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 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 129
 130 static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
 131 {
 132         struct kvec iov = {
 133                 .iov_base       = xdr->head[0].iov_base + base,
 134                 .iov_len        = len - base,
 135         };
 136         struct msghdr msg = {
 137                 .msg_name       = addr,
 138                 .msg_namelen    = addrlen,
 139                 .msg_flags      = XS_SENDMSG_FLAGS,
 140         };
 141
 142         if (xdr->len > len)
 143                 msg.msg_flags |= MSG_MORE;
 144
 145         if (likely(iov.iov_len))
 146                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 147         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 148 }
 149
 150 static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
 151 {
 152         struct kvec iov = {
 153                 .iov_base       = xdr->tail[0].iov_base + base,
 154                 .iov_len        = len - base,
 155         };
 156         struct msghdr msg = {
 157                 .msg_flags      = XS_SENDMSG_FLAGS,
 158         };
 159
 160         return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 161 }
 162
 163 /**
 164  * xs_sendpages - write pages directly to a socket
 165  * @sock: socket to send on
 166  * @addr: UDP only -- address of destination
 167  * @addrlen: UDP only -- length of destination address
 168  * @xdr: buffer containing this request
 169  * @base: starting position in the buffer
 170  *
 171  */
 172 static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 173 {
 174         struct page **ppage = xdr->pages;
 175         unsigned int len, pglen = xdr->page_len;
 176         int err, ret = 0;
 177
 178         if (unlikely(!sock))
 179                 return -ENOTCONN;
 180
 181         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 182
 183         len = xdr->head[0].iov_len;
 184         if (base < len || (addr != NULL && base == 0)) {
 185                 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
 186                 if (ret == 0)
 187                         ret = err;
 188                 else if (err > 0)
 189                         ret += err;
 190                 if (err != (len - base))
 191                         goto out;
 192                 base = 0;
 193         } else
 194                 base -= len;
 195
 196         if (unlikely(pglen == 0))
 197                 goto copy_tail;
 198         if (unlikely(base >= pglen)) {
 199                 base -= pglen;
 200                 goto copy_tail;
 201         }
 202         if (base || xdr->page_base) {
 203                 pglen -= base;
 204                 base += xdr->page_base;
 205                 ppage += base >> PAGE_CACHE_SHIFT;
 206                 base &= ~PAGE_CACHE_MASK;
 207         }
 208
 209         do {
 210                 int flags = XS_SENDMSG_FLAGS;
 211
 212                 len = PAGE_CACHE_SIZE;
 213                 if (base)
 214                         len -= base;
 215                 if (pglen < len)
 216                         len = pglen;
 217
 218                 if (pglen != len || xdr->tail[0].iov_len != 0)
 219                         flags |= MSG_MORE;
 220
 221                 err = kernel_sendpage(sock, *ppage, base, len, flags);
 222                 if (ret == 0)
 223                         ret = err;
 224                 else if (err > 0)
 225                         ret += err;
 226                 if (err != len)
 227                         goto out;
 228                 base = 0;
 229                 ppage++;
 230         } while ((pglen -= len) != 0);
 231 copy_tail:
 232         len = xdr->tail[0].iov_len;
 233         if (base < len) {
 234                 err = xs_send_tail(sock, xdr, base, len);
 235                 if (ret == 0)
 236                         ret = err;
 237                 else if (err > 0)
 238                         ret += err;
 239         }
 240 out:
 241         return ret;
 242 }
 243
 244 /**
 245  * xs_nospace - place task on wait queue if transmit was incomplete
 246  * @task: task to put to sleep
 247  *
 248  */
 249 static void xs_nospace(struct rpc_task *task)
 250 {
 251         struct rpc_rqst *req = task->tk_rqstp;
 252         struct rpc_xprt *xprt = req->rq_xprt;
 253
 254         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 255                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 256                         req->rq_slen);
 257
 258         if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
 259                 /* Protect against races with write_space */
 260                 spin_lock_bh(&xprt->transport_lock);
 261
 262                 /* Don't race with disconnect */
 263                 if (!xprt_connected(xprt))
 264                         task->tk_status = -ENOTCONN;
 265                 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
 266                         xprt_wait_for_buffer_space(task);
 267
 268                 spin_unlock_bh(&xprt->transport_lock);
 269         } else
 270                 /* Keep holding the socket if it is blocked */
 271                 rpc_delay(task, HZ>>4);
 272 }
 273
 274 /**
 275  * xs_udp_send_request - write an RPC request to a UDP socket
 276  * @task: address of RPC task that manages the state of an RPC request
 277  *
 278  * Return values:
 279  *        0:    The request has been sent
 280  *   EAGAIN:    The socket was blocked, please call again later to
 281  *              complete the request
 282  * ENOTCONN:    Caller needs to invoke connect logic then call again
 283  *    other:    Some other error occured, the request was not sent
 284  */
 285 static int xs_udp_send_request(struct rpc_task *task)
 286 {
 287         struct rpc_rqst *req = task->tk_rqstp;
 288         struct rpc_xprt *xprt = req->rq_xprt;
 289         struct xdr_buf *xdr = &req->rq_snd_buf;
 290         int status;
 291
 292         xs_pktdump("packet data:",
 293                                 req->rq_svec->iov_base,
 294                                 req->rq_svec->iov_len);
 295
 296         req->rq_xtime = jiffies;
 297         status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
 298                                 sizeof(xprt->addr), xdr, req->rq_bytes_sent);
 299
 300         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 301                         xdr->len - req->rq_bytes_sent, status);
 302
 303         if (likely(status >= (int) req->rq_slen))
 304                 return 0;
 305
 306         /* Still some bytes left; set up for a retry later. */
 307         if (status > 0)
 308                 status = -EAGAIN;
 309
 310         switch (status) {
 311         case -ENETUNREACH:
 312         case -EPIPE:
 313         case -ECONNREFUSED:
 314                 /* When the server has died, an ICMP port unreachable message
 315                  * prompts ECONNREFUSED. */
 316                 break;
 317         case -EAGAIN:
 318                 xs_nospace(task);
 319                 break;
 320         default:
 321                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 322                         -status);
 323                 break;
 324         }
 325
 326         return status;
 327 }
 328
 329 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 330 {
 331         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 332         rpc_fraghdr *base = buf->head[0].iov_base;
 333         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 334 }
 335
 336 /**
 337  * xs_tcp_send_request - write an RPC request to a TCP socket
 338  * @task: address of RPC task that manages the state of an RPC request
 339  *
 340  * Return values:
 341  *        0:    The request has been sent
 342  *   EAGAIN:    The socket was blocked, please call again later to
 343  *              complete the request
 344  * ENOTCONN:    Caller needs to invoke connect logic then call again
 345  *    other:    Some other error occured, the request was not sent
 346  *
 347  * XXX: In the case of soft timeouts, should we eventually give up
 348  *      if sendmsg is not able to make progress?
 349  */
 350 static int xs_tcp_send_request(struct rpc_task *task)
 351 {
 352         struct rpc_rqst *req = task->tk_rqstp;
 353         struct rpc_xprt *xprt = req->rq_xprt;
 354         struct xdr_buf *xdr = &req->rq_snd_buf;
 355         int status, retry = 0;
 356
 357         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 358
 359         xs_pktdump("packet data:",
 360                                 req->rq_svec->iov_base,
 361                                 req->rq_svec->iov_len);
 362
 363         /* Continue transmitting the packet/record. We must be careful
 364          * to cope with writespace callbacks arriving _after_ we have
 365          * called sendmsg(). */
 366         while (1) {
 367                 req->rq_xtime = jiffies;
 368                 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
 369                                                 req->rq_bytes_sent);
 370
 371                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 372                                 xdr->len - req->rq_bytes_sent, status);
 373
 374                 if (unlikely(status < 0))
 375                         break;
 376
 377                 /* If we've sent the entire packet, immediately
 378                  * reset the count of bytes sent. */
 379                 req->rq_bytes_sent += status;
 380                 task->tk_bytes_sent += status;
 381                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 382                         req->rq_bytes_sent = 0;
 383                         return 0;
 384                 }
 385
 386                 status = -EAGAIN;
 387                 if (retry++ > XS_SENDMSG_RETRY)
 388                         break;
 389         }
 390
 391         switch (status) {
 392         case -EAGAIN:
 393                 xs_nospace(task);
 394                 break;
 395         case -ECONNREFUSED:
 396         case -ECONNRESET:
 397         case -ENOTCONN:
 398         case -EPIPE:
 399                 status = -ENOTCONN;
 400                 break;
 401         default:
 402                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 403                         -status);
 404                 xprt_disconnect(xprt);
 405                 break;
 406         }
 407
 408         return status;
 409 }
 410
 411 /**
 412  * xs_tcp_release_xprt - clean up after a tcp transmission
 413  * @xprt: transport
 414  * @task: rpc task
 415  *
 416  * This cleans up if an error causes us to abort the transmission of a request.
 417  * In this case, the socket may need to be reset in order to avoid confusing
 418  * the server.
 419  */
 420 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 421 {
 422         struct rpc_rqst *req;
 423
 424         if (task != xprt->snd_task)
 425                 return;
 426         if (task == NULL)
 427                 goto out_release;
 428         req = task->tk_rqstp;
 429         if (req->rq_bytes_sent == 0)
 430                 goto out_release;
 431         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 432                 goto out_release;
 433         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 434 out_release:
 435         xprt_release_xprt(xprt, task);
 436 }
 437
 438 /**
 439  * xs_close - close a socket
 440  * @xprt: transport
 441  *
 442  * This is used when all requests are complete; ie, no DRC state remains
 443  * on the server we want to save.
 444  */
 445 static void xs_close(struct rpc_xprt *xprt)
 446 {
 447         struct socket *sock = xprt->sock;
 448         struct sock *sk = xprt->inet;
 449
 450         if (!sk)
 451                 goto clear_close_wait;
 452
 453         dprintk("RPC:      xs_close xprt %p\n", xprt);
 454
 455         write_lock_bh(&sk->sk_callback_lock);
 456         xprt->inet = NULL;
 457         xprt->sock = NULL;
 458
 459         sk->sk_user_data = NULL;
 460         sk->sk_data_ready = xprt->old_data_ready;
 461         sk->sk_state_change = xprt->old_state_change;
 462         sk->sk_write_space = xprt->old_write_space;
 463         write_unlock_bh(&sk->sk_callback_lock);
 464
 465         sk->sk_no_check = 0;
 466
 467         sock_release(sock);
 468 clear_close_wait:
 469         smp_mb__before_clear_bit();
 470         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 471         smp_mb__after_clear_bit();
 472 }
 473
 474 /**
 475  * xs_destroy - prepare to shutdown a transport
 476  * @xprt: doomed transport
 477  *
 478  */
 479 static void xs_destroy(struct rpc_xprt *xprt)
 480 {
 481         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 482
 483         cancel_delayed_work(&xprt->connect_worker);
 484         flush_scheduled_work();
 485
 486         xprt_disconnect(xprt);
 487         xs_close(xprt);
 488         kfree(xprt->slot);
 489 }
 490
 491 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 492 {
 493         return (struct rpc_xprt *) sk->sk_user_data;
 494 }
 495
 496 /**
 497  * xs_udp_data_ready - "data ready" callback for UDP sockets
 498  * @sk: socket with data to read
 499  * @len: how much data to read
 500  *
 501  */
 502 static void xs_udp_data_ready(struct sock *sk, int len)
 503 {
 504         struct rpc_task *task;
 505         struct rpc_xprt *xprt;
 506         struct rpc_rqst *rovr;
 507         struct sk_buff *skb;
 508         int err, repsize, copied;
 509         u32 _xid, *xp;
 510
 511         read_lock(&sk->sk_callback_lock);
 512         dprintk("RPC:      xs_udp_data_ready...\n");
 513         if (!(xprt = xprt_from_sock(sk)))
 514                 goto out;
 515
 516         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 517                 goto out;
 518
 519         if (xprt->shutdown)
 520                 goto dropit;
 521
 522         repsize = skb->len - sizeof(struct udphdr);
 523         if (repsize < 4) {
 524                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 525                 goto dropit;
 526         }
 527
 528         /* Copy the XID from the skb... */
 529         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 530                                 sizeof(_xid), &_xid);
 531         if (xp == NULL)
 532                 goto dropit;
 533
 534         /* Look up and lock the request corresponding to the given XID */
 535         spin_lock(&xprt->transport_lock);
 536         rovr = xprt_lookup_rqst(xprt, *xp);
 537         if (!rovr)
 538                 goto out_unlock;
 539         task = rovr->rq_task;
 540
 541         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 542                 copied = repsize;
 543
 544         /* Suck it into the iovec, verify checksum if not done by hw. */
 545         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 546                 goto out_unlock;
 547
 548         /* Something worked... */
 549         dst_confirm(skb->dst);
 550
 551         xprt_adjust_cwnd(task, copied);
 552         xprt_update_rtt(task);
 553         xprt_complete_rqst(task, copied);
 554
 555  out_unlock:
 556         spin_unlock(&xprt->transport_lock);
 557  dropit:
 558         skb_free_datagram(sk, skb);
 559  out:
 560         read_unlock(&sk->sk_callback_lock);
 561 }
 562
 563 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 564 {
 565         if (len > desc->count)
 566                 len = desc->count;
 567         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 568                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 569                                 len, desc->count);
 570                 return 0;
 571         }
 572         desc->offset += len;
 573         desc->count -= len;
 574         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 575                         len, desc->count);
 576         return len;
 577 }
 578
 579 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 580 {
 581         size_t len, used;
 582         char *p;
 583
 584         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 585         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 586         used = xs_tcp_copy_data(desc, p, len);
 587         xprt->tcp_offset += used;
 588         if (used != len)
 589                 return;
 590
 591         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 592         if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 593                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 594         else
 595                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 596         xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 597
 598         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 599         xprt->tcp_offset = 0;
 600
 601         /* Sanity check of the record length */
 602         if (unlikely(xprt->tcp_reclen < 4)) {
 603                 dprintk("RPC:      invalid TCP record fragment length\n");
 604                 xprt_disconnect(xprt);
 605                 return;
 606         }
 607         dprintk("RPC:      reading TCP record fragment of length %d\n",
 608                         xprt->tcp_reclen);
 609 }
 610
 611 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 612 {
 613         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 614                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 615         if (xprt->tcp_offset == xprt->tcp_reclen) {
 616                 xprt->tcp_flags |= XPRT_COPY_RECM;
 617                 xprt->tcp_offset = 0;
 618                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 619                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 620                         xprt->tcp_flags |= XPRT_COPY_XID;
 621                         xprt->tcp_copied = 0;
 622                 }
 623         }
 624 }
 625
 626 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 627 {
 628         size_t len, used;
 629         char *p;
 630
 631         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 632         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 633         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 634         used = xs_tcp_copy_data(desc, p, len);
 635         xprt->tcp_offset += used;
 636         if (used != len)
 637                 return;
 638         xprt->tcp_flags &= ~XPRT_COPY_XID;
 639         xprt->tcp_flags |= XPRT_COPY_DATA;
 640         xprt->tcp_copied = 4;
 641         dprintk("RPC:      reading reply for XID %08x\n",
 642                                                 ntohl(xprt->tcp_xid));
 643         xs_tcp_check_recm(xprt);
 644 }
 645
 646 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 647 {
 648         struct rpc_rqst *req;
 649         struct xdr_buf *rcvbuf;
 650         size_t len;
 651         ssize_t r;
 652
 653         /* Find and lock the request corresponding to this xid */
 654         spin_lock(&xprt->transport_lock);
 655         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 656         if (!req) {
 657                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 658                 dprintk("RPC:      XID %08x request not found!\n",
 659                                 ntohl(xprt->tcp_xid));
 660                 spin_unlock(&xprt->transport_lock);
 661                 return;
 662         }
 663
 664         rcvbuf = &req->rq_private_buf;
 665         len = desc->count;
 666         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 667                 skb_reader_t my_desc;
 668
 669                 len = xprt->tcp_reclen - xprt->tcp_offset;
 670                 memcpy(&my_desc, desc, sizeof(my_desc));
 671                 my_desc.count = len;
 672                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 673                                           &my_desc, xs_tcp_copy_data);
 674                 desc->count -= r;
 675                 desc->offset += r;
 676         } else
 677                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 678                                           desc, xs_tcp_copy_data);
 679
 680         if (r > 0) {
 681                 xprt->tcp_copied += r;
 682                 xprt->tcp_offset += r;
 683         }
 684         if (r != len) {
 685                 /* Error when copying to the receive buffer,
 686                  * usually because we weren't able to allocate
 687                  * additional buffer pages. All we can do now
 688                  * is turn off XPRT_COPY_DATA, so the request
 689                  * will not receive any additional updates,
 690                  * and time out.
 691                  * Any remaining data from this record will
 692                  * be discarded.
 693                  */
 694                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 695                 dprintk("RPC:      XID %08x truncated request\n",
 696                                 ntohl(xprt->tcp_xid));
 697                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 698                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 699                 goto out;
 700         }
 701
 702         dprintk("RPC:      XID %08x read %Zd bytes\n",
 703                         ntohl(xprt->tcp_xid), r);
 704         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 705                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 706
 707         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 708                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 709         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 710                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 711                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 712         }
 713
 714 out:
 715         if (!(xprt->tcp_flags & XPRT_COPY_DATA))
 716                 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
 717         spin_unlock(&xprt->transport_lock);
 718         xs_tcp_check_recm(xprt);
 719 }
 720
 721 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 722 {
 723         size_t len;
 724
 725         len = xprt->tcp_reclen - xprt->tcp_offset;
 726         if (len > desc->count)
 727                 len = desc->count;
 728         desc->count -= len;
 729         desc->offset += len;
 730         xprt->tcp_offset += len;
 731         dprintk("RPC:      discarded %Zu bytes\n", len);
 732         xs_tcp_check_recm(xprt);
 733 }
 734
 735 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 736 {
 737         struct rpc_xprt *xprt = rd_desc->arg.data;
 738         skb_reader_t desc = {
 739                 .skb    = skb,
 740                 .offset = offset,
 741                 .count  = len,
 742                 .csum   = 0
 743         };
 744
 745         dprintk("RPC:      xs_tcp_data_recv started\n");
 746         do {
 747                 /* Read in a new fragment marker if necessary */
 748                 /* Can we ever really expect to get completely empty fragments? */
 749                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 750                         xs_tcp_read_fraghdr(xprt, &desc);
 751                         continue;
 752                 }
 753                 /* Read in the xid if necessary */
 754                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 755                         xs_tcp_read_xid(xprt, &desc);
 756                         continue;
 757                 }
 758                 /* Read in the request data */
 759                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 760                         xs_tcp_read_request(xprt, &desc);
 761                         continue;
 762                 }
 763                 /* Skip over any trailing bytes on short reads */
 764                 xs_tcp_read_discard(xprt, &desc);
 765         } while (desc.count);
 766         dprintk("RPC:      xs_tcp_data_recv done\n");
 767         return len - desc.count;
 768 }
 769
 770 /**
 771  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 772  * @sk: socket with data to read
 773  * @bytes: how much data to read
 774  *
 775  */
 776 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 777 {
 778         struct rpc_xprt *xprt;
 779         read_descriptor_t rd_desc;
 780
 781         read_lock(&sk->sk_callback_lock);
 782         dprintk("RPC:      xs_tcp_data_ready...\n");
 783         if (!(xprt = xprt_from_sock(sk)))
 784                 goto out;
 785         if (xprt->shutdown)
 786                 goto out;
 787
 788         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 789         rd_desc.arg.data = xprt;
 790         rd_desc.count = 65536;
 791         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 792 out:
 793         read_unlock(&sk->sk_callback_lock);
 794 }
 795
 796 /**
 797  * xs_tcp_state_change - callback to handle TCP socket state changes
 798  * @sk: socket whose state has changed
 799  *
 800  */
 801 static void xs_tcp_state_change(struct sock *sk)
 802 {
 803         struct rpc_xprt *xprt;
 804
 805         read_lock(&sk->sk_callback_lock);
 806         if (!(xprt = xprt_from_sock(sk)))
 807                 goto out;
 808         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 809         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 810                                 sk->sk_state, xprt_connected(xprt),
 811                                 sock_flag(sk, SOCK_DEAD),
 812                                 sock_flag(sk, SOCK_ZAPPED));
 813
 814         switch (sk->sk_state) {
 815         case TCP_ESTABLISHED:
 816                 spin_lock_bh(&xprt->transport_lock);
 817                 if (!xprt_test_and_set_connected(xprt)) {
 818                         /* Reset TCP record info */
 819                         xprt->tcp_offset = 0;
 820                         xprt->tcp_reclen = 0;
 821                         xprt->tcp_copied = 0;
 822                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 823                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 824                         xprt_wake_pending_tasks(xprt, 0);
 825                 }
 826                 spin_unlock_bh(&xprt->transport_lock);
 827                 break;
 828         case TCP_SYN_SENT:
 829         case TCP_SYN_RECV:
 830                 break;
 831         case TCP_CLOSE_WAIT:
 832                 /* Try to schedule an autoclose RPC calls */
 833                 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 834                 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 835                         schedule_work(&xprt->task_cleanup);
 836         default:
 837                 xprt_disconnect(xprt);
 838         }
 839  out:
 840         read_unlock(&sk->sk_callback_lock);
 841 }
 842
 843 /**
 844  * xs_udp_write_space - callback invoked when socket buffer space
 845  *                             becomes available
 846  * @sk: socket whose state has changed
 847  *
 848  * Called when more output buffer space is available for this socket.
 849  * We try not to wake our writers until they can make "significant"
 850  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 851  * with a bunch of small requests.
 852  */
 853 static void xs_udp_write_space(struct sock *sk)
 854 {
 855         read_lock(&sk->sk_callback_lock);
 856
 857         /* from net/core/sock.c:sock_def_write_space */
 858         if (sock_writeable(sk)) {
 859                 struct socket *sock;
 860                 struct rpc_xprt *xprt;
 861
 862                 if (unlikely(!(sock = sk->sk_socket)))
 863                         goto out;
 864                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 865                         goto out;
 866                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 867                         goto out;
 868
 869                 xprt_write_space(xprt);
 870         }
 871
 872  out:
 873         read_unlock(&sk->sk_callback_lock);
 874 }
 875
 876 /**
 877  * xs_tcp_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_tcp_write_space(struct sock *sk)
 887 {
 888         read_lock(&sk->sk_callback_lock);
 889
 890         /* from net/core/stream.c:sk_stream_write_space */
 891         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(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 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 910 {
 911         struct sock *sk = xprt->inet;
 912
 913         if (xprt->rcvsize) {
 914                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 915                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 916         }
 917         if (xprt->sndsize) {
 918                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 919                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 920                 sk->sk_write_space(sk);
 921         }
 922 }
 923
 924 /**
 925  * xs_udp_set_buffer_size - set send and receive limits
 926  * @xprt: generic transport
 927  * @sndsize: requested size of send buffer, in bytes
 928  * @rcvsize: requested size of receive buffer, in bytes
 929  *
 930  * Set socket send and receive buffer size limits.
 931  */
 932 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
 933 {
 934         xprt->sndsize = 0;
 935         if (sndsize)
 936                 xprt->sndsize = sndsize + 1024;
 937         xprt->rcvsize = 0;
 938         if (rcvsize)
 939                 xprt->rcvsize = rcvsize + 1024;
 940
 941         xs_udp_do_set_buffer_size(xprt);
 942 }
 943
 944 /**
 945  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
 946  * @task: task that timed out
 947  *
 948  * Adjust the congestion window after a retransmit timeout has occurred.
 949  */
 950 static void xs_udp_timer(struct rpc_task *task)
 951 {
 952         xprt_adjust_cwnd(task, -ETIMEDOUT);
 953 }
 954
 955 static unsigned short xs_get_random_port(void)
 956 {
 957         unsigned short range = xprt_max_resvport - xprt_min_resvport;
 958         unsigned short rand = (unsigned short) net_random() % range;
 959         return rand + xprt_min_resvport;
 960 }
 961
 962 /**
 963  * xs_set_port - reset the port number in the remote endpoint address
 964  * @xprt: generic transport
 965  * @port: new port number
 966  *
 967  */
 968 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
 969 {
 970         dprintk("RPC:      setting port for xprt %p to %u\n", xprt, port);
 971         xprt->addr.sin_port = htons(port);
 972 }
 973
 974 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
 975 {
 976         struct sockaddr_in myaddr = {
 977                 .sin_family = AF_INET,
 978         };
 979         int err;
 980         unsigned short port = xprt->port;
 981
 982         do {
 983                 myaddr.sin_port = htons(port);
 984                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
 985                                                 sizeof(myaddr));
 986                 if (err == 0) {
 987                         xprt->port = port;
 988                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
 989                                         port);
 990                         return 0;
 991                 }
 992                 if (port <= xprt_min_resvport)
 993                         port = xprt_max_resvport;
 994                 else
 995                         port--;
 996         } while (err == -EADDRINUSE && port != xprt->port);
 997
 998         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
 999         return err;
1000 }
1001
1002 /**
1003  * xs_udp_connect_worker - set up a UDP socket
1004  * @args: RPC transport to connect
1005  *
1006  * Invoked by a work queue tasklet.
1007  */
1008 static void xs_udp_connect_worker(void *args)
1009 {
1010         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
1011         struct socket *sock = xprt->sock;
1012         int err, status = -EIO;
1013
1014         if (xprt->shutdown || xprt->addr.sin_port == 0)
1015                 goto out;
1016
1017         dprintk("RPC:      xs_udp_connect_worker for xprt %p\n", xprt);
1018
1019         /* Start by resetting any existing state */
1020         xs_close(xprt);
1021
1022         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
1023                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
1024                 goto out;
1025         }
1026
1027         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1028                 sock_release(sock);
1029                 goto out;
1030         }
1031
1032         if (!xprt->inet) {
1033                 struct sock *sk = sock->sk;
1034
1035                 write_lock_bh(&sk->sk_callback_lock);
1036
1037                 sk->sk_user_data = xprt;
1038                 xprt->old_data_ready = sk->sk_data_ready;
1039                 xprt->old_state_change = sk->sk_state_change;
1040                 xprt->old_write_space = sk->sk_write_space;
1041                 sk->sk_data_ready = xs_udp_data_ready;
1042                 sk->sk_write_space = xs_udp_write_space;
1043                 sk->sk_no_check = UDP_CSUM_NORCV;
1044                 sk->sk_allocation = GFP_ATOMIC;
1045
1046                 xprt_set_connected(xprt);
1047
1048                 /* Reset to new socket */
1049                 xprt->sock = sock;
1050                 xprt->inet = sk;
1051
1052                 write_unlock_bh(&sk->sk_callback_lock);
1053         }
1054         xs_udp_do_set_buffer_size(xprt);
1055         status = 0;
1056 out:
1057         xprt_wake_pending_tasks(xprt, status);
1058         xprt_clear_connecting(xprt);
1059 }
1060
1061 /*
1062  * We need to preserve the port number so the reply cache on the server can
1063  * find our cached RPC replies when we get around to reconnecting.
1064  */
1065 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1066 {
1067         int result;
1068         struct socket *sock = xprt->sock;
1069         struct sockaddr any;
1070
1071         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1072
1073         /*
1074          * Disconnect the transport socket by doing a connect operation
1075          * with AF_UNSPEC.  This should return immediately...
1076          */
1077         memset(&any, 0, sizeof(any));
1078         any.sa_family = AF_UNSPEC;
1079         result = kernel_connect(sock, &any, sizeof(any), 0);
1080         if (result)
1081                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1082                                 result);
1083 }
1084
1085 /**
1086  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1087  * @args: RPC transport to connect
1088  *
1089  * Invoked by a work queue tasklet.
1090  */
1091 static void xs_tcp_connect_worker(void *args)
1092 {
1093         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1094         struct socket *sock = xprt->sock;
1095         int err, status = -EIO;
1096
1097         if (xprt->shutdown || xprt->addr.sin_port == 0)
1098                 goto out;
1099
1100         dprintk("RPC:      xs_tcp_connect_worker for xprt %p\n", xprt);
1101
1102         if (!xprt->sock) {
1103                 /* start from scratch */
1104                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1105                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1106                         goto out;
1107                 }
1108
1109                 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1110                         sock_release(sock);
1111                         goto out;
1112                 }
1113         } else
1114                 /* "close" the socket, preserving the local port */
1115                 xs_tcp_reuse_connection(xprt);
1116
1117         if (!xprt->inet) {
1118                 struct sock *sk = sock->sk;
1119
1120                 write_lock_bh(&sk->sk_callback_lock);
1121
1122                 sk->sk_user_data = xprt;
1123                 xprt->old_data_ready = sk->sk_data_ready;
1124                 xprt->old_state_change = sk->sk_state_change;
1125                 xprt->old_write_space = sk->sk_write_space;
1126                 sk->sk_data_ready = xs_tcp_data_ready;
1127                 sk->sk_state_change = xs_tcp_state_change;
1128                 sk->sk_write_space = xs_tcp_write_space;
1129                 sk->sk_allocation = GFP_ATOMIC;
1130
1131                 /* socket options */
1132                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1133                 sock_reset_flag(sk, SOCK_LINGER);
1134                 tcp_sk(sk)->linger2 = 0;
1135                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1136
1137                 xprt_clear_connected(xprt);
1138
1139                 /* Reset to new socket */
1140                 xprt->sock = sock;
1141                 xprt->inet = sk;
1142
1143                 write_unlock_bh(&sk->sk_callback_lock);
1144         }
1145
1146         /* Tell the socket layer to start connecting... */
1147         xprt->stat.connect_count++;
1148         xprt->stat.connect_start = jiffies;
1149         status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
1150                         sizeof(xprt->addr), O_NONBLOCK);
1151         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1152                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1153         if (status < 0) {
1154                 switch (status) {
1155                         case -EINPROGRESS:
1156                         case -EALREADY:
1157                                 goto out_clear;
1158                         case -ECONNREFUSED:
1159                         case -ECONNRESET:
1160                                 /* retry with existing socket, after a delay */
1161                                 break;
1162                         default:
1163                                 /* get rid of existing socket, and retry */
1164                                 xs_close(xprt);
1165                                 break;
1166                 }
1167         }
1168 out:
1169         xprt_wake_pending_tasks(xprt, status);
1170 out_clear:
1171         xprt_clear_connecting(xprt);
1172 }
1173
1174 /**
1175  * xs_connect - connect a socket to a remote endpoint
1176  * @task: address of RPC task that manages state of connect request
1177  *
1178  * TCP: If the remote end dropped the connection, delay reconnecting.
1179  *
1180  * UDP socket connects are synchronous, but we use a work queue anyway
1181  * to guarantee that even unprivileged user processes can set up a
1182  * socket on a privileged port.
1183  *
1184  * If a UDP socket connect fails, the delay behavior here prevents
1185  * retry floods (hard mounts).
1186  */
1187 static void xs_connect(struct rpc_task *task)
1188 {
1189         struct rpc_xprt *xprt = task->tk_xprt;
1190
1191         if (xprt_test_and_set_connecting(xprt))
1192                 return;
1193
1194         if (xprt->sock != NULL) {
1195                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1196                                 xprt, xprt->reestablish_timeout / HZ);
1197                 schedule_delayed_work(&xprt->connect_worker,
1198                                         xprt->reestablish_timeout);
1199                 xprt->reestablish_timeout <<= 1;
1200                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1201                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1202         } else {
1203                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1204                 schedule_work(&xprt->connect_worker);
1205
1206                 /* flush_scheduled_work can sleep... */
1207                 if (!RPC_IS_ASYNC(task))
1208                         flush_scheduled_work();
1209         }
1210 }
1211
1212 /**
1213  * xs_udp_print_stats - display UDP socket-specifc stats
1214  * @xprt: rpc_xprt struct containing statistics
1215  * @seq: output file
1216  *
1217  */
1218 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1219 {
1220         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
1221                         xprt->port,
1222                         xprt->stat.bind_count,
1223                         xprt->stat.sends,
1224                         xprt->stat.recvs,
1225                         xprt->stat.bad_xids,
1226                         xprt->stat.req_u,
1227                         xprt->stat.bklog_u);
1228 }
1229
1230 /**
1231  * xs_tcp_print_stats - display TCP socket-specifc stats
1232  * @xprt: rpc_xprt struct containing statistics
1233  * @seq: output file
1234  *
1235  */
1236 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
1237 {
1238         long idle_time = 0;
1239
1240         if (xprt_connected(xprt))
1241                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
1242
1243         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
1244                         xprt->port,
1245                         xprt->stat.bind_count,
1246                         xprt->stat.connect_count,
1247                         xprt->stat.connect_time,
1248                         idle_time,
1249                         xprt->stat.sends,
1250                         xprt->stat.recvs,
1251                         xprt->stat.bad_xids,
1252                         xprt->stat.req_u,
1253                         xprt->stat.bklog_u);
1254 }
1255
1256 static struct rpc_xprt_ops xs_udp_ops = {
1257         .set_buffer_size        = xs_udp_set_buffer_size,
1258         .reserve_xprt           = xprt_reserve_xprt_cong,
1259         .release_xprt           = xprt_release_xprt_cong,
1260         .set_port               = xs_set_port,
1261         .connect                = xs_connect,
1262         .buf_alloc              = rpc_malloc,
1263         .buf_free               = rpc_free,
1264         .send_request           = xs_udp_send_request,
1265         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1266         .timer                  = xs_udp_timer,
1267         .release_request        = xprt_release_rqst_cong,
1268         .close                  = xs_close,
1269         .destroy                = xs_destroy,
1270         .print_stats            = xs_udp_print_stats,
1271 };
1272
1273 static struct rpc_xprt_ops xs_tcp_ops = {
1274         .reserve_xprt           = xprt_reserve_xprt,
1275         .release_xprt           = xs_tcp_release_xprt,
1276         .set_port               = xs_set_port,
1277         .connect                = xs_connect,
1278         .buf_alloc              = rpc_malloc,
1279         .buf_free               = rpc_free,
1280         .send_request           = xs_tcp_send_request,
1281         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1282         .close                  = xs_close,
1283         .destroy                = xs_destroy,
1284         .print_stats            = xs_tcp_print_stats,
1285 };
1286
1287 /**
1288  * xs_setup_udp - Set up transport to use a UDP socket
1289  * @xprt: transport to set up
1290  * @to:   timeout parameters
1291  *
1292  */
1293 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1294 {
1295         size_t slot_table_size;
1296
1297         dprintk("RPC:      setting up udp-ipv4 transport...\n");
1298
1299         xprt->max_reqs = xprt_udp_slot_table_entries;
1300         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1301         xprt->slot = kzalloc(slot_table_size, GFP_KERNEL);
1302         if (xprt->slot == NULL)
1303                 return -ENOMEM;
1304
1305         xprt->prot = IPPROTO_UDP;
1306         xprt->port = xs_get_random_port();
1307         xprt->tsh_size = 0;
1308         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1309         /* XXX: header size can vary due to auth type, IPv6, etc. */
1310         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1311
1312         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1313         xprt->bind_timeout = XS_BIND_TO;
1314         xprt->connect_timeout = XS_UDP_CONN_TO;
1315         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1316         xprt->idle_timeout = XS_IDLE_DISC_TO;
1317
1318         xprt->ops = &xs_udp_ops;
1319
1320         if (to)
1321                 xprt->timeout = *to;
1322         else
1323                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1324
1325         return 0;
1326 }
1327
1328 /**
1329  * xs_setup_tcp - Set up transport to use a TCP socket
1330  * @xprt: transport to set up
1331  * @to: timeout parameters
1332  *
1333  */
1334 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1335 {
1336         size_t slot_table_size;
1337
1338         dprintk("RPC:      setting up tcp-ipv4 transport...\n");
1339
1340         xprt->max_reqs = xprt_tcp_slot_table_entries;
1341         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1342         xprt->slot = kzalloc(slot_table_size, GFP_KERNEL);
1343         if (xprt->slot == NULL)
1344                 return -ENOMEM;
1345
1346         xprt->prot = IPPROTO_TCP;
1347         xprt->port = xs_get_random_port();
1348         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1349         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1350         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1351
1352         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1353         xprt->bind_timeout = XS_BIND_TO;
1354         xprt->connect_timeout = XS_TCP_CONN_TO;
1355         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1356         xprt->idle_timeout = XS_IDLE_DISC_TO;
1357
1358         xprt->ops = &xs_tcp_ops;
1359
1360         if (to)
1361                 xprt->timeout = *to;
1362         else
1363                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1364
1365         return 0;
1366 }