SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  37 #include <linux/module.h>
  38
  39 #include <linux/types.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/workqueue.h>
  42 #include <linux/random.h>
  43
  44 #include <linux/sunrpc/clnt.h>
  45
  46 /*
  47  * Local variables
  48  */
  49
  50 #ifdef RPC_DEBUG
  51 # undef  RPC_DEBUG_DATA
  52 # define RPCDBG_FACILITY        RPCDBG_XPRT
  53 #endif
  54
  55 #define XPRT_MAX_BACKOFF        (8)
  56
  57 /*
  58  * Local functions
  59  */
  60 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  61 static inline void      do_xprt_reserve(struct rpc_task *);
  62 static void     xprt_connect_status(struct rpc_task *task);
  63 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
  64
  65 static int      xprt_clear_backlog(struct rpc_xprt *xprt);
  66
  67 /*
  68  * Serialize write access to sockets, in order to prevent different
  69  * requests from interfering with each other.
  70  * Also prevents TCP socket connects from colliding with writes.
  71  */
  72 static int
  73 __xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
  74 {
  75         struct rpc_rqst *req = task->tk_rqstp;
  76
  77         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
  78                 if (task == xprt->snd_task)
  79                         return 1;
  80                 goto out_sleep;
  81         }
  82         if (xprt->nocong || __xprt_get_cong(xprt, task)) {
  83                 xprt->snd_task = task;
  84                 if (req) {
  85                         req->rq_bytes_sent = 0;
  86                         req->rq_ntrans++;
  87                 }
  88                 return 1;
  89         }
  90         smp_mb__before_clear_bit();
  91         clear_bit(XPRT_LOCKED, &xprt->state);
  92         smp_mb__after_clear_bit();
  93 out_sleep:
  94         dprintk("RPC: %4d failed to lock socket %p\n", task->tk_pid, xprt);
  95         task->tk_timeout = 0;
  96         task->tk_status = -EAGAIN;
  97         if (req && req->rq_ntrans)
  98                 rpc_sleep_on(&xprt->resend, task, NULL, NULL);
  99         else
 100                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
 101         return 0;
 102 }
 103
 104 static inline int
 105 xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 106 {
 107         int retval;
 108
 109         spin_lock_bh(&xprt->transport_lock);
 110         retval = __xprt_lock_write(xprt, task);
 111         spin_unlock_bh(&xprt->transport_lock);
 112         return retval;
 113 }
 114
 115
 116 static void
 117 __xprt_lock_write_next(struct rpc_xprt *xprt)
 118 {
 119         struct rpc_task *task;
 120
 121         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 122                 return;
 123         if (!xprt->nocong && RPCXPRT_CONGESTED(xprt))
 124                 goto out_unlock;
 125         task = rpc_wake_up_next(&xprt->resend);
 126         if (!task) {
 127                 task = rpc_wake_up_next(&xprt->sending);
 128                 if (!task)
 129                         goto out_unlock;
 130         }
 131         if (xprt->nocong || __xprt_get_cong(xprt, task)) {
 132                 struct rpc_rqst *req = task->tk_rqstp;
 133                 xprt->snd_task = task;
 134                 if (req) {
 135                         req->rq_bytes_sent = 0;
 136                         req->rq_ntrans++;
 137                 }
 138                 return;
 139         }
 140 out_unlock:
 141         smp_mb__before_clear_bit();
 142         clear_bit(XPRT_LOCKED, &xprt->state);
 143         smp_mb__after_clear_bit();
 144 }
 145
 146 /*
 147  * Releases the socket for use by other requests.
 148  */
 149 static void
 150 __xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 151 {
 152         if (xprt->snd_task == task) {
 153                 xprt->snd_task = NULL;
 154                 smp_mb__before_clear_bit();
 155                 clear_bit(XPRT_LOCKED, &xprt->state);
 156                 smp_mb__after_clear_bit();
 157                 __xprt_lock_write_next(xprt);
 158         }
 159 }
 160
 161 static inline void
 162 xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 163 {
 164         spin_lock_bh(&xprt->transport_lock);
 165         __xprt_release_write(xprt, task);
 166         spin_unlock_bh(&xprt->transport_lock);
 167 }
 168
 169 /*
 170  * Van Jacobson congestion avoidance. Check if the congestion window
 171  * overflowed. Put the task to sleep if this is the case.
 172  */
 173 static int
 174 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 175 {
 176         struct rpc_rqst *req = task->tk_rqstp;
 177
 178         if (req->rq_cong)
 179                 return 1;
 180         dprintk("RPC: %4d xprt_cwnd_limited cong = %ld cwnd = %ld\n",
 181                         task->tk_pid, xprt->cong, xprt->cwnd);
 182         if (RPCXPRT_CONGESTED(xprt))
 183                 return 0;
 184         req->rq_cong = 1;
 185         xprt->cong += RPC_CWNDSCALE;
 186         return 1;
 187 }
 188
 189 /*
 190  * Adjust the congestion window, and wake up the next task
 191  * that has been sleeping due to congestion
 192  */
 193 static void
 194 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 195 {
 196         if (!req->rq_cong)
 197                 return;
 198         req->rq_cong = 0;
 199         xprt->cong -= RPC_CWNDSCALE;
 200         __xprt_lock_write_next(xprt);
 201 }
 202
 203 /*
 204  * Adjust RPC congestion window
 205  * We use a time-smoothed congestion estimator to avoid heavy oscillation.
 206  */
 207 static void
 208 xprt_adjust_cwnd(struct rpc_xprt *xprt, int result)
 209 {
 210         unsigned long   cwnd;
 211
 212         cwnd = xprt->cwnd;
 213         if (result >= 0 && cwnd <= xprt->cong) {
 214                 /* The (cwnd >> 1) term makes sure
 215                  * the result gets rounded properly. */
 216                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 217                 if (cwnd > RPC_MAXCWND(xprt))
 218                         cwnd = RPC_MAXCWND(xprt);
 219                 __xprt_lock_write_next(xprt);
 220         } else if (result == -ETIMEDOUT) {
 221                 cwnd >>= 1;
 222                 if (cwnd < RPC_CWNDSCALE)
 223                         cwnd = RPC_CWNDSCALE;
 224         }
 225         dprintk("RPC:      cong %ld, cwnd was %ld, now %ld\n",
 226                         xprt->cong, xprt->cwnd, cwnd);
 227         xprt->cwnd = cwnd;
 228 }
 229
 230 /**
 231  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 232  * @xprt: transport with waiting tasks
 233  * @status: result code to plant in each task before waking it
 234  *
 235  */
 236 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 237 {
 238         if (status < 0)
 239                 rpc_wake_up_status(&xprt->pending, status);
 240         else
 241                 rpc_wake_up(&xprt->pending);
 242 }
 243
 244 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 245 {
 246         struct rpc_timeout *to = &req->rq_xprt->timeout;
 247
 248         req->rq_majortimeo = req->rq_timeout;
 249         if (to->to_exponential)
 250                 req->rq_majortimeo <<= to->to_retries;
 251         else
 252                 req->rq_majortimeo += to->to_increment * to->to_retries;
 253         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 254                 req->rq_majortimeo = to->to_maxval;
 255         req->rq_majortimeo += jiffies;
 256 }
 257
 258 /**
 259  * xprt_adjust_timeout - adjust timeout values for next retransmit
 260  * @req: RPC request containing parameters to use for the adjustment
 261  *
 262  */
 263 int xprt_adjust_timeout(struct rpc_rqst *req)
 264 {
 265         struct rpc_xprt *xprt = req->rq_xprt;
 266         struct rpc_timeout *to = &xprt->timeout;
 267         int status = 0;
 268
 269         if (time_before(jiffies, req->rq_majortimeo)) {
 270                 if (to->to_exponential)
 271                         req->rq_timeout <<= 1;
 272                 else
 273                         req->rq_timeout += to->to_increment;
 274                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 275                         req->rq_timeout = to->to_maxval;
 276                 req->rq_retries++;
 277                 pprintk("RPC: %lu retrans\n", jiffies);
 278         } else {
 279                 req->rq_timeout = to->to_initval;
 280                 req->rq_retries = 0;
 281                 xprt_reset_majortimeo(req);
 282                 /* Reset the RTT counters == "slow start" */
 283                 spin_lock_bh(&xprt->transport_lock);
 284                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 285                 spin_unlock_bh(&xprt->transport_lock);
 286                 pprintk("RPC: %lu timeout\n", jiffies);
 287                 status = -ETIMEDOUT;
 288         }
 289
 290         if (req->rq_timeout == 0) {
 291                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 292                 req->rq_timeout = 5 * HZ;
 293         }
 294         return status;
 295 }
 296
 297 static void
 298 xprt_socket_autoclose(void *args)
 299 {
 300         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
 301
 302         xprt_disconnect(xprt);
 303         xprt->ops->close(xprt);
 304         xprt_release_write(xprt, NULL);
 305 }
 306
 307 /**
 308  * xprt_disconnect - mark a transport as disconnected
 309  * @xprt: transport to flag for disconnect
 310  *
 311  */
 312 void xprt_disconnect(struct rpc_xprt *xprt)
 313 {
 314         dprintk("RPC:      disconnected transport %p\n", xprt);
 315         spin_lock_bh(&xprt->transport_lock);
 316         xprt_clear_connected(xprt);
 317         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 318         spin_unlock_bh(&xprt->transport_lock);
 319 }
 320
 321 static void
 322 xprt_init_autodisconnect(unsigned long data)
 323 {
 324         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 325
 326         spin_lock(&xprt->transport_lock);
 327         if (!list_empty(&xprt->recv) || xprt->shutdown)
 328                 goto out_abort;
 329         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 330                 goto out_abort;
 331         spin_unlock(&xprt->transport_lock);
 332         /* Let keventd close the socket */
 333         if (xprt_connecting(xprt))
 334                 xprt_release_write(xprt, NULL);
 335         else
 336                 schedule_work(&xprt->task_cleanup);
 337         return;
 338 out_abort:
 339         spin_unlock(&xprt->transport_lock);
 340 }
 341
 342 /**
 343  * xprt_connect - schedule a transport connect operation
 344  * @task: RPC task that is requesting the connect
 345  *
 346  */
 347 void xprt_connect(struct rpc_task *task)
 348 {
 349         struct rpc_xprt *xprt = task->tk_xprt;
 350
 351         dprintk("RPC: %4d xprt_connect xprt %p %s connected\n", task->tk_pid,
 352                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 353
 354         if (xprt->shutdown) {
 355                 task->tk_status = -EIO;
 356                 return;
 357         }
 358         if (!xprt->addr.sin_port) {
 359                 task->tk_status = -EIO;
 360                 return;
 361         }
 362         if (!xprt_lock_write(xprt, task))
 363                 return;
 364         if (xprt_connected(xprt))
 365                 xprt_release_write(xprt, task);
 366         else {
 367                 if (task->tk_rqstp)
 368                         task->tk_rqstp->rq_bytes_sent = 0;
 369
 370                 task->tk_timeout = RPC_CONNECT_TIMEOUT;
 371                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
 372                 xprt->ops->connect(task);
 373         }
 374         return;
 375 }
 376
 377 static void xprt_connect_status(struct rpc_task *task)
 378 {
 379         struct rpc_xprt *xprt = task->tk_xprt;
 380
 381         if (task->tk_status >= 0) {
 382                 dprintk("RPC: %4d xprt_connect_status: connection established\n",
 383                                 task->tk_pid);
 384                 return;
 385         }
 386
 387         switch (task->tk_status) {
 388         case -ECONNREFUSED:
 389         case -ECONNRESET:
 390                 dprintk("RPC: %4d xprt_connect_status: server %s refused connection\n",
 391                                 task->tk_pid, task->tk_client->cl_server);
 392                 break;
 393         case -ENOTCONN:
 394                 dprintk("RPC: %4d xprt_connect_status: connection broken\n",
 395                                 task->tk_pid);
 396                 break;
 397         case -ETIMEDOUT:
 398                 dprintk("RPC: %4d xprt_connect_status: connect attempt timed out\n",
 399                                 task->tk_pid);
 400                 break;
 401         default:
 402                 dprintk("RPC: %4d xprt_connect_status: error %d connecting to server %s\n",
 403                                 task->tk_pid, -task->tk_status, task->tk_client->cl_server);
 404                 xprt_release_write(xprt, task);
 405                 task->tk_status = -EIO;
 406                 return;
 407         }
 408
 409         /* if soft mounted, just cause this RPC to fail */
 410         if (RPC_IS_SOFT(task)) {
 411                 xprt_release_write(xprt, task);
 412                 task->tk_status = -EIO;
 413         }
 414 }
 415
 416 /**
 417  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 418  * @xprt: transport on which the original request was transmitted
 419  * @xid: RPC XID of incoming reply
 420  *
 421  */
 422 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
 423 {
 424         struct list_head *pos;
 425         struct rpc_rqst *req = NULL;
 426
 427         list_for_each(pos, &xprt->recv) {
 428                 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
 429                 if (entry->rq_xid == xid) {
 430                         req = entry;
 431                         break;
 432                 }
 433         }
 434         return req;
 435 }
 436
 437 /**
 438  * xprt_complete_rqst - called when reply processing is complete
 439  * @xprt: controlling transport
 440  * @req: RPC request that just completed
 441  * @copied: actual number of bytes received from the transport
 442  *
 443  */
 444 void xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
 445 {
 446         struct rpc_task *task = req->rq_task;
 447         struct rpc_clnt *clnt = task->tk_client;
 448
 449         /* Adjust congestion window */
 450         if (!xprt->nocong) {
 451                 unsigned timer = task->tk_msg.rpc_proc->p_timer;
 452                 xprt_adjust_cwnd(xprt, copied);
 453                 __xprt_put_cong(xprt, req);
 454                 if (timer) {
 455                         if (req->rq_ntrans == 1)
 456                                 rpc_update_rtt(clnt->cl_rtt, timer,
 457                                                 (long)jiffies - req->rq_xtime);
 458                         rpc_set_timeo(clnt->cl_rtt, timer, req->rq_ntrans - 1);
 459                 }
 460         }
 461
 462 #ifdef RPC_PROFILE
 463         /* Profile only reads for now */
 464         if (copied > 1024) {
 465                 static unsigned long    nextstat;
 466                 static unsigned long    pkt_rtt, pkt_len, pkt_cnt;
 467
 468                 pkt_cnt++;
 469                 pkt_len += req->rq_slen + copied;
 470                 pkt_rtt += jiffies - req->rq_xtime;
 471                 if (time_before(nextstat, jiffies)) {
 472                         printk("RPC: %lu %ld cwnd\n", jiffies, xprt->cwnd);
 473                         printk("RPC: %ld %ld %ld %ld stat\n",
 474                                         jiffies, pkt_cnt, pkt_len, pkt_rtt);
 475                         pkt_rtt = pkt_len = pkt_cnt = 0;
 476                         nextstat = jiffies + 5 * HZ;
 477                 }
 478         }
 479 #endif
 480
 481         dprintk("RPC: %4d has input (%d bytes)\n", task->tk_pid, copied);
 482         list_del_init(&req->rq_list);
 483         req->rq_received = req->rq_private_buf.len = copied;
 484
 485         /* ... and wake up the process. */
 486         rpc_wake_up_task(task);
 487         return;
 488 }
 489
 490 /*
 491  * RPC receive timeout handler.
 492  */
 493 static void
 494 xprt_timer(struct rpc_task *task)
 495 {
 496         struct rpc_rqst *req = task->tk_rqstp;
 497         struct rpc_xprt *xprt = req->rq_xprt;
 498
 499         spin_lock(&xprt->transport_lock);
 500         if (req->rq_received)
 501                 goto out;
 502
 503         xprt_adjust_cwnd(req->rq_xprt, -ETIMEDOUT);
 504         __xprt_put_cong(xprt, req);
 505
 506         dprintk("RPC: %4d xprt_timer (%s request)\n",
 507                 task->tk_pid, req ? "pending" : "backlogged");
 508
 509         task->tk_status  = -ETIMEDOUT;
 510 out:
 511         task->tk_timeout = 0;
 512         rpc_wake_up_task(task);
 513         spin_unlock(&xprt->transport_lock);
 514 }
 515
 516 /**
 517  * xprt_prepare_transmit - reserve the transport before sending a request
 518  * @task: RPC task about to send a request
 519  *
 520  */
 521 int xprt_prepare_transmit(struct rpc_task *task)
 522 {
 523         struct rpc_rqst *req = task->tk_rqstp;
 524         struct rpc_xprt *xprt = req->rq_xprt;
 525         int err = 0;
 526
 527         dprintk("RPC: %4d xprt_prepare_transmit\n", task->tk_pid);
 528
 529         if (xprt->shutdown)
 530                 return -EIO;
 531
 532         spin_lock_bh(&xprt->transport_lock);
 533         if (req->rq_received && !req->rq_bytes_sent) {
 534                 err = req->rq_received;
 535                 goto out_unlock;
 536         }
 537         if (!__xprt_lock_write(xprt, task)) {
 538                 err = -EAGAIN;
 539                 goto out_unlock;
 540         }
 541
 542         if (!xprt_connected(xprt)) {
 543                 err = -ENOTCONN;
 544                 goto out_unlock;
 545         }
 546 out_unlock:
 547         spin_unlock_bh(&xprt->transport_lock);
 548         return err;
 549 }
 550
 551 /**
 552  * xprt_transmit - send an RPC request on a transport
 553  * @task: controlling RPC task
 554  *
 555  * We have to copy the iovec because sendmsg fiddles with its contents.
 556  */
 557 void xprt_transmit(struct rpc_task *task)
 558 {
 559         struct rpc_clnt *clnt = task->tk_client;
 560         struct rpc_rqst *req = task->tk_rqstp;
 561         struct rpc_xprt *xprt = req->rq_xprt;
 562         int status;
 563
 564         dprintk("RPC: %4d xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 565
 566         smp_rmb();
 567         if (!req->rq_received) {
 568                 if (list_empty(&req->rq_list)) {
 569                         spin_lock_bh(&xprt->transport_lock);
 570                         /* Update the softirq receive buffer */
 571                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 572                                         sizeof(req->rq_private_buf));
 573                         /* Add request to the receive list */
 574                         list_add_tail(&req->rq_list, &xprt->recv);
 575                         spin_unlock_bh(&xprt->transport_lock);
 576                         xprt_reset_majortimeo(req);
 577                         /* Turn off autodisconnect */
 578                         del_singleshot_timer_sync(&xprt->timer);
 579                 }
 580         } else if (!req->rq_bytes_sent)
 581                 return;
 582
 583         status = xprt->ops->send_request(task);
 584         if (!status)
 585                 goto out_receive;
 586
 587         /* Note: at this point, task->tk_sleeping has not yet been set,
 588          *       hence there is no danger of the waking up task being put on
 589          *       schedq, and being picked up by a parallel run of rpciod().
 590          */
 591         task->tk_status = status;
 592
 593         switch (status) {
 594         case -ECONNREFUSED:
 595                 task->tk_timeout = RPC_REESTABLISH_TIMEOUT;
 596                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
 597         case -EAGAIN:
 598         case -ENOTCONN:
 599                 return;
 600         default:
 601                 if (xprt->stream)
 602                         xprt_disconnect(xprt);
 603         }
 604         xprt_release_write(xprt, task);
 605         return;
 606  out_receive:
 607         dprintk("RPC: %4d xmit complete\n", task->tk_pid);
 608         /* Set the task's receive timeout value */
 609         spin_lock_bh(&xprt->transport_lock);
 610         if (!xprt->nocong) {
 611                 int timer = task->tk_msg.rpc_proc->p_timer;
 612                 task->tk_timeout = rpc_calc_rto(clnt->cl_rtt, timer);
 613                 task->tk_timeout <<= rpc_ntimeo(clnt->cl_rtt, timer) + req->rq_retries;
 614                 if (task->tk_timeout > xprt->timeout.to_maxval || task->tk_timeout == 0)
 615                         task->tk_timeout = xprt->timeout.to_maxval;
 616         } else
 617                 task->tk_timeout = req->rq_timeout;
 618         /* Don't race with disconnect */
 619         if (!xprt_connected(xprt))
 620                 task->tk_status = -ENOTCONN;
 621         else if (!req->rq_received)
 622                 rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer);
 623         __xprt_release_write(xprt, task);
 624         spin_unlock_bh(&xprt->transport_lock);
 625 }
 626
 627 static inline void do_xprt_reserve(struct rpc_task *task)
 628 {
 629         struct rpc_xprt *xprt = task->tk_xprt;
 630
 631         task->tk_status = 0;
 632         if (task->tk_rqstp)
 633                 return;
 634         if (!list_empty(&xprt->free)) {
 635                 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
 636                 list_del_init(&req->rq_list);
 637                 task->tk_rqstp = req;
 638                 xprt_request_init(task, xprt);
 639                 return;
 640         }
 641         dprintk("RPC:      waiting for request slot\n");
 642         task->tk_status = -EAGAIN;
 643         task->tk_timeout = 0;
 644         rpc_sleep_on(&xprt->backlog, task, NULL, NULL);
 645 }
 646
 647 /**
 648  * xprt_reserve - allocate an RPC request slot
 649  * @task: RPC task requesting a slot allocation
 650  *
 651  * If no more slots are available, place the task on the transport's
 652  * backlog queue.
 653  */
 654 void xprt_reserve(struct rpc_task *task)
 655 {
 656         struct rpc_xprt *xprt = task->tk_xprt;
 657
 658         task->tk_status = -EIO;
 659         if (!xprt->shutdown) {
 660                 spin_lock(&xprt->reserve_lock);
 661                 do_xprt_reserve(task);
 662                 spin_unlock(&xprt->reserve_lock);
 663         }
 664 }
 665
 666 static inline u32 xprt_alloc_xid(struct rpc_xprt *xprt)
 667 {
 668         return xprt->xid++;
 669 }
 670
 671 static inline void xprt_init_xid(struct rpc_xprt *xprt)
 672 {
 673         get_random_bytes(&xprt->xid, sizeof(xprt->xid));
 674 }
 675
 676 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
 677 {
 678         struct rpc_rqst *req = task->tk_rqstp;
 679
 680         req->rq_timeout = xprt->timeout.to_initval;
 681         req->rq_task    = task;
 682         req->rq_xprt    = xprt;
 683         req->rq_xid     = xprt_alloc_xid(xprt);
 684         dprintk("RPC: %4d reserved req %p xid %08x\n", task->tk_pid,
 685                         req, ntohl(req->rq_xid));
 686 }
 687
 688 /**
 689  * xprt_release - release an RPC request slot
 690  * @task: task which is finished with the slot
 691  *
 692  */
 693 void xprt_release(struct rpc_task *task)
 694 {
 695         struct rpc_xprt *xprt = task->tk_xprt;
 696         struct rpc_rqst *req;
 697
 698         if (!(req = task->tk_rqstp))
 699                 return;
 700         spin_lock_bh(&xprt->transport_lock);
 701         __xprt_release_write(xprt, task);
 702         __xprt_put_cong(xprt, req);
 703         if (!list_empty(&req->rq_list))
 704                 list_del(&req->rq_list);
 705         xprt->last_used = jiffies;
 706         if (list_empty(&xprt->recv) && !xprt->shutdown)
 707                 mod_timer(&xprt->timer,
 708                                 xprt->last_used + RPC_IDLE_DISCONNECT_TIMEOUT);
 709         spin_unlock_bh(&xprt->transport_lock);
 710         task->tk_rqstp = NULL;
 711         memset(req, 0, sizeof(*req));   /* mark unused */
 712
 713         dprintk("RPC: %4d release request %p\n", task->tk_pid, req);
 714
 715         spin_lock(&xprt->reserve_lock);
 716         list_add(&req->rq_list, &xprt->free);
 717         xprt_clear_backlog(xprt);
 718         spin_unlock(&xprt->reserve_lock);
 719 }
 720
 721 /**
 722  * xprt_set_timeout - set constant RPC timeout
 723  * @to: RPC timeout parameters to set up
 724  * @retr: number of retries
 725  * @incr: amount of increase after each retry
 726  *
 727  */
 728 void xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
 729 {
 730         to->to_initval   =
 731         to->to_increment = incr;
 732         to->to_maxval    = to->to_initval + (incr * retr);
 733         to->to_retries   = retr;
 734         to->to_exponential = 0;
 735 }
 736
 737 static struct rpc_xprt *xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
 738 {
 739         int result;
 740         struct rpc_xprt *xprt;
 741         struct rpc_rqst *req;
 742
 743         if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL)
 744                 return ERR_PTR(-ENOMEM);
 745         memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */
 746
 747         xprt->addr = *ap;
 748
 749         switch (proto) {
 750         case IPPROTO_UDP:
 751                 result = xs_setup_udp(xprt, to);
 752                 break;
 753         case IPPROTO_TCP:
 754                 result = xs_setup_tcp(xprt, to);
 755                 break;
 756         default:
 757                 printk(KERN_ERR "RPC: unrecognized transport protocol: %d\n",
 758                                 proto);
 759                 result = -EIO;
 760                 break;
 761         }
 762         if (result) {
 763                 kfree(xprt);
 764                 return ERR_PTR(result);
 765         }
 766
 767         spin_lock_init(&xprt->transport_lock);
 768         spin_lock_init(&xprt->reserve_lock);
 769         init_waitqueue_head(&xprt->cong_wait);
 770
 771         INIT_LIST_HEAD(&xprt->free);
 772         INIT_LIST_HEAD(&xprt->recv);
 773         INIT_WORK(&xprt->task_cleanup, xprt_socket_autoclose, xprt);
 774         init_timer(&xprt->timer);
 775         xprt->timer.function = xprt_init_autodisconnect;
 776         xprt->timer.data = (unsigned long) xprt;
 777         xprt->last_used = jiffies;
 778
 779         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
 780         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
 781         rpc_init_wait_queue(&xprt->resend, "xprt_resend");
 782         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
 783
 784         /* initialize free list */
 785         for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
 786                 list_add(&req->rq_list, &xprt->free);
 787
 788         xprt_init_xid(xprt);
 789
 790         dprintk("RPC:      created transport %p with %u slots\n", xprt,
 791                         xprt->max_reqs);
 792
 793         return xprt;
 794 }
 795
 796 /**
 797  * xprt_create_proto - create an RPC client transport
 798  * @proto: requested transport protocol
 799  * @sap: remote peer's address
 800  * @to: timeout parameters for new transport
 801  *
 802  */
 803 struct rpc_xprt *xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to)
 804 {
 805         struct rpc_xprt *xprt;
 806
 807         xprt = xprt_setup(proto, sap, to);
 808         if (IS_ERR(xprt))
 809                 dprintk("RPC:      xprt_create_proto failed\n");
 810         else
 811                 dprintk("RPC:      xprt_create_proto created xprt %p\n", xprt);
 812         return xprt;
 813 }
 814
 815 static void xprt_shutdown(struct rpc_xprt *xprt)
 816 {
 817         xprt->shutdown = 1;
 818         rpc_wake_up(&xprt->sending);
 819         rpc_wake_up(&xprt->resend);
 820         xprt_wake_pending_tasks(xprt, -EIO);
 821         rpc_wake_up(&xprt->backlog);
 822         wake_up(&xprt->cong_wait);
 823         del_timer_sync(&xprt->timer);
 824 }
 825
 826 static int xprt_clear_backlog(struct rpc_xprt *xprt) {
 827         rpc_wake_up_next(&xprt->backlog);
 828         wake_up(&xprt->cong_wait);
 829         return 1;
 830 }
 831
 832 /**
 833  * xprt_destroy - destroy an RPC transport, killing off all requests.
 834  * @xprt: transport to destroy
 835  *
 836  */
 837 int xprt_destroy(struct rpc_xprt *xprt)
 838 {
 839         dprintk("RPC:      destroying transport %p\n", xprt);
 840         xprt_shutdown(xprt);
 841         xprt->ops->destroy(xprt);
 842         kfree(xprt);
 843
 844         return 0;
 845 }