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         ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
 178
 179         if (unlikely(!sock))
 180                 return -ENOTCONN;
 181
 182         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 183
 184         len = xdr->head[0].iov_len;
 185         if (base < len || (addr != NULL && base == 0)) {
 186                 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
 187                 if (ret == 0)
 188                         ret = err;
 189                 else if (err > 0)
 190                         ret += err;
 191                 if (err != (len - base))
 192                         goto out;
 193                 base = 0;
 194         } else
 195                 base -= len;
 196
 197         if (unlikely(pglen == 0))
 198                 goto copy_tail;
 199         if (unlikely(base >= pglen)) {
 200                 base -= pglen;
 201                 goto copy_tail;
 202         }
 203         if (base || xdr->page_base) {
 204                 pglen -= base;
 205                 base += xdr->page_base;
 206                 ppage += base >> PAGE_CACHE_SHIFT;
 207                 base &= ~PAGE_CACHE_MASK;
 208         }
 209
 210         sendpage = sock->ops->sendpage ? : sock_no_sendpage;
 211         do {
 212                 int flags = XS_SENDMSG_FLAGS;
 213
 214                 len = PAGE_CACHE_SIZE;
 215                 if (base)
 216                         len -= base;
 217                 if (pglen < len)
 218                         len = pglen;
 219
 220                 if (pglen != len || xdr->tail[0].iov_len != 0)
 221                         flags |= MSG_MORE;
 222
 223                 /* Hmm... We might be dealing with highmem pages */
 224                 if (PageHighMem(*ppage))
 225                         sendpage = sock_no_sendpage;
 226                 err = sendpage(sock, *ppage, base, len, flags);
 227                 if (ret == 0)
 228                         ret = err;
 229                 else if (err > 0)
 230                         ret += err;
 231                 if (err != len)
 232                         goto out;
 233                 base = 0;
 234                 ppage++;
 235         } while ((pglen -= len) != 0);
 236 copy_tail:
 237         len = xdr->tail[0].iov_len;
 238         if (base < len) {
 239                 err = xs_send_tail(sock, xdr, base, len);
 240                 if (ret == 0)
 241                         ret = err;
 242                 else if (err > 0)
 243                         ret += err;
 244         }
 245 out:
 246         return ret;
 247 }
 248
 249 /**
 250  * xs_nospace - place task on wait queue if transmit was incomplete
 251  * @task: task to put to sleep
 252  *
 253  */
 254 static void xs_nospace(struct rpc_task *task)
 255 {
 256         struct rpc_rqst *req = task->tk_rqstp;
 257         struct rpc_xprt *xprt = req->rq_xprt;
 258
 259         dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
 260                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 261                         req->rq_slen);
 262
 263         if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
 264                 /* Protect against races with write_space */
 265                 spin_lock_bh(&xprt->transport_lock);
 266
 267                 /* Don't race with disconnect */
 268                 if (!xprt_connected(xprt))
 269                         task->tk_status = -ENOTCONN;
 270                 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
 271                         xprt_wait_for_buffer_space(task);
 272
 273                 spin_unlock_bh(&xprt->transport_lock);
 274         } else
 275                 /* Keep holding the socket if it is blocked */
 276                 rpc_delay(task, HZ>>4);
 277 }
 278
 279 /**
 280  * xs_udp_send_request - write an RPC request to a UDP socket
 281  * @task: address of RPC task that manages the state of an RPC request
 282  *
 283  * Return values:
 284  *        0:    The request has been sent
 285  *   EAGAIN:    The socket was blocked, please call again later to
 286  *              complete the request
 287  * ENOTCONN:    Caller needs to invoke connect logic then call again
 288  *    other:    Some other error occured, the request was not sent
 289  */
 290 static int xs_udp_send_request(struct rpc_task *task)
 291 {
 292         struct rpc_rqst *req = task->tk_rqstp;
 293         struct rpc_xprt *xprt = req->rq_xprt;
 294         struct xdr_buf *xdr = &req->rq_snd_buf;
 295         int status;
 296
 297         xs_pktdump("packet data:",
 298                                 req->rq_svec->iov_base,
 299                                 req->rq_svec->iov_len);
 300
 301         req->rq_xtime = jiffies;
 302         status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
 303                                 sizeof(xprt->addr), xdr, req->rq_bytes_sent);
 304
 305         dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
 306                         xdr->len - req->rq_bytes_sent, status);
 307
 308         if (likely(status >= (int) req->rq_slen))
 309                 return 0;
 310
 311         /* Still some bytes left; set up for a retry later. */
 312         if (status > 0)
 313                 status = -EAGAIN;
 314
 315         switch (status) {
 316         case -ENETUNREACH:
 317         case -EPIPE:
 318         case -ECONNREFUSED:
 319                 /* When the server has died, an ICMP port unreachable message
 320                  * prompts ECONNREFUSED. */
 321                 break;
 322         case -EAGAIN:
 323                 xs_nospace(task);
 324                 break;
 325         default:
 326                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 327                         -status);
 328                 break;
 329         }
 330
 331         return status;
 332 }
 333
 334 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 335 {
 336         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 337         rpc_fraghdr *base = buf->head[0].iov_base;
 338         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 339 }
 340
 341 /**
 342  * xs_tcp_send_request - write an RPC request to a TCP socket
 343  * @task: address of RPC task that manages the state of an RPC request
 344  *
 345  * Return values:
 346  *        0:    The request has been sent
 347  *   EAGAIN:    The socket was blocked, please call again later to
 348  *              complete the request
 349  * ENOTCONN:    Caller needs to invoke connect logic then call again
 350  *    other:    Some other error occured, the request was not sent
 351  *
 352  * XXX: In the case of soft timeouts, should we eventually give up
 353  *      if sendmsg is not able to make progress?
 354  */
 355 static int xs_tcp_send_request(struct rpc_task *task)
 356 {
 357         struct rpc_rqst *req = task->tk_rqstp;
 358         struct rpc_xprt *xprt = req->rq_xprt;
 359         struct xdr_buf *xdr = &req->rq_snd_buf;
 360         int status, retry = 0;
 361
 362         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 363
 364         xs_pktdump("packet data:",
 365                                 req->rq_svec->iov_base,
 366                                 req->rq_svec->iov_len);
 367
 368         /* Continue transmitting the packet/record. We must be careful
 369          * to cope with writespace callbacks arriving _after_ we have
 370          * called sendmsg(). */
 371         while (1) {
 372                 req->rq_xtime = jiffies;
 373                 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
 374                                                 req->rq_bytes_sent);
 375
 376                 dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
 377                                 xdr->len - req->rq_bytes_sent, status);
 378
 379                 if (unlikely(status < 0))
 380                         break;
 381
 382                 /* If we've sent the entire packet, immediately
 383                  * reset the count of bytes sent. */
 384                 req->rq_bytes_sent += status;
 385                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 386                         req->rq_bytes_sent = 0;
 387                         return 0;
 388                 }
 389
 390                 status = -EAGAIN;
 391                 if (retry++ > XS_SENDMSG_RETRY)
 392                         break;
 393         }
 394
 395         switch (status) {
 396         case -EAGAIN:
 397                 xs_nospace(task);
 398                 break;
 399         case -ECONNREFUSED:
 400         case -ECONNRESET:
 401         case -ENOTCONN:
 402         case -EPIPE:
 403                 status = -ENOTCONN;
 404                 break;
 405         default:
 406                 dprintk("RPC:      sendmsg returned unrecognized error %d\n",
 407                         -status);
 408                 xprt_disconnect(xprt);
 409                 break;
 410         }
 411
 412         return status;
 413 }
 414
 415 /**
 416  * xs_close - close a socket
 417  * @xprt: transport
 418  *
 419  * This is used when all requests are complete; ie, no DRC state remains
 420  * on the server we want to save.
 421  */
 422 static void xs_close(struct rpc_xprt *xprt)
 423 {
 424         struct socket *sock = xprt->sock;
 425         struct sock *sk = xprt->inet;
 426
 427         if (!sk)
 428                 return;
 429
 430         dprintk("RPC:      xs_close xprt %p\n", xprt);
 431
 432         write_lock_bh(&sk->sk_callback_lock);
 433         xprt->inet = NULL;
 434         xprt->sock = NULL;
 435
 436         sk->sk_user_data = NULL;
 437         sk->sk_data_ready = xprt->old_data_ready;
 438         sk->sk_state_change = xprt->old_state_change;
 439         sk->sk_write_space = xprt->old_write_space;
 440         write_unlock_bh(&sk->sk_callback_lock);
 441
 442         sk->sk_no_check = 0;
 443
 444         sock_release(sock);
 445 }
 446
 447 /**
 448  * xs_destroy - prepare to shutdown a transport
 449  * @xprt: doomed transport
 450  *
 451  */
 452 static void xs_destroy(struct rpc_xprt *xprt)
 453 {
 454         dprintk("RPC:      xs_destroy xprt %p\n", xprt);
 455
 456         cancel_delayed_work(&xprt->connect_worker);
 457         flush_scheduled_work();
 458
 459         xprt_disconnect(xprt);
 460         xs_close(xprt);
 461         kfree(xprt->slot);
 462 }
 463
 464 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 465 {
 466         return (struct rpc_xprt *) sk->sk_user_data;
 467 }
 468
 469 /**
 470  * xs_udp_data_ready - "data ready" callback for UDP sockets
 471  * @sk: socket with data to read
 472  * @len: how much data to read
 473  *
 474  */
 475 static void xs_udp_data_ready(struct sock *sk, int len)
 476 {
 477         struct rpc_task *task;
 478         struct rpc_xprt *xprt;
 479         struct rpc_rqst *rovr;
 480         struct sk_buff *skb;
 481         int err, repsize, copied;
 482         u32 _xid, *xp;
 483
 484         read_lock(&sk->sk_callback_lock);
 485         dprintk("RPC:      xs_udp_data_ready...\n");
 486         if (!(xprt = xprt_from_sock(sk)))
 487                 goto out;
 488
 489         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 490                 goto out;
 491
 492         if (xprt->shutdown)
 493                 goto dropit;
 494
 495         repsize = skb->len - sizeof(struct udphdr);
 496         if (repsize < 4) {
 497                 dprintk("RPC:      impossible RPC reply size %d!\n", repsize);
 498                 goto dropit;
 499         }
 500
 501         /* Copy the XID from the skb... */
 502         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 503                                 sizeof(_xid), &_xid);
 504         if (xp == NULL)
 505                 goto dropit;
 506
 507         /* Look up and lock the request corresponding to the given XID */
 508         spin_lock(&xprt->transport_lock);
 509         rovr = xprt_lookup_rqst(xprt, *xp);
 510         if (!rovr)
 511                 goto out_unlock;
 512         task = rovr->rq_task;
 513
 514         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 515                 copied = repsize;
 516
 517         /* Suck it into the iovec, verify checksum if not done by hw. */
 518         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 519                 goto out_unlock;
 520
 521         /* Something worked... */
 522         dst_confirm(skb->dst);
 523
 524         xprt_adjust_cwnd(task, copied);
 525         xprt_update_rtt(task);
 526         xprt_complete_rqst(task, copied);
 527
 528  out_unlock:
 529         spin_unlock(&xprt->transport_lock);
 530  dropit:
 531         skb_free_datagram(sk, skb);
 532  out:
 533         read_unlock(&sk->sk_callback_lock);
 534 }
 535
 536 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 537 {
 538         if (len > desc->count)
 539                 len = desc->count;
 540         if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
 541                 dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
 542                                 len, desc->count);
 543                 return 0;
 544         }
 545         desc->offset += len;
 546         desc->count -= len;
 547         dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
 548                         len, desc->count);
 549         return len;
 550 }
 551
 552 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 553 {
 554         size_t len, used;
 555         char *p;
 556
 557         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 558         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 559         used = xs_tcp_copy_data(desc, p, len);
 560         xprt->tcp_offset += used;
 561         if (used != len)
 562                 return;
 563
 564         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 565         if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 566                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 567         else
 568                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 569         xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 570
 571         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 572         xprt->tcp_offset = 0;
 573
 574         /* Sanity check of the record length */
 575         if (unlikely(xprt->tcp_reclen < 4)) {
 576                 dprintk("RPC:      invalid TCP record fragment length\n");
 577                 xprt_disconnect(xprt);
 578                 return;
 579         }
 580         dprintk("RPC:      reading TCP record fragment of length %d\n",
 581                         xprt->tcp_reclen);
 582 }
 583
 584 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
 585 {
 586         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
 587                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
 588         if (xprt->tcp_offset == xprt->tcp_reclen) {
 589                 xprt->tcp_flags |= XPRT_COPY_RECM;
 590                 xprt->tcp_offset = 0;
 591                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 592                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 593                         xprt->tcp_flags |= XPRT_COPY_XID;
 594                         xprt->tcp_copied = 0;
 595                 }
 596         }
 597 }
 598
 599 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 600 {
 601         size_t len, used;
 602         char *p;
 603
 604         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 605         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 606         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 607         used = xs_tcp_copy_data(desc, p, len);
 608         xprt->tcp_offset += used;
 609         if (used != len)
 610                 return;
 611         xprt->tcp_flags &= ~XPRT_COPY_XID;
 612         xprt->tcp_flags |= XPRT_COPY_DATA;
 613         xprt->tcp_copied = 4;
 614         dprintk("RPC:      reading reply for XID %08x\n",
 615                                                 ntohl(xprt->tcp_xid));
 616         xs_tcp_check_recm(xprt);
 617 }
 618
 619 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 620 {
 621         struct rpc_rqst *req;
 622         struct xdr_buf *rcvbuf;
 623         size_t len;
 624         ssize_t r;
 625
 626         /* Find and lock the request corresponding to this xid */
 627         spin_lock(&xprt->transport_lock);
 628         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 629         if (!req) {
 630                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 631                 dprintk("RPC:      XID %08x request not found!\n",
 632                                 ntohl(xprt->tcp_xid));
 633                 spin_unlock(&xprt->transport_lock);
 634                 return;
 635         }
 636
 637         rcvbuf = &req->rq_private_buf;
 638         len = desc->count;
 639         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 640                 skb_reader_t my_desc;
 641
 642                 len = xprt->tcp_reclen - xprt->tcp_offset;
 643                 memcpy(&my_desc, desc, sizeof(my_desc));
 644                 my_desc.count = len;
 645                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 646                                           &my_desc, xs_tcp_copy_data);
 647                 desc->count -= r;
 648                 desc->offset += r;
 649         } else
 650                 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 651                                           desc, xs_tcp_copy_data);
 652
 653         if (r > 0) {
 654                 xprt->tcp_copied += r;
 655                 xprt->tcp_offset += r;
 656         }
 657         if (r != len) {
 658                 /* Error when copying to the receive buffer,
 659                  * usually because we weren't able to allocate
 660                  * additional buffer pages. All we can do now
 661                  * is turn off XPRT_COPY_DATA, so the request
 662                  * will not receive any additional updates,
 663                  * and time out.
 664                  * Any remaining data from this record will
 665                  * be discarded.
 666                  */
 667                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 668                 dprintk("RPC:      XID %08x truncated request\n",
 669                                 ntohl(xprt->tcp_xid));
 670                 dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 671                                 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 672                 goto out;
 673         }
 674
 675         dprintk("RPC:      XID %08x read %Zd bytes\n",
 676                         ntohl(xprt->tcp_xid), r);
 677         dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
 678                         xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
 679
 680         if (xprt->tcp_copied == req->rq_private_buf.buflen)
 681                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 682         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 683                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 684                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 685         }
 686
 687 out:
 688         if (!(xprt->tcp_flags & XPRT_COPY_DATA))
 689                 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
 690         spin_unlock(&xprt->transport_lock);
 691         xs_tcp_check_recm(xprt);
 692 }
 693
 694 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 695 {
 696         size_t len;
 697
 698         len = xprt->tcp_reclen - xprt->tcp_offset;
 699         if (len > desc->count)
 700                 len = desc->count;
 701         desc->count -= len;
 702         desc->offset += len;
 703         xprt->tcp_offset += len;
 704         dprintk("RPC:      discarded %Zu bytes\n", len);
 705         xs_tcp_check_recm(xprt);
 706 }
 707
 708 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
 709 {
 710         struct rpc_xprt *xprt = rd_desc->arg.data;
 711         skb_reader_t desc = {
 712                 .skb    = skb,
 713                 .offset = offset,
 714                 .count  = len,
 715                 .csum   = 0
 716         };
 717
 718         dprintk("RPC:      xs_tcp_data_recv started\n");
 719         do {
 720                 /* Read in a new fragment marker if necessary */
 721                 /* Can we ever really expect to get completely empty fragments? */
 722                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 723                         xs_tcp_read_fraghdr(xprt, &desc);
 724                         continue;
 725                 }
 726                 /* Read in the xid if necessary */
 727                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 728                         xs_tcp_read_xid(xprt, &desc);
 729                         continue;
 730                 }
 731                 /* Read in the request data */
 732                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 733                         xs_tcp_read_request(xprt, &desc);
 734                         continue;
 735                 }
 736                 /* Skip over any trailing bytes on short reads */
 737                 xs_tcp_read_discard(xprt, &desc);
 738         } while (desc.count);
 739         dprintk("RPC:      xs_tcp_data_recv done\n");
 740         return len - desc.count;
 741 }
 742
 743 /**
 744  * xs_tcp_data_ready - "data ready" callback for TCP sockets
 745  * @sk: socket with data to read
 746  * @bytes: how much data to read
 747  *
 748  */
 749 static void xs_tcp_data_ready(struct sock *sk, int bytes)
 750 {
 751         struct rpc_xprt *xprt;
 752         read_descriptor_t rd_desc;
 753
 754         read_lock(&sk->sk_callback_lock);
 755         dprintk("RPC:      xs_tcp_data_ready...\n");
 756         if (!(xprt = xprt_from_sock(sk)))
 757                 goto out;
 758         if (xprt->shutdown)
 759                 goto out;
 760
 761         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
 762         rd_desc.arg.data = xprt;
 763         rd_desc.count = 65536;
 764         tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
 765 out:
 766         read_unlock(&sk->sk_callback_lock);
 767 }
 768
 769 /**
 770  * xs_tcp_state_change - callback to handle TCP socket state changes
 771  * @sk: socket whose state has changed
 772  *
 773  */
 774 static void xs_tcp_state_change(struct sock *sk)
 775 {
 776         struct rpc_xprt *xprt;
 777
 778         read_lock(&sk->sk_callback_lock);
 779         if (!(xprt = xprt_from_sock(sk)))
 780                 goto out;
 781         dprintk("RPC:      xs_tcp_state_change client %p...\n", xprt);
 782         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 783                                 sk->sk_state, xprt_connected(xprt),
 784                                 sock_flag(sk, SOCK_DEAD),
 785                                 sock_flag(sk, SOCK_ZAPPED));
 786
 787         switch (sk->sk_state) {
 788         case TCP_ESTABLISHED:
 789                 spin_lock_bh(&xprt->transport_lock);
 790                 if (!xprt_test_and_set_connected(xprt)) {
 791                         /* Reset TCP record info */
 792                         xprt->tcp_offset = 0;
 793                         xprt->tcp_reclen = 0;
 794                         xprt->tcp_copied = 0;
 795                         xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 796                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 797                         xprt_wake_pending_tasks(xprt, 0);
 798                 }
 799                 spin_unlock_bh(&xprt->transport_lock);
 800                 break;
 801         case TCP_SYN_SENT:
 802         case TCP_SYN_RECV:
 803                 break;
 804         default:
 805                 xprt_disconnect(xprt);
 806                 break;
 807         }
 808  out:
 809         read_unlock(&sk->sk_callback_lock);
 810 }
 811
 812 /**
 813  * xs_udp_write_space - callback invoked when socket buffer space
 814  *                             becomes available
 815  * @sk: socket whose state has changed
 816  *
 817  * Called when more output buffer space is available for this socket.
 818  * We try not to wake our writers until they can make "significant"
 819  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 820  * with a bunch of small requests.
 821  */
 822 static void xs_udp_write_space(struct sock *sk)
 823 {
 824         read_lock(&sk->sk_callback_lock);
 825
 826         /* from net/core/sock.c:sock_def_write_space */
 827         if (sock_writeable(sk)) {
 828                 struct socket *sock;
 829                 struct rpc_xprt *xprt;
 830
 831                 if (unlikely(!(sock = sk->sk_socket)))
 832                         goto out;
 833                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 834                         goto out;
 835                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 836                         goto out;
 837
 838                 xprt_write_space(xprt);
 839         }
 840
 841  out:
 842         read_unlock(&sk->sk_callback_lock);
 843 }
 844
 845 /**
 846  * xs_tcp_write_space - callback invoked when socket buffer space
 847  *                             becomes available
 848  * @sk: socket whose state has changed
 849  *
 850  * Called when more output buffer space is available for this socket.
 851  * We try not to wake our writers until they can make "significant"
 852  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
 853  * with a bunch of small requests.
 854  */
 855 static void xs_tcp_write_space(struct sock *sk)
 856 {
 857         read_lock(&sk->sk_callback_lock);
 858
 859         /* from net/core/stream.c:sk_stream_write_space */
 860         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
 861                 struct socket *sock;
 862                 struct rpc_xprt *xprt;
 863
 864                 if (unlikely(!(sock = sk->sk_socket)))
 865                         goto out;
 866                 if (unlikely(!(xprt = xprt_from_sock(sk))))
 867                         goto out;
 868                 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 869                         goto out;
 870
 871                 xprt_write_space(xprt);
 872         }
 873
 874  out:
 875         read_unlock(&sk->sk_callback_lock);
 876 }
 877
 878 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
 879 {
 880         struct sock *sk = xprt->inet;
 881
 882         if (xprt->rcvsize) {
 883                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 884                 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
 885         }
 886         if (xprt->sndsize) {
 887                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 888                 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
 889                 sk->sk_write_space(sk);
 890         }
 891 }
 892
 893 /**
 894  * xs_udp_set_buffer_size - set send and receive limits
 895  * @xprt: generic transport
 896  * @sndsize: requested size of send buffer, in bytes
 897  * @rcvsize: requested size of receive buffer, in bytes
 898  *
 899  * Set socket send and receive buffer size limits.
 900  */
 901 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
 902 {
 903         xprt->sndsize = 0;
 904         if (sndsize)
 905                 xprt->sndsize = sndsize + 1024;
 906         xprt->rcvsize = 0;
 907         if (rcvsize)
 908                 xprt->rcvsize = rcvsize + 1024;
 909
 910         xs_udp_do_set_buffer_size(xprt);
 911 }
 912
 913 /**
 914  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
 915  * @task: task that timed out
 916  *
 917  * Adjust the congestion window after a retransmit timeout has occurred.
 918  */
 919 static void xs_udp_timer(struct rpc_task *task)
 920 {
 921         xprt_adjust_cwnd(task, -ETIMEDOUT);
 922 }
 923
 924 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
 925 {
 926         struct sockaddr_in myaddr = {
 927                 .sin_family = AF_INET,
 928         };
 929         int err;
 930         unsigned short port = xprt->port;
 931
 932         do {
 933                 myaddr.sin_port = htons(port);
 934                 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
 935                                                 sizeof(myaddr));
 936                 if (err == 0) {
 937                         xprt->port = port;
 938                         dprintk("RPC:      xs_bindresvport bound to port %u\n",
 939                                         port);
 940                         return 0;
 941                 }
 942                 if (port <= xprt_min_resvport)
 943                         port = xprt_max_resvport;
 944                 else
 945                         port--;
 946         } while (err == -EADDRINUSE && port != xprt->port);
 947
 948         dprintk("RPC:      can't bind to reserved port (%d).\n", -err);
 949         return err;
 950 }
 951
 952 /**
 953  * xs_udp_connect_worker - set up a UDP socket
 954  * @args: RPC transport to connect
 955  *
 956  * Invoked by a work queue tasklet.
 957  */
 958 static void xs_udp_connect_worker(void *args)
 959 {
 960         struct rpc_xprt *xprt = (struct rpc_xprt *) args;
 961         struct socket *sock = xprt->sock;
 962         int err, status = -EIO;
 963
 964         if (xprt->shutdown || xprt->addr.sin_port == 0)
 965                 goto out;
 966
 967         dprintk("RPC:      xs_udp_connect_worker for xprt %p\n", xprt);
 968
 969         /* Start by resetting any existing state */
 970         xs_close(xprt);
 971
 972         if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
 973                 dprintk("RPC:      can't create UDP transport socket (%d).\n", -err);
 974                 goto out;
 975         }
 976
 977         if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
 978                 sock_release(sock);
 979                 goto out;
 980         }
 981
 982         if (!xprt->inet) {
 983                 struct sock *sk = sock->sk;
 984
 985                 write_lock_bh(&sk->sk_callback_lock);
 986
 987                 sk->sk_user_data = xprt;
 988                 xprt->old_data_ready = sk->sk_data_ready;
 989                 xprt->old_state_change = sk->sk_state_change;
 990                 xprt->old_write_space = sk->sk_write_space;
 991                 sk->sk_data_ready = xs_udp_data_ready;
 992                 sk->sk_write_space = xs_udp_write_space;
 993                 sk->sk_no_check = UDP_CSUM_NORCV;
 994                 sk->sk_allocation = GFP_ATOMIC;
 995
 996                 xprt_set_connected(xprt);
 997
 998                 /* Reset to new socket */
 999                 xprt->sock = sock;
1000                 xprt->inet = sk;
1001
1002                 write_unlock_bh(&sk->sk_callback_lock);
1003         }
1004         xs_udp_do_set_buffer_size(xprt);
1005         status = 0;
1006 out:
1007         xprt_wake_pending_tasks(xprt, status);
1008         xprt_clear_connecting(xprt);
1009 }
1010
1011 /*
1012  * We need to preserve the port number so the reply cache on the server can
1013  * find our cached RPC replies when we get around to reconnecting.
1014  */
1015 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1016 {
1017         int result;
1018         struct socket *sock = xprt->sock;
1019         struct sockaddr any;
1020
1021         dprintk("RPC:      disconnecting xprt %p to reuse port\n", xprt);
1022
1023         /*
1024          * Disconnect the transport socket by doing a connect operation
1025          * with AF_UNSPEC.  This should return immediately...
1026          */
1027         memset(&any, 0, sizeof(any));
1028         any.sa_family = AF_UNSPEC;
1029         result = sock->ops->connect(sock, &any, sizeof(any), 0);
1030         if (result)
1031                 dprintk("RPC:      AF_UNSPEC connect return code %d\n",
1032                                 result);
1033 }
1034
1035 /**
1036  * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1037  * @args: RPC transport to connect
1038  *
1039  * Invoked by a work queue tasklet.
1040  */
1041 static void xs_tcp_connect_worker(void *args)
1042 {
1043         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1044         struct socket *sock = xprt->sock;
1045         int err, status = -EIO;
1046
1047         if (xprt->shutdown || xprt->addr.sin_port == 0)
1048                 goto out;
1049
1050         dprintk("RPC:      xs_tcp_connect_worker for xprt %p\n", xprt);
1051
1052         if (!xprt->sock) {
1053                 /* start from scratch */
1054                 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1055                         dprintk("RPC:      can't create TCP transport socket (%d).\n", -err);
1056                         goto out;
1057                 }
1058
1059                 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1060                         sock_release(sock);
1061                         goto out;
1062                 }
1063         } else
1064                 /* "close" the socket, preserving the local port */
1065                 xs_tcp_reuse_connection(xprt);
1066
1067         if (!xprt->inet) {
1068                 struct sock *sk = sock->sk;
1069
1070                 write_lock_bh(&sk->sk_callback_lock);
1071
1072                 sk->sk_user_data = xprt;
1073                 xprt->old_data_ready = sk->sk_data_ready;
1074                 xprt->old_state_change = sk->sk_state_change;
1075                 xprt->old_write_space = sk->sk_write_space;
1076                 sk->sk_data_ready = xs_tcp_data_ready;
1077                 sk->sk_state_change = xs_tcp_state_change;
1078                 sk->sk_write_space = xs_tcp_write_space;
1079                 sk->sk_allocation = GFP_ATOMIC;
1080
1081                 /* socket options */
1082                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1083                 sock_reset_flag(sk, SOCK_LINGER);
1084                 tcp_sk(sk)->linger2 = 0;
1085                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1086
1087                 xprt_clear_connected(xprt);
1088
1089                 /* Reset to new socket */
1090                 xprt->sock = sock;
1091                 xprt->inet = sk;
1092
1093                 write_unlock_bh(&sk->sk_callback_lock);
1094         }
1095
1096         /* Tell the socket layer to start connecting... */
1097         status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
1098                         sizeof(xprt->addr), O_NONBLOCK);
1099         dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
1100                         xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1101         if (status < 0) {
1102                 switch (status) {
1103                         case -EINPROGRESS:
1104                         case -EALREADY:
1105                                 goto out_clear;
1106                         case -ECONNREFUSED:
1107                         case -ECONNRESET:
1108                                 /* retry with existing socket, after a delay */
1109                                 break;
1110                         default:
1111                                 /* get rid of existing socket, and retry */
1112                                 xs_close(xprt);
1113                                 break;
1114                 }
1115         }
1116 out:
1117         xprt_wake_pending_tasks(xprt, status);
1118 out_clear:
1119         xprt_clear_connecting(xprt);
1120 }
1121
1122 /**
1123  * xs_connect - connect a socket to a remote endpoint
1124  * @task: address of RPC task that manages state of connect request
1125  *
1126  * TCP: If the remote end dropped the connection, delay reconnecting.
1127  *
1128  * UDP socket connects are synchronous, but we use a work queue anyway
1129  * to guarantee that even unprivileged user processes can set up a
1130  * socket on a privileged port.
1131  *
1132  * If a UDP socket connect fails, the delay behavior here prevents
1133  * retry floods (hard mounts).
1134  */
1135 static void xs_connect(struct rpc_task *task)
1136 {
1137         struct rpc_xprt *xprt = task->tk_xprt;
1138
1139         if (xprt_test_and_set_connecting(xprt))
1140                 return;
1141
1142         if (xprt->sock != NULL) {
1143                 dprintk("RPC:      xs_connect delayed xprt %p for %lu seconds\n",
1144                                 xprt, xprt->reestablish_timeout / HZ);
1145                 schedule_delayed_work(&xprt->connect_worker,
1146                                         xprt->reestablish_timeout);
1147                 xprt->reestablish_timeout <<= 1;
1148                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1149                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1150         } else {
1151                 dprintk("RPC:      xs_connect scheduled xprt %p\n", xprt);
1152                 schedule_work(&xprt->connect_worker);
1153
1154                 /* flush_scheduled_work can sleep... */
1155                 if (!RPC_IS_ASYNC(task))
1156                         flush_scheduled_work();
1157         }
1158 }
1159
1160 static struct rpc_xprt_ops xs_udp_ops = {
1161         .set_buffer_size        = xs_udp_set_buffer_size,
1162         .reserve_xprt           = xprt_reserve_xprt_cong,
1163         .release_xprt           = xprt_release_xprt_cong,
1164         .connect                = xs_connect,
1165         .buf_alloc              = rpc_malloc,
1166         .buf_free               = rpc_free,
1167         .send_request           = xs_udp_send_request,
1168         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
1169         .timer                  = xs_udp_timer,
1170         .release_request        = xprt_release_rqst_cong,
1171         .close                  = xs_close,
1172         .destroy                = xs_destroy,
1173 };
1174
1175 static struct rpc_xprt_ops xs_tcp_ops = {
1176         .reserve_xprt           = xprt_reserve_xprt,
1177         .release_xprt           = xprt_release_xprt,
1178         .connect                = xs_connect,
1179         .buf_alloc              = rpc_malloc,
1180         .buf_free               = rpc_free,
1181         .send_request           = xs_tcp_send_request,
1182         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
1183         .close                  = xs_close,
1184         .destroy                = xs_destroy,
1185 };
1186
1187 /**
1188  * xs_setup_udp - Set up transport to use a UDP socket
1189  * @xprt: transport to set up
1190  * @to:   timeout parameters
1191  *
1192  */
1193 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1194 {
1195         size_t slot_table_size;
1196
1197         dprintk("RPC:      setting up udp-ipv4 transport...\n");
1198
1199         xprt->max_reqs = xprt_udp_slot_table_entries;
1200         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1201         xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1202         if (xprt->slot == NULL)
1203                 return -ENOMEM;
1204         memset(xprt->slot, 0, slot_table_size);
1205
1206         xprt->prot = IPPROTO_UDP;
1207         xprt->port = xprt_max_resvport;
1208         xprt->tsh_size = 0;
1209         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1210         /* XXX: header size can vary due to auth type, IPv6, etc. */
1211         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1212
1213         INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1214         xprt->bind_timeout = XS_BIND_TO;
1215         xprt->connect_timeout = XS_UDP_CONN_TO;
1216         xprt->reestablish_timeout = XS_UDP_REEST_TO;
1217         xprt->idle_timeout = XS_IDLE_DISC_TO;
1218
1219         xprt->ops = &xs_udp_ops;
1220
1221         if (to)
1222                 xprt->timeout = *to;
1223         else
1224                 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1225
1226         return 0;
1227 }
1228
1229 /**
1230  * xs_setup_tcp - Set up transport to use a TCP socket
1231  * @xprt: transport to set up
1232  * @to: timeout parameters
1233  *
1234  */
1235 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1236 {
1237         size_t slot_table_size;
1238
1239         dprintk("RPC:      setting up tcp-ipv4 transport...\n");
1240
1241         xprt->max_reqs = xprt_tcp_slot_table_entries;
1242         slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1243         xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1244         if (xprt->slot == NULL)
1245                 return -ENOMEM;
1246         memset(xprt->slot, 0, slot_table_size);
1247
1248         xprt->prot = IPPROTO_TCP;
1249         xprt->port = xprt_max_resvport;
1250         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1251         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1252         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1253
1254         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1255         xprt->bind_timeout = XS_BIND_TO;
1256         xprt->connect_timeout = XS_TCP_CONN_TO;
1257         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1258         xprt->idle_timeout = XS_IDLE_DISC_TO;
1259
1260         xprt->ops = &xs_tcp_ops;
1261
1262         if (to)
1263                 xprt->timeout = *to;
1264         else
1265                 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1266
1267         return 0;
1268 }