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_class_release(const struct xprt_class *t)
157 module_put(t->owner);
160 static const struct xprt_class *
161 xprt_class_find_by_ident_locked(int ident)
163 const struct xprt_class *t;
165 list_for_each_entry(t, &xprt_list, list) {
166 if (t->ident != ident)
168 if (!try_module_get(t->owner))
175 static const struct xprt_class *
176 xprt_class_find_by_ident(int ident)
178 const struct xprt_class *t;
180 spin_lock(&xprt_list_lock);
181 t = xprt_class_find_by_ident_locked(ident);
182 spin_unlock(&xprt_list_lock);
186 static const struct xprt_class *
187 xprt_class_find_by_netid_locked(const char *netid)
189 const struct xprt_class *t;
192 list_for_each_entry(t, &xprt_list, list) {
193 for (i = 0; t->netid[i][0] != '\0'; i++) {
194 if (strcmp(t->netid[i], netid) != 0)
196 if (!try_module_get(t->owner))
204 static const struct xprt_class *
205 xprt_class_find_by_netid(const char *netid)
207 const struct xprt_class *t;
209 spin_lock(&xprt_list_lock);
210 t = xprt_class_find_by_netid_locked(netid);
212 spin_unlock(&xprt_list_lock);
213 request_module("rpc%s", netid);
214 spin_lock(&xprt_list_lock);
215 t = xprt_class_find_by_netid_locked(netid);
217 spin_unlock(&xprt_list_lock);
222 * xprt_find_transport_ident - convert a netid into a transport identifier
223 * @netid: transport to load
226 * > 0: transport identifier
227 * -ENOENT: transport module not available
229 int xprt_find_transport_ident(const char *netid)
231 const struct xprt_class *t;
234 t = xprt_class_find_by_netid(netid);
238 xprt_class_release(t);
241 EXPORT_SYMBOL_GPL(xprt_find_transport_ident);
243 static void xprt_clear_locked(struct rpc_xprt *xprt)
245 xprt->snd_task = NULL;
246 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
247 smp_mb__before_atomic();
248 clear_bit(XPRT_LOCKED, &xprt->state);
249 smp_mb__after_atomic();
251 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
255 * xprt_reserve_xprt - serialize write access to transports
256 * @task: task that is requesting access to the transport
257 * @xprt: pointer to the target transport
259 * This prevents mixing the payload of separate requests, and prevents
260 * transport connects from colliding with writes. No congestion control
263 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
265 struct rpc_rqst *req = task->tk_rqstp;
267 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
268 if (task == xprt->snd_task)
272 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
274 xprt->snd_task = task;
277 trace_xprt_reserve_xprt(xprt, task);
281 xprt_clear_locked(xprt);
283 task->tk_status = -EAGAIN;
284 if (RPC_IS_SOFT(task))
285 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
286 xprt_request_timeout(req));
288 rpc_sleep_on(&xprt->sending, task, NULL);
291 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
294 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
296 return test_bit(XPRT_CWND_WAIT, &xprt->state);
300 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
302 if (!list_empty(&xprt->xmit_queue)) {
303 /* Peek at head of queue to see if it can make progress */
304 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
308 set_bit(XPRT_CWND_WAIT, &xprt->state);
312 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
314 if (!RPCXPRT_CONGESTED(xprt))
315 clear_bit(XPRT_CWND_WAIT, &xprt->state);
319 * xprt_reserve_xprt_cong - serialize write access to transports
320 * @task: task that is requesting access to the transport
322 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
323 * integrated into the decision of whether a request is allowed to be
324 * woken up and given access to the transport.
325 * Note that the lock is only granted if we know there are free slots.
327 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
329 struct rpc_rqst *req = task->tk_rqstp;
331 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
332 if (task == xprt->snd_task)
337 xprt->snd_task = task;
340 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
342 if (!xprt_need_congestion_window_wait(xprt)) {
343 xprt->snd_task = task;
347 xprt_clear_locked(xprt);
349 task->tk_status = -EAGAIN;
350 if (RPC_IS_SOFT(task))
351 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
352 xprt_request_timeout(req));
354 rpc_sleep_on(&xprt->sending, task, NULL);
357 trace_xprt_reserve_cong(xprt, task);
360 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
362 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
366 if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
368 spin_lock(&xprt->transport_lock);
369 retval = xprt->ops->reserve_xprt(xprt, task);
370 spin_unlock(&xprt->transport_lock);
374 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
376 struct rpc_xprt *xprt = data;
378 xprt->snd_task = task;
382 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
384 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
386 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
388 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
389 __xprt_lock_write_func, xprt))
392 xprt_clear_locked(xprt);
395 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
397 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
399 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
401 if (xprt_need_congestion_window_wait(xprt))
403 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
404 __xprt_lock_write_func, xprt))
407 xprt_clear_locked(xprt);
411 * xprt_release_xprt - allow other requests to use a transport
412 * @xprt: transport with other tasks potentially waiting
413 * @task: task that is releasing access to the transport
415 * Note that "task" can be NULL. No congestion control is provided.
417 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
419 if (xprt->snd_task == task) {
420 xprt_clear_locked(xprt);
421 __xprt_lock_write_next(xprt);
423 trace_xprt_release_xprt(xprt, task);
425 EXPORT_SYMBOL_GPL(xprt_release_xprt);
428 * xprt_release_xprt_cong - allow other requests to use a transport
429 * @xprt: transport with other tasks potentially waiting
430 * @task: task that is releasing access to the transport
432 * Note that "task" can be NULL. Another task is awoken to use the
433 * transport if the transport's congestion window allows it.
435 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
437 if (xprt->snd_task == task) {
438 xprt_clear_locked(xprt);
439 __xprt_lock_write_next_cong(xprt);
441 trace_xprt_release_cong(xprt, task);
443 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
445 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
447 if (xprt->snd_task != task)
449 spin_lock(&xprt->transport_lock);
450 xprt->ops->release_xprt(xprt, task);
451 spin_unlock(&xprt->transport_lock);
455 * Van Jacobson congestion avoidance. Check if the congestion window
456 * overflowed. Put the task to sleep if this is the case.
459 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
463 trace_xprt_get_cong(xprt, req->rq_task);
464 if (RPCXPRT_CONGESTED(xprt)) {
465 xprt_set_congestion_window_wait(xprt);
469 xprt->cong += RPC_CWNDSCALE;
474 * Adjust the congestion window, and wake up the next task
475 * that has been sleeping due to congestion
478 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
483 xprt->cong -= RPC_CWNDSCALE;
484 xprt_test_and_clear_congestion_window_wait(xprt);
485 trace_xprt_put_cong(xprt, req->rq_task);
486 __xprt_lock_write_next_cong(xprt);
490 * xprt_request_get_cong - Request congestion control credits
491 * @xprt: pointer to transport
492 * @req: pointer to RPC request
494 * Useful for transports that require congestion control.
497 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
503 spin_lock(&xprt->transport_lock);
504 ret = __xprt_get_cong(xprt, req) != 0;
505 spin_unlock(&xprt->transport_lock);
508 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
511 * xprt_release_rqst_cong - housekeeping when request is complete
512 * @task: RPC request that recently completed
514 * Useful for transports that require congestion control.
516 void xprt_release_rqst_cong(struct rpc_task *task)
518 struct rpc_rqst *req = task->tk_rqstp;
520 __xprt_put_cong(req->rq_xprt, req);
522 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
524 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
526 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
527 __xprt_lock_write_next_cong(xprt);
531 * Clear the congestion window wait flag and wake up the next
532 * entry on xprt->sending
535 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
537 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
538 spin_lock(&xprt->transport_lock);
539 __xprt_lock_write_next_cong(xprt);
540 spin_unlock(&xprt->transport_lock);
545 * xprt_adjust_cwnd - adjust transport congestion window
546 * @xprt: pointer to xprt
547 * @task: recently completed RPC request used to adjust window
548 * @result: result code of completed RPC request
550 * The transport code maintains an estimate on the maximum number of out-
551 * standing RPC requests, using a smoothed version of the congestion
552 * avoidance implemented in 44BSD. This is basically the Van Jacobson
553 * congestion algorithm: If a retransmit occurs, the congestion window is
554 * halved; otherwise, it is incremented by 1/cwnd when
556 * - a reply is received and
557 * - a full number of requests are outstanding and
558 * - the congestion window hasn't been updated recently.
560 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
562 struct rpc_rqst *req = task->tk_rqstp;
563 unsigned long cwnd = xprt->cwnd;
565 if (result >= 0 && cwnd <= xprt->cong) {
566 /* The (cwnd >> 1) term makes sure
567 * the result gets rounded properly. */
568 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
569 if (cwnd > RPC_MAXCWND(xprt))
570 cwnd = RPC_MAXCWND(xprt);
571 __xprt_lock_write_next_cong(xprt);
572 } else if (result == -ETIMEDOUT) {
574 if (cwnd < RPC_CWNDSCALE)
575 cwnd = RPC_CWNDSCALE;
577 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
578 xprt->cong, xprt->cwnd, cwnd);
580 __xprt_put_cong(xprt, req);
582 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
585 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
586 * @xprt: transport with waiting tasks
587 * @status: result code to plant in each task before waking it
590 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
593 rpc_wake_up_status(&xprt->pending, status);
595 rpc_wake_up(&xprt->pending);
597 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
600 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
603 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
604 * we don't in general want to force a socket disconnection due to
605 * an incomplete RPC call transmission.
607 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
609 set_bit(XPRT_WRITE_SPACE, &xprt->state);
611 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
614 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
616 if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
617 __xprt_lock_write_next(xprt);
618 dprintk("RPC: write space: waking waiting task on "
626 * xprt_write_space - wake the task waiting for transport output buffer space
627 * @xprt: transport with waiting tasks
629 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
631 bool xprt_write_space(struct rpc_xprt *xprt)
635 if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
637 spin_lock(&xprt->transport_lock);
638 ret = xprt_clear_write_space_locked(xprt);
639 spin_unlock(&xprt->transport_lock);
642 EXPORT_SYMBOL_GPL(xprt_write_space);
644 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
646 s64 delta = ktime_to_ns(ktime_get() - abstime);
647 return likely(delta >= 0) ?
648 jiffies - nsecs_to_jiffies(delta) :
649 jiffies + nsecs_to_jiffies(-delta);
652 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
654 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
655 unsigned long majortimeo = req->rq_timeout;
657 if (to->to_exponential)
658 majortimeo <<= to->to_retries;
660 majortimeo += to->to_increment * to->to_retries;
661 if (majortimeo > to->to_maxval || majortimeo == 0)
662 majortimeo = to->to_maxval;
666 static void xprt_reset_majortimeo(struct rpc_rqst *req)
668 req->rq_majortimeo += xprt_calc_majortimeo(req);
671 static void xprt_reset_minortimeo(struct rpc_rqst *req)
673 req->rq_minortimeo += req->rq_timeout;
676 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
678 unsigned long time_init;
679 struct rpc_xprt *xprt = req->rq_xprt;
681 if (likely(xprt && xprt_connected(xprt)))
684 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
685 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
686 req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
687 req->rq_minortimeo = time_init + req->rq_timeout;
691 * xprt_adjust_timeout - adjust timeout values for next retransmit
692 * @req: RPC request containing parameters to use for the adjustment
695 int xprt_adjust_timeout(struct rpc_rqst *req)
697 struct rpc_xprt *xprt = req->rq_xprt;
698 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
701 if (time_before(jiffies, req->rq_majortimeo)) {
702 if (time_before(jiffies, req->rq_minortimeo))
704 if (to->to_exponential)
705 req->rq_timeout <<= 1;
707 req->rq_timeout += to->to_increment;
708 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
709 req->rq_timeout = to->to_maxval;
712 req->rq_timeout = to->to_initval;
714 xprt_reset_majortimeo(req);
715 /* Reset the RTT counters == "slow start" */
716 spin_lock(&xprt->transport_lock);
717 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
718 spin_unlock(&xprt->transport_lock);
721 xprt_reset_minortimeo(req);
723 if (req->rq_timeout == 0) {
724 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
725 req->rq_timeout = 5 * HZ;
730 static void xprt_autoclose(struct work_struct *work)
732 struct rpc_xprt *xprt =
733 container_of(work, struct rpc_xprt, task_cleanup);
734 unsigned int pflags = memalloc_nofs_save();
736 trace_xprt_disconnect_auto(xprt);
737 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
738 xprt->ops->close(xprt);
739 xprt_release_write(xprt, NULL);
740 wake_up_bit(&xprt->state, XPRT_LOCKED);
741 memalloc_nofs_restore(pflags);
745 * xprt_disconnect_done - mark a transport as disconnected
746 * @xprt: transport to flag for disconnect
749 void xprt_disconnect_done(struct rpc_xprt *xprt)
751 trace_xprt_disconnect_done(xprt);
752 spin_lock(&xprt->transport_lock);
753 xprt_clear_connected(xprt);
754 xprt_clear_write_space_locked(xprt);
755 xprt_clear_congestion_window_wait_locked(xprt);
756 xprt_wake_pending_tasks(xprt, -ENOTCONN);
757 spin_unlock(&xprt->transport_lock);
759 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
762 * xprt_force_disconnect - force a transport to disconnect
763 * @xprt: transport to disconnect
766 void xprt_force_disconnect(struct rpc_xprt *xprt)
768 trace_xprt_disconnect_force(xprt);
770 /* Don't race with the test_bit() in xprt_clear_locked() */
771 spin_lock(&xprt->transport_lock);
772 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
773 /* Try to schedule an autoclose RPC call */
774 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
775 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
776 else if (xprt->snd_task)
777 rpc_wake_up_queued_task_set_status(&xprt->pending,
778 xprt->snd_task, -ENOTCONN);
779 spin_unlock(&xprt->transport_lock);
781 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
784 xprt_connect_cookie(struct rpc_xprt *xprt)
786 return READ_ONCE(xprt->connect_cookie);
790 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
792 struct rpc_rqst *req = task->tk_rqstp;
793 struct rpc_xprt *xprt = req->rq_xprt;
795 return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
796 !xprt_connected(xprt);
800 * xprt_conditional_disconnect - force a transport to disconnect
801 * @xprt: transport to disconnect
802 * @cookie: 'connection cookie'
804 * This attempts to break the connection if and only if 'cookie' matches
805 * the current transport 'connection cookie'. It ensures that we don't
806 * try to break the connection more than once when we need to retransmit
807 * a batch of RPC requests.
810 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
812 /* Don't race with the test_bit() in xprt_clear_locked() */
813 spin_lock(&xprt->transport_lock);
814 if (cookie != xprt->connect_cookie)
816 if (test_bit(XPRT_CLOSING, &xprt->state))
818 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
819 /* Try to schedule an autoclose RPC call */
820 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
821 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
822 xprt_wake_pending_tasks(xprt, -EAGAIN);
824 spin_unlock(&xprt->transport_lock);
828 xprt_has_timer(const struct rpc_xprt *xprt)
830 return xprt->idle_timeout != 0;
834 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
835 __must_hold(&xprt->transport_lock)
837 xprt->last_used = jiffies;
838 if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
839 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
843 xprt_init_autodisconnect(struct timer_list *t)
845 struct rpc_xprt *xprt = from_timer(xprt, t, timer);
847 if (!RB_EMPTY_ROOT(&xprt->recv_queue))
849 /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
850 xprt->last_used = jiffies;
851 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
853 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
856 bool xprt_lock_connect(struct rpc_xprt *xprt,
857 struct rpc_task *task,
862 spin_lock(&xprt->transport_lock);
863 if (!test_bit(XPRT_LOCKED, &xprt->state))
865 if (xprt->snd_task != task)
867 xprt->snd_task = cookie;
870 spin_unlock(&xprt->transport_lock);
874 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
876 spin_lock(&xprt->transport_lock);
877 if (xprt->snd_task != cookie)
879 if (!test_bit(XPRT_LOCKED, &xprt->state))
881 xprt->snd_task =NULL;
882 xprt->ops->release_xprt(xprt, NULL);
883 xprt_schedule_autodisconnect(xprt);
885 spin_unlock(&xprt->transport_lock);
886 wake_up_bit(&xprt->state, XPRT_LOCKED);
890 * xprt_connect - schedule a transport connect operation
891 * @task: RPC task that is requesting the connect
894 void xprt_connect(struct rpc_task *task)
896 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
898 trace_xprt_connect(xprt);
900 if (!xprt_bound(xprt)) {
901 task->tk_status = -EAGAIN;
904 if (!xprt_lock_write(xprt, task))
907 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
908 trace_xprt_disconnect_cleanup(xprt);
909 xprt->ops->close(xprt);
912 if (!xprt_connected(xprt)) {
913 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
914 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
915 xprt_request_timeout(task->tk_rqstp));
917 if (test_bit(XPRT_CLOSING, &xprt->state))
919 if (xprt_test_and_set_connecting(xprt))
922 if (!xprt_connected(xprt)) {
923 xprt->stat.connect_start = jiffies;
924 xprt->ops->connect(xprt, task);
926 xprt_clear_connecting(xprt);
928 rpc_wake_up_queued_task(&xprt->pending, task);
931 xprt_release_write(xprt, task);
935 * xprt_reconnect_delay - compute the wait before scheduling a connect
936 * @xprt: transport instance
939 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
941 unsigned long start, now = jiffies;
943 start = xprt->stat.connect_start + xprt->reestablish_timeout;
944 if (time_after(start, now))
948 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
951 * xprt_reconnect_backoff - compute the new re-establish timeout
952 * @xprt: transport instance
953 * @init_to: initial reestablish timeout
956 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
958 xprt->reestablish_timeout <<= 1;
959 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
960 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
961 if (xprt->reestablish_timeout < init_to)
962 xprt->reestablish_timeout = init_to;
964 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
966 enum xprt_xid_rb_cmp {
971 static enum xprt_xid_rb_cmp
972 xprt_xid_cmp(__be32 xid1, __be32 xid2)
976 if ((__force u32)xid1 < (__force u32)xid2)
981 static struct rpc_rqst *
982 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
984 struct rb_node *n = xprt->recv_queue.rb_node;
985 struct rpc_rqst *req;
988 req = rb_entry(n, struct rpc_rqst, rq_recv);
989 switch (xprt_xid_cmp(xid, req->rq_xid)) {
1004 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
1006 struct rb_node **p = &xprt->recv_queue.rb_node;
1007 struct rb_node *n = NULL;
1008 struct rpc_rqst *req;
1010 while (*p != NULL) {
1012 req = rb_entry(n, struct rpc_rqst, rq_recv);
1013 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
1021 WARN_ON_ONCE(new != req);
1025 rb_link_node(&new->rq_recv, n, p);
1026 rb_insert_color(&new->rq_recv, &xprt->recv_queue);
1030 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
1032 rb_erase(&req->rq_recv, &xprt->recv_queue);
1036 * xprt_lookup_rqst - find an RPC request corresponding to an XID
1037 * @xprt: transport on which the original request was transmitted
1038 * @xid: RPC XID of incoming reply
1040 * Caller holds xprt->queue_lock.
1042 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
1044 struct rpc_rqst *entry;
1046 entry = xprt_request_rb_find(xprt, xid);
1047 if (entry != NULL) {
1048 trace_xprt_lookup_rqst(xprt, xid, 0);
1049 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
1053 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
1055 trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
1056 xprt->stat.bad_xids++;
1059 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1062 xprt_is_pinned_rqst(struct rpc_rqst *req)
1064 return atomic_read(&req->rq_pin) != 0;
1068 * xprt_pin_rqst - Pin a request on the transport receive list
1069 * @req: Request to pin
1071 * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1072 * so should be holding xprt->queue_lock.
1074 void xprt_pin_rqst(struct rpc_rqst *req)
1076 atomic_inc(&req->rq_pin);
1078 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1081 * xprt_unpin_rqst - Unpin a request on the transport receive list
1082 * @req: Request to pin
1084 * Caller should be holding xprt->queue_lock.
1086 void xprt_unpin_rqst(struct rpc_rqst *req)
1088 if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1089 atomic_dec(&req->rq_pin);
1092 if (atomic_dec_and_test(&req->rq_pin))
1093 wake_up_var(&req->rq_pin);
1095 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1097 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1099 wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1103 xprt_request_data_received(struct rpc_task *task)
1105 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1106 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1110 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1112 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1113 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1117 * xprt_request_enqueue_receive - Add an request to the receive queue
1122 xprt_request_enqueue_receive(struct rpc_task *task)
1124 struct rpc_rqst *req = task->tk_rqstp;
1125 struct rpc_xprt *xprt = req->rq_xprt;
1127 if (!xprt_request_need_enqueue_receive(task, req))
1130 xprt_request_prepare(task->tk_rqstp);
1131 spin_lock(&xprt->queue_lock);
1133 /* Update the softirq receive buffer */
1134 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1135 sizeof(req->rq_private_buf));
1137 /* Add request to the receive list */
1138 xprt_request_rb_insert(xprt, req);
1139 set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1140 spin_unlock(&xprt->queue_lock);
1142 /* Turn off autodisconnect */
1143 del_singleshot_timer_sync(&xprt->timer);
1147 * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1150 * Caller must hold xprt->queue_lock.
1153 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1155 struct rpc_rqst *req = task->tk_rqstp;
1157 if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1158 xprt_request_rb_remove(req->rq_xprt, req);
1162 * xprt_update_rtt - Update RPC RTT statistics
1163 * @task: RPC request that recently completed
1165 * Caller holds xprt->queue_lock.
1167 void xprt_update_rtt(struct rpc_task *task)
1169 struct rpc_rqst *req = task->tk_rqstp;
1170 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1171 unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1172 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1175 if (req->rq_ntrans == 1)
1176 rpc_update_rtt(rtt, timer, m);
1177 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1180 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1183 * xprt_complete_rqst - called when reply processing is complete
1184 * @task: RPC request that recently completed
1185 * @copied: actual number of bytes received from the transport
1187 * Caller holds xprt->queue_lock.
1189 void xprt_complete_rqst(struct rpc_task *task, int copied)
1191 struct rpc_rqst *req = task->tk_rqstp;
1192 struct rpc_xprt *xprt = req->rq_xprt;
1196 req->rq_private_buf.len = copied;
1197 /* Ensure all writes are done before we update */
1198 /* req->rq_reply_bytes_recvd */
1200 req->rq_reply_bytes_recvd = copied;
1201 xprt_request_dequeue_receive_locked(task);
1202 rpc_wake_up_queued_task(&xprt->pending, task);
1204 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1206 static void xprt_timer(struct rpc_task *task)
1208 struct rpc_rqst *req = task->tk_rqstp;
1209 struct rpc_xprt *xprt = req->rq_xprt;
1211 if (task->tk_status != -ETIMEDOUT)
1214 trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1215 if (!req->rq_reply_bytes_recvd) {
1216 if (xprt->ops->timer)
1217 xprt->ops->timer(xprt, task);
1219 task->tk_status = 0;
1223 * xprt_wait_for_reply_request_def - wait for reply
1224 * @task: pointer to rpc_task
1226 * Set a request's retransmit timeout based on the transport's
1227 * default timeout parameters. Used by transports that don't adjust
1228 * the retransmit timeout based on round-trip time estimation,
1229 * and put the task to sleep on the pending queue.
1231 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1233 struct rpc_rqst *req = task->tk_rqstp;
1235 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1236 xprt_request_timeout(req));
1238 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1241 * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1242 * @task: pointer to rpc_task
1244 * Set a request's retransmit timeout using the RTT estimator,
1245 * and put the task to sleep on the pending queue.
1247 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1249 int timer = task->tk_msg.rpc_proc->p_timer;
1250 struct rpc_clnt *clnt = task->tk_client;
1251 struct rpc_rtt *rtt = clnt->cl_rtt;
1252 struct rpc_rqst *req = task->tk_rqstp;
1253 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1254 unsigned long timeout;
1256 timeout = rpc_calc_rto(rtt, timer);
1257 timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1258 if (timeout > max_timeout || timeout == 0)
1259 timeout = max_timeout;
1260 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1263 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1266 * xprt_request_wait_receive - wait for the reply to an RPC request
1267 * @task: RPC task about to send a request
1270 void xprt_request_wait_receive(struct rpc_task *task)
1272 struct rpc_rqst *req = task->tk_rqstp;
1273 struct rpc_xprt *xprt = req->rq_xprt;
1275 if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1278 * Sleep on the pending queue if we're expecting a reply.
1279 * The spinlock ensures atomicity between the test of
1280 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1282 spin_lock(&xprt->queue_lock);
1283 if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1284 xprt->ops->wait_for_reply_request(task);
1286 * Send an extra queue wakeup call if the
1287 * connection was dropped in case the call to
1288 * rpc_sleep_on() raced.
1290 if (xprt_request_retransmit_after_disconnect(task))
1291 rpc_wake_up_queued_task_set_status(&xprt->pending,
1294 spin_unlock(&xprt->queue_lock);
1298 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1300 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1304 * xprt_request_enqueue_transmit - queue a task for transmission
1305 * @task: pointer to rpc_task
1307 * Add a task to the transmission queue.
1310 xprt_request_enqueue_transmit(struct rpc_task *task)
1312 struct rpc_rqst *pos, *req = task->tk_rqstp;
1313 struct rpc_xprt *xprt = req->rq_xprt;
1315 if (xprt_request_need_enqueue_transmit(task, req)) {
1316 req->rq_bytes_sent = 0;
1317 spin_lock(&xprt->queue_lock);
1319 * Requests that carry congestion control credits are added
1320 * to the head of the list to avoid starvation issues.
1323 xprt_clear_congestion_window_wait(xprt);
1324 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1327 /* Note: req is added _before_ pos */
1328 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1329 INIT_LIST_HEAD(&req->rq_xmit2);
1332 } else if (RPC_IS_SWAPPER(task)) {
1333 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1334 if (pos->rq_cong || pos->rq_bytes_sent)
1336 if (RPC_IS_SWAPPER(pos->rq_task))
1338 /* Note: req is added _before_ pos */
1339 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1340 INIT_LIST_HEAD(&req->rq_xmit2);
1343 } else if (!req->rq_seqno) {
1344 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1345 if (pos->rq_task->tk_owner != task->tk_owner)
1347 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1348 INIT_LIST_HEAD(&req->rq_xmit);
1352 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1353 INIT_LIST_HEAD(&req->rq_xmit2);
1355 atomic_long_inc(&xprt->xmit_queuelen);
1356 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1357 spin_unlock(&xprt->queue_lock);
1362 * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1363 * @task: pointer to rpc_task
1365 * Remove a task from the transmission queue
1366 * Caller must hold xprt->queue_lock
1369 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1371 struct rpc_rqst *req = task->tk_rqstp;
1373 if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1375 if (!list_empty(&req->rq_xmit)) {
1376 list_del(&req->rq_xmit);
1377 if (!list_empty(&req->rq_xmit2)) {
1378 struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1379 struct rpc_rqst, rq_xmit2);
1380 list_del(&req->rq_xmit2);
1381 list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1384 list_del(&req->rq_xmit2);
1385 atomic_long_dec(&req->rq_xprt->xmit_queuelen);
1389 * xprt_request_dequeue_transmit - remove a task from the transmission queue
1390 * @task: pointer to rpc_task
1392 * Remove a task from the transmission queue
1395 xprt_request_dequeue_transmit(struct rpc_task *task)
1397 struct rpc_rqst *req = task->tk_rqstp;
1398 struct rpc_xprt *xprt = req->rq_xprt;
1400 spin_lock(&xprt->queue_lock);
1401 xprt_request_dequeue_transmit_locked(task);
1402 spin_unlock(&xprt->queue_lock);
1406 * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1407 * @task: pointer to rpc_task
1409 * Remove a task from the transmit and receive queues, and ensure that
1410 * it is not pinned by the receive work item.
1413 xprt_request_dequeue_xprt(struct rpc_task *task)
1415 struct rpc_rqst *req = task->tk_rqstp;
1416 struct rpc_xprt *xprt = req->rq_xprt;
1418 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1419 test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1420 xprt_is_pinned_rqst(req)) {
1421 spin_lock(&xprt->queue_lock);
1422 xprt_request_dequeue_transmit_locked(task);
1423 xprt_request_dequeue_receive_locked(task);
1424 while (xprt_is_pinned_rqst(req)) {
1425 set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1426 spin_unlock(&xprt->queue_lock);
1427 xprt_wait_on_pinned_rqst(req);
1428 spin_lock(&xprt->queue_lock);
1429 clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1431 spin_unlock(&xprt->queue_lock);
1436 * xprt_request_prepare - prepare an encoded request for transport
1437 * @req: pointer to rpc_rqst
1439 * Calls into the transport layer to do whatever is needed to prepare
1440 * the request for transmission or receive.
1443 xprt_request_prepare(struct rpc_rqst *req)
1445 struct rpc_xprt *xprt = req->rq_xprt;
1447 if (xprt->ops->prepare_request)
1448 xprt->ops->prepare_request(req);
1452 * xprt_request_need_retransmit - Test if a task needs retransmission
1453 * @task: pointer to rpc_task
1455 * Test for whether a connection breakage requires the task to retransmit
1458 xprt_request_need_retransmit(struct rpc_task *task)
1460 return xprt_request_retransmit_after_disconnect(task);
1464 * xprt_prepare_transmit - reserve the transport before sending a request
1465 * @task: RPC task about to send a request
1468 bool xprt_prepare_transmit(struct rpc_task *task)
1470 struct rpc_rqst *req = task->tk_rqstp;
1471 struct rpc_xprt *xprt = req->rq_xprt;
1473 if (!xprt_lock_write(xprt, task)) {
1474 /* Race breaker: someone may have transmitted us */
1475 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1476 rpc_wake_up_queued_task_set_status(&xprt->sending,
1484 void xprt_end_transmit(struct rpc_task *task)
1486 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1488 xprt_inject_disconnect(xprt);
1489 xprt_release_write(xprt, task);
1493 * xprt_request_transmit - send an RPC request on a transport
1494 * @req: pointer to request to transmit
1495 * @snd_task: RPC task that owns the transport lock
1497 * This performs the transmission of a single request.
1498 * Note that if the request is not the same as snd_task, then it
1499 * does need to be pinned.
1500 * Returns '0' on success.
1503 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1505 struct rpc_xprt *xprt = req->rq_xprt;
1506 struct rpc_task *task = req->rq_task;
1507 unsigned int connect_cookie;
1508 int is_retrans = RPC_WAS_SENT(task);
1511 if (!req->rq_bytes_sent) {
1512 if (xprt_request_data_received(task)) {
1516 /* Verify that our message lies in the RPCSEC_GSS window */
1517 if (rpcauth_xmit_need_reencode(task)) {
1521 if (RPC_SIGNALLED(task)) {
1522 status = -ERESTARTSYS;
1528 * Update req->rq_ntrans before transmitting to avoid races with
1529 * xprt_update_rtt(), which needs to know that it is recording a
1530 * reply to the first transmission.
1534 trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1535 connect_cookie = xprt->connect_cookie;
1536 status = xprt->ops->send_request(req);
1539 trace_xprt_transmit(req, status);
1544 task->tk_client->cl_stats->rpcretrans++;
1545 trace_xprt_retransmit(req);
1548 xprt_inject_disconnect(xprt);
1550 task->tk_flags |= RPC_TASK_SENT;
1551 spin_lock(&xprt->transport_lock);
1554 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1555 xprt->stat.bklog_u += xprt->backlog.qlen;
1556 xprt->stat.sending_u += xprt->sending.qlen;
1557 xprt->stat.pending_u += xprt->pending.qlen;
1558 spin_unlock(&xprt->transport_lock);
1560 req->rq_connect_cookie = connect_cookie;
1562 trace_xprt_transmit(req, status);
1563 xprt_request_dequeue_transmit(task);
1564 rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1569 * xprt_transmit - send an RPC request on a transport
1570 * @task: controlling RPC task
1572 * Attempts to drain the transmit queue. On exit, either the transport
1573 * signalled an error that needs to be handled before transmission can
1574 * resume, or @task finished transmitting, and detected that it already
1578 xprt_transmit(struct rpc_task *task)
1580 struct rpc_rqst *next, *req = task->tk_rqstp;
1581 struct rpc_xprt *xprt = req->rq_xprt;
1582 int counter, status;
1584 spin_lock(&xprt->queue_lock);
1586 while (!list_empty(&xprt->xmit_queue)) {
1587 if (++counter == 20)
1589 next = list_first_entry(&xprt->xmit_queue,
1590 struct rpc_rqst, rq_xmit);
1591 xprt_pin_rqst(next);
1592 spin_unlock(&xprt->queue_lock);
1593 status = xprt_request_transmit(next, task);
1594 if (status == -EBADMSG && next != req)
1596 spin_lock(&xprt->queue_lock);
1597 xprt_unpin_rqst(next);
1599 if (!xprt_request_data_received(task) ||
1600 test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1602 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1603 task->tk_status = status;
1606 spin_unlock(&xprt->queue_lock);
1609 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1611 set_bit(XPRT_CONGESTED, &xprt->state);
1612 rpc_sleep_on(&xprt->backlog, task, NULL);
1615 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1617 if (rpc_wake_up_next(&xprt->backlog) == NULL)
1618 clear_bit(XPRT_CONGESTED, &xprt->state);
1621 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1625 if (!test_bit(XPRT_CONGESTED, &xprt->state))
1627 spin_lock(&xprt->reserve_lock);
1628 if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1629 rpc_sleep_on(&xprt->backlog, task, NULL);
1632 spin_unlock(&xprt->reserve_lock);
1637 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1639 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1641 if (xprt->num_reqs >= xprt->max_reqs)
1644 spin_unlock(&xprt->reserve_lock);
1645 req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1646 spin_lock(&xprt->reserve_lock);
1650 req = ERR_PTR(-ENOMEM);
1655 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1657 if (xprt->num_reqs > xprt->min_reqs) {
1665 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1667 struct rpc_rqst *req;
1669 spin_lock(&xprt->reserve_lock);
1670 if (!list_empty(&xprt->free)) {
1671 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1672 list_del(&req->rq_list);
1675 req = xprt_dynamic_alloc_slot(xprt);
1678 switch (PTR_ERR(req)) {
1680 dprintk("RPC: dynamic allocation of request slot "
1681 "failed! Retrying\n");
1682 task->tk_status = -ENOMEM;
1685 xprt_add_backlog(xprt, task);
1686 dprintk("RPC: waiting for request slot\n");
1689 task->tk_status = -EAGAIN;
1691 spin_unlock(&xprt->reserve_lock);
1694 xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1696 spin_unlock(&xprt->reserve_lock);
1698 task->tk_status = 0;
1699 task->tk_rqstp = req;
1701 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1703 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1705 spin_lock(&xprt->reserve_lock);
1706 if (!xprt_dynamic_free_slot(xprt, req)) {
1707 memset(req, 0, sizeof(*req)); /* mark unused */
1708 list_add(&req->rq_list, &xprt->free);
1710 xprt_wake_up_backlog(xprt);
1711 spin_unlock(&xprt->reserve_lock);
1713 EXPORT_SYMBOL_GPL(xprt_free_slot);
1715 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1717 struct rpc_rqst *req;
1718 while (!list_empty(&xprt->free)) {
1719 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1720 list_del(&req->rq_list);
1725 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1726 unsigned int num_prealloc,
1727 unsigned int max_alloc)
1729 struct rpc_xprt *xprt;
1730 struct rpc_rqst *req;
1733 xprt = kzalloc(size, GFP_KERNEL);
1737 xprt_init(xprt, net);
1739 for (i = 0; i < num_prealloc; i++) {
1740 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1743 list_add(&req->rq_list, &xprt->free);
1745 if (max_alloc > num_prealloc)
1746 xprt->max_reqs = max_alloc;
1748 xprt->max_reqs = num_prealloc;
1749 xprt->min_reqs = num_prealloc;
1750 xprt->num_reqs = num_prealloc;
1759 EXPORT_SYMBOL_GPL(xprt_alloc);
1761 void xprt_free(struct rpc_xprt *xprt)
1763 put_net(xprt->xprt_net);
1764 xprt_free_all_slots(xprt);
1765 kfree_rcu(xprt, rcu);
1767 EXPORT_SYMBOL_GPL(xprt_free);
1770 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1772 req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1776 xprt_alloc_xid(struct rpc_xprt *xprt)
1780 spin_lock(&xprt->reserve_lock);
1781 xid = (__force __be32)xprt->xid++;
1782 spin_unlock(&xprt->reserve_lock);
1787 xprt_init_xid(struct rpc_xprt *xprt)
1789 xprt->xid = prandom_u32();
1793 xprt_request_init(struct rpc_task *task)
1795 struct rpc_xprt *xprt = task->tk_xprt;
1796 struct rpc_rqst *req = task->tk_rqstp;
1798 req->rq_task = task;
1799 req->rq_xprt = xprt;
1800 req->rq_buffer = NULL;
1801 req->rq_xid = xprt_alloc_xid(xprt);
1802 xprt_init_connect_cookie(req, xprt);
1803 req->rq_snd_buf.len = 0;
1804 req->rq_snd_buf.buflen = 0;
1805 req->rq_rcv_buf.len = 0;
1806 req->rq_rcv_buf.buflen = 0;
1807 req->rq_snd_buf.bvec = NULL;
1808 req->rq_rcv_buf.bvec = NULL;
1809 req->rq_release_snd_buf = NULL;
1810 xprt_init_majortimeo(task, req);
1812 trace_xprt_reserve(req);
1816 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1818 xprt->ops->alloc_slot(xprt, task);
1819 if (task->tk_rqstp != NULL)
1820 xprt_request_init(task);
1824 * xprt_reserve - allocate an RPC request slot
1825 * @task: RPC task requesting a slot allocation
1827 * If the transport is marked as being congested, or if no more
1828 * slots are available, place the task on the transport's
1831 void xprt_reserve(struct rpc_task *task)
1833 struct rpc_xprt *xprt = task->tk_xprt;
1835 task->tk_status = 0;
1836 if (task->tk_rqstp != NULL)
1839 task->tk_status = -EAGAIN;
1840 if (!xprt_throttle_congested(xprt, task))
1841 xprt_do_reserve(xprt, task);
1845 * xprt_retry_reserve - allocate an RPC request slot
1846 * @task: RPC task requesting a slot allocation
1848 * If no more slots are available, place the task on the transport's
1850 * Note that the only difference with xprt_reserve is that we now
1851 * ignore the value of the XPRT_CONGESTED flag.
1853 void xprt_retry_reserve(struct rpc_task *task)
1855 struct rpc_xprt *xprt = task->tk_xprt;
1857 task->tk_status = 0;
1858 if (task->tk_rqstp != NULL)
1861 task->tk_status = -EAGAIN;
1862 xprt_do_reserve(xprt, task);
1866 * xprt_release - release an RPC request slot
1867 * @task: task which is finished with the slot
1870 void xprt_release(struct rpc_task *task)
1872 struct rpc_xprt *xprt;
1873 struct rpc_rqst *req = task->tk_rqstp;
1876 if (task->tk_client) {
1877 xprt = task->tk_xprt;
1878 xprt_release_write(xprt, task);
1883 xprt = req->rq_xprt;
1884 xprt_request_dequeue_xprt(task);
1885 spin_lock(&xprt->transport_lock);
1886 xprt->ops->release_xprt(xprt, task);
1887 if (xprt->ops->release_request)
1888 xprt->ops->release_request(task);
1889 xprt_schedule_autodisconnect(xprt);
1890 spin_unlock(&xprt->transport_lock);
1892 xprt->ops->buf_free(task);
1893 xdr_free_bvec(&req->rq_rcv_buf);
1894 xdr_free_bvec(&req->rq_snd_buf);
1895 if (req->rq_cred != NULL)
1896 put_rpccred(req->rq_cred);
1897 task->tk_rqstp = NULL;
1898 if (req->rq_release_snd_buf)
1899 req->rq_release_snd_buf(req);
1901 if (likely(!bc_prealloc(req)))
1902 xprt->ops->free_slot(xprt, req);
1904 xprt_free_bc_request(req);
1907 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1909 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1911 struct xdr_buf *xbufp = &req->rq_snd_buf;
1913 task->tk_rqstp = req;
1914 req->rq_task = task;
1915 xprt_init_connect_cookie(req, req->rq_xprt);
1917 * Set up the xdr_buf length.
1918 * This also indicates that the buffer is XDR encoded already.
1920 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1921 xbufp->tail[0].iov_len;
1925 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1927 kref_init(&xprt->kref);
1929 spin_lock_init(&xprt->transport_lock);
1930 spin_lock_init(&xprt->reserve_lock);
1931 spin_lock_init(&xprt->queue_lock);
1933 INIT_LIST_HEAD(&xprt->free);
1934 xprt->recv_queue = RB_ROOT;
1935 INIT_LIST_HEAD(&xprt->xmit_queue);
1936 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1937 spin_lock_init(&xprt->bc_pa_lock);
1938 INIT_LIST_HEAD(&xprt->bc_pa_list);
1939 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1940 INIT_LIST_HEAD(&xprt->xprt_switch);
1942 xprt->last_used = jiffies;
1943 xprt->cwnd = RPC_INITCWND;
1944 xprt->bind_index = 0;
1946 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1947 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1948 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1949 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1951 xprt_init_xid(xprt);
1953 xprt->xprt_net = get_net(net);
1957 * xprt_create_transport - create an RPC transport
1958 * @args: rpc transport creation arguments
1961 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1963 struct rpc_xprt *xprt;
1964 const struct xprt_class *t;
1966 t = xprt_class_find_by_ident(args->ident);
1968 dprintk("RPC: transport (%d) not supported\n", args->ident);
1969 return ERR_PTR(-EIO);
1972 xprt = t->setup(args);
1973 xprt_class_release(t);
1977 if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1978 xprt->idle_timeout = 0;
1979 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1980 if (xprt_has_timer(xprt))
1981 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1983 timer_setup(&xprt->timer, NULL, 0);
1985 if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1987 return ERR_PTR(-EINVAL);
1989 xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1990 if (xprt->servername == NULL) {
1992 return ERR_PTR(-ENOMEM);
1995 rpc_xprt_debugfs_register(xprt);
1997 trace_xprt_create(xprt);
2002 static void xprt_destroy_cb(struct work_struct *work)
2004 struct rpc_xprt *xprt =
2005 container_of(work, struct rpc_xprt, task_cleanup);
2007 trace_xprt_destroy(xprt);
2009 rpc_xprt_debugfs_unregister(xprt);
2010 rpc_destroy_wait_queue(&xprt->binding);
2011 rpc_destroy_wait_queue(&xprt->pending);
2012 rpc_destroy_wait_queue(&xprt->sending);
2013 rpc_destroy_wait_queue(&xprt->backlog);
2014 kfree(xprt->servername);
2016 * Destroy any existing back channel
2018 xprt_destroy_backchannel(xprt, UINT_MAX);
2021 * Tear down transport state and free the rpc_xprt
2023 xprt->ops->destroy(xprt);
2027 * xprt_destroy - destroy an RPC transport, killing off all requests.
2028 * @xprt: transport to destroy
2031 static void xprt_destroy(struct rpc_xprt *xprt)
2034 * Exclude transport connect/disconnect handlers and autoclose
2036 wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
2038 del_timer_sync(&xprt->timer);
2041 * Destroy sockets etc from the system workqueue so they can
2042 * safely flush receive work running on rpciod.
2044 INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
2045 schedule_work(&xprt->task_cleanup);
2048 static void xprt_destroy_kref(struct kref *kref)
2050 xprt_destroy(container_of(kref, struct rpc_xprt, kref));
2054 * xprt_get - return a reference to an RPC transport.
2055 * @xprt: pointer to the transport
2058 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2060 if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2064 EXPORT_SYMBOL_GPL(xprt_get);
2067 * xprt_put - release a reference to an RPC transport.
2068 * @xprt: pointer to the transport
2071 void xprt_put(struct rpc_xprt *xprt)
2074 kref_put(&xprt->kref, xprt_destroy_kref);
2076 EXPORT_SYMBOL_GPL(xprt_put);