net/sunrpc/xprt.c

   1 /*
   2  *  linux/net/sunrpc/xprt.c
   3  *
   4  *  This is a generic RPC call interface supporting congestion avoidance,
   5  *  and asynchronous calls.
   6  *
   7  *  The interface works like this:
   8  *
   9  *  -   When a process places a call, it allocates a request slot if
  10  *      one is available. Otherwise, it sleeps on the backlog queue
  11  *      (xprt_reserve).
  12  *  -   Next, the caller puts together the RPC message, stuffs it into
  13  *      the request struct, and calls xprt_call().
  14  *  -   xprt_call transmits the message and installs the caller on the
  15  *      socket's wait list. At the same time, it installs a timer that
  16  *      is run after the packet's timeout has expired.
  17  *  -   When a packet arrives, the data_ready handler walks the list of
  18  *      pending requests for that socket. If a matching XID is found, the
  19  *      caller is woken up, and the timer removed.
  20  *  -   When no reply arrives within the timeout interval, the timer is
  21  *      fired by the kernel and runs xprt_timer(). It either adjusts the
  22  *      timeout values (minor timeout) or wakes up the caller with a status
  23  *      of -ETIMEDOUT.
  24  *  -   When the caller receives a notification from RPC that a reply arrived,
  25  *      it should release the RPC slot, and process the reply.
  26  *      If the call timed out, it may choose to retry the operation by
  27  *      adjusting the initial timeout value, and simply calling rpc_call
  28  *      again.
  29  *
  30  *  Support for async RPC is done through a set of RPC-specific scheduling
  31  *  primitives that `transparently' work for processes as well as async
  32  *  tasks that rely on callbacks.
  33  *
  34  *  Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
  35  *
  36  *  TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
  37  *  TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
  38  *  TCP NFS related read + write fixes
  39  *   (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  40  *
  41  *  Rewrite of larges part of the code in order to stabilize TCP stuff.
  42  *  Fix behaviour when socket buffer is full.
  43  *   (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  44  */
  45
  46 #define __KERNEL_SYSCALLS__
  47
  48 #include <linux/version.h>
  49 #include <linux/types.h>
  50 #include <linux/slab.h>
  51 #include <linux/capability.h>
  52 #include <linux/sched.h>
  53 #include <linux/errno.h>
  54 #include <linux/socket.h>
  55 #include <linux/in.h>
  56 #include <linux/net.h>
  57 #include <linux/mm.h>
  58 #include <linux/udp.h>
  59 #include <linux/unistd.h>
  60 #include <linux/sunrpc/clnt.h>
  61 #include <linux/file.h>
  62
  63 #include <net/sock.h>
  64 #include <net/checksum.h>
  65 #include <net/udp.h>
  66 #include <net/tcp.h>
  67
  68 #include <asm/uaccess.h>
  69
  70 /*
  71  * Local variables
  72  */
  73
  74 #ifdef RPC_DEBUG
  75 # undef  RPC_DEBUG_DATA
  76 # define RPCDBG_FACILITY        RPCDBG_XPRT
  77 #endif
  78
  79 #define XPRT_MAX_BACKOFF        (8)
  80
  81 /*
  82  * Local functions
  83  */
  84 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  85 static void     do_xprt_transmit(struct rpc_task *);
  86 static inline void      do_xprt_reserve(struct rpc_task *);
  87 static void     xprt_disconnect(struct rpc_xprt *);
  88 static void     xprt_connect_status(struct rpc_task *task);
  89 static struct socket *xprt_create_socket(int, struct rpc_timeout *, int);
  90 static int      xprt_bind_socket(struct rpc_xprt *, struct socket *);
  91 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
  92
  93 #ifdef RPC_DEBUG_DATA
  94 /*
  95  * Print the buffer contents (first 128 bytes only--just enough for
  96  * diropres return).
  97  */
  98 static void
  99 xprt_pktdump(char *msg, u32 *packet, unsigned int count)
 100 {
 101         u8      *buf = (u8 *) packet;
 102         int     j;
 103
 104         dprintk("RPC:      %s\n", msg);
 105         for (j = 0; j < count && j < 128; j += 4) {
 106                 if (!(j & 31)) {
 107                         if (j)
 108                                 dprintk("\n");
 109                         dprintk("0x%04x ", j);
 110                 }
 111                 dprintk("%02x%02x%02x%02x ",
 112                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 113         }
 114         dprintk("\n");
 115 }
 116 #else
 117 static inline void
 118 xprt_pktdump(char *msg, u32 *packet, unsigned int count)
 119 {
 120         /* NOP */
 121 }
 122 #endif
 123
 124 /*
 125  * Look up RPC transport given an INET socket
 126  */
 127 static inline struct rpc_xprt *
 128 xprt_from_sock(struct sock *sk)
 129 {
 130         return (struct rpc_xprt *) sk->user_data;
 131 }
 132
 133 /*
 134  * Serialize write access to sockets, in order to prevent different
 135  * requests from interfering with each other.
 136  * Also prevents TCP socket connections from colliding with writes.
 137  */
 138 static int
 139 __xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 140 {
 141         struct rpc_rqst *req = task->tk_rqstp;
 142         if (!xprt->snd_task) {
 143                 if (xprt->nocong || __xprt_get_cong(xprt, task)) {
 144                         xprt->snd_task = task;
 145                         if (req) {
 146                                 req->rq_bytes_sent = 0;
 147                                 req->rq_ntrans++;
 148                         }
 149                 }
 150         }
 151         if (xprt->snd_task != task) {
 152                 dprintk("RPC: %4d TCP write queue full\n", task->tk_pid);
 153                 task->tk_timeout = 0;
 154                 task->tk_status = -EAGAIN;
 155                 if (req && req->rq_ntrans)
 156                         rpc_sleep_on(&xprt->resend, task, NULL, NULL);
 157                 else
 158                         rpc_sleep_on(&xprt->sending, task, NULL, NULL);
 159         }
 160         return xprt->snd_task == task;
 161 }
 162
 163 static inline int
 164 xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 165 {
 166         int retval;
 167         spin_lock_bh(&xprt->sock_lock);
 168         retval = __xprt_lock_write(xprt, task);
 169         spin_unlock_bh(&xprt->sock_lock);
 170         return retval;
 171 }
 172
 173 static void
 174 __xprt_lock_write_next(struct rpc_xprt *xprt)
 175 {
 176         struct rpc_task *task;
 177
 178         if (xprt->snd_task)
 179                 return;
 180         task = rpc_wake_up_next(&xprt->resend);
 181         if (!task) {
 182                 if (!xprt->nocong && RPCXPRT_CONGESTED(xprt))
 183                         return;
 184                 task = rpc_wake_up_next(&xprt->sending);
 185                 if (!task)
 186                         return;
 187         }
 188         if (xprt->nocong || __xprt_get_cong(xprt, task)) {
 189                 struct rpc_rqst *req = task->tk_rqstp;
 190                 xprt->snd_task = task;
 191                 if (req) {
 192                         req->rq_bytes_sent = 0;
 193                         req->rq_ntrans++;
 194                 }
 195         }
 196 }
 197
 198 /*
 199  * Releases the socket for use by other requests.
 200  */
 201 static void
 202 __xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 203 {
 204         if (xprt->snd_task == task)
 205                 xprt->snd_task = NULL;
 206         __xprt_lock_write_next(xprt);
 207 }
 208
 209 static inline void
 210 xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 211 {
 212         spin_lock_bh(&xprt->sock_lock);
 213         __xprt_release_write(xprt, task);
 214         spin_unlock_bh(&xprt->sock_lock);
 215 }
 216
 217 /*
 218  * Write data to socket.
 219  */
 220 static inline int
 221 xprt_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req)
 222 {
 223         struct socket   *sock = xprt->sock;
 224         struct msghdr   msg;
 225         struct xdr_buf  *xdr = &req->rq_snd_buf;
 226         struct iovec    niv[MAX_IOVEC];
 227         unsigned int    niov, slen, skip;
 228         mm_segment_t    oldfs;
 229         int             result;
 230
 231         if (!sock)
 232                 return -ENOTCONN;
 233
 234         xprt_pktdump("packet data:",
 235                                 req->rq_svec->iov_base,
 236                                 req->rq_svec->iov_len);
 237
 238         /* Dont repeat bytes */
 239         skip = req->rq_bytes_sent;
 240         slen = xdr->len - skip;
 241         oldfs = get_fs(); set_fs(get_ds());
 242         do {
 243                 unsigned int slen_part, n;
 244
 245                 niov = xdr_kmap(niv, xdr, skip);
 246                 if (!niov) {
 247                         result = -EAGAIN;
 248                         break;
 249                 }
 250
 251                 msg.msg_flags   = MSG_DONTWAIT|MSG_NOSIGNAL;
 252                 msg.msg_iov     = niv;
 253                 msg.msg_iovlen  = niov;
 254                 msg.msg_name    = (struct sockaddr *) &xprt->addr;
 255                 msg.msg_namelen = sizeof(xprt->addr);
 256                 msg.msg_control = NULL;
 257                 msg.msg_controllen = 0;
 258
 259                 slen_part = 0;
 260                 for (n = 0; n < niov; n++)
 261                         slen_part += niv[n].iov_len;
 262
 263                 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 264                 result = sock_sendmsg(sock, &msg, slen_part);
 265
 266                 xdr_kunmap(xdr, skip, niov);
 267
 268                 skip += slen_part;
 269                 slen -= slen_part;
 270         } while (result >= 0 && slen);
 271         set_fs(oldfs);
 272
 273         dprintk("RPC:      xprt_sendmsg(%d) = %d\n", slen, result);
 274
 275         if (result >= 0)
 276                 return result;
 277
 278         switch (result) {
 279         case -ECONNREFUSED:
 280                 /* When the server has died, an ICMP port unreachable message
 281                  * prompts ECONNREFUSED.
 282                  */
 283         case -EAGAIN:
 284                 break;
 285         case -ECONNRESET:
 286         case -ENOTCONN:
 287         case -EPIPE:
 288                 /* connection broken */
 289                 if (xprt->stream)
 290                         result = -ENOTCONN;
 291                 break;
 292         default:
 293                 printk(KERN_NOTICE "RPC: sendmsg returned error %d\n", -result);
 294         }
 295         return result;
 296 }
 297
 298 /*
 299  * Van Jacobson congestion avoidance. Check if the congestion window
 300  * overflowed. Put the task to sleep if this is the case.
 301  */
 302 static int
 303 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 304 {
 305         struct rpc_rqst *req = task->tk_rqstp;
 306
 307         if (req->rq_cong)
 308                 return 1;
 309         dprintk("RPC: %4d xprt_cwnd_limited cong = %ld cwnd = %ld\n",
 310                         task->tk_pid, xprt->cong, xprt->cwnd);
 311         if (RPCXPRT_CONGESTED(xprt))
 312                 return 0;
 313         req->rq_cong = 1;
 314         xprt->cong += RPC_CWNDSCALE;
 315         return 1;
 316 }
 317
 318 /*
 319  * Adjust the congestion window, and wake up the next task
 320  * that has been sleeping due to congestion
 321  */
 322 static void
 323 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 324 {
 325         if (!req->rq_cong)
 326                 return;
 327         req->rq_cong = 0;
 328         xprt->cong -= RPC_CWNDSCALE;
 329         __xprt_lock_write_next(xprt);
 330 }
 331
 332 /*
 333  * Adjust RPC congestion window
 334  * We use a time-smoothed congestion estimator to avoid heavy oscillation.
 335  */
 336 static void
 337 xprt_adjust_cwnd(struct rpc_xprt *xprt, int result)
 338 {
 339         unsigned long   cwnd;
 340
 341         cwnd = xprt->cwnd;
 342         if (result >= 0 && cwnd <= xprt->cong) {
 343                 /* The (cwnd >> 1) term makes sure
 344                  * the result gets rounded properly. */
 345                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 346                 if (cwnd > RPC_MAXCWND)
 347                         cwnd = RPC_MAXCWND;
 348                 __xprt_lock_write_next(xprt);
 349         } else if (result == -ETIMEDOUT) {
 350                 cwnd >>= 1;
 351                 if (cwnd < RPC_CWNDSCALE)
 352                         cwnd = RPC_CWNDSCALE;
 353         }
 354         dprintk("RPC:      cong %ld, cwnd was %ld, now %ld\n",
 355                         xprt->cong, xprt->cwnd, cwnd);
 356         xprt->cwnd = cwnd;
 357 }
 358
 359 /*
 360  * Adjust timeout values etc for next retransmit
 361  */
 362 int
 363 xprt_adjust_timeout(struct rpc_timeout *to)
 364 {
 365         if (to->to_retries > 0) {
 366                 if (to->to_exponential)
 367                         to->to_current <<= 1;
 368                 else
 369                         to->to_current += to->to_increment;
 370                 if (to->to_maxval && to->to_current >= to->to_maxval)
 371                         to->to_current = to->to_maxval;
 372         } else {
 373                 if (to->to_exponential)
 374                         to->to_initval <<= 1;
 375                 else
 376                         to->to_initval += to->to_increment;
 377                 if (to->to_maxval && to->to_initval >= to->to_maxval)
 378                         to->to_initval = to->to_maxval;
 379                 to->to_current = to->to_initval;
 380         }
 381
 382         if (!to->to_current) {
 383                 printk(KERN_WARNING "xprt_adjust_timeout: to_current = 0!\n");
 384                 to->to_current = 5 * HZ;
 385         }
 386         pprintk("RPC: %lu %s\n", jiffies,
 387                         to->to_retries? "retrans" : "timeout");
 388         return to->to_retries-- > 0;
 389 }
 390
 391 /*
 392  * Close down a transport socket
 393  */
 394 static void
 395 xprt_close(struct rpc_xprt *xprt)
 396 {
 397         struct socket   *sock = xprt->sock;
 398         struct sock     *sk = xprt->inet;
 399
 400         if (!sk)
 401                 return;
 402
 403         write_lock_bh(&sk->callback_lock);
 404         xprt->inet = NULL;
 405         xprt->sock = NULL;
 406
 407         sk->user_data    = NULL;
 408         sk->data_ready   = xprt->old_data_ready;
 409         sk->state_change = xprt->old_state_change;
 410         sk->write_space  = xprt->old_write_space;
 411         write_unlock_bh(&sk->callback_lock);
 412
 413         xprt_disconnect(xprt);
 414         sk->no_check     = 0;
 415
 416         sock_release(sock);
 417 }
 418
 419 /*
 420  * Mark a transport as disconnected
 421  */
 422 static void
 423 xprt_disconnect(struct rpc_xprt *xprt)
 424 {
 425         dprintk("RPC:      disconnected transport %p\n", xprt);
 426         spin_lock_bh(&xprt->sock_lock);
 427         xprt_clear_connected(xprt);
 428         rpc_wake_up_status(&xprt->pending, -ENOTCONN);
 429         spin_unlock_bh(&xprt->sock_lock);
 430 }
 431
 432 /*
 433  * Reconnect a broken TCP connection.
 434  *
 435  */
 436 void
 437 xprt_connect(struct rpc_task *task)
 438 {
 439         struct rpc_xprt *xprt = task->tk_xprt;
 440         struct socket   *sock = xprt->sock;
 441         struct sock     *inet;
 442         int             status;
 443
 444         dprintk("RPC: %4d xprt_connect %p connected %d\n",
 445                                 task->tk_pid, xprt, xprt_connected(xprt));
 446         if (xprt->shutdown)
 447                 return;
 448
 449         if (!xprt->addr.sin_port) {
 450                 task->tk_status = -EIO;
 451                 return;
 452         }
 453
 454         if (!xprt_lock_write(xprt, task))
 455                 return;
 456         if (xprt_connected(xprt))
 457                 goto out_write;
 458
 459         if (task->tk_rqstp)
 460                 task->tk_rqstp->rq_bytes_sent = 0;
 461
 462         xprt_close(xprt);
 463         /* Create an unconnected socket */
 464         sock = xprt_create_socket(xprt->prot, &xprt->timeout, xprt->resvport);
 465         if (!sock) {
 466                 /* couldn't create socket or bind to reserved port;
 467                  * this is likely a permanent error, so cause an abort */
 468                 task->tk_status = -EIO;
 469                 goto out_write;
 470         }
 471         xprt_bind_socket(xprt, sock);
 472
 473         if (!xprt->stream)
 474                 goto out_write;
 475
 476         inet = sock->sk;
 477
 478         /* Now connect it asynchronously. */
 479         dprintk("RPC: %4d connecting new socket\n", task->tk_pid);
 480         status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
 481                                 sizeof(xprt->addr), O_NONBLOCK);
 482         dprintk("RPC: %4d connect status %d connected %d\n",
 483                 task->tk_pid, status, xprt_connected(xprt));
 484
 485         if (status >= 0)
 486                 return;
 487
 488         switch (status) {
 489         case -EALREADY:
 490         case -EINPROGRESS:
 491                 /* Protect against TCP socket state changes */
 492                 lock_sock(inet);
 493                 if (inet->state != TCP_ESTABLISHED) {
 494                         dprintk("RPC: %4d  waiting for connection\n",
 495                                         task->tk_pid);
 496                         task->tk_timeout = RPC_CONNECT_TIMEOUT;
 497                         /* if the socket is already closing, delay briefly */
 498                         if ((1<<inet->state) & ~(TCPF_SYN_SENT|TCPF_SYN_RECV))
 499                                 task->tk_timeout = RPC_REESTABLISH_TIMEOUT;
 500                         rpc_sleep_on(&xprt->pending, task, xprt_connect_status,
 501                                         NULL);
 502                 }
 503                 release_sock(inet);
 504                 break;
 505         case -ECONNREFUSED:
 506         case -ECONNRESET:
 507         case -ENOTCONN:
 508                 if (!task->tk_client->cl_softrtry) {
 509                         rpc_delay(task, RPC_REESTABLISH_TIMEOUT);
 510                         task->tk_status = -ENOTCONN;
 511                         break;
 512                 }
 513         default:
 514                 /* Report myriad other possible returns.  If this file
 515                  * system is soft mounted, just error out, like Solaris.  */
 516                 if (task->tk_client->cl_softrtry) {
 517                         printk(KERN_WARNING
 518                                         "RPC: error %d connecting to server %s, exiting\n",
 519                                         -status, task->tk_client->cl_server);
 520                         task->tk_status = -EIO;
 521                         goto out_write;
 522                 }
 523                 printk(KERN_WARNING "RPC: error %d connecting to server %s\n",
 524                                 -status, task->tk_client->cl_server);
 525                 /* This will prevent anybody else from connecting */
 526                 rpc_delay(task, RPC_REESTABLISH_TIMEOUT);
 527                 task->tk_status = status;
 528                 break;
 529         }
 530         return;
 531  out_write:
 532         xprt_release_write(xprt, task);
 533 }
 534
 535 /*
 536  * We arrive here when awoken from waiting on connection establishment.
 537  */
 538 static void
 539 xprt_connect_status(struct rpc_task *task)
 540 {
 541         struct rpc_xprt *xprt = task->tk_xprt;
 542
 543         if (task->tk_status >= 0) {
 544                 dprintk("RPC: %4d xprt_connect_status: connection established\n",
 545                                 task->tk_pid);
 546                 return;
 547         }
 548
 549         /* if soft mounted, cause this RPC to fail */
 550         if (task->tk_client->cl_softrtry)
 551                 task->tk_status = -EIO;
 552
 553         switch (task->tk_status) {
 554         case -ENOTCONN:
 555                 rpc_delay(task, RPC_REESTABLISH_TIMEOUT);
 556                 return;
 557         case -ETIMEDOUT:
 558                 dprintk("RPC: %4d xprt_connect_status: timed out\n",
 559                                 task->tk_pid);
 560                 break;
 561         default:
 562                 printk(KERN_ERR "RPC: error %d connecting to server %s\n",
 563                                 -task->tk_status, task->tk_client->cl_server);
 564         }
 565         xprt_release_write(xprt, task);
 566 }
 567
 568 /*
 569  * Look up the RPC request corresponding to a reply, and then lock it.
 570  */
 571 static inline struct rpc_rqst *
 572 xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
 573 {
 574         struct list_head *pos;
 575         struct rpc_rqst *req = NULL;
 576
 577         list_for_each(pos, &xprt->recv) {
 578                 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
 579                 if (entry->rq_xid == xid) {
 580                         req = entry;
 581                         break;
 582                 }
 583         }
 584         return req;
 585 }
 586
 587 /*
 588  * Complete reply received.
 589  * The TCP code relies on us to remove the request from xprt->pending.
 590  */
 591 static void
 592 xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
 593 {
 594         struct rpc_task *task = req->rq_task;
 595         struct rpc_clnt *clnt = task->tk_client;
 596
 597         /* Adjust congestion window */
 598         if (!xprt->nocong) {
 599                 int timer = rpcproc_timer(clnt, task->tk_msg.rpc_proc);
 600                 xprt_adjust_cwnd(xprt, copied);
 601                 __xprt_put_cong(xprt, req);
 602                 if (req->rq_ntrans == 1) {
 603                         if (timer)
 604                                 rpc_update_rtt(&clnt->cl_rtt, timer, (long)jiffies - req->rq_xtime);
 605                 }
 606                 rpc_set_timeo(&clnt->cl_rtt, timer, req->rq_ntrans - 1);
 607         }
 608
 609 #ifdef RPC_PROFILE
 610         /* Profile only reads for now */
 611         if (copied > 1024) {
 612                 static unsigned long    nextstat = 0;
 613                 static unsigned long    pkt_rtt = 0, pkt_len = 0, pkt_cnt = 0;
 614
 615                 pkt_cnt++;
 616                 pkt_len += req->rq_slen + copied;
 617                 pkt_rtt += jiffies - req->rq_xtime;
 618                 if (time_before(nextstat, jiffies)) {
 619                         printk("RPC: %lu %ld cwnd\n", jiffies, xprt->cwnd);
 620                         printk("RPC: %ld %ld %ld %ld stat\n",
 621                                         jiffies, pkt_cnt, pkt_len, pkt_rtt);
 622                         pkt_rtt = pkt_len = pkt_cnt = 0;
 623                         nextstat = jiffies + 5 * HZ;
 624                 }
 625         }
 626 #endif
 627
 628         dprintk("RPC: %4d has input (%d bytes)\n", task->tk_pid, copied);
 629         req->rq_received = copied;
 630         list_del_init(&req->rq_list);
 631
 632         /* ... and wake up the process. */
 633         rpc_wake_up_task(task);
 634         return;
 635 }
 636
 637 static size_t
 638 skb_read_bits(skb_reader_t *desc, void *to, size_t len)
 639 {
 640         if (len > desc->count)
 641                 len = desc->count;
 642         skb_copy_bits(desc->skb, desc->offset, to, len);
 643         desc->count -= len;
 644         desc->offset += len;
 645         return len;
 646 }
 647
 648 static size_t
 649 skb_read_and_csum_bits(skb_reader_t *desc, void *to, size_t len)
 650 {
 651         unsigned int csum2, pos;
 652
 653         if (len > desc->count)
 654                 len = desc->count;
 655         pos = desc->offset;
 656         csum2 = skb_copy_and_csum_bits(desc->skb, pos, to, len, 0);
 657         desc->csum = csum_block_add(desc->csum, csum2, pos);
 658         desc->count -= len;
 659         desc->offset += len;
 660         return len;
 661 }
 662
 663 /*
 664  * We have set things up such that we perform the checksum of the UDP
 665  * packet in parallel with the copies into the RPC client iovec.  -DaveM
 666  */
 667 static int
 668 csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 669 {
 670         skb_reader_t desc;
 671
 672         desc.skb = skb;
 673         desc.offset = sizeof(struct udphdr);
 674         desc.count = skb->len - desc.offset;
 675
 676         if (skb->ip_summed == CHECKSUM_UNNECESSARY)
 677                 goto no_checksum;
 678
 679         desc.csum = csum_partial(skb->data, desc.offset, skb->csum);
 680         xdr_partial_copy_from_skb(xdr, 0, &desc, skb_read_and_csum_bits);
 681         if (desc.offset != skb->len) {
 682                 unsigned int csum2;
 683                 csum2 = skb_checksum(skb, desc.offset, skb->len - desc.offset, 0);
 684                 desc.csum = csum_block_add(desc.csum, csum2, desc.offset);
 685         }
 686         if ((unsigned short)csum_fold(desc.csum))
 687                 return -1;
 688         return 0;
 689 no_checksum:
 690         xdr_partial_copy_from_skb(xdr, 0, &desc, skb_read_bits);
 691         return 0;
 692 }
 693
 694 /*
 695  * Input handler for RPC replies. Called from a bottom half and hence
 696  * atomic.
 697  */
 698 static void
 699 udp_data_ready(struct sock *sk, int len)
 700 {
 701         struct rpc_task *task;
 702         struct rpc_xprt *xprt;
 703         struct rpc_rqst *rovr;
 704         struct sk_buff  *skb;
 705         int             err, repsize, copied;
 706
 707         read_lock(&sk->callback_lock);
 708         dprintk("RPC:      udp_data_ready...\n");
 709         if (sk->dead || !(xprt = xprt_from_sock(sk))) {
 710                 printk("RPC:      udp_data_ready request not found!\n");
 711                 goto out;
 712         }
 713
 714         dprintk("RPC:      udp_data_ready client %p\n", xprt);
 715
 716         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 717                 goto out;
 718
 719         if (xprt->shutdown)
 720                 goto dropit;
 721
 722         repsize = skb->len - sizeof(struct udphdr);
 723         if (repsize < 4) {
 724                 printk("RPC: impossible RPC reply size %d!\n", repsize);
 725                 goto dropit;
 726         }
 727
 728         /* Look up and lock the request corresponding to the given XID */
 729         spin_lock(&xprt->sock_lock);
 730         rovr = xprt_lookup_rqst(xprt, *(u32 *) (skb->h.raw + sizeof(struct udphdr)));
 731         if (!rovr)
 732                 goto out_unlock;
 733         task = rovr->rq_task;
 734
 735         dprintk("RPC: %4d received reply\n", task->tk_pid);
 736         xprt_pktdump("packet data:",
 737                      (u32 *) (skb->h.raw+sizeof(struct udphdr)), repsize);
 738
 739         if ((copied = rovr->rq_private_buf.len) > repsize)
 740                 copied = repsize;
 741
 742         /* Suck it into the iovec, verify checksum if not done by hw. */
 743         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
 744                 goto out_unlock;
 745
 746         /* Something worked... */
 747         dst_confirm(skb->dst);
 748
 749         xprt_complete_rqst(xprt, rovr, copied);
 750
 751  out_unlock:
 752         spin_unlock(&xprt->sock_lock);
 753  dropit:
 754         skb_free_datagram(sk, skb);
 755  out:
 756         if (sk->sleep && waitqueue_active(sk->sleep))
 757                 wake_up_interruptible(sk->sleep);
 758         read_unlock(&sk->callback_lock);
 759 }
 760
 761 /*
 762  * Copy from an skb into memory and shrink the skb.
 763  */
 764 static inline size_t
 765 tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
 766 {
 767         if (len > desc->count)
 768                 len = desc->count;
 769         skb_copy_bits(desc->skb, desc->offset, p, len);
 770         desc->offset += len;
 771         desc->count -= len;
 772         return len;
 773 }
 774
 775 /*
 776  * TCP read fragment marker
 777  */
 778 static inline void
 779 tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
 780 {
 781         size_t len, used;
 782         char *p;
 783
 784         p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
 785         len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
 786         used = tcp_copy_data(desc, p, len);
 787         xprt->tcp_offset += used;
 788         if (used != len)
 789                 return;
 790         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
 791         if (xprt->tcp_reclen & 0x80000000)
 792                 xprt->tcp_flags |= XPRT_LAST_FRAG;
 793         else
 794                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
 795         xprt->tcp_reclen &= 0x7fffffff;
 796         xprt->tcp_flags &= ~XPRT_COPY_RECM;
 797         xprt->tcp_offset = 0;
 798         /* Sanity check of the record length */
 799         if (xprt->tcp_reclen < 4) {
 800                 printk(KERN_ERR "RPC: Invalid TCP record fragment length\n");
 801                 xprt_disconnect(xprt);
 802         }
 803         dprintk("RPC:      reading TCP record fragment of length %d\n",
 804                         xprt->tcp_reclen);
 805 }
 806
 807 static void
 808 tcp_check_recm(struct rpc_xprt *xprt)
 809 {
 810         if (xprt->tcp_offset == xprt->tcp_reclen) {
 811                 xprt->tcp_flags |= XPRT_COPY_RECM;
 812                 xprt->tcp_offset = 0;
 813                 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
 814                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 815                         xprt->tcp_flags |= XPRT_COPY_XID;
 816                         xprt->tcp_copied = 0;
 817                 }
 818         }
 819 }
 820
 821 /*
 822  * TCP read xid
 823  */
 824 static inline void
 825 tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
 826 {
 827         size_t len, used;
 828         char *p;
 829
 830         len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
 831         dprintk("RPC:      reading XID (%Zu bytes)\n", len);
 832         p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
 833         used = tcp_copy_data(desc, p, len);
 834         xprt->tcp_offset += used;
 835         if (used != len)
 836                 return;
 837         xprt->tcp_flags &= ~XPRT_COPY_XID;
 838         xprt->tcp_flags |= XPRT_COPY_DATA;
 839         xprt->tcp_copied = 4;
 840         dprintk("RPC:      reading reply for XID %08x\n", xprt->tcp_xid);
 841         tcp_check_recm(xprt);
 842 }
 843
 844 /*
 845  * TCP read and complete request
 846  */
 847 static inline void
 848 tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
 849 {
 850         struct rpc_rqst *req;
 851         struct xdr_buf *rcvbuf;
 852         size_t len;
 853
 854         /* Find and lock the request corresponding to this xid */
 855         spin_lock(&xprt->sock_lock);
 856         req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
 857         if (!req) {
 858                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 859                 dprintk("RPC:      XID %08x request not found!\n",
 860                                 xprt->tcp_xid);
 861                 spin_unlock(&xprt->sock_lock);
 862                 return;
 863         }
 864
 865         rcvbuf = &req->rq_private_buf;
 866         len = desc->count;
 867         if (len > xprt->tcp_reclen - xprt->tcp_offset) {
 868                 skb_reader_t my_desc;
 869
 870                 len = xprt->tcp_reclen - xprt->tcp_offset;
 871                 memcpy(&my_desc, desc, sizeof(my_desc));
 872                 my_desc.count = len;
 873                 xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 874                                           &my_desc, tcp_copy_data);
 875                 desc->count -= len;
 876                 desc->offset += len;
 877         } else
 878                 xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
 879                                           desc, tcp_copy_data);
 880         xprt->tcp_copied += len;
 881         xprt->tcp_offset += len;
 882
 883         if (xprt->tcp_copied == req->rq_private_buf.len)
 884                 xprt->tcp_flags &= ~XPRT_COPY_DATA;
 885         else if (xprt->tcp_offset == xprt->tcp_reclen) {
 886                 if (xprt->tcp_flags & XPRT_LAST_FRAG)
 887                         xprt->tcp_flags &= ~XPRT_COPY_DATA;
 888         }
 889
 890         if (!(xprt->tcp_flags & XPRT_COPY_DATA)) {
 891                 dprintk("RPC: %4d received reply complete\n",
 892                                 req->rq_task->tk_pid);
 893                 xprt_complete_rqst(xprt, req, xprt->tcp_copied);
 894         }
 895         spin_unlock(&xprt->sock_lock);
 896         tcp_check_recm(xprt);
 897 }
 898
 899 /*
 900  * TCP discard extra bytes from a short read
 901  */
 902 static inline void
 903 tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
 904 {
 905         size_t len;
 906
 907         len = xprt->tcp_reclen - xprt->tcp_offset;
 908         if (len > desc->count)
 909                 len = desc->count;
 910         desc->count -= len;
 911         desc->offset += len;
 912         xprt->tcp_offset += len;
 913         tcp_check_recm(xprt);
 914 }
 915
 916 /*
 917  * TCP record receive routine
 918  * We first have to grab the record marker, then the XID, then the data.
 919  */
 920 static int
 921 tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
 922                 unsigned int offset, size_t len)
 923 {
 924         struct rpc_xprt *xprt = (struct rpc_xprt *)rd_desc->buf;
 925         skb_reader_t desc = { skb, offset, len };
 926
 927         dprintk("RPC:      tcp_data_recv\n");
 928         do {
 929                 /* Read in a new fragment marker if necessary */
 930                 /* Can we ever really expect to get completely empty fragments? */
 931                 if (xprt->tcp_flags & XPRT_COPY_RECM) {
 932                         tcp_read_fraghdr(xprt, &desc);
 933                         continue;
 934                 }
 935                 /* Read in the xid if necessary */
 936                 if (xprt->tcp_flags & XPRT_COPY_XID) {
 937                         tcp_read_xid(xprt, &desc);
 938                         continue;
 939                 }
 940                 /* Read in the request data */
 941                 if (xprt->tcp_flags & XPRT_COPY_DATA) {
 942                         tcp_read_request(xprt, &desc);
 943                         continue;
 944                 }
 945                 /* Skip over any trailing bytes on short reads */
 946                 tcp_read_discard(xprt, &desc);
 947         } while (desc.count);
 948         dprintk("RPC:      tcp_data_recv done\n");
 949         return len - desc.count;
 950 }
 951
 952 static void tcp_data_ready(struct sock *sk, int bytes)
 953 {
 954         struct rpc_xprt *xprt;
 955         read_descriptor_t rd_desc;
 956
 957         read_lock(&sk->callback_lock);
 958         dprintk("RPC:      tcp_data_ready...\n");
 959         if (!(xprt = xprt_from_sock(sk))) {
 960                 printk("RPC:      tcp_data_ready socket info not found!\n");
 961                 goto out;
 962         }
 963         if (xprt->shutdown)
 964                 goto out;
 965
 966         /* We use rd_desc to pass struct xprt to tcp_data_recv */
 967         rd_desc.buf = (char *)xprt;
 968         rd_desc.count = 65536;
 969         tcp_read_sock(sk, &rd_desc, tcp_data_recv);
 970 out:
 971         read_unlock(&sk->callback_lock);
 972 }
 973
 974 static void
 975 tcp_state_change(struct sock *sk)
 976 {
 977         struct rpc_xprt *xprt;
 978
 979         read_lock(&sk->callback_lock);
 980         if (!(xprt = xprt_from_sock(sk)))
 981                 goto out;
 982         dprintk("RPC:      tcp_state_change client %p...\n", xprt);
 983         dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
 984                                 sk->state, xprt_connected(xprt),
 985                                 sk->dead, sk->zapped);
 986
 987         switch (sk->state) {
 988         case TCP_ESTABLISHED:
 989                 if (xprt_test_and_set_connected(xprt))
 990                         break;
 991
 992                 /* Reset TCP record info */
 993                 xprt->tcp_offset = 0;
 994                 xprt->tcp_reclen = 0;
 995                 xprt->tcp_copied = 0;
 996                 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
 997
 998                 spin_lock_bh(&xprt->sock_lock);
 999                 if (xprt->snd_task && xprt->snd_task->tk_rpcwait == &xprt->pending)
1000                         rpc_wake_up_task(xprt->snd_task);
1001                 spin_unlock_bh(&xprt->sock_lock);
1002                 break;
1003         case TCP_SYN_SENT:
1004         case TCP_SYN_RECV:
1005                 break;
1006         default:
1007                 xprt_disconnect(xprt);
1008                 break;
1009         }
1010  out:
1011         if (sk->sleep && waitqueue_active(sk->sleep))
1012                 wake_up_interruptible_all(sk->sleep);
1013         read_unlock(&sk->callback_lock);
1014 }
1015
1016 /*
1017  * Called when more output buffer space is available for this socket.
1018  * We try not to wake our writers until they can make "significant"
1019  * progress, otherwise we'll waste resources thrashing sock_sendmsg
1020  * with a bunch of small requests.
1021  */
1022 static void
1023 xprt_write_space(struct sock *sk)
1024 {
1025         struct rpc_xprt *xprt;
1026         struct socket   *sock;
1027
1028         read_lock(&sk->callback_lock);
1029         if (!(xprt = xprt_from_sock(sk)) || !(sock = sk->socket))
1030                 goto out;
1031         if (xprt->shutdown)
1032                 goto out;
1033
1034         /* Wait until we have enough socket memory */
1035         if (xprt->stream) {
1036                 /* from net/ipv4/tcp.c:tcp_write_space */
1037                 if (tcp_wspace(sk) < tcp_min_write_space(sk))
1038                         goto out;
1039         } else {
1040                 /* from net/core/sock.c:sock_def_write_space */
1041                 if (!sock_writeable(sk))
1042                         goto out;
1043         }
1044
1045         if (!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))
1046                 goto out;
1047
1048         spin_lock_bh(&xprt->sock_lock);
1049         if (xprt->snd_task && xprt->snd_task->tk_rpcwait == &xprt->pending)
1050                 rpc_wake_up_task(xprt->snd_task);
1051         spin_unlock_bh(&xprt->sock_lock);
1052         if (sk->sleep && waitqueue_active(sk->sleep))
1053                 wake_up_interruptible(sk->sleep);
1054 out:
1055         read_unlock(&sk->callback_lock);
1056 }
1057
1058 /*
1059  * RPC receive timeout handler.
1060  */
1061 static void
1062 xprt_timer(struct rpc_task *task)
1063 {
1064         struct rpc_rqst *req = task->tk_rqstp;
1065         struct rpc_xprt *xprt = req->rq_xprt;
1066
1067         spin_lock(&xprt->sock_lock);
1068         if (req->rq_received)
1069                 goto out;
1070
1071         xprt_adjust_cwnd(req->rq_xprt, -ETIMEDOUT);
1072         __xprt_put_cong(xprt, req);
1073
1074         dprintk("RPC: %4d xprt_timer (%s request)\n",
1075                 task->tk_pid, req ? "pending" : "backlogged");
1076
1077         task->tk_status  = -ETIMEDOUT;
1078 out:
1079         task->tk_timeout = 0;
1080         rpc_wake_up_task(task);
1081         spin_unlock(&xprt->sock_lock);
1082 }
1083
1084 /*
1085  * Place the actual RPC call.
1086  * We have to copy the iovec because sendmsg fiddles with its contents.
1087  */
1088 void
1089 xprt_transmit(struct rpc_task *task)
1090 {
1091         struct rpc_rqst *req = task->tk_rqstp;
1092         struct rpc_xprt *xprt = req->rq_xprt;
1093
1094         dprintk("RPC: %4d xprt_transmit(%x)\n", task->tk_pid,
1095                                 *(u32 *)(req->rq_svec[0].iov_base));
1096
1097         if (xprt->shutdown)
1098                 task->tk_status = -EIO;
1099
1100         if (task->tk_status < 0)
1101                 return;
1102
1103         if (task->tk_rpcwait)
1104                 rpc_remove_wait_queue(task);
1105
1106         /* set up everything as needed. */
1107         /* Write the record marker */
1108         if (xprt->stream) {
1109                 u32     *marker = req->rq_svec[0].iov_base;
1110
1111                 *marker = htonl(0x80000000|(req->rq_slen-sizeof(*marker)));
1112         }
1113
1114         spin_lock_bh(&xprt->sock_lock);
1115         if (req->rq_received != 0 && !req->rq_bytes_sent)
1116                 goto out_notrans;
1117
1118         if (!__xprt_lock_write(xprt, task))
1119                 goto out_notrans;
1120
1121         if (!xprt_connected(xprt)) {
1122                 task->tk_status = -ENOTCONN;
1123                 goto out_notrans;
1124         }
1125
1126         if (list_empty(&req->rq_list)) {
1127                 /* Update the softirq receive buffer */
1128                 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1129                                 sizeof(req->rq_private_buf));
1130                 list_add_tail(&req->rq_list, &xprt->recv);
1131         }
1132         spin_unlock_bh(&xprt->sock_lock);
1133
1134         do_xprt_transmit(task);
1135         return;
1136 out_notrans:
1137         spin_unlock_bh(&xprt->sock_lock);
1138 }
1139
1140 static void
1141 do_xprt_transmit(struct rpc_task *task)
1142 {
1143         struct rpc_clnt *clnt = task->tk_client;
1144         struct rpc_rqst *req = task->tk_rqstp;
1145         struct rpc_xprt *xprt = req->rq_xprt;
1146         int status, retry = 0;
1147
1148
1149         /* Continue transmitting the packet/record. We must be careful
1150          * to cope with writespace callbacks arriving _after_ we have
1151          * called xprt_sendmsg().
1152          */
1153         while (1) {
1154                 req->rq_xtime = jiffies;
1155                 status = xprt_sendmsg(xprt, req);
1156
1157                 if (status < 0)
1158                         break;
1159
1160                 if (xprt->stream) {
1161                         req->rq_bytes_sent += status;
1162
1163                         /* If we've sent the entire packet, immediately
1164                          * reset the count of bytes sent. */
1165                         if (req->rq_bytes_sent >= req->rq_slen) {
1166                                 req->rq_bytes_sent = 0;
1167                                 goto out_receive;
1168                         }
1169                 } else {
1170                         if (status >= req->rq_slen)
1171                                 goto out_receive;
1172                         status = -EAGAIN;
1173                         break;
1174                 }
1175
1176                 dprintk("RPC: %4d xmit incomplete (%d left of %d)\n",
1177                                 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
1178                                 req->rq_slen);
1179
1180                 status = -EAGAIN;
1181                 if (retry++ > 50)
1182                         break;
1183         }
1184
1185         /* If we're doing a resend and have received a reply already,
1186          * then exit early.
1187          * Note, though, that we can't do this if we've already started
1188          * resending down a TCP stream.
1189          */
1190         task->tk_status = status;
1191
1192         switch (status) {
1193         case -EAGAIN:
1194                 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
1195                         /* Protect against races with xprt_write_space */
1196                         spin_lock_bh(&xprt->sock_lock);
1197                         /* Don't race with disconnect */
1198                         if (!xprt_connected(xprt))
1199                                 task->tk_status = -ENOTCONN;
1200                         else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags)) {
1201                                 task->tk_timeout = req->rq_timeout.to_current;
1202                                 rpc_sleep_on(&xprt->pending, task, NULL, NULL);
1203                         }
1204                         spin_unlock_bh(&xprt->sock_lock);
1205                         return;
1206                 }
1207                 /* Keep holding the socket if it is blocked */
1208                 rpc_delay(task, HZ>>4);
1209                 return;
1210         case -ECONNREFUSED:
1211                 task->tk_timeout = RPC_REESTABLISH_TIMEOUT;
1212                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
1213         case -ENOTCONN:
1214                 return;
1215         default:
1216                 if (xprt->stream)
1217                         xprt_disconnect(xprt);
1218         }
1219         xprt_release_write(xprt, task);
1220         return;
1221  out_receive:
1222         dprintk("RPC: %4d xmit complete\n", task->tk_pid);
1223         spin_lock_bh(&xprt->sock_lock);
1224         /* Set the task's receive timeout value */
1225         if (!xprt->nocong) {
1226                 int timer = rpcproc_timer(clnt, task->tk_msg.rpc_proc);
1227                 task->tk_timeout = rpc_calc_rto(&clnt->cl_rtt, timer);
1228                 task->tk_timeout <<= rpc_ntimeo(&clnt->cl_rtt, timer);
1229                 task->tk_timeout <<= clnt->cl_timeout.to_retries
1230                         - req->rq_timeout.to_retries;
1231                 if (task->tk_timeout > req->rq_timeout.to_maxval)
1232                         task->tk_timeout = req->rq_timeout.to_maxval;
1233         } else
1234                 task->tk_timeout = req->rq_timeout.to_current;
1235         /* Don't race with disconnect */
1236         if (!xprt_connected(xprt))
1237                 task->tk_status = -ENOTCONN;
1238         else if (!req->rq_received)
1239                 rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer);
1240         __xprt_release_write(xprt, task);
1241         spin_unlock_bh(&xprt->sock_lock);
1242 }
1243
1244 /*
1245  * Reserve an RPC call slot.
1246  */
1247 static inline void
1248 do_xprt_reserve(struct rpc_task *task)
1249 {
1250         struct rpc_xprt *xprt = task->tk_xprt;
1251
1252         task->tk_status = 0;
1253         if (task->tk_rqstp)
1254                 return;
1255         if (xprt->free) {
1256                 struct rpc_rqst *req = xprt->free;
1257                 xprt->free = req->rq_next;
1258                 req->rq_next = NULL;
1259                 task->tk_rqstp = req;
1260                 xprt_request_init(task, xprt);
1261                 return;
1262         }
1263         dprintk("RPC:      waiting for request slot\n");
1264         task->tk_status = -EAGAIN;
1265         task->tk_timeout = 0;
1266         rpc_sleep_on(&xprt->backlog, task, NULL, NULL);
1267 }
1268
1269 void
1270 xprt_reserve(struct rpc_task *task)
1271 {
1272         struct rpc_xprt *xprt = task->tk_xprt;
1273
1274         task->tk_status = -EIO;
1275         if (!xprt->shutdown) {
1276                 spin_lock(&xprt->xprt_lock);
1277                 do_xprt_reserve(task);
1278                 spin_unlock(&xprt->xprt_lock);
1279         }
1280 }
1281
1282 /*
1283  * Allocate a 'unique' XID
1284  */
1285 static u32
1286 xprt_alloc_xid(void)
1287 {
1288         static spinlock_t xid_lock = SPIN_LOCK_UNLOCKED;
1289         static int need_init = 1;
1290         static u32 xid;
1291         u32 ret;
1292
1293         spin_lock(&xid_lock);
1294         if (unlikely(need_init)) {
1295                 xid = CURRENT_TIME << 12;
1296                 need_init = 0;
1297         }
1298         ret = xid++;
1299         spin_unlock(&xid_lock);
1300         return ret;
1301 }
1302
1303 /*
1304  * Initialize RPC request
1305  */
1306 static void
1307 xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1308 {
1309         struct rpc_rqst *req = task->tk_rqstp;
1310
1311         req->rq_timeout = xprt->timeout;
1312         req->rq_task    = task;
1313         req->rq_xprt    = xprt;
1314         req->rq_xid     = xprt_alloc_xid();
1315         INIT_LIST_HEAD(&req->rq_list);
1316         dprintk("RPC: %4d reserved req %p xid %08x\n", task->tk_pid,
1317                         req, req->rq_xid);
1318 }
1319
1320 /*
1321  * Release an RPC call slot
1322  */
1323 void
1324 xprt_release(struct rpc_task *task)
1325 {
1326         struct rpc_xprt *xprt = task->tk_xprt;
1327         struct rpc_rqst *req;
1328
1329         if (!(req = task->tk_rqstp))
1330                 return;
1331         spin_lock_bh(&xprt->sock_lock);
1332         __xprt_release_write(xprt, task);
1333         __xprt_put_cong(xprt, req);
1334         if (!list_empty(&req->rq_list))
1335                 list_del(&req->rq_list);
1336         spin_unlock_bh(&xprt->sock_lock);
1337         task->tk_rqstp = NULL;
1338         memset(req, 0, sizeof(*req));   /* mark unused */
1339
1340         dprintk("RPC: %4d release request %p\n", task->tk_pid, req);
1341
1342         spin_lock(&xprt->xprt_lock);
1343         req->rq_next = xprt->free;
1344         xprt->free   = req;
1345
1346         xprt_clear_backlog(xprt);
1347         spin_unlock(&xprt->xprt_lock);
1348 }
1349
1350 /*
1351  * Set default timeout parameters
1352  */
1353 void
1354 xprt_default_timeout(struct rpc_timeout *to, int proto)
1355 {
1356         if (proto == IPPROTO_UDP)
1357                 xprt_set_timeout(to, 5,  5 * HZ);
1358         else
1359                 xprt_set_timeout(to, 5, 60 * HZ);
1360 }
1361
1362 /*
1363  * Set constant timeout
1364  */
1365 void
1366 xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
1367 {
1368         to->to_current   =
1369         to->to_initval   =
1370         to->to_increment = incr;
1371         to->to_maxval    = incr * retr;
1372         to->to_retries   = retr;
1373         to->to_exponential = 0;
1374 }
1375
1376 /*
1377  * Initialize an RPC client
1378  */
1379 static struct rpc_xprt *
1380 xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
1381 {
1382         struct rpc_xprt *xprt;
1383         struct rpc_rqst *req;
1384         int             i;
1385
1386         dprintk("RPC:      setting up %s transport...\n",
1387                                 proto == IPPROTO_UDP? "UDP" : "TCP");
1388
1389         if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL)
1390                 return NULL;
1391         memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */
1392
1393         xprt->addr = *ap;
1394         xprt->prot = proto;
1395         xprt->stream = (proto == IPPROTO_TCP)? 1 : 0;
1396         if (xprt->stream) {
1397                 xprt->cwnd = RPC_MAXCWND;
1398                 xprt->nocong = 1;
1399         } else
1400                 xprt->cwnd = RPC_INITCWND;
1401         spin_lock_init(&xprt->sock_lock);
1402         spin_lock_init(&xprt->xprt_lock);
1403         init_waitqueue_head(&xprt->cong_wait);
1404
1405         INIT_LIST_HEAD(&xprt->recv);
1406
1407         /* Set timeout parameters */
1408         if (to) {
1409                 xprt->timeout = *to;
1410                 xprt->timeout.to_current = to->to_initval;
1411         } else
1412                 xprt_default_timeout(&xprt->timeout, xprt->prot);
1413
1414         INIT_RPC_WAITQ(&xprt->pending, "xprt_pending");
1415         INIT_RPC_WAITQ(&xprt->sending, "xprt_sending");
1416         INIT_RPC_WAITQ(&xprt->resend, "xprt_resend");
1417         INIT_RPC_WAITQ(&xprt->backlog, "xprt_backlog");
1418
1419         /* initialize free list */
1420         for (i = 0, req = xprt->slot; i < RPC_MAXREQS-1; i++, req++)
1421                 req->rq_next = req + 1;
1422         req->rq_next = NULL;
1423         xprt->free = xprt->slot;
1424
1425         /* Check whether we want to use a reserved port */
1426         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1427
1428         dprintk("RPC:      created transport %p\n", xprt);
1429
1430         return xprt;
1431 }
1432
1433 /*
1434  * Bind to a reserved port
1435  */
1436 static inline int
1437 xprt_bindresvport(struct socket *sock)
1438 {
1439         struct sockaddr_in myaddr;
1440         int             err, port;
1441         kernel_cap_t saved_cap = current->cap_effective;
1442
1443         /* Override capabilities.
1444          * They were checked in xprt_create_proto i.e. at mount time
1445          */
1446         cap_raise (current->cap_effective, CAP_NET_BIND_SERVICE);
1447
1448         memset(&myaddr, 0, sizeof(myaddr));
1449         myaddr.sin_family = AF_INET;
1450         port = 800;
1451         do {
1452                 myaddr.sin_port = htons(port);
1453                 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
1454                                                 sizeof(myaddr));
1455         } while (err == -EADDRINUSE && --port > 0);
1456         current->cap_effective = saved_cap;
1457
1458         if (err < 0)
1459                 printk("RPC: Can't bind to reserved port (%d).\n", -err);
1460
1461         return err;
1462 }
1463
1464 static int
1465 xprt_bind_socket(struct rpc_xprt *xprt, struct socket *sock)
1466 {
1467         struct sock     *sk = sock->sk;
1468
1469         if (xprt->inet)
1470                 return -EBUSY;
1471
1472         write_lock_bh(&sk->callback_lock);
1473         sk->user_data = xprt;
1474         xprt->old_data_ready = sk->data_ready;
1475         xprt->old_state_change = sk->state_change;
1476         xprt->old_write_space = sk->write_space;
1477         if (xprt->prot == IPPROTO_UDP) {
1478                 sk->data_ready = udp_data_ready;
1479                 sk->no_check = UDP_CSUM_NORCV;
1480                 xprt_set_connected(xprt);
1481         } else {
1482                 struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
1483                 tp->nonagle = 1;        /* disable Nagle's algorithm */
1484                 sk->data_ready = tcp_data_ready;
1485                 sk->state_change = tcp_state_change;
1486                 xprt_clear_connected(xprt);
1487         }
1488         sk->write_space = xprt_write_space;
1489
1490         /* Reset to new socket */
1491         xprt->sock = sock;
1492         xprt->inet = sk;
1493         write_unlock_bh(&sk->callback_lock);
1494
1495         return 0;
1496 }
1497
1498 /*
1499  * Set socket buffer length
1500  */
1501 void
1502 xprt_sock_setbufsize(struct rpc_xprt *xprt)
1503 {
1504         struct sock *sk = xprt->inet;
1505
1506         if (xprt->stream)
1507                 return;
1508         if (xprt->rcvsize) {
1509                 sk->userlocks |= SOCK_RCVBUF_LOCK;
1510                 sk->rcvbuf = xprt->rcvsize * RPC_MAXCONG * 2;
1511         }
1512         if (xprt->sndsize) {
1513                 sk->userlocks |= SOCK_SNDBUF_LOCK;
1514                 sk->sndbuf = xprt->sndsize * RPC_MAXCONG * 2;
1515                 sk->write_space(sk);
1516         }
1517 }
1518
1519 /*
1520  * Create a client socket given the protocol and peer address.
1521  */
1522 static struct socket *
1523 xprt_create_socket(int proto, struct rpc_timeout *to, int resvport)
1524 {
1525         struct socket   *sock;
1526         int             type, err;
1527
1528         dprintk("RPC:      xprt_create_socket(%s %d)\n",
1529                            (proto == IPPROTO_UDP)? "udp" : "tcp", proto);
1530
1531         type = (proto == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1532
1533         if ((err = sock_create(PF_INET, type, proto, &sock)) < 0) {
1534                 printk("RPC: can't create socket (%d).\n", -err);
1535                 goto failed;
1536         }
1537
1538         /* bind to a reserved port */
1539         if (resvport && xprt_bindresvport(sock) < 0)
1540                 goto failed;
1541
1542         return sock;
1543
1544 failed:
1545         sock_release(sock);
1546         return NULL;
1547 }
1548
1549 /*
1550  * Create an RPC client transport given the protocol and peer address.
1551  */
1552 struct rpc_xprt *
1553 xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to)
1554 {
1555         struct rpc_xprt *xprt;
1556
1557         xprt = xprt_setup(proto, sap, to);
1558         if (!xprt)
1559                 goto out_bad;
1560
1561         dprintk("RPC:      xprt_create_proto created xprt %p\n", xprt);
1562         return xprt;
1563 out_bad:
1564         dprintk("RPC:      xprt_create_proto failed\n");
1565         if (xprt)
1566                 kfree(xprt);
1567         return NULL;
1568 }
1569
1570 /*
1571  * Prepare for transport shutdown.
1572  */
1573 void
1574 xprt_shutdown(struct rpc_xprt *xprt)
1575 {
1576         xprt->shutdown = 1;
1577         rpc_wake_up(&xprt->sending);
1578         rpc_wake_up(&xprt->resend);
1579         rpc_wake_up(&xprt->pending);
1580         rpc_wake_up(&xprt->backlog);
1581         if (waitqueue_active(&xprt->cong_wait))
1582                 wake_up(&xprt->cong_wait);
1583 }
1584
1585 /*
1586  * Clear the xprt backlog queue
1587  */
1588 int
1589 xprt_clear_backlog(struct rpc_xprt *xprt) {
1590         rpc_wake_up_next(&xprt->backlog);
1591         if (waitqueue_active(&xprt->cong_wait))
1592                 wake_up(&xprt->cong_wait);
1593         return 1;
1594 }
1595
1596 /*
1597  * Destroy an RPC transport, killing off all requests.
1598  */
1599 int
1600 xprt_destroy(struct rpc_xprt *xprt)
1601 {
1602         dprintk("RPC:      destroying transport %p\n", xprt);
1603         xprt_shutdown(xprt);
1604         xprt_close(xprt);
1605         kfree(xprt);
1606
1607         return 0;
1608 }