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);
73 static void xprt_request_init(struct rpc_task *task);
75 static DEFINE_SPINLOCK(xprt_list_lock);
76 static LIST_HEAD(xprt_list);
78 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
80 unsigned long timeout = jiffies + req->rq_timeout;
82 if (time_before(timeout, req->rq_majortimeo))
84 return req->rq_majortimeo;
88 * xprt_register_transport - register a transport implementation
89 * @transport: transport to register
91 * If a transport implementation is loaded as a kernel module, it can
92 * call this interface to make itself known to the RPC client.
95 * 0: transport successfully registered
96 * -EEXIST: transport already registered
97 * -EINVAL: transport module being unloaded
99 int xprt_register_transport(struct xprt_class *transport)
101 struct xprt_class *t;
105 spin_lock(&xprt_list_lock);
106 list_for_each_entry(t, &xprt_list, list) {
107 /* don't register the same transport class twice */
108 if (t->ident == transport->ident)
112 list_add_tail(&transport->list, &xprt_list);
113 printk(KERN_INFO "RPC: Registered %s transport module.\n",
118 spin_unlock(&xprt_list_lock);
121 EXPORT_SYMBOL_GPL(xprt_register_transport);
124 * xprt_unregister_transport - unregister a transport implementation
125 * @transport: transport to unregister
128 * 0: transport successfully unregistered
129 * -ENOENT: transport never registered
131 int xprt_unregister_transport(struct xprt_class *transport)
133 struct xprt_class *t;
137 spin_lock(&xprt_list_lock);
138 list_for_each_entry(t, &xprt_list, list) {
139 if (t == transport) {
141 "RPC: Unregistered %s transport module.\n",
143 list_del_init(&transport->list);
150 spin_unlock(&xprt_list_lock);
153 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
156 xprt_class_release(const struct xprt_class *t)
158 module_put(t->owner);
161 static const struct xprt_class *
162 xprt_class_find_by_netid_locked(const char *netid)
164 const struct xprt_class *t;
167 list_for_each_entry(t, &xprt_list, list) {
168 for (i = 0; t->netid[i][0] != '\0'; i++) {
169 if (strcmp(t->netid[i], netid) != 0)
171 if (!try_module_get(t->owner))
179 static const struct xprt_class *
180 xprt_class_find_by_netid(const char *netid)
182 const struct xprt_class *t;
184 spin_lock(&xprt_list_lock);
185 t = xprt_class_find_by_netid_locked(netid);
187 spin_unlock(&xprt_list_lock);
188 request_module("rpc%s", netid);
189 spin_lock(&xprt_list_lock);
190 t = xprt_class_find_by_netid_locked(netid);
192 spin_unlock(&xprt_list_lock);
197 * xprt_load_transport - load a transport implementation
198 * @netid: transport to load
201 * 0: transport successfully loaded
202 * -ENOENT: transport module not available
204 int xprt_load_transport(const char *netid)
206 const struct xprt_class *t;
208 t = xprt_class_find_by_netid(netid);
211 xprt_class_release(t);
214 EXPORT_SYMBOL_GPL(xprt_load_transport);
216 static void xprt_clear_locked(struct rpc_xprt *xprt)
218 xprt->snd_task = NULL;
219 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
220 smp_mb__before_atomic();
221 clear_bit(XPRT_LOCKED, &xprt->state);
222 smp_mb__after_atomic();
224 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
228 * xprt_reserve_xprt - serialize write access to transports
229 * @task: task that is requesting access to the transport
230 * @xprt: pointer to the target transport
232 * This prevents mixing the payload of separate requests, and prevents
233 * transport connects from colliding with writes. No congestion control
236 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
238 struct rpc_rqst *req = task->tk_rqstp;
240 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
241 if (task == xprt->snd_task)
245 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
247 xprt->snd_task = task;
250 trace_xprt_reserve_xprt(xprt, task);
254 xprt_clear_locked(xprt);
256 task->tk_status = -EAGAIN;
257 if (RPC_IS_SOFT(task))
258 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
259 xprt_request_timeout(req));
261 rpc_sleep_on(&xprt->sending, task, NULL);
264 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
267 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
269 return test_bit(XPRT_CWND_WAIT, &xprt->state);
273 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
275 if (!list_empty(&xprt->xmit_queue)) {
276 /* Peek at head of queue to see if it can make progress */
277 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
281 set_bit(XPRT_CWND_WAIT, &xprt->state);
285 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
287 if (!RPCXPRT_CONGESTED(xprt))
288 clear_bit(XPRT_CWND_WAIT, &xprt->state);
292 * xprt_reserve_xprt_cong - serialize write access to transports
293 * @task: task that is requesting access to the transport
295 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
296 * integrated into the decision of whether a request is allowed to be
297 * woken up and given access to the transport.
298 * Note that the lock is only granted if we know there are free slots.
300 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
302 struct rpc_rqst *req = task->tk_rqstp;
304 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
305 if (task == xprt->snd_task)
310 xprt->snd_task = task;
313 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
315 if (!xprt_need_congestion_window_wait(xprt)) {
316 xprt->snd_task = task;
320 xprt_clear_locked(xprt);
322 task->tk_status = -EAGAIN;
323 if (RPC_IS_SOFT(task))
324 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
325 xprt_request_timeout(req));
327 rpc_sleep_on(&xprt->sending, task, NULL);
330 trace_xprt_reserve_cong(xprt, task);
333 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
335 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
339 if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
341 spin_lock(&xprt->transport_lock);
342 retval = xprt->ops->reserve_xprt(xprt, task);
343 spin_unlock(&xprt->transport_lock);
347 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
349 struct rpc_xprt *xprt = data;
351 xprt->snd_task = task;
355 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
357 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
359 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
361 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
362 __xprt_lock_write_func, xprt))
365 xprt_clear_locked(xprt);
368 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
370 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
372 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
374 if (xprt_need_congestion_window_wait(xprt))
376 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
377 __xprt_lock_write_func, xprt))
380 xprt_clear_locked(xprt);
384 * xprt_release_xprt - allow other requests to use a transport
385 * @xprt: transport with other tasks potentially waiting
386 * @task: task that is releasing access to the transport
388 * Note that "task" can be NULL. No congestion control is provided.
390 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
392 if (xprt->snd_task == task) {
393 xprt_clear_locked(xprt);
394 __xprt_lock_write_next(xprt);
396 trace_xprt_release_xprt(xprt, task);
398 EXPORT_SYMBOL_GPL(xprt_release_xprt);
401 * xprt_release_xprt_cong - allow other requests to use a transport
402 * @xprt: transport with other tasks potentially waiting
403 * @task: task that is releasing access to the transport
405 * Note that "task" can be NULL. Another task is awoken to use the
406 * transport if the transport's congestion window allows it.
408 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
410 if (xprt->snd_task == task) {
411 xprt_clear_locked(xprt);
412 __xprt_lock_write_next_cong(xprt);
414 trace_xprt_release_cong(xprt, task);
416 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
418 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
420 if (xprt->snd_task != task)
422 spin_lock(&xprt->transport_lock);
423 xprt->ops->release_xprt(xprt, task);
424 spin_unlock(&xprt->transport_lock);
428 * Van Jacobson congestion avoidance. Check if the congestion window
429 * overflowed. Put the task to sleep if this is the case.
432 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
436 trace_xprt_get_cong(xprt, req->rq_task);
437 if (RPCXPRT_CONGESTED(xprt)) {
438 xprt_set_congestion_window_wait(xprt);
442 xprt->cong += RPC_CWNDSCALE;
447 * Adjust the congestion window, and wake up the next task
448 * that has been sleeping due to congestion
451 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
456 xprt->cong -= RPC_CWNDSCALE;
457 xprt_test_and_clear_congestion_window_wait(xprt);
458 trace_xprt_put_cong(xprt, req->rq_task);
459 __xprt_lock_write_next_cong(xprt);
463 * xprt_request_get_cong - Request congestion control credits
464 * @xprt: pointer to transport
465 * @req: pointer to RPC request
467 * Useful for transports that require congestion control.
470 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
476 spin_lock(&xprt->transport_lock);
477 ret = __xprt_get_cong(xprt, req) != 0;
478 spin_unlock(&xprt->transport_lock);
481 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
484 * xprt_release_rqst_cong - housekeeping when request is complete
485 * @task: RPC request that recently completed
487 * Useful for transports that require congestion control.
489 void xprt_release_rqst_cong(struct rpc_task *task)
491 struct rpc_rqst *req = task->tk_rqstp;
493 __xprt_put_cong(req->rq_xprt, req);
495 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
497 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
499 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
500 __xprt_lock_write_next_cong(xprt);
504 * Clear the congestion window wait flag and wake up the next
505 * entry on xprt->sending
508 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
510 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
511 spin_lock(&xprt->transport_lock);
512 __xprt_lock_write_next_cong(xprt);
513 spin_unlock(&xprt->transport_lock);
518 * xprt_adjust_cwnd - adjust transport congestion window
519 * @xprt: pointer to xprt
520 * @task: recently completed RPC request used to adjust window
521 * @result: result code of completed RPC request
523 * The transport code maintains an estimate on the maximum number of out-
524 * standing RPC requests, using a smoothed version of the congestion
525 * avoidance implemented in 44BSD. This is basically the Van Jacobson
526 * congestion algorithm: If a retransmit occurs, the congestion window is
527 * halved; otherwise, it is incremented by 1/cwnd when
529 * - a reply is received and
530 * - a full number of requests are outstanding and
531 * - the congestion window hasn't been updated recently.
533 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
535 struct rpc_rqst *req = task->tk_rqstp;
536 unsigned long cwnd = xprt->cwnd;
538 if (result >= 0 && cwnd <= xprt->cong) {
539 /* The (cwnd >> 1) term makes sure
540 * the result gets rounded properly. */
541 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
542 if (cwnd > RPC_MAXCWND(xprt))
543 cwnd = RPC_MAXCWND(xprt);
544 __xprt_lock_write_next_cong(xprt);
545 } else if (result == -ETIMEDOUT) {
547 if (cwnd < RPC_CWNDSCALE)
548 cwnd = RPC_CWNDSCALE;
550 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
551 xprt->cong, xprt->cwnd, cwnd);
553 __xprt_put_cong(xprt, req);
555 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
558 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
559 * @xprt: transport with waiting tasks
560 * @status: result code to plant in each task before waking it
563 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
566 rpc_wake_up_status(&xprt->pending, status);
568 rpc_wake_up(&xprt->pending);
570 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
573 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
576 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
577 * we don't in general want to force a socket disconnection due to
578 * an incomplete RPC call transmission.
580 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
582 set_bit(XPRT_WRITE_SPACE, &xprt->state);
584 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
587 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
589 if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
590 __xprt_lock_write_next(xprt);
591 dprintk("RPC: write space: waking waiting task on "
599 * xprt_write_space - wake the task waiting for transport output buffer space
600 * @xprt: transport with waiting tasks
602 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
604 bool xprt_write_space(struct rpc_xprt *xprt)
608 if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
610 spin_lock(&xprt->transport_lock);
611 ret = xprt_clear_write_space_locked(xprt);
612 spin_unlock(&xprt->transport_lock);
615 EXPORT_SYMBOL_GPL(xprt_write_space);
617 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
619 s64 delta = ktime_to_ns(ktime_get() - abstime);
620 return likely(delta >= 0) ?
621 jiffies - nsecs_to_jiffies(delta) :
622 jiffies + nsecs_to_jiffies(-delta);
625 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
627 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
628 unsigned long majortimeo = req->rq_timeout;
630 if (to->to_exponential)
631 majortimeo <<= to->to_retries;
633 majortimeo += to->to_increment * to->to_retries;
634 if (majortimeo > to->to_maxval || majortimeo == 0)
635 majortimeo = to->to_maxval;
639 static void xprt_reset_majortimeo(struct rpc_rqst *req)
641 req->rq_majortimeo += xprt_calc_majortimeo(req);
644 static void xprt_reset_minortimeo(struct rpc_rqst *req)
646 req->rq_minortimeo += req->rq_timeout;
649 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
651 unsigned long time_init;
652 struct rpc_xprt *xprt = req->rq_xprt;
654 if (likely(xprt && xprt_connected(xprt)))
657 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
658 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
659 req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
660 req->rq_minortimeo = time_init + req->rq_timeout;
664 * xprt_adjust_timeout - adjust timeout values for next retransmit
665 * @req: RPC request containing parameters to use for the adjustment
668 int xprt_adjust_timeout(struct rpc_rqst *req)
670 struct rpc_xprt *xprt = req->rq_xprt;
671 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
674 if (time_before(jiffies, req->rq_majortimeo)) {
675 if (time_before(jiffies, req->rq_minortimeo))
677 if (to->to_exponential)
678 req->rq_timeout <<= 1;
680 req->rq_timeout += to->to_increment;
681 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
682 req->rq_timeout = to->to_maxval;
685 req->rq_timeout = to->to_initval;
687 xprt_reset_majortimeo(req);
688 /* Reset the RTT counters == "slow start" */
689 spin_lock(&xprt->transport_lock);
690 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
691 spin_unlock(&xprt->transport_lock);
694 xprt_reset_minortimeo(req);
696 if (req->rq_timeout == 0) {
697 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
698 req->rq_timeout = 5 * HZ;
703 static void xprt_autoclose(struct work_struct *work)
705 struct rpc_xprt *xprt =
706 container_of(work, struct rpc_xprt, task_cleanup);
707 unsigned int pflags = memalloc_nofs_save();
709 trace_xprt_disconnect_auto(xprt);
710 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
711 xprt->ops->close(xprt);
712 xprt_release_write(xprt, NULL);
713 wake_up_bit(&xprt->state, XPRT_LOCKED);
714 memalloc_nofs_restore(pflags);
718 * xprt_disconnect_done - mark a transport as disconnected
719 * @xprt: transport to flag for disconnect
722 void xprt_disconnect_done(struct rpc_xprt *xprt)
724 trace_xprt_disconnect_done(xprt);
725 spin_lock(&xprt->transport_lock);
726 xprt_clear_connected(xprt);
727 xprt_clear_write_space_locked(xprt);
728 xprt_clear_congestion_window_wait_locked(xprt);
729 xprt_wake_pending_tasks(xprt, -ENOTCONN);
730 spin_unlock(&xprt->transport_lock);
732 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
735 * xprt_schedule_autoclose_locked - Try to schedule an autoclose RPC call
736 * @xprt: transport to disconnect
738 static void xprt_schedule_autoclose_locked(struct rpc_xprt *xprt)
740 if (test_and_set_bit(XPRT_CLOSE_WAIT, &xprt->state))
742 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
743 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
744 else if (xprt->snd_task && !test_bit(XPRT_SND_IS_COOKIE, &xprt->state))
745 rpc_wake_up_queued_task_set_status(&xprt->pending,
746 xprt->snd_task, -ENOTCONN);
750 * xprt_force_disconnect - force a transport to disconnect
751 * @xprt: transport to disconnect
754 void xprt_force_disconnect(struct rpc_xprt *xprt)
756 trace_xprt_disconnect_force(xprt);
758 /* Don't race with the test_bit() in xprt_clear_locked() */
759 spin_lock(&xprt->transport_lock);
760 xprt_schedule_autoclose_locked(xprt);
761 spin_unlock(&xprt->transport_lock);
763 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
766 xprt_connect_cookie(struct rpc_xprt *xprt)
768 return READ_ONCE(xprt->connect_cookie);
772 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
774 struct rpc_rqst *req = task->tk_rqstp;
775 struct rpc_xprt *xprt = req->rq_xprt;
777 return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
778 !xprt_connected(xprt);
782 * xprt_conditional_disconnect - force a transport to disconnect
783 * @xprt: transport to disconnect
784 * @cookie: 'connection cookie'
786 * This attempts to break the connection if and only if 'cookie' matches
787 * the current transport 'connection cookie'. It ensures that we don't
788 * try to break the connection more than once when we need to retransmit
789 * a batch of RPC requests.
792 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
794 /* Don't race with the test_bit() in xprt_clear_locked() */
795 spin_lock(&xprt->transport_lock);
796 if (cookie != xprt->connect_cookie)
798 if (test_bit(XPRT_CLOSING, &xprt->state))
800 xprt_schedule_autoclose_locked(xprt);
802 spin_unlock(&xprt->transport_lock);
806 xprt_has_timer(const struct rpc_xprt *xprt)
808 return xprt->idle_timeout != 0;
812 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
813 __must_hold(&xprt->transport_lock)
815 xprt->last_used = jiffies;
816 if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
817 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
821 xprt_init_autodisconnect(struct timer_list *t)
823 struct rpc_xprt *xprt = from_timer(xprt, t, timer);
825 if (!RB_EMPTY_ROOT(&xprt->recv_queue))
827 /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
828 xprt->last_used = jiffies;
829 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
831 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
834 bool xprt_lock_connect(struct rpc_xprt *xprt,
835 struct rpc_task *task,
840 spin_lock(&xprt->transport_lock);
841 if (!test_bit(XPRT_LOCKED, &xprt->state))
843 if (xprt->snd_task != task)
845 set_bit(XPRT_SND_IS_COOKIE, &xprt->state);
846 xprt->snd_task = cookie;
849 spin_unlock(&xprt->transport_lock);
852 EXPORT_SYMBOL_GPL(xprt_lock_connect);
854 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
856 spin_lock(&xprt->transport_lock);
857 if (xprt->snd_task != cookie)
859 if (!test_bit(XPRT_LOCKED, &xprt->state))
861 xprt->snd_task =NULL;
862 clear_bit(XPRT_SND_IS_COOKIE, &xprt->state);
863 xprt->ops->release_xprt(xprt, NULL);
864 xprt_schedule_autodisconnect(xprt);
866 spin_unlock(&xprt->transport_lock);
867 wake_up_bit(&xprt->state, XPRT_LOCKED);
869 EXPORT_SYMBOL_GPL(xprt_unlock_connect);
872 * xprt_connect - schedule a transport connect operation
873 * @task: RPC task that is requesting the connect
876 void xprt_connect(struct rpc_task *task)
878 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
880 trace_xprt_connect(xprt);
882 if (!xprt_bound(xprt)) {
883 task->tk_status = -EAGAIN;
886 if (!xprt_lock_write(xprt, task))
889 if (!xprt_connected(xprt) && !test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
890 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
891 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
892 xprt_request_timeout(task->tk_rqstp));
894 if (test_bit(XPRT_CLOSING, &xprt->state))
896 if (xprt_test_and_set_connecting(xprt))
899 if (!xprt_connected(xprt)) {
900 xprt->stat.connect_start = jiffies;
901 xprt->ops->connect(xprt, task);
903 xprt_clear_connecting(xprt);
905 rpc_wake_up_queued_task(&xprt->pending, task);
908 xprt_release_write(xprt, task);
912 * xprt_reconnect_delay - compute the wait before scheduling a connect
913 * @xprt: transport instance
916 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
918 unsigned long start, now = jiffies;
920 start = xprt->stat.connect_start + xprt->reestablish_timeout;
921 if (time_after(start, now))
925 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
928 * xprt_reconnect_backoff - compute the new re-establish timeout
929 * @xprt: transport instance
930 * @init_to: initial reestablish timeout
933 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
935 xprt->reestablish_timeout <<= 1;
936 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
937 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
938 if (xprt->reestablish_timeout < init_to)
939 xprt->reestablish_timeout = init_to;
941 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
943 enum xprt_xid_rb_cmp {
948 static enum xprt_xid_rb_cmp
949 xprt_xid_cmp(__be32 xid1, __be32 xid2)
953 if ((__force u32)xid1 < (__force u32)xid2)
958 static struct rpc_rqst *
959 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
961 struct rb_node *n = xprt->recv_queue.rb_node;
962 struct rpc_rqst *req;
965 req = rb_entry(n, struct rpc_rqst, rq_recv);
966 switch (xprt_xid_cmp(xid, req->rq_xid)) {
981 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
983 struct rb_node **p = &xprt->recv_queue.rb_node;
984 struct rb_node *n = NULL;
985 struct rpc_rqst *req;
989 req = rb_entry(n, struct rpc_rqst, rq_recv);
990 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
998 WARN_ON_ONCE(new != req);
1002 rb_link_node(&new->rq_recv, n, p);
1003 rb_insert_color(&new->rq_recv, &xprt->recv_queue);
1007 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
1009 rb_erase(&req->rq_recv, &xprt->recv_queue);
1013 * xprt_lookup_rqst - find an RPC request corresponding to an XID
1014 * @xprt: transport on which the original request was transmitted
1015 * @xid: RPC XID of incoming reply
1017 * Caller holds xprt->queue_lock.
1019 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
1021 struct rpc_rqst *entry;
1023 entry = xprt_request_rb_find(xprt, xid);
1024 if (entry != NULL) {
1025 trace_xprt_lookup_rqst(xprt, xid, 0);
1026 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
1030 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
1032 trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
1033 xprt->stat.bad_xids++;
1036 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1039 xprt_is_pinned_rqst(struct rpc_rqst *req)
1041 return atomic_read(&req->rq_pin) != 0;
1045 * xprt_pin_rqst - Pin a request on the transport receive list
1046 * @req: Request to pin
1048 * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1049 * so should be holding xprt->queue_lock.
1051 void xprt_pin_rqst(struct rpc_rqst *req)
1053 atomic_inc(&req->rq_pin);
1055 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1058 * xprt_unpin_rqst - Unpin a request on the transport receive list
1059 * @req: Request to pin
1061 * Caller should be holding xprt->queue_lock.
1063 void xprt_unpin_rqst(struct rpc_rqst *req)
1065 if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1066 atomic_dec(&req->rq_pin);
1069 if (atomic_dec_and_test(&req->rq_pin))
1070 wake_up_var(&req->rq_pin);
1072 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1074 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1076 wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1080 xprt_request_data_received(struct rpc_task *task)
1082 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1083 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1087 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1089 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1090 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1094 * xprt_request_enqueue_receive - Add an request to the receive queue
1099 xprt_request_enqueue_receive(struct rpc_task *task)
1101 struct rpc_rqst *req = task->tk_rqstp;
1102 struct rpc_xprt *xprt = req->rq_xprt;
1104 if (!xprt_request_need_enqueue_receive(task, req))
1107 xprt_request_prepare(task->tk_rqstp);
1108 spin_lock(&xprt->queue_lock);
1110 /* Update the softirq receive buffer */
1111 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1112 sizeof(req->rq_private_buf));
1114 /* Add request to the receive list */
1115 xprt_request_rb_insert(xprt, req);
1116 set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1117 spin_unlock(&xprt->queue_lock);
1119 /* Turn off autodisconnect */
1120 del_singleshot_timer_sync(&xprt->timer);
1124 * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1127 * Caller must hold xprt->queue_lock.
1130 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1132 struct rpc_rqst *req = task->tk_rqstp;
1134 if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1135 xprt_request_rb_remove(req->rq_xprt, req);
1139 * xprt_update_rtt - Update RPC RTT statistics
1140 * @task: RPC request that recently completed
1142 * Caller holds xprt->queue_lock.
1144 void xprt_update_rtt(struct rpc_task *task)
1146 struct rpc_rqst *req = task->tk_rqstp;
1147 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1148 unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1149 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1152 if (req->rq_ntrans == 1)
1153 rpc_update_rtt(rtt, timer, m);
1154 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1157 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1160 * xprt_complete_rqst - called when reply processing is complete
1161 * @task: RPC request that recently completed
1162 * @copied: actual number of bytes received from the transport
1164 * Caller holds xprt->queue_lock.
1166 void xprt_complete_rqst(struct rpc_task *task, int copied)
1168 struct rpc_rqst *req = task->tk_rqstp;
1169 struct rpc_xprt *xprt = req->rq_xprt;
1173 req->rq_private_buf.len = copied;
1174 /* Ensure all writes are done before we update */
1175 /* req->rq_reply_bytes_recvd */
1177 req->rq_reply_bytes_recvd = copied;
1178 xprt_request_dequeue_receive_locked(task);
1179 rpc_wake_up_queued_task(&xprt->pending, task);
1181 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1183 static void xprt_timer(struct rpc_task *task)
1185 struct rpc_rqst *req = task->tk_rqstp;
1186 struct rpc_xprt *xprt = req->rq_xprt;
1188 if (task->tk_status != -ETIMEDOUT)
1191 trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1192 if (!req->rq_reply_bytes_recvd) {
1193 if (xprt->ops->timer)
1194 xprt->ops->timer(xprt, task);
1196 task->tk_status = 0;
1200 * xprt_wait_for_reply_request_def - wait for reply
1201 * @task: pointer to rpc_task
1203 * Set a request's retransmit timeout based on the transport's
1204 * default timeout parameters. Used by transports that don't adjust
1205 * the retransmit timeout based on round-trip time estimation,
1206 * and put the task to sleep on the pending queue.
1208 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1210 struct rpc_rqst *req = task->tk_rqstp;
1212 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1213 xprt_request_timeout(req));
1215 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1218 * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1219 * @task: pointer to rpc_task
1221 * Set a request's retransmit timeout using the RTT estimator,
1222 * and put the task to sleep on the pending queue.
1224 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1226 int timer = task->tk_msg.rpc_proc->p_timer;
1227 struct rpc_clnt *clnt = task->tk_client;
1228 struct rpc_rtt *rtt = clnt->cl_rtt;
1229 struct rpc_rqst *req = task->tk_rqstp;
1230 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1231 unsigned long timeout;
1233 timeout = rpc_calc_rto(rtt, timer);
1234 timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1235 if (timeout > max_timeout || timeout == 0)
1236 timeout = max_timeout;
1237 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1240 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1243 * xprt_request_wait_receive - wait for the reply to an RPC request
1244 * @task: RPC task about to send a request
1247 void xprt_request_wait_receive(struct rpc_task *task)
1249 struct rpc_rqst *req = task->tk_rqstp;
1250 struct rpc_xprt *xprt = req->rq_xprt;
1252 if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1255 * Sleep on the pending queue if we're expecting a reply.
1256 * The spinlock ensures atomicity between the test of
1257 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1259 spin_lock(&xprt->queue_lock);
1260 if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1261 xprt->ops->wait_for_reply_request(task);
1263 * Send an extra queue wakeup call if the
1264 * connection was dropped in case the call to
1265 * rpc_sleep_on() raced.
1267 if (xprt_request_retransmit_after_disconnect(task))
1268 rpc_wake_up_queued_task_set_status(&xprt->pending,
1271 spin_unlock(&xprt->queue_lock);
1275 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1277 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1281 * xprt_request_enqueue_transmit - queue a task for transmission
1282 * @task: pointer to rpc_task
1284 * Add a task to the transmission queue.
1287 xprt_request_enqueue_transmit(struct rpc_task *task)
1289 struct rpc_rqst *pos, *req = task->tk_rqstp;
1290 struct rpc_xprt *xprt = req->rq_xprt;
1292 if (xprt_request_need_enqueue_transmit(task, req)) {
1293 req->rq_bytes_sent = 0;
1294 spin_lock(&xprt->queue_lock);
1296 * Requests that carry congestion control credits are added
1297 * to the head of the list to avoid starvation issues.
1300 xprt_clear_congestion_window_wait(xprt);
1301 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1304 /* Note: req is added _before_ pos */
1305 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1306 INIT_LIST_HEAD(&req->rq_xmit2);
1309 } else if (!req->rq_seqno) {
1310 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1311 if (pos->rq_task->tk_owner != task->tk_owner)
1313 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1314 INIT_LIST_HEAD(&req->rq_xmit);
1318 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1319 INIT_LIST_HEAD(&req->rq_xmit2);
1321 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1322 spin_unlock(&xprt->queue_lock);
1327 * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1328 * @task: pointer to rpc_task
1330 * Remove a task from the transmission queue
1331 * Caller must hold xprt->queue_lock
1334 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1336 struct rpc_rqst *req = task->tk_rqstp;
1338 if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1340 if (!list_empty(&req->rq_xmit)) {
1341 list_del(&req->rq_xmit);
1342 if (!list_empty(&req->rq_xmit2)) {
1343 struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1344 struct rpc_rqst, rq_xmit2);
1345 list_del(&req->rq_xmit2);
1346 list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1349 list_del(&req->rq_xmit2);
1353 * xprt_request_dequeue_transmit - remove a task from the transmission queue
1354 * @task: pointer to rpc_task
1356 * Remove a task from the transmission queue
1359 xprt_request_dequeue_transmit(struct rpc_task *task)
1361 struct rpc_rqst *req = task->tk_rqstp;
1362 struct rpc_xprt *xprt = req->rq_xprt;
1364 spin_lock(&xprt->queue_lock);
1365 xprt_request_dequeue_transmit_locked(task);
1366 spin_unlock(&xprt->queue_lock);
1370 * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1371 * @task: pointer to rpc_task
1373 * Remove a task from the transmit and receive queues, and ensure that
1374 * it is not pinned by the receive work item.
1377 xprt_request_dequeue_xprt(struct rpc_task *task)
1379 struct rpc_rqst *req = task->tk_rqstp;
1380 struct rpc_xprt *xprt = req->rq_xprt;
1382 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1383 test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1384 xprt_is_pinned_rqst(req)) {
1385 spin_lock(&xprt->queue_lock);
1386 xprt_request_dequeue_transmit_locked(task);
1387 xprt_request_dequeue_receive_locked(task);
1388 while (xprt_is_pinned_rqst(req)) {
1389 set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1390 spin_unlock(&xprt->queue_lock);
1391 xprt_wait_on_pinned_rqst(req);
1392 spin_lock(&xprt->queue_lock);
1393 clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1395 spin_unlock(&xprt->queue_lock);
1400 * xprt_request_prepare - prepare an encoded request for transport
1401 * @req: pointer to rpc_rqst
1403 * Calls into the transport layer to do whatever is needed to prepare
1404 * the request for transmission or receive.
1407 xprt_request_prepare(struct rpc_rqst *req)
1409 struct rpc_xprt *xprt = req->rq_xprt;
1411 if (xprt->ops->prepare_request)
1412 xprt->ops->prepare_request(req);
1416 * xprt_request_need_retransmit - Test if a task needs retransmission
1417 * @task: pointer to rpc_task
1419 * Test for whether a connection breakage requires the task to retransmit
1422 xprt_request_need_retransmit(struct rpc_task *task)
1424 return xprt_request_retransmit_after_disconnect(task);
1428 * xprt_prepare_transmit - reserve the transport before sending a request
1429 * @task: RPC task about to send a request
1432 bool xprt_prepare_transmit(struct rpc_task *task)
1434 struct rpc_rqst *req = task->tk_rqstp;
1435 struct rpc_xprt *xprt = req->rq_xprt;
1437 if (!xprt_lock_write(xprt, task)) {
1438 /* Race breaker: someone may have transmitted us */
1439 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1440 rpc_wake_up_queued_task_set_status(&xprt->sending,
1448 void xprt_end_transmit(struct rpc_task *task)
1450 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1452 xprt_inject_disconnect(xprt);
1453 xprt_release_write(xprt, task);
1457 * xprt_request_transmit - send an RPC request on a transport
1458 * @req: pointer to request to transmit
1459 * @snd_task: RPC task that owns the transport lock
1461 * This performs the transmission of a single request.
1462 * Note that if the request is not the same as snd_task, then it
1463 * does need to be pinned.
1464 * Returns '0' on success.
1467 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1469 struct rpc_xprt *xprt = req->rq_xprt;
1470 struct rpc_task *task = req->rq_task;
1471 unsigned int connect_cookie;
1472 int is_retrans = RPC_WAS_SENT(task);
1475 if (!req->rq_bytes_sent) {
1476 if (xprt_request_data_received(task)) {
1480 /* Verify that our message lies in the RPCSEC_GSS window */
1481 if (rpcauth_xmit_need_reencode(task)) {
1485 if (RPC_SIGNALLED(task)) {
1486 status = -ERESTARTSYS;
1492 * Update req->rq_ntrans before transmitting to avoid races with
1493 * xprt_update_rtt(), which needs to know that it is recording a
1494 * reply to the first transmission.
1498 trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1499 connect_cookie = xprt->connect_cookie;
1500 status = xprt->ops->send_request(req);
1503 trace_xprt_transmit(req, status);
1508 task->tk_client->cl_stats->rpcretrans++;
1510 xprt_inject_disconnect(xprt);
1512 task->tk_flags |= RPC_TASK_SENT;
1513 spin_lock(&xprt->transport_lock);
1516 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1517 xprt->stat.bklog_u += xprt->backlog.qlen;
1518 xprt->stat.sending_u += xprt->sending.qlen;
1519 xprt->stat.pending_u += xprt->pending.qlen;
1520 spin_unlock(&xprt->transport_lock);
1522 req->rq_connect_cookie = connect_cookie;
1524 trace_xprt_transmit(req, status);
1525 xprt_request_dequeue_transmit(task);
1526 rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1531 * xprt_transmit - send an RPC request on a transport
1532 * @task: controlling RPC task
1534 * Attempts to drain the transmit queue. On exit, either the transport
1535 * signalled an error that needs to be handled before transmission can
1536 * resume, or @task finished transmitting, and detected that it already
1540 xprt_transmit(struct rpc_task *task)
1542 struct rpc_rqst *next, *req = task->tk_rqstp;
1543 struct rpc_xprt *xprt = req->rq_xprt;
1546 spin_lock(&xprt->queue_lock);
1548 next = list_first_entry_or_null(&xprt->xmit_queue,
1549 struct rpc_rqst, rq_xmit);
1552 xprt_pin_rqst(next);
1553 spin_unlock(&xprt->queue_lock);
1554 status = xprt_request_transmit(next, task);
1555 if (status == -EBADMSG && next != req)
1557 spin_lock(&xprt->queue_lock);
1558 xprt_unpin_rqst(next);
1560 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1561 task->tk_status = status;
1564 /* Was @task transmitted, and has it received a reply? */
1565 if (xprt_request_data_received(task) &&
1566 !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1568 cond_resched_lock(&xprt->queue_lock);
1570 spin_unlock(&xprt->queue_lock);
1573 static void xprt_complete_request_init(struct rpc_task *task)
1576 xprt_request_init(task);
1579 void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1581 set_bit(XPRT_CONGESTED, &xprt->state);
1582 rpc_sleep_on(&xprt->backlog, task, xprt_complete_request_init);
1584 EXPORT_SYMBOL_GPL(xprt_add_backlog);
1586 static bool __xprt_set_rq(struct rpc_task *task, void *data)
1588 struct rpc_rqst *req = data;
1590 if (task->tk_rqstp == NULL) {
1591 memset(req, 0, sizeof(*req)); /* mark unused */
1592 task->tk_rqstp = req;
1598 bool xprt_wake_up_backlog(struct rpc_xprt *xprt, struct rpc_rqst *req)
1600 if (rpc_wake_up_first(&xprt->backlog, __xprt_set_rq, req) == NULL) {
1601 clear_bit(XPRT_CONGESTED, &xprt->state);
1606 EXPORT_SYMBOL_GPL(xprt_wake_up_backlog);
1608 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1612 if (!test_bit(XPRT_CONGESTED, &xprt->state))
1614 spin_lock(&xprt->reserve_lock);
1615 if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1616 xprt_add_backlog(xprt, task);
1619 spin_unlock(&xprt->reserve_lock);
1624 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1626 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1627 gfp_t gfp_mask = GFP_KERNEL;
1629 if (xprt->num_reqs >= xprt->max_reqs)
1632 spin_unlock(&xprt->reserve_lock);
1633 if (current->flags & PF_WQ_WORKER)
1634 gfp_mask |= __GFP_NORETRY | __GFP_NOWARN;
1635 req = kzalloc(sizeof(*req), gfp_mask);
1636 spin_lock(&xprt->reserve_lock);
1640 req = ERR_PTR(-ENOMEM);
1645 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1647 if (xprt->num_reqs > xprt->min_reqs) {
1655 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1657 struct rpc_rqst *req;
1659 spin_lock(&xprt->reserve_lock);
1660 if (!list_empty(&xprt->free)) {
1661 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1662 list_del(&req->rq_list);
1665 req = xprt_dynamic_alloc_slot(xprt);
1668 switch (PTR_ERR(req)) {
1670 dprintk("RPC: dynamic allocation of request slot "
1671 "failed! Retrying\n");
1672 task->tk_status = -ENOMEM;
1675 xprt_add_backlog(xprt, task);
1676 dprintk("RPC: waiting for request slot\n");
1679 task->tk_status = -EAGAIN;
1681 spin_unlock(&xprt->reserve_lock);
1684 xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1686 spin_unlock(&xprt->reserve_lock);
1688 task->tk_status = 0;
1689 task->tk_rqstp = req;
1691 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1693 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1695 spin_lock(&xprt->reserve_lock);
1696 if (!xprt_wake_up_backlog(xprt, req) &&
1697 !xprt_dynamic_free_slot(xprt, req)) {
1698 memset(req, 0, sizeof(*req)); /* mark unused */
1699 list_add(&req->rq_list, &xprt->free);
1701 spin_unlock(&xprt->reserve_lock);
1703 EXPORT_SYMBOL_GPL(xprt_free_slot);
1705 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1707 struct rpc_rqst *req;
1708 while (!list_empty(&xprt->free)) {
1709 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1710 list_del(&req->rq_list);
1715 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1716 unsigned int num_prealloc,
1717 unsigned int max_alloc)
1719 struct rpc_xprt *xprt;
1720 struct rpc_rqst *req;
1723 xprt = kzalloc(size, GFP_KERNEL);
1727 xprt_init(xprt, net);
1729 for (i = 0; i < num_prealloc; i++) {
1730 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1733 list_add(&req->rq_list, &xprt->free);
1735 if (max_alloc > num_prealloc)
1736 xprt->max_reqs = max_alloc;
1738 xprt->max_reqs = num_prealloc;
1739 xprt->min_reqs = num_prealloc;
1740 xprt->num_reqs = num_prealloc;
1749 EXPORT_SYMBOL_GPL(xprt_alloc);
1751 void xprt_free(struct rpc_xprt *xprt)
1753 put_net(xprt->xprt_net);
1754 xprt_free_all_slots(xprt);
1755 kfree_rcu(xprt, rcu);
1757 EXPORT_SYMBOL_GPL(xprt_free);
1760 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1762 req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1766 xprt_alloc_xid(struct rpc_xprt *xprt)
1770 spin_lock(&xprt->reserve_lock);
1771 xid = (__force __be32)xprt->xid++;
1772 spin_unlock(&xprt->reserve_lock);
1777 xprt_init_xid(struct rpc_xprt *xprt)
1779 xprt->xid = prandom_u32();
1783 xprt_request_init(struct rpc_task *task)
1785 struct rpc_xprt *xprt = task->tk_xprt;
1786 struct rpc_rqst *req = task->tk_rqstp;
1788 req->rq_task = task;
1789 req->rq_xprt = xprt;
1790 req->rq_buffer = NULL;
1791 req->rq_xid = xprt_alloc_xid(xprt);
1792 xprt_init_connect_cookie(req, xprt);
1793 req->rq_snd_buf.len = 0;
1794 req->rq_snd_buf.buflen = 0;
1795 req->rq_rcv_buf.len = 0;
1796 req->rq_rcv_buf.buflen = 0;
1797 req->rq_snd_buf.bvec = NULL;
1798 req->rq_rcv_buf.bvec = NULL;
1799 req->rq_release_snd_buf = NULL;
1800 xprt_init_majortimeo(task, req);
1802 trace_xprt_reserve(req);
1806 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1808 xprt->ops->alloc_slot(xprt, task);
1809 if (task->tk_rqstp != NULL)
1810 xprt_request_init(task);
1814 * xprt_reserve - allocate an RPC request slot
1815 * @task: RPC task requesting a slot allocation
1817 * If the transport is marked as being congested, or if no more
1818 * slots are available, place the task on the transport's
1821 void xprt_reserve(struct rpc_task *task)
1823 struct rpc_xprt *xprt = task->tk_xprt;
1825 task->tk_status = 0;
1826 if (task->tk_rqstp != NULL)
1829 task->tk_status = -EAGAIN;
1830 if (!xprt_throttle_congested(xprt, task))
1831 xprt_do_reserve(xprt, task);
1835 * xprt_retry_reserve - allocate an RPC request slot
1836 * @task: RPC task requesting a slot allocation
1838 * If no more slots are available, place the task on the transport's
1840 * Note that the only difference with xprt_reserve is that we now
1841 * ignore the value of the XPRT_CONGESTED flag.
1843 void xprt_retry_reserve(struct rpc_task *task)
1845 struct rpc_xprt *xprt = task->tk_xprt;
1847 task->tk_status = 0;
1848 if (task->tk_rqstp != NULL)
1851 task->tk_status = -EAGAIN;
1852 xprt_do_reserve(xprt, task);
1856 * xprt_release - release an RPC request slot
1857 * @task: task which is finished with the slot
1860 void xprt_release(struct rpc_task *task)
1862 struct rpc_xprt *xprt;
1863 struct rpc_rqst *req = task->tk_rqstp;
1866 if (task->tk_client) {
1867 xprt = task->tk_xprt;
1868 xprt_release_write(xprt, task);
1873 xprt = req->rq_xprt;
1874 xprt_request_dequeue_xprt(task);
1875 spin_lock(&xprt->transport_lock);
1876 xprt->ops->release_xprt(xprt, task);
1877 if (xprt->ops->release_request)
1878 xprt->ops->release_request(task);
1879 xprt_schedule_autodisconnect(xprt);
1880 spin_unlock(&xprt->transport_lock);
1882 xprt->ops->buf_free(task);
1883 xdr_free_bvec(&req->rq_rcv_buf);
1884 xdr_free_bvec(&req->rq_snd_buf);
1885 if (req->rq_cred != NULL)
1886 put_rpccred(req->rq_cred);
1887 if (req->rq_release_snd_buf)
1888 req->rq_release_snd_buf(req);
1890 task->tk_rqstp = NULL;
1891 if (likely(!bc_prealloc(req)))
1892 xprt->ops->free_slot(xprt, req);
1894 xprt_free_bc_request(req);
1897 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1899 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1901 struct xdr_buf *xbufp = &req->rq_snd_buf;
1903 task->tk_rqstp = req;
1904 req->rq_task = task;
1905 xprt_init_connect_cookie(req, req->rq_xprt);
1907 * Set up the xdr_buf length.
1908 * This also indicates that the buffer is XDR encoded already.
1910 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1911 xbufp->tail[0].iov_len;
1915 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1917 kref_init(&xprt->kref);
1919 spin_lock_init(&xprt->transport_lock);
1920 spin_lock_init(&xprt->reserve_lock);
1921 spin_lock_init(&xprt->queue_lock);
1923 INIT_LIST_HEAD(&xprt->free);
1924 xprt->recv_queue = RB_ROOT;
1925 INIT_LIST_HEAD(&xprt->xmit_queue);
1926 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1927 spin_lock_init(&xprt->bc_pa_lock);
1928 INIT_LIST_HEAD(&xprt->bc_pa_list);
1929 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1930 INIT_LIST_HEAD(&xprt->xprt_switch);
1932 xprt->last_used = jiffies;
1933 xprt->cwnd = RPC_INITCWND;
1934 xprt->bind_index = 0;
1936 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1937 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1938 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1939 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1941 xprt_init_xid(xprt);
1943 xprt->xprt_net = get_net(net);
1947 * xprt_create_transport - create an RPC transport
1948 * @args: rpc transport creation arguments
1951 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1953 struct rpc_xprt *xprt;
1954 struct xprt_class *t;
1956 spin_lock(&xprt_list_lock);
1957 list_for_each_entry(t, &xprt_list, list) {
1958 if (t->ident == args->ident) {
1959 spin_unlock(&xprt_list_lock);
1963 spin_unlock(&xprt_list_lock);
1964 dprintk("RPC: transport (%d) not supported\n", args->ident);
1965 return ERR_PTR(-EIO);
1968 xprt = t->setup(args);
1971 if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1972 xprt->idle_timeout = 0;
1973 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1974 if (xprt_has_timer(xprt))
1975 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1977 timer_setup(&xprt->timer, NULL, 0);
1979 if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1981 return ERR_PTR(-EINVAL);
1983 xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1984 if (xprt->servername == NULL) {
1986 return ERR_PTR(-ENOMEM);
1989 rpc_xprt_debugfs_register(xprt);
1991 trace_xprt_create(xprt);
1996 static void xprt_destroy_cb(struct work_struct *work)
1998 struct rpc_xprt *xprt =
1999 container_of(work, struct rpc_xprt, task_cleanup);
2001 trace_xprt_destroy(xprt);
2003 rpc_xprt_debugfs_unregister(xprt);
2004 rpc_destroy_wait_queue(&xprt->binding);
2005 rpc_destroy_wait_queue(&xprt->pending);
2006 rpc_destroy_wait_queue(&xprt->sending);
2007 rpc_destroy_wait_queue(&xprt->backlog);
2008 kfree(xprt->servername);
2010 * Destroy any existing back channel
2012 xprt_destroy_backchannel(xprt, UINT_MAX);
2015 * Tear down transport state and free the rpc_xprt
2017 xprt->ops->destroy(xprt);
2021 * xprt_destroy - destroy an RPC transport, killing off all requests.
2022 * @xprt: transport to destroy
2025 static void xprt_destroy(struct rpc_xprt *xprt)
2028 * Exclude transport connect/disconnect handlers and autoclose
2030 wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
2033 * xprt_schedule_autodisconnect() can run after XPRT_LOCKED
2034 * is cleared. We use ->transport_lock to ensure the mod_timer()
2035 * can only run *before* del_time_sync(), never after.
2037 spin_lock(&xprt->transport_lock);
2038 del_timer_sync(&xprt->timer);
2039 spin_unlock(&xprt->transport_lock);
2042 * Destroy sockets etc from the system workqueue so they can
2043 * safely flush receive work running on rpciod.
2045 INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
2046 schedule_work(&xprt->task_cleanup);
2049 static void xprt_destroy_kref(struct kref *kref)
2051 xprt_destroy(container_of(kref, struct rpc_xprt, kref));
2055 * xprt_get - return a reference to an RPC transport.
2056 * @xprt: pointer to the transport
2059 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2061 if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2065 EXPORT_SYMBOL_GPL(xprt_get);
2068 * xprt_put - release a reference to an RPC transport.
2069 * @xprt: pointer to the transport
2072 void xprt_put(struct rpc_xprt *xprt)
2075 kref_put(&xprt->kref, xprt_destroy_kref);
2077 EXPORT_SYMBOL_GPL(xprt_put);