net/sunrpc/xprt.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/net/sunrpc/xprt.c
   4  *
   5  *  This is a generic RPC call interface supporting congestion avoidance,
   6  *  and asynchronous calls.
   7  *
   8  *  The interface works like this:
   9  *
  10  *  -   When a process places a call, it allocates a request slot if
  11  *      one is available. Otherwise, it sleeps on the backlog queue
  12  *      (xprt_reserve).
  13  *  -   Next, the caller puts together the RPC message, stuffs it into
  14  *      the request struct, and calls xprt_transmit().
  15  *  -   xprt_transmit sends the message and installs the caller on the
  16  *      transport's wait list. At the same time, if a reply is expected,
  17  *      it installs a timer that is run after the packet's timeout has
  18  *      expired.
  19  *  -   When a packet arrives, the data_ready handler walks the list of
  20  *      pending requests for that transport. If a matching XID is found, the
  21  *      caller is woken up, and the timer removed.
  22  *  -   When no reply arrives within the timeout interval, the timer is
  23  *      fired by the kernel and runs xprt_timer(). It either adjusts the
  24  *      timeout values (minor timeout) or wakes up the caller with a status
  25  *      of -ETIMEDOUT.
  26  *  -   When the caller receives a notification from RPC that a reply arrived,
  27  *      it should release the RPC slot, and process the reply.
  28  *      If the call timed out, it may choose to retry the operation by
  29  *      adjusting the initial timeout value, and simply calling rpc_call
  30  *      again.
  31  *
  32  *  Support for async RPC is done through a set of RPC-specific scheduling
  33  *  primitives that `transparently' work for processes as well as async
  34  *  tasks that rely on callbacks.
  35  *
  36  *  Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
  37  *
  38  *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
  39  */
  40
  41 #include <linux/module.h>
  42
  43 #include <linux/types.h>
  44 #include <linux/interrupt.h>
  45 #include <linux/workqueue.h>
  46 #include <linux/net.h>
  47 #include <linux/ktime.h>
  48
  49 #include <linux/sunrpc/clnt.h>
  50 #include <linux/sunrpc/metrics.h>
  51 #include <linux/sunrpc/bc_xprt.h>
  52 #include <linux/rcupdate.h>
  53 #include <linux/sched/mm.h>
  54
  55 #include <trace/events/sunrpc.h>
  56
  57 #include "sunrpc.h"
  58
  59 /*
  60  * Local variables
  61  */
  62
  63 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  64 # define RPCDBG_FACILITY        RPCDBG_XPRT
  65 #endif
  66
  67 /*
  68  * Local functions
  69  */
  70 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
  71 static __be32   xprt_alloc_xid(struct rpc_xprt *xprt);
  72 static void      xprt_destroy(struct rpc_xprt *xprt);
  73
  74 static DEFINE_SPINLOCK(xprt_list_lock);
  75 static LIST_HEAD(xprt_list);
  76
  77 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
  78 {
  79         unsigned long timeout = jiffies + req->rq_timeout;
  80
  81         if (time_before(timeout, req->rq_majortimeo))
  82                 return timeout;
  83         return req->rq_majortimeo;
  84 }
  85
  86 /**
  87  * xprt_register_transport - register a transport implementation
  88  * @transport: transport to register
  89  *
  90  * If a transport implementation is loaded as a kernel module, it can
  91  * call this interface to make itself known to the RPC client.
  92  *
  93  * Returns:
  94  * 0:           transport successfully registered
  95  * -EEXIST:     transport already registered
  96  * -EINVAL:     transport module being unloaded
  97  */
  98 int xprt_register_transport(struct xprt_class *transport)
  99 {
 100         struct xprt_class *t;
 101         int result;
 102
 103         result = -EEXIST;
 104         spin_lock(&xprt_list_lock);
 105         list_for_each_entry(t, &xprt_list, list) {
 106                 /* don't register the same transport class twice */
 107                 if (t->ident == transport->ident)
 108                         goto out;
 109         }
 110
 111         list_add_tail(&transport->list, &xprt_list);
 112         printk(KERN_INFO "RPC: Registered %s transport module.\n",
 113                transport->name);
 114         result = 0;
 115
 116 out:
 117         spin_unlock(&xprt_list_lock);
 118         return result;
 119 }
 120 EXPORT_SYMBOL_GPL(xprt_register_transport);
 121
 122 /**
 123  * xprt_unregister_transport - unregister a transport implementation
 124  * @transport: transport to unregister
 125  *
 126  * Returns:
 127  * 0:           transport successfully unregistered
 128  * -ENOENT:     transport never registered
 129  */
 130 int xprt_unregister_transport(struct xprt_class *transport)
 131 {
 132         struct xprt_class *t;
 133         int result;
 134
 135         result = 0;
 136         spin_lock(&xprt_list_lock);
 137         list_for_each_entry(t, &xprt_list, list) {
 138                 if (t == transport) {
 139                         printk(KERN_INFO
 140                                 "RPC: Unregistered %s transport module.\n",
 141                                 transport->name);
 142                         list_del_init(&transport->list);
 143                         goto out;
 144                 }
 145         }
 146         result = -ENOENT;
 147
 148 out:
 149         spin_unlock(&xprt_list_lock);
 150         return result;
 151 }
 152 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
 153
 154 /**
 155  * xprt_load_transport - load a transport implementation
 156  * @transport_name: transport to load
 157  *
 158  * Returns:
 159  * 0:           transport successfully loaded
 160  * -ENOENT:     transport module not available
 161  */
 162 int xprt_load_transport(const char *transport_name)
 163 {
 164         struct xprt_class *t;
 165         int result;
 166
 167         result = 0;
 168         spin_lock(&xprt_list_lock);
 169         list_for_each_entry(t, &xprt_list, list) {
 170                 if (strcmp(t->name, transport_name) == 0) {
 171                         spin_unlock(&xprt_list_lock);
 172                         goto out;
 173                 }
 174         }
 175         spin_unlock(&xprt_list_lock);
 176         result = request_module("xprt%s", transport_name);
 177 out:
 178         return result;
 179 }
 180 EXPORT_SYMBOL_GPL(xprt_load_transport);
 181
 182 static void xprt_clear_locked(struct rpc_xprt *xprt)
 183 {
 184         xprt->snd_task = NULL;
 185         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
 186                 smp_mb__before_atomic();
 187                 clear_bit(XPRT_LOCKED, &xprt->state);
 188                 smp_mb__after_atomic();
 189         } else
 190                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 191 }
 192
 193 /**
 194  * xprt_reserve_xprt - serialize write access to transports
 195  * @task: task that is requesting access to the transport
 196  * @xprt: pointer to the target transport
 197  *
 198  * This prevents mixing the payload of separate requests, and prevents
 199  * transport connects from colliding with writes.  No congestion control
 200  * is provided.
 201  */
 202 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 203 {
 204         struct rpc_rqst *req = task->tk_rqstp;
 205
 206         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 207                 if (task == xprt->snd_task)
 208                         goto out_locked;
 209                 goto out_sleep;
 210         }
 211         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 212                 goto out_unlock;
 213         xprt->snd_task = task;
 214
 215 out_locked:
 216         trace_xprt_reserve_xprt(xprt, task);
 217         return 1;
 218
 219 out_unlock:
 220         xprt_clear_locked(xprt);
 221 out_sleep:
 222         task->tk_status = -EAGAIN;
 223         if  (RPC_IS_SOFT(task))
 224                 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
 225                                 xprt_request_timeout(req));
 226         else
 227                 rpc_sleep_on(&xprt->sending, task, NULL);
 228         return 0;
 229 }
 230 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 231
 232 static bool
 233 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
 234 {
 235         return test_bit(XPRT_CWND_WAIT, &xprt->state);
 236 }
 237
 238 static void
 239 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
 240 {
 241         if (!list_empty(&xprt->xmit_queue)) {
 242                 /* Peek at head of queue to see if it can make progress */
 243                 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
 244                                         rq_xmit)->rq_cong)
 245                         return;
 246         }
 247         set_bit(XPRT_CWND_WAIT, &xprt->state);
 248 }
 249
 250 static void
 251 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
 252 {
 253         if (!RPCXPRT_CONGESTED(xprt))
 254                 clear_bit(XPRT_CWND_WAIT, &xprt->state);
 255 }
 256
 257 /*
 258  * xprt_reserve_xprt_cong - serialize write access to transports
 259  * @task: task that is requesting access to the transport
 260  *
 261  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
 262  * integrated into the decision of whether a request is allowed to be
 263  * woken up and given access to the transport.
 264  * Note that the lock is only granted if we know there are free slots.
 265  */
 266 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 267 {
 268         struct rpc_rqst *req = task->tk_rqstp;
 269
 270         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 271                 if (task == xprt->snd_task)
 272                         goto out_locked;
 273                 goto out_sleep;
 274         }
 275         if (req == NULL) {
 276                 xprt->snd_task = task;
 277                 goto out_locked;
 278         }
 279         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 280                 goto out_unlock;
 281         if (!xprt_need_congestion_window_wait(xprt)) {
 282                 xprt->snd_task = task;
 283                 goto out_locked;
 284         }
 285 out_unlock:
 286         xprt_clear_locked(xprt);
 287 out_sleep:
 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 out_locked:
 296         trace_xprt_reserve_cong(xprt, task);
 297         return 1;
 298 }
 299 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
 300
 301 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 302 {
 303         int retval;
 304
 305         if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
 306                 return 1;
 307         spin_lock(&xprt->transport_lock);
 308         retval = xprt->ops->reserve_xprt(xprt, task);
 309         spin_unlock(&xprt->transport_lock);
 310         return retval;
 311 }
 312
 313 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
 314 {
 315         struct rpc_xprt *xprt = data;
 316
 317         xprt->snd_task = task;
 318         return true;
 319 }
 320
 321 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 322 {
 323         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 324                 return;
 325         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 326                 goto out_unlock;
 327         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 328                                 __xprt_lock_write_func, xprt))
 329                 return;
 330 out_unlock:
 331         xprt_clear_locked(xprt);
 332 }
 333
 334 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 335 {
 336         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 337                 return;
 338         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
 339                 goto out_unlock;
 340         if (xprt_need_congestion_window_wait(xprt))
 341                 goto out_unlock;
 342         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 343                                 __xprt_lock_write_func, xprt))
 344                 return;
 345 out_unlock:
 346         xprt_clear_locked(xprt);
 347 }
 348
 349 /**
 350  * xprt_release_xprt - allow other requests to use a transport
 351  * @xprt: transport with other tasks potentially waiting
 352  * @task: task that is releasing access to the transport
 353  *
 354  * Note that "task" can be NULL.  No congestion control is provided.
 355  */
 356 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 357 {
 358         if (xprt->snd_task == task) {
 359                 xprt_clear_locked(xprt);
 360                 __xprt_lock_write_next(xprt);
 361         }
 362         trace_xprt_release_xprt(xprt, task);
 363 }
 364 EXPORT_SYMBOL_GPL(xprt_release_xprt);
 365
 366 /**
 367  * xprt_release_xprt_cong - allow other requests to use a transport
 368  * @xprt: transport with other tasks potentially waiting
 369  * @task: task that is releasing access to the transport
 370  *
 371  * Note that "task" can be NULL.  Another task is awoken to use the
 372  * transport if the transport's congestion window allows it.
 373  */
 374 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 375 {
 376         if (xprt->snd_task == task) {
 377                 xprt_clear_locked(xprt);
 378                 __xprt_lock_write_next_cong(xprt);
 379         }
 380         trace_xprt_release_cong(xprt, task);
 381 }
 382 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
 383
 384 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 385 {
 386         if (xprt->snd_task != task)
 387                 return;
 388         spin_lock(&xprt->transport_lock);
 389         xprt->ops->release_xprt(xprt, task);
 390         spin_unlock(&xprt->transport_lock);
 391 }
 392
 393 /*
 394  * Van Jacobson congestion avoidance. Check if the congestion window
 395  * overflowed. Put the task to sleep if this is the case.
 396  */
 397 static int
 398 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 399 {
 400         if (req->rq_cong)
 401                 return 1;
 402         trace_xprt_get_cong(xprt, req->rq_task);
 403         if (RPCXPRT_CONGESTED(xprt)) {
 404                 xprt_set_congestion_window_wait(xprt);
 405                 return 0;
 406         }
 407         req->rq_cong = 1;
 408         xprt->cong += RPC_CWNDSCALE;
 409         return 1;
 410 }
 411
 412 /*
 413  * Adjust the congestion window, and wake up the next task
 414  * that has been sleeping due to congestion
 415  */
 416 static void
 417 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 418 {
 419         if (!req->rq_cong)
 420                 return;
 421         req->rq_cong = 0;
 422         xprt->cong -= RPC_CWNDSCALE;
 423         xprt_test_and_clear_congestion_window_wait(xprt);
 424         trace_xprt_put_cong(xprt, req->rq_task);
 425         __xprt_lock_write_next_cong(xprt);
 426 }
 427
 428 /**
 429  * xprt_request_get_cong - Request congestion control credits
 430  * @xprt: pointer to transport
 431  * @req: pointer to RPC request
 432  *
 433  * Useful for transports that require congestion control.
 434  */
 435 bool
 436 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 437 {
 438         bool ret = false;
 439
 440         if (req->rq_cong)
 441                 return true;
 442         spin_lock(&xprt->transport_lock);
 443         ret = __xprt_get_cong(xprt, req) != 0;
 444         spin_unlock(&xprt->transport_lock);
 445         return ret;
 446 }
 447 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
 448
 449 /**
 450  * xprt_release_rqst_cong - housekeeping when request is complete
 451  * @task: RPC request that recently completed
 452  *
 453  * Useful for transports that require congestion control.
 454  */
 455 void xprt_release_rqst_cong(struct rpc_task *task)
 456 {
 457         struct rpc_rqst *req = task->tk_rqstp;
 458
 459         __xprt_put_cong(req->rq_xprt, req);
 460 }
 461 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
 462
 463 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
 464 {
 465         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
 466                 __xprt_lock_write_next_cong(xprt);
 467 }
 468
 469 /*
 470  * Clear the congestion window wait flag and wake up the next
 471  * entry on xprt->sending
 472  */
 473 static void
 474 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
 475 {
 476         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
 477                 spin_lock(&xprt->transport_lock);
 478                 __xprt_lock_write_next_cong(xprt);
 479                 spin_unlock(&xprt->transport_lock);
 480         }
 481 }
 482
 483 /**
 484  * xprt_adjust_cwnd - adjust transport congestion window
 485  * @xprt: pointer to xprt
 486  * @task: recently completed RPC request used to adjust window
 487  * @result: result code of completed RPC request
 488  *
 489  * The transport code maintains an estimate on the maximum number of out-
 490  * standing RPC requests, using a smoothed version of the congestion
 491  * avoidance implemented in 44BSD. This is basically the Van Jacobson
 492  * congestion algorithm: If a retransmit occurs, the congestion window is
 493  * halved; otherwise, it is incremented by 1/cwnd when
 494  *
 495  *      -       a reply is received and
 496  *      -       a full number of requests are outstanding and
 497  *      -       the congestion window hasn't been updated recently.
 498  */
 499 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
 500 {
 501         struct rpc_rqst *req = task->tk_rqstp;
 502         unsigned long cwnd = xprt->cwnd;
 503
 504         if (result >= 0 && cwnd <= xprt->cong) {
 505                 /* The (cwnd >> 1) term makes sure
 506                  * the result gets rounded properly. */
 507                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 508                 if (cwnd > RPC_MAXCWND(xprt))
 509                         cwnd = RPC_MAXCWND(xprt);
 510                 __xprt_lock_write_next_cong(xprt);
 511         } else if (result == -ETIMEDOUT) {
 512                 cwnd >>= 1;
 513                 if (cwnd < RPC_CWNDSCALE)
 514                         cwnd = RPC_CWNDSCALE;
 515         }
 516         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
 517                         xprt->cong, xprt->cwnd, cwnd);
 518         xprt->cwnd = cwnd;
 519         __xprt_put_cong(xprt, req);
 520 }
 521 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
 522
 523 /**
 524  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
 525  * @xprt: transport with waiting tasks
 526  * @status: result code to plant in each task before waking it
 527  *
 528  */
 529 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 530 {
 531         if (status < 0)
 532                 rpc_wake_up_status(&xprt->pending, status);
 533         else
 534                 rpc_wake_up(&xprt->pending);
 535 }
 536 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 537
 538 /**
 539  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
 540  * @xprt: transport
 541  *
 542  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
 543  * we don't in general want to force a socket disconnection due to
 544  * an incomplete RPC call transmission.
 545  */
 546 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
 547 {
 548         set_bit(XPRT_WRITE_SPACE, &xprt->state);
 549 }
 550 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 551
 552 static bool
 553 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
 554 {
 555         if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
 556                 __xprt_lock_write_next(xprt);
 557                 dprintk("RPC:       write space: waking waiting task on "
 558                                 "xprt %p\n", xprt);
 559                 return true;
 560         }
 561         return false;
 562 }
 563
 564 /**
 565  * xprt_write_space - wake the task waiting for transport output buffer space
 566  * @xprt: transport with waiting tasks
 567  *
 568  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
 569  */
 570 bool xprt_write_space(struct rpc_xprt *xprt)
 571 {
 572         bool ret;
 573
 574         if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
 575                 return false;
 576         spin_lock(&xprt->transport_lock);
 577         ret = xprt_clear_write_space_locked(xprt);
 578         spin_unlock(&xprt->transport_lock);
 579         return ret;
 580 }
 581 EXPORT_SYMBOL_GPL(xprt_write_space);
 582
 583 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
 584 {
 585         s64 delta = ktime_to_ns(ktime_get() - abstime);
 586         return likely(delta >= 0) ?
 587                 jiffies - nsecs_to_jiffies(delta) :
 588                 jiffies + nsecs_to_jiffies(-delta);
 589 }
 590
 591 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
 592 {
 593         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 594         unsigned long majortimeo = req->rq_timeout;
 595
 596         if (to->to_exponential)
 597                 majortimeo <<= to->to_retries;
 598         else
 599                 majortimeo += to->to_increment * to->to_retries;
 600         if (majortimeo > to->to_maxval || majortimeo == 0)
 601                 majortimeo = to->to_maxval;
 602         return majortimeo;
 603 }
 604
 605 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 606 {
 607         req->rq_majortimeo += xprt_calc_majortimeo(req);
 608 }
 609
 610 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
 611 {
 612         unsigned long time_init;
 613         struct rpc_xprt *xprt = req->rq_xprt;
 614
 615         if (likely(xprt && xprt_connected(xprt)))
 616                 time_init = jiffies;
 617         else
 618                 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
 619         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
 620         req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
 621 }
 622
 623 /**
 624  * xprt_adjust_timeout - adjust timeout values for next retransmit
 625  * @req: RPC request containing parameters to use for the adjustment
 626  *
 627  */
 628 int xprt_adjust_timeout(struct rpc_rqst *req)
 629 {
 630         struct rpc_xprt *xprt = req->rq_xprt;
 631         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
 632         int status = 0;
 633
 634         if (time_before(jiffies, req->rq_majortimeo)) {
 635                 if (to->to_exponential)
 636                         req->rq_timeout <<= 1;
 637                 else
 638                         req->rq_timeout += to->to_increment;
 639                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
 640                         req->rq_timeout = to->to_maxval;
 641                 req->rq_retries++;
 642         } else {
 643                 req->rq_timeout = to->to_initval;
 644                 req->rq_retries = 0;
 645                 xprt_reset_majortimeo(req);
 646                 /* Reset the RTT counters == "slow start" */
 647                 spin_lock(&xprt->transport_lock);
 648                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
 649                 spin_unlock(&xprt->transport_lock);
 650                 status = -ETIMEDOUT;
 651         }
 652
 653         if (req->rq_timeout == 0) {
 654                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
 655                 req->rq_timeout = 5 * HZ;
 656         }
 657         return status;
 658 }
 659
 660 static void xprt_autoclose(struct work_struct *work)
 661 {
 662         struct rpc_xprt *xprt =
 663                 container_of(work, struct rpc_xprt, task_cleanup);
 664         unsigned int pflags = memalloc_nofs_save();
 665
 666         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 667         xprt->ops->close(xprt);
 668         xprt_release_write(xprt, NULL);
 669         wake_up_bit(&xprt->state, XPRT_LOCKED);
 670         memalloc_nofs_restore(pflags);
 671 }
 672
 673 /**
 674  * xprt_disconnect_done - mark a transport as disconnected
 675  * @xprt: transport to flag for disconnect
 676  *
 677  */
 678 void xprt_disconnect_done(struct rpc_xprt *xprt)
 679 {
 680         dprintk("RPC:       disconnected transport %p\n", xprt);
 681         spin_lock(&xprt->transport_lock);
 682         xprt_clear_connected(xprt);
 683         xprt_clear_write_space_locked(xprt);
 684         xprt_clear_congestion_window_wait_locked(xprt);
 685         xprt_wake_pending_tasks(xprt, -ENOTCONN);
 686         spin_unlock(&xprt->transport_lock);
 687 }
 688 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
 689
 690 /**
 691  * xprt_force_disconnect - force a transport to disconnect
 692  * @xprt: transport to disconnect
 693  *
 694  */
 695 void xprt_force_disconnect(struct rpc_xprt *xprt)
 696 {
 697         /* Don't race with the test_bit() in xprt_clear_locked() */
 698         spin_lock(&xprt->transport_lock);
 699         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 700         /* Try to schedule an autoclose RPC call */
 701         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 702                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 703         else if (xprt->snd_task)
 704                 rpc_wake_up_queued_task_set_status(&xprt->pending,
 705                                 xprt->snd_task, -ENOTCONN);
 706         spin_unlock(&xprt->transport_lock);
 707 }
 708 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
 709
 710 static unsigned int
 711 xprt_connect_cookie(struct rpc_xprt *xprt)
 712 {
 713         return READ_ONCE(xprt->connect_cookie);
 714 }
 715
 716 static bool
 717 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
 718 {
 719         struct rpc_rqst *req = task->tk_rqstp;
 720         struct rpc_xprt *xprt = req->rq_xprt;
 721
 722         return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
 723                 !xprt_connected(xprt);
 724 }
 725
 726 /**
 727  * xprt_conditional_disconnect - force a transport to disconnect
 728  * @xprt: transport to disconnect
 729  * @cookie: 'connection cookie'
 730  *
 731  * This attempts to break the connection if and only if 'cookie' matches
 732  * the current transport 'connection cookie'. It ensures that we don't
 733  * try to break the connection more than once when we need to retransmit
 734  * a batch of RPC requests.
 735  *
 736  */
 737 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
 738 {
 739         /* Don't race with the test_bit() in xprt_clear_locked() */
 740         spin_lock(&xprt->transport_lock);
 741         if (cookie != xprt->connect_cookie)
 742                 goto out;
 743         if (test_bit(XPRT_CLOSING, &xprt->state))
 744                 goto out;
 745         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 746         /* Try to schedule an autoclose RPC call */
 747         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
 748                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 749         xprt_wake_pending_tasks(xprt, -EAGAIN);
 750 out:
 751         spin_unlock(&xprt->transport_lock);
 752 }
 753
 754 static bool
 755 xprt_has_timer(const struct rpc_xprt *xprt)
 756 {
 757         return xprt->idle_timeout != 0;
 758 }
 759
 760 static void
 761 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
 762         __must_hold(&xprt->transport_lock)
 763 {
 764         xprt->last_used = jiffies;
 765         if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
 766                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
 767 }
 768
 769 static void
 770 xprt_init_autodisconnect(struct timer_list *t)
 771 {
 772         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
 773
 774         if (!RB_EMPTY_ROOT(&xprt->recv_queue))
 775                 return;
 776         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
 777         xprt->last_used = jiffies;
 778         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 779                 return;
 780         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
 781 }
 782
 783 bool xprt_lock_connect(struct rpc_xprt *xprt,
 784                 struct rpc_task *task,
 785                 void *cookie)
 786 {
 787         bool ret = false;
 788
 789         spin_lock(&xprt->transport_lock);
 790         if (!test_bit(XPRT_LOCKED, &xprt->state))
 791                 goto out;
 792         if (xprt->snd_task != task)
 793                 goto out;
 794         xprt->snd_task = cookie;
 795         ret = true;
 796 out:
 797         spin_unlock(&xprt->transport_lock);
 798         return ret;
 799 }
 800
 801 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
 802 {
 803         spin_lock(&xprt->transport_lock);
 804         if (xprt->snd_task != cookie)
 805                 goto out;
 806         if (!test_bit(XPRT_LOCKED, &xprt->state))
 807                 goto out;
 808         xprt->snd_task =NULL;
 809         xprt->ops->release_xprt(xprt, NULL);
 810         xprt_schedule_autodisconnect(xprt);
 811 out:
 812         spin_unlock(&xprt->transport_lock);
 813         wake_up_bit(&xprt->state, XPRT_LOCKED);
 814 }
 815
 816 /**
 817  * xprt_connect - schedule a transport connect operation
 818  * @task: RPC task that is requesting the connect
 819  *
 820  */
 821 void xprt_connect(struct rpc_task *task)
 822 {
 823         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
 824
 825         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
 826                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
 827
 828         if (!xprt_bound(xprt)) {
 829                 task->tk_status = -EAGAIN;
 830                 return;
 831         }
 832         if (!xprt_lock_write(xprt, task))
 833                 return;
 834
 835         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
 836                 xprt->ops->close(xprt);
 837
 838         if (!xprt_connected(xprt)) {
 839                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
 840                 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
 841                                 xprt_request_timeout(task->tk_rqstp));
 842
 843                 if (test_bit(XPRT_CLOSING, &xprt->state))
 844                         return;
 845                 if (xprt_test_and_set_connecting(xprt))
 846                         return;
 847                 /* Race breaker */
 848                 if (!xprt_connected(xprt)) {
 849                         xprt->stat.connect_start = jiffies;
 850                         xprt->ops->connect(xprt, task);
 851                 } else {
 852                         xprt_clear_connecting(xprt);
 853                         task->tk_status = 0;
 854                         rpc_wake_up_queued_task(&xprt->pending, task);
 855                 }
 856         }
 857         xprt_release_write(xprt, task);
 858 }
 859
 860 /**
 861  * xprt_reconnect_delay - compute the wait before scheduling a connect
 862  * @xprt: transport instance
 863  *
 864  */
 865 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
 866 {
 867         unsigned long start, now = jiffies;
 868
 869         start = xprt->stat.connect_start + xprt->reestablish_timeout;
 870         if (time_after(start, now))
 871                 return start - now;
 872         return 0;
 873 }
 874 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
 875
 876 /**
 877  * xprt_reconnect_backoff - compute the new re-establish timeout
 878  * @xprt: transport instance
 879  * @init_to: initial reestablish timeout
 880  *
 881  */
 882 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
 883 {
 884         xprt->reestablish_timeout <<= 1;
 885         if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
 886                 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
 887         if (xprt->reestablish_timeout < init_to)
 888                 xprt->reestablish_timeout = init_to;
 889 }
 890 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
 891
 892 enum xprt_xid_rb_cmp {
 893         XID_RB_EQUAL,
 894         XID_RB_LEFT,
 895         XID_RB_RIGHT,
 896 };
 897 static enum xprt_xid_rb_cmp
 898 xprt_xid_cmp(__be32 xid1, __be32 xid2)
 899 {
 900         if (xid1 == xid2)
 901                 return XID_RB_EQUAL;
 902         if ((__force u32)xid1 < (__force u32)xid2)
 903                 return XID_RB_LEFT;
 904         return XID_RB_RIGHT;
 905 }
 906
 907 static struct rpc_rqst *
 908 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
 909 {
 910         struct rb_node *n = xprt->recv_queue.rb_node;
 911         struct rpc_rqst *req;
 912
 913         while (n != NULL) {
 914                 req = rb_entry(n, struct rpc_rqst, rq_recv);
 915                 switch (xprt_xid_cmp(xid, req->rq_xid)) {
 916                 case XID_RB_LEFT:
 917                         n = n->rb_left;
 918                         break;
 919                 case XID_RB_RIGHT:
 920                         n = n->rb_right;
 921                         break;
 922                 case XID_RB_EQUAL:
 923                         return req;
 924                 }
 925         }
 926         return NULL;
 927 }
 928
 929 static void
 930 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
 931 {
 932         struct rb_node **p = &xprt->recv_queue.rb_node;
 933         struct rb_node *n = NULL;
 934         struct rpc_rqst *req;
 935
 936         while (*p != NULL) {
 937                 n = *p;
 938                 req = rb_entry(n, struct rpc_rqst, rq_recv);
 939                 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
 940                 case XID_RB_LEFT:
 941                         p = &n->rb_left;
 942                         break;
 943                 case XID_RB_RIGHT:
 944                         p = &n->rb_right;
 945                         break;
 946                 case XID_RB_EQUAL:
 947                         WARN_ON_ONCE(new != req);
 948                         return;
 949                 }
 950         }
 951         rb_link_node(&new->rq_recv, n, p);
 952         rb_insert_color(&new->rq_recv, &xprt->recv_queue);
 953 }
 954
 955 static void
 956 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
 957 {
 958         rb_erase(&req->rq_recv, &xprt->recv_queue);
 959 }
 960
 961 /**
 962  * xprt_lookup_rqst - find an RPC request corresponding to an XID
 963  * @xprt: transport on which the original request was transmitted
 964  * @xid: RPC XID of incoming reply
 965  *
 966  * Caller holds xprt->queue_lock.
 967  */
 968 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
 969 {
 970         struct rpc_rqst *entry;
 971
 972         entry = xprt_request_rb_find(xprt, xid);
 973         if (entry != NULL) {
 974                 trace_xprt_lookup_rqst(xprt, xid, 0);
 975                 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
 976                 return entry;
 977         }
 978
 979         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
 980                         ntohl(xid));
 981         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
 982         xprt->stat.bad_xids++;
 983         return NULL;
 984 }
 985 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
 986
 987 static bool
 988 xprt_is_pinned_rqst(struct rpc_rqst *req)
 989 {
 990         return atomic_read(&req->rq_pin) != 0;
 991 }
 992
 993 /**
 994  * xprt_pin_rqst - Pin a request on the transport receive list
 995  * @req: Request to pin
 996  *
 997  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
 998  * so should be holding xprt->queue_lock.
 999  */
1000 void xprt_pin_rqst(struct rpc_rqst *req)
1001 {
1002         atomic_inc(&req->rq_pin);
1003 }
1004 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1005
1006 /**
1007  * xprt_unpin_rqst - Unpin a request on the transport receive list
1008  * @req: Request to pin
1009  *
1010  * Caller should be holding xprt->queue_lock.
1011  */
1012 void xprt_unpin_rqst(struct rpc_rqst *req)
1013 {
1014         if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1015                 atomic_dec(&req->rq_pin);
1016                 return;
1017         }
1018         if (atomic_dec_and_test(&req->rq_pin))
1019                 wake_up_var(&req->rq_pin);
1020 }
1021 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1022
1023 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1024 {
1025         wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1026 }
1027
1028 static bool
1029 xprt_request_data_received(struct rpc_task *task)
1030 {
1031         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1032                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1033 }
1034
1035 static bool
1036 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1037 {
1038         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1039                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1040 }
1041
1042 /**
1043  * xprt_request_enqueue_receive - Add an request to the receive queue
1044  * @task: RPC task
1045  *
1046  */
1047 void
1048 xprt_request_enqueue_receive(struct rpc_task *task)
1049 {
1050         struct rpc_rqst *req = task->tk_rqstp;
1051         struct rpc_xprt *xprt = req->rq_xprt;
1052
1053         if (!xprt_request_need_enqueue_receive(task, req))
1054                 return;
1055
1056         xprt_request_prepare(task->tk_rqstp);
1057         spin_lock(&xprt->queue_lock);
1058
1059         /* Update the softirq receive buffer */
1060         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1061                         sizeof(req->rq_private_buf));
1062
1063         /* Add request to the receive list */
1064         xprt_request_rb_insert(xprt, req);
1065         set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1066         spin_unlock(&xprt->queue_lock);
1067
1068         /* Turn off autodisconnect */
1069         del_singleshot_timer_sync(&xprt->timer);
1070 }
1071
1072 /**
1073  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1074  * @task: RPC task
1075  *
1076  * Caller must hold xprt->queue_lock.
1077  */
1078 static void
1079 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1080 {
1081         struct rpc_rqst *req = task->tk_rqstp;
1082
1083         if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1084                 xprt_request_rb_remove(req->rq_xprt, req);
1085 }
1086
1087 /**
1088  * xprt_update_rtt - Update RPC RTT statistics
1089  * @task: RPC request that recently completed
1090  *
1091  * Caller holds xprt->queue_lock.
1092  */
1093 void xprt_update_rtt(struct rpc_task *task)
1094 {
1095         struct rpc_rqst *req = task->tk_rqstp;
1096         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1097         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1098         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1099
1100         if (timer) {
1101                 if (req->rq_ntrans == 1)
1102                         rpc_update_rtt(rtt, timer, m);
1103                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1104         }
1105 }
1106 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1107
1108 /**
1109  * xprt_complete_rqst - called when reply processing is complete
1110  * @task: RPC request that recently completed
1111  * @copied: actual number of bytes received from the transport
1112  *
1113  * Caller holds xprt->queue_lock.
1114  */
1115 void xprt_complete_rqst(struct rpc_task *task, int copied)
1116 {
1117         struct rpc_rqst *req = task->tk_rqstp;
1118         struct rpc_xprt *xprt = req->rq_xprt;
1119
1120         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
1121                         task->tk_pid, ntohl(req->rq_xid), copied);
1122         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
1123
1124         xprt->stat.recvs++;
1125
1126         req->rq_private_buf.len = copied;
1127         /* Ensure all writes are done before we update */
1128         /* req->rq_reply_bytes_recvd */
1129         smp_wmb();
1130         req->rq_reply_bytes_recvd = copied;
1131         xprt_request_dequeue_receive_locked(task);
1132         rpc_wake_up_queued_task(&xprt->pending, task);
1133 }
1134 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1135
1136 static void xprt_timer(struct rpc_task *task)
1137 {
1138         struct rpc_rqst *req = task->tk_rqstp;
1139         struct rpc_xprt *xprt = req->rq_xprt;
1140
1141         if (task->tk_status != -ETIMEDOUT)
1142                 return;
1143
1144         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1145         if (!req->rq_reply_bytes_recvd) {
1146                 if (xprt->ops->timer)
1147                         xprt->ops->timer(xprt, task);
1148         } else
1149                 task->tk_status = 0;
1150 }
1151
1152 /**
1153  * xprt_wait_for_reply_request_def - wait for reply
1154  * @task: pointer to rpc_task
1155  *
1156  * Set a request's retransmit timeout based on the transport's
1157  * default timeout parameters.  Used by transports that don't adjust
1158  * the retransmit timeout based on round-trip time estimation,
1159  * and put the task to sleep on the pending queue.
1160  */
1161 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1162 {
1163         struct rpc_rqst *req = task->tk_rqstp;
1164
1165         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1166                         xprt_request_timeout(req));
1167 }
1168 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1169
1170 /**
1171  * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1172  * @task: pointer to rpc_task
1173  *
1174  * Set a request's retransmit timeout using the RTT estimator,
1175  * and put the task to sleep on the pending queue.
1176  */
1177 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1178 {
1179         int timer = task->tk_msg.rpc_proc->p_timer;
1180         struct rpc_clnt *clnt = task->tk_client;
1181         struct rpc_rtt *rtt = clnt->cl_rtt;
1182         struct rpc_rqst *req = task->tk_rqstp;
1183         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1184         unsigned long timeout;
1185
1186         timeout = rpc_calc_rto(rtt, timer);
1187         timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1188         if (timeout > max_timeout || timeout == 0)
1189                 timeout = max_timeout;
1190         rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1191                         jiffies + timeout);
1192 }
1193 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1194
1195 /**
1196  * xprt_request_wait_receive - wait for the reply to an RPC request
1197  * @task: RPC task about to send a request
1198  *
1199  */
1200 void xprt_request_wait_receive(struct rpc_task *task)
1201 {
1202         struct rpc_rqst *req = task->tk_rqstp;
1203         struct rpc_xprt *xprt = req->rq_xprt;
1204
1205         if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1206                 return;
1207         /*
1208          * Sleep on the pending queue if we're expecting a reply.
1209          * The spinlock ensures atomicity between the test of
1210          * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1211          */
1212         spin_lock(&xprt->queue_lock);
1213         if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1214                 xprt->ops->wait_for_reply_request(task);
1215                 /*
1216                  * Send an extra queue wakeup call if the
1217                  * connection was dropped in case the call to
1218                  * rpc_sleep_on() raced.
1219                  */
1220                 if (xprt_request_retransmit_after_disconnect(task))
1221                         rpc_wake_up_queued_task_set_status(&xprt->pending,
1222                                         task, -ENOTCONN);
1223         }
1224         spin_unlock(&xprt->queue_lock);
1225 }
1226
1227 static bool
1228 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1229 {
1230         return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1231 }
1232
1233 /**
1234  * xprt_request_enqueue_transmit - queue a task for transmission
1235  * @task: pointer to rpc_task
1236  *
1237  * Add a task to the transmission queue.
1238  */
1239 void
1240 xprt_request_enqueue_transmit(struct rpc_task *task)
1241 {
1242         struct rpc_rqst *pos, *req = task->tk_rqstp;
1243         struct rpc_xprt *xprt = req->rq_xprt;
1244
1245         if (xprt_request_need_enqueue_transmit(task, req)) {
1246                 req->rq_bytes_sent = 0;
1247                 spin_lock(&xprt->queue_lock);
1248                 /*
1249                  * Requests that carry congestion control credits are added
1250                  * to the head of the list to avoid starvation issues.
1251                  */
1252                 if (req->rq_cong) {
1253                         xprt_clear_congestion_window_wait(xprt);
1254                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1255                                 if (pos->rq_cong)
1256                                         continue;
1257                                 /* Note: req is added _before_ pos */
1258                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1259                                 INIT_LIST_HEAD(&req->rq_xmit2);
1260                                 trace_xprt_enq_xmit(task, 1);
1261                                 goto out;
1262                         }
1263                 } else if (RPC_IS_SWAPPER(task)) {
1264                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1265                                 if (pos->rq_cong || pos->rq_bytes_sent)
1266                                         continue;
1267                                 if (RPC_IS_SWAPPER(pos->rq_task))
1268                                         continue;
1269                                 /* Note: req is added _before_ pos */
1270                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1271                                 INIT_LIST_HEAD(&req->rq_xmit2);
1272                                 trace_xprt_enq_xmit(task, 2);
1273                                 goto out;
1274                         }
1275                 } else if (!req->rq_seqno) {
1276                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1277                                 if (pos->rq_task->tk_owner != task->tk_owner)
1278                                         continue;
1279                                 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1280                                 INIT_LIST_HEAD(&req->rq_xmit);
1281                                 trace_xprt_enq_xmit(task, 3);
1282                                 goto out;
1283                         }
1284                 }
1285                 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1286                 INIT_LIST_HEAD(&req->rq_xmit2);
1287                 trace_xprt_enq_xmit(task, 4);
1288 out:
1289                 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1290                 spin_unlock(&xprt->queue_lock);
1291         }
1292 }
1293
1294 /**
1295  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1296  * @task: pointer to rpc_task
1297  *
1298  * Remove a task from the transmission queue
1299  * Caller must hold xprt->queue_lock
1300  */
1301 static void
1302 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1303 {
1304         struct rpc_rqst *req = task->tk_rqstp;
1305
1306         if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1307                 return;
1308         if (!list_empty(&req->rq_xmit)) {
1309                 list_del(&req->rq_xmit);
1310                 if (!list_empty(&req->rq_xmit2)) {
1311                         struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1312                                         struct rpc_rqst, rq_xmit2);
1313                         list_del(&req->rq_xmit2);
1314                         list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1315                 }
1316         } else
1317                 list_del(&req->rq_xmit2);
1318 }
1319
1320 /**
1321  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1322  * @task: pointer to rpc_task
1323  *
1324  * Remove a task from the transmission queue
1325  */
1326 static void
1327 xprt_request_dequeue_transmit(struct rpc_task *task)
1328 {
1329         struct rpc_rqst *req = task->tk_rqstp;
1330         struct rpc_xprt *xprt = req->rq_xprt;
1331
1332         spin_lock(&xprt->queue_lock);
1333         xprt_request_dequeue_transmit_locked(task);
1334         spin_unlock(&xprt->queue_lock);
1335 }
1336
1337 /**
1338  * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1339  * @task: pointer to rpc_task
1340  *
1341  * Remove a task from the transmit and receive queues, and ensure that
1342  * it is not pinned by the receive work item.
1343  */
1344 void
1345 xprt_request_dequeue_xprt(struct rpc_task *task)
1346 {
1347         struct rpc_rqst *req = task->tk_rqstp;
1348         struct rpc_xprt *xprt = req->rq_xprt;
1349
1350         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1351             test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1352             xprt_is_pinned_rqst(req)) {
1353                 spin_lock(&xprt->queue_lock);
1354                 xprt_request_dequeue_transmit_locked(task);
1355                 xprt_request_dequeue_receive_locked(task);
1356                 while (xprt_is_pinned_rqst(req)) {
1357                         set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1358                         spin_unlock(&xprt->queue_lock);
1359                         xprt_wait_on_pinned_rqst(req);
1360                         spin_lock(&xprt->queue_lock);
1361                         clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1362                 }
1363                 spin_unlock(&xprt->queue_lock);
1364         }
1365 }
1366
1367 /**
1368  * xprt_request_prepare - prepare an encoded request for transport
1369  * @req: pointer to rpc_rqst
1370  *
1371  * Calls into the transport layer to do whatever is needed to prepare
1372  * the request for transmission or receive.
1373  */
1374 void
1375 xprt_request_prepare(struct rpc_rqst *req)
1376 {
1377         struct rpc_xprt *xprt = req->rq_xprt;
1378
1379         if (xprt->ops->prepare_request)
1380                 xprt->ops->prepare_request(req);
1381 }
1382
1383 /**
1384  * xprt_request_need_retransmit - Test if a task needs retransmission
1385  * @task: pointer to rpc_task
1386  *
1387  * Test for whether a connection breakage requires the task to retransmit
1388  */
1389 bool
1390 xprt_request_need_retransmit(struct rpc_task *task)
1391 {
1392         return xprt_request_retransmit_after_disconnect(task);
1393 }
1394
1395 /**
1396  * xprt_prepare_transmit - reserve the transport before sending a request
1397  * @task: RPC task about to send a request
1398  *
1399  */
1400 bool xprt_prepare_transmit(struct rpc_task *task)
1401 {
1402         struct rpc_rqst *req = task->tk_rqstp;
1403         struct rpc_xprt *xprt = req->rq_xprt;
1404
1405         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1406
1407         if (!xprt_lock_write(xprt, task)) {
1408                 /* Race breaker: someone may have transmitted us */
1409                 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1410                         rpc_wake_up_queued_task_set_status(&xprt->sending,
1411                                         task, 0);
1412                 return false;
1413
1414         }
1415         return true;
1416 }
1417
1418 void xprt_end_transmit(struct rpc_task *task)
1419 {
1420         xprt_release_write(task->tk_rqstp->rq_xprt, task);
1421 }
1422
1423 /**
1424  * xprt_request_transmit - send an RPC request on a transport
1425  * @req: pointer to request to transmit
1426  * @snd_task: RPC task that owns the transport lock
1427  *
1428  * This performs the transmission of a single request.
1429  * Note that if the request is not the same as snd_task, then it
1430  * does need to be pinned.
1431  * Returns '0' on success.
1432  */
1433 static int
1434 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1435 {
1436         struct rpc_xprt *xprt = req->rq_xprt;
1437         struct rpc_task *task = req->rq_task;
1438         unsigned int connect_cookie;
1439         int is_retrans = RPC_WAS_SENT(task);
1440         int status;
1441
1442         if (!req->rq_bytes_sent) {
1443                 if (xprt_request_data_received(task)) {
1444                         status = 0;
1445                         goto out_dequeue;
1446                 }
1447                 /* Verify that our message lies in the RPCSEC_GSS window */
1448                 if (rpcauth_xmit_need_reencode(task)) {
1449                         status = -EBADMSG;
1450                         goto out_dequeue;
1451                 }
1452                 if (RPC_SIGNALLED(task)) {
1453                         status = -ERESTARTSYS;
1454                         goto out_dequeue;
1455                 }
1456         }
1457
1458         /*
1459          * Update req->rq_ntrans before transmitting to avoid races with
1460          * xprt_update_rtt(), which needs to know that it is recording a
1461          * reply to the first transmission.
1462          */
1463         req->rq_ntrans++;
1464
1465         connect_cookie = xprt->connect_cookie;
1466         status = xprt->ops->send_request(req);
1467         if (status != 0) {
1468                 req->rq_ntrans--;
1469                 trace_xprt_transmit(req, status);
1470                 return status;
1471         }
1472
1473         if (is_retrans)
1474                 task->tk_client->cl_stats->rpcretrans++;
1475
1476         xprt_inject_disconnect(xprt);
1477
1478         task->tk_flags |= RPC_TASK_SENT;
1479         spin_lock(&xprt->transport_lock);
1480
1481         xprt->stat.sends++;
1482         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1483         xprt->stat.bklog_u += xprt->backlog.qlen;
1484         xprt->stat.sending_u += xprt->sending.qlen;
1485         xprt->stat.pending_u += xprt->pending.qlen;
1486         spin_unlock(&xprt->transport_lock);
1487
1488         req->rq_connect_cookie = connect_cookie;
1489 out_dequeue:
1490         trace_xprt_transmit(req, status);
1491         xprt_request_dequeue_transmit(task);
1492         rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1493         return status;
1494 }
1495
1496 /**
1497  * xprt_transmit - send an RPC request on a transport
1498  * @task: controlling RPC task
1499  *
1500  * Attempts to drain the transmit queue. On exit, either the transport
1501  * signalled an error that needs to be handled before transmission can
1502  * resume, or @task finished transmitting, and detected that it already
1503  * received a reply.
1504  */
1505 void
1506 xprt_transmit(struct rpc_task *task)
1507 {
1508         struct rpc_rqst *next, *req = task->tk_rqstp;
1509         struct rpc_xprt *xprt = req->rq_xprt;
1510         int status;
1511
1512         spin_lock(&xprt->queue_lock);
1513         while (!list_empty(&xprt->xmit_queue)) {
1514                 next = list_first_entry(&xprt->xmit_queue,
1515                                 struct rpc_rqst, rq_xmit);
1516                 xprt_pin_rqst(next);
1517                 spin_unlock(&xprt->queue_lock);
1518                 status = xprt_request_transmit(next, task);
1519                 if (status == -EBADMSG && next != req)
1520                         status = 0;
1521                 cond_resched();
1522                 spin_lock(&xprt->queue_lock);
1523                 xprt_unpin_rqst(next);
1524                 if (status == 0) {
1525                         if (!xprt_request_data_received(task) ||
1526                             test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1527                                 continue;
1528                 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1529                         task->tk_status = status;
1530                 break;
1531         }
1532         spin_unlock(&xprt->queue_lock);
1533 }
1534
1535 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1536 {
1537         set_bit(XPRT_CONGESTED, &xprt->state);
1538         rpc_sleep_on(&xprt->backlog, task, NULL);
1539 }
1540
1541 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1542 {
1543         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1544                 clear_bit(XPRT_CONGESTED, &xprt->state);
1545 }
1546
1547 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1548 {
1549         bool ret = false;
1550
1551         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1552                 goto out;
1553         spin_lock(&xprt->reserve_lock);
1554         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1555                 rpc_sleep_on(&xprt->backlog, task, NULL);
1556                 ret = true;
1557         }
1558         spin_unlock(&xprt->reserve_lock);
1559 out:
1560         return ret;
1561 }
1562
1563 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1564 {
1565         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1566
1567         if (xprt->num_reqs >= xprt->max_reqs)
1568                 goto out;
1569         ++xprt->num_reqs;
1570         spin_unlock(&xprt->reserve_lock);
1571         req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1572         spin_lock(&xprt->reserve_lock);
1573         if (req != NULL)
1574                 goto out;
1575         --xprt->num_reqs;
1576         req = ERR_PTR(-ENOMEM);
1577 out:
1578         return req;
1579 }
1580
1581 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1582 {
1583         if (xprt->num_reqs > xprt->min_reqs) {
1584                 --xprt->num_reqs;
1585                 kfree(req);
1586                 return true;
1587         }
1588         return false;
1589 }
1590
1591 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1592 {
1593         struct rpc_rqst *req;
1594
1595         spin_lock(&xprt->reserve_lock);
1596         if (!list_empty(&xprt->free)) {
1597                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1598                 list_del(&req->rq_list);
1599                 goto out_init_req;
1600         }
1601         req = xprt_dynamic_alloc_slot(xprt);
1602         if (!IS_ERR(req))
1603                 goto out_init_req;
1604         switch (PTR_ERR(req)) {
1605         case -ENOMEM:
1606                 dprintk("RPC:       dynamic allocation of request slot "
1607                                 "failed! Retrying\n");
1608                 task->tk_status = -ENOMEM;
1609                 break;
1610         case -EAGAIN:
1611                 xprt_add_backlog(xprt, task);
1612                 dprintk("RPC:       waiting for request slot\n");
1613                 /* fall through */
1614         default:
1615                 task->tk_status = -EAGAIN;
1616         }
1617         spin_unlock(&xprt->reserve_lock);
1618         return;
1619 out_init_req:
1620         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1621                                      xprt->num_reqs);
1622         spin_unlock(&xprt->reserve_lock);
1623
1624         task->tk_status = 0;
1625         task->tk_rqstp = req;
1626 }
1627 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1628
1629 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1630 {
1631         spin_lock(&xprt->reserve_lock);
1632         if (!xprt_dynamic_free_slot(xprt, req)) {
1633                 memset(req, 0, sizeof(*req));   /* mark unused */
1634                 list_add(&req->rq_list, &xprt->free);
1635         }
1636         xprt_wake_up_backlog(xprt);
1637         spin_unlock(&xprt->reserve_lock);
1638 }
1639 EXPORT_SYMBOL_GPL(xprt_free_slot);
1640
1641 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1642 {
1643         struct rpc_rqst *req;
1644         while (!list_empty(&xprt->free)) {
1645                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1646                 list_del(&req->rq_list);
1647                 kfree(req);
1648         }
1649 }
1650
1651 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1652                 unsigned int num_prealloc,
1653                 unsigned int max_alloc)
1654 {
1655         struct rpc_xprt *xprt;
1656         struct rpc_rqst *req;
1657         int i;
1658
1659         xprt = kzalloc(size, GFP_KERNEL);
1660         if (xprt == NULL)
1661                 goto out;
1662
1663         xprt_init(xprt, net);
1664
1665         for (i = 0; i < num_prealloc; i++) {
1666                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1667                 if (!req)
1668                         goto out_free;
1669                 list_add(&req->rq_list, &xprt->free);
1670         }
1671         if (max_alloc > num_prealloc)
1672                 xprt->max_reqs = max_alloc;
1673         else
1674                 xprt->max_reqs = num_prealloc;
1675         xprt->min_reqs = num_prealloc;
1676         xprt->num_reqs = num_prealloc;
1677
1678         return xprt;
1679
1680 out_free:
1681         xprt_free(xprt);
1682 out:
1683         return NULL;
1684 }
1685 EXPORT_SYMBOL_GPL(xprt_alloc);
1686
1687 void xprt_free(struct rpc_xprt *xprt)
1688 {
1689         put_net(xprt->xprt_net);
1690         xprt_free_all_slots(xprt);
1691         kfree_rcu(xprt, rcu);
1692 }
1693 EXPORT_SYMBOL_GPL(xprt_free);
1694
1695 static void
1696 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1697 {
1698         req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1699 }
1700
1701 static __be32
1702 xprt_alloc_xid(struct rpc_xprt *xprt)
1703 {
1704         __be32 xid;
1705
1706         spin_lock(&xprt->reserve_lock);
1707         xid = (__force __be32)xprt->xid++;
1708         spin_unlock(&xprt->reserve_lock);
1709         return xid;
1710 }
1711
1712 static void
1713 xprt_init_xid(struct rpc_xprt *xprt)
1714 {
1715         xprt->xid = prandom_u32();
1716 }
1717
1718 static void
1719 xprt_request_init(struct rpc_task *task)
1720 {
1721         struct rpc_xprt *xprt = task->tk_xprt;
1722         struct rpc_rqst *req = task->tk_rqstp;
1723
1724         req->rq_task    = task;
1725         req->rq_xprt    = xprt;
1726         req->rq_buffer  = NULL;
1727         req->rq_xid     = xprt_alloc_xid(xprt);
1728         xprt_init_connect_cookie(req, xprt);
1729         req->rq_snd_buf.len = 0;
1730         req->rq_snd_buf.buflen = 0;
1731         req->rq_rcv_buf.len = 0;
1732         req->rq_rcv_buf.buflen = 0;
1733         req->rq_snd_buf.bvec = NULL;
1734         req->rq_rcv_buf.bvec = NULL;
1735         req->rq_release_snd_buf = NULL;
1736         xprt_init_majortimeo(task, req);
1737         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1738                         req, ntohl(req->rq_xid));
1739 }
1740
1741 static void
1742 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1743 {
1744         xprt->ops->alloc_slot(xprt, task);
1745         if (task->tk_rqstp != NULL)
1746                 xprt_request_init(task);
1747 }
1748
1749 /**
1750  * xprt_reserve - allocate an RPC request slot
1751  * @task: RPC task requesting a slot allocation
1752  *
1753  * If the transport is marked as being congested, or if no more
1754  * slots are available, place the task on the transport's
1755  * backlog queue.
1756  */
1757 void xprt_reserve(struct rpc_task *task)
1758 {
1759         struct rpc_xprt *xprt = task->tk_xprt;
1760
1761         task->tk_status = 0;
1762         if (task->tk_rqstp != NULL)
1763                 return;
1764
1765         task->tk_status = -EAGAIN;
1766         if (!xprt_throttle_congested(xprt, task))
1767                 xprt_do_reserve(xprt, task);
1768 }
1769
1770 /**
1771  * xprt_retry_reserve - allocate an RPC request slot
1772  * @task: RPC task requesting a slot allocation
1773  *
1774  * If no more slots are available, place the task on the transport's
1775  * backlog queue.
1776  * Note that the only difference with xprt_reserve is that we now
1777  * ignore the value of the XPRT_CONGESTED flag.
1778  */
1779 void xprt_retry_reserve(struct rpc_task *task)
1780 {
1781         struct rpc_xprt *xprt = task->tk_xprt;
1782
1783         task->tk_status = 0;
1784         if (task->tk_rqstp != NULL)
1785                 return;
1786
1787         task->tk_status = -EAGAIN;
1788         xprt_do_reserve(xprt, task);
1789 }
1790
1791 /**
1792  * xprt_release - release an RPC request slot
1793  * @task: task which is finished with the slot
1794  *
1795  */
1796 void xprt_release(struct rpc_task *task)
1797 {
1798         struct rpc_xprt *xprt;
1799         struct rpc_rqst *req = task->tk_rqstp;
1800
1801         if (req == NULL) {
1802                 if (task->tk_client) {
1803                         xprt = task->tk_xprt;
1804                         xprt_release_write(xprt, task);
1805                 }
1806                 return;
1807         }
1808
1809         xprt = req->rq_xprt;
1810         xprt_request_dequeue_xprt(task);
1811         spin_lock(&xprt->transport_lock);
1812         xprt->ops->release_xprt(xprt, task);
1813         if (xprt->ops->release_request)
1814                 xprt->ops->release_request(task);
1815         xprt_schedule_autodisconnect(xprt);
1816         spin_unlock(&xprt->transport_lock);
1817         if (req->rq_buffer)
1818                 xprt->ops->buf_free(task);
1819         xprt_inject_disconnect(xprt);
1820         xdr_free_bvec(&req->rq_rcv_buf);
1821         xdr_free_bvec(&req->rq_snd_buf);
1822         if (req->rq_cred != NULL)
1823                 put_rpccred(req->rq_cred);
1824         task->tk_rqstp = NULL;
1825         if (req->rq_release_snd_buf)
1826                 req->rq_release_snd_buf(req);
1827
1828         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1829         if (likely(!bc_prealloc(req)))
1830                 xprt->ops->free_slot(xprt, req);
1831         else
1832                 xprt_free_bc_request(req);
1833 }
1834
1835 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1836 void
1837 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1838 {
1839         struct xdr_buf *xbufp = &req->rq_snd_buf;
1840
1841         task->tk_rqstp = req;
1842         req->rq_task = task;
1843         xprt_init_connect_cookie(req, req->rq_xprt);
1844         /*
1845          * Set up the xdr_buf length.
1846          * This also indicates that the buffer is XDR encoded already.
1847          */
1848         xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1849                 xbufp->tail[0].iov_len;
1850 }
1851 #endif
1852
1853 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1854 {
1855         kref_init(&xprt->kref);
1856
1857         spin_lock_init(&xprt->transport_lock);
1858         spin_lock_init(&xprt->reserve_lock);
1859         spin_lock_init(&xprt->queue_lock);
1860
1861         INIT_LIST_HEAD(&xprt->free);
1862         xprt->recv_queue = RB_ROOT;
1863         INIT_LIST_HEAD(&xprt->xmit_queue);
1864 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1865         spin_lock_init(&xprt->bc_pa_lock);
1866         INIT_LIST_HEAD(&xprt->bc_pa_list);
1867 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1868         INIT_LIST_HEAD(&xprt->xprt_switch);
1869
1870         xprt->last_used = jiffies;
1871         xprt->cwnd = RPC_INITCWND;
1872         xprt->bind_index = 0;
1873
1874         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1875         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1876         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1877         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1878
1879         xprt_init_xid(xprt);
1880
1881         xprt->xprt_net = get_net(net);
1882 }
1883
1884 /**
1885  * xprt_create_transport - create an RPC transport
1886  * @args: rpc transport creation arguments
1887  *
1888  */
1889 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1890 {
1891         struct rpc_xprt *xprt;
1892         struct xprt_class *t;
1893
1894         spin_lock(&xprt_list_lock);
1895         list_for_each_entry(t, &xprt_list, list) {
1896                 if (t->ident == args->ident) {
1897                         spin_unlock(&xprt_list_lock);
1898                         goto found;
1899                 }
1900         }
1901         spin_unlock(&xprt_list_lock);
1902         dprintk("RPC: transport (%d) not supported\n", args->ident);
1903         return ERR_PTR(-EIO);
1904
1905 found:
1906         xprt = t->setup(args);
1907         if (IS_ERR(xprt)) {
1908                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1909                                 -PTR_ERR(xprt));
1910                 goto out;
1911         }
1912         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1913                 xprt->idle_timeout = 0;
1914         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1915         if (xprt_has_timer(xprt))
1916                 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1917         else
1918                 timer_setup(&xprt->timer, NULL, 0);
1919
1920         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1921                 xprt_destroy(xprt);
1922                 return ERR_PTR(-EINVAL);
1923         }
1924         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1925         if (xprt->servername == NULL) {
1926                 xprt_destroy(xprt);
1927                 return ERR_PTR(-ENOMEM);
1928         }
1929
1930         rpc_xprt_debugfs_register(xprt);
1931
1932         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1933                         xprt->max_reqs);
1934 out:
1935         return xprt;
1936 }
1937
1938 static void xprt_destroy_cb(struct work_struct *work)
1939 {
1940         struct rpc_xprt *xprt =
1941                 container_of(work, struct rpc_xprt, task_cleanup);
1942
1943         rpc_xprt_debugfs_unregister(xprt);
1944         rpc_destroy_wait_queue(&xprt->binding);
1945         rpc_destroy_wait_queue(&xprt->pending);
1946         rpc_destroy_wait_queue(&xprt->sending);
1947         rpc_destroy_wait_queue(&xprt->backlog);
1948         kfree(xprt->servername);
1949         /*
1950          * Destroy any existing back channel
1951          */
1952         xprt_destroy_backchannel(xprt, UINT_MAX);
1953
1954         /*
1955          * Tear down transport state and free the rpc_xprt
1956          */
1957         xprt->ops->destroy(xprt);
1958 }
1959
1960 /**
1961  * xprt_destroy - destroy an RPC transport, killing off all requests.
1962  * @xprt: transport to destroy
1963  *
1964  */
1965 static void xprt_destroy(struct rpc_xprt *xprt)
1966 {
1967         dprintk("RPC:       destroying transport %p\n", xprt);
1968
1969         /*
1970          * Exclude transport connect/disconnect handlers and autoclose
1971          */
1972         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1973
1974         del_timer_sync(&xprt->timer);
1975
1976         /*
1977          * Destroy sockets etc from the system workqueue so they can
1978          * safely flush receive work running on rpciod.
1979          */
1980         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1981         schedule_work(&xprt->task_cleanup);
1982 }
1983
1984 static void xprt_destroy_kref(struct kref *kref)
1985 {
1986         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1987 }
1988
1989 /**
1990  * xprt_get - return a reference to an RPC transport.
1991  * @xprt: pointer to the transport
1992  *
1993  */
1994 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1995 {
1996         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1997                 return xprt;
1998         return NULL;
1999 }
2000 EXPORT_SYMBOL_GPL(xprt_get);
2001
2002 /**
2003  * xprt_put - release a reference to an RPC transport.
2004  * @xprt: pointer to the transport
2005  *
2006  */
2007 void xprt_put(struct rpc_xprt *xprt)
2008 {
2009         if (xprt != NULL)
2010                 kref_put(&xprt->kref, xprt_destroy_kref);
2011 }
2012 EXPORT_SYMBOL_GPL(xprt_put);