git.oblomov.eu Git - linux-2.6/blob - 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 inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
 172 {
 173         struct kvec iov = {
 174                 .iov_base       = xdr->head[0].iov_base + base,
 175                 .iov_len        = len - base,
 176         };
 177         struct msghdr msg = {
 178                 .msg_name       = addr,
 179                 .msg_namelen    = addrlen,
 180                 .msg_flags      = XS_SENDMSG_FLAGS,
 181         };
 182
 183         if (xdr->len > len)
 184                 msg.msg_flags |= MSG_MORE;
 185
 186         if (likely(iov.iov_len))
 187                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 188         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 189 }
 190
 191 static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
 192 {
 193         struct kvec iov = {
 194                 .iov_base       = xdr->tail[0].iov_base + base,
 195                 .iov_len        = len - base,
 196         };
 197         struct msghdr msg = {
 198                 .msg_flags      = XS_SENDMSG_FLAGS,
 199         };
 200
 201         return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 202 }
 203
 204 /**
 205  * xs_sendpages - write pages directly to a socket
 206  * @sock: socket to send on
 207  * @addr: UDP only -- address of destination
 208  * @addrlen: UDP only -- length of destination address
 209  * @xdr: buffer containing this request
 210  * @base: starting position in the buffer
 211  *
 212  */
 213 static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 214 {
 215         struct page **ppage = xdr->pages;
 216         unsigned int len, pglen = xdr->page_len;
 217         int err, ret = 0;
 218
 219         if (unlikely(!sock))
 220                 return -ENOTCONN;
 221
 222         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 223
 224         len = xdr->head[0].iov_len;
 225         if (base < len || (addr != NULL && base == 0)) {
 226                 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
 227                 if (ret == 0)
 228                         ret = err;
 229                 else if (err > 0)
 230                         ret += err;
 231                 if (err != (len - base))
 232                         goto out;
 233                 base = 0;
 234         } else
 235                 base -= len;
 236
 237         if (unlikely(pglen == 0))
 238                 goto copy_tail;
 239         if (unlikely(base >= pglen)) {
 240                 base -= pglen;
 241                 goto copy_tail;
 242         }
 243         if (base || xdr->page_base) {
 244                 pglen -= base;
 245                 base += xdr->page_base;
 246                 ppage += base >> PAGE_CACHE_SHIFT;
 247                 base &= ~PAGE_CACHE_MASK;
 248         }
 249
 250         do {
 251                 int flags = XS_SENDMSG_FLAGS;
 252
 253                 len = PAGE_CACHE_SIZE;
 254                 if (base)
 255                         len -= base;
 256                 if (pglen < len)
 257                         len = pglen;
 258
 259                 if (pglen != len || xdr->tail[0].iov_len != 0)
 260                         flags |= MSG_MORE;
 261
 262                 err = kernel_sendpage(sock, *ppage, base, len, flags);
 263                 if (ret == 0)
 264                         ret = err;
 265                 else if (err > 0)
 266                         ret += err;
 267                 if (err != len)
 268                         goto out;
 269                 base = 0;
 270                 ppage++;
 271         } while ((pglen -= len) != 0);
 272 copy_tail:
 273         len = xdr->tail[0].iov_len;
 274         if (base < len) {
 275                 err = xs_send_tail(sock, xdr, base, len);
 276                 if (ret == 0)
 277                         ret = err;
 278                 else if (err > 0)
 279                         ret += err;
 280         }
 281 out:
 282         return ret;
 283 }
 284
 285 /**
 286  * xs_nospace - place task on wait queue if transmit was incomplete
 287  * @task: task to put to sleep
 288  *
 289  */
 290 static void xs_nospace(struct rpc_task *task)
 291 {
 292         struct rpc_rqst *req = task->tk_rqstp;
 293         struct rpc_xprt *xprt = req->rq_xprt;
 294
 295         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 296                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 297                         req->rq_slen);
 298
 299         if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
 300                 /* Protect against races with write_space */
 301                 spin_lock_bh(&xprt->transport_lock);
 302
 303                 /* Don't race with disconnect */
 304                 if (!xprt_connected(xprt))
 305                         task->tk_status = -ENOTCONN;
 306                 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
 307                         xprt_wait_for_buffer_space(task);
 308
 309                 spin_unlock_bh(&xprt->transport_lock);
 310         } else
 311                 /* Keep holding the socket if it is blocked */
 312                 rpc_delay(task, HZ>>4);
 313 }
 314
 315 /**
 316  * xs_udp_send_request - write an RPC request to a UDP socket
 317  * @task: address of RPC task that manages the state of an RPC request
 318  *
 319  * Return values:
 320  *        0:    The request has been sent
 321  *   EAGAIN:    The socket was blocked, please call again later to
 322  *              complete the request
 323  * ENOTCONN:    Caller needs to invoke connect logic then call again
 324  *    other:    Some other error occured, the request was not sent
 325  */
 326 static int xs_udp_send_request(struct rpc_task *task)
 327 {
 328         struct rpc_rqst *req = task->tk_rqstp;
 329         struct rpc_xprt *xprt = req->rq_xprt;
 330         struct xdr_buf *xdr = &req->rq_snd_buf;
 331         int status;
 332
 333         xs_pktdump("packet data:",
 334                                 req->rq_svec->iov_base,
 335                                 req->rq_svec->iov_len);
 336
 337         req->rq_xtime = jiffies;
 338         status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
 339                                 xprt->addrlen, xdr, req->rq_bytes_sent);
 340
 341         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 342                         xdr->len - req->rq_bytes_sent, status);
 343
 344         if (likely(status >= (int) req->rq_slen))
 345                 return 0;
 346
 347         /* Still some bytes left; set up for a retry later. */
 348         if (status > 0)
 349                 status = -EAGAIN;
 350
 351         switch (status) {
 352         case -ENETUNREACH:
 353         case -EPIPE:
 354         case -ECONNREFUSED:
 355                 /* When the server has died, an ICMP port unreachable message
 356                  * prompts ECONNREFUSED. */
 357                 break;
 358         case -EAGAIN:
 359                 xs_nospace(task);
 360                 break;
 361         default:
 362                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 363                         -status);
 364                 break;
 365         }
 366
 367         return status;
 368 }
 369
 370 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 371 {
 372         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 373         rpc_fraghdr *base = buf->head[0].iov_base;
 374         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 375 }
 376
 377 /**
 378  * xs_tcp_send_request - write an RPC request to a TCP socket
 379  * @task: address of RPC task that manages the state of an RPC request
 380  *
 381  * Return values:
 382  *        0:    The request has been sent
 383  *   EAGAIN:    The socket was blocked, please call again later to
 384  *              complete the request
 385  * ENOTCONN:    Caller needs to invoke connect logic then call again
 386  *    other:    Some other error occured, the request was not sent
 387  *
 388  * XXX: In the case of soft timeouts, should we eventually give up
 389  *      if sendmsg is not able to make progress?
 390  */
 391 static int xs_tcp_send_request(struct rpc_task *task)
 392 {
 393         struct rpc_rqst *req = task->tk_rqstp;
 394         struct rpc_xprt *xprt = req->rq_xprt;
 395         struct xdr_buf *xdr = &req->rq_snd_buf;
 396         int status, retry = 0;
 397
 398         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 399
 400         xs_pktdump("packet data:",
 401                                 req->rq_svec->iov_base,
 402                                 req->rq_svec->iov_len);
 403
 404         /* Continue transmitting the packet/record. We must be careful
 405          * to cope with writespace callbacks arriving _after_ we have
 406          * called sendmsg(). */
 407         while (1) {
 408                 req->rq_xtime = jiffies;
 409                 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
 410                                                 req->rq_bytes_sent);
 411
 412                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 413                                 xdr->len - req->rq_bytes_sent, status);
 414
 415                 if (unlikely(status < 0))
 416                         break;
 417
 418                 /* If we've sent the entire packet, immediately
 419                  * reset the count of bytes sent. */
 420                 req->rq_bytes_sent += status;
 421                 task->tk_bytes_sent += status;
 422                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 423                         req->rq_bytes_sent = 0;
 424                         return 0;
 425                 }
 426
 427                 status = -EAGAIN;
 428                 if (retry++ > XS_SENDMSG_RETRY)
 429                         break;
 430         }
 431
 432         switch (status) {
 433         case -EAGAIN:
 434                 xs_nospace(task);
 435                 break;
 436         case -ECONNREFUSED:
 437         case -ECONNRESET:
 438         case -ENOTCONN:
 439         case -EPIPE:
 440                 status = -ENOTCONN;
 441                 break;
 442         default:
 443                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 444                         -status);
 445                 xprt_disconnect(xprt);
 446                 break;
 447         }
 448
 449         return status;
 450 }
 451
 452 /**
 453  * xs_tcp_release_xprt - clean up after a tcp transmission
 454  * @xprt: transport
 455  * @task: rpc task
 456  *
 457  * This cleans up if an error causes us to abort the transmission of a request.
 458  * In this case, the socket may need to be reset in order to avoid confusing
 459  * the server.
 460  */
 461 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 462 {
 463         struct rpc_rqst *req;
 464
 465         if (task != xprt->snd_task)
 466                 return;
 467         if (task == NULL)
 468                 goto out_release;
 469         req = task->tk_rqstp;
 470         if (req->rq_bytes_sent == 0)
 471                 goto out_release;
 472         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 473                 goto out_release;
 474         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 475 out_release:
 476         xprt_release_xprt(xprt, task);
 477 }
 478
 479 /**
 480  * xs_close - close a socket
 481  * @xprt: transport
 482  *
 483  * This is used when all requests are complete; ie, no DRC state remains
 484  * on the server we want to save.
 485  */
 486 static void xs_close(struct rpc_xprt *xprt)
 487 {
 488         struct socket *sock = xprt->sock;
 489         struct sock *sk = xprt->inet;
 490
 491         if (!sk)
 492                 goto clear_close_wait;
 493
 494         dprintk("RPC:      xs_close xprt %p\n", xprt);
 495
 496         write_lock_bh(&sk->sk_callback_lock);
 497         xprt->inet = NULL;
 498         xprt->sock = NULL;
 499
 500         sk->sk_user_data = NULL;
 501         sk->sk_data_ready = xprt->old_data_ready;
 502         sk->sk_state_change = xprt->old_state_change;
 503         sk->sk_write_space = xprt->old_write_space;
 504         write_unlock_bh(&sk->sk_callback_lock);
 505
 506         sk->sk_no_check = 0;
 507
 508         sock_release(sock);
 509 clear_close_wait:
 510         smp_mb__before_clear_bit();
 511         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 512         smp_mb__after_clear_bit();
 513 }
 514
 515 /**
 516  * xs_destroy - prepare to shutdown a transport
 517  * @xprt: doomed transport
 518  *
 519  */
 520 static void xs_destroy(struct rpc_xprt *xprt)
 521 {
 522         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 523
 524         cancel_delayed_work(&xprt->connect_worker);
 525         flush_scheduled_work();
 526
 527         xprt_disconnect(xprt);
 528         xs_close(xprt);
 529         xs_free_peer_addresses(xprt);
 530         kfree(xprt->slot);
 531 }
 532
 533 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 534 {
 535         return (struct rpc_xprt *) sk->sk_user_data;
 536 }
 537
 538 /**
 539  * xs_udp_data_ready - "data ready" callback for UDP sockets
 540  * @sk: socket with data to read
 541  * @len: how much data to read
 542  *
 543  */
 544 static void xs_udp_data_ready(struct sock *sk, int len)
 545 {
 546         struct rpc_task *task;
 547         struct rpc_xprt *xprt;
 548         struct rpc_rqst *rovr;
 549         struct sk_buff *skb;
 550         int err, repsize, copied;
 551         u32 _xid;
 552         __be32 *xp;
 553
 554         read_lock(&sk->sk_callback_lock);
 555         dprintk("RPC:      xs_udp_data_ready...\n");
 556         if (!(xprt = xprt_from_sock(sk)))
 557                 goto out;
 558
 559         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 560                 goto out;
 561
 562         if (xprt->shutdown)
 563                 goto dropit;
 564
 565         repsize = skb->len - sizeof(struct udphdr);
 566         if (repsize < 4) {
 567                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 568                 goto dropit;
 569         }
 570
 571         /* Copy the XID from the skb... */
 572         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 573                                 sizeof(_xid), &_xid);
 574         if (xp == NULL)
 575                 goto dropit;
 576
 577         /* Look up and lock the request corresponding to the given XID */
 578         spin_lock(&xprt->transport_lock);
 579         rovr = xprt_lookup_rqst(xprt, *xp);
 580         if (!rovr)
 581                 goto out_unlock;
 582         task = rovr->rq_task;
 583
 584         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 585                 copied = repsize;
 586
 587         /* Suck it into the iovec, verify checksum if not done by hw. */
 588         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 589                 goto out_unlock;
 590
 591         /* Something worked... */
 592         dst_confirm(skb->dst);
 593
 594         xprt_adjust_cwnd(task, copied);
 595         xprt_update_rtt(task);
 596         xprt_complete_rqst(task, copied);
 597
 598  out_unlock:
 599         spin_unlock(&xprt->transport_lock);
 600  dropit:
 601         skb_free_datagram(sk, skb);
 602  out:
 603         read_unlock(&sk->sk_callback_lock);
 604 }
 605
 606 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 607 {
 608         if (len > desc->count)
 609                 len = desc->count;
 610         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 611                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 612                                 len, desc->count);
 613                 return 0;
 614         }
 615         desc->offset += len;
 616         desc->count -= len;
 617         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 618                         len, desc->count);
 619         return len;
 620 }
 621
 622 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 623 {
 624         size_t len, used;
 625         char *p;
 626
 627         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 628         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 629         used = xs_tcp_copy_data(desc, p, len);
 630         xprt->tcp_offset += used;
 631         if (used != len)
 632                 return;
 633
 634         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 635         if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 636                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 637         else
 638                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 639         xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 640
 641         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 642         xprt->tcp_offset = 0;
 643
 644         /* Sanity check of the record length */
 645         if (unlikely(xprt->tcp_reclen < 4)) {
 646                 dprintk("RPC:      invalid TCP record fragment length\n");
 647                 xprt_disconnect(xprt);
 648                 return;
 649         }
 650         dprintk("RPC:      reading TCP record fragment of length %d\n",
 651                         xprt->tcp_reclen);
 652 }
 653
 654 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 655 {
 656         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 657                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 658         if (xprt->tcp_offset == xprt->tcp_reclen) {
 659                 xprt->tcp_flags |= XPRT_COPY_RECM;
 660                 xprt->tcp_offset = 0;
 661                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 662                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 663                         xprt->tcp_flags |= XPRT_COPY_XID;
 664                         xprt->tcp_copied = 0;
 665                 }
 666         }
 667 }
 668
 669 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 670 {
 671         size_t len, used;
 672         char *p;
 673
 674         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 675         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 676         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 677         used = xs_tcp_copy_data(desc, p, len);
 678         xprt->tcp_offset += used;
 679         if (used != len)
 680                 return;
 681         xprt->tcp_flags &= ~XPRT_COPY_XID;
 682         xprt->tcp_flags |= XPRT_COPY_DATA;
 683         xprt->tcp_copied = 4;
 684         dprintk("RPC:      reading reply for XID %08x\n",
 685                                                 ntohl(xprt->tcp_xid));
 686         xs_tcp_check_recm(xprt);
 687 }
 688
 689 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 690 {
 691         struct rpc_rqst *req;
 692         struct xdr_buf *rcvbuf;
 693         size_t len;
 694         ssize_t r;
 695
 696         /* Find and lock the request corresponding to this xid */
 697         spin_lock(&xprt->transport_lock);
 698         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 699         if (!req) {
 700                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 701                 dprintk("RPC:      XID %08x request not found!\n",
 702                                 ntohl(xprt->tcp_xid));
 703                 spin_unlock(&xprt->transport_lock);
 704                 return;
 705         }
 706
 707         rcvbuf = &req->rq_private_buf;
 708         len = desc->count;
 709         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 710                 skb_reader_t my_desc;
 711
 712                 len = xprt->tcp_reclen - xprt->tcp_offset;
 713                 memcpy(&my_desc, desc, sizeof(my_desc));
 714                 my_desc.count = len;
 715                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 716                                           &my_desc, xs_tcp_copy_data);
 717                 desc->count -= r;
 718                 desc->offset += r;
 719         } else
 720                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 721                                           desc, xs_tcp_copy_data);
 722
 723         if (r > 0) {
 724                 xprt->tcp_copied += r;
 725                 xprt->tcp_offset += r;
 726         }
 727         if (r != len) {
 728                 /* Error when copying to the receive buffer,
 729                  * usually because we weren't able to allocate
 730                  * additional buffer pages. All we can do now
 731                  * is turn off XPRT_COPY_DATA, so the request
 732                  * will not receive any additional updates,
 733                  * and time out.
 734                  * Any remaining data from this record will
 735                  * be discarded.
 736                  */
 737                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 738                 dprintk("RPC:      XID %08x truncated request\n",
 739                                 ntohl(xprt->tcp_xid));
 740                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 741                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 742                 goto out;
 743         }
 744
 745         dprintk("RPC:      XID %08x read %Zd bytes\n",
 746                         ntohl(xprt->tcp_xid), r);
 747         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 748                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 749
 750         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 751                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 752         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 753                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 754                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 755         }
 756
 757 out:
 758         if (!(xprt->tcp_flags & XPRT_COPY_DATA))
 759                 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
 760         spin_unlock(&xprt->transport_lock);
 761         xs_tcp_check_recm(xprt);
 762 }
 763
 764 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 765 {
 766         size_t len;
 767
 768         len = xprt->tcp_reclen - xprt->tcp_offset;
 769         if (len > desc->count)
 770                 len = desc->count;
 771         desc->count -= len;
 772         desc->offset += len;
 773         xprt->tcp_offset += len;
 774         dprintk("RPC:      discarded %Zu bytes\n", len);
 775         xs_tcp_check_recm(xprt);
 776 }
 777
 778 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 779 {
 780         struct rpc_xprt *xprt = rd_desc->arg.data;
 781         skb_reader_t desc = {
 782                 .skb    = skb,
 783                 .offset = offset,
 784                 .count  = len,
 785                 .csum   = 0
 786         };
 787
 788         dprintk("RPC:      xs_tcp_data_recv started\n");
 789         do {
 790                 /* Read in a new fragment marker if necessary */
 791                 /* Can we ever really expect to get completely empty fragments? */
 792                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 793                         xs_tcp_read_fraghdr(xprt, &desc);
 794                         continue;
 795                 }
 796                 /* Read in the xid if necessary */
 797                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 798                         xs_tcp_read_xid(xprt, &desc);
 799                         continue;
 800                 }
 801                 /* Read in the request data */
 802                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 803                         xs_tcp_read_request(xprt, &desc);
 804                         continue;
 805                 }
 806                 /* Skip over any trailing bytes on short reads */
 807                 xs_tcp_read_discard(xprt, &desc);
 808         } while (desc.count);
 809         dprintk("RPC:      xs_tcp_data_recv done\n");
 810         return len - desc.count;
 811 }
 812
 813 /**
 814  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 815  * @sk: socket with data to read
 816  * @bytes: how much data to read
 817  *
 818  */
 819 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 820 {
 821         struct rpc_xprt *xprt;
 822         read_descriptor_t rd_desc;
 823
 824         read_lock(&sk->sk_callback_lock);
 825         dprintk("RPC:      xs_tcp_data_ready...\n");
 826         if (!(xprt = xprt_from_sock(sk)))
 827                 goto out;
 828         if (xprt->shutdown)
 829                 goto out;
 830
 831         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 832         rd_desc.arg.data = xprt;
 833         rd_desc.count = 65536;
 834         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 835 out:
 836         read_unlock(&sk->sk_callback_lock);
 837 }
 838
 839 /**
 840  * xs_tcp_state_change - callback to handle TCP socket state changes
 841  * @sk: socket whose state has changed
 842  *
 843  */
 844 static void xs_tcp_state_change(struct sock *sk)
 845 {
 846         struct rpc_xprt *xprt;
 847
 848         read_lock(&sk->sk_callback_lock);
 849         if (!(xprt = xprt_from_sock(sk)))
 850                 goto out;
 851         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 852         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 853                                 sk->sk_state, xprt_connected(xprt),
 854                                 sock_flag(sk, SOCK_DEAD),
 855                                 sock_flag(sk, SOCK_ZAPPED));
 856
 857         switch (sk->sk_state) {
 858         case TCP_ESTABLISHED:
 859                 spin_lock_bh(&xprt->transport_lock);
 860                 if (!xprt_test_and_set_connected(xprt)) {
 861                         /* Reset TCP record info */
 862                         xprt->tcp_offset = 0;
 863                         xprt->tcp_reclen = 0;
 864                         xprt->tcp_copied = 0;
 865                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 866                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 867                         xprt_wake_pending_tasks(xprt, 0);
 868                 }
 869                 spin_unlock_bh(&xprt->transport_lock);
 870                 break;
 871         case TCP_SYN_SENT:
 872         case TCP_SYN_RECV:
 873                 break;
 874         case TCP_CLOSE_WAIT:
 875                 /* Try to schedule an autoclose RPC calls */
 876                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 877                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 878                         schedule_work(&xprt->task_cleanup);
 879         default:
 880                 xprt_disconnect(xprt);
 881         }
 882  out:
 883         read_unlock(&sk->sk_callback_lock);
 884 }
 885
 886 /**
 887  * xs_udp_write_space - callback invoked when socket buffer space
 888  *                             becomes available
 889  * @sk: socket whose state has changed
 890  *
 891  * Called when more output buffer space is available for this socket.
 892  * We try not to wake our writers until they can make "significant"
 893  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 894  * with a bunch of small requests.
 895  */
 896 static void xs_udp_write_space(struct sock *sk)
 897 {
 898         read_lock(&sk->sk_callback_lock);
 899
 900         /* from net/core/sock.c:sock_def_write_space */
 901         if (sock_writeable(sk)) {
 902                 struct socket *sock;
 903                 struct rpc_xprt *xprt;
 904
 905                 if (unlikely(!(sock = sk->sk_socket)))
 906                         goto out;
 907                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 908                         goto out;
 909                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 910                         goto out;
 911
 912                 xprt_write_space(xprt);
 913         }
 914
 915  out:
 916         read_unlock(&sk->sk_callback_lock);
 917 }
 918
 919 /**
 920  * xs_tcp_write_space - callback invoked when socket buffer space
 921  *                             becomes available
 922  * @sk: socket whose state has changed
 923  *
 924  * Called when more output buffer space is available for this socket.
 925  * We try not to wake our writers until they can make "significant"
 926  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 927  * with a bunch of small requests.
 928  */
 929 static void xs_tcp_write_space(struct sock *sk)
 930 {
 931         read_lock(&sk->sk_callback_lock);
 932
 933         /* from net/core/stream.c:sk_stream_write_space */
 934         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 935                 struct socket *sock;
 936                 struct rpc_xprt *xprt;
 937
 938                 if (unlikely(!(sock = sk->sk_socket)))
 939                         goto out;
 940                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 941                         goto out;
 942                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 943                         goto out;
 944
 945                 xprt_write_space(xprt);
 946         }
 947
 948  out:
 949         read_unlock(&sk->sk_callback_lock);
 950 }
 951
 952 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 953 {
 954         struct sock *sk = xprt->inet;
 955
 956         if (xprt->rcvsize) {
 957                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 958                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 959         }
 960         if (xprt->sndsize) {
 961                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 962                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 963                 sk->sk_write_space(sk);
 964         }
 965 }
 966
 967 /**
 968  * xs_udp_set_buffer_size - set send and receive limits
 969  * @xprt: generic transport
 970  * @sndsize: requested size of send buffer, in bytes
 971  * @rcvsize: requested size of receive buffer, in bytes
 972  *
 973  * Set socket send and receive buffer size limits.
 974  */
 975 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
 976 {
 977         xprt->sndsize = 0;
 978         if (sndsize)
 979                 xprt->sndsize = sndsize + 1024;
 980         xprt->rcvsize = 0;
 981         if (rcvsize)
 982                 xprt->rcvsize = rcvsize + 1024;
 983
 984         xs_udp_do_set_buffer_size(xprt);
 985 }
 986
 987 /**
 988  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
 989  * @task: task that timed out
 990  *
 991  * Adjust the congestion window after a retransmit timeout has occurred.
 992  */
 993 static void xs_udp_timer(struct rpc_task *task)
 994 {
 995         xprt_adjust_cwnd(task, -ETIMEDOUT);
 996 }
 997
 998 static unsigned short xs_get_random_port(void)
 999 {
1000         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1001         unsigned short rand = (unsigned short) net_random() % range;
1002         return rand + xprt_min_resvport;
1003 }
1004
1005 /**
1006  * xs_print_peer_address - format an IPv4 address for printing
1007  * @xprt: generic transport
1008  * @format: flags field indicating which parts of the address to render
1009  */
1010 static char *xs_print_peer_address(struct rpc_xprt *xprt, enum rpc_display_format_t format)
1011 {
1012         if (xprt->address_strings[format] != NULL)
1013                 return xprt->address_strings[format];
1014         else
1015                 return "unprintable";
1016 }
1017
1018 /**
1019  * xs_set_port - reset the port number in the remote endpoint address
1020  * @xprt: generic transport
1021  * @port: new port number
1022  *
1023  */
1024 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1025 {
1026         struct sockaddr_in *sap = (struct sockaddr_in *) &xprt->addr;
1027
1028         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
1029
1030         sap->sin_port = htons(port);
1031 }
1032
1033 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
1034 {
1035         struct sockaddr_in myaddr = {
1036                 .sin_family = AF_INET,
1037         };
1038         int err;
1039         unsigned short port = xprt->port;
1040
1041         do {
1042                 myaddr.sin_port = htons(port);
1043                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1044                                                 sizeof(myaddr));
1045                 if (err == 0) {
1046                         xprt->port = port;
1047                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
1048                                         port);
1049                         return 0;
1050                 }
1051                 if (port <= xprt_min_resvport)
1052                         port = xprt_max_resvport;
1053                 else
1054                         port--;
1055         } while (err == -EADDRINUSE && port != xprt->port);
1056
1057         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
1058         return err;
1059 }
1060
1061 /**
1062  * xs_udp_connect_worker - set up a UDP socket
1063  * @args: RPC transport to connect
1064  *
1065  * Invoked by a work queue tasklet.
1066  */
1067 static void xs_udp_connect_worker(void *args)
1068 {
1069         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
1070         struct socket *sock = xprt->sock;
1071         int err, status = -EIO;
1072
1073         if (xprt->shutdown || !xprt_bound(xprt))
1074                 goto out;
1075
1076         /* Start by resetting any existing state */
1077         xs_close(xprt);
1078
1079         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1080                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1081                 goto out;
1082         }
1083
1084         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1085                 sock_release(sock);
1086                 goto out;
1087         }
1088
1089         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1090                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1091
1092         if (!xprt->inet) {
1093                 struct sock *sk = sock->sk;
1094
1095                 write_lock_bh(&sk->sk_callback_lock);
1096
1097                 sk->sk_user_data = xprt;
1098                 xprt->old_data_ready = sk->sk_data_ready;
1099                 xprt->old_state_change = sk->sk_state_change;
1100                 xprt->old_write_space = sk->sk_write_space;
1101                 sk->sk_data_ready = xs_udp_data_ready;
1102                 sk->sk_write_space = xs_udp_write_space;
1103                 sk->sk_no_check = UDP_CSUM_NORCV;
1104                 sk->sk_allocation = GFP_ATOMIC;
1105
1106                 xprt_set_connected(xprt);
1107
1108                 /* Reset to new socket */
1109                 xprt->sock = sock;
1110                 xprt->inet = sk;
1111
1112                 write_unlock_bh(&sk->sk_callback_lock);
1113         }
1114         xs_udp_do_set_buffer_size(xprt);
1115         status = 0;
1116 out:
1117         xprt_wake_pending_tasks(xprt, status);
1118         xprt_clear_connecting(xprt);
1119 }
1120
1121 /*
1122  * We need to preserve the port number so the reply cache on the server can
1123  * find our cached RPC replies when we get around to reconnecting.
1124  */
1125 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1126 {
1127         int result;
1128         struct socket *sock = xprt->sock;
1129         struct sockaddr any;
1130
1131         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1132
1133         /*
1134          * Disconnect the transport socket by doing a connect operation
1135          * with AF_UNSPEC.  This should return immediately...
1136          */
1137         memset(&any, 0, sizeof(any));
1138         any.sa_family = AF_UNSPEC;
1139         result = kernel_connect(sock, &any, sizeof(any), 0);
1140         if (result)
1141                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1142                                 result);
1143 }
1144
1145 /**
1146  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1147  * @args: RPC transport to connect
1148  *
1149  * Invoked by a work queue tasklet.
1150  */
1151 static void xs_tcp_connect_worker(void *args)
1152 {
1153         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1154         struct socket *sock = xprt->sock;
1155         int err, status = -EIO;
1156
1157         if (xprt->shutdown || !xprt_bound(xprt))
1158                 goto out;
1159
1160         if (!xprt->sock) {
1161                 /* start from scratch */
1162                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1163                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1164                         goto out;
1165                 }
1166
1167                 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1168                         sock_release(sock);
1169                         goto out;
1170                 }
1171         } else
1172                 /* "close" the socket, preserving the local port */
1173                 xs_tcp_reuse_connection(xprt);
1174
1175         dprintk("RPC:      worker connecting xprt %p to address: %s\n",
1176                         xprt, xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1177
1178         if (!xprt->inet) {
1179                 struct sock *sk = sock->sk;
1180
1181                 write_lock_bh(&sk->sk_callback_lock);
1182
1183                 sk->sk_user_data = xprt;
1184                 xprt->old_data_ready = sk->sk_data_ready;
1185                 xprt->old_state_change = sk->sk_state_change;
1186                 xprt->old_write_space = sk->sk_write_space;
1187                 sk->sk_data_ready = xs_tcp_data_ready;
1188                 sk->sk_state_change = xs_tcp_state_change;
1189                 sk->sk_write_space = xs_tcp_write_space;
1190                 sk->sk_allocation = GFP_ATOMIC;
1191
1192                 /* socket options */
1193                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1194                 sock_reset_flag(sk, SOCK_LINGER);
1195                 tcp_sk(sk)->linger2 = 0;
1196                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1197
1198                 xprt_clear_connected(xprt);
1199
1200                 /* Reset to new socket */
1201                 xprt->sock = sock;
1202                 xprt->inet = sk;
1203
1204                 write_unlock_bh(&sk->sk_callback_lock);
1205         }
1206
1207         /* Tell the socket layer to start connecting... */
1208         xprt->stat.connect_count++;
1209         xprt->stat.connect_start = jiffies;
1210         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1211                         xprt->addrlen, O_NONBLOCK);
1212         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1213                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1214         if (status < 0) {
1215                 switch (status) {
1216                         case -EINPROGRESS:
1217                         case -EALREADY:
1218                                 goto out_clear;
1219                         case -ECONNREFUSED:
1220                         case -ECONNRESET:
1221                                 /* retry with existing socket, after a delay */
1222                                 break;
1223                         default:
1224                                 /* get rid of existing socket, and retry */
1225                                 xs_close(xprt);
1226                                 break;
1227                 }
1228         }
1229 out:
1230         xprt_wake_pending_tasks(xprt, status);
1231 out_clear:
1232         xprt_clear_connecting(xprt);
1233 }
1234
1235 /**
1236  * xs_connect - connect a socket to a remote endpoint
1237  * @task: address of RPC task that manages state of connect request
1238  *
1239  * TCP: If the remote end dropped the connection, delay reconnecting.
1240  *
1241  * UDP socket connects are synchronous, but we use a work queue anyway
1242  * to guarantee that even unprivileged user processes can set up a
1243  * socket on a privileged port.
1244  *
1245  * If a UDP socket connect fails, the delay behavior here prevents
1246  * retry floods (hard mounts).
1247  */
1248 static void xs_connect(struct rpc_task *task)
1249 {
1250         struct rpc_xprt *xprt = task->tk_xprt;
1251
1252         if (xprt_test_and_set_connecting(xprt))
1253                 return;
1254
1255         if (xprt->sock != NULL) {
1256                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1257                                 xprt, xprt->reestablish_timeout / HZ);
1258                 schedule_delayed_work(&xprt->connect_worker,
1259                                         xprt->reestablish_timeout);
1260                 xprt->reestablish_timeout <<= 1;
1261                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1262                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1263         } else {
1264                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1265                 schedule_work(&xprt->connect_worker);
1266
1267                 /* flush_scheduled_work can sleep... */
1268                 if (!RPC_IS_ASYNC(task))
1269                         flush_scheduled_work();
1270         }
1271 }
1272
1273 /**
1274  * xs_udp_print_stats - display UDP socket-specifc stats
1275  * @xprt: rpc_xprt struct containing statistics
1276  * @seq: output file
1277  *
1278  */
1279 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1280 {
1281         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1282                         xprt->port,
1283                         xprt->stat.bind_count,
1284                         xprt->stat.sends,
1285                         xprt->stat.recvs,
1286                         xprt->stat.bad_xids,
1287                         xprt->stat.req_u,
1288                         xprt->stat.bklog_u);
1289 }
1290
1291 /**
1292  * xs_tcp_print_stats - display TCP socket-specifc stats
1293  * @xprt: rpc_xprt struct containing statistics
1294  * @seq: output file
1295  *
1296  */
1297 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1298 {
1299         long idle_time = 0;
1300
1301         if (xprt_connected(xprt))
1302                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1303
1304         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1305                         xprt->port,
1306                         xprt->stat.bind_count,
1307                         xprt->stat.connect_count,
1308                         xprt->stat.connect_time,
1309                         idle_time,
1310                         xprt->stat.sends,
1311                         xprt->stat.recvs,
1312                         xprt->stat.bad_xids,
1313                         xprt->stat.req_u,
1314                         xprt->stat.bklog_u);
1315 }
1316
1317 static struct rpc_xprt_ops xs_udp_ops = {
1318         .set_buffer_size        = xs_udp_set_buffer_size,
1319         .print_addr             = xs_print_peer_address,
1320         .reserve_xprt           = xprt_reserve_xprt_cong,
1321         .release_xprt           = xprt_release_xprt_cong,
1322         .rpcbind                = rpc_getport,
1323         .set_port               = xs_set_port,
1324         .connect                = xs_connect,
1325         .buf_alloc              = rpc_malloc,
1326         .buf_free               = rpc_free,
1327         .send_request           = xs_udp_send_request,
1328         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1329         .timer                  = xs_udp_timer,
1330         .release_request        = xprt_release_rqst_cong,
1331         .close                  = xs_close,
1332         .destroy                = xs_destroy,
1333         .print_stats            = xs_udp_print_stats,
1334 };
1335
1336 static struct rpc_xprt_ops xs_tcp_ops = {
1337         .print_addr             = xs_print_peer_address,
1338         .reserve_xprt           = xprt_reserve_xprt,
1339         .release_xprt           = xs_tcp_release_xprt,
1340         .rpcbind                = rpc_getport,
1341         .set_port               = xs_set_port,
1342         .connect                = xs_connect,
1343         .buf_alloc              = rpc_malloc,
1344         .buf_free               = rpc_free,
1345         .send_request           = xs_tcp_send_request,
1346         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1347         .close                  = xs_close,
1348         .destroy                = xs_destroy,
1349         .print_stats            = xs_tcp_print_stats,
1350 };
1351
1352 /**
1353  * xs_setup_udp - Set up transport to use a UDP socket
1354  * @xprt: transport to set up
1355  * @to:   timeout parameters
1356  *
1357  */
1358 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1359 {
1360         size_t slot_table_size;
1361         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
1362
1363         xprt->max_reqs = xprt_udp_slot_table_entries;
1364         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1365         xprt->slot = kzalloc(slot_table_size, GFP_KERNEL);
1366         if (xprt->slot == NULL)
1367                 return -ENOMEM;
1368
1369         if (ntohs(addr->sin_port != 0))
1370                 xprt_set_bound(xprt);
1371         xprt->port = xs_get_random_port();
1372
1373         xprt->prot = IPPROTO_UDP;
1374         xprt->tsh_size = 0;
1375         /* XXX: header size can vary due to auth type, IPv6, etc. */
1376         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1377
1378         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1379         xprt->bind_timeout = XS_BIND_TO;
1380         xprt->connect_timeout = XS_UDP_CONN_TO;
1381         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1382         xprt->idle_timeout = XS_IDLE_DISC_TO;
1383
1384         xprt->ops = &xs_udp_ops;
1385
1386         if (to)
1387                 xprt->timeout = *to;
1388         else
1389                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1390
1391         xs_format_peer_addresses(xprt);
1392         dprintk("RPC:      set up transport to address %s\n",
1393                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1394
1395         return 0;
1396 }
1397
1398 /**
1399  * xs_setup_tcp - Set up transport to use a TCP socket
1400  * @xprt: transport to set up
1401  * @to: timeout parameters
1402  *
1403  */
1404 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1405 {
1406         size_t slot_table_size;
1407         struct sockaddr_in *addr = (struct sockaddr_in *) &xprt->addr;
1408
1409         xprt->max_reqs = xprt_tcp_slot_table_entries;
1410         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1411         xprt->slot = kzalloc(slot_table_size, GFP_KERNEL);
1412         if (xprt->slot == NULL)
1413                 return -ENOMEM;
1414
1415         if (ntohs(addr->sin_port) != 0)
1416                 xprt_set_bound(xprt);
1417         xprt->port = xs_get_random_port();
1418
1419         xprt->prot = IPPROTO_TCP;
1420         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1421         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1422
1423         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1424         xprt->bind_timeout = XS_BIND_TO;
1425         xprt->connect_timeout = XS_TCP_CONN_TO;
1426         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1427         xprt->idle_timeout = XS_IDLE_DISC_TO;
1428
1429         xprt->ops = &xs_tcp_ops;
1430
1431         if (to)
1432                 xprt->timeout = *to;
1433         else
1434                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1435
1436         xs_format_peer_addresses(xprt);
1437         dprintk("RPC:      set up transport to address %s\n",
1438                         xs_print_peer_address(xprt, RPC_DISPLAY_ALL));
1439
1440         return 0;
1441 }