1/* 2 * linux/net/sunrpc/xprtsock.c 3 * 4 * Client-side transport implementation for sockets. 5 * 6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com> 7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com> 8 * TCP NFS related read + write fixes 9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 10 * 11 * Rewrite of larges part of the code in order to stabilize TCP stuff. 12 * Fix behaviour when socket buffer is full. 13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no> 14 * 15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com> 16 */ 17 18#include <linux/types.h> 19#include <linux/slab.h> 20#include <linux/capability.h> 21#include <linux/pagemap.h> 22#include <linux/errno.h> 23#include <linux/socket.h> 24#include <linux/in.h> 25#include <linux/net.h> 26#include <linux/mm.h> 27#include <linux/udp.h> 28#include <linux/tcp.h> 29#include <linux/sunrpc/clnt.h> 30#include <linux/sunrpc/sched.h> 31#include <linux/file.h> 32 33#include <net/sock.h> 34#include <net/checksum.h> 35#include <net/udp.h> 36#include <net/tcp.h> 37 38/* 39 * xprtsock tunables 40 */ 41unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 42unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE; 43 44unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT; 45unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT; 46 47/* 48 * We can register our own files under /proc/sys/sunrpc by 49 * calling register_sysctl_table() again. The files in that 50 * directory become the union of all files registered there. 51 * 52 * We simply need to make sure that we don't collide with 53 * someone else's file names! 54 */ 55 56#ifdef RPC_DEBUG 57 58static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 59static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 60static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT; 61static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT; 62 63static struct ctl_table_header *sunrpc_table_header; 64 65static ctl_table xs_tunables_table[] = { 66 { 67 .ctl_name = CTL_SLOTTABLE_UDP, 68 .procname = "udp_slot_table_entries", 69 .data = &xprt_udp_slot_table_entries, 70 .maxlen = sizeof(unsigned int), 71 .mode = 0644, 72 .proc_handler = &proc_dointvec_minmax, 73 .strategy = &sysctl_intvec, 74 .extra1 = &min_slot_table_size, 75 .extra2 = &max_slot_table_size 76 }, 77 { 78 .ctl_name = CTL_SLOTTABLE_TCP, 79 .procname = "tcp_slot_table_entries", 80 .data = &xprt_tcp_slot_table_entries, 81 .maxlen = sizeof(unsigned int), 82 .mode = 0644, 83 .proc_handler = &proc_dointvec_minmax, 84 .strategy = &sysctl_intvec, 85 .extra1 = &min_slot_table_size, 86 .extra2 = &max_slot_table_size 87 }, 88 { 89 .ctl_name = CTL_MIN_RESVPORT, 90 .procname = "min_resvport", 91 .data = &xprt_min_resvport, 92 .maxlen = sizeof(unsigned int), 93 .mode = 0644, 94 .proc_handler = &proc_dointvec_minmax, 95 .strategy = &sysctl_intvec, 96 .extra1 = &xprt_min_resvport_limit, 97 .extra2 = &xprt_max_resvport_limit 98 }, 99 { 100 .ctl_name = CTL_MAX_RESVPORT, 101 .procname = "max_resvport", 102 .data = &xprt_max_resvport, 103 .maxlen = sizeof(unsigned int), 104 .mode = 0644, 105 .proc_handler = &proc_dointvec_minmax, 106 .strategy = &sysctl_intvec, 107 .extra1 = &xprt_min_resvport_limit, 108 .extra2 = &xprt_max_resvport_limit 109 }, 110 { 111 .ctl_name = 0, 112 }, 113}; 114 115static ctl_table sunrpc_table[] = { 116 { 117 .ctl_name = CTL_SUNRPC, 118 .procname = "sunrpc", 119 .mode = 0555, 120 .child = xs_tunables_table 121 }, 122 { 123 .ctl_name = 0, 124 }, 125}; 126 127#endif 128 129/* 130 * How many times to try sending a request on a socket before waiting 131 * for the socket buffer to clear. 132 */ 133#define XS_SENDMSG_RETRY (10U) 134 135/* 136 * Time out for an RPC UDP socket connect. UDP socket connects are 137 * synchronous, but we set a timeout anyway in case of resource 138 * exhaustion on the local host. 139 */ 140#define XS_UDP_CONN_TO (5U * HZ) 141 142/* 143 * Wait duration for an RPC TCP connection to be established. Solaris 144 * NFS over TCP uses 60 seconds, for example, which is in line with how 145 * long a server takes to reboot. 146 */ 147#define XS_TCP_CONN_TO (60U * HZ) 148 149/* 150 * Wait duration for a reply from the RPC portmapper. 151 */ 152#define XS_BIND_TO (60U * HZ) 153 154/* 155 * Delay if a UDP socket connect error occurs. This is most likely some 156 * kind of resource problem on the local host. 157 */ 158#define XS_UDP_REEST_TO (2U * HZ) 159 160/* 161 * The reestablish timeout allows clients to delay for a bit before attempting 162 * to reconnect to a server that just dropped our connection. 163 * 164 * We implement an exponential backoff when trying to reestablish a TCP 165 * transport connection with the server. Some servers like to drop a TCP 166 * connection when they are overworked, so we start with a short timeout and 167 * increase over time if the server is down or not responding. 168 */ 169#define XS_TCP_INIT_REEST_TO (3U * HZ) 170#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ) 171 172/* 173 * TCP idle timeout; client drops the transport socket if it is idle 174 * for this long. Note that we also timeout UDP sockets to prevent 175 * holding port numbers when there is no RPC traffic. 176 */ 177#define XS_IDLE_DISC_TO (5U * 60 * HZ) 178 179#ifdef RPC_DEBUG 180# undef RPC_DEBUG_DATA 181# define RPCDBG_FACILITY RPCDBG_TRANS 182#endif 183 184#ifdef RPC_DEBUG_DATA 185static void xs_pktdump(char *msg, u32 *packet, unsigned int count) 186{ 187 u8 *buf = (u8 *) packet; 188 int j; 189 190 dprintk("RPC: %s\n", msg); 191 for (j = 0; j < count && j < 128; j += 4) { 192 if (!(j & 31)) { 193 if (j) 194 dprintk("\n"); 195 dprintk("0x%04x ", j); 196 } 197 dprintk("%02x%02x%02x%02x ", 198 buf[j], buf[j+1], buf[j+2], buf[j+3]); 199 } 200 dprintk("\n"); 201} 202#else 203static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count) 204{ 205 /* NOP */ 206} 207#endif 208 209struct sock_xprt { 210 struct rpc_xprt xprt; 211 212 /* 213 * Network layer 214 */ 215 struct socket * sock; 216 struct sock * inet; 217 218 /* 219 * State of TCP reply receive 220 */ 221 __be32 tcp_fraghdr, 222 tcp_xid; 223 224 u32 tcp_offset, 225 tcp_reclen; 226 227 unsigned long tcp_copied, 228 tcp_flags; 229 230 /* 231 * Connection of transports 232 */ 233 struct delayed_work connect_worker; 234 unsigned short port; 235 236 /* 237 * UDP socket buffer size parameters 238 */ 239 size_t rcvsize, 240 sndsize; 241 242 /* 243 * Saved socket callback addresses 244 */ 245 void (*old_data_ready)(struct sock *, int); 246 void (*old_state_change)(struct sock *); 247 void (*old_write_space)(struct sock *); 248}; 249 250/* 251 * TCP receive state flags 252 */ 253#define TCP_RCV_LAST_FRAG (1UL << 0) 254#define TCP_RCV_COPY_FRAGHDR (1UL << 1) 255#define TCP_RCV_COPY_XID (1UL << 2) 256#define TCP_RCV_COPY_DATA (1UL << 3) 257 258static void xs_format_peer_addresses(struct rpc_xprt *xprt) 259{ 260 struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr; 261 char *buf; 262 263 buf = kzalloc(20, GFP_KERNEL); 264 if (buf) { 265 snprintf(buf, 20, "%u.%u.%u.%u", 266 NIPQUAD(addr->sin_addr.s_addr)); 267 } 268 xprt->address_strings[RPC_DISPLAY_ADDR] = buf; 269 270 buf = kzalloc(8, GFP_KERNEL); 271 if (buf) { 272 snprintf(buf, 8, "%u", 273 ntohs(addr->sin_port)); 274 } 275 xprt->address_strings[RPC_DISPLAY_PORT] = buf; 276 277 if (xprt->prot == IPPROTO_UDP) 278 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp"; 279 else 280 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp"; 281 282 buf = kzalloc(48, GFP_KERNEL); 283 if (buf) { 284 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s", 285 NIPQUAD(addr->sin_addr.s_addr), 286 ntohs(addr->sin_port), 287 xprt->prot == IPPROTO_UDP ? "udp" : "tcp"); 288 } 289 xprt->address_strings[RPC_DISPLAY_ALL] = buf; 290} 291 292static void xs_free_peer_addresses(struct rpc_xprt *xprt) 293{ 294 kfree(xprt->address_strings[RPC_DISPLAY_ADDR]); 295 kfree(xprt->address_strings[RPC_DISPLAY_PORT]); 296 kfree(xprt->address_strings[RPC_DISPLAY_ALL]); 297} 298 299#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL) 300 301static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more) 302{ 303 struct msghdr msg = { 304 .msg_name = addr, 305 .msg_namelen = addrlen, 306 .msg_flags = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0), 307 }; 308 struct kvec iov = { 309 .iov_base = vec->iov_base + base, 310 .iov_len = vec->iov_len - base, 311 }; 312 313 if (iov.iov_len != 0) 314 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len); 315 return kernel_sendmsg(sock, &msg, NULL, 0, 0); 316} 317 318static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more) 319{ 320 struct page **ppage; 321 unsigned int remainder; 322 int err, sent = 0; 323 324 remainder = xdr->page_len - base; 325 base += xdr->page_base; 326 ppage = xdr->pages + (base >> PAGE_SHIFT); 327 base &= ~PAGE_MASK; 328 for(;;) { 329 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder); 330 int flags = XS_SENDMSG_FLAGS; 331 332 remainder -= len; 333 if (remainder != 0 || more) 334 flags |= MSG_MORE; 335 err = sock->ops->sendpage(sock, *ppage, base, len, flags); 336 if (remainder == 0 || err != len) 337 break; 338 sent += err; 339 ppage++; 340 base = 0; 341 } 342 if (sent == 0) 343 return err; 344 if (err > 0) 345 sent += err; 346 return sent; 347} 348 349/** 350 * xs_sendpages - write pages directly to a socket 351 * @sock: socket to send on 352 * @addr: UDP only -- address of destination 353 * @addrlen: UDP only -- length of destination address 354 * @xdr: buffer containing this request 355 * @base: starting position in the buffer 356 * 357 */ 358static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base) 359{ 360 unsigned int remainder = xdr->len - base; 361 int err, sent = 0; 362 363 if (unlikely(!sock)) 364 return -ENOTCONN; 365 366 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags); 367 if (base != 0) { 368 addr = NULL; 369 addrlen = 0; 370 } 371 372 if (base < xdr->head[0].iov_len || addr != NULL) { 373 unsigned int len = xdr->head[0].iov_len - base; 374 remainder -= len; 375 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0); 376 if (remainder == 0 || err != len) 377 goto out; 378 sent += err; 379 base = 0; 380 } else 381 base -= xdr->head[0].iov_len; 382 383 if (base < xdr->page_len) { 384 unsigned int len = xdr->page_len - base; 385 remainder -= len; 386 err = xs_send_pagedata(sock, xdr, base, remainder != 0); 387 if (remainder == 0 || err != len) 388 goto out; 389 sent += err; 390 base = 0; 391 } else 392 base -= xdr->page_len; 393 394 if (base >= xdr->tail[0].iov_len) 395 return sent; 396 err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0); 397out: 398 if (sent == 0) 399 return err; 400 if (err > 0) 401 sent += err; 402 return sent; 403} 404 405/** 406 * xs_nospace - place task on wait queue if transmit was incomplete 407 * @task: task to put to sleep 408 * 409 */ 410static void xs_nospace(struct rpc_task *task) 411{ 412 struct rpc_rqst *req = task->tk_rqstp; 413 struct rpc_xprt *xprt = req->rq_xprt; 414 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 415 416 dprintk("RPC: %5u xmit incomplete (%u left of %u)\n", 417 task->tk_pid, req->rq_slen - req->rq_bytes_sent, 418 req->rq_slen); 419 420 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) { 421 /* Protect against races with write_space */ 422 spin_lock_bh(&xprt->transport_lock); 423 424 /* Don't race with disconnect */ 425 if (!xprt_connected(xprt)) 426 task->tk_status = -ENOTCONN; 427 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags)) 428 xprt_wait_for_buffer_space(task); 429 430 spin_unlock_bh(&xprt->transport_lock); 431 } else 432 /* Keep holding the socket if it is blocked */ 433 rpc_delay(task, HZ>>4); 434} 435 436/** 437 * xs_udp_send_request - write an RPC request to a UDP socket 438 * @task: address of RPC task that manages the state of an RPC request 439 * 440 * Return values: 441 * 0: The request has been sent 442 * EAGAIN: The socket was blocked, please call again later to 443 * complete the request 444 * ENOTCONN: Caller needs to invoke connect logic then call again 445 * other: Some other error occured, the request was not sent 446 */ 447static int xs_udp_send_request(struct rpc_task *task) 448{ 449 struct rpc_rqst *req = task->tk_rqstp; 450 struct rpc_xprt *xprt = req->rq_xprt; 451 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 452 struct xdr_buf *xdr = &req->rq_snd_buf; 453 int status; 454 455 xs_pktdump("packet data:", 456 req->rq_svec->iov_base, 457 req->rq_svec->iov_len); 458 459 req->rq_xtime = jiffies; 460 status = xs_sendpages(transport->sock, 461 (struct sockaddr *) &xprt->addr, 462 xprt->addrlen, xdr, 463 req->rq_bytes_sent); 464 465 dprintk("RPC: xs_udp_send_request(%u) = %d\n", 466 xdr->len - req->rq_bytes_sent, status); 467 468 if (likely(status >= (int) req->rq_slen)) 469 return 0; 470 471 /* Still some bytes left; set up for a retry later. */ 472 if (status > 0) 473 status = -EAGAIN; 474 475 switch (status) { 476 case -ENETUNREACH: 477 case -EPIPE: 478 case -ECONNREFUSED: 479 /* When the server has died, an ICMP port unreachable message 480 * prompts ECONNREFUSED. */ 481 break; 482 case -EAGAIN: 483 xs_nospace(task); 484 break; 485 default: 486 dprintk("RPC: sendmsg returned unrecognized error %d\n", 487 -status); 488 break; 489 } 490 491 return status; 492} 493 494static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf) 495{ 496 u32 reclen = buf->len - sizeof(rpc_fraghdr); 497 rpc_fraghdr *base = buf->head[0].iov_base; 498 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen); 499} 500 501static int xs_tcp_send_request(struct rpc_task *task) 502{ 503 struct rpc_rqst *req = task->tk_rqstp; 504 struct rpc_xprt *xprt = req->rq_xprt; 505 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 506 struct xdr_buf *xdr = &req->rq_snd_buf; 507 int status, retry = 0; 508 509 xs_encode_tcp_record_marker(&req->rq_snd_buf); 510 511 xs_pktdump("packet data:", 512 req->rq_svec->iov_base, 513 req->rq_svec->iov_len); 514 515 /* Continue transmitting the packet/record. We must be careful 516 * to cope with writespace callbacks arriving _after_ we have 517 * called sendmsg(). */ 518 while (1) { 519 req->rq_xtime = jiffies; 520 status = xs_sendpages(transport->sock, 521 NULL, 0, xdr, req->rq_bytes_sent); 522 523 dprintk("RPC: xs_tcp_send_request(%u) = %d\n", 524 xdr->len - req->rq_bytes_sent, status); 525 526 if (unlikely(status < 0)) 527 break; 528 529 /* If we've sent the entire packet, immediately 530 * reset the count of bytes sent. */ 531 req->rq_bytes_sent += status; 532 task->tk_bytes_sent += status; 533 if (likely(req->rq_bytes_sent >= req->rq_slen)) { 534 req->rq_bytes_sent = 0; 535 return 0; 536 } 537 538 status = -EAGAIN; 539 if (retry++ > XS_SENDMSG_RETRY) 540 break; 541 } 542 543 switch (status) { 544 case -EAGAIN: 545 xs_nospace(task); 546 break; 547 case -ECONNREFUSED: 548 case -ECONNRESET: 549 case -ENOTCONN: 550 case -EPIPE: 551 status = -ENOTCONN; 552 break; 553 default: 554 dprintk("RPC: sendmsg returned unrecognized error %d\n", 555 -status); 556 xprt_disconnect(xprt); 557 break; 558 } 559 560 return status; 561} 562 563/** 564 * xs_tcp_release_xprt - clean up after a tcp transmission 565 * @xprt: transport 566 * @task: rpc task 567 * 568 * This cleans up if an error causes us to abort the transmission of a request. 569 * In this case, the socket may need to be reset in order to avoid confusing 570 * the server. 571 */ 572static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task) 573{ 574 struct rpc_rqst *req; 575 576 if (task != xprt->snd_task) 577 return; 578 if (task == NULL) 579 goto out_release; 580 req = task->tk_rqstp; 581 if (req->rq_bytes_sent == 0) 582 goto out_release; 583 if (req->rq_bytes_sent == req->rq_snd_buf.len) 584 goto out_release; 585 set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state); 586out_release: 587 xprt_release_xprt(xprt, task); 588} 589 590/** 591 * xs_close - close a socket 592 * @xprt: transport 593 * 594 * This is used when all requests are complete; ie, no DRC state remains 595 * on the server we want to save. 596 */ 597static void xs_close(struct rpc_xprt *xprt) 598{ 599 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 600 struct socket *sock = transport->sock; 601 struct sock *sk = transport->inet; 602 603 if (!sk) 604 goto clear_close_wait; 605 606 dprintk("RPC: xs_close xprt %p\n", xprt); 607 608 write_lock_bh(&sk->sk_callback_lock); 609 transport->inet = NULL; 610 transport->sock = NULL; 611 612 sk->sk_user_data = NULL; 613 sk->sk_data_ready = transport->old_data_ready; 614 sk->sk_state_change = transport->old_state_change; 615 sk->sk_write_space = transport->old_write_space; 616 write_unlock_bh(&sk->sk_callback_lock); 617 618 sk->sk_no_check = 0; 619 620 sock_release(sock); 621clear_close_wait: 622 smp_mb__before_clear_bit(); 623 clear_bit(XPRT_CLOSE_WAIT, &xprt->state); 624 smp_mb__after_clear_bit(); 625} 626 627/** 628 * xs_destroy - prepare to shutdown a transport 629 * @xprt: doomed transport 630 * 631 */ 632static void xs_destroy(struct rpc_xprt *xprt) 633{ 634 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 635 636 dprintk("RPC: xs_destroy xprt %p\n", xprt); 637 638 cancel_delayed_work(&transport->connect_worker); 639 flush_scheduled_work(); 640 641 xprt_disconnect(xprt); 642 xs_close(xprt); 643 xs_free_peer_addresses(xprt); 644 kfree(xprt->slot); 645 kfree(xprt); 646} 647 648static inline struct rpc_xprt *xprt_from_sock(struct sock *sk) 649{ 650 return (struct rpc_xprt *) sk->sk_user_data; 651} 652 653/** 654 * xs_udp_data_ready - "data ready" callback for UDP sockets 655 * @sk: socket with data to read 656 * @len: how much data to read 657 * 658 */ 659static void xs_udp_data_ready(struct sock *sk, int len) 660{ 661 struct rpc_task *task; 662 struct rpc_xprt *xprt; 663 struct rpc_rqst *rovr; 664 struct sk_buff *skb; 665 int err, repsize, copied; 666 u32 _xid; 667 __be32 *xp; 668 669 read_lock(&sk->sk_callback_lock); 670 dprintk("RPC: xs_udp_data_ready...\n"); 671 if (!(xprt = xprt_from_sock(sk))) 672 goto out; 673 674 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL) 675 goto out; 676 677 if (xprt->shutdown) 678 goto dropit; 679 680 repsize = skb->len - sizeof(struct udphdr); 681 if (repsize < 4) { 682 dprintk("RPC: impossible RPC reply size %d!\n", repsize); 683 goto dropit; 684 } 685 686 /* Copy the XID from the skb... */ 687 xp = skb_header_pointer(skb, sizeof(struct udphdr), 688 sizeof(_xid), &_xid); 689 if (xp == NULL) 690 goto dropit; 691 692 /* Look up and lock the request corresponding to the given XID */ 693 spin_lock(&xprt->transport_lock); 694 rovr = xprt_lookup_rqst(xprt, *xp); 695 if (!rovr) 696 goto out_unlock; 697 task = rovr->rq_task; 698 699 if ((copied = rovr->rq_private_buf.buflen) > repsize) 700 copied = repsize; 701 702 /* Suck it into the iovec, verify checksum if not done by hw. */ 703 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) 704 goto out_unlock; 705 706 /* Something worked... */ 707 dst_confirm(skb->dst); 708 709 xprt_adjust_cwnd(task, copied); 710 xprt_update_rtt(task); 711 xprt_complete_rqst(task, copied); 712 713 out_unlock: 714 spin_unlock(&xprt->transport_lock); 715 dropit: 716 skb_free_datagram(sk, skb); 717 out: 718 read_unlock(&sk->sk_callback_lock); 719} 720 721static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 722{ 723 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 724 size_t len, used; 725 char *p; 726 727 p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset; 728 len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset; 729 used = xdr_skb_read_bits(desc, p, len); 730 transport->tcp_offset += used; 731 if (used != len) 732 return; 733 734 transport->tcp_reclen = ntohl(transport->tcp_fraghdr); 735 if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT) 736 transport->tcp_flags |= TCP_RCV_LAST_FRAG; 737 else 738 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG; 739 transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK; 740 741 transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR; 742 transport->tcp_offset = 0; 743 744 /* Sanity check of the record length */ 745 if (unlikely(transport->tcp_reclen < 4)) { 746 dprintk("RPC: invalid TCP record fragment length\n"); 747 xprt_disconnect(xprt); 748 return; 749 } 750 dprintk("RPC: reading TCP record fragment of length %d\n", 751 transport->tcp_reclen); 752} 753 754static void xs_tcp_check_fraghdr(struct sock_xprt *transport) 755{ 756 if (transport->tcp_offset == transport->tcp_reclen) { 757 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR; 758 transport->tcp_offset = 0; 759 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) { 760 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 761 transport->tcp_flags |= TCP_RCV_COPY_XID; 762 transport->tcp_copied = 0; 763 } 764 } 765} 766 767static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc) 768{ 769 size_t len, used; 770 char *p; 771 772 len = sizeof(transport->tcp_xid) - transport->tcp_offset; 773 dprintk("RPC: reading XID (%Zu bytes)\n", len); 774 p = ((char *) &transport->tcp_xid) + transport->tcp_offset; 775 used = xdr_skb_read_bits(desc, p, len); 776 transport->tcp_offset += used; 777 if (used != len) 778 return; 779 transport->tcp_flags &= ~TCP_RCV_COPY_XID; 780 transport->tcp_flags |= TCP_RCV_COPY_DATA; 781 transport->tcp_copied = 4; 782 dprintk("RPC: reading reply for XID %08x\n", 783 ntohl(transport->tcp_xid)); 784 xs_tcp_check_fraghdr(transport); 785} 786 787static inline void xs_tcp_read_request(struct rpc_xprt *xprt, struct xdr_skb_reader *desc) 788{ 789 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 790 struct rpc_rqst *req; 791 struct xdr_buf *rcvbuf; 792 size_t len; 793 ssize_t r; 794 795 /* Find and lock the request corresponding to this xid */ 796 spin_lock(&xprt->transport_lock); 797 req = xprt_lookup_rqst(xprt, transport->tcp_xid); 798 if (!req) { 799 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 800 dprintk("RPC: XID %08x request not found!\n", 801 ntohl(transport->tcp_xid)); 802 spin_unlock(&xprt->transport_lock); 803 return; 804 } 805 806 rcvbuf = &req->rq_private_buf; 807 len = desc->count; 808 if (len > transport->tcp_reclen - transport->tcp_offset) { 809 struct xdr_skb_reader my_desc; 810 811 len = transport->tcp_reclen - transport->tcp_offset; 812 memcpy(&my_desc, desc, sizeof(my_desc)); 813 my_desc.count = len; 814 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 815 &my_desc, xdr_skb_read_bits); 816 desc->count -= r; 817 desc->offset += r; 818 } else 819 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied, 820 desc, xdr_skb_read_bits); 821 822 if (r > 0) { 823 transport->tcp_copied += r; 824 transport->tcp_offset += r; 825 } 826 if (r != len) { 827 /* Error when copying to the receive buffer, 828 * usually because we weren't able to allocate 829 * additional buffer pages. All we can do now 830 * is turn off TCP_RCV_COPY_DATA, so the request 831 * will not receive any additional updates, 832 * and time out. 833 * Any remaining data from this record will 834 * be discarded. 835 */ 836 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 837 dprintk("RPC: XID %08x truncated request\n", 838 ntohl(transport->tcp_xid)); 839 dprintk("RPC: xprt = %p, tcp_copied = %lu, " 840 "tcp_offset = %u, tcp_reclen = %u\n", 841 xprt, transport->tcp_copied, 842 transport->tcp_offset, transport->tcp_reclen); 843 goto out; 844 } 845 846 dprintk("RPC: XID %08x read %Zd bytes\n", 847 ntohl(transport->tcp_xid), r); 848 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, " 849 "tcp_reclen = %u\n", xprt, transport->tcp_copied, 850 transport->tcp_offset, transport->tcp_reclen); 851 852 if (transport->tcp_copied == req->rq_private_buf.buflen) 853 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 854 else if (transport->tcp_offset == transport->tcp_reclen) { 855 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) 856 transport->tcp_flags &= ~TCP_RCV_COPY_DATA; 857 } 858 859out: 860 if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) 861 xprt_complete_rqst(req->rq_task, transport->tcp_copied); 862 spin_unlock(&xprt->transport_lock); 863 xs_tcp_check_fraghdr(transport); 864} 865 866static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc) 867{ 868 size_t len; 869 870 len = transport->tcp_reclen - transport->tcp_offset; 871 if (len > desc->count) 872 len = desc->count; 873 desc->count -= len; 874 desc->offset += len; 875 transport->tcp_offset += len; 876 dprintk("RPC: discarded %Zu bytes\n", len); 877 xs_tcp_check_fraghdr(transport); 878} 879 880static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len) 881{ 882 struct rpc_xprt *xprt = rd_desc->arg.data; 883 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 884 struct xdr_skb_reader desc = { 885 .skb = skb, 886 .offset = offset, 887 .count = len, 888 }; 889 890 dprintk("RPC: xs_tcp_data_recv started\n"); 891 do { 892 /* Read in a new fragment marker if necessary */ 893 /* Can we ever really expect to get completely empty fragments? */ 894 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) { 895 xs_tcp_read_fraghdr(xprt, &desc); 896 continue; 897 } 898 /* Read in the xid if necessary */ 899 if (transport->tcp_flags & TCP_RCV_COPY_XID) { 900 xs_tcp_read_xid(transport, &desc); 901 continue; 902 } 903 /* Read in the request data */ 904 if (transport->tcp_flags & TCP_RCV_COPY_DATA) { 905 xs_tcp_read_request(xprt, &desc); 906 continue; 907 } 908 /* Skip over any trailing bytes on short reads */ 909 xs_tcp_read_discard(transport, &desc); 910 } while (desc.count); 911 dprintk("RPC: xs_tcp_data_recv done\n"); 912 return len - desc.count; 913} 914 915/** 916 * xs_tcp_data_ready - "data ready" callback for TCP sockets 917 * @sk: socket with data to read 918 * @bytes: how much data to read 919 * 920 */ 921static void xs_tcp_data_ready(struct sock *sk, int bytes) 922{ 923 struct rpc_xprt *xprt; 924 read_descriptor_t rd_desc; 925 926 dprintk("RPC: xs_tcp_data_ready...\n"); 927 928 read_lock(&sk->sk_callback_lock); 929 if (!(xprt = xprt_from_sock(sk))) 930 goto out; 931 if (xprt->shutdown) 932 goto out; 933 934 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */ 935 rd_desc.arg.data = xprt; 936 rd_desc.count = 65536; 937 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv); 938out: 939 read_unlock(&sk->sk_callback_lock); 940} 941 942/** 943 * xs_tcp_state_change - callback to handle TCP socket state changes 944 * @sk: socket whose state has changed 945 * 946 */ 947static void xs_tcp_state_change(struct sock *sk) 948{ 949 struct rpc_xprt *xprt; 950 951 read_lock(&sk->sk_callback_lock); 952 if (!(xprt = xprt_from_sock(sk))) 953 goto out; 954 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt); 955 dprintk("RPC: state %x conn %d dead %d zapped %d\n", 956 sk->sk_state, xprt_connected(xprt), 957 sock_flag(sk, SOCK_DEAD), 958 sock_flag(sk, SOCK_ZAPPED)); 959 960 switch (sk->sk_state) { 961 case TCP_ESTABLISHED: 962 spin_lock_bh(&xprt->transport_lock); 963 if (!xprt_test_and_set_connected(xprt)) { 964 struct sock_xprt *transport = container_of(xprt, 965 struct sock_xprt, xprt); 966 967 /* Reset TCP record info */ 968 transport->tcp_offset = 0; 969 transport->tcp_reclen = 0; 970 transport->tcp_copied = 0; 971 transport->tcp_flags = 972 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID; 973 974 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 975 xprt_wake_pending_tasks(xprt, 0); 976 } 977 spin_unlock_bh(&xprt->transport_lock); 978 break; 979 case TCP_SYN_SENT: 980 case TCP_SYN_RECV: 981 break; 982 case TCP_CLOSE_WAIT: 983 /* Try to schedule an autoclose RPC calls */ 984 set_bit(XPRT_CLOSE_WAIT, &xprt->state); 985 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0) 986 schedule_work(&xprt->task_cleanup); 987 default: 988 xprt_disconnect(xprt); 989 } 990 out: 991 read_unlock(&sk->sk_callback_lock); 992} 993 994/** 995 * xs_udp_write_space - callback invoked when socket buffer space 996 * becomes available 997 * @sk: socket whose state has changed 998 * 999 * Called when more output buffer space is available for this socket. 1000 * We try not to wake our writers until they can make "significant" 1001 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1002 * with a bunch of small requests. 1003 */ 1004static void xs_udp_write_space(struct sock *sk) 1005{ 1006 read_lock(&sk->sk_callback_lock); 1007 1008 /* from net/core/sock.c:sock_def_write_space */ 1009 if (sock_writeable(sk)) { 1010 struct socket *sock; 1011 struct rpc_xprt *xprt; 1012 1013 if (unlikely(!(sock = sk->sk_socket))) 1014 goto out; 1015 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1016 goto out; 1017 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))) 1018 goto out; 1019 1020 xprt_write_space(xprt); 1021 } 1022 1023 out: 1024 read_unlock(&sk->sk_callback_lock); 1025} 1026 1027/** 1028 * xs_tcp_write_space - callback invoked when socket buffer space 1029 * becomes available 1030 * @sk: socket whose state has changed 1031 * 1032 * Called when more output buffer space is available for this socket. 1033 * We try not to wake our writers until they can make "significant" 1034 * progress, otherwise we'll waste resources thrashing kernel_sendmsg 1035 * with a bunch of small requests. 1036 */ 1037static void xs_tcp_write_space(struct sock *sk) 1038{ 1039 read_lock(&sk->sk_callback_lock); 1040 1041 /* from net/core/stream.c:sk_stream_write_space */ 1042 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) { 1043 struct socket *sock; 1044 struct rpc_xprt *xprt; 1045 1046 if (unlikely(!(sock = sk->sk_socket))) 1047 goto out; 1048 if (unlikely(!(xprt = xprt_from_sock(sk)))) 1049 goto out; 1050 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))) 1051 goto out; 1052 1053 xprt_write_space(xprt); 1054 } 1055 1056 out: 1057 read_unlock(&sk->sk_callback_lock); 1058} 1059 1060static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt) 1061{ 1062 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1063 struct sock *sk = transport->inet; 1064 1065 if (transport->rcvsize) { 1066 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1067 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2; 1068 } 1069 if (transport->sndsize) { 1070 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1071 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2; 1072 sk->sk_write_space(sk); 1073 } 1074} 1075 1076/** 1077 * xs_udp_set_buffer_size - set send and receive limits 1078 * @xprt: generic transport 1079 * @sndsize: requested size of send buffer, in bytes 1080 * @rcvsize: requested size of receive buffer, in bytes 1081 * 1082 * Set socket send and receive buffer size limits. 1083 */ 1084static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize) 1085{ 1086 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1087 1088 transport->sndsize = 0; 1089 if (sndsize) 1090 transport->sndsize = sndsize + 1024; 1091 transport->rcvsize = 0; 1092 if (rcvsize) 1093 transport->rcvsize = rcvsize + 1024; 1094 1095 xs_udp_do_set_buffer_size(xprt); 1096} 1097 1098/** 1099 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport 1100 * @task: task that timed out 1101 * 1102 * Adjust the congestion window after a retransmit timeout has occurred. 1103 */ 1104static void xs_udp_timer(struct rpc_task *task) 1105{ 1106 xprt_adjust_cwnd(task, -ETIMEDOUT); 1107} 1108 1109static unsigned short xs_get_random_port(void) 1110{ 1111 unsigned short range = xprt_max_resvport - xprt_min_resvport; 1112 unsigned short rand = (unsigned short) net_random() % range; 1113 return rand + xprt_min_resvport; 1114} 1115 1116/** 1117 * xs_set_port - reset the port number in the remote endpoint address 1118 * @xprt: generic transport 1119 * @port: new port number 1120 * 1121 */ 1122static void xs_set_port(struct rpc_xprt *xprt, unsigned short port) 1123{ 1124 struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr; 1125 1126 dprintk("RPC: setting port for xprt %p to %u\n", xprt, port); 1127 1128 sap->sin_port = htons(port); 1129} 1130 1131static int xs_bindresvport(struct sock_xprt *transport, struct socket *sock) 1132{ 1133 struct sockaddr_in myaddr = { 1134 .sin_family = AF_INET, 1135 }; 1136 int err; 1137 unsigned short port = transport->port; 1138 1139 do { 1140 myaddr.sin_port = htons(port); 1141 err = kernel_bind(sock, (struct sockaddr *) &myaddr, 1142 sizeof(myaddr)); 1143 if (err == 0) { 1144 transport->port = port; 1145 dprintk("RPC: xs_bindresvport bound to port %u\n", 1146 port); 1147 return 0; 1148 } 1149 if (port <= xprt_min_resvport) 1150 port = xprt_max_resvport; 1151 else 1152 port--; 1153 } while (err == -EADDRINUSE && port != transport->port); 1154 1155 dprintk("RPC: can't bind to reserved port (%d).\n", -err); 1156 return err; 1157} 1158 1159#ifdef CONFIG_DEBUG_LOCK_ALLOC 1160static struct lock_class_key xs_key[2]; 1161static struct lock_class_key xs_slock_key[2]; 1162 1163static inline void xs_reclassify_socket(struct socket *sock) 1164{ 1165 struct sock *sk = sock->sk; 1166 BUG_ON(sk->sk_lock.owner != NULL); 1167 switch (sk->sk_family) { 1168 case AF_INET: 1169 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFS", 1170 &xs_slock_key[0], "sk_lock-AF_INET-NFS", &xs_key[0]); 1171 break; 1172 1173 case AF_INET6: 1174 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFS", 1175 &xs_slock_key[1], "sk_lock-AF_INET6-NFS", &xs_key[1]); 1176 break; 1177 1178 default: 1179 BUG(); 1180 } 1181} 1182#else 1183static inline void xs_reclassify_socket(struct socket *sock) 1184{ 1185} 1186#endif 1187 1188/** 1189 * xs_udp_connect_worker - set up a UDP socket 1190 * @work: RPC transport to connect 1191 * 1192 * Invoked by a work queue tasklet. 1193 */ 1194static void xs_udp_connect_worker(struct work_struct *work) 1195{ 1196 struct sock_xprt *transport = 1197 container_of(work, struct sock_xprt, connect_worker.work); 1198 struct rpc_xprt *xprt = &transport->xprt; 1199 struct socket *sock = transport->sock; 1200 int err, status = -EIO; 1201 1202 if (xprt->shutdown || !xprt_bound(xprt)) 1203 goto out; 1204 1205 /* Start by resetting any existing state */ 1206 xs_close(xprt); 1207 1208 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) { 1209 dprintk("RPC: can't create UDP transport socket (%d).\n", -err); 1210 goto out; 1211 } 1212 xs_reclassify_socket(sock); 1213 1214 if (xprt->resvport && xs_bindresvport(transport, sock) < 0) { 1215 sock_release(sock); 1216 goto out; 1217 } 1218 1219 dprintk("RPC: worker connecting xprt %p to address: %s\n", 1220 xprt, xprt->address_strings[RPC_DISPLAY_ALL]); 1221 1222 if (!transport->inet) { 1223 struct sock *sk = sock->sk; 1224 1225 write_lock_bh(&sk->sk_callback_lock); 1226 1227 sk->sk_user_data = xprt; 1228 transport->old_data_ready = sk->sk_data_ready; 1229 transport->old_state_change = sk->sk_state_change; 1230 transport->old_write_space = sk->sk_write_space; 1231 sk->sk_data_ready = xs_udp_data_ready; 1232 sk->sk_write_space = xs_udp_write_space; 1233 sk->sk_no_check = UDP_CSUM_NORCV; 1234 sk->sk_allocation = GFP_ATOMIC; 1235 1236 xprt_set_connected(xprt); 1237 1238 /* Reset to new socket */ 1239 transport->sock = sock; 1240 transport->inet = sk; 1241 1242 write_unlock_bh(&sk->sk_callback_lock); 1243 } 1244 xs_udp_do_set_buffer_size(xprt); 1245 status = 0; 1246out: 1247 xprt_wake_pending_tasks(xprt, status); 1248 xprt_clear_connecting(xprt); 1249} 1250 1251/* 1252 * We need to preserve the port number so the reply cache on the server can 1253 * find our cached RPC replies when we get around to reconnecting. 1254 */ 1255static void xs_tcp_reuse_connection(struct rpc_xprt *xprt) 1256{ 1257 int result; 1258 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1259 struct sockaddr any; 1260 1261 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt); 1262 1263 /* 1264 * Disconnect the transport socket by doing a connect operation 1265 * with AF_UNSPEC. This should return immediately... 1266 */ 1267 memset(&any, 0, sizeof(any)); 1268 any.sa_family = AF_UNSPEC; 1269 result = kernel_connect(transport->sock, &any, sizeof(any), 0); 1270 if (result) 1271 dprintk("RPC: AF_UNSPEC connect return code %d\n", 1272 result); 1273} 1274 1275/** 1276 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint 1277 * @work: RPC transport to connect 1278 * 1279 * Invoked by a work queue tasklet. 1280 */ 1281static void xs_tcp_connect_worker(struct work_struct *work) 1282{ 1283 struct sock_xprt *transport = 1284 container_of(work, struct sock_xprt, connect_worker.work); 1285 struct rpc_xprt *xprt = &transport->xprt; 1286 struct socket *sock = transport->sock; 1287 int err, status = -EIO; 1288 1289 if (xprt->shutdown || !xprt_bound(xprt)) 1290 goto out; 1291 1292 if (!sock) { 1293 /* start from scratch */ 1294 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) { 1295 dprintk("RPC: can't create TCP transport " 1296 "socket (%d).\n", -err); 1297 goto out; 1298 } 1299 xs_reclassify_socket(sock); 1300 1301 if (xprt->resvport && xs_bindresvport(transport, sock) < 0) { 1302 sock_release(sock); 1303 goto out; 1304 } 1305 } else 1306 /* "close" the socket, preserving the local port */ 1307 xs_tcp_reuse_connection(xprt); 1308 1309 dprintk("RPC: worker connecting xprt %p to address: %s\n", 1310 xprt, xprt->address_strings[RPC_DISPLAY_ALL]); 1311 1312 if (!transport->inet) { 1313 struct sock *sk = sock->sk; 1314 1315 write_lock_bh(&sk->sk_callback_lock); 1316 1317 sk->sk_user_data = xprt; 1318 transport->old_data_ready = sk->sk_data_ready; 1319 transport->old_state_change = sk->sk_state_change; 1320 transport->old_write_space = sk->sk_write_space; 1321 sk->sk_data_ready = xs_tcp_data_ready; 1322 sk->sk_state_change = xs_tcp_state_change; 1323 sk->sk_write_space = xs_tcp_write_space; 1324 sk->sk_allocation = GFP_ATOMIC; 1325 1326 /* socket options */ 1327 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 1328 sock_reset_flag(sk, SOCK_LINGER); 1329 tcp_sk(sk)->linger2 = 0; 1330 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1331 1332 xprt_clear_connected(xprt); 1333 1334 /* Reset to new socket */ 1335 transport->sock = sock; 1336 transport->inet = sk; 1337 1338 write_unlock_bh(&sk->sk_callback_lock); 1339 } 1340 1341 /* Tell the socket layer to start connecting... */ 1342 xprt->stat.connect_count++; 1343 xprt->stat.connect_start = jiffies; 1344 status = kernel_connect(sock, (struct sockaddr *) &xprt->addr, 1345 xprt->addrlen, O_NONBLOCK); 1346 dprintk("RPC: %p connect status %d connected %d sock state %d\n", 1347 xprt, -status, xprt_connected(xprt), 1348 sock->sk->sk_state); 1349 if (status < 0) { 1350 switch (status) { 1351 case -EINPROGRESS: 1352 case -EALREADY: 1353 goto out_clear; 1354 case -ECONNREFUSED: 1355 case -ECONNRESET: 1356 /* retry with existing socket, after a delay */ 1357 break; 1358 default: 1359 /* get rid of existing socket, and retry */ 1360 xs_close(xprt); 1361 break; 1362 } 1363 } 1364out: 1365 xprt_wake_pending_tasks(xprt, status); 1366out_clear: 1367 xprt_clear_connecting(xprt); 1368} 1369 1370/** 1371 * xs_connect - connect a socket to a remote endpoint 1372 * @task: address of RPC task that manages state of connect request 1373 * 1374 * TCP: If the remote end dropped the connection, delay reconnecting. 1375 * 1376 * UDP socket connects are synchronous, but we use a work queue anyway 1377 * to guarantee that even unprivileged user processes can set up a 1378 * socket on a privileged port. 1379 * 1380 * If a UDP socket connect fails, the delay behavior here prevents 1381 * retry floods (hard mounts). 1382 */ 1383static void xs_connect(struct rpc_task *task) 1384{ 1385 struct rpc_xprt *xprt = task->tk_xprt; 1386 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1387 1388 if (xprt_test_and_set_connecting(xprt)) 1389 return; 1390 1391 if (transport->sock != NULL) { 1392 dprintk("RPC: xs_connect delayed xprt %p for %lu " 1393 "seconds\n", 1394 xprt, xprt->reestablish_timeout / HZ); 1395 schedule_delayed_work(&transport->connect_worker, 1396 xprt->reestablish_timeout); 1397 xprt->reestablish_timeout <<= 1; 1398 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO) 1399 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO; 1400 } else { 1401 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt); 1402 schedule_delayed_work(&transport->connect_worker, 0); 1403 1404 /* flush_scheduled_work can sleep... */ 1405 if (!RPC_IS_ASYNC(task)) 1406 flush_scheduled_work(); 1407 } 1408} 1409 1410/** 1411 * xs_udp_print_stats - display UDP socket-specifc stats 1412 * @xprt: rpc_xprt struct containing statistics 1413 * @seq: output file 1414 * 1415 */ 1416static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 1417{ 1418 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1419 1420 seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n", 1421 transport->port, 1422 xprt->stat.bind_count, 1423 xprt->stat.sends, 1424 xprt->stat.recvs, 1425 xprt->stat.bad_xids, 1426 xprt->stat.req_u, 1427 xprt->stat.bklog_u); 1428} 1429 1430/** 1431 * xs_tcp_print_stats - display TCP socket-specifc stats 1432 * @xprt: rpc_xprt struct containing statistics 1433 * @seq: output file 1434 * 1435 */ 1436static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq) 1437{ 1438 struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt); 1439 long idle_time = 0; 1440 1441 if (xprt_connected(xprt)) 1442 idle_time = (long)(jiffies - xprt->last_used) / HZ; 1443 1444 seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n", 1445 transport->port, 1446 xprt->stat.bind_count, 1447 xprt->stat.connect_count, 1448 xprt->stat.connect_time, 1449 idle_time, 1450 xprt->stat.sends, 1451 xprt->stat.recvs, 1452 xprt->stat.bad_xids, 1453 xprt->stat.req_u, 1454 xprt->stat.bklog_u); 1455} 1456 1457static struct rpc_xprt_ops xs_udp_ops = { 1458 .set_buffer_size = xs_udp_set_buffer_size, 1459 .reserve_xprt = xprt_reserve_xprt_cong, 1460 .release_xprt = xprt_release_xprt_cong, 1461 .rpcbind = rpcb_getport, 1462 .set_port = xs_set_port, 1463 .connect = xs_connect, 1464 .buf_alloc = rpc_malloc, 1465 .buf_free = rpc_free, 1466 .send_request = xs_udp_send_request, 1467 .set_retrans_timeout = xprt_set_retrans_timeout_rtt, 1468 .timer = xs_udp_timer, 1469 .release_request = xprt_release_rqst_cong, 1470 .close = xs_close, 1471 .destroy = xs_destroy, 1472 .print_stats = xs_udp_print_stats, 1473}; 1474 1475static struct rpc_xprt_ops xs_tcp_ops = { 1476 .reserve_xprt = xprt_reserve_xprt, 1477 .release_xprt = xs_tcp_release_xprt, 1478 .rpcbind = rpcb_getport, 1479 .set_port = xs_set_port, 1480 .connect = xs_connect, 1481 .buf_alloc = rpc_malloc, 1482 .buf_free = rpc_free, 1483 .send_request = xs_tcp_send_request, 1484 .set_retrans_timeout = xprt_set_retrans_timeout_def, 1485 .close = xs_close, 1486 .destroy = xs_destroy, 1487 .print_stats = xs_tcp_print_stats, 1488}; 1489 1490static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size) 1491{ 1492 struct rpc_xprt *xprt; 1493 struct sock_xprt *new; 1494 1495 if (addrlen > sizeof(xprt->addr)) { 1496 dprintk("RPC: xs_setup_xprt: address too large\n"); 1497 return ERR_PTR(-EBADF); 1498 } 1499 1500 new = kzalloc(sizeof(*new), GFP_KERNEL); 1501 if (new == NULL) { 1502 dprintk("RPC: xs_setup_xprt: couldn't allocate " 1503 "rpc_xprt\n"); 1504 return ERR_PTR(-ENOMEM); 1505 } 1506 xprt = &new->xprt; 1507 1508 xprt->max_reqs = slot_table_size; 1509 xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL); 1510 if (xprt->slot == NULL) { 1511 kfree(xprt); 1512 dprintk("RPC: xs_setup_xprt: couldn't allocate slot " 1513 "table\n"); 1514 return ERR_PTR(-ENOMEM); 1515 } 1516 1517 memcpy(&xprt->addr, addr, addrlen); 1518 xprt->addrlen = addrlen; 1519 new->port = xs_get_random_port(); 1520 1521 return xprt; 1522} 1523 1524/** 1525 * xs_setup_udp - Set up transport to use a UDP socket 1526 * @addr: address of remote server 1527 * @addrlen: length of address in bytes 1528 * @to: timeout parameters 1529 * 1530 */ 1531struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to) 1532{ 1533 struct rpc_xprt *xprt; 1534 struct sock_xprt *transport; 1535 1536 xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries); 1537 if (IS_ERR(xprt)) 1538 return xprt; 1539 transport = container_of(xprt, struct sock_xprt, xprt); 1540 1541 if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0) 1542 xprt_set_bound(xprt); 1543 1544 xprt->prot = IPPROTO_UDP; 1545 xprt->tsh_size = 0; 1546 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 1547 1548 INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_connect_worker); 1549 xprt->bind_timeout = XS_BIND_TO; 1550 xprt->connect_timeout = XS_UDP_CONN_TO; 1551 xprt->reestablish_timeout = XS_UDP_REEST_TO; 1552 xprt->idle_timeout = XS_IDLE_DISC_TO; 1553 1554 xprt->ops = &xs_udp_ops; 1555 1556 if (to) 1557 xprt->timeout = *to; 1558 else 1559 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ); 1560 1561 xs_format_peer_addresses(xprt); 1562 dprintk("RPC: set up transport to address %s\n", 1563 xprt->address_strings[RPC_DISPLAY_ALL]); 1564 1565 return xprt; 1566} 1567 1568/** 1569 * xs_setup_tcp - Set up transport to use a TCP socket 1570 * @addr: address of remote server 1571 * @addrlen: length of address in bytes 1572 * @to: timeout parameters 1573 * 1574 */ 1575struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to) 1576{ 1577 struct rpc_xprt *xprt; 1578 struct sock_xprt *transport; 1579 1580 xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries); 1581 if (IS_ERR(xprt)) 1582 return xprt; 1583 transport = container_of(xprt, struct sock_xprt, xprt); 1584 1585 if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0) 1586 xprt_set_bound(xprt); 1587 1588 xprt->prot = IPPROTO_TCP; 1589 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32); 1590 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE; 1591 1592 INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_connect_worker); 1593 xprt->bind_timeout = XS_BIND_TO; 1594 xprt->connect_timeout = XS_TCP_CONN_TO; 1595 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO; 1596 xprt->idle_timeout = XS_IDLE_DISC_TO; 1597 1598 xprt->ops = &xs_tcp_ops; 1599 1600 if (to) 1601 xprt->timeout = *to; 1602 else 1603 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ); 1604 1605 xs_format_peer_addresses(xprt); 1606 dprintk("RPC: set up transport to address %s\n", 1607 xprt->address_strings[RPC_DISPLAY_ALL]); 1608 1609 return xprt; 1610} 1611 1612/** 1613 * init_socket_xprt - set up xprtsock's sysctls 1614 * 1615 */ 1616int init_socket_xprt(void) 1617{ 1618#ifdef RPC_DEBUG 1619 if (!sunrpc_table_header) 1620 sunrpc_table_header = register_sysctl_table(sunrpc_table); 1621#endif 1622 1623 return 0; 1624} 1625 1626/** 1627 * cleanup_socket_xprt - remove xprtsock's sysctls 1628 * 1629 */ 1630void cleanup_socket_xprt(void) 1631{ 1632#ifdef RPC_DEBUG 1633 if (sunrpc_table_header) { 1634 unregister_sysctl_table(sunrpc_table_header); 1635 sunrpc_table_header = NULL; 1636 } 1637#endif 1638} 1639