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, if a reply is expected,
  16  *      it installs a timer that is run after the packet's timeout has
  17  *      expired.
  18  *  -   When a packet arrives, the data_ready handler walks the list of
  19  *      pending requests for that transport. If a matching XID is found, the
  20  *      caller is woken up, and the timer removed.
  21  *  -   When no reply arrives within the timeout interval, the timer is
  22  *      fired by the kernel and runs xprt_timer(). It either adjusts the
  23  *      timeout values (minor timeout) or wakes up the caller with a status
  24  *      of -ETIMEDOUT.
  25  *  -   When the caller receives a notification from RPC that a reply arrived,
  26  *      it should release the RPC slot, and process the reply.
  27  *      If the call timed out, it may choose to retry the operation by
  28  *      adjusting the initial timeout value, and simply calling rpc_call
  29  *      again.
  30  *
  31  *  Support for async RPC is done through a set of RPC-specific scheduling
  32  *  primitives that `transparently' work for processes as well as async
  33  *  tasks that rely on callbacks.
  34  *
  35  *  Copyright (C) 1995-1997, Olaf Kirch <[email protected]>
  36  *
  37  *  Transport switch API copyright (C) 2005, Chuck Lever <[email protected]>
  38  */
  39
  40 #include <linux/module.h>
  41
  42 #include <linux/types.h>
  43 #include <linux/interrupt.h>
  44 #include <linux/workqueue.h>
  45 #include <linux/net.h>
  46 #include <linux/ktime.h>
  47
  48 #include <linux/sunrpc/clnt.h>
  49 #include <linux/sunrpc/metrics.h>
  50 #include <linux/sunrpc/bc_xprt.h>
  51
  52 #include <trace/events/sunrpc.h>
  53
  54 #include "sunrpc.h"
  55
  56 /*
  57  * Local variables
  58  */
  59
  60 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  61 # define RPCDBG_FACILITY        RPCDBG_XPRT
  62 #endif
  63
  64 /*
  65  * Local functions
  66  */
  67 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
  68 static void     xprt_request_init(struct rpc_task *, struct rpc_xprt *);
  69 static void     xprt_connect_status(struct rpc_task *task);
  70 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
  71 static void      xprt_destroy(struct rpc_xprt *xprt);
  72
  73 static DEFINE_SPINLOCK(xprt_list_lock);
  74 static LIST_HEAD(xprt_list);
  75
  76 /**
  77  * xprt_register_transport - register a transport implementation
  78  * @transport: transport to register
  79  *
  80  * If a transport implementation is loaded as a kernel module, it can
  81  * call this interface to make itself known to the RPC client.
  82  *
  83  * Returns:
  84  * 0:           transport successfully registered
  85  * -EEXIST:     transport already registered
  86  * -EINVAL:     transport module being unloaded
  87  */
  88 int xprt_register_transport(struct xprt_class *transport)
  89 {
  90         struct xprt_class *t;
  91         int result;
  92
  93         result = -EEXIST;
  94         spin_lock(&xprt_list_lock);
  95         list_for_each_entry(t, &xprt_list, list) {
  96                 /* don't register the same transport class twice */
  97                 if (t->ident == transport->ident)
  98                         goto out;
  99         }
 100
 101         list_add_tail(&transport->list, &xprt_list);
 102         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 103                transport->name);
 104         result = 0;
 105
 106 out:
 107         spin_unlock(&xprt_list_lock);
 108         return result;
 109 }
 110 EXPORT_SYMBOL_GPL(xprt_register_transport);
 111
 112 /**
 113  * xprt_unregister_transport - unregister a transport implementation
 114  * @transport: transport to unregister
 115  *
 116  * Returns:
 117  * 0:           transport successfully unregistered
 118  * -ENOENT:     transport never registered
 119  */
 120 int xprt_unregister_transport(struct xprt_class *transport)
 121 {
 122         struct xprt_class *t;
 123         int result;
 124
 125         result = 0;
 126         spin_lock(&xprt_list_lock);
 127         list_for_each_entry(t, &xprt_list, list) {
 128                 if (t == transport) {
 129                         printk(KERN_INFO
 130                                 "RPC: Unregistered %s transport module.\n",
 131                                 transport->name);
 132                         list_del_init(&transport->list);
 133                         goto out;
 134                 }
 135         }
 136         result = -ENOENT;
 137
 138 out:
 139         spin_unlock(&xprt_list_lock);
 140         return result;
 141 }
 142 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 143
 144 /**
 145  * xprt_load_transport - load a transport implementation
 146  * @transport_name: transport to load
 147  *
 148  * Returns:
 149  * 0:           transport successfully loaded
 150  * -ENOENT:     transport module not available
 151  */
 152 int xprt_load_transport(const char *transport_name)
 153 {
 154         struct xprt_class *t;
 155         int result;
 156
 157         result = 0;
 158         spin_lock(&xprt_list_lock);
 159         list_for_each_entry(t, &xprt_list, list) {
 160                 if (strcmp(t->name, transport_name) == 0) {
 161                         spin_unlock(&xprt_list_lock);
 162                         goto out;
 163                 }
 164         }
 165         spin_unlock(&xprt_list_lock);
 166         result = request_module("xprt%s", transport_name);
 167 out:
 168         return result;
 169 }
 170 EXPORT_SYMBOL_GPL(xprt_load_transport);
 171
 172 /**
 173  * xprt_reserve_xprt - serialize write access to transports
 174  * @task: task that is requesting access to the transport
 175  * @xprt: pointer to the target transport
 176  *
 177  * This prevents mixing the payload of separate requests, and prevents
 178  * transport connects from colliding with writes.  No congestion control
 179  * is provided.
 180  */
 181 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 182 {
 183         struct rpc_rqst *req = task->tk_rqstp;
 184         int priority;
 185
 186         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 187                 if (task == xprt->snd_task)
 188                         return 1;
 189                 goto out_sleep;
 190         }
 191         xprt->snd_task = task;
 192         if (req != NULL)
 193                 req->rq_ntrans++;
 194
 195         return 1;
 196
 197 out_sleep:
 198         dprintk("RPC: %5u failed to lock transport %p\n",
 199                         task->tk_pid, xprt);
 200         task->tk_timeout = 0;
 201         task->tk_status = -EAGAIN;
 202         if (req == NULL)
 203                 priority = RPC_PRIORITY_LOW;
 204         else if (!req->rq_ntrans)
 205                 priority = RPC_PRIORITY_NORMAL;
 206         else
 207                 priority = RPC_PRIORITY_HIGH;
 208         rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
 209         return 0;
 210 }
 211 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 212
 213 static void xprt_clear_locked(struct rpc_xprt *xprt)
 214 {
 215         xprt->snd_task = NULL;
 216         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 217                 smp_mb__before_atomic();
 218                 clear_bit(XPRT_LOCKED, &xprt->state);
 219                 smp_mb__after_atomic();
 220         } else
 221                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 222 }
 223
 224 /*
 225  * xprt_reserve_xprt_cong - serialize write access to transports
 226  * @task: task that is requesting access to the transport
 227  *
 228  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 229  * integrated into the decision of whether a request is allowed to be
 230  * woken up and given access to the transport.
 231  */
 232 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 233 {
 234         struct rpc_rqst *req = task->tk_rqstp;
 235         int priority;
 236
 237         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 238                 if (task == xprt->snd_task)
 239                         return 1;
 240                 goto out_sleep;
 241         }
 242         if (req == NULL) {
 243                 xprt->snd_task = task;
 244                 return 1;
 245         }
 246         if (__xprt_get_cong(xprt, task)) {
 247                 xprt->snd_task = task;
 248                 req->rq_ntrans++;
 249                 return 1;
 250         }
 251         xprt_clear_locked(xprt);
 252 out_sleep:
 253         dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
 254         task->tk_timeout = 0;
 255         task->tk_status = -EAGAIN;
 256         if (req == NULL)
 257                 priority = RPC_PRIORITY_LOW;
 258         else if (!req->rq_ntrans)
 259                 priority = RPC_PRIORITY_NORMAL;
 260         else
 261                 priority = RPC_PRIORITY_HIGH;
 262         rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
 263         return 0;
 264 }
 265 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 266
 267 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 268 {
 269         int retval;
 270
 271         spin_lock_bh(&xprt->transport_lock);
 272         retval = xprt->ops->reserve_xprt(xprt, task);
 273         spin_unlock_bh(&xprt->transport_lock);
 274         return retval;
 275 }
 276
 277 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 278 {
 279         struct rpc_xprt *xprt = data;
 280         struct rpc_rqst *req;
 281
 282         req = task->tk_rqstp;
 283         xprt->snd_task = task;
 284         if (req)
 285                 req->rq_ntrans++;
 286         return true;
 287 }
 288
 289 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 290 {
 291         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 292                 return;
 293
 294         if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_func, xprt))
 295                 return;
 296         xprt_clear_locked(xprt);
 297 }
 298
 299 static bool __xprt_lock_write_cong_func(struct rpc_task *task, void *data)
 300 {
 301         struct rpc_xprt *xprt = data;
 302         struct rpc_rqst *req;
 303
 304         req = task->tk_rqstp;
 305         if (req == NULL) {
 306                 xprt->snd_task = task;
 307                 return true;
 308         }
 309         if (__xprt_get_cong(xprt, task)) {
 310                 xprt->snd_task = task;
 311                 req->rq_ntrans++;
 312                 return true;
 313         }
 314         return false;
 315 }
 316
 317 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 318 {
 319         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 320                 return;
 321         if (RPCXPRT_CONGESTED(xprt))
 322                 goto out_unlock;
 323         if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_cong_func, xprt))
 324                 return;
 325 out_unlock:
 326         xprt_clear_locked(xprt);
 327 }
 328
 329 static void xprt_task_clear_bytes_sent(struct rpc_task *task)
 330 {
 331         if (task != NULL) {
 332                 struct rpc_rqst *req = task->tk_rqstp;
 333                 if (req != NULL)
 334                         req->rq_bytes_sent = 0;
 335         }
 336 }
 337
 338 /**
 339  * xprt_release_xprt - allow other requests to use a transport
 340  * @xprt: transport with other tasks potentially waiting
 341  * @task: task that is releasing access to the transport
 342  *
 343  * Note that "task" can be NULL.  No congestion control is provided.
 344  */
 345 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 346 {
 347         if (xprt->snd_task == task) {
 348                 xprt_task_clear_bytes_sent(task);
 349                 xprt_clear_locked(xprt);
 350                 __xprt_lock_write_next(xprt);
 351         }
 352 }
 353 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 354
 355 /**
 356  * xprt_release_xprt_cong - allow other requests to use a transport
 357  * @xprt: transport with other tasks potentially waiting
 358  * @task: task that is releasing access to the transport
 359  *
 360  * Note that "task" can be NULL.  Another task is awoken to use the
 361  * transport if the transport's congestion window allows it.
 362  */
 363 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 364 {
 365         if (xprt->snd_task == task) {
 366                 xprt_task_clear_bytes_sent(task);
 367                 xprt_clear_locked(xprt);
 368                 __xprt_lock_write_next_cong(xprt);
 369         }
 370 }
 371 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 372
 373 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 374 {
 375         spin_lock_bh(&xprt->transport_lock);
 376         xprt->ops->release_xprt(xprt, task);
 377         spin_unlock_bh(&xprt->transport_lock);
 378 }
 379
 380 /*
 381  * Van Jacobson congestion avoidance. Check if the congestion window
 382  * overflowed. Put the task to sleep if this is the case.
 383  */
 384 static int
 385 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 386 {
 387         struct rpc_rqst *req = task->tk_rqstp;
 388
 389         if (req->rq_cong)
 390                 return 1;
 391         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
 392                         task->tk_pid, xprt->cong, xprt->cwnd);
 393         if (RPCXPRT_CONGESTED(xprt))
 394                 return 0;
 395         req->rq_cong = 1;
 396         xprt->cong += RPC_CWNDSCALE;
 397         return 1;
 398 }
 399
 400 /*
 401  * Adjust the congestion window, and wake up the next task
 402  * that has been sleeping due to congestion
 403  */
 404 static void
 405 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 406 {
 407         if (!req->rq_cong)
 408                 return;
 409         req->rq_cong = 0;
 410         xprt->cong -= RPC_CWNDSCALE;
 411         __xprt_lock_write_next_cong(xprt);
 412 }
 413
 414 /**
 415  * xprt_release_rqst_cong - housekeeping when request is complete
 416  * @task: RPC request that recently completed
 417  *
 418  * Useful for transports that require congestion control.
 419  */
 420 void xprt_release_rqst_cong(struct rpc_task *task)
 421 {
 422         struct rpc_rqst *req = task->tk_rqstp;
 423
 424         __xprt_put_cong(req->rq_xprt, req);
 425 }
 426 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 427
 428 /**
 429  * xprt_adjust_cwnd - adjust transport congestion window
 430  * @xprt: pointer to xprt
 431  * @task: recently completed RPC request used to adjust window
 432  * @result: result code of completed RPC request
 433  *
 434  * The transport code maintains an estimate on the maximum number of out-
 435  * standing RPC requests, using a smoothed version of the congestion
 436  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 437  * congestion algorithm: If a retransmit occurs, the congestion window is
 438  * halved; otherwise, it is incremented by 1/cwnd when
 439  *
 440  *      -       a reply is received and
 441  *      -       a full number of requests are outstanding and
 442  *      -       the congestion window hasn't been updated recently.
 443  */
 444 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 445 {
 446         struct rpc_rqst *req = task->tk_rqstp;
 447         unsigned long cwnd = xprt->cwnd;
 448
 449         if (result >= 0 && cwnd <= xprt->cong) {
 450                 /* The (cwnd >> 1) term makes sure
 451                  * the result gets rounded properly. */
 452                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 453                 if (cwnd > RPC_MAXCWND(xprt))
 454                         cwnd = RPC_MAXCWND(xprt);
 455                 __xprt_lock_write_next_cong(xprt);
 456         } else if (result == -ETIMEDOUT) {
 457                 cwnd >>= 1;
 458                 if (cwnd < RPC_CWNDSCALE)
 459                         cwnd = RPC_CWNDSCALE;
 460         }
 461         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 462                         xprt->cong, xprt->cwnd, cwnd);
 463         xprt->cwnd = cwnd;
 464         __xprt_put_cong(xprt, req);
 465 }
 466 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 467
 468 /**
 469  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 470  * @xprt: transport with waiting tasks
 471  * @status: result code to plant in each task before waking it
 472  *
 473  */
 474 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 475 {
 476         if (status < 0)
 477                 rpc_wake_up_status(&xprt->pending, status);
 478         else
 479                 rpc_wake_up(&xprt->pending);
 480 }
 481 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 482
 483 /**
 484  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 485  * @task: task to be put to sleep
 486  * @action: function pointer to be executed after wait
 487  *
 488  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 489  * we don't in general want to force a socket disconnection due to
 490  * an incomplete RPC call transmission.
 491  */
 492 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
 493 {
 494         struct rpc_rqst *req = task->tk_rqstp;
 495         struct rpc_xprt *xprt = req->rq_xprt;
 496
 497         task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
 498         rpc_sleep_on(&xprt->pending, task, action);
 499 }
 500 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 501
 502 /**
 503  * xprt_write_space - wake the task waiting for transport output buffer space
 504  * @xprt: transport with waiting tasks
 505  *
 506  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 507  */
 508 void xprt_write_space(struct rpc_xprt *xprt)
 509 {
 510         spin_lock_bh(&xprt->transport_lock);
 511         if (xprt->snd_task) {
 512                 dprintk("RPC:       write space: waking waiting task on "
 513                                 "xprt %p\n", xprt);
 514                 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
 515         }
 516         spin_unlock_bh(&xprt->transport_lock);
 517 }
 518 EXPORT_SYMBOL_GPL(xprt_write_space);
 519
 520 /**
 521  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
 522  * @task: task whose timeout is to be set
 523  *
 524  * Set a request's retransmit timeout based on the transport's
 525  * default timeout parameters.  Used by transports that don't adjust
 526  * the retransmit timeout based on round-trip time estimation.
 527  */
 528 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 529 {
 530         task->tk_timeout = task->tk_rqstp->rq_timeout;
 531 }
 532 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
 533
 534 /**
 535  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
 536  * @task: task whose timeout is to be set
 537  *
 538  * Set a request's retransmit timeout using the RTT estimator.
 539  */
 540 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
 541 {
 542         int timer = task->tk_msg.rpc_proc->p_timer;
 543         struct rpc_clnt *clnt = task->tk_client;
 544         struct rpc_rtt *rtt = clnt->cl_rtt;
 545         struct rpc_rqst *req = task->tk_rqstp;
 546         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
 547
 548         task->tk_timeout = rpc_calc_rto(rtt, timer);
 549         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
 550         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
 551                 task->tk_timeout = max_timeout;
 552 }
 553 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
 554
 555 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 556 {
 557         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 558
 559         req->rq_majortimeo = req->rq_timeout;
 560         if (to->to_exponential)
 561                 req->rq_majortimeo <<= to->to_retries;
 562         else
 563                 req->rq_majortimeo += to->to_increment * to->to_retries;
 564         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 565                 req->rq_majortimeo = to->to_maxval;
 566         req->rq_majortimeo += jiffies;
 567 }
 568
 569 /**
 570  * xprt_adjust_timeout - adjust timeout values for next retransmit
 571  * @req: RPC request containing parameters to use for the adjustment
 572  *
 573  */
 574 int xprt_adjust_timeout(struct rpc_rqst *req)
 575 {
 576         struct rpc_xprt *xprt = req->rq_xprt;
 577         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 578         int status = 0;
 579
 580         if (time_before(jiffies, req->rq_majortimeo)) {
 581                 if (to->to_exponential)
 582                         req->rq_timeout <<= 1;
 583                 else
 584                         req->rq_timeout += to->to_increment;
 585                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 586                         req->rq_timeout = to->to_maxval;
 587                 req->rq_retries++;
 588         } else {
 589                 req->rq_timeout = to->to_initval;
 590                 req->rq_retries = 0;
 591                 xprt_reset_majortimeo(req);
 592                 /* Reset the RTT counters == "slow start" */
 593                 spin_lock_bh(&xprt->transport_lock);
 594                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 595                 spin_unlock_bh(&xprt->transport_lock);
 596                 status = -ETIMEDOUT;
 597         }
 598
 599         if (req->rq_timeout == 0) {
 600                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 601                 req->rq_timeout = 5 * HZ;
 602         }
 603         return status;
 604 }
 605
 606 static void xprt_autoclose(struct work_struct *work)
 607 {
 608         struct rpc_xprt *xprt =
 609                 container_of(work, struct rpc_xprt, task_cleanup);
 610
 611         xprt->ops->close(xprt);
 612         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 613         xprt_release_write(xprt, NULL);
 614 }
 615
 616 /**
 617  * xprt_disconnect_done - mark a transport as disconnected
 618  * @xprt: transport to flag for disconnect
 619  *
 620  */
 621 void xprt_disconnect_done(struct rpc_xprt *xprt)
 622 {
 623         dprintk("RPC:       disconnected transport %p\n", xprt);
 624         spin_lock_bh(&xprt->transport_lock);
 625         xprt_clear_connected(xprt);
 626         xprt_wake_pending_tasks(xprt, -EAGAIN);
 627         spin_unlock_bh(&xprt->transport_lock);
 628 }
 629 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 630
 631 /**
 632  * xprt_force_disconnect - force a transport to disconnect
 633  * @xprt: transport to disconnect
 634  *
 635  */
 636 void xprt_force_disconnect(struct rpc_xprt *xprt)
 637 {
 638         /* Don't race with the test_bit() in xprt_clear_locked() */
 639         spin_lock_bh(&xprt->transport_lock);
 640         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 641         /* Try to schedule an autoclose RPC call */
 642         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 643                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 644         xprt_wake_pending_tasks(xprt, -EAGAIN);
 645         spin_unlock_bh(&xprt->transport_lock);
 646 }
 647
 648 /**
 649  * xprt_conditional_disconnect - force a transport to disconnect
 650  * @xprt: transport to disconnect
 651  * @cookie: 'connection cookie'
 652  *
 653  * This attempts to break the connection if and only if 'cookie' matches
 654  * the current transport 'connection cookie'. It ensures that we don't
 655  * try to break the connection more than once when we need to retransmit
 656  * a batch of RPC requests.
 657  *
 658  */
 659 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 660 {
 661         /* Don't race with the test_bit() in xprt_clear_locked() */
 662         spin_lock_bh(&xprt->transport_lock);
 663         if (cookie != xprt->connect_cookie)
 664                 goto out;
 665         if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
 666                 goto out;
 667         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 668         /* Try to schedule an autoclose RPC call */
 669         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 670                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 671         xprt_wake_pending_tasks(xprt, -EAGAIN);
 672 out:
 673         spin_unlock_bh(&xprt->transport_lock);
 674 }
 675
 676 static void
 677 xprt_init_autodisconnect(unsigned long data)
 678 {
 679         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 680
 681         spin_lock(&xprt->transport_lock);
 682         if (!list_empty(&xprt->recv))
 683                 goto out_abort;
 684         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 685                 goto out_abort;
 686         spin_unlock(&xprt->transport_lock);
 687         queue_work(rpciod_workqueue, &xprt->task_cleanup);
 688         return;
 689 out_abort:
 690         spin_unlock(&xprt->transport_lock);
 691 }
 692
 693 bool xprt_lock_connect(struct rpc_xprt *xprt,
 694                 struct rpc_task *task,
 695                 void *cookie)
 696 {
 697         bool ret = false;
 698
 699         spin_lock_bh(&xprt->transport_lock);
 700         if (!test_bit(XPRT_LOCKED, &xprt->state))
 701                 goto out;
 702         if (xprt->snd_task != task)
 703                 goto out;
 704         xprt_task_clear_bytes_sent(task);
 705         xprt->snd_task = cookie;
 706         ret = true;
 707 out:
 708         spin_unlock_bh(&xprt->transport_lock);
 709         return ret;
 710 }
 711
 712 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 713 {
 714         spin_lock_bh(&xprt->transport_lock);
 715         if (xprt->snd_task != cookie)
 716                 goto out;
 717         if (!test_bit(XPRT_LOCKED, &xprt->state))
 718                 goto out;
 719         xprt->snd_task =NULL;
 720         xprt->ops->release_xprt(xprt, NULL);
 721 out:
 722         spin_unlock_bh(&xprt->transport_lock);
 723 }
 724
 725 /**
 726  * xprt_connect - schedule a transport connect operation
 727  * @task: RPC task that is requesting the connect
 728  *
 729  */
 730 void xprt_connect(struct rpc_task *task)
 731 {
 732         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 733
 734         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 735                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 736
 737         if (!xprt_bound(xprt)) {
 738                 task->tk_status = -EAGAIN;
 739                 return;
 740         }
 741         if (!xprt_lock_write(xprt, task))
 742                 return;
 743
 744         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 745                 xprt->ops->close(xprt);
 746
 747         if (!xprt_connected(xprt)) {
 748                 task->tk_rqstp->rq_bytes_sent = 0;
 749                 task->tk_timeout = task->tk_rqstp->rq_timeout;
 750                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
 751
 752                 if (test_bit(XPRT_CLOSING, &xprt->state))
 753                         return;
 754                 if (xprt_test_and_set_connecting(xprt))
 755                         return;
 756                 xprt->stat.connect_start = jiffies;
 757                 xprt->ops->connect(xprt, task);
 758         }
 759         xprt_release_write(xprt, task);
 760 }
 761
 762 static void xprt_connect_status(struct rpc_task *task)
 763 {
 764         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 765
 766         if (task->tk_status == 0) {
 767                 xprt->stat.connect_count++;
 768                 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
 769                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
 770                                 task->tk_pid);
 771                 return;
 772         }
 773
 774         switch (task->tk_status) {
 775         case -ECONNREFUSED:
 776         case -ECONNRESET:
 777         case -ECONNABORTED:
 778         case -ENETUNREACH:
 779         case -EHOSTUNREACH:
 780         case -EPIPE:
 781         case -EAGAIN:
 782                 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
 783                 break;
 784         case -ETIMEDOUT:
 785                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
 786                                 "out\n", task->tk_pid);
 787                 break;
 788         default:
 789                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
 790                                 "server %s\n", task->tk_pid, -task->tk_status,
 791                                 xprt->servername);
 792                 task->tk_status = -EIO;
 793         }
 794 }
 795
 796 /**
 797  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 798  * @xprt: transport on which the original request was transmitted
 799  * @xid: RPC XID of incoming reply
 800  *
 801  */
 802 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 803 {
 804         struct rpc_rqst *entry;
 805
 806         list_for_each_entry(entry, &xprt->recv, rq_list)
 807                 if (entry->rq_xid == xid) {
 808                         trace_xprt_lookup_rqst(xprt, xid, 0);
 809                         return entry;
 810                 }
 811
 812         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 813                         ntohl(xid));
 814         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 815         xprt->stat.bad_xids++;
 816         return NULL;
 817 }
 818 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 819
 820 static void xprt_update_rtt(struct rpc_task *task)
 821 {
 822         struct rpc_rqst *req = task->tk_rqstp;
 823         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 824         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
 825         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
 826
 827         if (timer) {
 828                 if (req->rq_ntrans == 1)
 829                         rpc_update_rtt(rtt, timer, m);
 830                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 831         }
 832 }
 833
 834 /**
 835  * xprt_complete_rqst - called when reply processing is complete
 836  * @task: RPC request that recently completed
 837  * @copied: actual number of bytes received from the transport
 838  *
 839  * Caller holds transport lock.
 840  */
 841 void xprt_complete_rqst(struct rpc_task *task, int copied)
 842 {
 843         struct rpc_rqst *req = task->tk_rqstp;
 844         struct rpc_xprt *xprt = req->rq_xprt;
 845
 846         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 847                         task->tk_pid, ntohl(req->rq_xid), copied);
 848         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
 849
 850         xprt->stat.recvs++;
 851         req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
 852         if (xprt->ops->timer != NULL)
 853                 xprt_update_rtt(task);
 854
 855         list_del_init(&req->rq_list);
 856         req->rq_private_buf.len = copied;
 857         /* Ensure all writes are done before we update */
 858         /* req->rq_reply_bytes_recvd */
 859         smp_wmb();
 860         req->rq_reply_bytes_recvd = copied;
 861         rpc_wake_up_queued_task(&xprt->pending, task);
 862 }
 863 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 864
 865 static void xprt_timer(struct rpc_task *task)
 866 {
 867         struct rpc_rqst *req = task->tk_rqstp;
 868         struct rpc_xprt *xprt = req->rq_xprt;
 869
 870         if (task->tk_status != -ETIMEDOUT)
 871                 return;
 872         dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
 873
 874         spin_lock_bh(&xprt->transport_lock);
 875         if (!req->rq_reply_bytes_recvd) {
 876                 if (xprt->ops->timer)
 877                         xprt->ops->timer(xprt, task);
 878         } else
 879                 task->tk_status = 0;
 880         spin_unlock_bh(&xprt->transport_lock);
 881 }
 882
 883 static inline int xprt_has_timer(struct rpc_xprt *xprt)
 884 {
 885         return xprt->idle_timeout != 0;
 886 }
 887
 888 /**
 889  * xprt_prepare_transmit - reserve the transport before sending a request
 890  * @task: RPC task about to send a request
 891  *
 892  */
 893 bool xprt_prepare_transmit(struct rpc_task *task)
 894 {
 895         struct rpc_rqst *req = task->tk_rqstp;
 896         struct rpc_xprt *xprt = req->rq_xprt;
 897         bool ret = false;
 898
 899         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
 900
 901         spin_lock_bh(&xprt->transport_lock);
 902         if (!req->rq_bytes_sent) {
 903                 if (req->rq_reply_bytes_recvd) {
 904                         task->tk_status = req->rq_reply_bytes_recvd;
 905                         goto out_unlock;
 906                 }
 907                 if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
 908                     && xprt_connected(xprt)
 909                     && req->rq_connect_cookie == xprt->connect_cookie) {
 910                         xprt->ops->set_retrans_timeout(task);
 911                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
 912                         goto out_unlock;
 913                 }
 914         }
 915         if (!xprt->ops->reserve_xprt(xprt, task)) {
 916                 task->tk_status = -EAGAIN;
 917                 goto out_unlock;
 918         }
 919         ret = true;
 920 out_unlock:
 921         spin_unlock_bh(&xprt->transport_lock);
 922         return ret;
 923 }
 924
 925 void xprt_end_transmit(struct rpc_task *task)
 926 {
 927         xprt_release_write(task->tk_rqstp->rq_xprt, task);
 928 }
 929
 930 /**
 931  * xprt_transmit - send an RPC request on a transport
 932  * @task: controlling RPC task
 933  *
 934  * We have to copy the iovec because sendmsg fiddles with its contents.
 935  */
 936 void xprt_transmit(struct rpc_task *task)
 937 {
 938         struct rpc_rqst *req = task->tk_rqstp;
 939         struct rpc_xprt *xprt = req->rq_xprt;
 940         int status, numreqs;
 941
 942         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 943
 944         if (!req->rq_reply_bytes_recvd) {
 945                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
 946                         /*
 947                          * Add to the list only if we're expecting a reply
 948                          */
 949                         spin_lock_bh(&xprt->transport_lock);
 950                         /* Update the softirq receive buffer */
 951                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 952                                         sizeof(req->rq_private_buf));
 953                         /* Add request to the receive list */
 954                         list_add_tail(&req->rq_list, &xprt->recv);
 955                         spin_unlock_bh(&xprt->transport_lock);
 956                         xprt_reset_majortimeo(req);
 957                         /* Turn off autodisconnect */
 958                         del_singleshot_timer_sync(&xprt->timer);
 959                 }
 960         } else if (!req->rq_bytes_sent)
 961                 return;
 962
 963         req->rq_xtime = ktime_get();
 964         status = xprt->ops->send_request(task);
 965         trace_xprt_transmit(xprt, req->rq_xid, status);
 966         if (status != 0) {
 967                 task->tk_status = status;
 968                 return;
 969         }
 970
 971         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
 972         task->tk_flags |= RPC_TASK_SENT;
 973         spin_lock_bh(&xprt->transport_lock);
 974
 975         xprt->ops->set_retrans_timeout(task);
 976
 977         numreqs = atomic_read(&xprt->num_reqs);
 978         if (numreqs > xprt->stat.max_slots)
 979                 xprt->stat.max_slots = numreqs;
 980         xprt->stat.sends++;
 981         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
 982         xprt->stat.bklog_u += xprt->backlog.qlen;
 983         xprt->stat.sending_u += xprt->sending.qlen;
 984         xprt->stat.pending_u += xprt->pending.qlen;
 985
 986         /* Don't race with disconnect */
 987         if (!xprt_connected(xprt))
 988                 task->tk_status = -ENOTCONN;
 989         else {
 990                 /*
 991                  * Sleep on the pending queue since
 992                  * we're expecting a reply.
 993                  */
 994                 if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task))
 995                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
 996                 req->rq_connect_cookie = xprt->connect_cookie;
 997         }
 998         spin_unlock_bh(&xprt->transport_lock);
 999 }
1000
1001 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1002 {
1003         set_bit(XPRT_CONGESTED, &xprt->state);
1004         rpc_sleep_on(&xprt->backlog, task, NULL);
1005 }
1006
1007 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1008 {
1009         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1010                 clear_bit(XPRT_CONGESTED, &xprt->state);
1011 }
1012
1013 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1014 {
1015         bool ret = false;
1016
1017         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1018                 goto out;
1019         spin_lock(&xprt->reserve_lock);
1020         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1021                 rpc_sleep_on(&xprt->backlog, task, NULL);
1022                 ret = true;
1023         }
1024         spin_unlock(&xprt->reserve_lock);
1025 out:
1026         return ret;
1027 }
1028
1029 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
1030 {
1031         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1032
1033         if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
1034                 goto out;
1035         req = kzalloc(sizeof(struct rpc_rqst), gfp_flags);
1036         if (req != NULL)
1037                 goto out;
1038         atomic_dec(&xprt->num_reqs);
1039         req = ERR_PTR(-ENOMEM);
1040 out:
1041         return req;
1042 }
1043
1044 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1045 {
1046         if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
1047                 kfree(req);
1048                 return true;
1049         }
1050         return false;
1051 }
1052
1053 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1054 {
1055         struct rpc_rqst *req;
1056
1057         spin_lock(&xprt->reserve_lock);
1058         if (!list_empty(&xprt->free)) {
1059                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1060                 list_del(&req->rq_list);
1061                 goto out_init_req;
1062         }
1063         req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT|__GFP_NOWARN);
1064         if (!IS_ERR(req))
1065                 goto out_init_req;
1066         switch (PTR_ERR(req)) {
1067         case -ENOMEM:
1068                 dprintk("RPC:       dynamic allocation of request slot "
1069                                 "failed! Retrying\n");
1070                 task->tk_status = -ENOMEM;
1071                 break;
1072         case -EAGAIN:
1073                 xprt_add_backlog(xprt, task);
1074                 dprintk("RPC:       waiting for request slot\n");
1075         default:
1076                 task->tk_status = -EAGAIN;
1077         }
1078         spin_unlock(&xprt->reserve_lock);
1079         return;
1080 out_init_req:
1081         task->tk_status = 0;
1082         task->tk_rqstp = req;
1083         xprt_request_init(task, xprt);
1084         spin_unlock(&xprt->reserve_lock);
1085 }
1086 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1087
1088 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1089 {
1090         /* Note: grabbing the xprt_lock_write() ensures that we throttle
1091          * new slot allocation if the transport is congested (i.e. when
1092          * reconnecting a stream transport or when out of socket write
1093          * buffer space).
1094          */
1095         if (xprt_lock_write(xprt, task)) {
1096                 xprt_alloc_slot(xprt, task);
1097                 xprt_release_write(xprt, task);
1098         }
1099 }
1100 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1101
1102 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1103 {
1104         spin_lock(&xprt->reserve_lock);
1105         if (!xprt_dynamic_free_slot(xprt, req)) {
1106                 memset(req, 0, sizeof(*req));   /* mark unused */
1107                 list_add(&req->rq_list, &xprt->free);
1108         }
1109         xprt_wake_up_backlog(xprt);
1110         spin_unlock(&xprt->reserve_lock);
1111 }
1112
1113 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1114 {
1115         struct rpc_rqst *req;
1116         while (!list_empty(&xprt->free)) {
1117                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1118                 list_del(&req->rq_list);
1119                 kfree(req);
1120         }
1121 }
1122
1123 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1124                 unsigned int num_prealloc,
1125                 unsigned int max_alloc)
1126 {
1127         struct rpc_xprt *xprt;
1128         struct rpc_rqst *req;
1129         int i;
1130
1131         xprt = kzalloc(size, GFP_KERNEL);
1132         if (xprt == NULL)
1133                 goto out;
1134
1135         xprt_init(xprt, net);
1136
1137         for (i = 0; i < num_prealloc; i++) {
1138                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1139                 if (!req)
1140                         goto out_free;
1141                 list_add(&req->rq_list, &xprt->free);
1142         }
1143         if (max_alloc > num_prealloc)
1144                 xprt->max_reqs = max_alloc;
1145         else
1146                 xprt->max_reqs = num_prealloc;
1147         xprt->min_reqs = num_prealloc;
1148         atomic_set(&xprt->num_reqs, num_prealloc);
1149
1150         return xprt;
1151
1152 out_free:
1153         xprt_free(xprt);
1154 out:
1155         return NULL;
1156 }
1157 EXPORT_SYMBOL_GPL(xprt_alloc);
1158
1159 void xprt_free(struct rpc_xprt *xprt)
1160 {
1161         put_net(xprt->xprt_net);
1162         xprt_free_all_slots(xprt);
1163         kfree(xprt);
1164 }
1165 EXPORT_SYMBOL_GPL(xprt_free);
1166
1167 /**
1168  * xprt_reserve - allocate an RPC request slot
1169  * @task: RPC task requesting a slot allocation
1170  *
1171  * If the transport is marked as being congested, or if no more
1172  * slots are available, place the task on the transport's
1173  * backlog queue.
1174  */
1175 void xprt_reserve(struct rpc_task *task)
1176 {
1177         struct rpc_xprt *xprt;
1178
1179         task->tk_status = 0;
1180         if (task->tk_rqstp != NULL)
1181                 return;
1182
1183         task->tk_timeout = 0;
1184         task->tk_status = -EAGAIN;
1185         rcu_read_lock();
1186         xprt = rcu_dereference(task->tk_client->cl_xprt);
1187         if (!xprt_throttle_congested(xprt, task))
1188                 xprt->ops->alloc_slot(xprt, task);
1189         rcu_read_unlock();
1190 }
1191
1192 /**
1193  * xprt_retry_reserve - allocate an RPC request slot
1194  * @task: RPC task requesting a slot allocation
1195  *
1196  * If no more slots are available, place the task on the transport's
1197  * backlog queue.
1198  * Note that the only difference with xprt_reserve is that we now
1199  * ignore the value of the XPRT_CONGESTED flag.
1200  */
1201 void xprt_retry_reserve(struct rpc_task *task)
1202 {
1203         struct rpc_xprt *xprt;
1204
1205         task->tk_status = 0;
1206         if (task->tk_rqstp != NULL)
1207                 return;
1208
1209         task->tk_timeout = 0;
1210         task->tk_status = -EAGAIN;
1211         rcu_read_lock();
1212         xprt = rcu_dereference(task->tk_client->cl_xprt);
1213         xprt->ops->alloc_slot(xprt, task);
1214         rcu_read_unlock();
1215 }
1216
1217 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1218 {
1219         return (__force __be32)xprt->xid++;
1220 }
1221
1222 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1223 {
1224         xprt->xid = prandom_u32();
1225 }
1226
1227 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1228 {
1229         struct rpc_rqst *req = task->tk_rqstp;
1230
1231         INIT_LIST_HEAD(&req->rq_list);
1232         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1233         req->rq_task    = task;
1234         req->rq_xprt    = xprt;
1235         req->rq_buffer  = NULL;
1236         req->rq_xid     = xprt_alloc_xid(xprt);
1237         req->rq_connect_cookie = xprt->connect_cookie - 1;
1238         req->rq_bytes_sent = 0;
1239         req->rq_snd_buf.len = 0;
1240         req->rq_snd_buf.buflen = 0;
1241         req->rq_rcv_buf.len = 0;
1242         req->rq_rcv_buf.buflen = 0;
1243         req->rq_release_snd_buf = NULL;
1244         xprt_reset_majortimeo(req);
1245         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1246                         req, ntohl(req->rq_xid));
1247 }
1248
1249 /**
1250  * xprt_release - release an RPC request slot
1251  * @task: task which is finished with the slot
1252  *
1253  */
1254 void xprt_release(struct rpc_task *task)
1255 {
1256         struct rpc_xprt *xprt;
1257         struct rpc_rqst *req = task->tk_rqstp;
1258
1259         if (req == NULL) {
1260                 if (task->tk_client) {
1261                         rcu_read_lock();
1262                         xprt = rcu_dereference(task->tk_client->cl_xprt);
1263                         if (xprt->snd_task == task)
1264                                 xprt_release_write(xprt, task);
1265                         rcu_read_unlock();
1266                 }
1267                 return;
1268         }
1269
1270         xprt = req->rq_xprt;
1271         if (task->tk_ops->rpc_count_stats != NULL)
1272                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1273         else if (task->tk_client)
1274                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1275         spin_lock_bh(&xprt->transport_lock);
1276         xprt->ops->release_xprt(xprt, task);
1277         if (xprt->ops->release_request)
1278                 xprt->ops->release_request(task);
1279         if (!list_empty(&req->rq_list))
1280                 list_del(&req->rq_list);
1281         xprt->last_used = jiffies;
1282         if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1283                 mod_timer(&xprt->timer,
1284                                 xprt->last_used + xprt->idle_timeout);
1285         spin_unlock_bh(&xprt->transport_lock);
1286         if (req->rq_buffer)
1287                 xprt->ops->buf_free(req->rq_buffer);
1288         if (req->rq_cred != NULL)
1289                 put_rpccred(req->rq_cred);
1290         task->tk_rqstp = NULL;
1291         if (req->rq_release_snd_buf)
1292                 req->rq_release_snd_buf(req);
1293
1294         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1295         if (likely(!bc_prealloc(req)))
1296                 xprt_free_slot(xprt, req);
1297         else
1298                 xprt_free_bc_request(req);
1299 }
1300
1301 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1302 {
1303         atomic_set(&xprt->count, 1);
1304
1305         spin_lock_init(&xprt->transport_lock);
1306         spin_lock_init(&xprt->reserve_lock);
1307
1308         INIT_LIST_HEAD(&xprt->free);
1309         INIT_LIST_HEAD(&xprt->recv);
1310 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1311         spin_lock_init(&xprt->bc_pa_lock);
1312         INIT_LIST_HEAD(&xprt->bc_pa_list);
1313 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1314
1315         xprt->last_used = jiffies;
1316         xprt->cwnd = RPC_INITCWND;
1317         xprt->bind_index = 0;
1318
1319         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1320         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1321         rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1322         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1323
1324         xprt_init_xid(xprt);
1325
1326         xprt->xprt_net = get_net(net);
1327 }
1328
1329 /**
1330  * xprt_create_transport - create an RPC transport
1331  * @args: rpc transport creation arguments
1332  *
1333  */
1334 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1335 {
1336         struct rpc_xprt *xprt;
1337         struct xprt_class *t;
1338
1339         spin_lock(&xprt_list_lock);
1340         list_for_each_entry(t, &xprt_list, list) {
1341                 if (t->ident == args->ident) {
1342                         spin_unlock(&xprt_list_lock);
1343                         goto found;
1344                 }
1345         }
1346         spin_unlock(&xprt_list_lock);
1347         dprintk("RPC: transport (%d) not supported\n", args->ident);
1348         return ERR_PTR(-EIO);
1349
1350 found:
1351         xprt = t->setup(args);
1352         if (IS_ERR(xprt)) {
1353                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1354                                 -PTR_ERR(xprt));
1355                 goto out;
1356         }
1357         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1358                 xprt->idle_timeout = 0;
1359         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1360         if (xprt_has_timer(xprt))
1361                 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1362                             (unsigned long)xprt);
1363         else
1364                 init_timer(&xprt->timer);
1365
1366         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1367                 xprt_destroy(xprt);
1368                 return ERR_PTR(-EINVAL);
1369         }
1370         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1371         if (xprt->servername == NULL) {
1372                 xprt_destroy(xprt);
1373                 return ERR_PTR(-ENOMEM);
1374         }
1375
1376         rpc_xprt_debugfs_register(xprt);
1377
1378         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1379                         xprt->max_reqs);
1380 out:
1381         return xprt;
1382 }
1383
1384 /**
1385  * xprt_destroy - destroy an RPC transport, killing off all requests.
1386  * @xprt: transport to destroy
1387  *
1388  */
1389 static void xprt_destroy(struct rpc_xprt *xprt)
1390 {
1391         dprintk("RPC:       destroying transport %p\n", xprt);
1392         del_timer_sync(&xprt->timer);
1393
1394         rpc_xprt_debugfs_unregister(xprt);
1395         rpc_destroy_wait_queue(&xprt->binding);
1396         rpc_destroy_wait_queue(&xprt->pending);
1397         rpc_destroy_wait_queue(&xprt->sending);
1398         rpc_destroy_wait_queue(&xprt->backlog);
1399         cancel_work_sync(&xprt->task_cleanup);
1400         kfree(xprt->servername);
1401         /*
1402          * Tear down transport state and free the rpc_xprt
1403          */
1404         xprt->ops->destroy(xprt);
1405 }
1406
1407 /**
1408  * xprt_put - release a reference to an RPC transport.
1409  * @xprt: pointer to the transport
1410  *
1411  */
1412 void xprt_put(struct rpc_xprt *xprt)
1413 {
1414         if (atomic_dec_and_test(&xprt->count))
1415                 xprt_destroy(xprt);
1416 }