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 /*
 151  * TCP receive state flags
 152  */
 153 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 154 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 155 #define TCP_RCV_COPY_XID        (1UL << 2)
 156 #define TCP_RCV_COPY_DATA       (1UL << 3)
 157
 158 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
 159 {
 160         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
 161         char *buf;
 162
 163         buf = kzalloc(20, GFP_KERNEL);
 164         if (buf) {
 165                 snprintf(buf, 20, "%u.%u.%u.%u",
 166                                 NIPQUAD(addr->sin_addr.s_addr));
 167         }
 168         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 169
 170         buf = kzalloc(8, GFP_KERNEL);
 171         if (buf) {
 172                 snprintf(buf, 8, "%u",
 173                                 ntohs(addr->sin_port));
 174         }
 175         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 176
 177         if (xprt->prot == IPPROTO_UDP)
 178                 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
 179         else
 180                 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
 181
 182         buf = kzalloc(48, GFP_KERNEL);
 183         if (buf) {
 184                 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
 185                         NIPQUAD(addr->sin_addr.s_addr),
 186                         ntohs(addr->sin_port),
 187                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 188         }
 189         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 190 }
 191
 192 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 193 {
 194         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
 195         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 196         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
 197 }
 198
 199 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 200
 201 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 202 {
 203         struct msghdr msg = {
 204                 .msg_name       = addr,
 205                 .msg_namelen    = addrlen,
 206                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 207         };
 208         struct kvec iov = {
 209                 .iov_base       = vec->iov_base + base,
 210                 .iov_len        = vec->iov_len - base,
 211         };
 212
 213         if (iov.iov_len != 0)
 214                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 215         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 216 }
 217
 218 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 219 {
 220         struct page **ppage;
 221         unsigned int remainder;
 222         int err, sent = 0;
 223
 224         remainder = xdr->page_len - base;
 225         base += xdr->page_base;
 226         ppage = xdr->pages + (base >> PAGE_SHIFT);
 227         base &= ~PAGE_MASK;
 228         for(;;) {
 229                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 230                 int flags = XS_SENDMSG_FLAGS;
 231
 232                 remainder -= len;
 233                 if (remainder != 0 || more)
 234                         flags |= MSG_MORE;
 235                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 236                 if (remainder == 0 || err != len)
 237                         break;
 238                 sent += err;
 239                 ppage++;
 240                 base = 0;
 241         }
 242         if (sent == 0)
 243                 return err;
 244         if (err > 0)
 245                 sent += err;
 246         return sent;
 247 }
 248
 249 /**
 250  * xs_sendpages - write pages directly to a socket
 251  * @sock: socket to send on
 252  * @addr: UDP only -- address of destination
 253  * @addrlen: UDP only -- length of destination address
 254  * @xdr: buffer containing this request
 255  * @base: starting position in the buffer
 256  *
 257  */
 258 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 259 {
 260         unsigned int remainder = xdr->len - base;
 261         int err, sent = 0;
 262
 263         if (unlikely(!sock))
 264                 return -ENOTCONN;
 265
 266         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 267         if (base != 0) {
 268                 addr = NULL;
 269                 addrlen = 0;
 270         }
 271
 272         if (base < xdr->head[0].iov_len || addr != NULL) {
 273                 unsigned int len = xdr->head[0].iov_len - base;
 274                 remainder -= len;
 275                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 276                 if (remainder == 0 || err != len)
 277                         goto out;
 278                 sent += err;
 279                 base = 0;
 280         } else
 281                 base -= xdr->head[0].iov_len;
 282
 283         if (base < xdr->page_len) {
 284                 unsigned int len = xdr->page_len - base;
 285                 remainder -= len;
 286                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 287                 if (remainder == 0 || err != len)
 288                         goto out;
 289                 sent += err;
 290                 base = 0;
 291         } else
 292                 base -= xdr->page_len;
 293
 294         if (base >= xdr->tail[0].iov_len)
 295                 return sent;
 296         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 297 out:
 298         if (sent == 0)
 299                 return err;
 300         if (err > 0)
 301                 sent += err;
 302         return sent;
 303 }
 304
 305 /**
 306  * xs_nospace - place task on wait queue if transmit was incomplete
 307  * @task: task to put to sleep
 308  *
 309  */
 310 static void xs_nospace(struct rpc_task *task)
 311 {
 312         struct rpc_rqst *req = task->tk_rqstp;
 313         struct rpc_xprt *xprt = req->rq_xprt;
 314         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 315
 316         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 317                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 318                         req->rq_slen);
 319
 320         if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 321                 /* Protect against races with write_space */
 322                 spin_lock_bh(&xprt->transport_lock);
 323
 324                 /* Don't race with disconnect */
 325                 if (!xprt_connected(xprt))
 326                         task->tk_status = -ENOTCONN;
 327                 else if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
 328                         xprt_wait_for_buffer_space(task);
 329
 330                 spin_unlock_bh(&xprt->transport_lock);
 331         } else
 332                 /* Keep holding the socket if it is blocked */
 333                 rpc_delay(task, HZ>>4);
 334 }
 335
 336 /**
 337  * xs_udp_send_request - write an RPC request to a UDP socket
 338  * @task: address of RPC task that manages the state of an RPC request
 339  *
 340  * Return values:
 341  *        0:    The request has been sent
 342  *   EAGAIN:    The socket was blocked, please call again later to
 343  *              complete the request
 344  * ENOTCONN:    Caller needs to invoke connect logic then call again
 345  *    other:    Some other error occured, the request was not sent
 346  */
 347 static int xs_udp_send_request(struct rpc_task *task)
 348 {
 349         struct rpc_rqst *req = task->tk_rqstp;
 350         struct rpc_xprt *xprt = req->rq_xprt;
 351         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 352         struct xdr_buf *xdr = &req->rq_snd_buf;
 353         int status;
 354
 355         xs_pktdump("packet data:",
 356                                 req->rq_svec->iov_base,
 357                                 req->rq_svec->iov_len);
 358
 359         req->rq_xtime = jiffies;
 360         status = xs_sendpages(transport->sock,
 361                               (struct sockaddr *) &xprt->addr,
 362                               xprt->addrlen, xdr,
 363                               req->rq_bytes_sent);
 364
 365         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 366                         xdr->len - req->rq_bytes_sent, status);
 367
 368         if (likely(status >= (int) req->rq_slen))
 369                 return 0;
 370
 371         /* Still some bytes left; set up for a retry later. */
 372         if (status > 0)
 373                 status = -EAGAIN;
 374
 375         switch (status) {
 376         case -ENETUNREACH:
 377         case -EPIPE:
 378         case -ECONNREFUSED:
 379                 /* When the server has died, an ICMP port unreachable message
 380                  * prompts ECONNREFUSED. */
 381                 break;
 382         case -EAGAIN:
 383                 xs_nospace(task);
 384                 break;
 385         default:
 386                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 387                         -status);
 388                 break;
 389         }
 390
 391         return status;
 392 }
 393
 394 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 395 {
 396         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 397         rpc_fraghdr *base = buf->head[0].iov_base;
 398         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 399 }
 400
 401 /**
 402  * xs_tcp_send_request - write an RPC request to a TCP socket
 403  * @task: address of RPC task that manages the state of an RPC request
 404  *
 405  * Return values:
 406  *        0:    The request has been sent
 407  *   EAGAIN:    The socket was blocked, please call again later to
 408  *              complete the request
 409  * ENOTCONN:    Caller needs to invoke connect logic then call again
 410  *    other:    Some other error occured, the request was not sent
 411  *
 412  * XXX: In the case of soft timeouts, should we eventually give up
 413  *      if sendmsg is not able to make progress?
 414  */
 415 static int xs_tcp_send_request(struct rpc_task *task)
 416 {
 417         struct rpc_rqst *req = task->tk_rqstp;
 418         struct rpc_xprt *xprt = req->rq_xprt;
 419         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 420         struct xdr_buf *xdr = &req->rq_snd_buf;
 421         int status, retry = 0;
 422
 423         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 424
 425         xs_pktdump("packet data:",
 426                                 req->rq_svec->iov_base,
 427                                 req->rq_svec->iov_len);
 428
 429         /* Continue transmitting the packet/record. We must be careful
 430          * to cope with writespace callbacks arriving _after_ we have
 431          * called sendmsg(). */
 432         while (1) {
 433                 req->rq_xtime = jiffies;
 434                 status = xs_sendpages(transport->sock,
 435                                         NULL, 0, xdr, req->rq_bytes_sent);
 436
 437                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 438                                 xdr->len - req->rq_bytes_sent, status);
 439
 440                 if (unlikely(status < 0))
 441                         break;
 442
 443                 /* If we've sent the entire packet, immediately
 444                  * reset the count of bytes sent. */
 445                 req->rq_bytes_sent += status;
 446                 task->tk_bytes_sent += status;
 447                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 448                         req->rq_bytes_sent = 0;
 449                         return 0;
 450                 }
 451
 452                 status = -EAGAIN;
 453                 if (retry++ > XS_SENDMSG_RETRY)
 454                         break;
 455         }
 456
 457         switch (status) {
 458         case -EAGAIN:
 459                 xs_nospace(task);
 460                 break;
 461         case -ECONNREFUSED:
 462         case -ECONNRESET:
 463         case -ENOTCONN:
 464         case -EPIPE:
 465                 status = -ENOTCONN;
 466                 break;
 467         default:
 468                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 469                         -status);
 470                 xprt_disconnect(xprt);
 471                 break;
 472         }
 473
 474         return status;
 475 }
 476
 477 /**
 478  * xs_tcp_release_xprt - clean up after a tcp transmission
 479  * @xprt: transport
 480  * @task: rpc task
 481  *
 482  * This cleans up if an error causes us to abort the transmission of a request.
 483  * In this case, the socket may need to be reset in order to avoid confusing
 484  * the server.
 485  */
 486 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 487 {
 488         struct rpc_rqst *req;
 489
 490         if (task != xprt->snd_task)
 491                 return;
 492         if (task == NULL)
 493                 goto out_release;
 494         req = task->tk_rqstp;
 495         if (req->rq_bytes_sent == 0)
 496                 goto out_release;
 497         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 498                 goto out_release;
 499         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 500 out_release:
 501         xprt_release_xprt(xprt, task);
 502 }
 503
 504 /**
 505  * xs_close - close a socket
 506  * @xprt: transport
 507  *
 508  * This is used when all requests are complete; ie, no DRC state remains
 509  * on the server we want to save.
 510  */
 511 static void xs_close(struct rpc_xprt *xprt)
 512 {
 513         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 514         struct socket *sock = transport->sock;
 515         struct sock *sk = transport->inet;
 516
 517         if (!sk)
 518                 goto clear_close_wait;
 519
 520         dprintk("RPC:      xs_close xprt %p\n", xprt);
 521
 522         write_lock_bh(&sk->sk_callback_lock);
 523         transport->inet = NULL;
 524         transport->sock = NULL;
 525
 526         sk->sk_user_data = NULL;
 527         sk->sk_data_ready = xprt->old_data_ready;
 528         sk->sk_state_change = xprt->old_state_change;
 529         sk->sk_write_space = xprt->old_write_space;
 530         write_unlock_bh(&sk->sk_callback_lock);
 531
 532         sk->sk_no_check = 0;
 533
 534         sock_release(sock);
 535 clear_close_wait:
 536         smp_mb__before_clear_bit();
 537         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 538         smp_mb__after_clear_bit();
 539 }
 540
 541 /**
 542  * xs_destroy - prepare to shutdown a transport
 543  * @xprt: doomed transport
 544  *
 545  */
 546 static void xs_destroy(struct rpc_xprt *xprt)
 547 {
 548         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 549
 550         cancel_delayed_work(&xprt->connect_worker);
 551         flush_scheduled_work();
 552
 553         xprt_disconnect(xprt);
 554         xs_close(xprt);
 555         xs_free_peer_addresses(xprt);
 556         kfree(xprt->slot);
 557         kfree(xprt);
 558 }
 559
 560 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 561 {
 562         return (struct rpc_xprt *) sk->sk_user_data;
 563 }
 564
 565 /**
 566  * xs_udp_data_ready - "data ready" callback for UDP sockets
 567  * @sk: socket with data to read
 568  * @len: how much data to read
 569  *
 570  */
 571 static void xs_udp_data_ready(struct sock *sk, int len)
 572 {
 573         struct rpc_task *task;
 574         struct rpc_xprt *xprt;
 575         struct rpc_rqst *rovr;
 576         struct sk_buff *skb;
 577         int err, repsize, copied;
 578         u32 _xid;
 579         __be32 *xp;
 580
 581         read_lock(&sk->sk_callback_lock);
 582         dprintk("RPC:      xs_udp_data_ready...\n");
 583         if (!(xprt = xprt_from_sock(sk)))
 584                 goto out;
 585
 586         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 587                 goto out;
 588
 589         if (xprt->shutdown)
 590                 goto dropit;
 591
 592         repsize = skb->len - sizeof(struct udphdr);
 593         if (repsize < 4) {
 594                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 595                 goto dropit;
 596         }
 597
 598         /* Copy the XID from the skb... */
 599         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 600                                 sizeof(_xid), &_xid);
 601         if (xp == NULL)
 602                 goto dropit;
 603
 604         /* Look up and lock the request corresponding to the given XID */
 605         spin_lock(&xprt->transport_lock);
 606         rovr = xprt_lookup_rqst(xprt, *xp);
 607         if (!rovr)
 608                 goto out_unlock;
 609         task = rovr->rq_task;
 610
 611         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 612                 copied = repsize;
 613
 614         /* Suck it into the iovec, verify checksum if not done by hw. */
 615         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 616                 goto out_unlock;
 617
 618         /* Something worked... */
 619         dst_confirm(skb->dst);
 620
 621         xprt_adjust_cwnd(task, copied);
 622         xprt_update_rtt(task);
 623         xprt_complete_rqst(task, copied);
 624
 625  out_unlock:
 626         spin_unlock(&xprt->transport_lock);
 627  dropit:
 628         skb_free_datagram(sk, skb);
 629  out:
 630         read_unlock(&sk->sk_callback_lock);
 631 }
 632
 633 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 634 {
 635         if (len > desc->count)
 636                 len = desc->count;
 637         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 638                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 639                                 len, desc->count);
 640                 return 0;
 641         }
 642         desc->offset += len;
 643         desc->count -= len;
 644         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 645                         len, desc->count);
 646         return len;
 647 }
 648
 649 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 650 {
 651         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 652         size_t len, used;
 653         char *p;
 654
 655         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 656         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 657         used = xs_tcp_copy_data(desc, p, len);
 658         transport->tcp_offset += used;
 659         if (used != len)
 660                 return;
 661
 662         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 663         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 664                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 665         else
 666                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 667         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 668
 669         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 670         transport->tcp_offset = 0;
 671
 672         /* Sanity check of the record length */
 673         if (unlikely(transport->tcp_reclen < 4)) {
 674                 dprintk("RPC:      invalid TCP record fragment length\n");
 675                 xprt_disconnect(xprt);
 676                 return;
 677         }
 678         dprintk("RPC:      reading TCP record fragment of length %d\n",
 679                         transport->tcp_reclen);
 680 }
 681
 682 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 683 {
 684         if (transport->tcp_offset == transport->tcp_reclen) {
 685                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 686                 transport->tcp_offset = 0;
 687                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 688                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 689                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 690                         transport->tcp_copied = 0;
 691                 }
 692         }
 693 }
 694
 695 static inline void xs_tcp_read_xid(struct sock_xprt *transport, skb_reader_t *desc)
 696 {
 697         size_t len, used;
 698         char *p;
 699
 700         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 701         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 702         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 703         used = xs_tcp_copy_data(desc, p, len);
 704         transport->tcp_offset += used;
 705         if (used != len)
 706                 return;
 707         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 708         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 709         transport->tcp_copied = 4;
 710         dprintk("RPC:      reading reply for XID %08x\n",
 711                         ntohl(transport->tcp_xid));
 712         xs_tcp_check_fraghdr(transport);
 713 }
 714
 715 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 716 {
 717         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 718         struct rpc_rqst *req;
 719         struct xdr_buf *rcvbuf;
 720         size_t len;
 721         ssize_t r;
 722
 723         /* Find and lock the request corresponding to this xid */
 724         spin_lock(&xprt->transport_lock);
 725         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
 726         if (!req) {
 727                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 728                 dprintk("RPC:      XID %08x request not found!\n",
 729                                 ntohl(transport->tcp_xid));
 730                 spin_unlock(&xprt->transport_lock);
 731                 return;
 732         }
 733
 734         rcvbuf = &req->rq_private_buf;
 735         len = desc->count;
 736         if (len > transport->tcp_reclen - transport->tcp_offset) {
 737                 skb_reader_t my_desc;
 738
 739                 len = transport->tcp_reclen - transport->tcp_offset;
 740                 memcpy(&my_desc, desc, sizeof(my_desc));
 741                 my_desc.count = len;
 742                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 743                                           &my_desc, xs_tcp_copy_data);
 744                 desc->count -= r;
 745                 desc->offset += r;
 746         } else
 747                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
 748                                           desc, xs_tcp_copy_data);
 749
 750         if (r > 0) {
 751                 transport->tcp_copied += r;
 752                 transport->tcp_offset += r;
 753         }
 754         if (r != len) {
 755                 /* Error when copying to the receive buffer,
 756                  * usually because we weren't able to allocate
 757                  * additional buffer pages. All we can do now
 758                  * is turn off TCP_RCV_COPY_DATA, so the request
 759                  * will not receive any additional updates,
 760                  * and time out.
 761                  * Any remaining data from this record will
 762                  * be discarded.
 763                  */
 764                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 765                 dprintk("RPC:      XID %08x truncated request\n",
 766                                 ntohl(transport->tcp_xid));
 767                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 768                                 xprt, transport->tcp_copied, transport->tcp_offset,
 769                                         transport->tcp_reclen);
 770                 goto out;
 771         }
 772
 773         dprintk("RPC:      XID %08x read %Zd bytes\n",
 774                         ntohl(transport->tcp_xid), r);
 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
 779         if (transport->tcp_copied == req->rq_private_buf.buflen)
 780                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 781         else if (transport->tcp_offset == transport->tcp_reclen) {
 782                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
 783                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 784         }
 785
 786 out:
 787         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
 788                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
 789         spin_unlock(&xprt->transport_lock);
 790         xs_tcp_check_fraghdr(transport);
 791 }
 792
 793 static inline void xs_tcp_read_discard(struct sock_xprt *transport, skb_reader_t *desc)
 794 {
 795         size_t len;
 796
 797         len = transport->tcp_reclen - transport->tcp_offset;
 798         if (len > desc->count)
 799                 len = desc->count;
 800         desc->count -= len;
 801         desc->offset += len;
 802         transport->tcp_offset += len;
 803         dprintk("RPC:      discarded %Zu bytes\n", len);
 804         xs_tcp_check_fraghdr(transport);
 805 }
 806
 807 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 808 {
 809         struct rpc_xprt *xprt = rd_desc->arg.data;
 810         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 811         skb_reader_t desc = {
 812                 .skb    = skb,
 813                 .offset = offset,
 814                 .count  = len,
 815         };
 816
 817         dprintk("RPC:      xs_tcp_data_recv started\n");
 818         do {
 819                 /* Read in a new fragment marker if necessary */
 820                 /* Can we ever really expect to get completely empty fragments? */
 821                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
 822                         xs_tcp_read_fraghdr(xprt, &desc);
 823                         continue;
 824                 }
 825                 /* Read in the xid if necessary */
 826                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
 827                         xs_tcp_read_xid(transport, &desc);
 828                         continue;
 829                 }
 830                 /* Read in the request data */
 831                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
 832                         xs_tcp_read_request(xprt, &desc);
 833                         continue;
 834                 }
 835                 /* Skip over any trailing bytes on short reads */
 836                 xs_tcp_read_discard(transport, &desc);
 837         } while (desc.count);
 838         dprintk("RPC:      xs_tcp_data_recv done\n");
 839         return len - desc.count;
 840 }
 841
 842 /**
 843  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 844  * @sk: socket with data to read
 845  * @bytes: how much data to read
 846  *
 847  */
 848 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 849 {
 850         struct rpc_xprt *xprt;
 851         read_descriptor_t rd_desc;
 852
 853         read_lock(&sk->sk_callback_lock);
 854         dprintk("RPC:      xs_tcp_data_ready...\n");
 855         if (!(xprt = xprt_from_sock(sk)))
 856                 goto out;
 857         if (xprt->shutdown)
 858                 goto out;
 859
 860         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 861         rd_desc.arg.data = xprt;
 862         rd_desc.count = 65536;
 863         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 864 out:
 865         read_unlock(&sk->sk_callback_lock);
 866 }
 867
 868 /**
 869  * xs_tcp_state_change - callback to handle TCP socket state changes
 870  * @sk: socket whose state has changed
 871  *
 872  */
 873 static void xs_tcp_state_change(struct sock *sk)
 874 {
 875         struct rpc_xprt *xprt;
 876
 877         read_lock(&sk->sk_callback_lock);
 878         if (!(xprt = xprt_from_sock(sk)))
 879                 goto out;
 880         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 881         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 882                                 sk->sk_state, xprt_connected(xprt),
 883                                 sock_flag(sk, SOCK_DEAD),
 884                                 sock_flag(sk, SOCK_ZAPPED));
 885
 886         switch (sk->sk_state) {
 887         case TCP_ESTABLISHED:
 888                 spin_lock_bh(&xprt->transport_lock);
 889                 if (!xprt_test_and_set_connected(xprt)) {
 890                         struct sock_xprt *transport = container_of(xprt,
 891                                         struct sock_xprt, xprt);
 892
 893                         /* Reset TCP record info */
 894                         transport->tcp_offset = 0;
 895                         transport->tcp_reclen = 0;
 896                         transport->tcp_copied = 0;
 897                         transport->tcp_flags =
 898                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
 899
 900                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 901                         xprt_wake_pending_tasks(xprt, 0);
 902                 }
 903                 spin_unlock_bh(&xprt->transport_lock);
 904                 break;
 905         case TCP_SYN_SENT:
 906         case TCP_SYN_RECV:
 907                 break;
 908         case TCP_CLOSE_WAIT:
 909                 /* Try to schedule an autoclose RPC calls */
 910                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 911                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 912                         schedule_work(&xprt->task_cleanup);
 913         default:
 914                 xprt_disconnect(xprt);
 915         }
 916  out:
 917         read_unlock(&sk->sk_callback_lock);
 918 }
 919
 920 /**
 921  * xs_udp_write_space - callback invoked when socket buffer space
 922  *                             becomes available
 923  * @sk: socket whose state has changed
 924  *
 925  * Called when more output buffer space is available for this socket.
 926  * We try not to wake our writers until they can make "significant"
 927  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 928  * with a bunch of small requests.
 929  */
 930 static void xs_udp_write_space(struct sock *sk)
 931 {
 932         read_lock(&sk->sk_callback_lock);
 933
 934         /* from net/core/sock.c:sock_def_write_space */
 935         if (sock_writeable(sk)) {
 936                 struct socket *sock;
 937                 struct rpc_xprt *xprt;
 938
 939                 if (unlikely(!(sock = sk->sk_socket)))
 940                         goto out;
 941                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 942                         goto out;
 943                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 944                         goto out;
 945
 946                 xprt_write_space(xprt);
 947         }
 948
 949  out:
 950         read_unlock(&sk->sk_callback_lock);
 951 }
 952
 953 /**
 954  * xs_tcp_write_space - callback invoked when socket buffer space
 955  *                             becomes available
 956  * @sk: socket whose state has changed
 957  *
 958  * Called when more output buffer space is available for this socket.
 959  * We try not to wake our writers until they can make "significant"
 960  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 961  * with a bunch of small requests.
 962  */
 963 static void xs_tcp_write_space(struct sock *sk)
 964 {
 965         read_lock(&sk->sk_callback_lock);
 966
 967         /* from net/core/stream.c:sk_stream_write_space */
 968         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 969                 struct socket *sock;
 970                 struct rpc_xprt *xprt;
 971
 972                 if (unlikely(!(sock = sk->sk_socket)))
 973                         goto out;
 974                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 975                         goto out;
 976                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 977                         goto out;
 978
 979                 xprt_write_space(xprt);
 980         }
 981
 982  out:
 983         read_unlock(&sk->sk_callback_lock);
 984 }
 985
 986 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 987 {
 988         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 989         struct sock *sk = transport->inet;
 990
 991         if (xprt->rcvsize) {
 992                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 993                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 994         }
 995         if (xprt->sndsize) {
 996                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 997                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 998                 sk->sk_write_space(sk);
 999         }
1000 }
1001
1002 /**
1003  * xs_udp_set_buffer_size - set send and receive limits
1004  * @xprt: generic transport
1005  * @sndsize: requested size of send buffer, in bytes
1006  * @rcvsize: requested size of receive buffer, in bytes
1007  *
1008  * Set socket send and receive buffer size limits.
1009  */
1010 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1011 {
1012         xprt->sndsize = 0;
1013         if (sndsize)
1014                 xprt->sndsize = sndsize + 1024;
1015         xprt->rcvsize = 0;
1016         if (rcvsize)
1017                 xprt->rcvsize = rcvsize + 1024;
1018
1019         xs_udp_do_set_buffer_size(xprt);
1020 }
1021
1022 /**
1023  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1024  * @task: task that timed out
1025  *
1026  * Adjust the congestion window after a retransmit timeout has occurred.
1027  */
1028 static void xs_udp_timer(struct rpc_task *task)
1029 {
1030         xprt_adjust_cwnd(task, -ETIMEDOUT);
1031 }
1032
1033 static unsigned short xs_get_random_port(void)
1034 {
1035         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1036         unsigned short rand = (unsigned short) net_random() % range;
1037         return rand + xprt_min_resvport;
1038 }
1039
1040 /**
1041  * xs_print_peer_address - format an IPv4 address for printing
1042  * @xprt: generic transport
1043  * @format: flags field indicating which parts of the address to render
1044  */
1045 static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format)
1046 {
1047         if (xprt->address_strings[format] != NULL)
1048                 return xprt->address_strings[format];
1049         else
1050                 return "unprintable";
1051 }
1052
1053 /**
1054  * xs_set_port - reset the port number in the remote endpoint address
1055  * @xprt: generic transport
1056  * @port: new port number
1057  *
1058  */
1059 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1060 {
1061         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1062
1063         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
1064
1065         sap->sin_port = htons(port);
1066 }
1067
1068 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
1069 {
1070         struct sockaddr_in myaddr = {
1071                 .sin_family = AF_INET,
1072         };
1073         int err;
1074         unsigned short port = xprt->port;
1075
1076         do {
1077                 myaddr.sin_port = htons(port);
1078                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1079                                                 sizeof(myaddr));
1080                 if (err == 0) {
1081                         xprt->port = port;
1082                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
1083                                         port);
1084                         return 0;
1085                 }
1086                 if (port <= xprt_min_resvport)
1087                         port = xprt_max_resvport;
1088                 else
1089                         port--;
1090         } while (err == -EADDRINUSE && port != xprt->port);
1091
1092         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
1093         return err;
1094 }
1095
1096 /**
1097  * xs_udp_connect_worker - set up a UDP socket
1098  * @args: RPC transport to connect
1099  *
1100  * Invoked by a work queue tasklet.
1101  */
1102 static void xs_udp_connect_worker(void *args)
1103 {
1104         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
1105         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1106         struct socket *sock = transport->sock;
1107         int err, status = -EIO;
1108
1109         if (xprt->shutdown || !xprt_bound(xprt))
1110                 goto out;
1111
1112         /* Start by resetting any existing state */
1113         xs_close(xprt);
1114
1115         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1116                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1117                 goto out;
1118         }
1119
1120         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1121                 sock_release(sock);
1122                 goto out;
1123         }
1124
1125         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1126                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1127
1128         if (!transport->inet) {
1129                 struct sock *sk = sock->sk;
1130
1131                 write_lock_bh(&sk->sk_callback_lock);
1132
1133                 sk->sk_user_data = xprt;
1134                 xprt->old_data_ready = sk->sk_data_ready;
1135                 xprt->old_state_change = sk->sk_state_change;
1136                 xprt->old_write_space = sk->sk_write_space;
1137                 sk->sk_data_ready = xs_udp_data_ready;
1138                 sk->sk_write_space = xs_udp_write_space;
1139                 sk->sk_no_check = UDP_CSUM_NORCV;
1140                 sk->sk_allocation = GFP_ATOMIC;
1141
1142                 xprt_set_connected(xprt);
1143
1144                 /* Reset to new socket */
1145                 transport->sock = sock;
1146                 transport->inet = sk;
1147
1148                 write_unlock_bh(&sk->sk_callback_lock);
1149         }
1150         xs_udp_do_set_buffer_size(xprt);
1151         status = 0;
1152 out:
1153         xprt_wake_pending_tasks(xprt, status);
1154         xprt_clear_connecting(xprt);
1155 }
1156
1157 /*
1158  * We need to preserve the port number so the reply cache on the server can
1159  * find our cached RPC replies when we get around to reconnecting.
1160  */
1161 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1162 {
1163         int result;
1164         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1165         struct sockaddr any;
1166
1167         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1168
1169         /*
1170          * Disconnect the transport socket by doing a connect operation
1171          * with AF_UNSPEC.  This should return immediately...
1172          */
1173         memset(&any, 0, sizeof(any));
1174         any.sa_family = AF_UNSPEC;
1175         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1176         if (result)
1177                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1178                                 result);
1179 }
1180
1181 /**
1182  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1183  * @args: RPC transport to connect
1184  *
1185  * Invoked by a work queue tasklet.
1186  */
1187 static void xs_tcp_connect_worker(void *args)
1188 {
1189         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1190         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1191         struct socket *sock = transport->sock;
1192         int err, status = -EIO;
1193
1194         if (xprt->shutdown || !xprt_bound(xprt))
1195                 goto out;
1196
1197         if (!sock) {
1198                 /* start from scratch */
1199                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1200                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1201                         goto out;
1202                 }
1203
1204                 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1205                         sock_release(sock);
1206                         goto out;
1207                 }
1208         } else
1209                 /* "close" the socket, preserving the local port */
1210                 xs_tcp_reuse_connection(xprt);
1211
1212         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1213                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1214
1215         if (!transport->inet) {
1216                 struct sock *sk = sock->sk;
1217
1218                 write_lock_bh(&sk->sk_callback_lock);
1219
1220                 sk->sk_user_data = xprt;
1221                 xprt->old_data_ready = sk->sk_data_ready;
1222                 xprt->old_state_change = sk->sk_state_change;
1223                 xprt->old_write_space = sk->sk_write_space;
1224                 sk->sk_data_ready = xs_tcp_data_ready;
1225                 sk->sk_state_change = xs_tcp_state_change;
1226                 sk->sk_write_space = xs_tcp_write_space;
1227                 sk->sk_allocation = GFP_ATOMIC;
1228
1229                 /* socket options */
1230                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1231                 sock_reset_flag(sk, SOCK_LINGER);
1232                 tcp_sk(sk)->linger2 = 0;
1233                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1234
1235                 xprt_clear_connected(xprt);
1236
1237                 /* Reset to new socket */
1238                 transport->sock = sock;
1239                 transport->inet = sk;
1240
1241                 write_unlock_bh(&sk->sk_callback_lock);
1242         }
1243
1244         /* Tell the socket layer to start connecting... */
1245         xprt->stat.connect_count++;
1246         xprt->stat.connect_start = jiffies;
1247         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1248                         xprt->addrlen, O_NONBLOCK);
1249         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1250                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1251         if (status < 0) {
1252                 switch (status) {
1253                         case -EINPROGRESS:
1254                         case -EALREADY:
1255                                 goto out_clear;
1256                         case -ECONNREFUSED:
1257                         case -ECONNRESET:
1258                                 /* retry with existing socket, after a delay */
1259                                 break;
1260                         default:
1261                                 /* get rid of existing socket, and retry */
1262                                 xs_close(xprt);
1263                                 break;
1264                 }
1265         }
1266 out:
1267         xprt_wake_pending_tasks(xprt, status);
1268 out_clear:
1269         xprt_clear_connecting(xprt);
1270 }
1271
1272 /**
1273  * xs_connect - connect a socket to a remote endpoint
1274  * @task: address of RPC task that manages state of connect request
1275  *
1276  * TCP: If the remote end dropped the connection, delay reconnecting.
1277  *
1278  * UDP socket connects are synchronous, but we use a work queue anyway
1279  * to guarantee that even unprivileged user processes can set up a
1280  * socket on a privileged port.
1281  *
1282  * If a UDP socket connect fails, the delay behavior here prevents
1283  * retry floods (hard mounts).
1284  */
1285 static void xs_connect(struct rpc_task *task)
1286 {
1287         struct rpc_xprt *xprt = task->tk_xprt;
1288         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1289
1290         if (xprt_test_and_set_connecting(xprt))
1291                 return;
1292
1293         if (transport->sock != NULL) {
1294                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1295                                 xprt, xprt->reestablish_timeout / HZ);
1296                 schedule_delayed_work(&xprt->connect_worker,
1297                                         xprt->reestablish_timeout);
1298                 xprt->reestablish_timeout <<= 1;
1299                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1300                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1301         } else {
1302                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1303                 schedule_work(&xprt->connect_worker);
1304
1305                 /* flush_scheduled_work can sleep... */
1306                 if (!RPC_IS_ASYNC(task))
1307                         flush_scheduled_work();
1308         }
1309 }
1310
1311 /**
1312  * xs_udp_print_stats - display UDP socket-specifc stats
1313  * @xprt: rpc_xprt struct containing statistics
1314  * @seq: output file
1315  *
1316  */
1317 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1318 {
1319         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1320                         xprt->port,
1321                         xprt->stat.bind_count,
1322                         xprt->stat.sends,
1323                         xprt->stat.recvs,
1324                         xprt->stat.bad_xids,
1325                         xprt->stat.req_u,
1326                         xprt->stat.bklog_u);
1327 }
1328
1329 /**
1330  * xs_tcp_print_stats - display TCP socket-specifc stats
1331  * @xprt: rpc_xprt struct containing statistics
1332  * @seq: output file
1333  *
1334  */
1335 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1336 {
1337         long idle_time = 0;
1338
1339         if (xprt_connected(xprt))
1340                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1341
1342         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1343                         xprt->port,
1344                         xprt->stat.bind_count,
1345                         xprt->stat.connect_count,
1346                         xprt->stat.connect_time,
1347                         idle_time,
1348                         xprt->stat.sends,
1349                         xprt->stat.recvs,
1350                         xprt->stat.bad_xids,
1351                         xprt->stat.req_u,
1352                         xprt->stat.bklog_u);
1353 }
1354
1355 static struct rpc_xprt_ops xs_udp_ops = {
1356         .set_buffer_size        = xs_udp_set_buffer_size,
1357         .print_addr             = xs_print_peer_address,
1358         .reserve_xprt           = xprt_reserve_xprt_cong,
1359         .release_xprt           = xprt_release_xprt_cong,
1360         .rpcbind                = rpc_getport,
1361         .set_port               = xs_set_port,
1362         .connect                = xs_connect,
1363         .buf_alloc              = rpc_malloc,
1364         .buf_free               = rpc_free,
1365         .send_request           = xs_udp_send_request,
1366         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1367         .timer                  = xs_udp_timer,
1368         .release_request        = xprt_release_rqst_cong,
1369         .close                  = xs_close,
1370         .destroy                = xs_destroy,
1371         .print_stats            = xs_udp_print_stats,
1372 };
1373
1374 static struct rpc_xprt_ops xs_tcp_ops = {
1375         .print_addr             = xs_print_peer_address,
1376         .reserve_xprt           = xprt_reserve_xprt,
1377         .release_xprt           = xs_tcp_release_xprt,
1378         .rpcbind                = rpc_getport,
1379         .set_port               = xs_set_port,
1380         .connect                = xs_connect,
1381         .buf_alloc              = rpc_malloc,
1382         .buf_free               = rpc_free,
1383         .send_request           = xs_tcp_send_request,
1384         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1385         .close                  = xs_close,
1386         .destroy                = xs_destroy,
1387         .print_stats            = xs_tcp_print_stats,
1388 };
1389
1390 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1391 {
1392         struct rpc_xprt *xprt;
1393         struct sock_xprt *new;
1394
1395         if (addrlen > sizeof(xprt->addr)) {
1396                 dprintk("RPC:      xs_setup_xprt: address too large\n");
1397                 return ERR_PTR(-EBADF);
1398         }
1399
1400         new = kzalloc(sizeof(*new), GFP_KERNEL);
1401         if (new == NULL) {
1402                 dprintk("RPC:      xs_setup_xprt: couldn't allocate rpc_xprt\n");
1403                 return ERR_PTR(-ENOMEM);
1404         }
1405         xprt = &new->xprt;
1406
1407         xprt->max_reqs = slot_table_size;
1408         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1409         if (xprt->slot == NULL) {
1410                 kfree(xprt);
1411                 dprintk("RPC:      xs_setup_xprt: couldn't allocate slot table\n");
1412                 return ERR_PTR(-ENOMEM);
1413         }
1414
1415         memcpy(&xprt->addr, addr, addrlen);
1416         xprt->addrlen = addrlen;
1417         xprt->port = xs_get_random_port();
1418
1419         return xprt;
1420 }
1421
1422 /**
1423  * xs_setup_udp - Set up transport to use a UDP socket
1424  * @addr: address of remote server
1425  * @addrlen: length of address in bytes
1426  * @to:   timeout parameters
1427  *
1428  */
1429 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1430 {
1431         struct rpc_xprt *xprt;
1432
1433         xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1434         if (IS_ERR(xprt))
1435                 return xprt;
1436
1437         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1438                 xprt_set_bound(xprt);
1439
1440         xprt->prot = IPPROTO_UDP;
1441         xprt->tsh_size = 0;
1442         /* XXX: header size can vary due to auth type, IPv6, etc. */
1443         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1444
1445         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1446         xprt->bind_timeout = XS_BIND_TO;
1447         xprt->connect_timeout = XS_UDP_CONN_TO;
1448         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1449         xprt->idle_timeout = XS_IDLE_DISC_TO;
1450
1451         xprt->ops = &xs_udp_ops;
1452
1453         if (to)
1454                 xprt->timeout = *to;
1455         else
1456                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1457
1458         xs_format_peer_addresses(xprt);
1459         dprintk("RPC:      set up transport to address %s\n",
1460                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1461
1462         return xprt;
1463 }
1464
1465 /**
1466  * xs_setup_tcp - Set up transport to use a TCP socket
1467  * @addr: address of remote server
1468  * @addrlen: length of address in bytes
1469  * @to: timeout parameters
1470  *
1471  */
1472 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1473 {
1474         struct rpc_xprt *xprt;
1475
1476         xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1477         if (IS_ERR(xprt))
1478                 return xprt;
1479
1480         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1481                 xprt_set_bound(xprt);
1482
1483         xprt->prot = IPPROTO_TCP;
1484         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1485         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1486
1487         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1488         xprt->bind_timeout = XS_BIND_TO;
1489         xprt->connect_timeout = XS_TCP_CONN_TO;
1490         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1491         xprt->idle_timeout = XS_IDLE_DISC_TO;
1492
1493         xprt->ops = &xs_tcp_ops;
1494
1495         if (to)
1496                 xprt->timeout = *to;
1497         else
1498                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1499
1500         xs_format_peer_addresses(xprt);
1501         dprintk("RPC:      set up transport to address %s\n",
1502                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1503
1504         return xprt;
1505 }