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 }
 522
 523 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 524 {
 525         return (struct rpc_xprt *) sk->sk_user_data;
 526 }
 527
 528 /**
 529  * xs_udp_data_ready - "data ready" callback for UDP sockets
 530  * @sk: socket with data to read
 531  * @len: how much data to read
 532  *
 533  */
 534 static void xs_udp_data_ready(struct sock *sk, int len)
 535 {
 536         struct rpc_task *task;
 537         struct rpc_xprt *xprt;
 538         struct rpc_rqst *rovr;
 539         struct sk_buff *skb;
 540         int err, repsize, copied;
 541         u32 _xid;
 542         __be32 *xp;
 543
 544         read_lock(&sk->sk_callback_lock);
 545         dprintk("RPC:      xs_udp_data_ready...\n");
 546         if (!(xprt = xprt_from_sock(sk)))
 547                 goto out;
 548
 549         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 550                 goto out;
 551
 552         if (xprt->shutdown)
 553                 goto dropit;
 554
 555         repsize = skb->len - sizeof(struct udphdr);
 556         if (repsize < 4) {
 557                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 558                 goto dropit;
 559         }
 560
 561         /* Copy the XID from the skb... */
 562         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 563                                 sizeof(_xid), &_xid);
 564         if (xp == NULL)
 565                 goto dropit;
 566
 567         /* Look up and lock the request corresponding to the given XID */
 568         spin_lock(&xprt->transport_lock);
 569         rovr = xprt_lookup_rqst(xprt, *xp);
 570         if (!rovr)
 571                 goto out_unlock;
 572         task = rovr->rq_task;
 573
 574         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 575                 copied = repsize;
 576
 577         /* Suck it into the iovec, verify checksum if not done by hw. */
 578         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 579                 goto out_unlock;
 580
 581         /* Something worked... */
 582         dst_confirm(skb->dst);
 583
 584         xprt_adjust_cwnd(task, copied);
 585         xprt_update_rtt(task);
 586         xprt_complete_rqst(task, copied);
 587
 588  out_unlock:
 589         spin_unlock(&xprt->transport_lock);
 590  dropit:
 591         skb_free_datagram(sk, skb);
 592  out:
 593         read_unlock(&sk->sk_callback_lock);
 594 }
 595
 596 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 597 {
 598         if (len > desc->count)
 599                 len = desc->count;
 600         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 601                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 602                                 len, desc->count);
 603                 return 0;
 604         }
 605         desc->offset += len;
 606         desc->count -= len;
 607         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 608                         len, desc->count);
 609         return len;
 610 }
 611
 612 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 613 {
 614         size_t len, used;
 615         char *p;
 616
 617         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 618         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 619         used = xs_tcp_copy_data(desc, p, len);
 620         xprt->tcp_offset += used;
 621         if (used != len)
 622                 return;
 623
 624         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 625         if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 626                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 627         else
 628                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 629         xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 630
 631         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 632         xprt->tcp_offset = 0;
 633
 634         /* Sanity check of the record length */
 635         if (unlikely(xprt->tcp_reclen < 4)) {
 636                 dprintk("RPC:      invalid TCP record fragment length\n");
 637                 xprt_disconnect(xprt);
 638                 return;
 639         }
 640         dprintk("RPC:      reading TCP record fragment of length %d\n",
 641                         xprt->tcp_reclen);
 642 }
 643
 644 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 645 {
 646         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 647                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 648         if (xprt->tcp_offset == xprt->tcp_reclen) {
 649                 xprt->tcp_flags |= XPRT_COPY_RECM;
 650                 xprt->tcp_offset = 0;
 651                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 652                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 653                         xprt->tcp_flags |= XPRT_COPY_XID;
 654                         xprt->tcp_copied = 0;
 655                 }
 656         }
 657 }
 658
 659 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 660 {
 661         size_t len, used;
 662         char *p;
 663
 664         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 665         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 666         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 667         used = xs_tcp_copy_data(desc, p, len);
 668         xprt->tcp_offset += used;
 669         if (used != len)
 670                 return;
 671         xprt->tcp_flags &= ~XPRT_COPY_XID;
 672         xprt->tcp_flags |= XPRT_COPY_DATA;
 673         xprt->tcp_copied = 4;
 674         dprintk("RPC:      reading reply for XID %08x\n",
 675                                                 ntohl(xprt->tcp_xid));
 676         xs_tcp_check_recm(xprt);
 677 }
 678
 679 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 680 {
 681         struct rpc_rqst *req;
 682         struct xdr_buf *rcvbuf;
 683         size_t len;
 684         ssize_t r;
 685
 686         /* Find and lock the request corresponding to this xid */
 687         spin_lock(&xprt->transport_lock);
 688         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 689         if (!req) {
 690                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 691                 dprintk("RPC:      XID %08x request not found!\n",
 692                                 ntohl(xprt->tcp_xid));
 693                 spin_unlock(&xprt->transport_lock);
 694                 return;
 695         }
 696
 697         rcvbuf = &req->rq_private_buf;
 698         len = desc->count;
 699         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 700                 skb_reader_t my_desc;
 701
 702                 len = xprt->tcp_reclen - xprt->tcp_offset;
 703                 memcpy(&my_desc, desc, sizeof(my_desc));
 704                 my_desc.count = len;
 705                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 706                                           &my_desc, xs_tcp_copy_data);
 707                 desc->count -= r;
 708                 desc->offset += r;
 709         } else
 710                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 711                                           desc, xs_tcp_copy_data);
 712
 713         if (r > 0) {
 714                 xprt->tcp_copied += r;
 715                 xprt->tcp_offset += r;
 716         }
 717         if (r != len) {
 718                 /* Error when copying to the receive buffer,
 719                  * usually because we weren't able to allocate
 720                  * additional buffer pages. All we can do now
 721                  * is turn off XPRT_COPY_DATA, so the request
 722                  * will not receive any additional updates,
 723                  * and time out.
 724                  * Any remaining data from this record will
 725                  * be discarded.
 726                  */
 727                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 728                 dprintk("RPC:      XID %08x truncated request\n",
 729                                 ntohl(xprt->tcp_xid));
 730                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 731                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 732                 goto out;
 733         }
 734
 735         dprintk("RPC:      XID %08x read %Zd bytes\n",
 736                         ntohl(xprt->tcp_xid), r);
 737         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 738                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 739
 740         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 741                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 742         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 743                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 744                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 745         }
 746
 747 out:
 748         if (!(xprt->tcp_flags & XPRT_COPY_DATA))
 749                 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
 750         spin_unlock(&xprt->transport_lock);
 751         xs_tcp_check_recm(xprt);
 752 }
 753
 754 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 755 {
 756         size_t len;
 757
 758         len = xprt->tcp_reclen - xprt->tcp_offset;
 759         if (len > desc->count)
 760                 len = desc->count;
 761         desc->count -= len;
 762         desc->offset += len;
 763         xprt->tcp_offset += len;
 764         dprintk("RPC:      discarded %Zu bytes\n", len);
 765         xs_tcp_check_recm(xprt);
 766 }
 767
 768 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 769 {
 770         struct rpc_xprt *xprt = rd_desc->arg.data;
 771         skb_reader_t desc = {
 772                 .skb    = skb,
 773                 .offset = offset,
 774                 .count  = len,
 775                 .csum   = 0
 776         };
 777
 778         dprintk("RPC:      xs_tcp_data_recv started\n");
 779         do {
 780                 /* Read in a new fragment marker if necessary */
 781                 /* Can we ever really expect to get completely empty fragments? */
 782                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 783                         xs_tcp_read_fraghdr(xprt, &desc);
 784                         continue;
 785                 }
 786                 /* Read in the xid if necessary */
 787                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 788                         xs_tcp_read_xid(xprt, &desc);
 789                         continue;
 790                 }
 791                 /* Read in the request data */
 792                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 793                         xs_tcp_read_request(xprt, &desc);
 794                         continue;
 795                 }
 796                 /* Skip over any trailing bytes on short reads */
 797                 xs_tcp_read_discard(xprt, &desc);
 798         } while (desc.count);
 799         dprintk("RPC:      xs_tcp_data_recv done\n");
 800         return len - desc.count;
 801 }
 802
 803 /**
 804  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 805  * @sk: socket with data to read
 806  * @bytes: how much data to read
 807  *
 808  */
 809 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 810 {
 811         struct rpc_xprt *xprt;
 812         read_descriptor_t rd_desc;
 813
 814         read_lock(&sk->sk_callback_lock);
 815         dprintk("RPC:      xs_tcp_data_ready...\n");
 816         if (!(xprt = xprt_from_sock(sk)))
 817                 goto out;
 818         if (xprt->shutdown)
 819                 goto out;
 820
 821         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 822         rd_desc.arg.data = xprt;
 823         rd_desc.count = 65536;
 824         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 825 out:
 826         read_unlock(&sk->sk_callback_lock);
 827 }
 828
 829 /**
 830  * xs_tcp_state_change - callback to handle TCP socket state changes
 831  * @sk: socket whose state has changed
 832  *
 833  */
 834 static void xs_tcp_state_change(struct sock *sk)
 835 {
 836         struct rpc_xprt *xprt;
 837
 838         read_lock(&sk->sk_callback_lock);
 839         if (!(xprt = xprt_from_sock(sk)))
 840                 goto out;
 841         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 842         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 843                                 sk->sk_state, xprt_connected(xprt),
 844                                 sock_flag(sk, SOCK_DEAD),
 845                                 sock_flag(sk, SOCK_ZAPPED));
 846
 847         switch (sk->sk_state) {
 848         case TCP_ESTABLISHED:
 849                 spin_lock_bh(&xprt->transport_lock);
 850                 if (!xprt_test_and_set_connected(xprt)) {
 851                         /* Reset TCP record info */
 852                         xprt->tcp_offset = 0;
 853                         xprt->tcp_reclen = 0;
 854                         xprt->tcp_copied = 0;
 855                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 856                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 857                         xprt_wake_pending_tasks(xprt, 0);
 858                 }
 859                 spin_unlock_bh(&xprt->transport_lock);
 860                 break;
 861         case TCP_SYN_SENT:
 862         case TCP_SYN_RECV:
 863                 break;
 864         case TCP_CLOSE_WAIT:
 865                 /* Try to schedule an autoclose RPC calls */
 866                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 867                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 868                         schedule_work(&xprt->task_cleanup);
 869         default:
 870                 xprt_disconnect(xprt);
 871         }
 872  out:
 873         read_unlock(&sk->sk_callback_lock);
 874 }
 875
 876 /**
 877  * xs_udp_write_space - callback invoked when socket buffer space
 878  *                             becomes available
 879  * @sk: socket whose state has changed
 880  *
 881  * Called when more output buffer space is available for this socket.
 882  * We try not to wake our writers until they can make "significant"
 883  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 884  * with a bunch of small requests.
 885  */
 886 static void xs_udp_write_space(struct sock *sk)
 887 {
 888         read_lock(&sk->sk_callback_lock);
 889
 890         /* from net/core/sock.c:sock_def_write_space */
 891         if (sock_writeable(sk)) {
 892                 struct socket *sock;
 893                 struct rpc_xprt *xprt;
 894
 895                 if (unlikely(!(sock = sk->sk_socket)))
 896                         goto out;
 897                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 898                         goto out;
 899                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 900                         goto out;
 901
 902                 xprt_write_space(xprt);
 903         }
 904
 905  out:
 906         read_unlock(&sk->sk_callback_lock);
 907 }
 908
 909 /**
 910  * xs_tcp_write_space - callback invoked when socket buffer space
 911  *                             becomes available
 912  * @sk: socket whose state has changed
 913  *
 914  * Called when more output buffer space is available for this socket.
 915  * We try not to wake our writers until they can make "significant"
 916  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 917  * with a bunch of small requests.
 918  */
 919 static void xs_tcp_write_space(struct sock *sk)
 920 {
 921         read_lock(&sk->sk_callback_lock);
 922
 923         /* from net/core/stream.c:sk_stream_write_space */
 924         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 925                 struct socket *sock;
 926                 struct rpc_xprt *xprt;
 927
 928                 if (unlikely(!(sock = sk->sk_socket)))
 929                         goto out;
 930                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 931                         goto out;
 932                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 933                         goto out;
 934
 935                 xprt_write_space(xprt);
 936         }
 937
 938  out:
 939         read_unlock(&sk->sk_callback_lock);
 940 }
 941
 942 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 943 {
 944         struct sock *sk = xprt->inet;
 945
 946         if (xprt->rcvsize) {
 947                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 948                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 949         }
 950         if (xprt->sndsize) {
 951                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 952                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 953                 sk->sk_write_space(sk);
 954         }
 955 }
 956
 957 /**
 958  * xs_udp_set_buffer_size - set send and receive limits
 959  * @xprt: generic transport
 960  * @sndsize: requested size of send buffer, in bytes
 961  * @rcvsize: requested size of receive buffer, in bytes
 962  *
 963  * Set socket send and receive buffer size limits.
 964  */
 965 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
 966 {
 967         xprt->sndsize = 0;
 968         if (sndsize)
 969                 xprt->sndsize = sndsize + 1024;
 970         xprt->rcvsize = 0;
 971         if (rcvsize)
 972                 xprt->rcvsize = rcvsize + 1024;
 973
 974         xs_udp_do_set_buffer_size(xprt);
 975 }
 976
 977 /**
 978  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
 979  * @task: task that timed out
 980  *
 981  * Adjust the congestion window after a retransmit timeout has occurred.
 982  */
 983 static void xs_udp_timer(struct rpc_task *task)
 984 {
 985         xprt_adjust_cwnd(task, -ETIMEDOUT);
 986 }
 987
 988 static unsigned short xs_get_random_port(void)
 989 {
 990         unsigned short range = xprt_max_resvport - xprt_min_resvport;
 991         unsigned short rand = (unsigned short) net_random() % range;
 992         return rand + xprt_min_resvport;
 993 }
 994
 995 /**
 996  * xs_print_peer_address - format an IPv4 address for printing
 997  * @xprt: generic transport
 998  * @format: flags field indicating which parts of the address to render
 999  */
1000 static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format)
1001 {
1002         if (xprt->address_strings[format] != NULL)
1003                 return xprt->address_strings[format];
1004         else
1005                 return "unprintable";
1006 }
1007
1008 /**
1009  * xs_set_port - reset the port number in the remote endpoint address
1010  * @xprt: generic transport
1011  * @port: new port number
1012  *
1013  */
1014 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1015 {
1016         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1017
1018         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
1019
1020         sap->sin_port = htons(port);
1021 }
1022
1023 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
1024 {
1025         struct sockaddr_in myaddr = {
1026                 .sin_family = AF_INET,
1027         };
1028         int err;
1029         unsigned short port = xprt->port;
1030
1031         do {
1032                 myaddr.sin_port = htons(port);
1033                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1034                                                 sizeof(myaddr));
1035                 if (err == 0) {
1036                         xprt->port = port;
1037                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
1038                                         port);
1039                         return 0;
1040                 }
1041                 if (port <= xprt_min_resvport)
1042                         port = xprt_max_resvport;
1043                 else
1044                         port--;
1045         } while (err == -EADDRINUSE && port != xprt->port);
1046
1047         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
1048         return err;
1049 }
1050
1051 /**
1052  * xs_udp_connect_worker - set up a UDP socket
1053  * @args: RPC transport to connect
1054  *
1055  * Invoked by a work queue tasklet.
1056  */
1057 static void xs_udp_connect_worker(void *args)
1058 {
1059         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
1060         struct socket *sock = xprt->sock;
1061         int err, status = -EIO;
1062
1063         if (xprt->shutdown || !xprt_bound(xprt))
1064                 goto out;
1065
1066         /* Start by resetting any existing state */
1067         xs_close(xprt);
1068
1069         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1070                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1071                 goto out;
1072         }
1073
1074         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1075                 sock_release(sock);
1076                 goto out;
1077         }
1078
1079         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1080                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1081
1082         if (!xprt->inet) {
1083                 struct sock *sk = sock->sk;
1084
1085                 write_lock_bh(&sk->sk_callback_lock);
1086
1087                 sk->sk_user_data = xprt;
1088                 xprt->old_data_ready = sk->sk_data_ready;
1089                 xprt->old_state_change = sk->sk_state_change;
1090                 xprt->old_write_space = sk->sk_write_space;
1091                 sk->sk_data_ready = xs_udp_data_ready;
1092                 sk->sk_write_space = xs_udp_write_space;
1093                 sk->sk_no_check = UDP_CSUM_NORCV;
1094                 sk->sk_allocation = GFP_ATOMIC;
1095
1096                 xprt_set_connected(xprt);
1097
1098                 /* Reset to new socket */
1099                 xprt->sock = sock;
1100                 xprt->inet = sk;
1101
1102                 write_unlock_bh(&sk->sk_callback_lock);
1103         }
1104         xs_udp_do_set_buffer_size(xprt);
1105         status = 0;
1106 out:
1107         xprt_wake_pending_tasks(xprt, status);
1108         xprt_clear_connecting(xprt);
1109 }
1110
1111 /*
1112  * We need to preserve the port number so the reply cache on the server can
1113  * find our cached RPC replies when we get around to reconnecting.
1114  */
1115 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1116 {
1117         int result;
1118         struct socket *sock = xprt->sock;
1119         struct sockaddr any;
1120
1121         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1122
1123         /*
1124          * Disconnect the transport socket by doing a connect operation
1125          * with AF_UNSPEC.  This should return immediately...
1126          */
1127         memset(&any, 0, sizeof(any));
1128         any.sa_family = AF_UNSPEC;
1129         result = kernel_connect(sock, &any, sizeof(any), 0);
1130         if (result)
1131                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1132                                 result);
1133 }
1134
1135 /**
1136  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1137  * @args: RPC transport to connect
1138  *
1139  * Invoked by a work queue tasklet.
1140  */
1141 static void xs_tcp_connect_worker(void *args)
1142 {
1143         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1144         struct socket *sock = xprt->sock;
1145         int err, status = -EIO;
1146
1147         if (xprt->shutdown || !xprt_bound(xprt))
1148                 goto out;
1149
1150         if (!xprt->sock) {
1151                 /* start from scratch */
1152                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1153                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1154                         goto out;
1155                 }
1156
1157                 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1158                         sock_release(sock);
1159                         goto out;
1160                 }
1161         } else
1162                 /* "close" the socket, preserving the local port */
1163                 xs_tcp_reuse_connection(xprt);
1164
1165         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1166                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1167
1168         if (!xprt->inet) {
1169                 struct sock *sk = sock->sk;
1170
1171                 write_lock_bh(&sk->sk_callback_lock);
1172
1173                 sk->sk_user_data = xprt;
1174                 xprt->old_data_ready = sk->sk_data_ready;
1175                 xprt->old_state_change = sk->sk_state_change;
1176                 xprt->old_write_space = sk->sk_write_space;
1177                 sk->sk_data_ready = xs_tcp_data_ready;
1178                 sk->sk_state_change = xs_tcp_state_change;
1179                 sk->sk_write_space = xs_tcp_write_space;
1180                 sk->sk_allocation = GFP_ATOMIC;
1181
1182                 /* socket options */
1183                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1184                 sock_reset_flag(sk, SOCK_LINGER);
1185                 tcp_sk(sk)->linger2 = 0;
1186                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1187
1188                 xprt_clear_connected(xprt);
1189
1190                 /* Reset to new socket */
1191                 xprt->sock = sock;
1192                 xprt->inet = sk;
1193
1194                 write_unlock_bh(&sk->sk_callback_lock);
1195         }
1196
1197         /* Tell the socket layer to start connecting... */
1198         xprt->stat.connect_count++;
1199         xprt->stat.connect_start = jiffies;
1200         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1201                         xprt->addrlen, O_NONBLOCK);
1202         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1203                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1204         if (status < 0) {
1205                 switch (status) {
1206                         case -EINPROGRESS:
1207                         case -EALREADY:
1208                                 goto out_clear;
1209                         case -ECONNREFUSED:
1210                         case -ECONNRESET:
1211                                 /* retry with existing socket, after a delay */
1212                                 break;
1213                         default:
1214                                 /* get rid of existing socket, and retry */
1215                                 xs_close(xprt);
1216                                 break;
1217                 }
1218         }
1219 out:
1220         xprt_wake_pending_tasks(xprt, status);
1221 out_clear:
1222         xprt_clear_connecting(xprt);
1223 }
1224
1225 /**
1226  * xs_connect - connect a socket to a remote endpoint
1227  * @task: address of RPC task that manages state of connect request
1228  *
1229  * TCP: If the remote end dropped the connection, delay reconnecting.
1230  *
1231  * UDP socket connects are synchronous, but we use a work queue anyway
1232  * to guarantee that even unprivileged user processes can set up a
1233  * socket on a privileged port.
1234  *
1235  * If a UDP socket connect fails, the delay behavior here prevents
1236  * retry floods (hard mounts).
1237  */
1238 static void xs_connect(struct rpc_task *task)
1239 {
1240         struct rpc_xprt *xprt = task->tk_xprt;
1241
1242         if (xprt_test_and_set_connecting(xprt))
1243                 return;
1244
1245         if (xprt->sock != NULL) {
1246                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1247                                 xprt, xprt->reestablish_timeout / HZ);
1248                 schedule_delayed_work(&xprt->connect_worker,
1249                                         xprt->reestablish_timeout);
1250                 xprt->reestablish_timeout <<= 1;
1251                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1252                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1253         } else {
1254                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1255                 schedule_work(&xprt->connect_worker);
1256
1257                 /* flush_scheduled_work can sleep... */
1258                 if (!RPC_IS_ASYNC(task))
1259                         flush_scheduled_work();
1260         }
1261 }
1262
1263 /**
1264  * xs_udp_print_stats - display UDP socket-specifc stats
1265  * @xprt: rpc_xprt struct containing statistics
1266  * @seq: output file
1267  *
1268  */
1269 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1270 {
1271         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1272                         xprt->port,
1273                         xprt->stat.bind_count,
1274                         xprt->stat.sends,
1275                         xprt->stat.recvs,
1276                         xprt->stat.bad_xids,
1277                         xprt->stat.req_u,
1278                         xprt->stat.bklog_u);
1279 }
1280
1281 /**
1282  * xs_tcp_print_stats - display TCP socket-specifc stats
1283  * @xprt: rpc_xprt struct containing statistics
1284  * @seq: output file
1285  *
1286  */
1287 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1288 {
1289         long idle_time = 0;
1290
1291         if (xprt_connected(xprt))
1292                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1293
1294         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1295                         xprt->port,
1296                         xprt->stat.bind_count,
1297                         xprt->stat.connect_count,
1298                         xprt->stat.connect_time,
1299                         idle_time,
1300                         xprt->stat.sends,
1301                         xprt->stat.recvs,
1302                         xprt->stat.bad_xids,
1303                         xprt->stat.req_u,
1304                         xprt->stat.bklog_u);
1305 }
1306
1307 static struct rpc_xprt_ops xs_udp_ops = {
1308         .set_buffer_size        = xs_udp_set_buffer_size,
1309         .print_addr             = xs_print_peer_address,
1310         .reserve_xprt           = xprt_reserve_xprt_cong,
1311         .release_xprt           = xprt_release_xprt_cong,
1312         .rpcbind                = rpc_getport,
1313         .set_port               = xs_set_port,
1314         .connect                = xs_connect,
1315         .buf_alloc              = rpc_malloc,
1316         .buf_free               = rpc_free,
1317         .send_request           = xs_udp_send_request,
1318         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1319         .timer                  = xs_udp_timer,
1320         .release_request        = xprt_release_rqst_cong,
1321         .close                  = xs_close,
1322         .destroy                = xs_destroy,
1323         .print_stats            = xs_udp_print_stats,
1324 };
1325
1326 static struct rpc_xprt_ops xs_tcp_ops = {
1327         .print_addr             = xs_print_peer_address,
1328         .reserve_xprt           = xprt_reserve_xprt,
1329         .release_xprt           = xs_tcp_release_xprt,
1330         .rpcbind                = rpc_getport,
1331         .set_port               = xs_set_port,
1332         .connect                = xs_connect,
1333         .buf_alloc              = rpc_malloc,
1334         .buf_free               = rpc_free,
1335         .send_request           = xs_tcp_send_request,
1336         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1337         .close                  = xs_close,
1338         .destroy                = xs_destroy,
1339         .print_stats            = xs_tcp_print_stats,
1340 };
1341
1342 /**
1343  * xs_setup_udp - Set up transport to use a UDP socket
1344  * @xprt: transport to set up
1345  * @to:   timeout parameters
1346  *
1347  */
1348 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1349 {
1350         size_t slot_table_size;
1351         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
1352
1353         xprt->max_reqs = xprt_udp_slot_table_entries;
1354         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1355         xprt->slot = kzalloc(slot_table_size, GFP_KERNEL);
1356         if (xprt->slot == NULL)
1357                 return -ENOMEM;
1358
1359         if (ntohs(addr->sin_port) != 0)
1360                 xprt_set_bound(xprt);
1361         xprt->port = xs_get_random_port();
1362
1363         xprt->prot = IPPROTO_UDP;
1364         xprt->tsh_size = 0;
1365         /* XXX: header size can vary due to auth type, IPv6, etc. */
1366         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1367
1368         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1369         xprt->bind_timeout = XS_BIND_TO;
1370         xprt->connect_timeout = XS_UDP_CONN_TO;
1371         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1372         xprt->idle_timeout = XS_IDLE_DISC_TO;
1373
1374         xprt->ops = &xs_udp_ops;
1375
1376         if (to)
1377                 xprt->timeout = *to;
1378         else
1379                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1380
1381         xs_format_peer_addresses(xprt);
1382         dprintk("RPC:      set up transport to address %s\n",
1383                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1384
1385         return 0;
1386 }
1387
1388 /**
1389  * xs_setup_tcp - Set up transport to use a TCP socket
1390  * @xprt: transport to set up
1391  * @to: timeout parameters
1392  *
1393  */
1394 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1395 {
1396         size_t slot_table_size;
1397         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
1398
1399         xprt->max_reqs = xprt_tcp_slot_table_entries;
1400         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1401         xprt->slot = kzalloc(slot_table_size, GFP_KERNEL);
1402         if (xprt->slot == NULL)
1403                 return -ENOMEM;
1404
1405         if (ntohs(addr->sin_port) != 0)
1406                 xprt_set_bound(xprt);
1407         xprt->port = xs_get_random_port();
1408
1409         xprt->prot = IPPROTO_TCP;
1410         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1411         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1412
1413         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1414         xprt->bind_timeout = XS_BIND_TO;
1415         xprt->connect_timeout = XS_TCP_CONN_TO;
1416         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1417         xprt->idle_timeout = XS_IDLE_DISC_TO;
1418
1419         xprt->ops = &xs_tcp_ops;
1420
1421         if (to)
1422                 xprt->timeout = *to;
1423         else
1424                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1425
1426         xs_format_peer_addresses(xprt);
1427         dprintk("RPC:      set up transport to address %s\n",
1428                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1429
1430         return 0;
1431 }