1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/net/sunrpc/xprt.c
5 * This is a generic RPC call interface supporting congestion avoidance,
6 * and asynchronous calls.
8 * The interface works like this:
10 * - When a process places a call, it allocates a request slot if
11 * one is available. Otherwise, it sleeps on the backlog queue
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
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
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
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.
36 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
38 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
41 #include <linux/module.h>
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>
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>
55 #include <trace/events/sunrpc.h>
63 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
64 # define RPCDBG_FACILITY RPCDBG_XPRT
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);
74 static DEFINE_SPINLOCK(xprt_list_lock);
75 static LIST_HEAD(xprt_list);
77 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
79 unsigned long timeout = jiffies + req->rq_timeout;
81 if (time_before(timeout, req->rq_majortimeo))
83 return req->rq_majortimeo;
87 * xprt_register_transport - register a transport implementation
88 * @transport: transport to register
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.
94 * 0: transport successfully registered
95 * -EEXIST: transport already registered
96 * -EINVAL: transport module being unloaded
98 int xprt_register_transport(struct xprt_class *transport)
100 struct xprt_class *t;
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)
111 list_add_tail(&transport->list, &xprt_list);
112 printk(KERN_INFO "RPC: Registered %s transport module.\n",
117 spin_unlock(&xprt_list_lock);
120 EXPORT_SYMBOL_GPL(xprt_register_transport);
123 * xprt_unregister_transport - unregister a transport implementation
124 * @transport: transport to unregister
127 * 0: transport successfully unregistered
128 * -ENOENT: transport never registered
130 int xprt_unregister_transport(struct xprt_class *transport)
132 struct xprt_class *t;
136 spin_lock(&xprt_list_lock);
137 list_for_each_entry(t, &xprt_list, list) {
138 if (t == transport) {
140 "RPC: Unregistered %s transport module.\n",
142 list_del_init(&transport->list);
149 spin_unlock(&xprt_list_lock);
152 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
155 * xprt_load_transport - load a transport implementation
156 * @transport_name: transport to load
159 * 0: transport successfully loaded
160 * -ENOENT: transport module not available
162 int xprt_load_transport(const char *transport_name)
164 struct xprt_class *t;
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);
175 spin_unlock(&xprt_list_lock);
176 result = request_module("xprt%s", transport_name);
180 EXPORT_SYMBOL_GPL(xprt_load_transport);
182 static void xprt_clear_locked(struct rpc_xprt *xprt)
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();
190 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
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
198 * This prevents mixing the payload of separate requests, and prevents
199 * transport connects from colliding with writes. No congestion control
202 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
204 struct rpc_rqst *req = task->tk_rqstp;
206 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
207 if (task == xprt->snd_task)
211 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
213 xprt->snd_task = task;
218 xprt_clear_locked(xprt);
220 dprintk("RPC: %5u failed to lock transport %p\n",
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));
227 rpc_sleep_on(&xprt->sending, task, NULL);
230 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
233 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
235 return test_bit(XPRT_CWND_WAIT, &xprt->state);
239 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
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,
247 set_bit(XPRT_CWND_WAIT, &xprt->state);
251 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
253 if (!RPCXPRT_CONGESTED(xprt))
254 clear_bit(XPRT_CWND_WAIT, &xprt->state);
258 * xprt_reserve_xprt_cong - serialize write access to transports
259 * @task: task that is requesting access to the transport
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.
266 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
268 struct rpc_rqst *req = task->tk_rqstp;
270 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
271 if (task == xprt->snd_task)
276 xprt->snd_task = task;
279 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
281 if (!xprt_need_congestion_window_wait(xprt)) {
282 xprt->snd_task = task;
286 xprt_clear_locked(xprt);
288 dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
289 task->tk_status = -EAGAIN;
290 if (RPC_IS_SOFT(task))
291 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
292 xprt_request_timeout(req));
294 rpc_sleep_on(&xprt->sending, task, NULL);
297 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
299 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
303 if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
305 spin_lock(&xprt->transport_lock);
306 retval = xprt->ops->reserve_xprt(xprt, task);
307 spin_unlock(&xprt->transport_lock);
311 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
313 struct rpc_xprt *xprt = data;
315 xprt->snd_task = task;
319 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
321 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
323 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
325 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
326 __xprt_lock_write_func, xprt))
329 xprt_clear_locked(xprt);
332 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
334 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
336 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
338 if (xprt_need_congestion_window_wait(xprt))
340 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
341 __xprt_lock_write_func, xprt))
344 xprt_clear_locked(xprt);
348 * xprt_release_xprt - allow other requests to use a transport
349 * @xprt: transport with other tasks potentially waiting
350 * @task: task that is releasing access to the transport
352 * Note that "task" can be NULL. No congestion control is provided.
354 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
356 if (xprt->snd_task == task) {
357 xprt_clear_locked(xprt);
358 __xprt_lock_write_next(xprt);
361 EXPORT_SYMBOL_GPL(xprt_release_xprt);
364 * xprt_release_xprt_cong - allow other requests to use a transport
365 * @xprt: transport with other tasks potentially waiting
366 * @task: task that is releasing access to the transport
368 * Note that "task" can be NULL. Another task is awoken to use the
369 * transport if the transport's congestion window allows it.
371 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
373 if (xprt->snd_task == task) {
374 xprt_clear_locked(xprt);
375 __xprt_lock_write_next_cong(xprt);
378 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
380 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
382 if (xprt->snd_task != task)
384 spin_lock(&xprt->transport_lock);
385 xprt->ops->release_xprt(xprt, task);
386 spin_unlock(&xprt->transport_lock);
390 * Van Jacobson congestion avoidance. Check if the congestion window
391 * overflowed. Put the task to sleep if this is the case.
394 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
398 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
399 req->rq_task->tk_pid, xprt->cong, xprt->cwnd);
400 if (RPCXPRT_CONGESTED(xprt)) {
401 xprt_set_congestion_window_wait(xprt);
405 xprt->cong += RPC_CWNDSCALE;
410 * Adjust the congestion window, and wake up the next task
411 * that has been sleeping due to congestion
414 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
419 xprt->cong -= RPC_CWNDSCALE;
420 xprt_test_and_clear_congestion_window_wait(xprt);
421 __xprt_lock_write_next_cong(xprt);
425 * xprt_request_get_cong - Request congestion control credits
426 * @xprt: pointer to transport
427 * @req: pointer to RPC request
429 * Useful for transports that require congestion control.
432 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
438 spin_lock(&xprt->transport_lock);
439 ret = __xprt_get_cong(xprt, req) != 0;
440 spin_unlock(&xprt->transport_lock);
443 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
446 * xprt_release_rqst_cong - housekeeping when request is complete
447 * @task: RPC request that recently completed
449 * Useful for transports that require congestion control.
451 void xprt_release_rqst_cong(struct rpc_task *task)
453 struct rpc_rqst *req = task->tk_rqstp;
455 __xprt_put_cong(req->rq_xprt, req);
457 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
460 * Clear the congestion window wait flag and wake up the next
461 * entry on xprt->sending
464 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
466 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
467 spin_lock(&xprt->transport_lock);
468 __xprt_lock_write_next_cong(xprt);
469 spin_unlock(&xprt->transport_lock);
474 * xprt_adjust_cwnd - adjust transport congestion window
475 * @xprt: pointer to xprt
476 * @task: recently completed RPC request used to adjust window
477 * @result: result code of completed RPC request
479 * The transport code maintains an estimate on the maximum number of out-
480 * standing RPC requests, using a smoothed version of the congestion
481 * avoidance implemented in 44BSD. This is basically the Van Jacobson
482 * congestion algorithm: If a retransmit occurs, the congestion window is
483 * halved; otherwise, it is incremented by 1/cwnd when
485 * - a reply is received and
486 * - a full number of requests are outstanding and
487 * - the congestion window hasn't been updated recently.
489 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
491 struct rpc_rqst *req = task->tk_rqstp;
492 unsigned long cwnd = xprt->cwnd;
494 if (result >= 0 && cwnd <= xprt->cong) {
495 /* The (cwnd >> 1) term makes sure
496 * the result gets rounded properly. */
497 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
498 if (cwnd > RPC_MAXCWND(xprt))
499 cwnd = RPC_MAXCWND(xprt);
500 __xprt_lock_write_next_cong(xprt);
501 } else if (result == -ETIMEDOUT) {
503 if (cwnd < RPC_CWNDSCALE)
504 cwnd = RPC_CWNDSCALE;
506 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
507 xprt->cong, xprt->cwnd, cwnd);
509 __xprt_put_cong(xprt, req);
511 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
514 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
515 * @xprt: transport with waiting tasks
516 * @status: result code to plant in each task before waking it
519 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
522 rpc_wake_up_status(&xprt->pending, status);
524 rpc_wake_up(&xprt->pending);
526 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
529 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
532 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
533 * we don't in general want to force a socket disconnection due to
534 * an incomplete RPC call transmission.
536 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
538 set_bit(XPRT_WRITE_SPACE, &xprt->state);
540 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
543 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
545 if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
546 __xprt_lock_write_next(xprt);
547 dprintk("RPC: write space: waking waiting task on "
555 * xprt_write_space - wake the task waiting for transport output buffer space
556 * @xprt: transport with waiting tasks
558 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
560 bool xprt_write_space(struct rpc_xprt *xprt)
564 if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
566 spin_lock(&xprt->transport_lock);
567 ret = xprt_clear_write_space_locked(xprt);
568 spin_unlock(&xprt->transport_lock);
571 EXPORT_SYMBOL_GPL(xprt_write_space);
573 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
575 s64 delta = ktime_to_ns(ktime_get() - abstime);
576 return likely(delta >= 0) ?
577 jiffies - nsecs_to_jiffies(delta) :
578 jiffies + nsecs_to_jiffies(-delta);
581 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
583 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
584 unsigned long majortimeo = req->rq_timeout;
586 if (to->to_exponential)
587 majortimeo <<= to->to_retries;
589 majortimeo += to->to_increment * to->to_retries;
590 if (majortimeo > to->to_maxval || majortimeo == 0)
591 majortimeo = to->to_maxval;
595 static void xprt_reset_majortimeo(struct rpc_rqst *req)
597 req->rq_majortimeo += xprt_calc_majortimeo(req);
600 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
602 unsigned long time_init;
603 struct rpc_xprt *xprt = req->rq_xprt;
605 if (likely(xprt && xprt_connected(xprt)))
608 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
609 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
610 req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
614 * xprt_adjust_timeout - adjust timeout values for next retransmit
615 * @req: RPC request containing parameters to use for the adjustment
618 int xprt_adjust_timeout(struct rpc_rqst *req)
620 struct rpc_xprt *xprt = req->rq_xprt;
621 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
624 if (time_before(jiffies, req->rq_majortimeo)) {
625 if (to->to_exponential)
626 req->rq_timeout <<= 1;
628 req->rq_timeout += to->to_increment;
629 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
630 req->rq_timeout = to->to_maxval;
633 req->rq_timeout = to->to_initval;
635 xprt_reset_majortimeo(req);
636 /* Reset the RTT counters == "slow start" */
637 spin_lock(&xprt->transport_lock);
638 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
639 spin_unlock(&xprt->transport_lock);
643 if (req->rq_timeout == 0) {
644 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
645 req->rq_timeout = 5 * HZ;
650 static void xprt_autoclose(struct work_struct *work)
652 struct rpc_xprt *xprt =
653 container_of(work, struct rpc_xprt, task_cleanup);
654 unsigned int pflags = memalloc_nofs_save();
656 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
657 xprt->ops->close(xprt);
658 xprt_release_write(xprt, NULL);
659 wake_up_bit(&xprt->state, XPRT_LOCKED);
660 memalloc_nofs_restore(pflags);
664 * xprt_disconnect_done - mark a transport as disconnected
665 * @xprt: transport to flag for disconnect
668 void xprt_disconnect_done(struct rpc_xprt *xprt)
670 dprintk("RPC: disconnected transport %p\n", xprt);
671 spin_lock(&xprt->transport_lock);
672 xprt_clear_connected(xprt);
673 xprt_clear_write_space_locked(xprt);
674 xprt_wake_pending_tasks(xprt, -ENOTCONN);
675 spin_unlock(&xprt->transport_lock);
677 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
680 * xprt_force_disconnect - force a transport to disconnect
681 * @xprt: transport to disconnect
684 void xprt_force_disconnect(struct rpc_xprt *xprt)
686 /* Don't race with the test_bit() in xprt_clear_locked() */
687 spin_lock(&xprt->transport_lock);
688 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
689 /* Try to schedule an autoclose RPC call */
690 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
691 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
692 else if (xprt->snd_task)
693 rpc_wake_up_queued_task_set_status(&xprt->pending,
694 xprt->snd_task, -ENOTCONN);
695 spin_unlock(&xprt->transport_lock);
697 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
700 xprt_connect_cookie(struct rpc_xprt *xprt)
702 return READ_ONCE(xprt->connect_cookie);
706 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
708 struct rpc_rqst *req = task->tk_rqstp;
709 struct rpc_xprt *xprt = req->rq_xprt;
711 return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
712 !xprt_connected(xprt);
716 * xprt_conditional_disconnect - force a transport to disconnect
717 * @xprt: transport to disconnect
718 * @cookie: 'connection cookie'
720 * This attempts to break the connection if and only if 'cookie' matches
721 * the current transport 'connection cookie'. It ensures that we don't
722 * try to break the connection more than once when we need to retransmit
723 * a batch of RPC requests.
726 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
728 /* Don't race with the test_bit() in xprt_clear_locked() */
729 spin_lock(&xprt->transport_lock);
730 if (cookie != xprt->connect_cookie)
732 if (test_bit(XPRT_CLOSING, &xprt->state))
734 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
735 /* Try to schedule an autoclose RPC call */
736 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
737 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
738 xprt_wake_pending_tasks(xprt, -EAGAIN);
740 spin_unlock(&xprt->transport_lock);
744 xprt_has_timer(const struct rpc_xprt *xprt)
746 return xprt->idle_timeout != 0;
750 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
751 __must_hold(&xprt->transport_lock)
753 xprt->last_used = jiffies;
754 if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
755 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
759 xprt_init_autodisconnect(struct timer_list *t)
761 struct rpc_xprt *xprt = from_timer(xprt, t, timer);
763 if (!RB_EMPTY_ROOT(&xprt->recv_queue))
765 /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
766 xprt->last_used = jiffies;
767 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
769 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
772 bool xprt_lock_connect(struct rpc_xprt *xprt,
773 struct rpc_task *task,
778 spin_lock(&xprt->transport_lock);
779 if (!test_bit(XPRT_LOCKED, &xprt->state))
781 if (xprt->snd_task != task)
783 xprt->snd_task = cookie;
786 spin_unlock(&xprt->transport_lock);
790 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
792 spin_lock(&xprt->transport_lock);
793 if (xprt->snd_task != cookie)
795 if (!test_bit(XPRT_LOCKED, &xprt->state))
797 xprt->snd_task =NULL;
798 xprt->ops->release_xprt(xprt, NULL);
799 xprt_schedule_autodisconnect(xprt);
801 spin_unlock(&xprt->transport_lock);
802 wake_up_bit(&xprt->state, XPRT_LOCKED);
806 * xprt_connect - schedule a transport connect operation
807 * @task: RPC task that is requesting the connect
810 void xprt_connect(struct rpc_task *task)
812 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
814 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
815 xprt, (xprt_connected(xprt) ? "is" : "is not"));
817 if (!xprt_bound(xprt)) {
818 task->tk_status = -EAGAIN;
821 if (!xprt_lock_write(xprt, task))
824 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
825 xprt->ops->close(xprt);
827 if (!xprt_connected(xprt)) {
828 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
829 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
830 xprt_request_timeout(task->tk_rqstp));
832 if (test_bit(XPRT_CLOSING, &xprt->state))
834 if (xprt_test_and_set_connecting(xprt))
837 if (!xprt_connected(xprt)) {
838 xprt->stat.connect_start = jiffies;
839 xprt->ops->connect(xprt, task);
841 xprt_clear_connecting(xprt);
843 rpc_wake_up_queued_task(&xprt->pending, task);
846 xprt_release_write(xprt, task);
850 * xprt_reconnect_delay - compute the wait before scheduling a connect
851 * @xprt: transport instance
854 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
856 unsigned long start, now = jiffies;
858 start = xprt->stat.connect_start + xprt->reestablish_timeout;
859 if (time_after(start, now))
863 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
866 * xprt_reconnect_backoff - compute the new re-establish timeout
867 * @xprt: transport instance
868 * @init_to: initial reestablish timeout
871 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
873 xprt->reestablish_timeout <<= 1;
874 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
875 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
876 if (xprt->reestablish_timeout < init_to)
877 xprt->reestablish_timeout = init_to;
879 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
881 enum xprt_xid_rb_cmp {
886 static enum xprt_xid_rb_cmp
887 xprt_xid_cmp(__be32 xid1, __be32 xid2)
891 if ((__force u32)xid1 < (__force u32)xid2)
896 static struct rpc_rqst *
897 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
899 struct rb_node *n = xprt->recv_queue.rb_node;
900 struct rpc_rqst *req;
903 req = rb_entry(n, struct rpc_rqst, rq_recv);
904 switch (xprt_xid_cmp(xid, req->rq_xid)) {
919 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
921 struct rb_node **p = &xprt->recv_queue.rb_node;
922 struct rb_node *n = NULL;
923 struct rpc_rqst *req;
927 req = rb_entry(n, struct rpc_rqst, rq_recv);
928 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
936 WARN_ON_ONCE(new != req);
940 rb_link_node(&new->rq_recv, n, p);
941 rb_insert_color(&new->rq_recv, &xprt->recv_queue);
945 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
947 rb_erase(&req->rq_recv, &xprt->recv_queue);
951 * xprt_lookup_rqst - find an RPC request corresponding to an XID
952 * @xprt: transport on which the original request was transmitted
953 * @xid: RPC XID of incoming reply
955 * Caller holds xprt->queue_lock.
957 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
959 struct rpc_rqst *entry;
961 entry = xprt_request_rb_find(xprt, xid);
963 trace_xprt_lookup_rqst(xprt, xid, 0);
964 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
968 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
970 trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
971 xprt->stat.bad_xids++;
974 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
977 xprt_is_pinned_rqst(struct rpc_rqst *req)
979 return atomic_read(&req->rq_pin) != 0;
983 * xprt_pin_rqst - Pin a request on the transport receive list
984 * @req: Request to pin
986 * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
987 * so should be holding xprt->queue_lock.
989 void xprt_pin_rqst(struct rpc_rqst *req)
991 atomic_inc(&req->rq_pin);
993 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
996 * xprt_unpin_rqst - Unpin a request on the transport receive list
997 * @req: Request to pin
999 * Caller should be holding xprt->queue_lock.
1001 void xprt_unpin_rqst(struct rpc_rqst *req)
1003 if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1004 atomic_dec(&req->rq_pin);
1007 if (atomic_dec_and_test(&req->rq_pin))
1008 wake_up_var(&req->rq_pin);
1010 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1012 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1014 wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1018 xprt_request_data_received(struct rpc_task *task)
1020 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1021 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1025 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1027 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1028 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1032 * xprt_request_enqueue_receive - Add an request to the receive queue
1037 xprt_request_enqueue_receive(struct rpc_task *task)
1039 struct rpc_rqst *req = task->tk_rqstp;
1040 struct rpc_xprt *xprt = req->rq_xprt;
1042 if (!xprt_request_need_enqueue_receive(task, req))
1045 xprt_request_prepare(task->tk_rqstp);
1046 spin_lock(&xprt->queue_lock);
1048 /* Update the softirq receive buffer */
1049 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1050 sizeof(req->rq_private_buf));
1052 /* Add request to the receive list */
1053 xprt_request_rb_insert(xprt, req);
1054 set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1055 spin_unlock(&xprt->queue_lock);
1057 /* Turn off autodisconnect */
1058 del_singleshot_timer_sync(&xprt->timer);
1062 * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1065 * Caller must hold xprt->queue_lock.
1068 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1070 struct rpc_rqst *req = task->tk_rqstp;
1072 if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1073 xprt_request_rb_remove(req->rq_xprt, req);
1077 * xprt_update_rtt - Update RPC RTT statistics
1078 * @task: RPC request that recently completed
1080 * Caller holds xprt->queue_lock.
1082 void xprt_update_rtt(struct rpc_task *task)
1084 struct rpc_rqst *req = task->tk_rqstp;
1085 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1086 unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1087 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1090 if (req->rq_ntrans == 1)
1091 rpc_update_rtt(rtt, timer, m);
1092 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1095 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1098 * xprt_complete_rqst - called when reply processing is complete
1099 * @task: RPC request that recently completed
1100 * @copied: actual number of bytes received from the transport
1102 * Caller holds xprt->queue_lock.
1104 void xprt_complete_rqst(struct rpc_task *task, int copied)
1106 struct rpc_rqst *req = task->tk_rqstp;
1107 struct rpc_xprt *xprt = req->rq_xprt;
1109 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
1110 task->tk_pid, ntohl(req->rq_xid), copied);
1111 trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
1115 req->rq_private_buf.len = copied;
1116 /* Ensure all writes are done before we update */
1117 /* req->rq_reply_bytes_recvd */
1119 req->rq_reply_bytes_recvd = copied;
1120 xprt_request_dequeue_receive_locked(task);
1121 rpc_wake_up_queued_task(&xprt->pending, task);
1123 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1125 static void xprt_timer(struct rpc_task *task)
1127 struct rpc_rqst *req = task->tk_rqstp;
1128 struct rpc_xprt *xprt = req->rq_xprt;
1130 if (task->tk_status != -ETIMEDOUT)
1133 trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1134 if (!req->rq_reply_bytes_recvd) {
1135 if (xprt->ops->timer)
1136 xprt->ops->timer(xprt, task);
1138 task->tk_status = 0;
1142 * xprt_wait_for_reply_request_def - wait for reply
1143 * @task: pointer to rpc_task
1145 * Set a request's retransmit timeout based on the transport's
1146 * default timeout parameters. Used by transports that don't adjust
1147 * the retransmit timeout based on round-trip time estimation,
1148 * and put the task to sleep on the pending queue.
1150 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1152 struct rpc_rqst *req = task->tk_rqstp;
1154 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1155 xprt_request_timeout(req));
1157 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1160 * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1161 * @task: pointer to rpc_task
1163 * Set a request's retransmit timeout using the RTT estimator,
1164 * and put the task to sleep on the pending queue.
1166 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1168 int timer = task->tk_msg.rpc_proc->p_timer;
1169 struct rpc_clnt *clnt = task->tk_client;
1170 struct rpc_rtt *rtt = clnt->cl_rtt;
1171 struct rpc_rqst *req = task->tk_rqstp;
1172 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1173 unsigned long timeout;
1175 timeout = rpc_calc_rto(rtt, timer);
1176 timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1177 if (timeout > max_timeout || timeout == 0)
1178 timeout = max_timeout;
1179 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1182 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1185 * xprt_request_wait_receive - wait for the reply to an RPC request
1186 * @task: RPC task about to send a request
1189 void xprt_request_wait_receive(struct rpc_task *task)
1191 struct rpc_rqst *req = task->tk_rqstp;
1192 struct rpc_xprt *xprt = req->rq_xprt;
1194 if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1197 * Sleep on the pending queue if we're expecting a reply.
1198 * The spinlock ensures atomicity between the test of
1199 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1201 spin_lock(&xprt->queue_lock);
1202 if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1203 xprt->ops->wait_for_reply_request(task);
1205 * Send an extra queue wakeup call if the
1206 * connection was dropped in case the call to
1207 * rpc_sleep_on() raced.
1209 if (xprt_request_retransmit_after_disconnect(task))
1210 rpc_wake_up_queued_task_set_status(&xprt->pending,
1213 spin_unlock(&xprt->queue_lock);
1217 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1219 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1223 * xprt_request_enqueue_transmit - queue a task for transmission
1224 * @task: pointer to rpc_task
1226 * Add a task to the transmission queue.
1229 xprt_request_enqueue_transmit(struct rpc_task *task)
1231 struct rpc_rqst *pos, *req = task->tk_rqstp;
1232 struct rpc_xprt *xprt = req->rq_xprt;
1234 if (xprt_request_need_enqueue_transmit(task, req)) {
1235 req->rq_bytes_sent = 0;
1236 spin_lock(&xprt->queue_lock);
1238 * Requests that carry congestion control credits are added
1239 * to the head of the list to avoid starvation issues.
1242 xprt_clear_congestion_window_wait(xprt);
1243 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1246 /* Note: req is added _before_ pos */
1247 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1248 INIT_LIST_HEAD(&req->rq_xmit2);
1249 trace_xprt_enq_xmit(task, 1);
1252 } else if (RPC_IS_SWAPPER(task)) {
1253 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1254 if (pos->rq_cong || pos->rq_bytes_sent)
1256 if (RPC_IS_SWAPPER(pos->rq_task))
1258 /* Note: req is added _before_ pos */
1259 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1260 INIT_LIST_HEAD(&req->rq_xmit2);
1261 trace_xprt_enq_xmit(task, 2);
1264 } else if (!req->rq_seqno) {
1265 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1266 if (pos->rq_task->tk_owner != task->tk_owner)
1268 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1269 INIT_LIST_HEAD(&req->rq_xmit);
1270 trace_xprt_enq_xmit(task, 3);
1274 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1275 INIT_LIST_HEAD(&req->rq_xmit2);
1276 trace_xprt_enq_xmit(task, 4);
1278 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1279 spin_unlock(&xprt->queue_lock);
1284 * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1285 * @task: pointer to rpc_task
1287 * Remove a task from the transmission queue
1288 * Caller must hold xprt->queue_lock
1291 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1293 struct rpc_rqst *req = task->tk_rqstp;
1295 if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1297 if (!list_empty(&req->rq_xmit)) {
1298 list_del(&req->rq_xmit);
1299 if (!list_empty(&req->rq_xmit2)) {
1300 struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1301 struct rpc_rqst, rq_xmit2);
1302 list_del(&req->rq_xmit2);
1303 list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1306 list_del(&req->rq_xmit2);
1310 * xprt_request_dequeue_transmit - remove a task from the transmission queue
1311 * @task: pointer to rpc_task
1313 * Remove a task from the transmission queue
1316 xprt_request_dequeue_transmit(struct rpc_task *task)
1318 struct rpc_rqst *req = task->tk_rqstp;
1319 struct rpc_xprt *xprt = req->rq_xprt;
1321 spin_lock(&xprt->queue_lock);
1322 xprt_request_dequeue_transmit_locked(task);
1323 spin_unlock(&xprt->queue_lock);
1327 * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1328 * @task: pointer to rpc_task
1330 * Remove a task from the transmit and receive queues, and ensure that
1331 * it is not pinned by the receive work item.
1334 xprt_request_dequeue_xprt(struct rpc_task *task)
1336 struct rpc_rqst *req = task->tk_rqstp;
1337 struct rpc_xprt *xprt = req->rq_xprt;
1339 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1340 test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1341 xprt_is_pinned_rqst(req)) {
1342 spin_lock(&xprt->queue_lock);
1343 xprt_request_dequeue_transmit_locked(task);
1344 xprt_request_dequeue_receive_locked(task);
1345 while (xprt_is_pinned_rqst(req)) {
1346 set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1347 spin_unlock(&xprt->queue_lock);
1348 xprt_wait_on_pinned_rqst(req);
1349 spin_lock(&xprt->queue_lock);
1350 clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1352 spin_unlock(&xprt->queue_lock);
1357 * xprt_request_prepare - prepare an encoded request for transport
1358 * @req: pointer to rpc_rqst
1360 * Calls into the transport layer to do whatever is needed to prepare
1361 * the request for transmission or receive.
1364 xprt_request_prepare(struct rpc_rqst *req)
1366 struct rpc_xprt *xprt = req->rq_xprt;
1368 if (xprt->ops->prepare_request)
1369 xprt->ops->prepare_request(req);
1373 * xprt_request_need_retransmit - Test if a task needs retransmission
1374 * @task: pointer to rpc_task
1376 * Test for whether a connection breakage requires the task to retransmit
1379 xprt_request_need_retransmit(struct rpc_task *task)
1381 return xprt_request_retransmit_after_disconnect(task);
1385 * xprt_prepare_transmit - reserve the transport before sending a request
1386 * @task: RPC task about to send a request
1389 bool xprt_prepare_transmit(struct rpc_task *task)
1391 struct rpc_rqst *req = task->tk_rqstp;
1392 struct rpc_xprt *xprt = req->rq_xprt;
1394 dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1396 if (!xprt_lock_write(xprt, task)) {
1397 /* Race breaker: someone may have transmitted us */
1398 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1399 rpc_wake_up_queued_task_set_status(&xprt->sending,
1407 void xprt_end_transmit(struct rpc_task *task)
1409 xprt_release_write(task->tk_rqstp->rq_xprt, task);
1413 * xprt_request_transmit - send an RPC request on a transport
1414 * @req: pointer to request to transmit
1415 * @snd_task: RPC task that owns the transport lock
1417 * This performs the transmission of a single request.
1418 * Note that if the request is not the same as snd_task, then it
1419 * does need to be pinned.
1420 * Returns '0' on success.
1423 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1425 struct rpc_xprt *xprt = req->rq_xprt;
1426 struct rpc_task *task = req->rq_task;
1427 unsigned int connect_cookie;
1428 int is_retrans = RPC_WAS_SENT(task);
1431 if (!req->rq_bytes_sent) {
1432 if (xprt_request_data_received(task)) {
1436 /* Verify that our message lies in the RPCSEC_GSS window */
1437 if (rpcauth_xmit_need_reencode(task)) {
1441 if (task->tk_ops->rpc_call_prepare_transmit) {
1442 task->tk_ops->rpc_call_prepare_transmit(task,
1444 status = task->tk_status;
1448 if (RPC_SIGNALLED(task)) {
1449 status = -ERESTARTSYS;
1455 * Update req->rq_ntrans before transmitting to avoid races with
1456 * xprt_update_rtt(), which needs to know that it is recording a
1457 * reply to the first transmission.
1461 connect_cookie = xprt->connect_cookie;
1462 status = xprt->ops->send_request(req);
1465 trace_xprt_transmit(req, status);
1470 task->tk_client->cl_stats->rpcretrans++;
1472 xprt_inject_disconnect(xprt);
1474 task->tk_flags |= RPC_TASK_SENT;
1475 spin_lock(&xprt->transport_lock);
1478 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1479 xprt->stat.bklog_u += xprt->backlog.qlen;
1480 xprt->stat.sending_u += xprt->sending.qlen;
1481 xprt->stat.pending_u += xprt->pending.qlen;
1482 spin_unlock(&xprt->transport_lock);
1484 req->rq_connect_cookie = connect_cookie;
1486 trace_xprt_transmit(req, status);
1487 xprt_request_dequeue_transmit(task);
1488 rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1493 * xprt_transmit - send an RPC request on a transport
1494 * @task: controlling RPC task
1496 * Attempts to drain the transmit queue. On exit, either the transport
1497 * signalled an error that needs to be handled before transmission can
1498 * resume, or @task finished transmitting, and detected that it already
1502 xprt_transmit(struct rpc_task *task)
1504 struct rpc_rqst *next, *req = task->tk_rqstp;
1505 struct rpc_xprt *xprt = req->rq_xprt;
1508 spin_lock(&xprt->queue_lock);
1509 while (!list_empty(&xprt->xmit_queue)) {
1510 next = list_first_entry(&xprt->xmit_queue,
1511 struct rpc_rqst, rq_xmit);
1512 xprt_pin_rqst(next);
1513 spin_unlock(&xprt->queue_lock);
1514 status = xprt_request_transmit(next, task);
1515 if (status == -EBADMSG && next != req)
1518 spin_lock(&xprt->queue_lock);
1519 xprt_unpin_rqst(next);
1521 if (!xprt_request_data_received(task) ||
1522 test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1524 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1525 task->tk_status = status;
1528 spin_unlock(&xprt->queue_lock);
1531 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1533 set_bit(XPRT_CONGESTED, &xprt->state);
1534 rpc_sleep_on(&xprt->backlog, task, NULL);
1537 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1539 if (rpc_wake_up_next(&xprt->backlog) == NULL)
1540 clear_bit(XPRT_CONGESTED, &xprt->state);
1543 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1547 if (!test_bit(XPRT_CONGESTED, &xprt->state))
1549 spin_lock(&xprt->reserve_lock);
1550 if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1551 rpc_sleep_on(&xprt->backlog, task, NULL);
1554 spin_unlock(&xprt->reserve_lock);
1559 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1561 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1563 if (xprt->num_reqs >= xprt->max_reqs)
1566 spin_unlock(&xprt->reserve_lock);
1567 req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1568 spin_lock(&xprt->reserve_lock);
1572 req = ERR_PTR(-ENOMEM);
1577 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1579 if (xprt->num_reqs > xprt->min_reqs) {
1587 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1589 struct rpc_rqst *req;
1591 spin_lock(&xprt->reserve_lock);
1592 if (!list_empty(&xprt->free)) {
1593 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1594 list_del(&req->rq_list);
1597 req = xprt_dynamic_alloc_slot(xprt);
1600 switch (PTR_ERR(req)) {
1602 dprintk("RPC: dynamic allocation of request slot "
1603 "failed! Retrying\n");
1604 task->tk_status = -ENOMEM;
1607 xprt_add_backlog(xprt, task);
1608 dprintk("RPC: waiting for request slot\n");
1611 task->tk_status = -EAGAIN;
1613 spin_unlock(&xprt->reserve_lock);
1616 xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1618 spin_unlock(&xprt->reserve_lock);
1620 task->tk_status = 0;
1621 task->tk_rqstp = req;
1623 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1625 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1627 spin_lock(&xprt->reserve_lock);
1628 if (!xprt_dynamic_free_slot(xprt, req)) {
1629 memset(req, 0, sizeof(*req)); /* mark unused */
1630 list_add(&req->rq_list, &xprt->free);
1632 xprt_wake_up_backlog(xprt);
1633 spin_unlock(&xprt->reserve_lock);
1635 EXPORT_SYMBOL_GPL(xprt_free_slot);
1637 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1639 struct rpc_rqst *req;
1640 while (!list_empty(&xprt->free)) {
1641 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1642 list_del(&req->rq_list);
1647 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1648 unsigned int num_prealloc,
1649 unsigned int max_alloc)
1651 struct rpc_xprt *xprt;
1652 struct rpc_rqst *req;
1655 xprt = kzalloc(size, GFP_KERNEL);
1659 xprt_init(xprt, net);
1661 for (i = 0; i < num_prealloc; i++) {
1662 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1665 list_add(&req->rq_list, &xprt->free);
1667 if (max_alloc > num_prealloc)
1668 xprt->max_reqs = max_alloc;
1670 xprt->max_reqs = num_prealloc;
1671 xprt->min_reqs = num_prealloc;
1672 xprt->num_reqs = num_prealloc;
1681 EXPORT_SYMBOL_GPL(xprt_alloc);
1683 void xprt_free(struct rpc_xprt *xprt)
1685 put_net(xprt->xprt_net);
1686 xprt_free_all_slots(xprt);
1687 kfree_rcu(xprt, rcu);
1689 EXPORT_SYMBOL_GPL(xprt_free);
1692 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1694 req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1698 xprt_alloc_xid(struct rpc_xprt *xprt)
1702 spin_lock(&xprt->reserve_lock);
1703 xid = (__force __be32)xprt->xid++;
1704 spin_unlock(&xprt->reserve_lock);
1709 xprt_init_xid(struct rpc_xprt *xprt)
1711 xprt->xid = prandom_u32();
1715 xprt_request_init(struct rpc_task *task)
1717 struct rpc_xprt *xprt = task->tk_xprt;
1718 struct rpc_rqst *req = task->tk_rqstp;
1720 req->rq_task = task;
1721 req->rq_xprt = xprt;
1722 req->rq_buffer = NULL;
1723 req->rq_xid = xprt_alloc_xid(xprt);
1724 xprt_init_connect_cookie(req, xprt);
1725 req->rq_snd_buf.len = 0;
1726 req->rq_snd_buf.buflen = 0;
1727 req->rq_rcv_buf.len = 0;
1728 req->rq_rcv_buf.buflen = 0;
1729 req->rq_snd_buf.bvec = NULL;
1730 req->rq_rcv_buf.bvec = NULL;
1731 req->rq_release_snd_buf = NULL;
1732 xprt_init_majortimeo(task, req);
1733 dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1734 req, ntohl(req->rq_xid));
1738 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1740 xprt->ops->alloc_slot(xprt, task);
1741 if (task->tk_rqstp != NULL)
1742 xprt_request_init(task);
1746 * xprt_reserve - allocate an RPC request slot
1747 * @task: RPC task requesting a slot allocation
1749 * If the transport is marked as being congested, or if no more
1750 * slots are available, place the task on the transport's
1753 void xprt_reserve(struct rpc_task *task)
1755 struct rpc_xprt *xprt = task->tk_xprt;
1757 task->tk_status = 0;
1758 if (task->tk_rqstp != NULL)
1761 task->tk_status = -EAGAIN;
1762 if (!xprt_throttle_congested(xprt, task))
1763 xprt_do_reserve(xprt, task);
1767 * xprt_retry_reserve - allocate an RPC request slot
1768 * @task: RPC task requesting a slot allocation
1770 * If no more slots are available, place the task on the transport's
1772 * Note that the only difference with xprt_reserve is that we now
1773 * ignore the value of the XPRT_CONGESTED flag.
1775 void xprt_retry_reserve(struct rpc_task *task)
1777 struct rpc_xprt *xprt = task->tk_xprt;
1779 task->tk_status = 0;
1780 if (task->tk_rqstp != NULL)
1783 task->tk_status = -EAGAIN;
1784 xprt_do_reserve(xprt, task);
1788 * xprt_release - release an RPC request slot
1789 * @task: task which is finished with the slot
1792 void xprt_release(struct rpc_task *task)
1794 struct rpc_xprt *xprt;
1795 struct rpc_rqst *req = task->tk_rqstp;
1798 if (task->tk_client) {
1799 xprt = task->tk_xprt;
1800 xprt_release_write(xprt, task);
1805 xprt = req->rq_xprt;
1806 xprt_request_dequeue_xprt(task);
1807 spin_lock(&xprt->transport_lock);
1808 xprt->ops->release_xprt(xprt, task);
1809 if (xprt->ops->release_request)
1810 xprt->ops->release_request(task);
1811 xprt_schedule_autodisconnect(xprt);
1812 spin_unlock(&xprt->transport_lock);
1814 xprt->ops->buf_free(task);
1815 xprt_inject_disconnect(xprt);
1816 xdr_free_bvec(&req->rq_rcv_buf);
1817 xdr_free_bvec(&req->rq_snd_buf);
1818 if (req->rq_cred != NULL)
1819 put_rpccred(req->rq_cred);
1820 task->tk_rqstp = NULL;
1821 if (req->rq_release_snd_buf)
1822 req->rq_release_snd_buf(req);
1824 dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1825 if (likely(!bc_prealloc(req)))
1826 xprt->ops->free_slot(xprt, req);
1828 xprt_free_bc_request(req);
1831 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1833 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1835 struct xdr_buf *xbufp = &req->rq_snd_buf;
1837 task->tk_rqstp = req;
1838 req->rq_task = task;
1839 xprt_init_connect_cookie(req, req->rq_xprt);
1841 * Set up the xdr_buf length.
1842 * This also indicates that the buffer is XDR encoded already.
1844 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1845 xbufp->tail[0].iov_len;
1849 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1851 kref_init(&xprt->kref);
1853 spin_lock_init(&xprt->transport_lock);
1854 spin_lock_init(&xprt->reserve_lock);
1855 spin_lock_init(&xprt->queue_lock);
1857 INIT_LIST_HEAD(&xprt->free);
1858 xprt->recv_queue = RB_ROOT;
1859 INIT_LIST_HEAD(&xprt->xmit_queue);
1860 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1861 spin_lock_init(&xprt->bc_pa_lock);
1862 INIT_LIST_HEAD(&xprt->bc_pa_list);
1863 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1864 INIT_LIST_HEAD(&xprt->xprt_switch);
1866 xprt->last_used = jiffies;
1867 xprt->cwnd = RPC_INITCWND;
1868 xprt->bind_index = 0;
1870 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1871 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1872 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1873 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1875 xprt_init_xid(xprt);
1877 xprt->xprt_net = get_net(net);
1881 * xprt_create_transport - create an RPC transport
1882 * @args: rpc transport creation arguments
1885 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1887 struct rpc_xprt *xprt;
1888 struct xprt_class *t;
1890 spin_lock(&xprt_list_lock);
1891 list_for_each_entry(t, &xprt_list, list) {
1892 if (t->ident == args->ident) {
1893 spin_unlock(&xprt_list_lock);
1897 spin_unlock(&xprt_list_lock);
1898 dprintk("RPC: transport (%d) not supported\n", args->ident);
1899 return ERR_PTR(-EIO);
1902 xprt = t->setup(args);
1904 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1908 if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1909 xprt->idle_timeout = 0;
1910 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1911 if (xprt_has_timer(xprt))
1912 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1914 timer_setup(&xprt->timer, NULL, 0);
1916 if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1918 return ERR_PTR(-EINVAL);
1920 xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1921 if (xprt->servername == NULL) {
1923 return ERR_PTR(-ENOMEM);
1926 rpc_xprt_debugfs_register(xprt);
1928 dprintk("RPC: created transport %p with %u slots\n", xprt,
1934 static void xprt_destroy_cb(struct work_struct *work)
1936 struct rpc_xprt *xprt =
1937 container_of(work, struct rpc_xprt, task_cleanup);
1939 rpc_xprt_debugfs_unregister(xprt);
1940 rpc_destroy_wait_queue(&xprt->binding);
1941 rpc_destroy_wait_queue(&xprt->pending);
1942 rpc_destroy_wait_queue(&xprt->sending);
1943 rpc_destroy_wait_queue(&xprt->backlog);
1944 kfree(xprt->servername);
1946 * Tear down transport state and free the rpc_xprt
1948 xprt->ops->destroy(xprt);
1952 * xprt_destroy - destroy an RPC transport, killing off all requests.
1953 * @xprt: transport to destroy
1956 static void xprt_destroy(struct rpc_xprt *xprt)
1958 dprintk("RPC: destroying transport %p\n", xprt);
1961 * Exclude transport connect/disconnect handlers and autoclose
1963 wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1965 del_timer_sync(&xprt->timer);
1968 * Destroy sockets etc from the system workqueue so they can
1969 * safely flush receive work running on rpciod.
1971 INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1972 schedule_work(&xprt->task_cleanup);
1975 static void xprt_destroy_kref(struct kref *kref)
1977 xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1981 * xprt_get - return a reference to an RPC transport.
1982 * @xprt: pointer to the transport
1985 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1987 if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1991 EXPORT_SYMBOL_GPL(xprt_get);
1994 * xprt_put - release a reference to an RPC transport.
1995 * @xprt: pointer to the transport
1998 void xprt_put(struct rpc_xprt *xprt)
2001 kref_put(&xprt->kref, xprt_destroy_kref);
2003 EXPORT_SYMBOL_GPL(xprt_put);