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_transmit().
  14  *  -   xprt_transmit sends the message and installs the caller on the
  15  *      transport'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 transport. 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  *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
  37  */
  38
  39 #include <linux/module.h>
  40
  41 #include <linux/types.h>
  42 #include <linux/interrupt.h>
  43 #include <linux/workqueue.h>
  44 #include <linux/random.h>
  45
  46 #include <linux/sunrpc/clnt.h>
  47
  48 /*
  49  * Local variables
  50  */
  51
  52 #ifdef RPC_DEBUG
  53 # undef  RPC_DEBUG_DATA
  54 # define RPCDBG_FACILITY        RPCDBG_XPRT
  55 #endif
  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 transports, in order to prevent different
  69  * requests from interfering with each other.
  70  * Also prevents transport 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 transport %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 transport 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 xprt_autoclose(void *args)
 298 {
 299         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
 300
 301         xprt_disconnect(xprt);
 302         xprt->ops->close(xprt);
 303         xprt_release_write(xprt, NULL);
 304 }
 305
 306 /**
 307  * xprt_disconnect - mark a transport as disconnected
 308  * @xprt: transport to flag for disconnect
 309  *
 310  */
 311 void xprt_disconnect(struct rpc_xprt *xprt)
 312 {
 313         dprintk("RPC:      disconnected transport %p\n", xprt);
 314         spin_lock_bh(&xprt->transport_lock);
 315         xprt_clear_connected(xprt);
 316         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 317         spin_unlock_bh(&xprt->transport_lock);
 318 }
 319
 320 static void
 321 xprt_init_autodisconnect(unsigned long data)
 322 {
 323         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 324
 325         spin_lock(&xprt->transport_lock);
 326         if (!list_empty(&xprt->recv) || xprt->shutdown)
 327                 goto out_abort;
 328         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 329                 goto out_abort;
 330         spin_unlock(&xprt->transport_lock);
 331         if (xprt_connecting(xprt))
 332                 xprt_release_write(xprt, NULL);
 333         else
 334                 schedule_work(&xprt->task_cleanup);
 335         return;
 336 out_abort:
 337         spin_unlock(&xprt->transport_lock);
 338 }
 339
 340 /**
 341  * xprt_connect - schedule a transport connect operation
 342  * @task: RPC task that is requesting the connect
 343  *
 344  */
 345 void xprt_connect(struct rpc_task *task)
 346 {
 347         struct rpc_xprt *xprt = task->tk_xprt;
 348
 349         dprintk("RPC: %4d xprt_connect xprt %p %s connected\n", task->tk_pid,
 350                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 351
 352         if (xprt->shutdown) {
 353                 task->tk_status = -EIO;
 354                 return;
 355         }
 356         if (!xprt->addr.sin_port) {
 357                 task->tk_status = -EIO;
 358                 return;
 359         }
 360         if (!xprt_lock_write(xprt, task))
 361                 return;
 362         if (xprt_connected(xprt))
 363                 xprt_release_write(xprt, task);
 364         else {
 365                 if (task->tk_rqstp)
 366                         task->tk_rqstp->rq_bytes_sent = 0;
 367
 368                 task->tk_timeout = RPC_CONNECT_TIMEOUT;
 369                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
 370                 xprt->ops->connect(task);
 371         }
 372         return;
 373 }
 374
 375 static void xprt_connect_status(struct rpc_task *task)
 376 {
 377         struct rpc_xprt *xprt = task->tk_xprt;
 378
 379         if (task->tk_status >= 0) {
 380                 dprintk("RPC: %4d xprt_connect_status: connection established\n",
 381                                 task->tk_pid);
 382                 return;
 383         }
 384
 385         switch (task->tk_status) {
 386         case -ECONNREFUSED:
 387         case -ECONNRESET:
 388                 dprintk("RPC: %4d xprt_connect_status: server %s refused connection\n",
 389                                 task->tk_pid, task->tk_client->cl_server);
 390                 break;
 391         case -ENOTCONN:
 392                 dprintk("RPC: %4d xprt_connect_status: connection broken\n",
 393                                 task->tk_pid);
 394                 break;
 395         case -ETIMEDOUT:
 396                 dprintk("RPC: %4d xprt_connect_status: connect attempt timed out\n",
 397                                 task->tk_pid);
 398                 break;
 399         default:
 400                 dprintk("RPC: %4d xprt_connect_status: error %d connecting to server %s\n",
 401                                 task->tk_pid, -task->tk_status, task->tk_client->cl_server);
 402                 xprt_release_write(xprt, task);
 403                 task->tk_status = -EIO;
 404                 return;
 405         }
 406
 407         /* if soft mounted, just cause this RPC to fail */
 408         if (RPC_IS_SOFT(task)) {
 409                 xprt_release_write(xprt, task);
 410                 task->tk_status = -EIO;
 411         }
 412 }
 413
 414 /**
 415  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 416  * @xprt: transport on which the original request was transmitted
 417  * @xid: RPC XID of incoming reply
 418  *
 419  */
 420 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
 421 {
 422         struct list_head *pos;
 423         struct rpc_rqst *req = NULL;
 424
 425         list_for_each(pos, &xprt->recv) {
 426                 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
 427                 if (entry->rq_xid == xid) {
 428                         req = entry;
 429                         break;
 430                 }
 431         }
 432         return req;
 433 }
 434
 435 /**
 436  * xprt_complete_rqst - called when reply processing is complete
 437  * @xprt: controlling transport
 438  * @req: RPC request that just completed
 439  * @copied: actual number of bytes received from the transport
 440  *
 441  */
 442 void xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
 443 {
 444         struct rpc_task *task = req->rq_task;
 445         struct rpc_clnt *clnt = task->tk_client;
 446
 447         /* Adjust congestion window */
 448         if (!xprt->nocong) {
 449                 unsigned timer = task->tk_msg.rpc_proc->p_timer;
 450                 xprt_adjust_cwnd(xprt, copied);
 451                 __xprt_put_cong(xprt, req);
 452                 if (timer) {
 453                         if (req->rq_ntrans == 1)
 454                                 rpc_update_rtt(clnt->cl_rtt, timer,
 455                                                 (long)jiffies - req->rq_xtime);
 456                         rpc_set_timeo(clnt->cl_rtt, timer, req->rq_ntrans - 1);
 457                 }
 458         }
 459
 460 #ifdef RPC_PROFILE
 461         /* Profile only reads for now */
 462         if (copied > 1024) {
 463                 static unsigned long    nextstat;
 464                 static unsigned long    pkt_rtt, pkt_len, pkt_cnt;
 465
 466                 pkt_cnt++;
 467                 pkt_len += req->rq_slen + copied;
 468                 pkt_rtt += jiffies - req->rq_xtime;
 469                 if (time_before(nextstat, jiffies)) {
 470                         printk("RPC: %lu %ld cwnd\n", jiffies, xprt->cwnd);
 471                         printk("RPC: %ld %ld %ld %ld stat\n",
 472                                         jiffies, pkt_cnt, pkt_len, pkt_rtt);
 473                         pkt_rtt = pkt_len = pkt_cnt = 0;
 474                         nextstat = jiffies + 5 * HZ;
 475                 }
 476         }
 477 #endif
 478
 479         dprintk("RPC: %4d has input (%d bytes)\n", task->tk_pid, copied);
 480         list_del_init(&req->rq_list);
 481         req->rq_received = req->rq_private_buf.len = copied;
 482
 483         /* ... and wake up the process. */
 484         rpc_wake_up_task(task);
 485         return;
 486 }
 487
 488 /*
 489  * RPC receive timeout handler.
 490  */
 491 static void
 492 xprt_timer(struct rpc_task *task)
 493 {
 494         struct rpc_rqst *req = task->tk_rqstp;
 495         struct rpc_xprt *xprt = req->rq_xprt;
 496
 497         spin_lock(&xprt->transport_lock);
 498         if (req->rq_received)
 499                 goto out;
 500
 501         xprt_adjust_cwnd(req->rq_xprt, -ETIMEDOUT);
 502         __xprt_put_cong(xprt, req);
 503
 504         dprintk("RPC: %4d xprt_timer (%s request)\n",
 505                 task->tk_pid, req ? "pending" : "backlogged");
 506
 507         task->tk_status  = -ETIMEDOUT;
 508 out:
 509         task->tk_timeout = 0;
 510         rpc_wake_up_task(task);
 511         spin_unlock(&xprt->transport_lock);
 512 }
 513
 514 /**
 515  * xprt_prepare_transmit - reserve the transport before sending a request
 516  * @task: RPC task about to send a request
 517  *
 518  */
 519 int xprt_prepare_transmit(struct rpc_task *task)
 520 {
 521         struct rpc_rqst *req = task->tk_rqstp;
 522         struct rpc_xprt *xprt = req->rq_xprt;
 523         int err = 0;
 524
 525         dprintk("RPC: %4d xprt_prepare_transmit\n", task->tk_pid);
 526
 527         if (xprt->shutdown)
 528                 return -EIO;
 529
 530         spin_lock_bh(&xprt->transport_lock);
 531         if (req->rq_received && !req->rq_bytes_sent) {
 532                 err = req->rq_received;
 533                 goto out_unlock;
 534         }
 535         if (!__xprt_lock_write(xprt, task)) {
 536                 err = -EAGAIN;
 537                 goto out_unlock;
 538         }
 539
 540         if (!xprt_connected(xprt)) {
 541                 err = -ENOTCONN;
 542                 goto out_unlock;
 543         }
 544 out_unlock:
 545         spin_unlock_bh(&xprt->transport_lock);
 546         return err;
 547 }
 548
 549 /**
 550  * xprt_transmit - send an RPC request on a transport
 551  * @task: controlling RPC task
 552  *
 553  * We have to copy the iovec because sendmsg fiddles with its contents.
 554  */
 555 void xprt_transmit(struct rpc_task *task)
 556 {
 557         struct rpc_clnt *clnt = task->tk_client;
 558         struct rpc_rqst *req = task->tk_rqstp;
 559         struct rpc_xprt *xprt = req->rq_xprt;
 560         int status;
 561
 562         dprintk("RPC: %4d xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 563
 564         smp_rmb();
 565         if (!req->rq_received) {
 566                 if (list_empty(&req->rq_list)) {
 567                         spin_lock_bh(&xprt->transport_lock);
 568                         /* Update the softirq receive buffer */
 569                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 570                                         sizeof(req->rq_private_buf));
 571                         /* Add request to the receive list */
 572                         list_add_tail(&req->rq_list, &xprt->recv);
 573                         spin_unlock_bh(&xprt->transport_lock);
 574                         xprt_reset_majortimeo(req);
 575                         /* Turn off autodisconnect */
 576                         del_singleshot_timer_sync(&xprt->timer);
 577                 }
 578         } else if (!req->rq_bytes_sent)
 579                 return;
 580
 581         status = xprt->ops->send_request(task);
 582         if (!status)
 583                 goto out_receive;
 584
 585         /* Note: at this point, task->tk_sleeping has not yet been set,
 586          *       hence there is no danger of the waking up task being put on
 587          *       schedq, and being picked up by a parallel run of rpciod().
 588          */
 589         task->tk_status = status;
 590
 591         switch (status) {
 592         case -ECONNREFUSED:
 593                 task->tk_timeout = RPC_REESTABLISH_TIMEOUT;
 594                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
 595         case -EAGAIN:
 596         case -ENOTCONN:
 597                 return;
 598         default:
 599                 if (xprt->stream)
 600                         xprt_disconnect(xprt);
 601         }
 602         xprt_release_write(xprt, task);
 603         return;
 604  out_receive:
 605         dprintk("RPC: %4d xmit complete\n", task->tk_pid);
 606         /* Set the task's receive timeout value */
 607         spin_lock_bh(&xprt->transport_lock);
 608         if (!xprt->nocong) {
 609                 int timer = task->tk_msg.rpc_proc->p_timer;
 610                 task->tk_timeout = rpc_calc_rto(clnt->cl_rtt, timer);
 611                 task->tk_timeout <<= rpc_ntimeo(clnt->cl_rtt, timer) + req->rq_retries;
 612                 if (task->tk_timeout > xprt->timeout.to_maxval || task->tk_timeout == 0)
 613                         task->tk_timeout = xprt->timeout.to_maxval;
 614         } else
 615                 task->tk_timeout = req->rq_timeout;
 616         /* Don't race with disconnect */
 617         if (!xprt_connected(xprt))
 618                 task->tk_status = -ENOTCONN;
 619         else if (!req->rq_received)
 620                 rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer);
 621         __xprt_release_write(xprt, task);
 622         spin_unlock_bh(&xprt->transport_lock);
 623 }
 624
 625 static inline void do_xprt_reserve(struct rpc_task *task)
 626 {
 627         struct rpc_xprt *xprt = task->tk_xprt;
 628
 629         task->tk_status = 0;
 630         if (task->tk_rqstp)
 631                 return;
 632         if (!list_empty(&xprt->free)) {
 633                 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
 634                 list_del_init(&req->rq_list);
 635                 task->tk_rqstp = req;
 636                 xprt_request_init(task, xprt);
 637                 return;
 638         }
 639         dprintk("RPC:      waiting for request slot\n");
 640         task->tk_status = -EAGAIN;
 641         task->tk_timeout = 0;
 642         rpc_sleep_on(&xprt->backlog, task, NULL, NULL);
 643 }
 644
 645 /**
 646  * xprt_reserve - allocate an RPC request slot
 647  * @task: RPC task requesting a slot allocation
 648  *
 649  * If no more slots are available, place the task on the transport's
 650  * backlog queue.
 651  */
 652 void xprt_reserve(struct rpc_task *task)
 653 {
 654         struct rpc_xprt *xprt = task->tk_xprt;
 655
 656         task->tk_status = -EIO;
 657         if (!xprt->shutdown) {
 658                 spin_lock(&xprt->reserve_lock);
 659                 do_xprt_reserve(task);
 660                 spin_unlock(&xprt->reserve_lock);
 661         }
 662 }
 663
 664 static inline u32 xprt_alloc_xid(struct rpc_xprt *xprt)
 665 {
 666         return xprt->xid++;
 667 }
 668
 669 static inline void xprt_init_xid(struct rpc_xprt *xprt)
 670 {
 671         get_random_bytes(&xprt->xid, sizeof(xprt->xid));
 672 }
 673
 674 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
 675 {
 676         struct rpc_rqst *req = task->tk_rqstp;
 677
 678         req->rq_timeout = xprt->timeout.to_initval;
 679         req->rq_task    = task;
 680         req->rq_xprt    = xprt;
 681         req->rq_xid     = xprt_alloc_xid(xprt);
 682         dprintk("RPC: %4d reserved req %p xid %08x\n", task->tk_pid,
 683                         req, ntohl(req->rq_xid));
 684 }
 685
 686 /**
 687  * xprt_release - release an RPC request slot
 688  * @task: task which is finished with the slot
 689  *
 690  */
 691 void xprt_release(struct rpc_task *task)
 692 {
 693         struct rpc_xprt *xprt = task->tk_xprt;
 694         struct rpc_rqst *req;
 695
 696         if (!(req = task->tk_rqstp))
 697                 return;
 698         spin_lock_bh(&xprt->transport_lock);
 699         __xprt_release_write(xprt, task);
 700         __xprt_put_cong(xprt, req);
 701         if (!list_empty(&req->rq_list))
 702                 list_del(&req->rq_list);
 703         xprt->last_used = jiffies;
 704         if (list_empty(&xprt->recv) && !xprt->shutdown)
 705                 mod_timer(&xprt->timer,
 706                                 xprt->last_used + RPC_IDLE_DISCONNECT_TIMEOUT);
 707         spin_unlock_bh(&xprt->transport_lock);
 708         task->tk_rqstp = NULL;
 709         memset(req, 0, sizeof(*req));   /* mark unused */
 710
 711         dprintk("RPC: %4d release request %p\n", task->tk_pid, req);
 712
 713         spin_lock(&xprt->reserve_lock);
 714         list_add(&req->rq_list, &xprt->free);
 715         xprt_clear_backlog(xprt);
 716         spin_unlock(&xprt->reserve_lock);
 717 }
 718
 719 /**
 720  * xprt_set_timeout - set constant RPC timeout
 721  * @to: RPC timeout parameters to set up
 722  * @retr: number of retries
 723  * @incr: amount of increase after each retry
 724  *
 725  */
 726 void xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
 727 {
 728         to->to_initval   =
 729         to->to_increment = incr;
 730         to->to_maxval    = to->to_initval + (incr * retr);
 731         to->to_retries   = retr;
 732         to->to_exponential = 0;
 733 }
 734
 735 static struct rpc_xprt *xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
 736 {
 737         int result;
 738         struct rpc_xprt *xprt;
 739         struct rpc_rqst *req;
 740
 741         if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL)
 742                 return ERR_PTR(-ENOMEM);
 743         memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */
 744
 745         xprt->addr = *ap;
 746
 747         switch (proto) {
 748         case IPPROTO_UDP:
 749                 result = xs_setup_udp(xprt, to);
 750                 break;
 751         case IPPROTO_TCP:
 752                 result = xs_setup_tcp(xprt, to);
 753                 break;
 754         default:
 755                 printk(KERN_ERR "RPC: unrecognized transport protocol: %d\n",
 756                                 proto);
 757                 result = -EIO;
 758                 break;
 759         }
 760         if (result) {
 761                 kfree(xprt);
 762                 return ERR_PTR(result);
 763         }
 764
 765         spin_lock_init(&xprt->transport_lock);
 766         spin_lock_init(&xprt->reserve_lock);
 767         init_waitqueue_head(&xprt->cong_wait);
 768
 769         INIT_LIST_HEAD(&xprt->free);
 770         INIT_LIST_HEAD(&xprt->recv);
 771         INIT_WORK(&xprt->task_cleanup, xprt_autoclose, xprt);
 772         init_timer(&xprt->timer);
 773         xprt->timer.function = xprt_init_autodisconnect;
 774         xprt->timer.data = (unsigned long) xprt;
 775         xprt->last_used = jiffies;
 776
 777         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
 778         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
 779         rpc_init_wait_queue(&xprt->resend, "xprt_resend");
 780         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
 781
 782         /* initialize free list */
 783         for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
 784                 list_add(&req->rq_list, &xprt->free);
 785
 786         xprt_init_xid(xprt);
 787
 788         dprintk("RPC:      created transport %p with %u slots\n", xprt,
 789                         xprt->max_reqs);
 790
 791         return xprt;
 792 }
 793
 794 /**
 795  * xprt_create_proto - create an RPC client transport
 796  * @proto: requested transport protocol
 797  * @sap: remote peer's address
 798  * @to: timeout parameters for new transport
 799  *
 800  */
 801 struct rpc_xprt *xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to)
 802 {
 803         struct rpc_xprt *xprt;
 804
 805         xprt = xprt_setup(proto, sap, to);
 806         if (IS_ERR(xprt))
 807                 dprintk("RPC:      xprt_create_proto failed\n");
 808         else
 809                 dprintk("RPC:      xprt_create_proto created xprt %p\n", xprt);
 810         return xprt;
 811 }
 812
 813 static void xprt_shutdown(struct rpc_xprt *xprt)
 814 {
 815         xprt->shutdown = 1;
 816         rpc_wake_up(&xprt->sending);
 817         rpc_wake_up(&xprt->resend);
 818         xprt_wake_pending_tasks(xprt, -EIO);
 819         rpc_wake_up(&xprt->backlog);
 820         wake_up(&xprt->cong_wait);
 821         del_timer_sync(&xprt->timer);
 822 }
 823
 824 static int xprt_clear_backlog(struct rpc_xprt *xprt) {
 825         rpc_wake_up_next(&xprt->backlog);
 826         wake_up(&xprt->cong_wait);
 827         return 1;
 828 }
 829
 830 /**
 831  * xprt_destroy - destroy an RPC transport, killing off all requests.
 832  * @xprt: transport to destroy
 833  *
 834  */
 835 int xprt_destroy(struct rpc_xprt *xprt)
 836 {
 837         dprintk("RPC:      destroying transport %p\n", xprt);
 838         xprt_shutdown(xprt);
 839         xprt->ops->destroy(xprt);
 840         kfree(xprt);
 841
 842         return 0;
 843 }