net/sunrpc/xprtsock.c

   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/sched.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/errno.h>
  24 #include <linux/socket.h>
  25 #include <linux/in.h>
  26 #include <linux/net.h>
  27 #include <linux/mm.h>
  28 #include <linux/udp.h>
  29 #include <linux/tcp.h>
  30 #include <linux/sunrpc/clnt.h>
  31 #include <linux/sunrpc/sched.h>
  32 #include <linux/file.h>
  33
  34 #include <net/sock.h>
  35 #include <net/checksum.h>
  36 #include <net/udp.h>
  37 #include <net/tcp.h>
  38
  39 /*
  40  * xprtsock tunables
  41  */
  42 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  43 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  44
  45 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  46 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  47
  48 /*
  49  * How many times to try sending a request on a socket before waiting
  50  * for the socket buffer to clear.
  51  */
  52 #define XS_SENDMSG_RETRY        (10U)
  53
  54 /*
  55  * Time out for an RPC UDP socket connect.  UDP socket connects are
  56  * synchronous, but we set a timeout anyway in case of resource
  57  * exhaustion on the local host.
  58  */
  59 #define XS_UDP_CONN_TO          (5U * HZ)
  60
  61 /*
  62  * Wait duration for an RPC TCP connection to be established.  Solaris
  63  * NFS over TCP uses 60 seconds, for example, which is in line with how
  64  * long a server takes to reboot.
  65  */
  66 #define XS_TCP_CONN_TO          (60U * HZ)
  67
  68 /*
  69  * Wait duration for a reply from the RPC portmapper.
  70  */
  71 #define XS_BIND_TO              (60U * HZ)
  72
  73 /*
  74  * Delay if a UDP socket connect error occurs.  This is most likely some
  75  * kind of resource problem on the local host.
  76  */
  77 #define XS_UDP_REEST_TO         (2U * HZ)
  78
  79 /*
  80  * The reestablish timeout allows clients to delay for a bit before attempting
  81  * to reconnect to a server that just dropped our connection.
  82  *
  83  * We implement an exponential backoff when trying to reestablish a TCP
  84  * transport connection with the server.  Some servers like to drop a TCP
  85  * connection when they are overworked, so we start with a short timeout and
  86  * increase over time if the server is down or not responding.
  87  */
  88 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
  89 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
  90
  91 /*
  92  * TCP idle timeout; client drops the transport socket if it is idle
  93  * for this long.  Note that we also timeout UDP sockets to prevent
  94  * holding port numbers when there is no RPC traffic.
  95  */
  96 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
  97
  98 #ifdef RPC_DEBUG
  99 # undef  RPC_DEBUG_DATA
 100 # define RPCDBG_FACILITY        RPCDBG_TRANS
 101 #endif
 102
 103 #ifdef RPC_DEBUG_DATA
 104 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 105 {
 106         u8 *buf = (u8 *) packet;
 107         int j;
 108
 109         dprintk("RPC:      %s\n", msg);
 110         for (j = 0; j < count && j < 128; j += 4) {
 111                 if (!(j & 31)) {
 112                         if (j)
 113                                 dprintk("\n");
 114                         dprintk("0x%04x ", j);
 115                 }
 116                 dprintk("%02x%02x%02x%02x ",
 117                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 118         }
 119         dprintk("\n");
 120 }
 121 #else
 122 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 123 {
 124         /* NOP */
 125 }
 126 #endif
 127
 128 struct sock_xprt {
 129         struct rpc_xprt         xprt;
 130
 131         /*
 132          * Network layer
 133          */
 134         struct socket *         sock;
 135         struct sock *           inet;
 136
 137         /*
 138          * State of TCP reply receive
 139          */
 140         __be32                  tcp_fraghdr,
 141                                 tcp_xid;
 142
 143         u32                     tcp_offset,
 144                                 tcp_reclen;
 145
 146         unsigned long           tcp_copied,
 147                                 tcp_flags;
 148
 149         /*
 150          * Connection of transports
 151          */
 152         struct work_struct      connect_worker;
 153         unsigned short          port;
 154 };
 155
 156 /*
 157  * TCP receive state flags
 158  */
 159 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 160 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 161 #define TCP_RCV_COPY_XID        (1UL << 2)
 162 #define TCP_RCV_COPY_DATA       (1UL << 3)
 163
 164 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
 165 {
 166         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
 167         char *buf;
 168
 169         buf = kzalloc(20, GFP_KERNEL);
 170         if (buf) {
 171                 snprintf(buf, 20, "%u.%u.%u.%u",
 172                                 NIPQUAD(addr->sin_addr.s_addr));
 173         }
 174         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 175
 176         buf = kzalloc(8, GFP_KERNEL);
 177         if (buf) {
 178                 snprintf(buf, 8, "%u",
 179                                 ntohs(addr->sin_port));
 180         }
 181         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 182
 183         if (xprt->prot == IPPROTO_UDP)
 184                 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
 185         else
 186                 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
 187
 188         buf = kzalloc(48, GFP_KERNEL);
 189         if (buf) {
 190                 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
 191                         NIPQUAD(addr->sin_addr.s_addr),
 192                         ntohs(addr->sin_port),
 193                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 194         }
 195         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 196 }
 197
 198 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 199 {
 200         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
 201         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 202         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
 203 }
 204
 205 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 206
 207 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 208 {
 209         struct msghdr msg = {
 210                 .msg_name       = addr,
 211                 .msg_namelen    = addrlen,
 212                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 213         };
 214         struct kvec iov = {
 215                 .iov_base       = vec->iov_base + base,
 216                 .iov_len        = vec->iov_len - base,
 217         };
 218
 219         if (iov.iov_len != 0)
 220                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 221         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 222 }
 223
 224 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 225 {
 226         struct page **ppage;
 227         unsigned int remainder;
 228         int err, sent = 0;
 229
 230         remainder = xdr->page_len - base;
 231         base += xdr->page_base;
 232         ppage = xdr->pages + (base >> PAGE_SHIFT);
 233         base &= ~PAGE_MASK;
 234         for(;;) {
 235                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 236                 int flags = XS_SENDMSG_FLAGS;
 237
 238                 remainder -= len;
 239                 if (remainder != 0 || more)
 240                         flags |= MSG_MORE;
 241                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 242                 if (remainder == 0 || err != len)
 243                         break;
 244                 sent += err;
 245                 ppage++;
 246                 base = 0;
 247         }
 248         if (sent == 0)
 249                 return err;
 250         if (err > 0)
 251                 sent += err;
 252         return sent;
 253 }
 254
 255 /**
 256  * xs_sendpages - write pages directly to a socket
 257  * @sock: socket to send on
 258  * @addr: UDP only -- address of destination
 259  * @addrlen: UDP only -- length of destination address
 260  * @xdr: buffer containing this request
 261  * @base: starting position in the buffer
 262  *
 263  */
 264 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 265 {
 266         unsigned int remainder = xdr->len - base;
 267         int err, sent = 0;
 268
 269         if (unlikely(!sock))
 270                 return -ENOTCONN;
 271
 272         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 273         if (base != 0) {
 274                 addr = NULL;
 275                 addrlen = 0;
 276         }
 277
 278         if (base < xdr->head[0].iov_len || addr != NULL) {
 279                 unsigned int len = xdr->head[0].iov_len - base;
 280                 remainder -= len;
 281                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 282                 if (remainder == 0 || err != len)
 283                         goto out;
 284                 sent += err;
 285                 base = 0;
 286         } else
 287                 base -= xdr->head[0].iov_len;
 288
 289         if (base < xdr->page_len) {
 290                 unsigned int len = xdr->page_len - base;
 291                 remainder -= len;
 292                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 293                 if (remainder == 0 || err != len)
 294                         goto out;
 295                 sent += err;
 296                 base = 0;
 297         } else
 298                 base -= xdr->page_len;
 299
 300         if (base >= xdr->tail[0].iov_len)
 301                 return sent;
 302         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 303 out:
 304         if (sent == 0)
 305                 return err;
 306         if (err > 0)
 307                 sent += err;
 308         return sent;
 309 }
 310
 311 /**
 312  * xs_nospace - place task on wait queue if transmit was incomplete
 313  * @task: task to put to sleep
 314  *
 315  */
 316 static void xs_nospace(struct rpc_task *task)
 317 {
 318         struct rpc_rqst *req = task->tk_rqstp;
 319         struct rpc_xprt *xprt = req->rq_xprt;
 320         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 321
 322         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 323                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 324                         req->rq_slen);
 325
 326         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 327                 /* Protect against races with write_space */
 328                 spin_lock_bh(&xprt->transport_lock);
 329
 330                 /* Don't race with disconnect */
 331                 if (!xprt_connected(xprt))
 332                         task->tk_status = -ENOTCONN;
 333                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 334                         xprt_wait_for_buffer_space(task);
 335
 336                 spin_unlock_bh(&xprt->transport_lock);
 337         } else
 338                 /* Keep holding the socket if it is blocked */
 339                 rpc_delay(task, HZ>>4);
 340 }
 341
 342 /**
 343  * xs_udp_send_request - write an RPC request to a UDP socket
 344  * @task: address of RPC task that manages the state of an RPC request
 345  *
 346  * Return values:
 347  *        0:    The request has been sent
 348  *   EAGAIN:    The socket was blocked, please call again later to
 349  *              complete the request
 350  * ENOTCONN:    Caller needs to invoke connect logic then call again
 351  *    other:    Some other error occured, the request was not sent
 352  */
 353 static int xs_udp_send_request(struct rpc_task *task)
 354 {
 355         struct rpc_rqst *req = task->tk_rqstp;
 356         struct rpc_xprt *xprt = req->rq_xprt;
 357         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 358         struct xdr_buf *xdr = &req->rq_snd_buf;
 359         int status;
 360
 361         xs_pktdump("packet data:",
 362                                 req->rq_svec->iov_base,
 363                                 req->rq_svec->iov_len);
 364
 365         req->rq_xtime = jiffies;
 366         status = xs_sendpages(transport->sock,
 367                               (struct sockaddr *) &xprt->addr,
 368                               xprt->addrlen, xdr,
 369                               req->rq_bytes_sent);
 370
 371         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 372                         xdr->len - req->rq_bytes_sent, status);
 373
 374         if (likely(status >= (int) req->rq_slen))
 375                 return 0;
 376
 377         /* Still some bytes left; set up for a retry later. */
 378         if (status > 0)
 379                 status = -EAGAIN;
 380
 381         switch (status) {
 382         case -ENETUNREACH:
 383         case -EPIPE:
 384         case -ECONNREFUSED:
 385                 /* When the server has died, an ICMP port unreachable message
 386                  * prompts ECONNREFUSED. */
 387                 break;
 388         case -EAGAIN:
 389                 xs_nospace(task);
 390                 break;
 391         default:
 392                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 393                         -status);
 394                 break;
 395         }
 396
 397         return status;
 398 }
 399
 400 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 401 {
 402         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 403         rpc_fraghdr *base = buf->head[0].iov_base;
 404         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 405 }
 406
 407 /**
 408  * xs_tcp_send_request - write an RPC request to a TCP socket
 409  * @task: address of RPC task that manages the state of an RPC request
 410  *
 411  * Return values:
 412  *        0:    The request has been sent
 413  *   EAGAIN:    The socket was blocked, please call again later to
 414  *              complete the request
 415  * ENOTCONN:    Caller needs to invoke connect logic then call again
 416  *    other:    Some other error occured, the request was not sent
 417  *
 418  * XXX: In the case of soft timeouts, should we eventually give up
 419  *      if sendmsg is not able to make progress?
 420  */
 421 static int xs_tcp_send_request(struct rpc_task *task)
 422 {
 423         struct rpc_rqst *req = task->tk_rqstp;
 424         struct rpc_xprt *xprt = req->rq_xprt;
 425         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 426         struct xdr_buf *xdr = &req->rq_snd_buf;
 427         int status, retry = 0;
 428
 429         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 430
 431         xs_pktdump("packet data:",
 432                                 req->rq_svec->iov_base,
 433                                 req->rq_svec->iov_len);
 434
 435         /* Continue transmitting the packet/record. We must be careful
 436          * to cope with writespace callbacks arriving _after_ we have
 437          * called sendmsg(). */
 438         while (1) {
 439                 req->rq_xtime = jiffies;
 440                 status = xs_sendpages(transport->sock,
 441                                         NULL, 0, xdr, req->rq_bytes_sent);
 442
 443                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 444                                 xdr->len - req->rq_bytes_sent, status);
 445
 446                 if (unlikely(status < 0))
 447                         break;
 448
 449                 /* If we've sent the entire packet, immediately
 450                  * reset the count of bytes sent. */
 451                 req->rq_bytes_sent += status;
 452                 task->tk_bytes_sent += status;
 453                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 454                         req->rq_bytes_sent = 0;
 455                         return 0;
 456                 }
 457
 458                 status = -EAGAIN;
 459                 if (retry++ > XS_SENDMSG_RETRY)
 460                         break;
 461         }
 462
 463         switch (status) {
 464         case -EAGAIN:
 465                 xs_nospace(task);
 466                 break;
 467         case -ECONNREFUSED:
 468         case -ECONNRESET:
 469         case -ENOTCONN:
 470         case -EPIPE:
 471                 status = -ENOTCONN;
 472                 break;
 473         default:
 474                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 475                         -status);
 476                 xprt_disconnect(xprt);
 477                 break;
 478         }
 479
 480         return status;
 481 }
 482
 483 /**
 484  * xs_tcp_release_xprt - clean up after a tcp transmission
 485  * @xprt: transport
 486  * @task: rpc task
 487  *
 488  * This cleans up if an error causes us to abort the transmission of a request.
 489  * In this case, the socket may need to be reset in order to avoid confusing
 490  * the server.
 491  */
 492 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 493 {
 494         struct rpc_rqst *req;
 495
 496         if (task != xprt->snd_task)
 497                 return;
 498         if (task == NULL)
 499                 goto out_release;
 500         req = task->tk_rqstp;
 501         if (req->rq_bytes_sent == 0)
 502                 goto out_release;
 503         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 504                 goto out_release;
 505         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 506 out_release:
 507         xprt_release_xprt(xprt, task);
 508 }
 509
 510 /**
 511  * xs_close - close a socket
 512  * @xprt: transport
 513  *
 514  * This is used when all requests are complete; ie, no DRC state remains
 515  * on the server we want to save.
 516  */
 517 static void xs_close(struct rpc_xprt *xprt)
 518 {
 519         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 520         struct socket *sock = transport->sock;
 521         struct sock *sk = transport->inet;
 522
 523         if (!sk)
 524                 goto clear_close_wait;
 525
 526         dprintk("RPC:      xs_close xprt %p\n", xprt);
 527
 528         write_lock_bh(&sk->sk_callback_lock);
 529         transport->inet = NULL;
 530         transport->sock = NULL;
 531
 532         sk->sk_user_data = NULL;
 533         sk->sk_data_ready = xprt->old_data_ready;
 534         sk->sk_state_change = xprt->old_state_change;
 535         sk->sk_write_space = xprt->old_write_space;
 536         write_unlock_bh(&sk->sk_callback_lock);
 537
 538         sk->sk_no_check = 0;
 539
 540         sock_release(sock);
 541 clear_close_wait:
 542         smp_mb__before_clear_bit();
 543         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 544         smp_mb__after_clear_bit();
 545 }
 546
 547 /**
 548  * xs_destroy - prepare to shutdown a transport
 549  * @xprt: doomed transport
 550  *
 551  */
 552 static void xs_destroy(struct rpc_xprt *xprt)
 553 {
 554         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 555
 556         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 557
 558         cancel_delayed_work(&transport->connect_worker);
 559         flush_scheduled_work();
 560
 561         xprt_disconnect(xprt);
 562         xs_close(xprt);
 563         xs_free_peer_addresses(xprt);
 564         kfree(xprt->slot);
 565         kfree(xprt);
 566 }
 567
 568 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 569 {
 570         return (struct rpc_xprt *) sk->sk_user_data;
 571 }
 572
 573 /**
 574  * xs_udp_data_ready - "data ready" callback for UDP sockets
 575  * @sk: socket with data to read
 576  * @len: how much data to read
 577  *
 578  */
 579 static void xs_udp_data_ready(struct sock *sk, int len)
 580 {
 581         struct rpc_task *task;
 582         struct rpc_xprt *xprt;
 583         struct rpc_rqst *rovr;
 584         struct sk_buff *skb;
 585         int err, repsize, copied;
 586         u32 _xid;
 587         __be32 *xp;
 588
 589         read_lock(&sk->sk_callback_lock);
 590         dprintk("RPC:      xs_udp_data_ready...\n");
 591         if (!(xprt = xprt_from_sock(sk)))
 592                 goto out;
 593
 594         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 595                 goto out;
 596
 597         if (xprt->shutdown)
 598                 goto dropit;
 599
 600         repsize = skb->len - sizeof(struct udphdr);
 601         if (repsize < 4) {
 602                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 603                 goto dropit;
 604         }
 605
 606         /* Copy the XID from the skb... */
 607         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 608                                 sizeof(_xid), &_xid);
 609         if (xp == NULL)
 610                 goto dropit;
 611
 612         /* Look up and lock the request corresponding to the given XID */
 613         spin_lock(&xprt->transport_lock);
 614         rovr = xprt_lookup_rqst(xprt, *xp);
 615         if (!rovr)
 616                 goto out_unlock;
 617         task = rovr->rq_task;
 618
 619         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 620                 copied = repsize;
 621
 622         /* Suck it into the iovec, verify checksum if not done by hw. */
 623         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 624                 goto out_unlock;
 625
 626         /* Something worked... */
 627         dst_confirm(skb->dst);
 628
 629         xprt_adjust_cwnd(task, copied);
 630         xprt_update_rtt(task);
 631         xprt_complete_rqst(task, copied);
 632
 633  out_unlock:
 634         spin_unlock(&xprt->transport_lock);
 635  dropit:
 636         skb_free_datagram(sk, skb);
 637  out:
 638         read_unlock(&sk->sk_callback_lock);
 639 }
 640
 641 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 642 {
 643         if (len > desc->count)
 644                 len = desc->count;
 645         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 646                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 647                                 len, desc->count);
 648                 return 0;
 649         }
 650         desc->offset += len;
 651         desc->count -= len;
 652         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 653                         len, desc->count);
 654         return len;
 655 }
 656
 657 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 658 {
 659         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 660         size_t len, used;
 661         char *p;
 662
 663         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 664         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 665         used = xs_tcp_copy_data(desc, p, len);
 666         transport->tcp_offset += used;
 667         if (used != len)
 668                 return;
 669
 670         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 671         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 672                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 673         else
 674                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 675         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 676
 677         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 678         transport->tcp_offset = 0;
 679
 680         /* Sanity check of the record length */
 681         if (unlikely(transport->tcp_reclen < 4)) {
 682                 dprintk("RPC:      invalid TCP record fragment length\n");
 683                 xprt_disconnect(xprt);
 684                 return;
 685         }
 686         dprintk("RPC:      reading TCP record fragment of length %d\n",
 687                         transport->tcp_reclen);
 688 }
 689
 690 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 691 {
 692         if (transport->tcp_offset == transport->tcp_reclen) {
 693                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 694                 transport->tcp_offset = 0;
 695                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 696                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 697                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 698                         transport->tcp_copied = 0;
 699                 }
 700         }
 701 }
 702
 703 static inline void xs_tcp_read_xid(struct sock_xprt *transport, skb_reader_t *desc)
 704 {
 705         size_t len, used;
 706         char *p;
 707
 708         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 709         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 710         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 711         used = xs_tcp_copy_data(desc, p, len);
 712         transport->tcp_offset += used;
 713         if (used != len)
 714                 return;
 715         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 716         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 717         transport->tcp_copied = 4;
 718         dprintk("RPC:      reading reply for XID %08x\n",
 719                         ntohl(transport->tcp_xid));
 720         xs_tcp_check_fraghdr(transport);
 721 }
 722
 723 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 724 {
 725         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 726         struct rpc_rqst *req;
 727         struct xdr_buf *rcvbuf;
 728         size_t len;
 729         ssize_t r;
 730
 731         /* Find and lock the request corresponding to this xid */
 732         spin_lock(&xprt->transport_lock);
 733         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 734         if (!req) {
 735                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 736                 dprintk("RPC:      XID %08x request not found!\n",
 737                                 ntohl(transport->tcp_xid));
 738                 spin_unlock(&xprt->transport_lock);
 739                 return;
 740         }
 741
 742         rcvbuf = &req->rq_private_buf;
 743         len = desc->count;
 744         if (len > transport->tcp_reclen - transport->tcp_offset) {
 745                 skb_reader_t my_desc;
 746
 747                 len = transport->tcp_reclen - transport->tcp_offset;
 748                 memcpy(&my_desc, desc, sizeof(my_desc));
 749                 my_desc.count = len;
 750                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 751                                           &my_desc, xs_tcp_copy_data);
 752                 desc->count -= r;
 753                 desc->offset += r;
 754         } else
 755                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 756                                           desc, xs_tcp_copy_data);
 757
 758         if (r > 0) {
 759                 transport->tcp_copied += r;
 760                 transport->tcp_offset += r;
 761         }
 762         if (r != len) {
 763                 /* Error when copying to the receive buffer,
 764                  * usually because we weren't able to allocate
 765                  * additional buffer pages. All we can do now
 766                  * is turn off TCP_RCV_COPY_DATA, so the request
 767                  * will not receive any additional updates,
 768                  * and time out.
 769                  * Any remaining data from this record will
 770                  * be discarded.
 771                  */
 772                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 773                 dprintk("RPC:      XID %08x truncated request\n",
 774                                 ntohl(transport->tcp_xid));
 775                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 776                                 xprt, transport->tcp_copied, transport->tcp_offset,
 777                                         transport->tcp_reclen);
 778                 goto out;
 779         }
 780
 781         dprintk("RPC:      XID %08x read %Zd bytes\n",
 782                         ntohl(transport->tcp_xid), r);
 783         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 784                         xprt, transport->tcp_copied, transport->tcp_offset,
 785                                 transport->tcp_reclen);
 786
 787         if (transport->tcp_copied == req->rq_private_buf.buflen)
 788                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 789         else if (transport->tcp_offset == transport->tcp_reclen) {
 790                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
 791                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 792         }
 793
 794 out:
 795         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
 796                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
 797         spin_unlock(&xprt->transport_lock);
 798         xs_tcp_check_fraghdr(transport);
 799 }
 800
 801 static inline void xs_tcp_read_discard(struct sock_xprt *transport, skb_reader_t *desc)
 802 {
 803         size_t len;
 804
 805         len = transport->tcp_reclen - transport->tcp_offset;
 806         if (len > desc->count)
 807                 len = desc->count;
 808         desc->count -= len;
 809         desc->offset += len;
 810         transport->tcp_offset += len;
 811         dprintk("RPC:      discarded %Zu bytes\n", len);
 812         xs_tcp_check_fraghdr(transport);
 813 }
 814
 815 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 816 {
 817         struct rpc_xprt *xprt = rd_desc->arg.data;
 818         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 819         skb_reader_t desc = {
 820                 .skb    = skb,
 821                 .offset = offset,
 822                 .count  = len,
 823         };
 824
 825         dprintk("RPC:      xs_tcp_data_recv started\n");
 826         do {
 827                 /* Read in a new fragment marker if necessary */
 828                 /* Can we ever really expect to get completely empty fragments? */
 829                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
 830                         xs_tcp_read_fraghdr(xprt, &desc);
 831                         continue;
 832                 }
 833                 /* Read in the xid if necessary */
 834                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
 835                         xs_tcp_read_xid(transport, &desc);
 836                         continue;
 837                 }
 838                 /* Read in the request data */
 839                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
 840                         xs_tcp_read_request(xprt, &desc);
 841                         continue;
 842                 }
 843                 /* Skip over any trailing bytes on short reads */
 844                 xs_tcp_read_discard(transport, &desc);
 845         } while (desc.count);
 846         dprintk("RPC:      xs_tcp_data_recv done\n");
 847         return len - desc.count;
 848 }
 849
 850 /**
 851  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 852  * @sk: socket with data to read
 853  * @bytes: how much data to read
 854  *
 855  */
 856 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 857 {
 858         struct rpc_xprt *xprt;
 859         read_descriptor_t rd_desc;
 860
 861         read_lock(&sk->sk_callback_lock);
 862         dprintk("RPC:      xs_tcp_data_ready...\n");
 863         if (!(xprt = xprt_from_sock(sk)))
 864                 goto out;
 865         if (xprt->shutdown)
 866                 goto out;
 867
 868         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 869         rd_desc.arg.data = xprt;
 870         rd_desc.count = 65536;
 871         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 872 out:
 873         read_unlock(&sk->sk_callback_lock);
 874 }
 875
 876 /**
 877  * xs_tcp_state_change - callback to handle TCP socket state changes
 878  * @sk: socket whose state has changed
 879  *
 880  */
 881 static void xs_tcp_state_change(struct sock *sk)
 882 {
 883         struct rpc_xprt *xprt;
 884
 885         read_lock(&sk->sk_callback_lock);
 886         if (!(xprt = xprt_from_sock(sk)))
 887                 goto out;
 888         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 889         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 890                                 sk->sk_state, xprt_connected(xprt),
 891                                 sock_flag(sk, SOCK_DEAD),
 892                                 sock_flag(sk, SOCK_ZAPPED));
 893
 894         switch (sk->sk_state) {
 895         case TCP_ESTABLISHED:
 896                 spin_lock_bh(&xprt->transport_lock);
 897                 if (!xprt_test_and_set_connected(xprt)) {
 898                         struct sock_xprt *transport = container_of(xprt,
 899                                         struct sock_xprt, xprt);
 900
 901                         /* Reset TCP record info */
 902                         transport->tcp_offset = 0;
 903                         transport->tcp_reclen = 0;
 904                         transport->tcp_copied = 0;
 905                         transport->tcp_flags =
 906                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
 907
 908                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 909                         xprt_wake_pending_tasks(xprt, 0);
 910                 }
 911                 spin_unlock_bh(&xprt->transport_lock);
 912                 break;
 913         case TCP_SYN_SENT:
 914         case TCP_SYN_RECV:
 915                 break;
 916         case TCP_CLOSE_WAIT:
 917                 /* Try to schedule an autoclose RPC calls */
 918                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 919                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 920                         schedule_work(&xprt->task_cleanup);
 921         default:
 922                 xprt_disconnect(xprt);
 923         }
 924  out:
 925         read_unlock(&sk->sk_callback_lock);
 926 }
 927
 928 /**
 929  * xs_udp_write_space - callback invoked when socket buffer space
 930  *                             becomes available
 931  * @sk: socket whose state has changed
 932  *
 933  * Called when more output buffer space is available for this socket.
 934  * We try not to wake our writers until they can make "significant"
 935  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 936  * with a bunch of small requests.
 937  */
 938 static void xs_udp_write_space(struct sock *sk)
 939 {
 940         read_lock(&sk->sk_callback_lock);
 941
 942         /* from net/core/sock.c:sock_def_write_space */
 943         if (sock_writeable(sk)) {
 944                 struct socket *sock;
 945                 struct rpc_xprt *xprt;
 946
 947                 if (unlikely(!(sock = sk->sk_socket)))
 948                         goto out;
 949                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 950                         goto out;
 951                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 952                         goto out;
 953
 954                 xprt_write_space(xprt);
 955         }
 956
 957  out:
 958         read_unlock(&sk->sk_callback_lock);
 959 }
 960
 961 /**
 962  * xs_tcp_write_space - callback invoked when socket buffer space
 963  *                             becomes available
 964  * @sk: socket whose state has changed
 965  *
 966  * Called when more output buffer space is available for this socket.
 967  * We try not to wake our writers until they can make "significant"
 968  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 969  * with a bunch of small requests.
 970  */
 971 static void xs_tcp_write_space(struct sock *sk)
 972 {
 973         read_lock(&sk->sk_callback_lock);
 974
 975         /* from net/core/stream.c:sk_stream_write_space */
 976         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 977                 struct socket *sock;
 978                 struct rpc_xprt *xprt;
 979
 980                 if (unlikely(!(sock = sk->sk_socket)))
 981                         goto out;
 982                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 983                         goto out;
 984                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 985                         goto out;
 986
 987                 xprt_write_space(xprt);
 988         }
 989
 990  out:
 991         read_unlock(&sk->sk_callback_lock);
 992 }
 993
 994 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 995 {
 996         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 997         struct sock *sk = transport->inet;
 998
 999         if (xprt->rcvsize) {
1000                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1001                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
1002         }
1003         if (xprt->sndsize) {
1004                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1005                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
1006                 sk->sk_write_space(sk);
1007         }
1008 }
1009
1010 /**
1011  * xs_udp_set_buffer_size - set send and receive limits
1012  * @xprt: generic transport
1013  * @sndsize: requested size of send buffer, in bytes
1014  * @rcvsize: requested size of receive buffer, in bytes
1015  *
1016  * Set socket send and receive buffer size limits.
1017  */
1018 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1019 {
1020         xprt->sndsize = 0;
1021         if (sndsize)
1022                 xprt->sndsize = sndsize + 1024;
1023         xprt->rcvsize = 0;
1024         if (rcvsize)
1025                 xprt->rcvsize = rcvsize + 1024;
1026
1027         xs_udp_do_set_buffer_size(xprt);
1028 }
1029
1030 /**
1031  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1032  * @task: task that timed out
1033  *
1034  * Adjust the congestion window after a retransmit timeout has occurred.
1035  */
1036 static void xs_udp_timer(struct rpc_task *task)
1037 {
1038         xprt_adjust_cwnd(task, -ETIMEDOUT);
1039 }
1040
1041 static unsigned short xs_get_random_port(void)
1042 {
1043         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1044         unsigned short rand = (unsigned short) net_random() % range;
1045         return rand + xprt_min_resvport;
1046 }
1047
1048 /**
1049  * xs_print_peer_address - format an IPv4 address for printing
1050  * @xprt: generic transport
1051  * @format: flags field indicating which parts of the address to render
1052  */
1053 static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format)
1054 {
1055         if (xprt->address_strings[format] != NULL)
1056                 return xprt->address_strings[format];
1057         else
1058                 return "unprintable";
1059 }
1060
1061 /**
1062  * xs_set_port - reset the port number in the remote endpoint address
1063  * @xprt: generic transport
1064  * @port: new port number
1065  *
1066  */
1067 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1068 {
1069         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1070
1071         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
1072
1073         sap->sin_port = htons(port);
1074 }
1075
1076 static int xs_bindresvport(struct sock_xprt *transport, struct socket *sock)
1077 {
1078         struct sockaddr_in myaddr = {
1079                 .sin_family = AF_INET,
1080         };
1081         int err;
1082         unsigned short port = transport->port;
1083
1084         do {
1085                 myaddr.sin_port = htons(port);
1086                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1087                                                 sizeof(myaddr));
1088                 if (err == 0) {
1089                         transport->port = port;
1090                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
1091                                         port);
1092                         return 0;
1093                 }
1094                 if (port <= xprt_min_resvport)
1095                         port = xprt_max_resvport;
1096                 else
1097                         port--;
1098         } while (err == -EADDRINUSE && port != transport->port);
1099
1100         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
1101         return err;
1102 }
1103
1104 /**
1105  * xs_udp_connect_worker - set up a UDP socket
1106  * @args: RPC transport to connect
1107  *
1108  * Invoked by a work queue tasklet.
1109  */
1110 static void xs_udp_connect_worker(void *args)
1111 {
1112         struct sock_xprt *transport = (struct sock_xprt *)args;
1113         struct rpc_xprt *xprt = &transport->xprt;
1114         struct socket *sock = transport->sock;
1115         int err, status = -EIO;
1116
1117         if (xprt->shutdown || !xprt_bound(xprt))
1118                 goto out;
1119
1120         /* Start by resetting any existing state */
1121         xs_close(xprt);
1122
1123         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1124                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1125                 goto out;
1126         }
1127
1128         if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1129                 sock_release(sock);
1130                 goto out;
1131         }
1132
1133         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1134                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1135
1136         if (!transport->inet) {
1137                 struct sock *sk = sock->sk;
1138
1139                 write_lock_bh(&sk->sk_callback_lock);
1140
1141                 sk->sk_user_data = xprt;
1142                 xprt->old_data_ready = sk->sk_data_ready;
1143                 xprt->old_state_change = sk->sk_state_change;
1144                 xprt->old_write_space = sk->sk_write_space;
1145                 sk->sk_data_ready = xs_udp_data_ready;
1146                 sk->sk_write_space = xs_udp_write_space;
1147                 sk->sk_no_check = UDP_CSUM_NORCV;
1148                 sk->sk_allocation = GFP_ATOMIC;
1149
1150                 xprt_set_connected(xprt);
1151
1152                 /* Reset to new socket */
1153                 transport->sock = sock;
1154                 transport->inet = sk;
1155
1156                 write_unlock_bh(&sk->sk_callback_lock);
1157         }
1158         xs_udp_do_set_buffer_size(xprt);
1159         status = 0;
1160 out:
1161         xprt_wake_pending_tasks(xprt, status);
1162         xprt_clear_connecting(xprt);
1163 }
1164
1165 /*
1166  * We need to preserve the port number so the reply cache on the server can
1167  * find our cached RPC replies when we get around to reconnecting.
1168  */
1169 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1170 {
1171         int result;
1172         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1173         struct sockaddr any;
1174
1175         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1176
1177         /*
1178          * Disconnect the transport socket by doing a connect operation
1179          * with AF_UNSPEC.  This should return immediately...
1180          */
1181         memset(&any, 0, sizeof(any));
1182         any.sa_family = AF_UNSPEC;
1183         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1184         if (result)
1185                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1186                                 result);
1187 }
1188
1189 /**
1190  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1191  * @args: RPC transport to connect
1192  *
1193  * Invoked by a work queue tasklet.
1194  */
1195 static void xs_tcp_connect_worker(void *args)
1196 {
1197         struct sock_xprt *transport = (struct sock_xprt *)args;
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         if (!sock) {
1206                 /* start from scratch */
1207                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1208                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1209                         goto out;
1210                 }
1211
1212                 if (xprt->resvport && xs_bindresvport(transport, sock) < 0) {
1213                         sock_release(sock);
1214                         goto out;
1215                 }
1216         } else
1217                 /* "close" the socket, preserving the local port */
1218                 xs_tcp_reuse_connection(xprt);
1219
1220         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1221                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1222
1223         if (!transport->inet) {
1224                 struct sock *sk = sock->sk;
1225
1226                 write_lock_bh(&sk->sk_callback_lock);
1227
1228                 sk->sk_user_data = xprt;
1229                 xprt->old_data_ready = sk->sk_data_ready;
1230                 xprt->old_state_change = sk->sk_state_change;
1231                 xprt->old_write_space = sk->sk_write_space;
1232                 sk->sk_data_ready = xs_tcp_data_ready;
1233                 sk->sk_state_change = xs_tcp_state_change;
1234                 sk->sk_write_space = xs_tcp_write_space;
1235                 sk->sk_allocation = GFP_ATOMIC;
1236
1237                 /* socket options */
1238                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1239                 sock_reset_flag(sk, SOCK_LINGER);
1240                 tcp_sk(sk)->linger2 = 0;
1241                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1242
1243                 xprt_clear_connected(xprt);
1244
1245                 /* Reset to new socket */
1246                 transport->sock = sock;
1247                 transport->inet = sk;
1248
1249                 write_unlock_bh(&sk->sk_callback_lock);
1250         }
1251
1252         /* Tell the socket layer to start connecting... */
1253         xprt->stat.connect_count++;
1254         xprt->stat.connect_start = jiffies;
1255         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1256                         xprt->addrlen, O_NONBLOCK);
1257         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1258                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1259         if (status < 0) {
1260                 switch (status) {
1261                         case -EINPROGRESS:
1262                         case -EALREADY:
1263                                 goto out_clear;
1264                         case -ECONNREFUSED:
1265                         case -ECONNRESET:
1266                                 /* retry with existing socket, after a delay */
1267                                 break;
1268                         default:
1269                                 /* get rid of existing socket, and retry */
1270                                 xs_close(xprt);
1271                                 break;
1272                 }
1273         }
1274 out:
1275         xprt_wake_pending_tasks(xprt, status);
1276 out_clear:
1277         xprt_clear_connecting(xprt);
1278 }
1279
1280 /**
1281  * xs_connect - connect a socket to a remote endpoint
1282  * @task: address of RPC task that manages state of connect request
1283  *
1284  * TCP: If the remote end dropped the connection, delay reconnecting.
1285  *
1286  * UDP socket connects are synchronous, but we use a work queue anyway
1287  * to guarantee that even unprivileged user processes can set up a
1288  * socket on a privileged port.
1289  *
1290  * If a UDP socket connect fails, the delay behavior here prevents
1291  * retry floods (hard mounts).
1292  */
1293 static void xs_connect(struct rpc_task *task)
1294 {
1295         struct rpc_xprt *xprt = task->tk_xprt;
1296         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1297
1298         if (xprt_test_and_set_connecting(xprt))
1299                 return;
1300
1301         if (transport->sock != NULL) {
1302                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1303                                 xprt, xprt->reestablish_timeout / HZ);
1304                 schedule_delayed_work(&transport->connect_worker,
1305                                         xprt->reestablish_timeout);
1306                 xprt->reestablish_timeout <<= 1;
1307                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1308                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1309         } else {
1310                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1311                 schedule_work(&transport->connect_worker);
1312
1313                 /* flush_scheduled_work can sleep... */
1314                 if (!RPC_IS_ASYNC(task))
1315                         flush_scheduled_work();
1316         }
1317 }
1318
1319 /**
1320  * xs_udp_print_stats - display UDP socket-specifc stats
1321  * @xprt: rpc_xprt struct containing statistics
1322  * @seq: output file
1323  *
1324  */
1325 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1326 {
1327         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1328
1329         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1330                         transport->port,
1331                         xprt->stat.bind_count,
1332                         xprt->stat.sends,
1333                         xprt->stat.recvs,
1334                         xprt->stat.bad_xids,
1335                         xprt->stat.req_u,
1336                         xprt->stat.bklog_u);
1337 }
1338
1339 /**
1340  * xs_tcp_print_stats - display TCP socket-specifc stats
1341  * @xprt: rpc_xprt struct containing statistics
1342  * @seq: output file
1343  *
1344  */
1345 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1346 {
1347         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1348         long idle_time = 0;
1349
1350         if (xprt_connected(xprt))
1351                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1352
1353         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1354                         transport->port,
1355                         xprt->stat.bind_count,
1356                         xprt->stat.connect_count,
1357                         xprt->stat.connect_time,
1358                         idle_time,
1359                         xprt->stat.sends,
1360                         xprt->stat.recvs,
1361                         xprt->stat.bad_xids,
1362                         xprt->stat.req_u,
1363                         xprt->stat.bklog_u);
1364 }
1365
1366 static struct rpc_xprt_ops xs_udp_ops = {
1367         .set_buffer_size        = xs_udp_set_buffer_size,
1368         .print_addr             = xs_print_peer_address,
1369         .reserve_xprt           = xprt_reserve_xprt_cong,
1370         .release_xprt           = xprt_release_xprt_cong,
1371         .rpcbind                = rpc_getport,
1372         .set_port               = xs_set_port,
1373         .connect                = xs_connect,
1374         .buf_alloc              = rpc_malloc,
1375         .buf_free               = rpc_free,
1376         .send_request           = xs_udp_send_request,
1377         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1378         .timer                  = xs_udp_timer,
1379         .release_request        = xprt_release_rqst_cong,
1380         .close                  = xs_close,
1381         .destroy                = xs_destroy,
1382         .print_stats            = xs_udp_print_stats,
1383 };
1384
1385 static struct rpc_xprt_ops xs_tcp_ops = {
1386         .print_addr             = xs_print_peer_address,
1387         .reserve_xprt           = xprt_reserve_xprt,
1388         .release_xprt           = xs_tcp_release_xprt,
1389         .rpcbind                = rpc_getport,
1390         .set_port               = xs_set_port,
1391         .connect                = xs_connect,
1392         .buf_alloc              = rpc_malloc,
1393         .buf_free               = rpc_free,
1394         .send_request           = xs_tcp_send_request,
1395         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1396         .close                  = xs_close,
1397         .destroy                = xs_destroy,
1398         .print_stats            = xs_tcp_print_stats,
1399 };
1400
1401 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1402 {
1403         struct rpc_xprt *xprt;
1404         struct sock_xprt *new;
1405
1406         if (addrlen > sizeof(xprt->addr)) {
1407                 dprintk("RPC:      xs_setup_xprt: address too large\n");
1408                 return ERR_PTR(-EBADF);
1409         }
1410
1411         new = kzalloc(sizeof(*new), GFP_KERNEL);
1412         if (new == NULL) {
1413                 dprintk("RPC:      xs_setup_xprt: couldn't allocate rpc_xprt\n");
1414                 return ERR_PTR(-ENOMEM);
1415         }
1416         xprt = &new->xprt;
1417
1418         xprt->max_reqs = slot_table_size;
1419         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1420         if (xprt->slot == NULL) {
1421                 kfree(xprt);
1422                 dprintk("RPC:      xs_setup_xprt: couldn't allocate slot table\n");
1423                 return ERR_PTR(-ENOMEM);
1424         }
1425
1426         memcpy(&xprt->addr, addr, addrlen);
1427         xprt->addrlen = addrlen;
1428         new->port = xs_get_random_port();
1429
1430         return xprt;
1431 }
1432
1433 /**
1434  * xs_setup_udp - Set up transport to use a UDP socket
1435  * @addr: address of remote server
1436  * @addrlen: length of address in bytes
1437  * @to:   timeout parameters
1438  *
1439  */
1440 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1441 {
1442         struct rpc_xprt *xprt;
1443         struct sock_xprt *transport;
1444
1445         xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1446         if (IS_ERR(xprt))
1447                 return xprt;
1448         transport = container_of(xprt, struct sock_xprt, xprt);
1449
1450         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1451                 xprt_set_bound(xprt);
1452
1453         xprt->prot = IPPROTO_UDP;
1454         xprt->tsh_size = 0;
1455         /* XXX: header size can vary due to auth type, IPv6, etc. */
1456         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1457
1458         INIT_WORK(&transport->connect_worker, xs_udp_connect_worker, transport);
1459         xprt->bind_timeout = XS_BIND_TO;
1460         xprt->connect_timeout = XS_UDP_CONN_TO;
1461         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1462         xprt->idle_timeout = XS_IDLE_DISC_TO;
1463
1464         xprt->ops = &xs_udp_ops;
1465
1466         if (to)
1467                 xprt->timeout = *to;
1468         else
1469                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1470
1471         xs_format_peer_addresses(xprt);
1472         dprintk("RPC:      set up transport to address %s\n",
1473                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1474
1475         return xprt;
1476 }
1477
1478 /**
1479  * xs_setup_tcp - Set up transport to use a TCP socket
1480  * @addr: address of remote server
1481  * @addrlen: length of address in bytes
1482  * @to: timeout parameters
1483  *
1484  */
1485 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1486 {
1487         struct rpc_xprt *xprt;
1488         struct sock_xprt *transport;
1489
1490         xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1491         if (IS_ERR(xprt))
1492                 return xprt;
1493         transport = container_of(xprt, struct sock_xprt, xprt);
1494
1495         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1496                 xprt_set_bound(xprt);
1497
1498         xprt->prot = IPPROTO_TCP;
1499         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1500         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1501
1502         INIT_WORK(&transport->connect_worker, xs_tcp_connect_worker, transport);
1503         xprt->bind_timeout = XS_BIND_TO;
1504         xprt->connect_timeout = XS_TCP_CONN_TO;
1505         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1506         xprt->idle_timeout = XS_IDLE_DISC_TO;
1507
1508         xprt->ops = &xs_tcp_ops;
1509
1510         if (to)
1511                 xprt->timeout = *to;
1512         else
1513                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1514
1515         xs_format_peer_addresses(xprt);
1516         dprintk("RPC:      set up transport to address %s\n",
1517                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1518
1519         return xprt;
1520 }