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 <okir@monad.swb.de>
  36  *
  37  *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
  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 /**
 330  * xprt_release_xprt - allow other requests to use a transport
 331  * @xprt: transport with other tasks potentially waiting
 332  * @task: task that is releasing access to the transport
 333  *
 334  * Note that "task" can be NULL.  No congestion control is provided.
 335  */
 336 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 337 {
 338         if (xprt->snd_task == task) {
 339                 if (task != NULL) {
 340                         struct rpc_rqst *req = task->tk_rqstp;
 341                         if (req != NULL)
 342                                 req->rq_bytes_sent = 0;
 343                 }
 344                 xprt_clear_locked(xprt);
 345                 __xprt_lock_write_next(xprt);
 346         }
 347 }
 348 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 349
 350 /**
 351  * xprt_release_xprt_cong - allow other requests to use a transport
 352  * @xprt: transport with other tasks potentially waiting
 353  * @task: task that is releasing access to the transport
 354  *
 355  * Note that "task" can be NULL.  Another task is awoken to use the
 356  * transport if the transport's congestion window allows it.
 357  */
 358 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 359 {
 360         if (xprt->snd_task == task) {
 361                 if (task != NULL) {
 362                         struct rpc_rqst *req = task->tk_rqstp;
 363                         if (req != NULL)
 364                                 req->rq_bytes_sent = 0;
 365                 }
 366                 xprt_clear_locked(xprt);
 367                 __xprt_lock_write_next_cong(xprt);
 368         }
 369 }
 370 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 371
 372 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 373 {
 374         spin_lock_bh(&xprt->transport_lock);
 375         xprt->ops->release_xprt(xprt, task);
 376         spin_unlock_bh(&xprt->transport_lock);
 377 }
 378
 379 /*
 380  * Van Jacobson congestion avoidance. Check if the congestion window
 381  * overflowed. Put the task to sleep if this is the case.
 382  */
 383 static int
 384 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 385 {
 386         struct rpc_rqst *req = task->tk_rqstp;
 387
 388         if (req->rq_cong)
 389                 return 1;
 390         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
 391                         task->tk_pid, xprt->cong, xprt->cwnd);
 392         if (RPCXPRT_CONGESTED(xprt))
 393                 return 0;
 394         req->rq_cong = 1;
 395         xprt->cong += RPC_CWNDSCALE;
 396         return 1;
 397 }
 398
 399 /*
 400  * Adjust the congestion window, and wake up the next task
 401  * that has been sleeping due to congestion
 402  */
 403 static void
 404 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 405 {
 406         if (!req->rq_cong)
 407                 return;
 408         req->rq_cong = 0;
 409         xprt->cong -= RPC_CWNDSCALE;
 410         __xprt_lock_write_next_cong(xprt);
 411 }
 412
 413 /**
 414  * xprt_release_rqst_cong - housekeeping when request is complete
 415  * @task: RPC request that recently completed
 416  *
 417  * Useful for transports that require congestion control.
 418  */
 419 void xprt_release_rqst_cong(struct rpc_task *task)
 420 {
 421         struct rpc_rqst *req = task->tk_rqstp;
 422
 423         __xprt_put_cong(req->rq_xprt, req);
 424 }
 425 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 426
 427 /**
 428  * xprt_adjust_cwnd - adjust transport congestion window
 429  * @xprt: pointer to xprt
 430  * @task: recently completed RPC request used to adjust window
 431  * @result: result code of completed RPC request
 432  *
 433  * The transport code maintains an estimate on the maximum number of out-
 434  * standing RPC requests, using a smoothed version of the congestion
 435  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 436  * congestion algorithm: If a retransmit occurs, the congestion window is
 437  * halved; otherwise, it is incremented by 1/cwnd when
 438  *
 439  *      -       a reply is received and
 440  *      -       a full number of requests are outstanding and
 441  *      -       the congestion window hasn't been updated recently.
 442  */
 443 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 444 {
 445         struct rpc_rqst *req = task->tk_rqstp;
 446         unsigned long cwnd = xprt->cwnd;
 447
 448         if (result >= 0 && cwnd <= xprt->cong) {
 449                 /* The (cwnd >> 1) term makes sure
 450                  * the result gets rounded properly. */
 451                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 452                 if (cwnd > RPC_MAXCWND(xprt))
 453                         cwnd = RPC_MAXCWND(xprt);
 454                 __xprt_lock_write_next_cong(xprt);
 455         } else if (result == -ETIMEDOUT) {
 456                 cwnd >>= 1;
 457                 if (cwnd < RPC_CWNDSCALE)
 458                         cwnd = RPC_CWNDSCALE;
 459         }
 460         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 461                         xprt->cong, xprt->cwnd, cwnd);
 462         xprt->cwnd = cwnd;
 463         __xprt_put_cong(xprt, req);
 464 }
 465 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 466
 467 /**
 468  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 469  * @xprt: transport with waiting tasks
 470  * @status: result code to plant in each task before waking it
 471  *
 472  */
 473 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 474 {
 475         if (status < 0)
 476                 rpc_wake_up_status(&xprt->pending, status);
 477         else
 478                 rpc_wake_up(&xprt->pending);
 479 }
 480 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 481
 482 /**
 483  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 484  * @task: task to be put to sleep
 485  * @action: function pointer to be executed after wait
 486  *
 487  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 488  * we don't in general want to force a socket disconnection due to
 489  * an incomplete RPC call transmission.
 490  */
 491 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
 492 {
 493         struct rpc_rqst *req = task->tk_rqstp;
 494         struct rpc_xprt *xprt = req->rq_xprt;
 495
 496         task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
 497         rpc_sleep_on(&xprt->pending, task, action);
 498 }
 499 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 500
 501 /**
 502  * xprt_write_space - wake the task waiting for transport output buffer space
 503  * @xprt: transport with waiting tasks
 504  *
 505  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 506  */
 507 void xprt_write_space(struct rpc_xprt *xprt)
 508 {
 509         spin_lock_bh(&xprt->transport_lock);
 510         if (xprt->snd_task) {
 511                 dprintk("RPC:       write space: waking waiting task on "
 512                                 "xprt %p\n", xprt);
 513                 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
 514         }
 515         spin_unlock_bh(&xprt->transport_lock);
 516 }
 517 EXPORT_SYMBOL_GPL(xprt_write_space);
 518
 519 /**
 520  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
 521  * @task: task whose timeout is to be set
 522  *
 523  * Set a request's retransmit timeout based on the transport's
 524  * default timeout parameters.  Used by transports that don't adjust
 525  * the retransmit timeout based on round-trip time estimation.
 526  */
 527 void xprt_set_retrans_timeout_def(struct rpc_task *task)
 528 {
 529         task->tk_timeout = task->tk_rqstp->rq_timeout;
 530 }
 531 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
 532
 533 /**
 534  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
 535  * @task: task whose timeout is to be set
 536  *
 537  * Set a request's retransmit timeout using the RTT estimator.
 538  */
 539 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
 540 {
 541         int timer = task->tk_msg.rpc_proc->p_timer;
 542         struct rpc_clnt *clnt = task->tk_client;
 543         struct rpc_rtt *rtt = clnt->cl_rtt;
 544         struct rpc_rqst *req = task->tk_rqstp;
 545         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
 546
 547         task->tk_timeout = rpc_calc_rto(rtt, timer);
 548         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
 549         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
 550                 task->tk_timeout = max_timeout;
 551 }
 552 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
 553
 554 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 555 {
 556         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 557
 558         req->rq_majortimeo = req->rq_timeout;
 559         if (to->to_exponential)
 560                 req->rq_majortimeo <<= to->to_retries;
 561         else
 562                 req->rq_majortimeo += to->to_increment * to->to_retries;
 563         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
 564                 req->rq_majortimeo = to->to_maxval;
 565         req->rq_majortimeo += jiffies;
 566 }
 567
 568 /**
 569  * xprt_adjust_timeout - adjust timeout values for next retransmit
 570  * @req: RPC request containing parameters to use for the adjustment
 571  *
 572  */
 573 int xprt_adjust_timeout(struct rpc_rqst *req)
 574 {
 575         struct rpc_xprt *xprt = req->rq_xprt;
 576         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 577         int status = 0;
 578
 579         if (time_before(jiffies, req->rq_majortimeo)) {
 580                 if (to->to_exponential)
 581                         req->rq_timeout <<= 1;
 582                 else
 583                         req->rq_timeout += to->to_increment;
 584                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 585                         req->rq_timeout = to->to_maxval;
 586                 req->rq_retries++;
 587         } else {
 588                 req->rq_timeout = to->to_initval;
 589                 req->rq_retries = 0;
 590                 xprt_reset_majortimeo(req);
 591                 /* Reset the RTT counters == "slow start" */
 592                 spin_lock_bh(&xprt->transport_lock);
 593                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 594                 spin_unlock_bh(&xprt->transport_lock);
 595                 status = -ETIMEDOUT;
 596         }
 597
 598         if (req->rq_timeout == 0) {
 599                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 600                 req->rq_timeout = 5 * HZ;
 601         }
 602         return status;
 603 }
 604
 605 static void xprt_autoclose(struct work_struct *work)
 606 {
 607         struct rpc_xprt *xprt =
 608                 container_of(work, struct rpc_xprt, task_cleanup);
 609
 610         xprt->ops->close(xprt);
 611         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 612         xprt_release_write(xprt, NULL);
 613 }
 614
 615 /**
 616  * xprt_disconnect_done - mark a transport as disconnected
 617  * @xprt: transport to flag for disconnect
 618  *
 619  */
 620 void xprt_disconnect_done(struct rpc_xprt *xprt)
 621 {
 622         dprintk("RPC:       disconnected transport %p\n", xprt);
 623         spin_lock_bh(&xprt->transport_lock);
 624         xprt_clear_connected(xprt);
 625         xprt_wake_pending_tasks(xprt, -EAGAIN);
 626         spin_unlock_bh(&xprt->transport_lock);
 627 }
 628 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 629
 630 /**
 631  * xprt_force_disconnect - force a transport to disconnect
 632  * @xprt: transport to disconnect
 633  *
 634  */
 635 void xprt_force_disconnect(struct rpc_xprt *xprt)
 636 {
 637         /* Don't race with the test_bit() in xprt_clear_locked() */
 638         spin_lock_bh(&xprt->transport_lock);
 639         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 640         /* Try to schedule an autoclose RPC call */
 641         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 642                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 643         xprt_wake_pending_tasks(xprt, -EAGAIN);
 644         spin_unlock_bh(&xprt->transport_lock);
 645 }
 646
 647 /**
 648  * xprt_conditional_disconnect - force a transport to disconnect
 649  * @xprt: transport to disconnect
 650  * @cookie: 'connection cookie'
 651  *
 652  * This attempts to break the connection if and only if 'cookie' matches
 653  * the current transport 'connection cookie'. It ensures that we don't
 654  * try to break the connection more than once when we need to retransmit
 655  * a batch of RPC requests.
 656  *
 657  */
 658 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 659 {
 660         /* Don't race with the test_bit() in xprt_clear_locked() */
 661         spin_lock_bh(&xprt->transport_lock);
 662         if (cookie != xprt->connect_cookie)
 663                 goto out;
 664         if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
 665                 goto out;
 666         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 667         /* Try to schedule an autoclose RPC call */
 668         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 669                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
 670         xprt_wake_pending_tasks(xprt, -EAGAIN);
 671 out:
 672         spin_unlock_bh(&xprt->transport_lock);
 673 }
 674
 675 static void
 676 xprt_init_autodisconnect(unsigned long data)
 677 {
 678         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 679
 680         spin_lock(&xprt->transport_lock);
 681         if (!list_empty(&xprt->recv))
 682                 goto out_abort;
 683         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 684                 goto out_abort;
 685         spin_unlock(&xprt->transport_lock);
 686         set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
 687         queue_work(rpciod_workqueue, &xprt->task_cleanup);
 688         return;
 689 out_abort:
 690         spin_unlock(&xprt->transport_lock);
 691 }
 692
 693 /**
 694  * xprt_connect - schedule a transport connect operation
 695  * @task: RPC task that is requesting the connect
 696  *
 697  */
 698 void xprt_connect(struct rpc_task *task)
 699 {
 700         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 701
 702         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 703                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 704
 705         if (!xprt_bound(xprt)) {
 706                 task->tk_status = -EAGAIN;
 707                 return;
 708         }
 709         if (!xprt_lock_write(xprt, task))
 710                 return;
 711
 712         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 713                 xprt->ops->close(xprt);
 714
 715         if (xprt_connected(xprt))
 716                 xprt_release_write(xprt, task);
 717         else {
 718                 task->tk_rqstp->rq_bytes_sent = 0;
 719                 task->tk_timeout = task->tk_rqstp->rq_timeout;
 720                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
 721
 722                 if (test_bit(XPRT_CLOSING, &xprt->state))
 723                         return;
 724                 if (xprt_test_and_set_connecting(xprt))
 725                         return;
 726                 xprt->stat.connect_start = jiffies;
 727                 xprt->ops->connect(xprt, task);
 728         }
 729 }
 730
 731 static void xprt_connect_status(struct rpc_task *task)
 732 {
 733         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 734
 735         if (task->tk_status == 0) {
 736                 xprt->stat.connect_count++;
 737                 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
 738                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
 739                                 task->tk_pid);
 740                 return;
 741         }
 742
 743         switch (task->tk_status) {
 744         case -ECONNREFUSED:
 745         case -ECONNRESET:
 746         case -ECONNABORTED:
 747         case -ENETUNREACH:
 748         case -EHOSTUNREACH:
 749         case -EPIPE:
 750         case -EAGAIN:
 751                 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
 752                 break;
 753         case -ETIMEDOUT:
 754                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
 755                                 "out\n", task->tk_pid);
 756                 break;
 757         default:
 758                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
 759                                 "server %s\n", task->tk_pid, -task->tk_status,
 760                                 xprt->servername);
 761                 xprt_release_write(xprt, task);
 762                 task->tk_status = -EIO;
 763         }
 764 }
 765
 766 /**
 767  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 768  * @xprt: transport on which the original request was transmitted
 769  * @xid: RPC XID of incoming reply
 770  *
 771  */
 772 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 773 {
 774         struct rpc_rqst *entry;
 775
 776         list_for_each_entry(entry, &xprt->recv, rq_list)
 777                 if (entry->rq_xid == xid) {
 778                         trace_xprt_lookup_rqst(xprt, xid, 0);
 779                         return entry;
 780                 }
 781
 782         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 783                         ntohl(xid));
 784         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 785         xprt->stat.bad_xids++;
 786         return NULL;
 787 }
 788 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 789
 790 static void xprt_update_rtt(struct rpc_task *task)
 791 {
 792         struct rpc_rqst *req = task->tk_rqstp;
 793         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
 794         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
 795         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
 796
 797         if (timer) {
 798                 if (req->rq_ntrans == 1)
 799                         rpc_update_rtt(rtt, timer, m);
 800                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
 801         }
 802 }
 803
 804 /**
 805  * xprt_complete_rqst - called when reply processing is complete
 806  * @task: RPC request that recently completed
 807  * @copied: actual number of bytes received from the transport
 808  *
 809  * Caller holds transport lock.
 810  */
 811 void xprt_complete_rqst(struct rpc_task *task, int copied)
 812 {
 813         struct rpc_rqst *req = task->tk_rqstp;
 814         struct rpc_xprt *xprt = req->rq_xprt;
 815
 816         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
 817                         task->tk_pid, ntohl(req->rq_xid), copied);
 818         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
 819
 820         xprt->stat.recvs++;
 821         req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
 822         if (xprt->ops->timer != NULL)
 823                 xprt_update_rtt(task);
 824
 825         list_del_init(&req->rq_list);
 826         req->rq_private_buf.len = copied;
 827         /* Ensure all writes are done before we update */
 828         /* req->rq_reply_bytes_recvd */
 829         smp_wmb();
 830         req->rq_reply_bytes_recvd = copied;
 831         rpc_wake_up_queued_task(&xprt->pending, task);
 832 }
 833 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
 834
 835 static void xprt_timer(struct rpc_task *task)
 836 {
 837         struct rpc_rqst *req = task->tk_rqstp;
 838         struct rpc_xprt *xprt = req->rq_xprt;
 839
 840         if (task->tk_status != -ETIMEDOUT)
 841                 return;
 842         dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
 843
 844         spin_lock_bh(&xprt->transport_lock);
 845         if (!req->rq_reply_bytes_recvd) {
 846                 if (xprt->ops->timer)
 847                         xprt->ops->timer(xprt, task);
 848         } else
 849                 task->tk_status = 0;
 850         spin_unlock_bh(&xprt->transport_lock);
 851 }
 852
 853 static inline int xprt_has_timer(struct rpc_xprt *xprt)
 854 {
 855         return xprt->idle_timeout != 0;
 856 }
 857
 858 /**
 859  * xprt_prepare_transmit - reserve the transport before sending a request
 860  * @task: RPC task about to send a request
 861  *
 862  */
 863 bool xprt_prepare_transmit(struct rpc_task *task)
 864 {
 865         struct rpc_rqst *req = task->tk_rqstp;
 866         struct rpc_xprt *xprt = req->rq_xprt;
 867         bool ret = false;
 868
 869         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
 870
 871         spin_lock_bh(&xprt->transport_lock);
 872         if (!req->rq_bytes_sent) {
 873                 if (req->rq_reply_bytes_recvd) {
 874                         task->tk_status = req->rq_reply_bytes_recvd;
 875                         goto out_unlock;
 876                 }
 877                 if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
 878                     && xprt_connected(xprt)
 879                     && req->rq_connect_cookie == xprt->connect_cookie) {
 880                         xprt->ops->set_retrans_timeout(task);
 881                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
 882                         goto out_unlock;
 883                 }
 884         }
 885         if (!xprt->ops->reserve_xprt(xprt, task)) {
 886                 task->tk_status = -EAGAIN;
 887                 goto out_unlock;
 888         }
 889         ret = true;
 890 out_unlock:
 891         spin_unlock_bh(&xprt->transport_lock);
 892         return ret;
 893 }
 894
 895 void xprt_end_transmit(struct rpc_task *task)
 896 {
 897         xprt_release_write(task->tk_rqstp->rq_xprt, task);
 898 }
 899
 900 /**
 901  * xprt_transmit - send an RPC request on a transport
 902  * @task: controlling RPC task
 903  *
 904  * We have to copy the iovec because sendmsg fiddles with its contents.
 905  */
 906 void xprt_transmit(struct rpc_task *task)
 907 {
 908         struct rpc_rqst *req = task->tk_rqstp;
 909         struct rpc_xprt *xprt = req->rq_xprt;
 910         int status, numreqs;
 911
 912         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 913
 914         if (!req->rq_reply_bytes_recvd) {
 915                 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
 916                         /*
 917                          * Add to the list only if we're expecting a reply
 918                          */
 919                         spin_lock_bh(&xprt->transport_lock);
 920                         /* Update the softirq receive buffer */
 921                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
 922                                         sizeof(req->rq_private_buf));
 923                         /* Add request to the receive list */
 924                         list_add_tail(&req->rq_list, &xprt->recv);
 925                         spin_unlock_bh(&xprt->transport_lock);
 926                         xprt_reset_majortimeo(req);
 927                         /* Turn off autodisconnect */
 928                         del_singleshot_timer_sync(&xprt->timer);
 929                 }
 930         } else if (!req->rq_bytes_sent)
 931                 return;
 932
 933         req->rq_xtime = ktime_get();
 934         status = xprt->ops->send_request(task);
 935         trace_xprt_transmit(xprt, req->rq_xid, status);
 936         if (status != 0) {
 937                 task->tk_status = status;
 938                 return;
 939         }
 940
 941         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
 942         task->tk_flags |= RPC_TASK_SENT;
 943         spin_lock_bh(&xprt->transport_lock);
 944
 945         xprt->ops->set_retrans_timeout(task);
 946
 947         numreqs = atomic_read(&xprt->num_reqs);
 948         if (numreqs > xprt->stat.max_slots)
 949                 xprt->stat.max_slots = numreqs;
 950         xprt->stat.sends++;
 951         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
 952         xprt->stat.bklog_u += xprt->backlog.qlen;
 953         xprt->stat.sending_u += xprt->sending.qlen;
 954         xprt->stat.pending_u += xprt->pending.qlen;
 955
 956         /* Don't race with disconnect */
 957         if (!xprt_connected(xprt))
 958                 task->tk_status = -ENOTCONN;
 959         else {
 960                 /*
 961                  * Sleep on the pending queue since
 962                  * we're expecting a reply.
 963                  */
 964                 if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task))
 965                         rpc_sleep_on(&xprt->pending, task, xprt_timer);
 966                 req->rq_connect_cookie = xprt->connect_cookie;
 967         }
 968         spin_unlock_bh(&xprt->transport_lock);
 969 }
 970
 971 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
 972 {
 973         set_bit(XPRT_CONGESTED, &xprt->state);
 974         rpc_sleep_on(&xprt->backlog, task, NULL);
 975 }
 976
 977 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
 978 {
 979         if (rpc_wake_up_next(&xprt->backlog) == NULL)
 980                 clear_bit(XPRT_CONGESTED, &xprt->state);
 981 }
 982
 983 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
 984 {
 985         bool ret = false;
 986
 987         if (!test_bit(XPRT_CONGESTED, &xprt->state))
 988                 goto out;
 989         spin_lock(&xprt->reserve_lock);
 990         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
 991                 rpc_sleep_on(&xprt->backlog, task, NULL);
 992                 ret = true;
 993         }
 994         spin_unlock(&xprt->reserve_lock);
 995 out:
 996         return ret;
 997 }
 998
 999 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
1000 {
1001         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1002
1003         if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
1004                 goto out;
1005         req = kzalloc(sizeof(struct rpc_rqst), gfp_flags);
1006         if (req != NULL)
1007                 goto out;
1008         atomic_dec(&xprt->num_reqs);
1009         req = ERR_PTR(-ENOMEM);
1010 out:
1011         return req;
1012 }
1013
1014 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1015 {
1016         if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
1017                 kfree(req);
1018                 return true;
1019         }
1020         return false;
1021 }
1022
1023 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1024 {
1025         struct rpc_rqst *req;
1026
1027         spin_lock(&xprt->reserve_lock);
1028         if (!list_empty(&xprt->free)) {
1029                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1030                 list_del(&req->rq_list);
1031                 goto out_init_req;
1032         }
1033         req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT|__GFP_NOWARN);
1034         if (!IS_ERR(req))
1035                 goto out_init_req;
1036         switch (PTR_ERR(req)) {
1037         case -ENOMEM:
1038                 dprintk("RPC:       dynamic allocation of request slot "
1039                                 "failed! Retrying\n");
1040                 task->tk_status = -ENOMEM;
1041                 break;
1042         case -EAGAIN:
1043                 xprt_add_backlog(xprt, task);
1044                 dprintk("RPC:       waiting for request slot\n");
1045         default:
1046                 task->tk_status = -EAGAIN;
1047         }
1048         spin_unlock(&xprt->reserve_lock);
1049         return;
1050 out_init_req:
1051         task->tk_status = 0;
1052         task->tk_rqstp = req;
1053         xprt_request_init(task, xprt);
1054         spin_unlock(&xprt->reserve_lock);
1055 }
1056 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1057
1058 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1059 {
1060         /* Note: grabbing the xprt_lock_write() ensures that we throttle
1061          * new slot allocation if the transport is congested (i.e. when
1062          * reconnecting a stream transport or when out of socket write
1063          * buffer space).
1064          */
1065         if (xprt_lock_write(xprt, task)) {
1066                 xprt_alloc_slot(xprt, task);
1067                 xprt_release_write(xprt, task);
1068         }
1069 }
1070 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1071
1072 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1073 {
1074         spin_lock(&xprt->reserve_lock);
1075         if (!xprt_dynamic_free_slot(xprt, req)) {
1076                 memset(req, 0, sizeof(*req));   /* mark unused */
1077                 list_add(&req->rq_list, &xprt->free);
1078         }
1079         xprt_wake_up_backlog(xprt);
1080         spin_unlock(&xprt->reserve_lock);
1081 }
1082
1083 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1084 {
1085         struct rpc_rqst *req;
1086         while (!list_empty(&xprt->free)) {
1087                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1088                 list_del(&req->rq_list);
1089                 kfree(req);
1090         }
1091 }
1092
1093 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1094                 unsigned int num_prealloc,
1095                 unsigned int max_alloc)
1096 {
1097         struct rpc_xprt *xprt;
1098         struct rpc_rqst *req;
1099         int i;
1100
1101         xprt = kzalloc(size, GFP_KERNEL);
1102         if (xprt == NULL)
1103                 goto out;
1104
1105         xprt_init(xprt, net);
1106
1107         for (i = 0; i < num_prealloc; i++) {
1108                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1109                 if (!req)
1110                         goto out_free;
1111                 list_add(&req->rq_list, &xprt->free);
1112         }
1113         if (max_alloc > num_prealloc)
1114                 xprt->max_reqs = max_alloc;
1115         else
1116                 xprt->max_reqs = num_prealloc;
1117         xprt->min_reqs = num_prealloc;
1118         atomic_set(&xprt->num_reqs, num_prealloc);
1119
1120         return xprt;
1121
1122 out_free:
1123         xprt_free(xprt);
1124 out:
1125         return NULL;
1126 }
1127 EXPORT_SYMBOL_GPL(xprt_alloc);
1128
1129 void xprt_free(struct rpc_xprt *xprt)
1130 {
1131         put_net(xprt->xprt_net);
1132         xprt_free_all_slots(xprt);
1133         kfree(xprt);
1134 }
1135 EXPORT_SYMBOL_GPL(xprt_free);
1136
1137 /**
1138  * xprt_reserve - allocate an RPC request slot
1139  * @task: RPC task requesting a slot allocation
1140  *
1141  * If the transport is marked as being congested, or if no more
1142  * slots are available, place the task on the transport's
1143  * backlog queue.
1144  */
1145 void xprt_reserve(struct rpc_task *task)
1146 {
1147         struct rpc_xprt *xprt;
1148
1149         task->tk_status = 0;
1150         if (task->tk_rqstp != NULL)
1151                 return;
1152
1153         task->tk_timeout = 0;
1154         task->tk_status = -EAGAIN;
1155         rcu_read_lock();
1156         xprt = rcu_dereference(task->tk_client->cl_xprt);
1157         if (!xprt_throttle_congested(xprt, task))
1158                 xprt->ops->alloc_slot(xprt, task);
1159         rcu_read_unlock();
1160 }
1161
1162 /**
1163  * xprt_retry_reserve - allocate an RPC request slot
1164  * @task: RPC task requesting a slot allocation
1165  *
1166  * If no more slots are available, place the task on the transport's
1167  * backlog queue.
1168  * Note that the only difference with xprt_reserve is that we now
1169  * ignore the value of the XPRT_CONGESTED flag.
1170  */
1171 void xprt_retry_reserve(struct rpc_task *task)
1172 {
1173         struct rpc_xprt *xprt;
1174
1175         task->tk_status = 0;
1176         if (task->tk_rqstp != NULL)
1177                 return;
1178
1179         task->tk_timeout = 0;
1180         task->tk_status = -EAGAIN;
1181         rcu_read_lock();
1182         xprt = rcu_dereference(task->tk_client->cl_xprt);
1183         xprt->ops->alloc_slot(xprt, task);
1184         rcu_read_unlock();
1185 }
1186
1187 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1188 {
1189         return (__force __be32)xprt->xid++;
1190 }
1191
1192 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1193 {
1194         xprt->xid = prandom_u32();
1195 }
1196
1197 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1198 {
1199         struct rpc_rqst *req = task->tk_rqstp;
1200
1201         INIT_LIST_HEAD(&req->rq_list);
1202         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1203         req->rq_task    = task;
1204         req->rq_xprt    = xprt;
1205         req->rq_buffer  = NULL;
1206         req->rq_xid     = xprt_alloc_xid(xprt);
1207         req->rq_connect_cookie = xprt->connect_cookie - 1;
1208         req->rq_bytes_sent = 0;
1209         req->rq_snd_buf.len = 0;
1210         req->rq_snd_buf.buflen = 0;
1211         req->rq_rcv_buf.len = 0;
1212         req->rq_rcv_buf.buflen = 0;
1213         req->rq_release_snd_buf = NULL;
1214         xprt_reset_majortimeo(req);
1215         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1216                         req, ntohl(req->rq_xid));
1217 }
1218
1219 /**
1220  * xprt_release - release an RPC request slot
1221  * @task: task which is finished with the slot
1222  *
1223  */
1224 void xprt_release(struct rpc_task *task)
1225 {
1226         struct rpc_xprt *xprt;
1227         struct rpc_rqst *req = task->tk_rqstp;
1228
1229         if (req == NULL) {
1230                 if (task->tk_client) {
1231                         rcu_read_lock();
1232                         xprt = rcu_dereference(task->tk_client->cl_xprt);
1233                         if (xprt->snd_task == task)
1234                                 xprt_release_write(xprt, task);
1235                         rcu_read_unlock();
1236                 }
1237                 return;
1238         }
1239
1240         xprt = req->rq_xprt;
1241         if (task->tk_ops->rpc_count_stats != NULL)
1242                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1243         else if (task->tk_client)
1244                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1245         spin_lock_bh(&xprt->transport_lock);
1246         xprt->ops->release_xprt(xprt, task);
1247         if (xprt->ops->release_request)
1248                 xprt->ops->release_request(task);
1249         if (!list_empty(&req->rq_list))
1250                 list_del(&req->rq_list);
1251         xprt->last_used = jiffies;
1252         if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1253                 mod_timer(&xprt->timer,
1254                                 xprt->last_used + xprt->idle_timeout);
1255         spin_unlock_bh(&xprt->transport_lock);
1256         if (req->rq_buffer)
1257                 xprt->ops->buf_free(req->rq_buffer);
1258         if (req->rq_cred != NULL)
1259                 put_rpccred(req->rq_cred);
1260         task->tk_rqstp = NULL;
1261         if (req->rq_release_snd_buf)
1262                 req->rq_release_snd_buf(req);
1263
1264         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1265         if (likely(!bc_prealloc(req)))
1266                 xprt_free_slot(xprt, req);
1267         else
1268                 xprt_free_bc_request(req);
1269 }
1270
1271 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1272 {
1273         atomic_set(&xprt->count, 1);
1274
1275         spin_lock_init(&xprt->transport_lock);
1276         spin_lock_init(&xprt->reserve_lock);
1277
1278         INIT_LIST_HEAD(&xprt->free);
1279         INIT_LIST_HEAD(&xprt->recv);
1280 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1281         spin_lock_init(&xprt->bc_pa_lock);
1282         INIT_LIST_HEAD(&xprt->bc_pa_list);
1283 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1284
1285         xprt->last_used = jiffies;
1286         xprt->cwnd = RPC_INITCWND;
1287         xprt->bind_index = 0;
1288
1289         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1290         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1291         rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1292         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1293
1294         xprt_init_xid(xprt);
1295
1296         xprt->xprt_net = get_net(net);
1297 }
1298
1299 /**
1300  * xprt_create_transport - create an RPC transport
1301  * @args: rpc transport creation arguments
1302  *
1303  */
1304 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1305 {
1306         int err;
1307         struct rpc_xprt *xprt;
1308         struct xprt_class *t;
1309
1310         spin_lock(&xprt_list_lock);
1311         list_for_each_entry(t, &xprt_list, list) {
1312                 if (t->ident == args->ident) {
1313                         spin_unlock(&xprt_list_lock);
1314                         goto found;
1315                 }
1316         }
1317         spin_unlock(&xprt_list_lock);
1318         dprintk("RPC: transport (%d) not supported\n", args->ident);
1319         return ERR_PTR(-EIO);
1320
1321 found:
1322         xprt = t->setup(args);
1323         if (IS_ERR(xprt)) {
1324                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1325                                 -PTR_ERR(xprt));
1326                 goto out;
1327         }
1328         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1329                 xprt->idle_timeout = 0;
1330         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1331         if (xprt_has_timer(xprt))
1332                 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1333                             (unsigned long)xprt);
1334         else
1335                 init_timer(&xprt->timer);
1336
1337         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1338                 xprt_destroy(xprt);
1339                 return ERR_PTR(-EINVAL);
1340         }
1341         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1342         if (xprt->servername == NULL) {
1343                 xprt_destroy(xprt);
1344                 return ERR_PTR(-ENOMEM);
1345         }
1346
1347         err = rpc_xprt_debugfs_register(xprt);
1348         if (err) {
1349                 xprt_destroy(xprt);
1350                 return ERR_PTR(err);
1351         }
1352
1353         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1354                         xprt->max_reqs);
1355 out:
1356         return xprt;
1357 }
1358
1359 /**
1360  * xprt_destroy - destroy an RPC transport, killing off all requests.
1361  * @xprt: transport to destroy
1362  *
1363  */
1364 static void xprt_destroy(struct rpc_xprt *xprt)
1365 {
1366         dprintk("RPC:       destroying transport %p\n", xprt);
1367         del_timer_sync(&xprt->timer);
1368
1369         rpc_xprt_debugfs_unregister(xprt);
1370         rpc_destroy_wait_queue(&xprt->binding);
1371         rpc_destroy_wait_queue(&xprt->pending);
1372         rpc_destroy_wait_queue(&xprt->sending);
1373         rpc_destroy_wait_queue(&xprt->backlog);
1374         cancel_work_sync(&xprt->task_cleanup);
1375         kfree(xprt->servername);
1376         /*
1377          * Tear down transport state and free the rpc_xprt
1378          */
1379         xprt->ops->destroy(xprt);
1380 }
1381
1382 /**
1383  * xprt_put - release a reference to an RPC transport.
1384  * @xprt: pointer to the transport
1385  *
1386  */
1387 void xprt_put(struct rpc_xprt *xprt)
1388 {
1389         if (atomic_dec_and_test(&xprt->count))
1390                 xprt_destroy(xprt);
1391 }