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 #include <linux/rcupdate.h>
  52 #include <linux/sched/mm.h>
  53
  54 #include <trace/events/sunrpc.h>
  55
  56 #include "sunrpc.h"
  57
  58 /*
  59  * Local variables
  60  */
  61
  62 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  63 # define RPCDBG_FACILITY        RPCDBG_XPRT
  64 #endif
  65
  66 /*
  67  * Local functions
  68  */
  69 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
  70 static __be32   xprt_alloc_xid(struct rpc_xprt *xprt);
  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 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
  77 {
  78         unsigned long timeout = jiffies + req->rq_timeout;
  79
  80         if (time_before(timeout, req->rq_majortimeo))
  81                 return timeout;
  82         return req->rq_majortimeo;
  83 }
  84
  85 /**
  86  * xprt_register_transport - register a transport implementation
  87  * @transport: transport to register
  88  *
  89  * If a transport implementation is loaded as a kernel module, it can
  90  * call this interface to make itself known to the RPC client.
  91  *
  92  * Returns:
  93  * 0:           transport successfully registered
  94  * -EEXIST:     transport already registered
  95  * -EINVAL:     transport module being unloaded
  96  */
  97 int xprt_register_transport(struct xprt_class *transport)
  98 {
  99         struct xprt_class *t;
 100         int result;
 101
 102         result = -EEXIST;
 103         spin_lock(&xprt_list_lock);
 104         list_for_each_entry(t, &xprt_list, list) {
 105                 /* don't register the same transport class twice */
 106                 if (t->ident == transport->ident)
 107                         goto out;
 108         }
 109
 110         list_add_tail(&transport->list, &xprt_list);
 111         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 112                transport->name);
 113         result = 0;
 114
 115 out:
 116         spin_unlock(&xprt_list_lock);
 117         return result;
 118 }
 119 EXPORT_SYMBOL_GPL(xprt_register_transport);
 120
 121 /**
 122  * xprt_unregister_transport - unregister a transport implementation
 123  * @transport: transport to unregister
 124  *
 125  * Returns:
 126  * 0:           transport successfully unregistered
 127  * -ENOENT:     transport never registered
 128  */
 129 int xprt_unregister_transport(struct xprt_class *transport)
 130 {
 131         struct xprt_class *t;
 132         int result;
 133
 134         result = 0;
 135         spin_lock(&xprt_list_lock);
 136         list_for_each_entry(t, &xprt_list, list) {
 137                 if (t == transport) {
 138                         printk(KERN_INFO
 139                                 "RPC: Unregistered %s transport module.\n",
 140                                 transport->name);
 141                         list_del_init(&transport->list);
 142                         goto out;
 143                 }
 144         }
 145         result = -ENOENT;
 146
 147 out:
 148         spin_unlock(&xprt_list_lock);
 149         return result;
 150 }
 151 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 152
 153 /**
 154  * xprt_load_transport - load a transport implementation
 155  * @transport_name: transport to load
 156  *
 157  * Returns:
 158  * 0:           transport successfully loaded
 159  * -ENOENT:     transport module not available
 160  */
 161 int xprt_load_transport(const char *transport_name)
 162 {
 163         struct xprt_class *t;
 164         int result;
 165
 166         result = 0;
 167         spin_lock(&xprt_list_lock);
 168         list_for_each_entry(t, &xprt_list, list) {
 169                 if (strcmp(t->name, transport_name) == 0) {
 170                         spin_unlock(&xprt_list_lock);
 171                         goto out;
 172                 }
 173         }
 174         spin_unlock(&xprt_list_lock);
 175         result = request_module("xprt%s", transport_name);
 176 out:
 177         return result;
 178 }
 179 EXPORT_SYMBOL_GPL(xprt_load_transport);
 180
 181 static void xprt_clear_locked(struct rpc_xprt *xprt)
 182 {
 183         xprt->snd_task = NULL;
 184         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 185                 smp_mb__before_atomic();
 186                 clear_bit(XPRT_LOCKED, &xprt->state);
 187                 smp_mb__after_atomic();
 188         } else
 189                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 190 }
 191
 192 /**
 193  * xprt_reserve_xprt - serialize write access to transports
 194  * @task: task that is requesting access to the transport
 195  * @xprt: pointer to the target transport
 196  *
 197  * This prevents mixing the payload of separate requests, and prevents
 198  * transport connects from colliding with writes.  No congestion control
 199  * is provided.
 200  */
 201 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 202 {
 203         struct rpc_rqst *req = task->tk_rqstp;
 204
 205         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 206                 if (task == xprt->snd_task)
 207                         return 1;
 208                 goto out_sleep;
 209         }
 210         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 211                 goto out_unlock;
 212         xprt->snd_task = task;
 213
 214         return 1;
 215
 216 out_unlock:
 217         xprt_clear_locked(xprt);
 218 out_sleep:
 219         dprintk("RPC: %5u failed to lock transport %p\n",
 220                         task->tk_pid, xprt);
 221         task->tk_status = -EAGAIN;
 222         if  (RPC_IS_SOFT(task))
 223                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 224                                 xprt_request_timeout(req));
 225         else
 226                 rpc_sleep_on(&xprt->sending, task, NULL);
 227         return 0;
 228 }
 229 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 230
 231 static bool
 232 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
 233 {
 234         return test_bit(XPRT_CWND_WAIT, &xprt->state);
 235 }
 236
 237 static void
 238 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
 239 {
 240         if (!list_empty(&xprt->xmit_queue)) {
 241                 /* Peek at head of queue to see if it can make progress */
 242                 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
 243                                         rq_xmit)->rq_cong)
 244                         return;
 245         }
 246         set_bit(XPRT_CWND_WAIT, &xprt->state);
 247 }
 248
 249 static void
 250 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
 251 {
 252         if (!RPCXPRT_CONGESTED(xprt))
 253                 clear_bit(XPRT_CWND_WAIT, &xprt->state);
 254 }
 255
 256 /*
 257  * xprt_reserve_xprt_cong - serialize write access to transports
 258  * @task: task that is requesting access to the transport
 259  *
 260  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 261  * integrated into the decision of whether a request is allowed to be
 262  * woken up and given access to the transport.
 263  * Note that the lock is only granted if we know there are free slots.
 264  */
 265 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 266 {
 267         struct rpc_rqst *req = task->tk_rqstp;
 268
 269         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 270                 if (task == xprt->snd_task)
 271                         return 1;
 272                 goto out_sleep;
 273         }
 274         if (req == NULL) {
 275                 xprt->snd_task = task;
 276                 return 1;
 277         }
 278         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 279                 goto out_unlock;
 280         if (!xprt_need_congestion_window_wait(xprt)) {
 281                 xprt->snd_task = task;
 282                 return 1;
 283         }
 284 out_unlock:
 285         xprt_clear_locked(xprt);
 286 out_sleep:
 287         dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
 288         task->tk_status = -EAGAIN;
 289         if (RPC_IS_SOFT(task))
 290                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 291                                 xprt_request_timeout(req));
 292         else
 293                 rpc_sleep_on(&xprt->sending, task, NULL);
 294         return 0;
 295 }
 296 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 297
 298 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 299 {
 300         int retval;
 301
 302         if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
 303                 return 1;
 304         spin_lock_bh(&xprt->transport_lock);
 305         retval = xprt->ops->reserve_xprt(xprt, task);
 306         spin_unlock_bh(&xprt->transport_lock);
 307         return retval;
 308 }
 309
 310 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 311 {
 312         struct rpc_xprt *xprt = data;
 313
 314         xprt->snd_task = task;
 315         return true;
 316 }
 317
 318 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 319 {
 320         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 321                 return;
 322         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 323                 goto out_unlock;
 324         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 325                                 __xprt_lock_write_func, xprt))
 326                 return;
 327 out_unlock:
 328         xprt_clear_locked(xprt);
 329 }
 330
 331 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 332 {
 333         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 334                 return;
 335         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 336                 goto out_unlock;
 337         if (xprt_need_congestion_window_wait(xprt))
 338                 goto out_unlock;
 339         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 340                                 __xprt_lock_write_func, xprt))
 341                 return;
 342 out_unlock:
 343         xprt_clear_locked(xprt);
 344 }
 345
 346 /**
 347  * xprt_release_xprt - allow other requests to use a transport
 348  * @xprt: transport with other tasks potentially waiting
 349  * @task: task that is releasing access to the transport
 350  *
 351  * Note that "task" can be NULL.  No congestion control is provided.
 352  */
 353 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 354 {
 355         if (xprt->snd_task == task) {
 356                 xprt_clear_locked(xprt);
 357                 __xprt_lock_write_next(xprt);
 358         }
 359 }
 360 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 361
 362 /**
 363  * xprt_release_xprt_cong - allow other requests to use a transport
 364  * @xprt: transport with other tasks potentially waiting
 365  * @task: task that is releasing access to the transport
 366  *
 367  * Note that "task" can be NULL.  Another task is awoken to use the
 368  * transport if the transport's congestion window allows it.
 369  */
 370 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 371 {
 372         if (xprt->snd_task == task) {
 373                 xprt_clear_locked(xprt);
 374                 __xprt_lock_write_next_cong(xprt);
 375         }
 376 }
 377 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 378
 379 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 380 {
 381         if (xprt->snd_task != task)
 382                 return;
 383         spin_lock_bh(&xprt->transport_lock);
 384         xprt->ops->release_xprt(xprt, task);
 385         spin_unlock_bh(&xprt->transport_lock);
 386 }
 387
 388 /*
 389  * Van Jacobson congestion avoidance. Check if the congestion window
 390  * overflowed. Put the task to sleep if this is the case.
 391  */
 392 static int
 393 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 394 {
 395         if (req->rq_cong)
 396                 return 1;
 397         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
 398                         req->rq_task->tk_pid, xprt->cong, xprt->cwnd);
 399         if (RPCXPRT_CONGESTED(xprt)) {
 400                 xprt_set_congestion_window_wait(xprt);
 401                 return 0;
 402         }
 403         req->rq_cong = 1;
 404         xprt->cong += RPC_CWNDSCALE;
 405         return 1;
 406 }
 407
 408 /*
 409  * Adjust the congestion window, and wake up the next task
 410  * that has been sleeping due to congestion
 411  */
 412 static void
 413 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 414 {
 415         if (!req->rq_cong)
 416                 return;
 417         req->rq_cong = 0;
 418         xprt->cong -= RPC_CWNDSCALE;
 419         xprt_test_and_clear_congestion_window_wait(xprt);
 420         __xprt_lock_write_next_cong(xprt);
 421 }
 422
 423 /**
 424  * xprt_request_get_cong - Request congestion control credits
 425  * @xprt: pointer to transport
 426  * @req: pointer to RPC request
 427  *
 428  * Useful for transports that require congestion control.
 429  */
 430 bool
 431 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 432 {
 433         bool ret = false;
 434
 435         if (req->rq_cong)
 436                 return true;
 437         spin_lock_bh(&xprt->transport_lock);
 438         ret = __xprt_get_cong(xprt, req) != 0;
 439         spin_unlock_bh(&xprt->transport_lock);
 440         return ret;
 441 }
 442 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
 443
 444 /**
 445  * xprt_release_rqst_cong - housekeeping when request is complete
 446  * @task: RPC request that recently completed
 447  *
 448  * Useful for transports that require congestion control.
 449  */
 450 void xprt_release_rqst_cong(struct rpc_task *task)
 451 {
 452         struct rpc_rqst *req = task->tk_rqstp;
 453
 454         __xprt_put_cong(req->rq_xprt, req);
 455 }
 456 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 457
 458 /*
 459  * Clear the congestion window wait flag and wake up the next
 460  * entry on xprt->sending
 461  */
 462 static void
 463 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
 464 {
 465         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
 466                 spin_lock_bh(&xprt->transport_lock);
 467                 __xprt_lock_write_next_cong(xprt);
 468                 spin_unlock_bh(&xprt->transport_lock);
 469         }
 470 }
 471
 472 /**
 473  * xprt_adjust_cwnd - adjust transport congestion window
 474  * @xprt: pointer to xprt
 475  * @task: recently completed RPC request used to adjust window
 476  * @result: result code of completed RPC request
 477  *
 478  * The transport code maintains an estimate on the maximum number of out-
 479  * standing RPC requests, using a smoothed version of the congestion
 480  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 481  * congestion algorithm: If a retransmit occurs, the congestion window is
 482  * halved; otherwise, it is incremented by 1/cwnd when
 483  *
 484  *      -       a reply is received and
 485  *      -       a full number of requests are outstanding and
 486  *      -       the congestion window hasn't been updated recently.
 487  */
 488 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 489 {
 490         struct rpc_rqst *req = task->tk_rqstp;
 491         unsigned long cwnd = xprt->cwnd;
 492
 493         if (result >= 0 && cwnd <= xprt->cong) {
 494                 /* The (cwnd >> 1) term makes sure
 495                  * the result gets rounded properly. */
 496                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 497                 if (cwnd > RPC_MAXCWND(xprt))
 498                         cwnd = RPC_MAXCWND(xprt);
 499                 __xprt_lock_write_next_cong(xprt);
 500         } else if (result == -ETIMEDOUT) {
 501                 cwnd >>= 1;
 502                 if (cwnd < RPC_CWNDSCALE)
 503                         cwnd = RPC_CWNDSCALE;
 504         }
 505         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 506                         xprt->cong, xprt->cwnd, cwnd);
 507         xprt->cwnd = cwnd;
 508         __xprt_put_cong(xprt, req);
 509 }
 510 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 511
 512 /**
 513  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 514  * @xprt: transport with waiting tasks
 515  * @status: result code to plant in each task before waking it
 516  *
 517  */
 518 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 519 {
 520         if (status < 0)
 521                 rpc_wake_up_status(&xprt->pending, status);
 522         else
 523                 rpc_wake_up(&xprt->pending);
 524 }
 525 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 526
 527 /**
 528  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 529  * @xprt: transport
 530  *
 531  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 532  * we don't in general want to force a socket disconnection due to
 533  * an incomplete RPC call transmission.
 534  */
 535 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
 536 {
 537         set_bit(XPRT_WRITE_SPACE, &xprt->state);
 538 }
 539 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 540
 541 static bool
 542 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
 543 {
 544         if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
 545                 __xprt_lock_write_next(xprt);
 546                 dprintk("RPC:       write space: waking waiting task on "
 547                                 "xprt %p\n", xprt);
 548                 return true;
 549         }
 550         return false;
 551 }
 552
 553 /**
 554  * xprt_write_space - wake the task waiting for transport output buffer space
 555  * @xprt: transport with waiting tasks
 556  *
 557  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 558  */
 559 bool xprt_write_space(struct rpc_xprt *xprt)
 560 {
 561         bool ret;
 562
 563         if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
 564                 return false;
 565         spin_lock_bh(&xprt->transport_lock);
 566         ret = xprt_clear_write_space_locked(xprt);
 567         spin_unlock_bh(&xprt->transport_lock);
 568         return ret;
 569 }
 570 EXPORT_SYMBOL_GPL(xprt_write_space);
 571
 572 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
 573 {
 574         s64 delta = ktime_to_ns(ktime_get() - abstime);
 575         return likely(delta >= 0) ?
 576                 jiffies - nsecs_to_jiffies(delta) :
 577                 jiffies + nsecs_to_jiffies(-delta);
 578 }
 579
 580 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
 581 {
 582         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 583         unsigned long majortimeo = req->rq_timeout;
 584
 585         if (to->to_exponential)
 586                 majortimeo <<= to->to_retries;
 587         else
 588                 majortimeo += to->to_increment * to->to_retries;
 589         if (majortimeo > to->to_maxval || majortimeo == 0)
 590                 majortimeo = to->to_maxval;
 591         return majortimeo;
 592 }
 593
 594 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 595 {
 596         req->rq_majortimeo += xprt_calc_majortimeo(req);
 597 }
 598
 599 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
 600 {
 601         unsigned long time_init;
 602         struct rpc_xprt *xprt = req->rq_xprt;
 603
 604         if (likely(xprt && xprt_connected(xprt)))
 605                 time_init = jiffies;
 606         else
 607                 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
 608         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
 609         req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
 610 }
 611
 612 /**
 613  * xprt_adjust_timeout - adjust timeout values for next retransmit
 614  * @req: RPC request containing parameters to use for the adjustment
 615  *
 616  */
 617 int xprt_adjust_timeout(struct rpc_rqst *req)
 618 {
 619         struct rpc_xprt *xprt = req->rq_xprt;
 620         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 621         int status = 0;
 622
 623         if (time_before(jiffies, req->rq_majortimeo)) {
 624                 if (to->to_exponential)
 625                         req->rq_timeout <<= 1;
 626                 else
 627                         req->rq_timeout += to->to_increment;
 628                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 629                         req->rq_timeout = to->to_maxval;
 630                 req->rq_retries++;
 631         } else {
 632                 req->rq_timeout = to->to_initval;
 633                 req->rq_retries = 0;
 634                 xprt_reset_majortimeo(req);
 635                 /* Reset the RTT counters == "slow start" */
 636                 spin_lock_bh(&xprt->transport_lock);
 637                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 638                 spin_unlock_bh(&xprt->transport_lock);
 639                 status = -ETIMEDOUT;
 640         }
 641
 642         if (req->rq_timeout == 0) {
 643                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 644                 req->rq_timeout = 5 * HZ;
 645         }
 646         return status;
 647 }
 648
 649 static void xprt_autoclose(struct work_struct *work)
 650 {
 651         struct rpc_xprt *xprt =
 652                 container_of(work, struct rpc_xprt, task_cleanup);
 653         unsigned int pflags = memalloc_nofs_save();
 654
 655         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 656         xprt->ops->close(xprt);
 657         xprt_release_write(xprt, NULL);
 658         wake_up_bit(&xprt->state, XPRT_LOCKED);
 659         memalloc_nofs_restore(pflags);
 660 }
 661
 662 /**
 663  * xprt_disconnect_done - mark a transport as disconnected
 664  * @xprt: transport to flag for disconnect
 665  *
 666  */
 667 void xprt_disconnect_done(struct rpc_xprt *xprt)
 668 {
 669         dprintk("RPC:       disconnected transport %p\n", xprt);
 670         spin_lock_bh(&xprt->transport_lock);
 671         xprt_clear_connected(xprt);
 672         xprt_clear_write_space_locked(xprt);
 673         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 674         spin_unlock_bh(&xprt->transport_lock);
 675 }
 676 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 677
 678 /**
 679  * xprt_force_disconnect - force a transport to disconnect
 680  * @xprt: transport to disconnect
 681  *
 682  */
 683 void xprt_force_disconnect(struct rpc_xprt *xprt)
 684 {
 685         /* Don't race with the test_bit() in xprt_clear_locked() */
 686         spin_lock_bh(&xprt->transport_lock);
 687         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 688         /* Try to schedule an autoclose RPC call */
 689         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 690                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 691         else if (xprt->snd_task)
 692                 rpc_wake_up_queued_task_set_status(&xprt->pending,
 693                                 xprt->snd_task, -ENOTCONN);
 694         spin_unlock_bh(&xprt->transport_lock);
 695 }
 696 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 697
 698 static unsigned int
 699 xprt_connect_cookie(struct rpc_xprt *xprt)
 700 {
 701         return READ_ONCE(xprt->connect_cookie);
 702 }
 703
 704 static bool
 705 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
 706 {
 707         struct rpc_rqst *req = task->tk_rqstp;
 708         struct rpc_xprt *xprt = req->rq_xprt;
 709
 710         return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
 711                 !xprt_connected(xprt);
 712 }
 713
 714 /**
 715  * xprt_conditional_disconnect - force a transport to disconnect
 716  * @xprt: transport to disconnect
 717  * @cookie: 'connection cookie'
 718  *
 719  * This attempts to break the connection if and only if 'cookie' matches
 720  * the current transport 'connection cookie'. It ensures that we don't
 721  * try to break the connection more than once when we need to retransmit
 722  * a batch of RPC requests.
 723  *
 724  */
 725 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 726 {
 727         /* Don't race with the test_bit() in xprt_clear_locked() */
 728         spin_lock_bh(&xprt->transport_lock);
 729         if (cookie != xprt->connect_cookie)
 730                 goto out;
 731         if (test_bit(XPRT_CLOSING, &xprt->state))
 732                 goto out;
 733         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 734         /* Try to schedule an autoclose RPC call */
 735         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 736                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 737         xprt_wake_pending_tasks(xprt, -EAGAIN);
 738 out:
 739         spin_unlock_bh(&xprt->transport_lock);
 740 }
 741
 742 static bool
 743 xprt_has_timer(const struct rpc_xprt *xprt)
 744 {
 745         return xprt->idle_timeout != 0;
 746 }
 747
 748 static void
 749 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 750         __must_hold(&xprt->transport_lock)
 751 {
 752         if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
 753                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 754 }
 755
 756 static void
 757 xprt_init_autodisconnect(struct timer_list *t)
 758 {
 759         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
 760
 761         spin_lock(&xprt->transport_lock);
 762         if (!RB_EMPTY_ROOT(&xprt->recv_queue))
 763                 goto out_abort;
 764         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 765         xprt->last_used = jiffies;
 766         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 767                 goto out_abort;
 768         spin_unlock(&xprt->transport_lock);
 769         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 770         return;
 771 out_abort:
 772         spin_unlock(&xprt->transport_lock);
 773 }
 774
 775 bool xprt_lock_connect(struct rpc_xprt *xprt,
 776                 struct rpc_task *task,
 777                 void *cookie)
 778 {
 779         bool ret = false;
 780
 781         spin_lock_bh(&xprt->transport_lock);
 782         if (!test_bit(XPRT_LOCKED, &xprt->state))
 783                 goto out;
 784         if (xprt->snd_task != task)
 785                 goto out;
 786         xprt->snd_task = cookie;
 787         ret = true;
 788 out:
 789         spin_unlock_bh(&xprt->transport_lock);
 790         return ret;
 791 }
 792
 793 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 794 {
 795         spin_lock_bh(&xprt->transport_lock);
 796         if (xprt->snd_task != cookie)
 797                 goto out;
 798         if (!test_bit(XPRT_LOCKED, &xprt->state))
 799                 goto out;
 800         xprt->snd_task =NULL;
 801         xprt->ops->release_xprt(xprt, NULL);
 802         xprt_schedule_autodisconnect(xprt);
 803 out:
 804         spin_unlock_bh(&xprt->transport_lock);
 805         wake_up_bit(&xprt->state, XPRT_LOCKED);
 806 }
 807
 808 /**
 809  * xprt_connect - schedule a transport connect operation
 810  * @task: RPC task that is requesting the connect
 811  *
 812  */
 813 void xprt_connect(struct rpc_task *task)
 814 {
 815         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 816
 817         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 818                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 819
 820         if (!xprt_bound(xprt)) {
 821                 task->tk_status = -EAGAIN;
 822                 return;
 823         }
 824         if (!xprt_lock_write(xprt, task))
 825                 return;
 826
 827         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 828                 xprt->ops->close(xprt);
 829
 830         if (!xprt_connected(xprt)) {
 831                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 832                 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
 833                                 xprt_request_timeout(task->tk_rqstp));
 834
 835                 if (test_bit(XPRT_CLOSING, &xprt->state))
 836                         return;
 837                 if (xprt_test_and_set_connecting(xprt))
 838                         return;
 839                 /* Race breaker */
 840                 if (!xprt_connected(xprt)) {
 841                         xprt->stat.connect_start = jiffies;
 842                         xprt->ops->connect(xprt, task);
 843                 } else {
 844                         xprt_clear_connecting(xprt);
 845                         task->tk_status = 0;
 846                         rpc_wake_up_queued_task(&xprt->pending, task);
 847                 }
 848         }
 849         xprt_release_write(xprt, task);
 850 }
 851
 852 enum xprt_xid_rb_cmp {
 853         XID_RB_EQUAL,
 854         XID_RB_LEFT,
 855         XID_RB_RIGHT,
 856 };
 857 static enum xprt_xid_rb_cmp
 858 xprt_xid_cmp(__be32 xid1, __be32 xid2)
 859 {
 860         if (xid1 == xid2)
 861                 return XID_RB_EQUAL;
 862         if ((__force u32)xid1 < (__force u32)xid2)
 863                 return XID_RB_LEFT;
 864         return XID_RB_RIGHT;
 865 }
 866
 867 static struct rpc_rqst *
 868 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
 869 {
 870         struct rb_node *n = xprt->recv_queue.rb_node;
 871         struct rpc_rqst *req;
 872
 873         while (n != NULL) {
 874                 req = rb_entry(n, struct rpc_rqst, rq_recv);
 875                 switch (xprt_xid_cmp(xid, req->rq_xid)) {
 876                 case XID_RB_LEFT:
 877                         n = n->rb_left;
 878                         break;
 879                 case XID_RB_RIGHT:
 880                         n = n->rb_right;
 881                         break;
 882                 case XID_RB_EQUAL:
 883                         return req;
 884                 }
 885         }
 886         return NULL;
 887 }
 888
 889 static void
 890 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
 891 {
 892         struct rb_node **p = &xprt->recv_queue.rb_node;
 893         struct rb_node *n = NULL;
 894         struct rpc_rqst *req;
 895
 896         while (*p != NULL) {
 897                 n = *p;
 898                 req = rb_entry(n, struct rpc_rqst, rq_recv);
 899                 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
 900                 case XID_RB_LEFT:
 901                         p = &n->rb_left;
 902                         break;
 903                 case XID_RB_RIGHT:
 904                         p = &n->rb_right;
 905                         break;
 906                 case XID_RB_EQUAL:
 907                         WARN_ON_ONCE(new != req);
 908                         return;
 909                 }
 910         }
 911         rb_link_node(&new->rq_recv, n, p);
 912         rb_insert_color(&new->rq_recv, &xprt->recv_queue);
 913 }
 914
 915 static void
 916 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
 917 {
 918         rb_erase(&req->rq_recv, &xprt->recv_queue);
 919 }
 920
 921 /**
 922  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 923  * @xprt: transport on which the original request was transmitted
 924  * @xid: RPC XID of incoming reply
 925  *
 926  * Caller holds xprt->queue_lock.
 927  */
 928 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 929 {
 930         struct rpc_rqst *entry;
 931
 932         entry = xprt_request_rb_find(xprt, xid);
 933         if (entry != NULL) {
 934                 trace_xprt_lookup_rqst(xprt, xid, 0);
 935                 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
 936                 return entry;
 937         }
 938
 939         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 940                         ntohl(xid));
 941         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 942         xprt->stat.bad_xids++;
 943         return NULL;
 944 }
 945 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 946
 947 static bool
 948 xprt_is_pinned_rqst(struct rpc_rqst *req)
 949 {
 950         return atomic_read(&req->rq_pin) != 0;
 951 }
 952
 953 /**
 954  * xprt_pin_rqst - Pin a request on the transport receive list
 955  * @req: Request to pin
 956  *
 957  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
 958  * so should be holding xprt->queue_lock.
 959  */
 960 void xprt_pin_rqst(struct rpc_rqst *req)
 961 {
 962         atomic_inc(&req->rq_pin);
 963 }
 964 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
 965
 966 /**
 967  * xprt_unpin_rqst - Unpin a request on the transport receive list
 968  * @req: Request to pin
 969  *
 970  * Caller should be holding xprt->queue_lock.
 971  */
 972 void xprt_unpin_rqst(struct rpc_rqst *req)
 973 {
 974         if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
 975                 atomic_dec(&req->rq_pin);
 976                 return;
 977         }
 978         if (atomic_dec_and_test(&req->rq_pin))
 979                 wake_up_var(&req->rq_pin);
 980 }
 981 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
 982
 983 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
 984 {
 985         wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
 986 }
 987
 988 static bool
 989 xprt_request_data_received(struct rpc_task *task)
 990 {
 991         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
 992                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
 993 }
 994
 995 static bool
 996 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
 997 {
 998         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
 999                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1000 }
1001
1002 /**
1003  * xprt_request_enqueue_receive - Add an request to the receive queue
1004  * @task: RPC task
1005  *
1006  */
1007 void
1008 xprt_request_enqueue_receive(struct rpc_task *task)
1009 {
1010         struct rpc_rqst *req = task->tk_rqstp;
1011         struct rpc_xprt *xprt = req->rq_xprt;
1012
1013         if (!xprt_request_need_enqueue_receive(task, req))
1014                 return;
1015         spin_lock(&xprt->queue_lock);
1016
1017         /* Update the softirq receive buffer */
1018         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1019                         sizeof(req->rq_private_buf));
1020
1021         /* Add request to the receive list */
1022         xprt_request_rb_insert(xprt, req);
1023         set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1024         spin_unlock(&xprt->queue_lock);
1025
1026         /* Turn off autodisconnect */
1027         del_singleshot_timer_sync(&xprt->timer);
1028 }
1029
1030 /**
1031  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1032  * @task: RPC task
1033  *
1034  * Caller must hold xprt->queue_lock.
1035  */
1036 static void
1037 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1038 {
1039         struct rpc_rqst *req = task->tk_rqstp;
1040
1041         if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1042                 xprt_request_rb_remove(req->rq_xprt, req);
1043 }
1044
1045 /**
1046  * xprt_update_rtt - Update RPC RTT statistics
1047  * @task: RPC request that recently completed
1048  *
1049  * Caller holds xprt->queue_lock.
1050  */
1051 void xprt_update_rtt(struct rpc_task *task)
1052 {
1053         struct rpc_rqst *req = task->tk_rqstp;
1054         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1055         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1056         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1057
1058         if (timer) {
1059                 if (req->rq_ntrans == 1)
1060                         rpc_update_rtt(rtt, timer, m);
1061                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1062         }
1063 }
1064 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1065
1066 /**
1067  * xprt_complete_rqst - called when reply processing is complete
1068  * @task: RPC request that recently completed
1069  * @copied: actual number of bytes received from the transport
1070  *
1071  * Caller holds xprt->queue_lock.
1072  */
1073 void xprt_complete_rqst(struct rpc_task *task, int copied)
1074 {
1075         struct rpc_rqst *req = task->tk_rqstp;
1076         struct rpc_xprt *xprt = req->rq_xprt;
1077
1078         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
1079                         task->tk_pid, ntohl(req->rq_xid), copied);
1080         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
1081
1082         xprt->stat.recvs++;
1083
1084         req->rq_private_buf.len = copied;
1085         /* Ensure all writes are done before we update */
1086         /* req->rq_reply_bytes_recvd */
1087         smp_wmb();
1088         req->rq_reply_bytes_recvd = copied;
1089         xprt_request_dequeue_receive_locked(task);
1090         rpc_wake_up_queued_task(&xprt->pending, task);
1091 }
1092 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1093
1094 static void xprt_timer(struct rpc_task *task)
1095 {
1096         struct rpc_rqst *req = task->tk_rqstp;
1097         struct rpc_xprt *xprt = req->rq_xprt;
1098
1099         if (task->tk_status != -ETIMEDOUT)
1100                 return;
1101
1102         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1103         if (!req->rq_reply_bytes_recvd) {
1104                 if (xprt->ops->timer)
1105                         xprt->ops->timer(xprt, task);
1106         } else
1107                 task->tk_status = 0;
1108 }
1109
1110 /**
1111  * xprt_wait_for_reply_request_def - wait for reply
1112  * @task: pointer to rpc_task
1113  *
1114  * Set a request's retransmit timeout based on the transport's
1115  * default timeout parameters.  Used by transports that don't adjust
1116  * the retransmit timeout based on round-trip time estimation,
1117  * and put the task to sleep on the pending queue.
1118  */
1119 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1120 {
1121         struct rpc_rqst *req = task->tk_rqstp;
1122
1123         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1124                         xprt_request_timeout(req));
1125 }
1126 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1127
1128 /**
1129  * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1130  * @task: pointer to rpc_task
1131  *
1132  * Set a request's retransmit timeout using the RTT estimator,
1133  * and put the task to sleep on the pending queue.
1134  */
1135 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1136 {
1137         int timer = task->tk_msg.rpc_proc->p_timer;
1138         struct rpc_clnt *clnt = task->tk_client;
1139         struct rpc_rtt *rtt = clnt->cl_rtt;
1140         struct rpc_rqst *req = task->tk_rqstp;
1141         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1142         unsigned long timeout;
1143
1144         timeout = rpc_calc_rto(rtt, timer);
1145         timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1146         if (timeout > max_timeout || timeout == 0)
1147                 timeout = max_timeout;
1148         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1149                         jiffies + timeout);
1150 }
1151 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1152
1153 /**
1154  * xprt_request_wait_receive - wait for the reply to an RPC request
1155  * @task: RPC task about to send a request
1156  *
1157  */
1158 void xprt_request_wait_receive(struct rpc_task *task)
1159 {
1160         struct rpc_rqst *req = task->tk_rqstp;
1161         struct rpc_xprt *xprt = req->rq_xprt;
1162
1163         if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1164                 return;
1165         /*
1166          * Sleep on the pending queue if we're expecting a reply.
1167          * The spinlock ensures atomicity between the test of
1168          * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1169          */
1170         spin_lock(&xprt->queue_lock);
1171         if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1172                 xprt->ops->wait_for_reply_request(task);
1173                 /*
1174                  * Send an extra queue wakeup call if the
1175                  * connection was dropped in case the call to
1176                  * rpc_sleep_on() raced.
1177                  */
1178                 if (xprt_request_retransmit_after_disconnect(task))
1179                         rpc_wake_up_queued_task_set_status(&xprt->pending,
1180                                         task, -ENOTCONN);
1181         }
1182         spin_unlock(&xprt->queue_lock);
1183 }
1184
1185 static bool
1186 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1187 {
1188         return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1189 }
1190
1191 /**
1192  * xprt_request_enqueue_transmit - queue a task for transmission
1193  * @task: pointer to rpc_task
1194  *
1195  * Add a task to the transmission queue.
1196  */
1197 void
1198 xprt_request_enqueue_transmit(struct rpc_task *task)
1199 {
1200         struct rpc_rqst *pos, *req = task->tk_rqstp;
1201         struct rpc_xprt *xprt = req->rq_xprt;
1202
1203         if (xprt_request_need_enqueue_transmit(task, req)) {
1204                 req->rq_bytes_sent = 0;
1205                 spin_lock(&xprt->queue_lock);
1206                 /*
1207                  * Requests that carry congestion control credits are added
1208                  * to the head of the list to avoid starvation issues.
1209                  */
1210                 if (req->rq_cong) {
1211                         xprt_clear_congestion_window_wait(xprt);
1212                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1213                                 if (pos->rq_cong)
1214                                         continue;
1215                                 /* Note: req is added _before_ pos */
1216                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1217                                 INIT_LIST_HEAD(&req->rq_xmit2);
1218                                 trace_xprt_enq_xmit(task, 1);
1219                                 goto out;
1220                         }
1221                 } else if (RPC_IS_SWAPPER(task)) {
1222                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1223                                 if (pos->rq_cong || pos->rq_bytes_sent)
1224                                         continue;
1225                                 if (RPC_IS_SWAPPER(pos->rq_task))
1226                                         continue;
1227                                 /* Note: req is added _before_ pos */
1228                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1229                                 INIT_LIST_HEAD(&req->rq_xmit2);
1230                                 trace_xprt_enq_xmit(task, 2);
1231                                 goto out;
1232                         }
1233                 } else if (!req->rq_seqno) {
1234                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1235                                 if (pos->rq_task->tk_owner != task->tk_owner)
1236                                         continue;
1237                                 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1238                                 INIT_LIST_HEAD(&req->rq_xmit);
1239                                 trace_xprt_enq_xmit(task, 3);
1240                                 goto out;
1241                         }
1242                 }
1243                 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1244                 INIT_LIST_HEAD(&req->rq_xmit2);
1245                 trace_xprt_enq_xmit(task, 4);
1246 out:
1247                 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1248                 spin_unlock(&xprt->queue_lock);
1249         }
1250 }
1251
1252 /**
1253  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1254  * @task: pointer to rpc_task
1255  *
1256  * Remove a task from the transmission queue
1257  * Caller must hold xprt->queue_lock
1258  */
1259 static void
1260 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1261 {
1262         struct rpc_rqst *req = task->tk_rqstp;
1263
1264         if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1265                 return;
1266         if (!list_empty(&req->rq_xmit)) {
1267                 list_del(&req->rq_xmit);
1268                 if (!list_empty(&req->rq_xmit2)) {
1269                         struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1270                                         struct rpc_rqst, rq_xmit2);
1271                         list_del(&req->rq_xmit2);
1272                         list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1273                 }
1274         } else
1275                 list_del(&req->rq_xmit2);
1276 }
1277
1278 /**
1279  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1280  * @task: pointer to rpc_task
1281  *
1282  * Remove a task from the transmission queue
1283  */
1284 static void
1285 xprt_request_dequeue_transmit(struct rpc_task *task)
1286 {
1287         struct rpc_rqst *req = task->tk_rqstp;
1288         struct rpc_xprt *xprt = req->rq_xprt;
1289
1290         spin_lock(&xprt->queue_lock);
1291         xprt_request_dequeue_transmit_locked(task);
1292         spin_unlock(&xprt->queue_lock);
1293 }
1294
1295 /**
1296  * xprt_request_prepare - prepare an encoded request for transport
1297  * @req: pointer to rpc_rqst
1298  *
1299  * Calls into the transport layer to do whatever is needed to prepare
1300  * the request for transmission or receive.
1301  */
1302 void
1303 xprt_request_prepare(struct rpc_rqst *req)
1304 {
1305         struct rpc_xprt *xprt = req->rq_xprt;
1306
1307         if (xprt->ops->prepare_request)
1308                 xprt->ops->prepare_request(req);
1309 }
1310
1311 /**
1312  * xprt_request_need_retransmit - Test if a task needs retransmission
1313  * @task: pointer to rpc_task
1314  *
1315  * Test for whether a connection breakage requires the task to retransmit
1316  */
1317 bool
1318 xprt_request_need_retransmit(struct rpc_task *task)
1319 {
1320         return xprt_request_retransmit_after_disconnect(task);
1321 }
1322
1323 /**
1324  * xprt_prepare_transmit - reserve the transport before sending a request
1325  * @task: RPC task about to send a request
1326  *
1327  */
1328 bool xprt_prepare_transmit(struct rpc_task *task)
1329 {
1330         struct rpc_rqst *req = task->tk_rqstp;
1331         struct rpc_xprt *xprt = req->rq_xprt;
1332
1333         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1334
1335         if (!xprt_lock_write(xprt, task)) {
1336                 /* Race breaker: someone may have transmitted us */
1337                 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1338                         rpc_wake_up_queued_task_set_status(&xprt->sending,
1339                                         task, 0);
1340                 return false;
1341
1342         }
1343         return true;
1344 }
1345
1346 void xprt_end_transmit(struct rpc_task *task)
1347 {
1348         xprt_release_write(task->tk_rqstp->rq_xprt, task);
1349 }
1350
1351 /**
1352  * xprt_request_transmit - send an RPC request on a transport
1353  * @req: pointer to request to transmit
1354  * @snd_task: RPC task that owns the transport lock
1355  *
1356  * This performs the transmission of a single request.
1357  * Note that if the request is not the same as snd_task, then it
1358  * does need to be pinned.
1359  * Returns '0' on success.
1360  */
1361 static int
1362 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1363 {
1364         struct rpc_xprt *xprt = req->rq_xprt;
1365         struct rpc_task *task = req->rq_task;
1366         unsigned int connect_cookie;
1367         int is_retrans = RPC_WAS_SENT(task);
1368         int status;
1369
1370         if (!req->rq_bytes_sent) {
1371                 if (xprt_request_data_received(task)) {
1372                         status = 0;
1373                         goto out_dequeue;
1374                 }
1375                 /* Verify that our message lies in the RPCSEC_GSS window */
1376                 if (rpcauth_xmit_need_reencode(task)) {
1377                         status = -EBADMSG;
1378                         goto out_dequeue;
1379                 }
1380                 if (task->tk_ops->rpc_call_prepare_transmit) {
1381                         task->tk_ops->rpc_call_prepare_transmit(task,
1382                                         task->tk_calldata);
1383                         status = task->tk_status;
1384                         if (status < 0)
1385                                 goto out_dequeue;
1386                 }
1387                 if (RPC_SIGNALLED(task)) {
1388                         status = -ERESTARTSYS;
1389                         goto out_dequeue;
1390                 }
1391         }
1392
1393         /*
1394          * Update req->rq_ntrans before transmitting to avoid races with
1395          * xprt_update_rtt(), which needs to know that it is recording a
1396          * reply to the first transmission.
1397          */
1398         req->rq_ntrans++;
1399
1400         connect_cookie = xprt->connect_cookie;
1401         status = xprt->ops->send_request(req);
1402         if (status != 0) {
1403                 req->rq_ntrans--;
1404                 trace_xprt_transmit(req, status);
1405                 return status;
1406         }
1407
1408         if (is_retrans)
1409                 task->tk_client->cl_stats->rpcretrans++;
1410
1411         xprt_inject_disconnect(xprt);
1412
1413         task->tk_flags |= RPC_TASK_SENT;
1414         spin_lock_bh(&xprt->transport_lock);
1415
1416         xprt->stat.sends++;
1417         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1418         xprt->stat.bklog_u += xprt->backlog.qlen;
1419         xprt->stat.sending_u += xprt->sending.qlen;
1420         xprt->stat.pending_u += xprt->pending.qlen;
1421         spin_unlock_bh(&xprt->transport_lock);
1422
1423         req->rq_connect_cookie = connect_cookie;
1424 out_dequeue:
1425         trace_xprt_transmit(req, status);
1426         xprt_request_dequeue_transmit(task);
1427         rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1428         return status;
1429 }
1430
1431 /**
1432  * xprt_transmit - send an RPC request on a transport
1433  * @task: controlling RPC task
1434  *
1435  * Attempts to drain the transmit queue. On exit, either the transport
1436  * signalled an error that needs to be handled before transmission can
1437  * resume, or @task finished transmitting, and detected that it already
1438  * received a reply.
1439  */
1440 void
1441 xprt_transmit(struct rpc_task *task)
1442 {
1443         struct rpc_rqst *next, *req = task->tk_rqstp;
1444         struct rpc_xprt *xprt = req->rq_xprt;
1445         int status;
1446
1447         spin_lock(&xprt->queue_lock);
1448         while (!list_empty(&xprt->xmit_queue)) {
1449                 next = list_first_entry(&xprt->xmit_queue,
1450                                 struct rpc_rqst, rq_xmit);
1451                 xprt_pin_rqst(next);
1452                 spin_unlock(&xprt->queue_lock);
1453                 status = xprt_request_transmit(next, task);
1454                 if (status == -EBADMSG && next != req)
1455                         status = 0;
1456                 cond_resched();
1457                 spin_lock(&xprt->queue_lock);
1458                 xprt_unpin_rqst(next);
1459                 if (status == 0) {
1460                         if (!xprt_request_data_received(task) ||
1461                             test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1462                                 continue;
1463                 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1464                         task->tk_status = status;
1465                 break;
1466         }
1467         spin_unlock(&xprt->queue_lock);
1468 }
1469
1470 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1471 {
1472         set_bit(XPRT_CONGESTED, &xprt->state);
1473         rpc_sleep_on(&xprt->backlog, task, NULL);
1474 }
1475
1476 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1477 {
1478         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1479                 clear_bit(XPRT_CONGESTED, &xprt->state);
1480 }
1481
1482 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1483 {
1484         bool ret = false;
1485
1486         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1487                 goto out;
1488         spin_lock(&xprt->reserve_lock);
1489         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1490                 rpc_sleep_on(&xprt->backlog, task, NULL);
1491                 ret = true;
1492         }
1493         spin_unlock(&xprt->reserve_lock);
1494 out:
1495         return ret;
1496 }
1497
1498 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1499 {
1500         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1501
1502         if (xprt->num_reqs >= xprt->max_reqs)
1503                 goto out;
1504         ++xprt->num_reqs;
1505         spin_unlock(&xprt->reserve_lock);
1506         req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1507         spin_lock(&xprt->reserve_lock);
1508         if (req != NULL)
1509                 goto out;
1510         --xprt->num_reqs;
1511         req = ERR_PTR(-ENOMEM);
1512 out:
1513         return req;
1514 }
1515
1516 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1517 {
1518         if (xprt->num_reqs > xprt->min_reqs) {
1519                 --xprt->num_reqs;
1520                 kfree(req);
1521                 return true;
1522         }
1523         return false;
1524 }
1525
1526 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1527 {
1528         struct rpc_rqst *req;
1529
1530         spin_lock(&xprt->reserve_lock);
1531         if (!list_empty(&xprt->free)) {
1532                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1533                 list_del(&req->rq_list);
1534                 goto out_init_req;
1535         }
1536         req = xprt_dynamic_alloc_slot(xprt);
1537         if (!IS_ERR(req))
1538                 goto out_init_req;
1539         switch (PTR_ERR(req)) {
1540         case -ENOMEM:
1541                 dprintk("RPC:       dynamic allocation of request slot "
1542                                 "failed! Retrying\n");
1543                 task->tk_status = -ENOMEM;
1544                 break;
1545         case -EAGAIN:
1546                 xprt_add_backlog(xprt, task);
1547                 dprintk("RPC:       waiting for request slot\n");
1548                 /* fall through */
1549         default:
1550                 task->tk_status = -EAGAIN;
1551         }
1552         spin_unlock(&xprt->reserve_lock);
1553         return;
1554 out_init_req:
1555         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1556                                      xprt->num_reqs);
1557         spin_unlock(&xprt->reserve_lock);
1558
1559         task->tk_status = 0;
1560         task->tk_rqstp = req;
1561 }
1562 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1563
1564 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1565 {
1566         spin_lock(&xprt->reserve_lock);
1567         if (!xprt_dynamic_free_slot(xprt, req)) {
1568                 memset(req, 0, sizeof(*req));   /* mark unused */
1569                 list_add(&req->rq_list, &xprt->free);
1570         }
1571         xprt_wake_up_backlog(xprt);
1572         spin_unlock(&xprt->reserve_lock);
1573 }
1574 EXPORT_SYMBOL_GPL(xprt_free_slot);
1575
1576 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1577 {
1578         struct rpc_rqst *req;
1579         while (!list_empty(&xprt->free)) {
1580                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1581                 list_del(&req->rq_list);
1582                 kfree(req);
1583         }
1584 }
1585
1586 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1587                 unsigned int num_prealloc,
1588                 unsigned int max_alloc)
1589 {
1590         struct rpc_xprt *xprt;
1591         struct rpc_rqst *req;
1592         int i;
1593
1594         xprt = kzalloc(size, GFP_KERNEL);
1595         if (xprt == NULL)
1596                 goto out;
1597
1598         xprt_init(xprt, net);
1599
1600         for (i = 0; i < num_prealloc; i++) {
1601                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1602                 if (!req)
1603                         goto out_free;
1604                 list_add(&req->rq_list, &xprt->free);
1605         }
1606         if (max_alloc > num_prealloc)
1607                 xprt->max_reqs = max_alloc;
1608         else
1609                 xprt->max_reqs = num_prealloc;
1610         xprt->min_reqs = num_prealloc;
1611         xprt->num_reqs = num_prealloc;
1612
1613         return xprt;
1614
1615 out_free:
1616         xprt_free(xprt);
1617 out:
1618         return NULL;
1619 }
1620 EXPORT_SYMBOL_GPL(xprt_alloc);
1621
1622 void xprt_free(struct rpc_xprt *xprt)
1623 {
1624         put_net(xprt->xprt_net);
1625         xprt_free_all_slots(xprt);
1626         kfree_rcu(xprt, rcu);
1627 }
1628 EXPORT_SYMBOL_GPL(xprt_free);
1629
1630 static void
1631 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1632 {
1633         req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1634 }
1635
1636 static __be32
1637 xprt_alloc_xid(struct rpc_xprt *xprt)
1638 {
1639         __be32 xid;
1640
1641         spin_lock(&xprt->reserve_lock);
1642         xid = (__force __be32)xprt->xid++;
1643         spin_unlock(&xprt->reserve_lock);
1644         return xid;
1645 }
1646
1647 static void
1648 xprt_init_xid(struct rpc_xprt *xprt)
1649 {
1650         xprt->xid = prandom_u32();
1651 }
1652
1653 static void
1654 xprt_request_init(struct rpc_task *task)
1655 {
1656         struct rpc_xprt *xprt = task->tk_xprt;
1657         struct rpc_rqst *req = task->tk_rqstp;
1658
1659         req->rq_task    = task;
1660         req->rq_xprt    = xprt;
1661         req->rq_buffer  = NULL;
1662         req->rq_xid     = xprt_alloc_xid(xprt);
1663         xprt_init_connect_cookie(req, xprt);
1664         req->rq_snd_buf.len = 0;
1665         req->rq_snd_buf.buflen = 0;
1666         req->rq_rcv_buf.len = 0;
1667         req->rq_rcv_buf.buflen = 0;
1668         req->rq_snd_buf.bvec = NULL;
1669         req->rq_rcv_buf.bvec = NULL;
1670         req->rq_release_snd_buf = NULL;
1671         xprt_init_majortimeo(task, req);
1672         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1673                         req, ntohl(req->rq_xid));
1674 }
1675
1676 static void
1677 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1678 {
1679         xprt->ops->alloc_slot(xprt, task);
1680         if (task->tk_rqstp != NULL)
1681                 xprt_request_init(task);
1682 }
1683
1684 /**
1685  * xprt_reserve - allocate an RPC request slot
1686  * @task: RPC task requesting a slot allocation
1687  *
1688  * If the transport is marked as being congested, or if no more
1689  * slots are available, place the task on the transport's
1690  * backlog queue.
1691  */
1692 void xprt_reserve(struct rpc_task *task)
1693 {
1694         struct rpc_xprt *xprt = task->tk_xprt;
1695
1696         task->tk_status = 0;
1697         if (task->tk_rqstp != NULL)
1698                 return;
1699
1700         task->tk_status = -EAGAIN;
1701         if (!xprt_throttle_congested(xprt, task))
1702                 xprt_do_reserve(xprt, task);
1703 }
1704
1705 /**
1706  * xprt_retry_reserve - allocate an RPC request slot
1707  * @task: RPC task requesting a slot allocation
1708  *
1709  * If no more slots are available, place the task on the transport's
1710  * backlog queue.
1711  * Note that the only difference with xprt_reserve is that we now
1712  * ignore the value of the XPRT_CONGESTED flag.
1713  */
1714 void xprt_retry_reserve(struct rpc_task *task)
1715 {
1716         struct rpc_xprt *xprt = task->tk_xprt;
1717
1718         task->tk_status = 0;
1719         if (task->tk_rqstp != NULL)
1720                 return;
1721
1722         task->tk_status = -EAGAIN;
1723         xprt_do_reserve(xprt, task);
1724 }
1725
1726 static void
1727 xprt_request_dequeue_all(struct rpc_task *task, struct rpc_rqst *req)
1728 {
1729         struct rpc_xprt *xprt = req->rq_xprt;
1730
1731         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1732             test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1733             xprt_is_pinned_rqst(req)) {
1734                 spin_lock(&xprt->queue_lock);
1735                 xprt_request_dequeue_transmit_locked(task);
1736                 xprt_request_dequeue_receive_locked(task);
1737                 while (xprt_is_pinned_rqst(req)) {
1738                         set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1739                         spin_unlock(&xprt->queue_lock);
1740                         xprt_wait_on_pinned_rqst(req);
1741                         spin_lock(&xprt->queue_lock);
1742                         clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1743                 }
1744                 spin_unlock(&xprt->queue_lock);
1745         }
1746 }
1747
1748 /**
1749  * xprt_release - release an RPC request slot
1750  * @task: task which is finished with the slot
1751  *
1752  */
1753 void xprt_release(struct rpc_task *task)
1754 {
1755         struct rpc_xprt *xprt;
1756         struct rpc_rqst *req = task->tk_rqstp;
1757
1758         if (req == NULL) {
1759                 if (task->tk_client) {
1760                         xprt = task->tk_xprt;
1761                         xprt_release_write(xprt, task);
1762                 }
1763                 return;
1764         }
1765
1766         xprt = req->rq_xprt;
1767         if (task->tk_ops->rpc_count_stats != NULL)
1768                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1769         else if (task->tk_client)
1770                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1771         xprt_request_dequeue_all(task, req);
1772         spin_lock_bh(&xprt->transport_lock);
1773         xprt->ops->release_xprt(xprt, task);
1774         if (xprt->ops->release_request)
1775                 xprt->ops->release_request(task);
1776         xprt->last_used = jiffies;
1777         xprt_schedule_autodisconnect(xprt);
1778         spin_unlock_bh(&xprt->transport_lock);
1779         if (req->rq_buffer)
1780                 xprt->ops->buf_free(task);
1781         xprt_inject_disconnect(xprt);
1782         xdr_free_bvec(&req->rq_rcv_buf);
1783         xdr_free_bvec(&req->rq_snd_buf);
1784         if (req->rq_cred != NULL)
1785                 put_rpccred(req->rq_cred);
1786         task->tk_rqstp = NULL;
1787         if (req->rq_release_snd_buf)
1788                 req->rq_release_snd_buf(req);
1789
1790         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1791         if (likely(!bc_prealloc(req)))
1792                 xprt->ops->free_slot(xprt, req);
1793         else
1794                 xprt_free_bc_request(req);
1795 }
1796
1797 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1798 void
1799 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1800 {
1801         struct xdr_buf *xbufp = &req->rq_snd_buf;
1802
1803         task->tk_rqstp = req;
1804         req->rq_task = task;
1805         xprt_init_connect_cookie(req, req->rq_xprt);
1806         /*
1807          * Set up the xdr_buf length.
1808          * This also indicates that the buffer is XDR encoded already.
1809          */
1810         xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1811                 xbufp->tail[0].iov_len;
1812 }
1813 #endif
1814
1815 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1816 {
1817         kref_init(&xprt->kref);
1818
1819         spin_lock_init(&xprt->transport_lock);
1820         spin_lock_init(&xprt->reserve_lock);
1821         spin_lock_init(&xprt->queue_lock);
1822
1823         INIT_LIST_HEAD(&xprt->free);
1824         xprt->recv_queue = RB_ROOT;
1825         INIT_LIST_HEAD(&xprt->xmit_queue);
1826 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1827         spin_lock_init(&xprt->bc_pa_lock);
1828         INIT_LIST_HEAD(&xprt->bc_pa_list);
1829 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1830         INIT_LIST_HEAD(&xprt->xprt_switch);
1831
1832         xprt->last_used = jiffies;
1833         xprt->cwnd = RPC_INITCWND;
1834         xprt->bind_index = 0;
1835
1836         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1837         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1838         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1839         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1840
1841         xprt_init_xid(xprt);
1842
1843         xprt->xprt_net = get_net(net);
1844 }
1845
1846 /**
1847  * xprt_create_transport - create an RPC transport
1848  * @args: rpc transport creation arguments
1849  *
1850  */
1851 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1852 {
1853         struct rpc_xprt *xprt;
1854         struct xprt_class *t;
1855
1856         spin_lock(&xprt_list_lock);
1857         list_for_each_entry(t, &xprt_list, list) {
1858                 if (t->ident == args->ident) {
1859                         spin_unlock(&xprt_list_lock);
1860                         goto found;
1861                 }
1862         }
1863         spin_unlock(&xprt_list_lock);
1864         dprintk("RPC: transport (%d) not supported\n", args->ident);
1865         return ERR_PTR(-EIO);
1866
1867 found:
1868         xprt = t->setup(args);
1869         if (IS_ERR(xprt)) {
1870                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1871                                 -PTR_ERR(xprt));
1872                 goto out;
1873         }
1874         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1875                 xprt->idle_timeout = 0;
1876         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1877         if (xprt_has_timer(xprt))
1878                 timer_setup(&xprt->timer,
1879                                 xprt_init_autodisconnect,
1880                                 TIMER_DEFERRABLE);
1881         else
1882                 timer_setup(&xprt->timer, NULL, 0);
1883
1884         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1885                 xprt_destroy(xprt);
1886                 return ERR_PTR(-EINVAL);
1887         }
1888         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1889         if (xprt->servername == NULL) {
1890                 xprt_destroy(xprt);
1891                 return ERR_PTR(-ENOMEM);
1892         }
1893
1894         rpc_xprt_debugfs_register(xprt);
1895
1896         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1897                         xprt->max_reqs);
1898 out:
1899         return xprt;
1900 }
1901
1902 static void xprt_destroy_cb(struct work_struct *work)
1903 {
1904         struct rpc_xprt *xprt =
1905                 container_of(work, struct rpc_xprt, task_cleanup);
1906
1907         rpc_xprt_debugfs_unregister(xprt);
1908         rpc_destroy_wait_queue(&xprt->binding);
1909         rpc_destroy_wait_queue(&xprt->pending);
1910         rpc_destroy_wait_queue(&xprt->sending);
1911         rpc_destroy_wait_queue(&xprt->backlog);
1912         kfree(xprt->servername);
1913         /*
1914          * Tear down transport state and free the rpc_xprt
1915          */
1916         xprt->ops->destroy(xprt);
1917 }
1918
1919 /**
1920  * xprt_destroy - destroy an RPC transport, killing off all requests.
1921  * @xprt: transport to destroy
1922  *
1923  */
1924 static void xprt_destroy(struct rpc_xprt *xprt)
1925 {
1926         dprintk("RPC:       destroying transport %p\n", xprt);
1927
1928         /*
1929          * Exclude transport connect/disconnect handlers and autoclose
1930          */
1931         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1932
1933         del_timer_sync(&xprt->timer);
1934
1935         /*
1936          * Destroy sockets etc from the system workqueue so they can
1937          * safely flush receive work running on rpciod.
1938          */
1939         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1940         schedule_work(&xprt->task_cleanup);
1941 }
1942
1943 static void xprt_destroy_kref(struct kref *kref)
1944 {
1945         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1946 }
1947
1948 /**
1949  * xprt_get - return a reference to an RPC transport.
1950  * @xprt: pointer to the transport
1951  *
1952  */
1953 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1954 {
1955         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1956                 return xprt;
1957         return NULL;
1958 }
1959 EXPORT_SYMBOL_GPL(xprt_get);
1960
1961 /**
1962  * xprt_put - release a reference to an RPC transport.
1963  * @xprt: pointer to the transport
1964  *
1965  */
1966 void xprt_put(struct rpc_xprt *xprt)
1967 {
1968         if (xprt != NULL)
1969                 kref_put(&xprt->kref, xprt_destroy_kref);
1970 }
1971 EXPORT_SYMBOL_GPL(xprt_put);