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 static void xs_format_peer_addresses(struct rpc_xprt *xprt)
 129 {
 130         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
 131         char *buf;
 132
 133         buf = kzalloc(20, GFP_KERNEL);
 134         if (buf) {
 135                 snprintf(buf, 20, "%u.%u.%u.%u",
 136                                 NIPQUAD(addr->sin_addr.s_addr));
 137         }
 138         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 139
 140         buf = kzalloc(8, GFP_KERNEL);
 141         if (buf) {
 142                 snprintf(buf, 8, "%u",
 143                                 ntohs(addr->sin_port));
 144         }
 145         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 146
 147         if (xprt->prot == IPPROTO_UDP)
 148                 xprt->address_strings[RPC_DISPLAY_PROTO] = "udp";
 149         else
 150                 xprt->address_strings[RPC_DISPLAY_PROTO] = "tcp";
 151
 152         buf = kzalloc(48, GFP_KERNEL);
 153         if (buf) {
 154                 snprintf(buf, 48, "addr=%u.%u.%u.%u port=%u proto=%s",
 155                         NIPQUAD(addr->sin_addr.s_addr),
 156                         ntohs(addr->sin_port),
 157                         xprt->prot == IPPROTO_UDP ? "udp" : "tcp");
 158         }
 159         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 160 }
 161
 162 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 163 {
 164         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
 165         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 166         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
 167 }
 168
 169 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 170
 171 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 172 {
 173         struct msghdr msg = {
 174                 .msg_name       = addr,
 175                 .msg_namelen    = addrlen,
 176                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 177         };
 178         struct kvec iov = {
 179                 .iov_base       = vec->iov_base + base,
 180                 .iov_len        = vec->iov_len - base,
 181         };
 182
 183         if (iov.iov_len != 0)
 184                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 185         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 186 }
 187
 188 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 189 {
 190         struct page **ppage;
 191         unsigned int remainder;
 192         int err, sent = 0;
 193
 194         remainder = xdr->page_len - base;
 195         base += xdr->page_base;
 196         ppage = xdr->pages + (base >> PAGE_SHIFT);
 197         base &= ~PAGE_MASK;
 198         for(;;) {
 199                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 200                 int flags = XS_SENDMSG_FLAGS;
 201
 202                 remainder -= len;
 203                 if (remainder != 0 || more)
 204                         flags |= MSG_MORE;
 205                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 206                 if (remainder == 0 || err != len)
 207                         break;
 208                 sent += err;
 209                 ppage++;
 210                 base = 0;
 211         }
 212         if (sent == 0)
 213                 return err;
 214         if (err > 0)
 215                 sent += err;
 216         return sent;
 217 }
 218
 219 /**
 220  * xs_sendpages - write pages directly to a socket
 221  * @sock: socket to send on
 222  * @addr: UDP only -- address of destination
 223  * @addrlen: UDP only -- length of destination address
 224  * @xdr: buffer containing this request
 225  * @base: starting position in the buffer
 226  *
 227  */
 228 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 229 {
 230         unsigned int remainder = xdr->len - base;
 231         int err, sent = 0;
 232
 233         if (unlikely(!sock))
 234                 return -ENOTCONN;
 235
 236         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 237         if (base != 0) {
 238                 addr = NULL;
 239                 addrlen = 0;
 240         }
 241
 242         if (base < xdr->head[0].iov_len || addr != NULL) {
 243                 unsigned int len = xdr->head[0].iov_len - base;
 244                 remainder -= len;
 245                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 246                 if (remainder == 0 || err != len)
 247                         goto out;
 248                 sent += err;
 249                 base = 0;
 250         } else
 251                 base -= xdr->head[0].iov_len;
 252
 253         if (base < xdr->page_len) {
 254                 unsigned int len = xdr->page_len - base;
 255                 remainder -= len;
 256                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 257                 if (remainder == 0 || err != len)
 258                         goto out;
 259                 sent += err;
 260                 base = 0;
 261         } else
 262                 base -= xdr->page_len;
 263
 264         if (base >= xdr->tail[0].iov_len)
 265                 return sent;
 266         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 267 out:
 268         if (sent == 0)
 269                 return err;
 270         if (err > 0)
 271                 sent += err;
 272         return sent;
 273 }
 274
 275 /**
 276  * xs_nospace - place task on wait queue if transmit was incomplete
 277  * @task: task to put to sleep
 278  *
 279  */
 280 static void xs_nospace(struct rpc_task *task)
 281 {
 282         struct rpc_rqst *req = task->tk_rqstp;
 283         struct rpc_xprt *xprt = req->rq_xprt;
 284
 285         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 286                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 287                         req->rq_slen);
 288
 289         if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
 290                 /* Protect against races with write_space */
 291                 spin_lock_bh(&xprt->transport_lock);
 292
 293                 /* Don't race with disconnect */
 294                 if (!xprt_connected(xprt))
 295                         task->tk_status = -ENOTCONN;
 296                 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
 297                         xprt_wait_for_buffer_space(task);
 298
 299                 spin_unlock_bh(&xprt->transport_lock);
 300         } else
 301                 /* Keep holding the socket if it is blocked */
 302                 rpc_delay(task, HZ>>4);
 303 }
 304
 305 /**
 306  * xs_udp_send_request - write an RPC request to a UDP socket
 307  * @task: address of RPC task that manages the state of an RPC request
 308  *
 309  * Return values:
 310  *        0:    The request has been sent
 311  *   EAGAIN:    The socket was blocked, please call again later to
 312  *              complete the request
 313  * ENOTCONN:    Caller needs to invoke connect logic then call again
 314  *    other:    Some other error occured, the request was not sent
 315  */
 316 static int xs_udp_send_request(struct rpc_task *task)
 317 {
 318         struct rpc_rqst *req = task->tk_rqstp;
 319         struct rpc_xprt *xprt = req->rq_xprt;
 320         struct xdr_buf *xdr = &req->rq_snd_buf;
 321         int status;
 322
 323         xs_pktdump("packet data:",
 324                                 req->rq_svec->iov_base,
 325                                 req->rq_svec->iov_len);
 326
 327         req->rq_xtime = jiffies;
 328         status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
 329                                 xprt->addrlen, xdr, req->rq_bytes_sent);
 330
 331         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 332                         xdr->len - req->rq_bytes_sent, status);
 333
 334         if (likely(status >= (int) req->rq_slen))
 335                 return 0;
 336
 337         /* Still some bytes left; set up for a retry later. */
 338         if (status > 0)
 339                 status = -EAGAIN;
 340
 341         switch (status) {
 342         case -ENETUNREACH:
 343         case -EPIPE:
 344         case -ECONNREFUSED:
 345                 /* When the server has died, an ICMP port unreachable message
 346                  * prompts ECONNREFUSED. */
 347                 break;
 348         case -EAGAIN:
 349                 xs_nospace(task);
 350                 break;
 351         default:
 352                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 353                         -status);
 354                 break;
 355         }
 356
 357         return status;
 358 }
 359
 360 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 361 {
 362         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 363         rpc_fraghdr *base = buf->head[0].iov_base;
 364         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 365 }
 366
 367 /**
 368  * xs_tcp_send_request - write an RPC request to a TCP socket
 369  * @task: address of RPC task that manages the state of an RPC request
 370  *
 371  * Return values:
 372  *        0:    The request has been sent
 373  *   EAGAIN:    The socket was blocked, please call again later to
 374  *              complete the request
 375  * ENOTCONN:    Caller needs to invoke connect logic then call again
 376  *    other:    Some other error occured, the request was not sent
 377  *
 378  * XXX: In the case of soft timeouts, should we eventually give up
 379  *      if sendmsg is not able to make progress?
 380  */
 381 static int xs_tcp_send_request(struct rpc_task *task)
 382 {
 383         struct rpc_rqst *req = task->tk_rqstp;
 384         struct rpc_xprt *xprt = req->rq_xprt;
 385         struct xdr_buf *xdr = &req->rq_snd_buf;
 386         int status, retry = 0;
 387
 388         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 389
 390         xs_pktdump("packet data:",
 391                                 req->rq_svec->iov_base,
 392                                 req->rq_svec->iov_len);
 393
 394         /* Continue transmitting the packet/record. We must be careful
 395          * to cope with writespace callbacks arriving _after_ we have
 396          * called sendmsg(). */
 397         while (1) {
 398                 req->rq_xtime = jiffies;
 399                 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
 400                                                 req->rq_bytes_sent);
 401
 402                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 403                                 xdr->len - req->rq_bytes_sent, status);
 404
 405                 if (unlikely(status < 0))
 406                         break;
 407
 408                 /* If we've sent the entire packet, immediately
 409                  * reset the count of bytes sent. */
 410                 req->rq_bytes_sent += status;
 411                 task->tk_bytes_sent += status;
 412                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 413                         req->rq_bytes_sent = 0;
 414                         return 0;
 415                 }
 416
 417                 status = -EAGAIN;
 418                 if (retry++ > XS_SENDMSG_RETRY)
 419                         break;
 420         }
 421
 422         switch (status) {
 423         case -EAGAIN:
 424                 xs_nospace(task);
 425                 break;
 426         case -ECONNREFUSED:
 427         case -ECONNRESET:
 428         case -ENOTCONN:
 429         case -EPIPE:
 430                 status = -ENOTCONN;
 431                 break;
 432         default:
 433                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 434                         -status);
 435                 xprt_disconnect(xprt);
 436                 break;
 437         }
 438
 439         return status;
 440 }
 441
 442 /**
 443  * xs_tcp_release_xprt - clean up after a tcp transmission
 444  * @xprt: transport
 445  * @task: rpc task
 446  *
 447  * This cleans up if an error causes us to abort the transmission of a request.
 448  * In this case, the socket may need to be reset in order to avoid confusing
 449  * the server.
 450  */
 451 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 452 {
 453         struct rpc_rqst *req;
 454
 455         if (task != xprt->snd_task)
 456                 return;
 457         if (task == NULL)
 458                 goto out_release;
 459         req = task->tk_rqstp;
 460         if (req->rq_bytes_sent == 0)
 461                 goto out_release;
 462         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 463                 goto out_release;
 464         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 465 out_release:
 466         xprt_release_xprt(xprt, task);
 467 }
 468
 469 /**
 470  * xs_close - close a socket
 471  * @xprt: transport
 472  *
 473  * This is used when all requests are complete; ie, no DRC state remains
 474  * on the server we want to save.
 475  */
 476 static void xs_close(struct rpc_xprt *xprt)
 477 {
 478         struct socket *sock = xprt->sock;
 479         struct sock *sk = xprt->inet;
 480
 481         if (!sk)
 482                 goto clear_close_wait;
 483
 484         dprintk("RPC:      xs_close xprt %p\n", xprt);
 485
 486         write_lock_bh(&sk->sk_callback_lock);
 487         xprt->inet = NULL;
 488         xprt->sock = NULL;
 489
 490         sk->sk_user_data = NULL;
 491         sk->sk_data_ready = xprt->old_data_ready;
 492         sk->sk_state_change = xprt->old_state_change;
 493         sk->sk_write_space = xprt->old_write_space;
 494         write_unlock_bh(&sk->sk_callback_lock);
 495
 496         sk->sk_no_check = 0;
 497
 498         sock_release(sock);
 499 clear_close_wait:
 500         smp_mb__before_clear_bit();
 501         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 502         smp_mb__after_clear_bit();
 503 }
 504
 505 /**
 506  * xs_destroy - prepare to shutdown a transport
 507  * @xprt: doomed transport
 508  *
 509  */
 510 static void xs_destroy(struct rpc_xprt *xprt)
 511 {
 512         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 513
 514         cancel_delayed_work(&xprt->connect_worker);
 515         flush_scheduled_work();
 516
 517         xprt_disconnect(xprt);
 518         xs_close(xprt);
 519         xs_free_peer_addresses(xprt);
 520         kfree(xprt->slot);
 521         kfree(xprt);
 522 }
 523
 524 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 525 {
 526         return (struct rpc_xprt *) sk->sk_user_data;
 527 }
 528
 529 /**
 530  * xs_udp_data_ready - "data ready" callback for UDP sockets
 531  * @sk: socket with data to read
 532  * @len: how much data to read
 533  *
 534  */
 535 static void xs_udp_data_ready(struct sock *sk, int len)
 536 {
 537         struct rpc_task *task;
 538         struct rpc_xprt *xprt;
 539         struct rpc_rqst *rovr;
 540         struct sk_buff *skb;
 541         int err, repsize, copied;
 542         u32 _xid;
 543         __be32 *xp;
 544
 545         read_lock(&sk->sk_callback_lock);
 546         dprintk("RPC:      xs_udp_data_ready...\n");
 547         if (!(xprt = xprt_from_sock(sk)))
 548                 goto out;
 549
 550         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 551                 goto out;
 552
 553         if (xprt->shutdown)
 554                 goto dropit;
 555
 556         repsize = skb->len - sizeof(struct udphdr);
 557         if (repsize < 4) {
 558                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 559                 goto dropit;
 560         }
 561
 562         /* Copy the XID from the skb... */
 563         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 564                                 sizeof(_xid), &_xid);
 565         if (xp == NULL)
 566                 goto dropit;
 567
 568         /* Look up and lock the request corresponding to the given XID */
 569         spin_lock(&xprt->transport_lock);
 570         rovr = xprt_lookup_rqst(xprt, *xp);
 571         if (!rovr)
 572                 goto out_unlock;
 573         task = rovr->rq_task;
 574
 575         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 576                 copied = repsize;
 577
 578         /* Suck it into the iovec, verify checksum if not done by hw. */
 579         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 580                 goto out_unlock;
 581
 582         /* Something worked... */
 583         dst_confirm(skb->dst);
 584
 585         xprt_adjust_cwnd(task, copied);
 586         xprt_update_rtt(task);
 587         xprt_complete_rqst(task, copied);
 588
 589  out_unlock:
 590         spin_unlock(&xprt->transport_lock);
 591  dropit:
 592         skb_free_datagram(sk, skb);
 593  out:
 594         read_unlock(&sk->sk_callback_lock);
 595 }
 596
 597 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 598 {
 599         if (len > desc->count)
 600                 len = desc->count;
 601         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 602                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 603                                 len, desc->count);
 604                 return 0;
 605         }
 606         desc->offset += len;
 607         desc->count -= len;
 608         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 609                         len, desc->count);
 610         return len;
 611 }
 612
 613 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 614 {
 615         size_t len, used;
 616         char *p;
 617
 618         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 619         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 620         used = xs_tcp_copy_data(desc, p, len);
 621         xprt->tcp_offset += used;
 622         if (used != len)
 623                 return;
 624
 625         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 626         if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 627                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 628         else
 629                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 630         xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 631
 632         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 633         xprt->tcp_offset = 0;
 634
 635         /* Sanity check of the record length */
 636         if (unlikely(xprt->tcp_reclen < 4)) {
 637                 dprintk("RPC:      invalid TCP record fragment length\n");
 638                 xprt_disconnect(xprt);
 639                 return;
 640         }
 641         dprintk("RPC:      reading TCP record fragment of length %d\n",
 642                         xprt->tcp_reclen);
 643 }
 644
 645 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 646 {
 647         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 648                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 649         if (xprt->tcp_offset == xprt->tcp_reclen) {
 650                 xprt->tcp_flags |= XPRT_COPY_RECM;
 651                 xprt->tcp_offset = 0;
 652                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 653                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 654                         xprt->tcp_flags |= XPRT_COPY_XID;
 655                         xprt->tcp_copied = 0;
 656                 }
 657         }
 658 }
 659
 660 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 661 {
 662         size_t len, used;
 663         char *p;
 664
 665         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 666         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 667         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 668         used = xs_tcp_copy_data(desc, p, len);
 669         xprt->tcp_offset += used;
 670         if (used != len)
 671                 return;
 672         xprt->tcp_flags &= ~XPRT_COPY_XID;
 673         xprt->tcp_flags |= XPRT_COPY_DATA;
 674         xprt->tcp_copied = 4;
 675         dprintk("RPC:      reading reply for XID %08x\n",
 676                                                 ntohl(xprt->tcp_xid));
 677         xs_tcp_check_recm(xprt);
 678 }
 679
 680 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 681 {
 682         struct rpc_rqst *req;
 683         struct xdr_buf *rcvbuf;
 684         size_t len;
 685         ssize_t r;
 686
 687         /* Find and lock the request corresponding to this xid */
 688         spin_lock(&xprt->transport_lock);
 689         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 690         if (!req) {
 691                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 692                 dprintk("RPC:      XID %08x request not found!\n",
 693                                 ntohl(xprt->tcp_xid));
 694                 spin_unlock(&xprt->transport_lock);
 695                 return;
 696         }
 697
 698         rcvbuf = &req->rq_private_buf;
 699         len = desc->count;
 700         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 701                 skb_reader_t my_desc;
 702
 703                 len = xprt->tcp_reclen - xprt->tcp_offset;
 704                 memcpy(&my_desc, desc, sizeof(my_desc));
 705                 my_desc.count = len;
 706                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 707                                           &my_desc, xs_tcp_copy_data);
 708                 desc->count -= r;
 709                 desc->offset += r;
 710         } else
 711                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 712                                           desc, xs_tcp_copy_data);
 713
 714         if (r > 0) {
 715                 xprt->tcp_copied += r;
 716                 xprt->tcp_offset += r;
 717         }
 718         if (r != len) {
 719                 /* Error when copying to the receive buffer,
 720                  * usually because we weren't able to allocate
 721                  * additional buffer pages. All we can do now
 722                  * is turn off XPRT_COPY_DATA, so the request
 723                  * will not receive any additional updates,
 724                  * and time out.
 725                  * Any remaining data from this record will
 726                  * be discarded.
 727                  */
 728                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 729                 dprintk("RPC:      XID %08x truncated request\n",
 730                                 ntohl(xprt->tcp_xid));
 731                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 732                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 733                 goto out;
 734         }
 735
 736         dprintk("RPC:      XID %08x read %Zd bytes\n",
 737                         ntohl(xprt->tcp_xid), r);
 738         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 739                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 740
 741         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 742                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 743         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 744                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 745                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 746         }
 747
 748 out:
 749         if (!(xprt->tcp_flags & XPRT_COPY_DATA))
 750                 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
 751         spin_unlock(&xprt->transport_lock);
 752         xs_tcp_check_recm(xprt);
 753 }
 754
 755 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 756 {
 757         size_t len;
 758
 759         len = xprt->tcp_reclen - xprt->tcp_offset;
 760         if (len > desc->count)
 761                 len = desc->count;
 762         desc->count -= len;
 763         desc->offset += len;
 764         xprt->tcp_offset += len;
 765         dprintk("RPC:      discarded %Zu bytes\n", len);
 766         xs_tcp_check_recm(xprt);
 767 }
 768
 769 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 770 {
 771         struct rpc_xprt *xprt = rd_desc->arg.data;
 772         skb_reader_t desc = {
 773                 .skb    = skb,
 774                 .offset = offset,
 775                 .count  = len,
 776                 .csum   = 0
 777         };
 778
 779         dprintk("RPC:      xs_tcp_data_recv started\n");
 780         do {
 781                 /* Read in a new fragment marker if necessary */
 782                 /* Can we ever really expect to get completely empty fragments? */
 783                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 784                         xs_tcp_read_fraghdr(xprt, &desc);
 785                         continue;
 786                 }
 787                 /* Read in the xid if necessary */
 788                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 789                         xs_tcp_read_xid(xprt, &desc);
 790                         continue;
 791                 }
 792                 /* Read in the request data */
 793                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 794                         xs_tcp_read_request(xprt, &desc);
 795                         continue;
 796                 }
 797                 /* Skip over any trailing bytes on short reads */
 798                 xs_tcp_read_discard(xprt, &desc);
 799         } while (desc.count);
 800         dprintk("RPC:      xs_tcp_data_recv done\n");
 801         return len - desc.count;
 802 }
 803
 804 /**
 805  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 806  * @sk: socket with data to read
 807  * @bytes: how much data to read
 808  *
 809  */
 810 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 811 {
 812         struct rpc_xprt *xprt;
 813         read_descriptor_t rd_desc;
 814
 815         read_lock(&sk->sk_callback_lock);
 816         dprintk("RPC:      xs_tcp_data_ready...\n");
 817         if (!(xprt = xprt_from_sock(sk)))
 818                 goto out;
 819         if (xprt->shutdown)
 820                 goto out;
 821
 822         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 823         rd_desc.arg.data = xprt;
 824         rd_desc.count = 65536;
 825         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 826 out:
 827         read_unlock(&sk->sk_callback_lock);
 828 }
 829
 830 /**
 831  * xs_tcp_state_change - callback to handle TCP socket state changes
 832  * @sk: socket whose state has changed
 833  *
 834  */
 835 static void xs_tcp_state_change(struct sock *sk)
 836 {
 837         struct rpc_xprt *xprt;
 838
 839         read_lock(&sk->sk_callback_lock);
 840         if (!(xprt = xprt_from_sock(sk)))
 841                 goto out;
 842         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 843         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 844                                 sk->sk_state, xprt_connected(xprt),
 845                                 sock_flag(sk, SOCK_DEAD),
 846                                 sock_flag(sk, SOCK_ZAPPED));
 847
 848         switch (sk->sk_state) {
 849         case TCP_ESTABLISHED:
 850                 spin_lock_bh(&xprt->transport_lock);
 851                 if (!xprt_test_and_set_connected(xprt)) {
 852                         /* Reset TCP record info */
 853                         xprt->tcp_offset = 0;
 854                         xprt->tcp_reclen = 0;
 855                         xprt->tcp_copied = 0;
 856                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 857                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 858                         xprt_wake_pending_tasks(xprt, 0);
 859                 }
 860                 spin_unlock_bh(&xprt->transport_lock);
 861                 break;
 862         case TCP_SYN_SENT:
 863         case TCP_SYN_RECV:
 864                 break;
 865         case TCP_CLOSE_WAIT:
 866                 /* Try to schedule an autoclose RPC calls */
 867                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 868                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 869                         schedule_work(&xprt->task_cleanup);
 870         default:
 871                 xprt_disconnect(xprt);
 872         }
 873  out:
 874         read_unlock(&sk->sk_callback_lock);
 875 }
 876
 877 /**
 878  * xs_udp_write_space - callback invoked when socket buffer space
 879  *                             becomes available
 880  * @sk: socket whose state has changed
 881  *
 882  * Called when more output buffer space is available for this socket.
 883  * We try not to wake our writers until they can make "significant"
 884  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 885  * with a bunch of small requests.
 886  */
 887 static void xs_udp_write_space(struct sock *sk)
 888 {
 889         read_lock(&sk->sk_callback_lock);
 890
 891         /* from net/core/sock.c:sock_def_write_space */
 892         if (sock_writeable(sk)) {
 893                 struct socket *sock;
 894                 struct rpc_xprt *xprt;
 895
 896                 if (unlikely(!(sock = sk->sk_socket)))
 897                         goto out;
 898                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 899                         goto out;
 900                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 901                         goto out;
 902
 903                 xprt_write_space(xprt);
 904         }
 905
 906  out:
 907         read_unlock(&sk->sk_callback_lock);
 908 }
 909
 910 /**
 911  * xs_tcp_write_space - callback invoked when socket buffer space
 912  *                             becomes available
 913  * @sk: socket whose state has changed
 914  *
 915  * Called when more output buffer space is available for this socket.
 916  * We try not to wake our writers until they can make "significant"
 917  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 918  * with a bunch of small requests.
 919  */
 920 static void xs_tcp_write_space(struct sock *sk)
 921 {
 922         read_lock(&sk->sk_callback_lock);
 923
 924         /* from net/core/stream.c:sk_stream_write_space */
 925         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 926                 struct socket *sock;
 927                 struct rpc_xprt *xprt;
 928
 929                 if (unlikely(!(sock = sk->sk_socket)))
 930                         goto out;
 931                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 932                         goto out;
 933                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 934                         goto out;
 935
 936                 xprt_write_space(xprt);
 937         }
 938
 939  out:
 940         read_unlock(&sk->sk_callback_lock);
 941 }
 942
 943 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 944 {
 945         struct sock *sk = xprt->inet;
 946
 947         if (xprt->rcvsize) {
 948                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 949                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 950         }
 951         if (xprt->sndsize) {
 952                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 953                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 954                 sk->sk_write_space(sk);
 955         }
 956 }
 957
 958 /**
 959  * xs_udp_set_buffer_size - set send and receive limits
 960  * @xprt: generic transport
 961  * @sndsize: requested size of send buffer, in bytes
 962  * @rcvsize: requested size of receive buffer, in bytes
 963  *
 964  * Set socket send and receive buffer size limits.
 965  */
 966 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
 967 {
 968         xprt->sndsize = 0;
 969         if (sndsize)
 970                 xprt->sndsize = sndsize + 1024;
 971         xprt->rcvsize = 0;
 972         if (rcvsize)
 973                 xprt->rcvsize = rcvsize + 1024;
 974
 975         xs_udp_do_set_buffer_size(xprt);
 976 }
 977
 978 /**
 979  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
 980  * @task: task that timed out
 981  *
 982  * Adjust the congestion window after a retransmit timeout has occurred.
 983  */
 984 static void xs_udp_timer(struct rpc_task *task)
 985 {
 986         xprt_adjust_cwnd(task, -ETIMEDOUT);
 987 }
 988
 989 static unsigned short xs_get_random_port(void)
 990 {
 991         unsigned short range = xprt_max_resvport - xprt_min_resvport;
 992         unsigned short rand = (unsigned short) net_random() % range;
 993         return rand + xprt_min_resvport;
 994 }
 995
 996 /**
 997  * xs_print_peer_address - format an IPv4 address for printing
 998  * @xprt: generic transport
 999  * @format: flags field indicating which parts of the address to render
1000  */
1001 static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format)
1002 {
1003         if (xprt->address_strings[format] != NULL)
1004                 return xprt->address_strings[format];
1005         else
1006                 return "unprintable";
1007 }
1008
1009 /**
1010  * xs_set_port - reset the port number in the remote endpoint address
1011  * @xprt: generic transport
1012  * @port: new port number
1013  *
1014  */
1015 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1016 {
1017         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1018
1019         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
1020
1021         sap->sin_port = htons(port);
1022 }
1023
1024 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
1025 {
1026         struct sockaddr_in myaddr = {
1027                 .sin_family = AF_INET,
1028         };
1029         int err;
1030         unsigned short port = xprt->port;
1031
1032         do {
1033                 myaddr.sin_port = htons(port);
1034                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1035                                                 sizeof(myaddr));
1036                 if (err == 0) {
1037                         xprt->port = port;
1038                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
1039                                         port);
1040                         return 0;
1041                 }
1042                 if (port <= xprt_min_resvport)
1043                         port = xprt_max_resvport;
1044                 else
1045                         port--;
1046         } while (err == -EADDRINUSE && port != xprt->port);
1047
1048         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
1049         return err;
1050 }
1051
1052 /**
1053  * xs_udp_connect_worker - set up a UDP socket
1054  * @args: RPC transport to connect
1055  *
1056  * Invoked by a work queue tasklet.
1057  */
1058 static void xs_udp_connect_worker(void *args)
1059 {
1060         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
1061         struct socket *sock = xprt->sock;
1062         int err, status = -EIO;
1063
1064         if (xprt->shutdown || !xprt_bound(xprt))
1065                 goto out;
1066
1067         /* Start by resetting any existing state */
1068         xs_close(xprt);
1069
1070         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1071                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1072                 goto out;
1073         }
1074
1075         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1076                 sock_release(sock);
1077                 goto out;
1078         }
1079
1080         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1081                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1082
1083         if (!xprt->inet) {
1084                 struct sock *sk = sock->sk;
1085
1086                 write_lock_bh(&sk->sk_callback_lock);
1087
1088                 sk->sk_user_data = xprt;
1089                 xprt->old_data_ready = sk->sk_data_ready;
1090                 xprt->old_state_change = sk->sk_state_change;
1091                 xprt->old_write_space = sk->sk_write_space;
1092                 sk->sk_data_ready = xs_udp_data_ready;
1093                 sk->sk_write_space = xs_udp_write_space;
1094                 sk->sk_no_check = UDP_CSUM_NORCV;
1095                 sk->sk_allocation = GFP_ATOMIC;
1096
1097                 xprt_set_connected(xprt);
1098
1099                 /* Reset to new socket */
1100                 xprt->sock = sock;
1101                 xprt->inet = sk;
1102
1103                 write_unlock_bh(&sk->sk_callback_lock);
1104         }
1105         xs_udp_do_set_buffer_size(xprt);
1106         status = 0;
1107 out:
1108         xprt_wake_pending_tasks(xprt, status);
1109         xprt_clear_connecting(xprt);
1110 }
1111
1112 /*
1113  * We need to preserve the port number so the reply cache on the server can
1114  * find our cached RPC replies when we get around to reconnecting.
1115  */
1116 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1117 {
1118         int result;
1119         struct socket *sock = xprt->sock;
1120         struct sockaddr any;
1121
1122         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1123
1124         /*
1125          * Disconnect the transport socket by doing a connect operation
1126          * with AF_UNSPEC.  This should return immediately...
1127          */
1128         memset(&any, 0, sizeof(any));
1129         any.sa_family = AF_UNSPEC;
1130         result = kernel_connect(sock, &any, sizeof(any), 0);
1131         if (result)
1132                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1133                                 result);
1134 }
1135
1136 /**
1137  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1138  * @args: RPC transport to connect
1139  *
1140  * Invoked by a work queue tasklet.
1141  */
1142 static void xs_tcp_connect_worker(void *args)
1143 {
1144         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1145         struct socket *sock = xprt->sock;
1146         int err, status = -EIO;
1147
1148         if (xprt->shutdown || !xprt_bound(xprt))
1149                 goto out;
1150
1151         if (!xprt->sock) {
1152                 /* start from scratch */
1153                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1154                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1155                         goto out;
1156                 }
1157
1158                 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1159                         sock_release(sock);
1160                         goto out;
1161                 }
1162         } else
1163                 /* "close" the socket, preserving the local port */
1164                 xs_tcp_reuse_connection(xprt);
1165
1166         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1167                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1168
1169         if (!xprt->inet) {
1170                 struct sock *sk = sock->sk;
1171
1172                 write_lock_bh(&sk->sk_callback_lock);
1173
1174                 sk->sk_user_data = xprt;
1175                 xprt->old_data_ready = sk->sk_data_ready;
1176                 xprt->old_state_change = sk->sk_state_change;
1177                 xprt->old_write_space = sk->sk_write_space;
1178                 sk->sk_data_ready = xs_tcp_data_ready;
1179                 sk->sk_state_change = xs_tcp_state_change;
1180                 sk->sk_write_space = xs_tcp_write_space;
1181                 sk->sk_allocation = GFP_ATOMIC;
1182
1183                 /* socket options */
1184                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1185                 sock_reset_flag(sk, SOCK_LINGER);
1186                 tcp_sk(sk)->linger2 = 0;
1187                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1188
1189                 xprt_clear_connected(xprt);
1190
1191                 /* Reset to new socket */
1192                 xprt->sock = sock;
1193                 xprt->inet = sk;
1194
1195                 write_unlock_bh(&sk->sk_callback_lock);
1196         }
1197
1198         /* Tell the socket layer to start connecting... */
1199         xprt->stat.connect_count++;
1200         xprt->stat.connect_start = jiffies;
1201         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1202                         xprt->addrlen, O_NONBLOCK);
1203         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1204                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1205         if (status < 0) {
1206                 switch (status) {
1207                         case -EINPROGRESS:
1208                         case -EALREADY:
1209                                 goto out_clear;
1210                         case -ECONNREFUSED:
1211                         case -ECONNRESET:
1212                                 /* retry with existing socket, after a delay */
1213                                 break;
1214                         default:
1215                                 /* get rid of existing socket, and retry */
1216                                 xs_close(xprt);
1217                                 break;
1218                 }
1219         }
1220 out:
1221         xprt_wake_pending_tasks(xprt, status);
1222 out_clear:
1223         xprt_clear_connecting(xprt);
1224 }
1225
1226 /**
1227  * xs_connect - connect a socket to a remote endpoint
1228  * @task: address of RPC task that manages state of connect request
1229  *
1230  * TCP: If the remote end dropped the connection, delay reconnecting.
1231  *
1232  * UDP socket connects are synchronous, but we use a work queue anyway
1233  * to guarantee that even unprivileged user processes can set up a
1234  * socket on a privileged port.
1235  *
1236  * If a UDP socket connect fails, the delay behavior here prevents
1237  * retry floods (hard mounts).
1238  */
1239 static void xs_connect(struct rpc_task *task)
1240 {
1241         struct rpc_xprt *xprt = task->tk_xprt;
1242
1243         if (xprt_test_and_set_connecting(xprt))
1244                 return;
1245
1246         if (xprt->sock != NULL) {
1247                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1248                                 xprt, xprt->reestablish_timeout / HZ);
1249                 schedule_delayed_work(&xprt->connect_worker,
1250                                         xprt->reestablish_timeout);
1251                 xprt->reestablish_timeout <<= 1;
1252                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1253                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1254         } else {
1255                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1256                 schedule_work(&xprt->connect_worker);
1257
1258                 /* flush_scheduled_work can sleep... */
1259                 if (!RPC_IS_ASYNC(task))
1260                         flush_scheduled_work();
1261         }
1262 }
1263
1264 /**
1265  * xs_udp_print_stats - display UDP socket-specifc stats
1266  * @xprt: rpc_xprt struct containing statistics
1267  * @seq: output file
1268  *
1269  */
1270 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1271 {
1272         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1273                         xprt->port,
1274                         xprt->stat.bind_count,
1275                         xprt->stat.sends,
1276                         xprt->stat.recvs,
1277                         xprt->stat.bad_xids,
1278                         xprt->stat.req_u,
1279                         xprt->stat.bklog_u);
1280 }
1281
1282 /**
1283  * xs_tcp_print_stats - display TCP socket-specifc stats
1284  * @xprt: rpc_xprt struct containing statistics
1285  * @seq: output file
1286  *
1287  */
1288 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1289 {
1290         long idle_time = 0;
1291
1292         if (xprt_connected(xprt))
1293                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1294
1295         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1296                         xprt->port,
1297                         xprt->stat.bind_count,
1298                         xprt->stat.connect_count,
1299                         xprt->stat.connect_time,
1300                         idle_time,
1301                         xprt->stat.sends,
1302                         xprt->stat.recvs,
1303                         xprt->stat.bad_xids,
1304                         xprt->stat.req_u,
1305                         xprt->stat.bklog_u);
1306 }
1307
1308 static struct rpc_xprt_ops xs_udp_ops = {
1309         .set_buffer_size        = xs_udp_set_buffer_size,
1310         .print_addr             = xs_print_peer_address,
1311         .reserve_xprt           = xprt_reserve_xprt_cong,
1312         .release_xprt           = xprt_release_xprt_cong,
1313         .rpcbind                = rpc_getport,
1314         .set_port               = xs_set_port,
1315         .connect                = xs_connect,
1316         .buf_alloc              = rpc_malloc,
1317         .buf_free               = rpc_free,
1318         .send_request           = xs_udp_send_request,
1319         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1320         .timer                  = xs_udp_timer,
1321         .release_request        = xprt_release_rqst_cong,
1322         .close                  = xs_close,
1323         .destroy                = xs_destroy,
1324         .print_stats            = xs_udp_print_stats,
1325 };
1326
1327 static struct rpc_xprt_ops xs_tcp_ops = {
1328         .print_addr             = xs_print_peer_address,
1329         .reserve_xprt           = xprt_reserve_xprt,
1330         .release_xprt           = xs_tcp_release_xprt,
1331         .rpcbind                = rpc_getport,
1332         .set_port               = xs_set_port,
1333         .connect                = xs_connect,
1334         .buf_alloc              = rpc_malloc,
1335         .buf_free               = rpc_free,
1336         .send_request           = xs_tcp_send_request,
1337         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1338         .close                  = xs_close,
1339         .destroy                = xs_destroy,
1340         .print_stats            = xs_tcp_print_stats,
1341 };
1342
1343 static struct rpc_xprt *xs_setup_xprt(struct sockaddr *addr, size_t addrlen, unsigned int slot_table_size)
1344 {
1345         struct rpc_xprt *xprt;
1346
1347         if (addrlen > sizeof(xprt->addr)) {
1348                 dprintk("RPC:      xs_setup_xprt: address too large\n");
1349                 return ERR_PTR(-EBADF);
1350         }
1351
1352         xprt = kzalloc(sizeof(struct rpc_xprt), GFP_KERNEL);
1353         if (xprt == NULL) {
1354                 dprintk("RPC:      xs_setup_xprt: couldn't allocate rpc_xprt\n");
1355                 return ERR_PTR(-ENOMEM);
1356         }
1357
1358         xprt->max_reqs = slot_table_size;
1359         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
1360         if (xprt->slot == NULL) {
1361                 kfree(xprt);
1362                 dprintk("RPC:      xs_setup_xprt: couldn't allocate slot table\n");
1363                 return ERR_PTR(-ENOMEM);
1364         }
1365
1366         memcpy(&xprt->addr, addr, addrlen);
1367         xprt->addrlen = addrlen;
1368         xprt->port = xs_get_random_port();
1369
1370         return xprt;
1371 }
1372
1373 /**
1374  * xs_setup_udp - Set up transport to use a UDP socket
1375  * @addr: address of remote server
1376  * @addrlen: length of address in bytes
1377  * @to:   timeout parameters
1378  *
1379  */
1380 struct rpc_xprt *xs_setup_udp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1381 {
1382         struct rpc_xprt *xprt;
1383
1384         xprt = xs_setup_xprt(addr, addrlen, xprt_udp_slot_table_entries);
1385         if (IS_ERR(xprt))
1386                 return xprt;
1387
1388         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1389                 xprt_set_bound(xprt);
1390
1391         xprt->prot = IPPROTO_UDP;
1392         xprt->tsh_size = 0;
1393         /* XXX: header size can vary due to auth type, IPv6, etc. */
1394         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1395
1396         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1397         xprt->bind_timeout = XS_BIND_TO;
1398         xprt->connect_timeout = XS_UDP_CONN_TO;
1399         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1400         xprt->idle_timeout = XS_IDLE_DISC_TO;
1401
1402         xprt->ops = &xs_udp_ops;
1403
1404         if (to)
1405                 xprt->timeout = *to;
1406         else
1407                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1408
1409         xs_format_peer_addresses(xprt);
1410         dprintk("RPC:      set up transport to address %s\n",
1411                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1412
1413         return xprt;
1414 }
1415
1416 /**
1417  * xs_setup_tcp - Set up transport to use a TCP socket
1418  * @addr: address of remote server
1419  * @addrlen: length of address in bytes
1420  * @to: timeout parameters
1421  *
1422  */
1423 struct rpc_xprt *xs_setup_tcp(struct sockaddr *addr, size_t addrlen, struct rpc_timeout *to)
1424 {
1425         struct rpc_xprt *xprt;
1426
1427         xprt = xs_setup_xprt(addr, addrlen, xprt_tcp_slot_table_entries);
1428         if (IS_ERR(xprt))
1429                 return xprt;
1430
1431         if (ntohs(((struct sockaddr_in *)addr)->sin_port) != 0)
1432                 xprt_set_bound(xprt);
1433
1434         xprt->prot = IPPROTO_TCP;
1435         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1436         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1437
1438         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1439         xprt->bind_timeout = XS_BIND_TO;
1440         xprt->connect_timeout = XS_TCP_CONN_TO;
1441         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1442         xprt->idle_timeout = XS_IDLE_DISC_TO;
1443
1444         xprt->ops = &xs_tcp_ops;
1445
1446         if (to)
1447                 xprt->timeout = *to;
1448         else
1449                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1450
1451         xs_format_peer_addresses(xprt);
1452         dprintk("RPC:      set up transport to address %s\n",
1453                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1454
1455         return xprt;
1456 }