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 #include <linux/sunrpc/metrics.h>
  48
  49 /*
  50  * Local variables
  51  */
  52
  53 #ifdef RPC_DEBUG
  54 # undef  RPC_DEBUG_DATA
  55 # define RPCDBG_FACILITY        RPCDBG_XPRT
  56 #endif
  57
  58 /*
  59  * Local functions
  60  */
  61 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  62 static inline void      do_xprt_reserve(struct rpc_task *);
  63 static void     xprt_connect_status(struct rpc_task *task);
  64 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
  65
  66 /*
  67  * The transport code maintains an estimate on the maximum number of out-
  68  * standing RPC requests, using a smoothed version of the congestion
  69  * avoidance implemented in 44BSD. This is basically the Van Jacobson
  70  * congestion algorithm: If a retransmit occurs, the congestion window is
  71  * halved; otherwise, it is incremented by 1/cwnd when
  72  *
  73  *      -       a reply is received and
  74  *      -       a full number of requests are outstanding and
  75  *      -       the congestion window hasn't been updated recently.
  76  */
  77 #define RPC_CWNDSHIFT           (8U)
  78 #define RPC_CWNDSCALE           (1U << RPC_CWNDSHIFT)
  79 #define RPC_INITCWND            RPC_CWNDSCALE
  80 #define RPC_MAXCWND(xprt)       ((xprt)->max_reqs << RPC_CWNDSHIFT)
  81
  82 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
  83
  84 /**
  85  * xprt_reserve_xprt - serialize write access to transports
  86  * @task: task that is requesting access to the transport
  87  *
  88  * This prevents mixing the payload of separate requests, and prevents
  89  * transport connects from colliding with writes.  No congestion control
  90  * is provided.
  91  */
  92 int xprt_reserve_xprt(struct rpc_task *task)
  93 {
  94         struct rpc_xprt *xprt = task->tk_xprt;
  95         struct rpc_rqst *req = task->tk_rqstp;
  96
  97         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
  98                 if (task == xprt->snd_task)
  99                         return 1;
 100                 if (task == NULL)
 101                         return 0;
 102                 goto out_sleep;
 103         }
 104         xprt->snd_task = task;
 105         if (req) {
 106                 req->rq_bytes_sent = 0;
 107                 req->rq_ntrans++;
 108         }
 109         return 1;
 110
 111 out_sleep:
 112         dprintk("RPC: %4d failed to lock transport %p\n",
 113                         task->tk_pid, xprt);
 114         task->tk_timeout = 0;
 115         task->tk_status = -EAGAIN;
 116         if (req && req->rq_ntrans)
 117                 rpc_sleep_on(&xprt->resend, task, NULL, NULL);
 118         else
 119                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
 120         return 0;
 121 }
 122
 123 static void xprt_clear_locked(struct rpc_xprt *xprt)
 124 {
 125         xprt->snd_task = NULL;
 126         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state) || xprt->shutdown) {
 127                 smp_mb__before_clear_bit();
 128                 clear_bit(XPRT_LOCKED, &xprt->state);
 129                 smp_mb__after_clear_bit();
 130         } else
 131                 schedule_work(&xprt->task_cleanup);
 132 }
 133
 134 /*
 135  * xprt_reserve_xprt_cong - serialize write access to transports
 136  * @task: task that is requesting access to the transport
 137  *
 138  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 139  * integrated into the decision of whether a request is allowed to be
 140  * woken up and given access to the transport.
 141  */
 142 int xprt_reserve_xprt_cong(struct rpc_task *task)
 143 {
 144         struct rpc_xprt *xprt = task->tk_xprt;
 145         struct rpc_rqst *req = task->tk_rqstp;
 146
 147         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 148                 if (task == xprt->snd_task)
 149                         return 1;
 150                 goto out_sleep;
 151         }
 152         if (__xprt_get_cong(xprt, task)) {
 153                 xprt->snd_task = task;
 154                 if (req) {
 155                         req->rq_bytes_sent = 0;
 156                         req->rq_ntrans++;
 157                 }
 158                 return 1;
 159         }
 160         xprt_clear_locked(xprt);
 161 out_sleep:
 162         dprintk("RPC: %4d failed to lock transport %p\n", task->tk_pid, xprt);
 163         task->tk_timeout = 0;
 164         task->tk_status = -EAGAIN;
 165         if (req && req->rq_ntrans)
 166                 rpc_sleep_on(&xprt->resend, task, NULL, NULL);
 167         else
 168                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
 169         return 0;
 170 }
 171
 172 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 173 {
 174         int retval;
 175
 176         spin_lock_bh(&xprt->transport_lock);
 177         retval = xprt->ops->reserve_xprt(task);
 178         spin_unlock_bh(&xprt->transport_lock);
 179         return retval;
 180 }
 181
 182 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 183 {
 184         struct rpc_task *task;
 185         struct rpc_rqst *req;
 186
 187         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 188                 return;
 189
 190         task = rpc_wake_up_next(&xprt->resend);
 191         if (!task) {
 192                 task = rpc_wake_up_next(&xprt->sending);
 193                 if (!task)
 194                         goto out_unlock;
 195         }
 196
 197         req = task->tk_rqstp;
 198         xprt->snd_task = task;
 199         if (req) {
 200                 req->rq_bytes_sent = 0;
 201                 req->rq_ntrans++;
 202         }
 203         return;
 204
 205 out_unlock:
 206         xprt_clear_locked(xprt);
 207 }
 208
 209 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 210 {
 211         struct rpc_task *task;
 212
 213         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 214                 return;
 215         if (RPCXPRT_CONGESTED(xprt))
 216                 goto out_unlock;
 217         task = rpc_wake_up_next(&xprt->resend);
 218         if (!task) {
 219                 task = rpc_wake_up_next(&xprt->sending);
 220                 if (!task)
 221                         goto out_unlock;
 222         }
 223         if (__xprt_get_cong(xprt, task)) {
 224                 struct rpc_rqst *req = task->tk_rqstp;
 225                 xprt->snd_task = task;
 226                 if (req) {
 227                         req->rq_bytes_sent = 0;
 228                         req->rq_ntrans++;
 229                 }
 230                 return;
 231         }
 232 out_unlock:
 233         xprt_clear_locked(xprt);
 234 }
 235
 236 /**
 237  * xprt_release_xprt - allow other requests to use a transport
 238  * @xprt: transport with other tasks potentially waiting
 239  * @task: task that is releasing access to the transport
 240  *
 241  * Note that "task" can be NULL.  No congestion control is provided.
 242  */
 243 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 244 {
 245         if (xprt->snd_task == task) {
 246                 xprt_clear_locked(xprt);
 247                 __xprt_lock_write_next(xprt);
 248         }
 249 }
 250
 251 /**
 252  * xprt_release_xprt_cong - allow other requests to use a transport
 253  * @xprt: transport with other tasks potentially waiting
 254  * @task: task that is releasing access to the transport
 255  *
 256  * Note that "task" can be NULL.  Another task is awoken to use the
 257  * transport if the transport's congestion window allows it.
 258  */
 259 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 260 {
 261         if (xprt->snd_task == task) {
 262                 xprt_clear_locked(xprt);
 263                 __xprt_lock_write_next_cong(xprt);
 264         }
 265 }
 266
 267 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 268 {
 269         spin_lock_bh(&xprt->transport_lock);
 270         xprt->ops->release_xprt(xprt, task);
 271         spin_unlock_bh(&xprt->transport_lock);
 272 }
 273
 274 /*
 275  * Van Jacobson congestion avoidance. Check if the congestion window
 276  * overflowed. Put the task to sleep if this is the case.
 277  */
 278 static int
 279 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 280 {
 281         struct rpc_rqst *req = task->tk_rqstp;
 282
 283         if (req->rq_cong)
 284                 return 1;
 285         dprintk("RPC: %4d xprt_cwnd_limited cong = %ld cwnd = %ld\n",
 286                         task->tk_pid, xprt->cong, xprt->cwnd);
 287         if (RPCXPRT_CONGESTED(xprt))
 288                 return 0;
 289         req->rq_cong = 1;
 290         xprt->cong += RPC_CWNDSCALE;
 291         return 1;
 292 }
 293
 294 /*
 295  * Adjust the congestion window, and wake up the next task
 296  * that has been sleeping due to congestion
 297  */
 298 static void
 299 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 300 {
 301         if (!req->rq_cong)
 302                 return;
 303         req->rq_cong = 0;
 304         xprt->cong -= RPC_CWNDSCALE;
 305         __xprt_lock_write_next_cong(xprt);
 306 }
 307
 308 /**
 309  * xprt_release_rqst_cong - housekeeping when request is complete
 310  * @task: RPC request that recently completed
 311  *
 312  * Useful for transports that require congestion control.
 313  */
 314 void xprt_release_rqst_cong(struct rpc_task *task)
 315 {
 316         __xprt_put_cong(task->tk_xprt, task->tk_rqstp);
 317 }
 318
 319 /**
 320  * xprt_adjust_cwnd - adjust transport congestion window
 321  * @task: recently completed RPC request used to adjust window
 322  * @result: result code of completed RPC request
 323  *
 324  * We use a time-smoothed congestion estimator to avoid heavy oscillation.
 325  */
 326 void xprt_adjust_cwnd(struct rpc_task *task, int result)
 327 {
 328         struct rpc_rqst *req = task->tk_rqstp;
 329         struct rpc_xprt *xprt = task->tk_xprt;
 330         unsigned long cwnd = xprt->cwnd;
 331
 332         if (result >= 0 && cwnd <= xprt->cong) {
 333                 /* The (cwnd >> 1) term makes sure
 334                  * the result gets rounded properly. */
 335                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 336                 if (cwnd > RPC_MAXCWND(xprt))
 337                         cwnd = RPC_MAXCWND(xprt);
 338                 __xprt_lock_write_next_cong(xprt);
 339         } else if (result == -ETIMEDOUT) {
 340                 cwnd >>= 1;
 341                 if (cwnd < RPC_CWNDSCALE)
 342                         cwnd = RPC_CWNDSCALE;
 343         }
 344         dprintk("RPC:      cong %ld, cwnd was %ld, now %ld\n",
 345                         xprt->cong, xprt->cwnd, cwnd);
 346         xprt->cwnd = cwnd;
 347         __xprt_put_cong(xprt, req);
 348 }
 349
 350 /**
 351  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 352  * @xprt: transport with waiting tasks
 353  * @status: result code to plant in each task before waking it
 354  *
 355  */
 356 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 357 {
 358         if (status < 0)
 359                 rpc_wake_up_status(&xprt->pending, status);
 360         else
 361                 rpc_wake_up(&xprt->pending);
 362 }
 363
 364 /**
 365  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 366  * @task: task to be put to sleep
 367  *
 368  */
 369 void xprt_wait_for_buffer_space(struct rpc_task *task)
 370 {
 371         struct rpc_rqst *req = task->tk_rqstp;
 372         struct rpc_xprt *xprt = req->rq_xprt;
 373
 374         task->tk_timeout = req->rq_timeout;
 375         rpc_sleep_on(&xprt->pending, task, NULL, NULL);
 376 }
 377
 378 /**
 379  * xprt_write_space - wake the task waiting for transport output buffer space
 380  * @xprt: transport with waiting tasks
 381  *
 382  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 383  */
 384 void xprt_write_space(struct rpc_xprt *xprt)
 385 {
 386         if (unlikely(xprt->shutdown))
 387                 return;
 388
 389         spin_lock_bh(&xprt->transport_lock);
 390         if (xprt->snd_task) {
 391                 dprintk("RPC:      write space: waking waiting task on xprt %p\n",
 392                                 xprt);
 393                 rpc_wake_up_task(xprt->snd_task);
 394         }
 395         spin_unlock_bh(&xprt->transport_lock);
 396 }
 397
 398 /**
 399  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
 400  * @task: task whose timeout is to be set
 401  *
 402  * Set a request's retransmit timeout based on the transport's
 403  * default timeout parameters.  Used by transports that don't adjust
 404  * the retransmit timeout based on round-trip time estimation.
 405  */
 406 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 407 {
 408         task->tk_timeout = task->tk_rqstp->rq_timeout;
 409 }
 410
 411 /*
 412  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
 413  * @task: task whose timeout is to be set
 414  *
 415  * Set a request's retransmit timeout using the RTT estimator.
 416  */
 417 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
 418 {
 419         int timer = task->tk_msg.rpc_proc->p_timer;
 420         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 421         struct rpc_rqst *req = task->tk_rqstp;
 422         unsigned long max_timeout = req->rq_xprt->timeout.to_maxval;
 423
 424         task->tk_timeout = rpc_calc_rto(rtt, timer);
 425         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
 426         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
 427                 task->tk_timeout = max_timeout;
 428 }
 429
 430 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 431 {
 432         struct rpc_timeout *to = &req->rq_xprt->timeout;
 433
 434         req->rq_majortimeo = req->rq_timeout;
 435         if (to->to_exponential)
 436                 req->rq_majortimeo <<= to->to_retries;
 437         else
 438                 req->rq_majortimeo += to->to_increment * to->to_retries;
 439         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 440                 req->rq_majortimeo = to->to_maxval;
 441         req->rq_majortimeo += jiffies;
 442 }
 443
 444 /**
 445  * xprt_adjust_timeout - adjust timeout values for next retransmit
 446  * @req: RPC request containing parameters to use for the adjustment
 447  *
 448  */
 449 int xprt_adjust_timeout(struct rpc_rqst *req)
 450 {
 451         struct rpc_xprt *xprt = req->rq_xprt;
 452         struct rpc_timeout *to = &xprt->timeout;
 453         int status = 0;
 454
 455         if (time_before(jiffies, req->rq_majortimeo)) {
 456                 if (to->to_exponential)
 457                         req->rq_timeout <<= 1;
 458                 else
 459                         req->rq_timeout += to->to_increment;
 460                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 461                         req->rq_timeout = to->to_maxval;
 462                 req->rq_retries++;
 463                 pprintk("RPC: %lu retrans\n", jiffies);
 464         } else {
 465                 req->rq_timeout = to->to_initval;
 466                 req->rq_retries = 0;
 467                 xprt_reset_majortimeo(req);
 468                 /* Reset the RTT counters == "slow start" */
 469                 spin_lock_bh(&xprt->transport_lock);
 470                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 471                 spin_unlock_bh(&xprt->transport_lock);
 472                 pprintk("RPC: %lu timeout\n", jiffies);
 473                 status = -ETIMEDOUT;
 474         }
 475
 476         if (req->rq_timeout == 0) {
 477                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 478                 req->rq_timeout = 5 * HZ;
 479         }
 480         return status;
 481 }
 482
 483 static void xprt_autoclose(void *args)
 484 {
 485         struct rpc_xprt *xprt = (struct rpc_xprt *)args;
 486
 487         xprt_disconnect(xprt);
 488         xprt->ops->close(xprt);
 489         xprt_release_write(xprt, NULL);
 490 }
 491
 492 /**
 493  * xprt_disconnect - mark a transport as disconnected
 494  * @xprt: transport to flag for disconnect
 495  *
 496  */
 497 void xprt_disconnect(struct rpc_xprt *xprt)
 498 {
 499         dprintk("RPC:      disconnected transport %p\n", xprt);
 500         spin_lock_bh(&xprt->transport_lock);
 501         xprt_clear_connected(xprt);
 502         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 503         spin_unlock_bh(&xprt->transport_lock);
 504 }
 505
 506 static void
 507 xprt_init_autodisconnect(unsigned long data)
 508 {
 509         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 510
 511         spin_lock(&xprt->transport_lock);
 512         if (!list_empty(&xprt->recv) || xprt->shutdown)
 513                 goto out_abort;
 514         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 515                 goto out_abort;
 516         spin_unlock(&xprt->transport_lock);
 517         if (xprt_connecting(xprt))
 518                 xprt_release_write(xprt, NULL);
 519         else
 520                 schedule_work(&xprt->task_cleanup);
 521         return;
 522 out_abort:
 523         spin_unlock(&xprt->transport_lock);
 524 }
 525
 526 /**
 527  * xprt_connect - schedule a transport connect operation
 528  * @task: RPC task that is requesting the connect
 529  *
 530  */
 531 void xprt_connect(struct rpc_task *task)
 532 {
 533         struct rpc_xprt *xprt = task->tk_xprt;
 534
 535         dprintk("RPC: %4d xprt_connect xprt %p %s connected\n", task->tk_pid,
 536                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 537
 538         if (!xprt->addr.sin_port) {
 539                 task->tk_status = -EIO;
 540                 return;
 541         }
 542         if (!xprt_lock_write(xprt, task))
 543                 return;
 544         if (xprt_connected(xprt))
 545                 xprt_release_write(xprt, task);
 546         else {
 547                 if (task->tk_rqstp)
 548                         task->tk_rqstp->rq_bytes_sent = 0;
 549
 550                 task->tk_timeout = xprt->connect_timeout;
 551                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
 552                 xprt->stat.connect_start = jiffies;
 553                 xprt->ops->connect(task);
 554         }
 555         return;
 556 }
 557
 558 static void xprt_connect_status(struct rpc_task *task)
 559 {
 560         struct rpc_xprt *xprt = task->tk_xprt;
 561
 562         if (task->tk_status >= 0) {
 563                 xprt->stat.connect_count++;
 564                 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
 565                 dprintk("RPC: %4d xprt_connect_status: connection established\n",
 566                                 task->tk_pid);
 567                 return;
 568         }
 569
 570         switch (task->tk_status) {
 571         case -ECONNREFUSED:
 572         case -ECONNRESET:
 573                 dprintk("RPC: %4d xprt_connect_status: server %s refused connection\n",
 574                                 task->tk_pid, task->tk_client->cl_server);
 575                 break;
 576         case -ENOTCONN:
 577                 dprintk("RPC: %4d xprt_connect_status: connection broken\n",
 578                                 task->tk_pid);
 579                 break;
 580         case -ETIMEDOUT:
 581                 dprintk("RPC: %4d xprt_connect_status: connect attempt timed out\n",
 582                                 task->tk_pid);
 583                 break;
 584         default:
 585                 dprintk("RPC: %4d xprt_connect_status: error %d connecting to server %s\n",
 586                                 task->tk_pid, -task->tk_status, task->tk_client->cl_server);
 587                 xprt_release_write(xprt, task);
 588                 task->tk_status = -EIO;
 589                 return;
 590         }
 591
 592         /* if soft mounted, just cause this RPC to fail */
 593         if (RPC_IS_SOFT(task)) {
 594                 xprt_release_write(xprt, task);
 595                 task->tk_status = -EIO;
 596         }
 597 }
 598
 599 /**
 600  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 601  * @xprt: transport on which the original request was transmitted
 602  * @xid: RPC XID of incoming reply
 603  *
 604  */
 605 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
 606 {
 607         struct list_head *pos;
 608
 609         list_for_each(pos, &xprt->recv) {
 610                 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
 611                 if (entry->rq_xid == xid)
 612                         return entry;
 613         }
 614         xprt->stat.bad_xids++;
 615         return NULL;
 616 }
 617
 618 /**
 619  * xprt_update_rtt - update an RPC client's RTT state after receiving a reply
 620  * @task: RPC request that recently completed
 621  *
 622  */
 623 void xprt_update_rtt(struct rpc_task *task)
 624 {
 625         struct rpc_rqst *req = task->tk_rqstp;
 626         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 627         unsigned timer = task->tk_msg.rpc_proc->p_timer;
 628
 629         if (timer) {
 630                 if (req->rq_ntrans == 1)
 631                         rpc_update_rtt(rtt, timer,
 632                                         (long)jiffies - req->rq_xtime);
 633                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 634         }
 635 }
 636
 637 /**
 638  * xprt_complete_rqst - called when reply processing is complete
 639  * @task: RPC request that recently completed
 640  * @copied: actual number of bytes received from the transport
 641  *
 642  * Caller holds transport lock.
 643  */
 644 void xprt_complete_rqst(struct rpc_task *task, int copied)
 645 {
 646         struct rpc_rqst *req = task->tk_rqstp;
 647
 648         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 649                         task->tk_pid, ntohl(req->rq_xid), copied);
 650
 651         task->tk_xprt->stat.recvs++;
 652         task->tk_rtt = (long)jiffies - req->rq_xtime;
 653
 654         list_del_init(&req->rq_list);
 655         req->rq_received = req->rq_private_buf.len = copied;
 656         rpc_wake_up_task(task);
 657 }
 658
 659 static void xprt_timer(struct rpc_task *task)
 660 {
 661         struct rpc_rqst *req = task->tk_rqstp;
 662         struct rpc_xprt *xprt = req->rq_xprt;
 663
 664         dprintk("RPC: %4d xprt_timer\n", task->tk_pid);
 665
 666         spin_lock(&xprt->transport_lock);
 667         if (!req->rq_received) {
 668                 if (xprt->ops->timer)
 669                         xprt->ops->timer(task);
 670                 task->tk_status = -ETIMEDOUT;
 671         }
 672         task->tk_timeout = 0;
 673         rpc_wake_up_task(task);
 674         spin_unlock(&xprt->transport_lock);
 675 }
 676
 677 /**
 678  * xprt_prepare_transmit - reserve the transport before sending a request
 679  * @task: RPC task about to send a request
 680  *
 681  */
 682 int xprt_prepare_transmit(struct rpc_task *task)
 683 {
 684         struct rpc_rqst *req = task->tk_rqstp;
 685         struct rpc_xprt *xprt = req->rq_xprt;
 686         int err = 0;
 687
 688         dprintk("RPC: %4d xprt_prepare_transmit\n", task->tk_pid);
 689
 690         spin_lock_bh(&xprt->transport_lock);
 691         if (req->rq_received && !req->rq_bytes_sent) {
 692                 err = req->rq_received;
 693                 goto out_unlock;
 694         }
 695         if (!xprt->ops->reserve_xprt(task)) {
 696                 err = -EAGAIN;
 697                 goto out_unlock;
 698         }
 699
 700         if (!xprt_connected(xprt)) {
 701                 err = -ENOTCONN;
 702                 goto out_unlock;
 703         }
 704 out_unlock:
 705         spin_unlock_bh(&xprt->transport_lock);
 706         return err;
 707 }
 708
 709 void
 710 xprt_abort_transmit(struct rpc_task *task)
 711 {
 712         struct rpc_xprt *xprt = task->tk_xprt;
 713
 714         xprt_release_write(xprt, task);
 715 }
 716
 717 /**
 718  * xprt_transmit - send an RPC request on a transport
 719  * @task: controlling RPC task
 720  *
 721  * We have to copy the iovec because sendmsg fiddles with its contents.
 722  */
 723 void xprt_transmit(struct rpc_task *task)
 724 {
 725         struct rpc_rqst *req = task->tk_rqstp;
 726         struct rpc_xprt *xprt = req->rq_xprt;
 727         int status;
 728
 729         dprintk("RPC: %4d xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 730
 731         smp_rmb();
 732         if (!req->rq_received) {
 733                 if (list_empty(&req->rq_list)) {
 734                         spin_lock_bh(&xprt->transport_lock);
 735                         /* Update the softirq receive buffer */
 736                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 737                                         sizeof(req->rq_private_buf));
 738                         /* Add request to the receive list */
 739                         list_add_tail(&req->rq_list, &xprt->recv);
 740                         spin_unlock_bh(&xprt->transport_lock);
 741                         xprt_reset_majortimeo(req);
 742                         /* Turn off autodisconnect */
 743                         del_singleshot_timer_sync(&xprt->timer);
 744                 }
 745         } else if (!req->rq_bytes_sent)
 746                 return;
 747
 748         status = xprt->ops->send_request(task);
 749         if (status == 0) {
 750                 dprintk("RPC: %4d xmit complete\n", task->tk_pid);
 751                 spin_lock_bh(&xprt->transport_lock);
 752
 753                 xprt->ops->set_retrans_timeout(task);
 754
 755                 xprt->stat.sends++;
 756                 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
 757                 xprt->stat.bklog_u += xprt->backlog.qlen;
 758
 759                 /* Don't race with disconnect */
 760                 if (!xprt_connected(xprt))
 761                         task->tk_status = -ENOTCONN;
 762                 else if (!req->rq_received)
 763                         rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer);
 764
 765                 xprt->ops->release_xprt(xprt, task);
 766                 spin_unlock_bh(&xprt->transport_lock);
 767                 return;
 768         }
 769
 770         /* Note: at this point, task->tk_sleeping has not yet been set,
 771          *       hence there is no danger of the waking up task being put on
 772          *       schedq, and being picked up by a parallel run of rpciod().
 773          */
 774         task->tk_status = status;
 775
 776         switch (status) {
 777         case -ECONNREFUSED:
 778                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
 779         case -EAGAIN:
 780         case -ENOTCONN:
 781                 return;
 782         default:
 783                 break;
 784         }
 785         xprt_release_write(xprt, task);
 786         return;
 787 }
 788
 789 static inline void do_xprt_reserve(struct rpc_task *task)
 790 {
 791         struct rpc_xprt *xprt = task->tk_xprt;
 792
 793         task->tk_status = 0;
 794         if (task->tk_rqstp)
 795                 return;
 796         if (!list_empty(&xprt->free)) {
 797                 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
 798                 list_del_init(&req->rq_list);
 799                 task->tk_rqstp = req;
 800                 xprt_request_init(task, xprt);
 801                 return;
 802         }
 803         dprintk("RPC:      waiting for request slot\n");
 804         task->tk_status = -EAGAIN;
 805         task->tk_timeout = 0;
 806         rpc_sleep_on(&xprt->backlog, task, NULL, NULL);
 807 }
 808
 809 /**
 810  * xprt_reserve - allocate an RPC request slot
 811  * @task: RPC task requesting a slot allocation
 812  *
 813  * If no more slots are available, place the task on the transport's
 814  * backlog queue.
 815  */
 816 void xprt_reserve(struct rpc_task *task)
 817 {
 818         struct rpc_xprt *xprt = task->tk_xprt;
 819
 820         task->tk_status = -EIO;
 821         spin_lock(&xprt->reserve_lock);
 822         do_xprt_reserve(task);
 823         spin_unlock(&xprt->reserve_lock);
 824 }
 825
 826 static inline u32 xprt_alloc_xid(struct rpc_xprt *xprt)
 827 {
 828         return xprt->xid++;
 829 }
 830
 831 static inline void xprt_init_xid(struct rpc_xprt *xprt)
 832 {
 833         get_random_bytes(&xprt->xid, sizeof(xprt->xid));
 834 }
 835
 836 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
 837 {
 838         struct rpc_rqst *req = task->tk_rqstp;
 839
 840         req->rq_timeout = xprt->timeout.to_initval;
 841         req->rq_task    = task;
 842         req->rq_xprt    = xprt;
 843         req->rq_buffer  = NULL;
 844         req->rq_bufsize = 0;
 845         req->rq_xid     = xprt_alloc_xid(xprt);
 846         req->rq_release_snd_buf = NULL;
 847         dprintk("RPC: %4d reserved req %p xid %08x\n", task->tk_pid,
 848                         req, ntohl(req->rq_xid));
 849 }
 850
 851 /**
 852  * xprt_release - release an RPC request slot
 853  * @task: task which is finished with the slot
 854  *
 855  */
 856 void xprt_release(struct rpc_task *task)
 857 {
 858         struct rpc_xprt *xprt = task->tk_xprt;
 859         struct rpc_rqst *req;
 860
 861         if (!(req = task->tk_rqstp))
 862                 return;
 863         rpc_count_iostats(task);
 864         spin_lock_bh(&xprt->transport_lock);
 865         xprt->ops->release_xprt(xprt, task);
 866         if (xprt->ops->release_request)
 867                 xprt->ops->release_request(task);
 868         if (!list_empty(&req->rq_list))
 869                 list_del(&req->rq_list);
 870         xprt->last_used = jiffies;
 871         if (list_empty(&xprt->recv))
 872                 mod_timer(&xprt->timer,
 873                                 xprt->last_used + xprt->idle_timeout);
 874         spin_unlock_bh(&xprt->transport_lock);
 875         xprt->ops->buf_free(task);
 876         task->tk_rqstp = NULL;
 877         if (req->rq_release_snd_buf)
 878                 req->rq_release_snd_buf(req);
 879         memset(req, 0, sizeof(*req));   /* mark unused */
 880
 881         dprintk("RPC: %4d release request %p\n", task->tk_pid, req);
 882
 883         spin_lock(&xprt->reserve_lock);
 884         list_add(&req->rq_list, &xprt->free);
 885         rpc_wake_up_next(&xprt->backlog);
 886         spin_unlock(&xprt->reserve_lock);
 887 }
 888
 889 /**
 890  * xprt_set_timeout - set constant RPC timeout
 891  * @to: RPC timeout parameters to set up
 892  * @retr: number of retries
 893  * @incr: amount of increase after each retry
 894  *
 895  */
 896 void xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
 897 {
 898         to->to_initval   =
 899         to->to_increment = incr;
 900         to->to_maxval    = to->to_initval + (incr * retr);
 901         to->to_retries   = retr;
 902         to->to_exponential = 0;
 903 }
 904
 905 static struct rpc_xprt *xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
 906 {
 907         int result;
 908         struct rpc_xprt *xprt;
 909         struct rpc_rqst *req;
 910
 911         if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL)
 912                 return ERR_PTR(-ENOMEM);
 913         memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */
 914
 915         xprt->addr = *ap;
 916
 917         switch (proto) {
 918         case IPPROTO_UDP:
 919                 result = xs_setup_udp(xprt, to);
 920                 break;
 921         case IPPROTO_TCP:
 922                 result = xs_setup_tcp(xprt, to);
 923                 break;
 924         default:
 925                 printk(KERN_ERR "RPC: unrecognized transport protocol: %d\n",
 926                                 proto);
 927                 result = -EIO;
 928                 break;
 929         }
 930         if (result) {
 931                 kfree(xprt);
 932                 return ERR_PTR(result);
 933         }
 934
 935         spin_lock_init(&xprt->transport_lock);
 936         spin_lock_init(&xprt->reserve_lock);
 937
 938         INIT_LIST_HEAD(&xprt->free);
 939         INIT_LIST_HEAD(&xprt->recv);
 940         INIT_WORK(&xprt->task_cleanup, xprt_autoclose, xprt);
 941         init_timer(&xprt->timer);
 942         xprt->timer.function = xprt_init_autodisconnect;
 943         xprt->timer.data = (unsigned long) xprt;
 944         xprt->last_used = jiffies;
 945         xprt->cwnd = RPC_INITCWND;
 946
 947         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
 948         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
 949         rpc_init_wait_queue(&xprt->resend, "xprt_resend");
 950         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
 951
 952         /* initialize free list */
 953         for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
 954                 list_add(&req->rq_list, &xprt->free);
 955
 956         xprt_init_xid(xprt);
 957
 958         dprintk("RPC:      created transport %p with %u slots\n", xprt,
 959                         xprt->max_reqs);
 960
 961         return xprt;
 962 }
 963
 964 /**
 965  * xprt_create_proto - create an RPC client transport
 966  * @proto: requested transport protocol
 967  * @sap: remote peer's address
 968  * @to: timeout parameters for new transport
 969  *
 970  */
 971 struct rpc_xprt *xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to)
 972 {
 973         struct rpc_xprt *xprt;
 974
 975         xprt = xprt_setup(proto, sap, to);
 976         if (IS_ERR(xprt))
 977                 dprintk("RPC:      xprt_create_proto failed\n");
 978         else
 979                 dprintk("RPC:      xprt_create_proto created xprt %p\n", xprt);
 980         return xprt;
 981 }
 982
 983 /**
 984  * xprt_destroy - destroy an RPC transport, killing off all requests.
 985  * @xprt: transport to destroy
 986  *
 987  */
 988 int xprt_destroy(struct rpc_xprt *xprt)
 989 {
 990         dprintk("RPC:      destroying transport %p\n", xprt);
 991         xprt->shutdown = 1;
 992         del_timer_sync(&xprt->timer);
 993         xprt->ops->destroy(xprt);
 994         kfree(xprt);
 995
 996         return 0;
 997 }