_max_mesg); 1194 1195 clear_bit(SK_DATA, &svsk->sk_flags); 1196 1197 /* Receive data. If we haven't got the record length yet, get 1198 * the next four bytes. Otherwise try to gobble up as much as 1199 * possible up to the complete record length. 1200 */ 1201 if (svsk->sk_tcplen < 4) { 1202 unsigned long want = 4 - svsk->sk_tcplen; 1203 struct kvec iov; 1204 1205 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen; 1206 iov.iov_len = want; 1207 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0) 1208 goto error; 1209 svsk->sk_tcplen += len; 1210 1211 if (len < want) { 1212 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n", 1213 len, want); 1214 svc_sock_received(svsk); 1215 return -EAGAIN; /* record header not complete */ 1216 } 1217 1218 svsk->sk_reclen = ntohl(svsk->sk_reclen); 1219 if (!(svsk->sk_reclen & 0x80000000)) { 1220 /* FIXME: technically, a record can be fragmented, 1221 * and non-terminal fragments will not have the top 1222 * bit set in the fragment length header. 1223 * But apparently no known nfs clients send fragmented 1224 * records. */ 1225 if (net_ratelimit()) 1226 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" 1227 " (non-terminal)\n", 1228 (unsigned long) svsk->sk_reclen); 1229 goto err_delete; 1230 } 1231 svsk->sk_reclen &= 0x7fffffff; 1232 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen); 1233 if (svsk->sk_reclen > serv->sv_max_mesg) { 1234 if (net_ratelimit()) 1235 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx" 1236 " (large)\n", 1237 (unsigned long) svsk->sk_reclen); 1238 goto err_delete; 1239 } 1240 } 1241 1242 /* Check whether enough data is available */ 1243 len = svc_recv_available(svsk); 1244 if (len < 0) 1245 goto error; 1246 1247 if (len < svsk->sk_reclen) { 1248 dprintk("svc: incomplete TCP record (%d of %d)\n", 1249 len, svsk->sk_reclen); 1250 svc_sock_received(svsk); 1251 return -EAGAIN; /* record not complete */ 1252 } 1253 len = svsk->sk_reclen; 1254 set_bit(SK_DATA, &svsk->sk_flags); 1255 1256 vec = rqstp->rq_vec; 1257 vec[0] = rqstp->rq_arg.head[0]; 1258 vlen = PAGE_SIZE; 1259 pnum = 1; 1260 while (vlen < len) { 1261 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]); 1262 vec[pnum].iov_len = PAGE_SIZE; 1263 pnum++; 1264 vlen += PAGE_SIZE; 1265 } 1266 rqstp->rq_respages = &rqstp->rq_pages[pnum]; 1267 1268 /* Now receive data */ 1269 len = svc_recvfrom(rqstp, vec, pnum, len); 1270 if (len < 0) 1271 goto error; 1272 1273 dprintk("svc: TCP complete record (%d bytes)\n", len); 1274 rqstp->rq_arg.len = len; 1275 rqstp->rq_arg.page_base = 0; 1276 if (len <= rqstp->rq_arg.head[0].iov_len) { 1277 rqstp->rq_arg.head[0].iov_len = len; 1278 rqstp->rq_arg.page_len = 0; 1279 } else { 1280 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len; 1281 } 1282 1283 rqstp->rq_skbuff = NULL; 1284 rqstp->rq_prot = IPPROTO_TCP; 1285 1286 /* Reset TCP read info */ 1287 svsk->sk_reclen = 0; 1288 svsk->sk_tcplen = 0; 1289 1290 svc_sock_received(svsk); 1291 if (serv->sv_stats) 1292 serv->sv_stats->nettcpcnt++; 1293 1294 return len; 1295 1296 err_delete: 1297 svc_delete_socket(svsk); 1298 return -EAGAIN; 1299 1300 error: 1301 if (len == -EAGAIN) { 1302 dprintk("RPC: TCP recvfrom got EAGAIN\n"); 1303 svc_sock_received(svsk); 1304 } else { 1305 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n", 1306 svsk->sk_server->sv_name, -len); 1307 goto err_delete; 1308 } 1309 1310 return len; 1311 } 1312 1313 /* 1314 * Send out data on TCP socket. 1315 */ 1316 static int 1317 svc_tcp_sendto(struct svc_rqst *rqstp) 1318 { 1319 struct xdr_buf *xbufp = &rqstp->rq_res; 1320 int sent; 1321 __be32 reclen; 1322 1323 /* Set up the first element of the reply kvec. 1324 * Any other kvecs that may be in use have been taken 1325 * care of by the server implementation itself. 1326 */ 1327 reclen = htonl(0x80000000|((xbufp->len ) - 4)); 1328 memcpy(xbufp->head[0].iov_base, &reclen, 4); 1329 1330 if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags)) 1331 return -ENOTCONN; 1332 1333 sent = svc_sendto(rqstp, &rqstp->rq_res); 1334 if (sent != xbufp->len) { 1335 printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n", 1336 rqstp->rq_sock->sk_server->sv_name, 1337 (sent<0)?"got error":"sent only", 1338 sent, xbufp->len); 1339 set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags); 1340 svc_sock_enqueue(rqstp->rq_sock); 1341 sent = -EAGAIN; 1342 } 1343 return sent; 1344 } 1345 1346 static void 1347 svc_tcp_init(struct svc_sock *svsk) 1348 { 1349 struct sock *sk = svsk->sk_sk; 1350 struct tcp_sock *tp = tcp_sk(sk); 1351 1352 svsk->sk_recvfrom = svc_tcp_recvfrom; 1353 svsk->sk_sendto = svc_tcp_sendto; 1354 1355 if (sk->sk_state == TCP_LISTEN) { 1356 dprintk("setting up TCP socket for listening\n"); 1357 sk->sk_data_ready = svc_tcp_listen_data_ready; 1358 set_bit(SK_CONN, &svsk->sk_flags); 1359 } else { 1360 dprintk("setting up TCP socket for reading\n"); 1361 sk->sk_state_change = svc_tcp_state_change; 1362 sk->sk_data_ready = svc_tcp_data_ready; 1363 sk->sk_write_space = svc_write_space; 1364 1365 svsk->sk_reclen = 0; 1366 svsk->sk_tcplen = 0; 1367 1368 tp->nonagle = 1; /* disable Nagle's algorithm */ 1369 1370 /* initialise setting must have enough space to 1371 * receive and respond to one request. 1372 * svc_tcp_recvfrom will re-adjust if necessary 1373 */ 1374 svc_sock_setbufsize(svsk->sk_sock, 1375 3 * svsk->sk_server->sv_max_mesg, 1376 3 * svsk->sk_server->sv_max_mesg); 1377 1378 set_bit(SK_CHNGBUF, &svsk->sk_flags); 1379 set_bit(SK_DATA, &svsk->sk_flags); 1380 if (sk->sk_state != TCP_ESTABLISHED) 1381 set_bit(SK_CLOSE, &svsk->sk_flags); 1382 } 1383 } 1384 1385 void 1386 svc_sock_update_bufs(struct svc_serv *serv) 1387 { 1388 /* 1389 * The number of server threads has changed. Update 1390 * rcvbuf and sndbuf accordingly on all sockets 1391 */ 1392 struct list_head *le; 1393 1394 spin_lock_bh(&serv->sv_lock); 1395 list_for_each(le, &serv->sv_permsocks) { 1396 struct svc_sock *svsk = 1397 list_entry(le, struct svc_sock, sk_list); 1398 set_bit(SK_CHNGBUF, &svsk->sk_flags); 1399 } 1400 list_for_each(le, &serv->sv_tempsocks) { 1401 struct svc_sock *svsk = 1402 list_entry(le, struct svc_sock, sk_list); 1403 set_bit(SK_CHNGBUF, &svsk->sk_flags); 1404 } 1405 spin_unlock_bh(&serv->sv_lock); 1406 } 1407 1408 /* 1409 * Receive the next request on any socket. This code is carefully 1410 * organised not to touch any cachelines in the shared svc_serv 1411 * structure, only cachelines in the local svc_pool. 1412 */ 1413 int 1414 svc_recv(struct svc_rqst *rqstp, long timeout) 1415 { 1416 struct svc_sock *svsk = NULL; 1417 struct svc_serv *serv = rqstp->rq_server; 1418 struct svc_pool *pool = rqstp->rq_pool; 1419 int len, i; 1420 int pages; 1421 struct xdr_buf *arg; 1422 DECLARE_WAITQUEUE(wait, current); 1423 1424 dprintk("svc: server %p waiting for data (to = %ld)\n", 1425 rqstp, timeout); 1426 1427 if (rqstp->rq_sock) 1428 printk(KERN_ERR 1429 "svc_recv: service %p, socket not NULL!\n", 1430 rqstp); 1431 if (waitqueue_active(&rqstp->rq_wait)) 1432 printk(KERN_ERR 1433 "svc_recv: service %p, wait queue active!\n", 1434 rqstp); 1435 1436 1437 /* now allocate needed pages. If we get a failure, sleep briefly */ 1438 pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE; 1439 for (i=0; i < pages ; i++) 1440 while (rqstp->rq_pages[i] == NULL) { 1441 struct page *p = alloc_page(GFP_KERNEL); 1442 if (!p) 1443 schedule_timeout_uninterruptible(msecs_to_jiffies(500)); 1444 rqstp->rq_pages[i] = p; 1445 } 1446 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */ 1447 BUG_ON(pages >= RPCSVC_MAXPAGES); 1448 1449 /* Make arg->head point to first page and arg->pages point to rest */ 1450 arg = &rqstp->rq_arg; 1451 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]); 1452 arg->head[0].iov_len = PAGE_SIZE; 1453 arg->pages = rqstp->rq_pages + 1; 1454 arg->page_base = 0; 1455 /* save at least one page for response */ 1456 arg->page_len = (pages-2)*PAGE_SIZE; 1457 arg->len = (pages-1)*PAGE_SIZE; 1458 arg->tail[0].iov_len = 0; 1459 1460 try_to_freeze(); 1461 cond_resched(); 1462 if (signalled()) 1463 return -EINTR; 1464 1465 spin_lock_bh(&pool->sp_lock); 1466 if ((svsk = svc_sock_dequeue(pool)) != NULL) { 1467 rqstp->rq_sock = svsk; 1468 atomic_inc(&svsk->sk_inuse); 1469 rqstp->rq_reserved = serv->sv_max_mesg; 1470 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved); 1471 } else { 1472 /* No data pending. Go to sleep */ 1473 svc_thread_enqueue(pool, rqstp); 1474 1475 /* 1476 * We have to be able to interrupt this wait 1477 * to bring down the daemons ... 1478 */ 1479 set_current_state(TASK_INTERRUPTIBLE); 1480 add_wait_queue(&rqstp->rq_wait, &wait); 1481 spin_unlock_bh(&pool->sp_lock); 1482 1483 schedule_timeout(timeout); 1484 1485 try_to_freeze(); 1486 1487 spin_lock_bh(&pool->sp_lock); 1488 remove_wait_queue(&rqstp->rq_wait, &wait); 1489 1490 if (!(svsk = rqstp->rq_sock)) { 1491 svc_thread_dequeue(pool, rqstp); 1492 spin_unlock_bh(&pool->sp_lock); 1493 dprintk("svc: server %p, no data yet\n", rqstp); 1494 return signalled()? -EINTR : -EAGAIN; 1495 } 1496 } 1497 spin_unlock_bh(&pool->sp_lock); 1498 1499 dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n", 1500 rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse)); 1501 len = svsk->sk_recvfrom(rqstp); 1502 dprintk("svc: got len=%d\n", len); 1503 1504 /* No data, incomplete (TCP) read, or accept() */ 1505 if (len == 0 || len == -EAGAIN) { 1506 rqstp->rq_res.len = 0; 1507 svc_sock_release(rqstp); 1508 return -EAGAIN; 1509 } 1510 svsk->sk_lastrecv = get_seconds(); 1511 clear_bit(SK_OLD, &svsk->sk_flags); 1512 1513 rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp)); 1514 rqstp->rq_chandle.defer = svc_defer; 1515 1516 if (serv->sv_stats) 1517 serv->sv_stats->netcnt++; 1518 return len; 1519 } 1520 1521 /* 1522 * Drop request 1523 */ 1524 void 1525 svc_drop(struct svc_rqst *rqstp) 1526 { 1527 dprintk("svc: socket %p dropped request\n", rqstp->rq_sock); 1528 svc_sock_release(rqstp); 1529 } 1530 1531 /* 1532 * Return reply to client. 1533 */ 1534 int 1535 svc_send(struct svc_rqst *rqstp) 1536 { 1537 struct svc_sock *svsk; 1538 int len; 1539 struct xdr_buf *xb; 1540 1541 if ((svsk = rqstp->rq_sock) == NULL) { 1542 printk(KERN_WARNING "NULL socket pointer in %s:%d\n", 1543 __FILE__, __LINE__); 1544 return -EFAULT; 1545 } 1546 1547 /* release the receive skb before sending the reply */ 1548 svc_release_skb(rqstp); 1549 1550 /* calculate over-all length */ 1551 xb = & rqstp->rq_res; 1552 xb->len = xb->head[0].iov_len + 1553 xb->page_len + 1554 xb->tail[0].iov_len; 1555 1556 /* Grab svsk->sk_mutex to serialize outgoing data. */ 1557 mutex_lock(&svsk->sk_mutex); 1558 if (test_bit(SK_DEAD, &svsk->sk_flags)) 1559 len = -ENOTCONN; 1560 else 1561 len = svsk->sk_sendto(rqstp); 1562 mutex_unlock(&svsk->sk_mutex); 1563 svc_sock_release(rqstp); 1564 1565 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN) 1566 return 0; 1567 return len; 1568 } 1569 1570 /* 1571 * Timer function to close old temporary sockets, using 1572 * a mark-and-sweep algorithm. 1573 */ 1574 static void 1575 svc_age_temp_sockets(unsigned long closure) 1576 { 1577 struct svc_serv *serv = (struct svc_serv *)closure; 1578 struct svc_sock *svsk; 1579 struct list_head *le, *next; 1580 LIST_HEAD(to_be_aged); 1581 1582 dprintk("svc_age_temp_sockets\n"); 1583 1584 if (!spin_trylock_bh(&serv->sv_lock)) { 1585 /* busy, try again 1 sec later */ 1586 dprintk("svc_age_temp_sockets: busy\n"); 1587 mod_timer(&serv->sv_temptimer, jiffies + HZ); 1588 return; 1589 } 1590 1591 list_for_each_safe(le, next, &serv->sv_tempsocks) { 1592 svsk = list_entry(le, struct svc_sock, sk_list); 1593 1594 if (!test_and_set_bit(SK_OLD, &svsk->sk_flags)) 1595 continue; 1596 if (atomic_read(&svsk->sk_inuse) > 1 || test_bit(SK_BUSY, &svsk->sk_flags)) 1597 continue; 1598 atomic_inc(&svsk->sk_inuse); 1599 list_move(le, &to_be_aged); 1600 set_bit(SK_CLOSE, &svsk->sk_flags); 1601 set_bit(SK_DETACHED, &svsk->sk_flags); 1602 } 1603 spin_unlock_bh(&serv->sv_lock); 1604 1605 while (!list_empty(&to_be_aged)) { 1606 le = to_be_aged.next; 1607 /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */ 1608 list_del_init(le); 1609 svsk = list_entry(le, struct svc_sock, sk_list); 1610 1611 dprintk("queuing svsk %p for closing, %lu seconds old\n", 1612 svsk, get_seconds() - svsk->sk_lastrecv); 1613 1614 /* a thread will dequeue and close it soon */ 1615 svc_sock_enqueue(svsk); 1616 svc_sock_put(svsk); 1617 } 1618 1619 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ); 1620 } 1621 1622 /* 1623 * Initialize socket for RPC use and create svc_sock struct 1624 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF. 1625 */ 1626 static struct svc_sock *svc_setup_socket(struct svc_serv *serv, 1627 struct socket *sock, 1628 int *errp, int flags) 1629 { 1630 struct svc_sock *svsk; 1631 struct sock *inet; 1632 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS); 1633 int is_temporary = flags & SVC_SOCK_TEMPORARY; 1634 1635 dprintk("svc: svc_setup_socket %p\n", sock); 1636 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) { 1637 *errp = -ENOMEM; 1638 return NULL; 1639 } 1640 1641 inet = sock->sk; 1642 1643 /* Register socket with portmapper */ 1644 if (*errp >= 0 && pmap_register) 1645 *errp = svc_register(serv, inet->sk_protocol, 1646 ntohs(inet_sk(inet)->sport)); 1647 1648 if (*errp < 0) { 1649 kfree(svsk); 1650 return NULL; 1651 } 1652 1653 set_bit(SK_BUSY, &svsk->sk_flags); 1654 inet->sk_user_data = svsk; 1655 svsk->sk_sock = sock; 1656 svsk->sk_sk = inet; 1657 svsk->sk_ostate = inet->sk_state_change; 1658 svsk->sk_odata = inet->sk_data_ready; 1659 svsk->sk_owspace = inet->sk_write_space; 1660 svsk->sk_server = serv; 1661 atomic_set(&svsk->sk_inuse, 1); 1662 svsk->sk_lastrecv = get_seconds(); 1663 spin_lock_init(&svsk->sk_lock); 1664 INIT_LIST_HEAD(&svsk->sk_deferred); 1665 INIT_LIST_HEAD(&svsk->sk_ready); 1666 mutex_init(&svsk->sk_mutex); 1667 1668 /* Initialize the socket */ 1669 if (sock->type == SOCK_DGRAM) 1670 svc_udp_init(svsk); 1671 else 1672 svc_tcp_init(svsk); 1673 1674 spin_lock_bh(&serv->sv_lock); 1675 if (is_temporary) { 1676 set_bit(SK_TEMP, &svsk->sk_flags); 1677 list_add(&svsk->sk_list, &serv->sv_tempsocks); 1678 serv->sv_tmpcnt++; 1679 if (serv->sv_temptimer.function == NULL) { 1680 /* setup timer to age temp sockets */ 1681 setup_timer(&serv->sv_temptimer, svc_age_temp_sockets, 1682 (unsigned long)serv); 1683 mod_timer(&serv->sv_temptimer, 1684 jiffies + svc_conn_age_period * HZ); 1685 } 1686 } else { 1687 clear_bit(SK_TEMP, &svsk->sk_flags); 1688 list_add(&svsk->sk_list, &serv->sv_permsocks); 1689 } 1690 spin_unlock_bh(&serv->sv_lock); 1691 1692 dprintk("svc: svc_setup_socket created %p (inet %p)\n", 1693 svsk, svsk->sk_sk); 1694 1695 return svsk; 1696 } 1697 1698 int svc_addsock(struct svc_serv *serv, 1699 int fd, 1700 char *name_return, 1701 int *proto) 1702 { 1703 int err = 0; 1704 struct socket *so = sockfd_lookup(fd, &err); 1705 struct svc_sock *svsk = NULL; 1706 1707 if (!so) 1708 return err; 1709 if (so->sk->sk_family != AF_INET) 1710 err = -EAFNOSUPPORT; 1711 else if (so->sk->sk_protocol != IPPROTO_TCP && 1712 so->sk->sk_protocol != IPPROTO_UDP) 1713 err = -EPROTONOSUPPORT; 1714 else if (so->state > SS_UNCONNECTED) 1715 err = -EISCONN; 1716 else { 1717 svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS); 1718 if (svsk) { 1719 svc_sock_received(svsk); 1720 err = 0; 1721 } 1722 } 1723 if (err) { 1724 sockfd_put(so); 1725 return err; 1726 } 1727 if (proto) *proto = so->sk->sk_protocol; 1728 return one_sock_name(name_return, svsk); 1729 } 1730 EXPORT_SYMBOL_GPL(svc_addsock); 1731 1732 /* 1733 * Create socket for RPC service. 1734 */ 1735 static int svc_create_socket(struct svc_serv *serv, int protocol, 1736 struct sockaddr *sin, int len, int flags) 1737 { 1738 struct svc_sock *svsk; 1739 struct socket *sock; 1740 int error; 1741 int type; 1742 char buf[RPC_MAX_ADDRBUFLEN]; 1743 1744 dprintk("svc: svc_create_socket(%s, %d, %s)\n", 1745 serv->sv_program->pg_name, protocol, 1746 __svc_print_addr(sin, buf, sizeof(buf))); 1747 1748 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) { 1749 printk(KERN_WARNING "svc: only UDP and TCP " 1750 "sockets supported\n"); 1751 return -EINVAL; 1752 } 1753 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM; 1754 1755 error = sock_create_kern(sin->sa_family, type, protocol, &sock); 1756 if (error < 0) 1757 return error; 1758 1759 svc_reclassify_socket(sock); 1760 1761 if (type == SOCK_STREAM) 1762 sock->sk->sk_reuse = 1; /* allow address reuse */ 1763 error = kernel_bind(sock, sin, len); 1764 if (error < 0) 1765 goto bummer; 1766 1767 if (protocol == IPPROTO_TCP) { 1768 if ((error = kernel_listen(sock, 64)) < 0) 1769 goto bummer; 1770 } 1771 1772 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) { 1773 svc_sock_received(svsk); 1774 return ntohs(inet_sk(svsk->sk_sk)->sport); 1775 } 1776 1777 bummer: 1778 dprintk("svc: svc_create_socket error = %d\n", -error); 1779 sock_release(sock); 1780 return error; 1781 } 1782 1783 /* 1784 * Remove a dead socket 1785 */ 1786 static void 1787 svc_delete_socket(struct svc_sock *svsk) 1788 { 1789 struct svc_serv *serv; 1790 struct sock *sk; 1791 1792 dprintk("svc: svc_delete_socket(%p)\n", svsk); 1793 1794 serv = svsk->sk_server; 1795 sk = svsk->sk_sk; 1796 1797 sk->sk_state_change = svsk->sk_ostate; 1798 sk->sk_data_ready = svsk->sk_odata; 1799 sk->sk_write_space = svsk->sk_owspace; 1800 1801 spin_lock_bh(&serv->sv_lock); 1802 1803 if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags)) 1804 list_del_init(&svsk->sk_list); 1805 /* 1806 * We used to delete the svc_sock from whichever list 1807 * it's sk_ready node was on, but we don't actually 1808 * need to. This is because the only time we're called 1809 * while still attached to a queue, the queue itself 1810 * is about to be destroyed (in svc_destroy). 1811 */ 1812 if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) { 1813 BUG_ON(atomic_read(&svsk->sk_inuse)<2); 1814 atomic_dec(&svsk->sk_inuse); 1815 if (test_bit(SK_TEMP, &svsk->sk_flags)) 1816 serv->sv_tmpcnt--; 1817 } 1818 1819 spin_unlock_bh(&serv->sv_lock); 1820 } 1821 1822 static void svc_close_socket(struct svc_sock *svsk) 1823 { 1824 set_bit(SK_CLOSE, &svsk->sk_flags); 1825 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) 1826 /* someone else will have to effect the close */ 1827 return; 1828 1829 atomic_inc(&svsk->sk_inuse); 1830 svc_delete_socket(svsk); 1831 clear_bit(SK_BUSY, &svsk->sk_flags); 1832 svc_sock_put(svsk); 1833 } 1834 1835 void svc_force_close_socket(struct svc_sock *svsk) 1836 { 1837 set_bit(SK_CLOSE, &svsk->sk_flags); 1838 if (test_bit(SK_BUSY, &svsk->sk_flags)) { 1839 /* Waiting to be processed, but no threads left, 1840 * So just remove it from the waiting list 1841 */ 1842 list_del_init(&svsk->sk_ready); 1843 clear_bit(SK_BUSY, &svsk->sk_flags); 1844 } 1845 svc_close_socket(svsk); 1846 } 1847 1848 /** 1849 * svc_makesock - Make a socket for nfsd and lockd 1850 * @serv: RPC server structure 1851 * @protocol: transport protocol to use 1852 * @port: port to use 1853 * @flags: requested socket characteristics 1854 * 1855 */ 1856 int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port, 1857 int flags) 1858 { 1859 struct sockaddr_in sin = { 1860 .sin_family = AF_INET, 1861 .sin_addr.s_addr = INADDR_ANY, 1862 .sin_port = htons(port), 1863 }; 1864 1865 dprintk("svc: creating socket proto = %d\n", protocol); 1866 return svc_create_socket(serv, protocol, (struct sockaddr *) &sin, 1867 sizeof(sin), flags); 1868 } 1869 1870 /* 1871 * Handle defer and revisit of requests 1872 */ 1873 1874 static void svc_revisit(struct cache_deferred_req *dreq, int too_many) 1875 { 1876 struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle); 1877 struct svc_sock *svsk; 1878 1879 if (too_many) { 1880 svc_sock_put(dr->svsk); 1881 kfree(dr); 1882 return; 1883 } 1884 dprintk("revisit queued\n"); 1885 svsk = dr->svsk; 1886 dr->svsk = NULL; 1887 spin_lock(&svsk->sk_lock); 1888 list_add(&dr->handle.recent, &svsk->sk_deferred); 1889 spin_unlock(&svsk->sk_lock); 1890 set_bit(SK_DEFERRED, &svsk->sk_flags); 1891 svc_sock_enqueue(svsk); 1892 svc_sock_put(svsk); 1893 } 1894 1895 static struct cache_deferred_req * 1896 svc_defer(struct cache_req *req) 1897 { 1898 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle); 1899 int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len); 1900 struct svc_deferred_req *dr; 1901 1902 if (rqstp->rq_arg.page_len) 1903 return NULL; /* if more than a page, give up FIXME */ 1904 if (rqstp->rq_deferred) { 1905 dr = rqstp->rq_deferred; 1906 rqstp->rq_deferred = NULL; 1907 } else { 1908 int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len; 1909 /* FIXME maybe discard if size too large */ 1910 dr = kmalloc(size, GFP_KERNEL); 1911 if (dr == NULL) 1912 return NULL; 1913 1914 dr->handle.owner = rqstp->rq_server; 1915 dr->prot = rqstp->rq_prot; 1916 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen); 1917 dr->addrlen = rqstp->rq_addrlen; 1918 dr->daddr = rqstp->rq_daddr; 1919 dr->argslen = rqstp->rq_arg.len >> 2; 1920 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2); 1921 } 1922 atomic_inc(&rqstp->rq_sock->sk_inuse); 1923 dr->svsk = rqstp->rq_sock; 1924 1925 dr->handle.revisit = svc_revisit; 1926 return &dr->handle; 1927 } 1928 1929 /* 1930 * recv data from a deferred request into an active one 1931 */ 1932 static int svc_deferred_recv(struct svc_rqst *rqstp) 1933 { 1934 struct svc_deferred_req *dr = rqstp->rq_deferred; 1935 1936 rqstp->rq_arg.head[0].iov_base = dr->args; 1937 rqstp->rq_arg.head[0].iov_len = dr->argslen<<2; 1938 rqstp->rq_arg.page_len = 0; 1939 rqstp->rq_arg.len = dr->argslen<<2; 1940 rqstp->rq_prot = dr->prot; 1941 memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen); 1942 rqstp->rq_addrlen = dr->addrlen; 1943 rqstp->rq_daddr = dr->daddr; 1944 rqstp->rq_respages = rqstp->rq_pages; 1945 return dr->argslen<<2; 1946 } 1947 1948 1949 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk) 1950 { 1951 struct svc_deferred_req *dr = NULL; 1952 1953 if (!test_bit(SK_DEFERRED, &svsk->sk_flags)) 1954 return NULL; 1955 spin_lock(&svsk->sk_lock); 1956 clear_bit(SK_DEFERRED, &svsk->sk_flags); 1957 if (!list_empty(&svsk->sk_deferred)) { 1958 dr = list_entry(svsk->sk_deferred.next, 1959 struct svc_deferred_req, 1960 handle.recent); 1961 list_del_init(&dr->handle.recent); 1962 set_bit(SK_DEFERRED, &svsk->sk_flags); 1963 } 1964 spin_unlock(&svsk->sk_lock); 1965 return dr; 1966 } 1967