1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/net/sunrpc/xprt.c
5 * This is a generic RPC call interface supporting congestion avoidance,
6 * and asynchronous calls.
8 * The interface works like this:
10 * - When a process places a call, it allocates a request slot if
11 * one is available. Otherwise, it sleeps on the backlog queue
13 * - Next, the caller puts together the RPC message, stuffs it into
14 * the request struct, and calls xprt_transmit().
15 * - xprt_transmit sends the message and installs the caller on the
16 * transport's wait list. At the same time, if a reply is expected,
17 * it installs a timer that is run after the packet's timeout has
19 * - When a packet arrives, the data_ready handler walks the list of
20 * pending requests for that transport. If a matching XID is found, the
21 * caller is woken up, and the timer removed.
22 * - When no reply arrives within the timeout interval, the timer is
23 * fired by the kernel and runs xprt_timer(). It either adjusts the
24 * timeout values (minor timeout) or wakes up the caller with a status
26 * - When the caller receives a notification from RPC that a reply arrived,
27 * it should release the RPC slot, and process the reply.
28 * If the call timed out, it may choose to retry the operation by
29 * adjusting the initial timeout value, and simply calling rpc_call
32 * Support for async RPC is done through a set of RPC-specific scheduling
33 * primitives that `transparently' work for processes as well as async
34 * tasks that rely on callbacks.
36 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
38 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
41 #include <linux/module.h>
43 #include <linux/types.h>
44 #include <linux/interrupt.h>
45 #include <linux/workqueue.h>
46 #include <linux/net.h>
47 #include <linux/ktime.h>
49 #include <linux/sunrpc/clnt.h>
50 #include <linux/sunrpc/metrics.h>
51 #include <linux/sunrpc/bc_xprt.h>
52 #include <linux/rcupdate.h>
53 #include <linux/sched/mm.h>
55 #include <trace/events/sunrpc.h>
63 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
64 # define RPCDBG_FACILITY RPCDBG_XPRT
70 static void xprt_init(struct rpc_xprt *xprt, struct net *net);
71 static __be32 xprt_alloc_xid(struct rpc_xprt *xprt);
72 static void xprt_destroy(struct rpc_xprt *xprt);
74 static DEFINE_SPINLOCK(xprt_list_lock);
75 static LIST_HEAD(xprt_list);
77 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
79 unsigned long timeout = jiffies + req->rq_timeout;
81 if (time_before(timeout, req->rq_majortimeo))
83 return req->rq_majortimeo;
87 * xprt_register_transport - register a transport implementation
88 * @transport: transport to register
90 * If a transport implementation is loaded as a kernel module, it can
91 * call this interface to make itself known to the RPC client.
94 * 0: transport successfully registered
95 * -EEXIST: transport already registered
96 * -EINVAL: transport module being unloaded
98 int xprt_register_transport(struct xprt_class *transport)
100 struct xprt_class *t;
104 spin_lock(&xprt_list_lock);
105 list_for_each_entry(t, &xprt_list, list) {
106 /* don't register the same transport class twice */
107 if (t->ident == transport->ident)
111 list_add_tail(&transport->list, &xprt_list);
112 printk(KERN_INFO "RPC: Registered %s transport module.\n",
117 spin_unlock(&xprt_list_lock);
120 EXPORT_SYMBOL_GPL(xprt_register_transport);
123 * xprt_unregister_transport - unregister a transport implementation
124 * @transport: transport to unregister
127 * 0: transport successfully unregistered
128 * -ENOENT: transport never registered
130 int xprt_unregister_transport(struct xprt_class *transport)
132 struct xprt_class *t;
136 spin_lock(&xprt_list_lock);
137 list_for_each_entry(t, &xprt_list, list) {
138 if (t == transport) {
140 "RPC: Unregistered %s transport module.\n",
142 list_del_init(&transport->list);
149 spin_unlock(&xprt_list_lock);
152 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
155 * xprt_load_transport - load a transport implementation
156 * @transport_name: transport to load
159 * 0: transport successfully loaded
160 * -ENOENT: transport module not available
162 int xprt_load_transport(const char *transport_name)
164 struct xprt_class *t;
168 spin_lock(&xprt_list_lock);
169 list_for_each_entry(t, &xprt_list, list) {
170 if (strcmp(t->name, transport_name) == 0) {
171 spin_unlock(&xprt_list_lock);
175 spin_unlock(&xprt_list_lock);
176 result = request_module("xprt%s", transport_name);
180 EXPORT_SYMBOL_GPL(xprt_load_transport);
182 static void xprt_clear_locked(struct rpc_xprt *xprt)
184 xprt->snd_task = NULL;
185 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
186 smp_mb__before_atomic();
187 clear_bit(XPRT_LOCKED, &xprt->state);
188 smp_mb__after_atomic();
190 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
194 * xprt_reserve_xprt - serialize write access to transports
195 * @task: task that is requesting access to the transport
196 * @xprt: pointer to the target transport
198 * This prevents mixing the payload of separate requests, and prevents
199 * transport connects from colliding with writes. No congestion control
202 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
204 struct rpc_rqst *req = task->tk_rqstp;
206 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
207 if (task == xprt->snd_task)
211 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
213 xprt->snd_task = task;
216 trace_xprt_reserve_xprt(xprt, task);
220 xprt_clear_locked(xprt);
222 task->tk_status = -EAGAIN;
223 if (RPC_IS_SOFT(task))
224 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
225 xprt_request_timeout(req));
227 rpc_sleep_on(&xprt->sending, task, NULL);
230 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
233 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
235 return test_bit(XPRT_CWND_WAIT, &xprt->state);
239 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
241 if (!list_empty(&xprt->xmit_queue)) {
242 /* Peek at head of queue to see if it can make progress */
243 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
247 set_bit(XPRT_CWND_WAIT, &xprt->state);
251 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
253 if (!RPCXPRT_CONGESTED(xprt))
254 clear_bit(XPRT_CWND_WAIT, &xprt->state);
258 * xprt_reserve_xprt_cong - serialize write access to transports
259 * @task: task that is requesting access to the transport
261 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
262 * integrated into the decision of whether a request is allowed to be
263 * woken up and given access to the transport.
264 * Note that the lock is only granted if we know there are free slots.
266 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
268 struct rpc_rqst *req = task->tk_rqstp;
270 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
271 if (task == xprt->snd_task)
276 xprt->snd_task = task;
279 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
281 if (!xprt_need_congestion_window_wait(xprt)) {
282 xprt->snd_task = task;
286 xprt_clear_locked(xprt);
288 task->tk_status = -EAGAIN;
289 if (RPC_IS_SOFT(task))
290 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
291 xprt_request_timeout(req));
293 rpc_sleep_on(&xprt->sending, task, NULL);
296 trace_xprt_reserve_cong(xprt, task);
299 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
301 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
305 if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
307 spin_lock(&xprt->transport_lock);
308 retval = xprt->ops->reserve_xprt(xprt, task);
309 spin_unlock(&xprt->transport_lock);
313 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
315 struct rpc_xprt *xprt = data;
317 xprt->snd_task = task;
321 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
323 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
325 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
327 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
328 __xprt_lock_write_func, xprt))
331 xprt_clear_locked(xprt);
334 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
336 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
338 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
340 if (xprt_need_congestion_window_wait(xprt))
342 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
343 __xprt_lock_write_func, xprt))
346 xprt_clear_locked(xprt);
350 * xprt_release_xprt - allow other requests to use a transport
351 * @xprt: transport with other tasks potentially waiting
352 * @task: task that is releasing access to the transport
354 * Note that "task" can be NULL. No congestion control is provided.
356 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
358 if (xprt->snd_task == task) {
359 xprt_clear_locked(xprt);
360 __xprt_lock_write_next(xprt);
362 trace_xprt_release_xprt(xprt, task);
364 EXPORT_SYMBOL_GPL(xprt_release_xprt);
367 * xprt_release_xprt_cong - allow other requests to use a transport
368 * @xprt: transport with other tasks potentially waiting
369 * @task: task that is releasing access to the transport
371 * Note that "task" can be NULL. Another task is awoken to use the
372 * transport if the transport's congestion window allows it.
374 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
376 if (xprt->snd_task == task) {
377 xprt_clear_locked(xprt);
378 __xprt_lock_write_next_cong(xprt);
380 trace_xprt_release_cong(xprt, task);
382 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
384 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
386 if (xprt->snd_task != task)
388 spin_lock(&xprt->transport_lock);
389 xprt->ops->release_xprt(xprt, task);
390 spin_unlock(&xprt->transport_lock);
394 * Van Jacobson congestion avoidance. Check if the congestion window
395 * overflowed. Put the task to sleep if this is the case.
398 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
402 trace_xprt_get_cong(xprt, req->rq_task);
403 if (RPCXPRT_CONGESTED(xprt)) {
404 xprt_set_congestion_window_wait(xprt);
408 xprt->cong += RPC_CWNDSCALE;
413 * Adjust the congestion window, and wake up the next task
414 * that has been sleeping due to congestion
417 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
422 xprt->cong -= RPC_CWNDSCALE;
423 xprt_test_and_clear_congestion_window_wait(xprt);
424 trace_xprt_put_cong(xprt, req->rq_task);
425 __xprt_lock_write_next_cong(xprt);
429 * xprt_request_get_cong - Request congestion control credits
430 * @xprt: pointer to transport
431 * @req: pointer to RPC request
433 * Useful for transports that require congestion control.
436 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
442 spin_lock(&xprt->transport_lock);
443 ret = __xprt_get_cong(xprt, req) != 0;
444 spin_unlock(&xprt->transport_lock);
447 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
450 * xprt_release_rqst_cong - housekeeping when request is complete
451 * @task: RPC request that recently completed
453 * Useful for transports that require congestion control.
455 void xprt_release_rqst_cong(struct rpc_task *task)
457 struct rpc_rqst *req = task->tk_rqstp;
459 __xprt_put_cong(req->rq_xprt, req);
461 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
463 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
465 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
466 __xprt_lock_write_next_cong(xprt);
470 * Clear the congestion window wait flag and wake up the next
471 * entry on xprt->sending
474 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
476 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
477 spin_lock(&xprt->transport_lock);
478 __xprt_lock_write_next_cong(xprt);
479 spin_unlock(&xprt->transport_lock);
484 * xprt_adjust_cwnd - adjust transport congestion window
485 * @xprt: pointer to xprt
486 * @task: recently completed RPC request used to adjust window
487 * @result: result code of completed RPC request
489 * The transport code maintains an estimate on the maximum number of out-
490 * standing RPC requests, using a smoothed version of the congestion
491 * avoidance implemented in 44BSD. This is basically the Van Jacobson
492 * congestion algorithm: If a retransmit occurs, the congestion window is
493 * halved; otherwise, it is incremented by 1/cwnd when
495 * - a reply is received and
496 * - a full number of requests are outstanding and
497 * - the congestion window hasn't been updated recently.
499 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
501 struct rpc_rqst *req = task->tk_rqstp;
502 unsigned long cwnd = xprt->cwnd;
504 if (result >= 0 && cwnd <= xprt->cong) {
505 /* The (cwnd >> 1) term makes sure
506 * the result gets rounded properly. */
507 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
508 if (cwnd > RPC_MAXCWND(xprt))
509 cwnd = RPC_MAXCWND(xprt);
510 __xprt_lock_write_next_cong(xprt);
511 } else if (result == -ETIMEDOUT) {
513 if (cwnd < RPC_CWNDSCALE)
514 cwnd = RPC_CWNDSCALE;
516 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
517 xprt->cong, xprt->cwnd, cwnd);
519 __xprt_put_cong(xprt, req);
521 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
524 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
525 * @xprt: transport with waiting tasks
526 * @status: result code to plant in each task before waking it
529 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
532 rpc_wake_up_status(&xprt->pending, status);
534 rpc_wake_up(&xprt->pending);
536 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
539 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
542 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
543 * we don't in general want to force a socket disconnection due to
544 * an incomplete RPC call transmission.
546 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
548 set_bit(XPRT_WRITE_SPACE, &xprt->state);
550 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
553 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
555 if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
556 __xprt_lock_write_next(xprt);
557 dprintk("RPC: write space: waking waiting task on "
565 * xprt_write_space - wake the task waiting for transport output buffer space
566 * @xprt: transport with waiting tasks
568 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
570 bool xprt_write_space(struct rpc_xprt *xprt)
574 if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
576 spin_lock(&xprt->transport_lock);
577 ret = xprt_clear_write_space_locked(xprt);
578 spin_unlock(&xprt->transport_lock);
581 EXPORT_SYMBOL_GPL(xprt_write_space);
583 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
585 s64 delta = ktime_to_ns(ktime_get() - abstime);
586 return likely(delta >= 0) ?
587 jiffies - nsecs_to_jiffies(delta) :
588 jiffies + nsecs_to_jiffies(-delta);
591 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
593 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
594 unsigned long majortimeo = req->rq_timeout;
596 if (to->to_exponential)
597 majortimeo <<= to->to_retries;
599 majortimeo += to->to_increment * to->to_retries;
600 if (majortimeo > to->to_maxval || majortimeo == 0)
601 majortimeo = to->to_maxval;
605 static void xprt_reset_majortimeo(struct rpc_rqst *req)
607 req->rq_majortimeo += xprt_calc_majortimeo(req);
610 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
612 unsigned long time_init;
613 struct rpc_xprt *xprt = req->rq_xprt;
615 if (likely(xprt && xprt_connected(xprt)))
618 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
619 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
620 req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
624 * xprt_adjust_timeout - adjust timeout values for next retransmit
625 * @req: RPC request containing parameters to use for the adjustment
628 int xprt_adjust_timeout(struct rpc_rqst *req)
630 struct rpc_xprt *xprt = req->rq_xprt;
631 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
634 if (time_before(jiffies, req->rq_majortimeo)) {
635 if (to->to_exponential)
636 req->rq_timeout <<= 1;
638 req->rq_timeout += to->to_increment;
639 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
640 req->rq_timeout = to->to_maxval;
643 req->rq_timeout = to->to_initval;
645 xprt_reset_majortimeo(req);
646 /* Reset the RTT counters == "slow start" */
647 spin_lock(&xprt->transport_lock);
648 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
649 spin_unlock(&xprt->transport_lock);
653 if (req->rq_timeout == 0) {
654 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
655 req->rq_timeout = 5 * HZ;
660 static void xprt_autoclose(struct work_struct *work)
662 struct rpc_xprt *xprt =
663 container_of(work, struct rpc_xprt, task_cleanup);
664 unsigned int pflags = memalloc_nofs_save();
666 trace_xprt_disconnect_auto(xprt);
667 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
668 xprt->ops->close(xprt);
669 xprt_release_write(xprt, NULL);
670 wake_up_bit(&xprt->state, XPRT_LOCKED);
671 memalloc_nofs_restore(pflags);
675 * xprt_disconnect_done - mark a transport as disconnected
676 * @xprt: transport to flag for disconnect
679 void xprt_disconnect_done(struct rpc_xprt *xprt)
681 trace_xprt_disconnect_done(xprt);
682 spin_lock(&xprt->transport_lock);
683 xprt_clear_connected(xprt);
684 xprt_clear_write_space_locked(xprt);
685 xprt_clear_congestion_window_wait_locked(xprt);
686 xprt_wake_pending_tasks(xprt, -ENOTCONN);
687 spin_unlock(&xprt->transport_lock);
689 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
692 * xprt_force_disconnect - force a transport to disconnect
693 * @xprt: transport to disconnect
696 void xprt_force_disconnect(struct rpc_xprt *xprt)
698 trace_xprt_disconnect_force(xprt);
700 /* Don't race with the test_bit() in xprt_clear_locked() */
701 spin_lock(&xprt->transport_lock);
702 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
703 /* Try to schedule an autoclose RPC call */
704 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
705 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
706 else if (xprt->snd_task)
707 rpc_wake_up_queued_task_set_status(&xprt->pending,
708 xprt->snd_task, -ENOTCONN);
709 spin_unlock(&xprt->transport_lock);
711 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
714 xprt_connect_cookie(struct rpc_xprt *xprt)
716 return READ_ONCE(xprt->connect_cookie);
720 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
722 struct rpc_rqst *req = task->tk_rqstp;
723 struct rpc_xprt *xprt = req->rq_xprt;
725 return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
726 !xprt_connected(xprt);
730 * xprt_conditional_disconnect - force a transport to disconnect
731 * @xprt: transport to disconnect
732 * @cookie: 'connection cookie'
734 * This attempts to break the connection if and only if 'cookie' matches
735 * the current transport 'connection cookie'. It ensures that we don't
736 * try to break the connection more than once when we need to retransmit
737 * a batch of RPC requests.
740 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
742 /* Don't race with the test_bit() in xprt_clear_locked() */
743 spin_lock(&xprt->transport_lock);
744 if (cookie != xprt->connect_cookie)
746 if (test_bit(XPRT_CLOSING, &xprt->state))
748 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
749 /* Try to schedule an autoclose RPC call */
750 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
751 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
752 xprt_wake_pending_tasks(xprt, -EAGAIN);
754 spin_unlock(&xprt->transport_lock);
758 xprt_has_timer(const struct rpc_xprt *xprt)
760 return xprt->idle_timeout != 0;
764 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
765 __must_hold(&xprt->transport_lock)
767 xprt->last_used = jiffies;
768 if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
769 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
773 xprt_init_autodisconnect(struct timer_list *t)
775 struct rpc_xprt *xprt = from_timer(xprt, t, timer);
777 if (!RB_EMPTY_ROOT(&xprt->recv_queue))
779 /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
780 xprt->last_used = jiffies;
781 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
783 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
786 bool xprt_lock_connect(struct rpc_xprt *xprt,
787 struct rpc_task *task,
792 spin_lock(&xprt->transport_lock);
793 if (!test_bit(XPRT_LOCKED, &xprt->state))
795 if (xprt->snd_task != task)
797 xprt->snd_task = cookie;
800 spin_unlock(&xprt->transport_lock);
804 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
806 spin_lock(&xprt->transport_lock);
807 if (xprt->snd_task != cookie)
809 if (!test_bit(XPRT_LOCKED, &xprt->state))
811 xprt->snd_task =NULL;
812 xprt->ops->release_xprt(xprt, NULL);
813 xprt_schedule_autodisconnect(xprt);
815 spin_unlock(&xprt->transport_lock);
816 wake_up_bit(&xprt->state, XPRT_LOCKED);
820 * xprt_connect - schedule a transport connect operation
821 * @task: RPC task that is requesting the connect
824 void xprt_connect(struct rpc_task *task)
826 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
828 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
829 xprt, (xprt_connected(xprt) ? "is" : "is not"));
831 if (!xprt_bound(xprt)) {
832 task->tk_status = -EAGAIN;
835 if (!xprt_lock_write(xprt, task))
838 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
839 trace_xprt_disconnect_cleanup(xprt);
840 xprt->ops->close(xprt);
843 if (!xprt_connected(xprt)) {
844 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
845 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
846 xprt_request_timeout(task->tk_rqstp));
848 if (test_bit(XPRT_CLOSING, &xprt->state))
850 if (xprt_test_and_set_connecting(xprt))
853 if (!xprt_connected(xprt)) {
854 xprt->stat.connect_start = jiffies;
855 xprt->ops->connect(xprt, task);
857 xprt_clear_connecting(xprt);
859 rpc_wake_up_queued_task(&xprt->pending, task);
862 xprt_release_write(xprt, task);
866 * xprt_reconnect_delay - compute the wait before scheduling a connect
867 * @xprt: transport instance
870 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
872 unsigned long start, now = jiffies;
874 start = xprt->stat.connect_start + xprt->reestablish_timeout;
875 if (time_after(start, now))
879 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
882 * xprt_reconnect_backoff - compute the new re-establish timeout
883 * @xprt: transport instance
884 * @init_to: initial reestablish timeout
887 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
889 xprt->reestablish_timeout <<= 1;
890 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
891 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
892 if (xprt->reestablish_timeout < init_to)
893 xprt->reestablish_timeout = init_to;
895 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
897 enum xprt_xid_rb_cmp {
902 static enum xprt_xid_rb_cmp
903 xprt_xid_cmp(__be32 xid1, __be32 xid2)
907 if ((__force u32)xid1 < (__force u32)xid2)
912 static struct rpc_rqst *
913 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
915 struct rb_node *n = xprt->recv_queue.rb_node;
916 struct rpc_rqst *req;
919 req = rb_entry(n, struct rpc_rqst, rq_recv);
920 switch (xprt_xid_cmp(xid, req->rq_xid)) {
935 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
937 struct rb_node **p = &xprt->recv_queue.rb_node;
938 struct rb_node *n = NULL;
939 struct rpc_rqst *req;
943 req = rb_entry(n, struct rpc_rqst, rq_recv);
944 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
952 WARN_ON_ONCE(new != req);
956 rb_link_node(&new->rq_recv, n, p);
957 rb_insert_color(&new->rq_recv, &xprt->recv_queue);
961 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
963 rb_erase(&req->rq_recv, &xprt->recv_queue);
967 * xprt_lookup_rqst - find an RPC request corresponding to an XID
968 * @xprt: transport on which the original request was transmitted
969 * @xid: RPC XID of incoming reply
971 * Caller holds xprt->queue_lock.
973 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
975 struct rpc_rqst *entry;
977 entry = xprt_request_rb_find(xprt, xid);
979 trace_xprt_lookup_rqst(xprt, xid, 0);
980 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
984 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
986 trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
987 xprt->stat.bad_xids++;
990 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
993 xprt_is_pinned_rqst(struct rpc_rqst *req)
995 return atomic_read(&req->rq_pin) != 0;
999 * xprt_pin_rqst - Pin a request on the transport receive list
1000 * @req: Request to pin
1002 * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1003 * so should be holding xprt->queue_lock.
1005 void xprt_pin_rqst(struct rpc_rqst *req)
1007 atomic_inc(&req->rq_pin);
1009 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1012 * xprt_unpin_rqst - Unpin a request on the transport receive list
1013 * @req: Request to pin
1015 * Caller should be holding xprt->queue_lock.
1017 void xprt_unpin_rqst(struct rpc_rqst *req)
1019 if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1020 atomic_dec(&req->rq_pin);
1023 if (atomic_dec_and_test(&req->rq_pin))
1024 wake_up_var(&req->rq_pin);
1026 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1028 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1030 wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1034 xprt_request_data_received(struct rpc_task *task)
1036 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1037 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1041 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1043 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1044 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1048 * xprt_request_enqueue_receive - Add an request to the receive queue
1053 xprt_request_enqueue_receive(struct rpc_task *task)
1055 struct rpc_rqst *req = task->tk_rqstp;
1056 struct rpc_xprt *xprt = req->rq_xprt;
1058 if (!xprt_request_need_enqueue_receive(task, req))
1061 xprt_request_prepare(task->tk_rqstp);
1062 spin_lock(&xprt->queue_lock);
1064 /* Update the softirq receive buffer */
1065 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1066 sizeof(req->rq_private_buf));
1068 /* Add request to the receive list */
1069 xprt_request_rb_insert(xprt, req);
1070 set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1071 spin_unlock(&xprt->queue_lock);
1073 /* Turn off autodisconnect */
1074 del_singleshot_timer_sync(&xprt->timer);
1078 * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1081 * Caller must hold xprt->queue_lock.
1084 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1086 struct rpc_rqst *req = task->tk_rqstp;
1088 if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1089 xprt_request_rb_remove(req->rq_xprt, req);
1093 * xprt_update_rtt - Update RPC RTT statistics
1094 * @task: RPC request that recently completed
1096 * Caller holds xprt->queue_lock.
1098 void xprt_update_rtt(struct rpc_task *task)
1100 struct rpc_rqst *req = task->tk_rqstp;
1101 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1102 unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1103 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1106 if (req->rq_ntrans == 1)
1107 rpc_update_rtt(rtt, timer, m);
1108 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1111 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1114 * xprt_complete_rqst - called when reply processing is complete
1115 * @task: RPC request that recently completed
1116 * @copied: actual number of bytes received from the transport
1118 * Caller holds xprt->queue_lock.
1120 void xprt_complete_rqst(struct rpc_task *task, int copied)
1122 struct rpc_rqst *req = task->tk_rqstp;
1123 struct rpc_xprt *xprt = req->rq_xprt;
1125 trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
1129 req->rq_private_buf.len = copied;
1130 /* Ensure all writes are done before we update */
1131 /* req->rq_reply_bytes_recvd */
1133 req->rq_reply_bytes_recvd = copied;
1134 xprt_request_dequeue_receive_locked(task);
1135 rpc_wake_up_queued_task(&xprt->pending, task);
1137 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1139 static void xprt_timer(struct rpc_task *task)
1141 struct rpc_rqst *req = task->tk_rqstp;
1142 struct rpc_xprt *xprt = req->rq_xprt;
1144 if (task->tk_status != -ETIMEDOUT)
1147 trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1148 if (!req->rq_reply_bytes_recvd) {
1149 if (xprt->ops->timer)
1150 xprt->ops->timer(xprt, task);
1152 task->tk_status = 0;
1156 * xprt_wait_for_reply_request_def - wait for reply
1157 * @task: pointer to rpc_task
1159 * Set a request's retransmit timeout based on the transport's
1160 * default timeout parameters. Used by transports that don't adjust
1161 * the retransmit timeout based on round-trip time estimation,
1162 * and put the task to sleep on the pending queue.
1164 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1166 struct rpc_rqst *req = task->tk_rqstp;
1168 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1169 xprt_request_timeout(req));
1171 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1174 * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1175 * @task: pointer to rpc_task
1177 * Set a request's retransmit timeout using the RTT estimator,
1178 * and put the task to sleep on the pending queue.
1180 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1182 int timer = task->tk_msg.rpc_proc->p_timer;
1183 struct rpc_clnt *clnt = task->tk_client;
1184 struct rpc_rtt *rtt = clnt->cl_rtt;
1185 struct rpc_rqst *req = task->tk_rqstp;
1186 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1187 unsigned long timeout;
1189 timeout = rpc_calc_rto(rtt, timer);
1190 timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1191 if (timeout > max_timeout || timeout == 0)
1192 timeout = max_timeout;
1193 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1196 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1199 * xprt_request_wait_receive - wait for the reply to an RPC request
1200 * @task: RPC task about to send a request
1203 void xprt_request_wait_receive(struct rpc_task *task)
1205 struct rpc_rqst *req = task->tk_rqstp;
1206 struct rpc_xprt *xprt = req->rq_xprt;
1208 if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1211 * Sleep on the pending queue if we're expecting a reply.
1212 * The spinlock ensures atomicity between the test of
1213 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1215 spin_lock(&xprt->queue_lock);
1216 if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1217 xprt->ops->wait_for_reply_request(task);
1219 * Send an extra queue wakeup call if the
1220 * connection was dropped in case the call to
1221 * rpc_sleep_on() raced.
1223 if (xprt_request_retransmit_after_disconnect(task))
1224 rpc_wake_up_queued_task_set_status(&xprt->pending,
1227 spin_unlock(&xprt->queue_lock);
1231 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1233 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1237 * xprt_request_enqueue_transmit - queue a task for transmission
1238 * @task: pointer to rpc_task
1240 * Add a task to the transmission queue.
1243 xprt_request_enqueue_transmit(struct rpc_task *task)
1245 struct rpc_rqst *pos, *req = task->tk_rqstp;
1246 struct rpc_xprt *xprt = req->rq_xprt;
1248 if (xprt_request_need_enqueue_transmit(task, req)) {
1249 req->rq_bytes_sent = 0;
1250 spin_lock(&xprt->queue_lock);
1252 * Requests that carry congestion control credits are added
1253 * to the head of the list to avoid starvation issues.
1256 xprt_clear_congestion_window_wait(xprt);
1257 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1260 /* Note: req is added _before_ pos */
1261 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1262 INIT_LIST_HEAD(&req->rq_xmit2);
1263 trace_xprt_enq_xmit(task, 1);
1266 } else if (RPC_IS_SWAPPER(task)) {
1267 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1268 if (pos->rq_cong || pos->rq_bytes_sent)
1270 if (RPC_IS_SWAPPER(pos->rq_task))
1272 /* Note: req is added _before_ pos */
1273 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1274 INIT_LIST_HEAD(&req->rq_xmit2);
1275 trace_xprt_enq_xmit(task, 2);
1278 } else if (!req->rq_seqno) {
1279 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1280 if (pos->rq_task->tk_owner != task->tk_owner)
1282 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1283 INIT_LIST_HEAD(&req->rq_xmit);
1284 trace_xprt_enq_xmit(task, 3);
1288 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1289 INIT_LIST_HEAD(&req->rq_xmit2);
1290 trace_xprt_enq_xmit(task, 4);
1292 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1293 spin_unlock(&xprt->queue_lock);
1298 * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1299 * @task: pointer to rpc_task
1301 * Remove a task from the transmission queue
1302 * Caller must hold xprt->queue_lock
1305 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1307 struct rpc_rqst *req = task->tk_rqstp;
1309 if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1311 if (!list_empty(&req->rq_xmit)) {
1312 list_del(&req->rq_xmit);
1313 if (!list_empty(&req->rq_xmit2)) {
1314 struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1315 struct rpc_rqst, rq_xmit2);
1316 list_del(&req->rq_xmit2);
1317 list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1320 list_del(&req->rq_xmit2);
1324 * xprt_request_dequeue_transmit - remove a task from the transmission queue
1325 * @task: pointer to rpc_task
1327 * Remove a task from the transmission queue
1330 xprt_request_dequeue_transmit(struct rpc_task *task)
1332 struct rpc_rqst *req = task->tk_rqstp;
1333 struct rpc_xprt *xprt = req->rq_xprt;
1335 spin_lock(&xprt->queue_lock);
1336 xprt_request_dequeue_transmit_locked(task);
1337 spin_unlock(&xprt->queue_lock);
1341 * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1342 * @task: pointer to rpc_task
1344 * Remove a task from the transmit and receive queues, and ensure that
1345 * it is not pinned by the receive work item.
1348 xprt_request_dequeue_xprt(struct rpc_task *task)
1350 struct rpc_rqst *req = task->tk_rqstp;
1351 struct rpc_xprt *xprt = req->rq_xprt;
1353 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1354 test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1355 xprt_is_pinned_rqst(req)) {
1356 spin_lock(&xprt->queue_lock);
1357 xprt_request_dequeue_transmit_locked(task);
1358 xprt_request_dequeue_receive_locked(task);
1359 while (xprt_is_pinned_rqst(req)) {
1360 set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1361 spin_unlock(&xprt->queue_lock);
1362 xprt_wait_on_pinned_rqst(req);
1363 spin_lock(&xprt->queue_lock);
1364 clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1366 spin_unlock(&xprt->queue_lock);
1371 * xprt_request_prepare - prepare an encoded request for transport
1372 * @req: pointer to rpc_rqst
1374 * Calls into the transport layer to do whatever is needed to prepare
1375 * the request for transmission or receive.
1378 xprt_request_prepare(struct rpc_rqst *req)
1380 struct rpc_xprt *xprt = req->rq_xprt;
1382 if (xprt->ops->prepare_request)
1383 xprt->ops->prepare_request(req);
1387 * xprt_request_need_retransmit - Test if a task needs retransmission
1388 * @task: pointer to rpc_task
1390 * Test for whether a connection breakage requires the task to retransmit
1393 xprt_request_need_retransmit(struct rpc_task *task)
1395 return xprt_request_retransmit_after_disconnect(task);
1399 * xprt_prepare_transmit - reserve the transport before sending a request
1400 * @task: RPC task about to send a request
1403 bool xprt_prepare_transmit(struct rpc_task *task)
1405 struct rpc_rqst *req = task->tk_rqstp;
1406 struct rpc_xprt *xprt = req->rq_xprt;
1408 dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1410 if (!xprt_lock_write(xprt, task)) {
1411 /* Race breaker: someone may have transmitted us */
1412 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1413 rpc_wake_up_queued_task_set_status(&xprt->sending,
1421 void xprt_end_transmit(struct rpc_task *task)
1423 xprt_release_write(task->tk_rqstp->rq_xprt, task);
1427 * xprt_request_transmit - send an RPC request on a transport
1428 * @req: pointer to request to transmit
1429 * @snd_task: RPC task that owns the transport lock
1431 * This performs the transmission of a single request.
1432 * Note that if the request is not the same as snd_task, then it
1433 * does need to be pinned.
1434 * Returns '0' on success.
1437 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1439 struct rpc_xprt *xprt = req->rq_xprt;
1440 struct rpc_task *task = req->rq_task;
1441 unsigned int connect_cookie;
1442 int is_retrans = RPC_WAS_SENT(task);
1445 if (!req->rq_bytes_sent) {
1446 if (xprt_request_data_received(task)) {
1450 /* Verify that our message lies in the RPCSEC_GSS window */
1451 if (rpcauth_xmit_need_reencode(task)) {
1455 if (RPC_SIGNALLED(task)) {
1456 status = -ERESTARTSYS;
1462 * Update req->rq_ntrans before transmitting to avoid races with
1463 * xprt_update_rtt(), which needs to know that it is recording a
1464 * reply to the first transmission.
1468 trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1469 connect_cookie = xprt->connect_cookie;
1470 status = xprt->ops->send_request(req);
1473 trace_xprt_transmit(req, status);
1478 task->tk_client->cl_stats->rpcretrans++;
1480 xprt_inject_disconnect(xprt);
1482 task->tk_flags |= RPC_TASK_SENT;
1483 spin_lock(&xprt->transport_lock);
1486 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1487 xprt->stat.bklog_u += xprt->backlog.qlen;
1488 xprt->stat.sending_u += xprt->sending.qlen;
1489 xprt->stat.pending_u += xprt->pending.qlen;
1490 spin_unlock(&xprt->transport_lock);
1492 req->rq_connect_cookie = connect_cookie;
1494 trace_xprt_transmit(req, status);
1495 xprt_request_dequeue_transmit(task);
1496 rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1501 * xprt_transmit - send an RPC request on a transport
1502 * @task: controlling RPC task
1504 * Attempts to drain the transmit queue. On exit, either the transport
1505 * signalled an error that needs to be handled before transmission can
1506 * resume, or @task finished transmitting, and detected that it already
1510 xprt_transmit(struct rpc_task *task)
1512 struct rpc_rqst *next, *req = task->tk_rqstp;
1513 struct rpc_xprt *xprt = req->rq_xprt;
1516 spin_lock(&xprt->queue_lock);
1517 while (!list_empty(&xprt->xmit_queue)) {
1518 next = list_first_entry(&xprt->xmit_queue,
1519 struct rpc_rqst, rq_xmit);
1520 xprt_pin_rqst(next);
1521 spin_unlock(&xprt->queue_lock);
1522 status = xprt_request_transmit(next, task);
1523 if (status == -EBADMSG && next != req)
1526 spin_lock(&xprt->queue_lock);
1527 xprt_unpin_rqst(next);
1529 if (!xprt_request_data_received(task) ||
1530 test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1532 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1533 task->tk_status = status;
1536 spin_unlock(&xprt->queue_lock);
1539 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1541 set_bit(XPRT_CONGESTED, &xprt->state);
1542 rpc_sleep_on(&xprt->backlog, task, NULL);
1545 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1547 if (rpc_wake_up_next(&xprt->backlog) == NULL)
1548 clear_bit(XPRT_CONGESTED, &xprt->state);
1551 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1555 if (!test_bit(XPRT_CONGESTED, &xprt->state))
1557 spin_lock(&xprt->reserve_lock);
1558 if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1559 rpc_sleep_on(&xprt->backlog, task, NULL);
1562 spin_unlock(&xprt->reserve_lock);
1567 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1569 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1571 if (xprt->num_reqs >= xprt->max_reqs)
1574 spin_unlock(&xprt->reserve_lock);
1575 req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1576 spin_lock(&xprt->reserve_lock);
1580 req = ERR_PTR(-ENOMEM);
1585 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1587 if (xprt->num_reqs > xprt->min_reqs) {
1595 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1597 struct rpc_rqst *req;
1599 spin_lock(&xprt->reserve_lock);
1600 if (!list_empty(&xprt->free)) {
1601 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1602 list_del(&req->rq_list);
1605 req = xprt_dynamic_alloc_slot(xprt);
1608 switch (PTR_ERR(req)) {
1610 dprintk("RPC: dynamic allocation of request slot "
1611 "failed! Retrying\n");
1612 task->tk_status = -ENOMEM;
1615 xprt_add_backlog(xprt, task);
1616 dprintk("RPC: waiting for request slot\n");
1619 task->tk_status = -EAGAIN;
1621 spin_unlock(&xprt->reserve_lock);
1624 xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1626 spin_unlock(&xprt->reserve_lock);
1628 task->tk_status = 0;
1629 task->tk_rqstp = req;
1631 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1633 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1635 spin_lock(&xprt->reserve_lock);
1636 if (!xprt_dynamic_free_slot(xprt, req)) {
1637 memset(req, 0, sizeof(*req)); /* mark unused */
1638 list_add(&req->rq_list, &xprt->free);
1640 xprt_wake_up_backlog(xprt);
1641 spin_unlock(&xprt->reserve_lock);
1643 EXPORT_SYMBOL_GPL(xprt_free_slot);
1645 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1647 struct rpc_rqst *req;
1648 while (!list_empty(&xprt->free)) {
1649 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1650 list_del(&req->rq_list);
1655 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1656 unsigned int num_prealloc,
1657 unsigned int max_alloc)
1659 struct rpc_xprt *xprt;
1660 struct rpc_rqst *req;
1663 xprt = kzalloc(size, GFP_KERNEL);
1667 xprt_init(xprt, net);
1669 for (i = 0; i < num_prealloc; i++) {
1670 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1673 list_add(&req->rq_list, &xprt->free);
1675 if (max_alloc > num_prealloc)
1676 xprt->max_reqs = max_alloc;
1678 xprt->max_reqs = num_prealloc;
1679 xprt->min_reqs = num_prealloc;
1680 xprt->num_reqs = num_prealloc;
1689 EXPORT_SYMBOL_GPL(xprt_alloc);
1691 void xprt_free(struct rpc_xprt *xprt)
1693 put_net(xprt->xprt_net);
1694 xprt_free_all_slots(xprt);
1695 kfree_rcu(xprt, rcu);
1697 EXPORT_SYMBOL_GPL(xprt_free);
1700 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1702 req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1706 xprt_alloc_xid(struct rpc_xprt *xprt)
1710 spin_lock(&xprt->reserve_lock);
1711 xid = (__force __be32)xprt->xid++;
1712 spin_unlock(&xprt->reserve_lock);
1717 xprt_init_xid(struct rpc_xprt *xprt)
1719 xprt->xid = prandom_u32();
1723 xprt_request_init(struct rpc_task *task)
1725 struct rpc_xprt *xprt = task->tk_xprt;
1726 struct rpc_rqst *req = task->tk_rqstp;
1728 req->rq_task = task;
1729 req->rq_xprt = xprt;
1730 req->rq_buffer = NULL;
1731 req->rq_xid = xprt_alloc_xid(xprt);
1732 xprt_init_connect_cookie(req, xprt);
1733 req->rq_snd_buf.len = 0;
1734 req->rq_snd_buf.buflen = 0;
1735 req->rq_rcv_buf.len = 0;
1736 req->rq_rcv_buf.buflen = 0;
1737 req->rq_snd_buf.bvec = NULL;
1738 req->rq_rcv_buf.bvec = NULL;
1739 req->rq_release_snd_buf = NULL;
1740 xprt_init_majortimeo(task, req);
1741 dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1742 req, ntohl(req->rq_xid));
1746 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1748 xprt->ops->alloc_slot(xprt, task);
1749 if (task->tk_rqstp != NULL)
1750 xprt_request_init(task);
1754 * xprt_reserve - allocate an RPC request slot
1755 * @task: RPC task requesting a slot allocation
1757 * If the transport is marked as being congested, or if no more
1758 * slots are available, place the task on the transport's
1761 void xprt_reserve(struct rpc_task *task)
1763 struct rpc_xprt *xprt = task->tk_xprt;
1765 task->tk_status = 0;
1766 if (task->tk_rqstp != NULL)
1769 task->tk_status = -EAGAIN;
1770 if (!xprt_throttle_congested(xprt, task))
1771 xprt_do_reserve(xprt, task);
1775 * xprt_retry_reserve - allocate an RPC request slot
1776 * @task: RPC task requesting a slot allocation
1778 * If no more slots are available, place the task on the transport's
1780 * Note that the only difference with xprt_reserve is that we now
1781 * ignore the value of the XPRT_CONGESTED flag.
1783 void xprt_retry_reserve(struct rpc_task *task)
1785 struct rpc_xprt *xprt = task->tk_xprt;
1787 task->tk_status = 0;
1788 if (task->tk_rqstp != NULL)
1791 task->tk_status = -EAGAIN;
1792 xprt_do_reserve(xprt, task);
1796 * xprt_release - release an RPC request slot
1797 * @task: task which is finished with the slot
1800 void xprt_release(struct rpc_task *task)
1802 struct rpc_xprt *xprt;
1803 struct rpc_rqst *req = task->tk_rqstp;
1806 if (task->tk_client) {
1807 xprt = task->tk_xprt;
1808 xprt_release_write(xprt, task);
1813 xprt = req->rq_xprt;
1814 xprt_request_dequeue_xprt(task);
1815 spin_lock(&xprt->transport_lock);
1816 xprt->ops->release_xprt(xprt, task);
1817 if (xprt->ops->release_request)
1818 xprt->ops->release_request(task);
1819 xprt_schedule_autodisconnect(xprt);
1820 spin_unlock(&xprt->transport_lock);
1822 xprt->ops->buf_free(task);
1823 xprt_inject_disconnect(xprt);
1824 xdr_free_bvec(&req->rq_rcv_buf);
1825 xdr_free_bvec(&req->rq_snd_buf);
1826 if (req->rq_cred != NULL)
1827 put_rpccred(req->rq_cred);
1828 task->tk_rqstp = NULL;
1829 if (req->rq_release_snd_buf)
1830 req->rq_release_snd_buf(req);
1832 dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1833 if (likely(!bc_prealloc(req)))
1834 xprt->ops->free_slot(xprt, req);
1836 xprt_free_bc_request(req);
1839 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1841 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1843 struct xdr_buf *xbufp = &req->rq_snd_buf;
1845 task->tk_rqstp = req;
1846 req->rq_task = task;
1847 xprt_init_connect_cookie(req, req->rq_xprt);
1849 * Set up the xdr_buf length.
1850 * This also indicates that the buffer is XDR encoded already.
1852 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1853 xbufp->tail[0].iov_len;
1857 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1859 kref_init(&xprt->kref);
1861 spin_lock_init(&xprt->transport_lock);
1862 spin_lock_init(&xprt->reserve_lock);
1863 spin_lock_init(&xprt->queue_lock);
1865 INIT_LIST_HEAD(&xprt->free);
1866 xprt->recv_queue = RB_ROOT;
1867 INIT_LIST_HEAD(&xprt->xmit_queue);
1868 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1869 spin_lock_init(&xprt->bc_pa_lock);
1870 INIT_LIST_HEAD(&xprt->bc_pa_list);
1871 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1872 INIT_LIST_HEAD(&xprt->xprt_switch);
1874 xprt->last_used = jiffies;
1875 xprt->cwnd = RPC_INITCWND;
1876 xprt->bind_index = 0;
1878 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1879 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1880 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1881 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1883 xprt_init_xid(xprt);
1885 xprt->xprt_net = get_net(net);
1889 * xprt_create_transport - create an RPC transport
1890 * @args: rpc transport creation arguments
1893 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1895 struct rpc_xprt *xprt;
1896 struct xprt_class *t;
1898 spin_lock(&xprt_list_lock);
1899 list_for_each_entry(t, &xprt_list, list) {
1900 if (t->ident == args->ident) {
1901 spin_unlock(&xprt_list_lock);
1905 spin_unlock(&xprt_list_lock);
1906 dprintk("RPC: transport (%d) not supported\n", args->ident);
1907 return ERR_PTR(-EIO);
1910 xprt = t->setup(args);
1913 if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1914 xprt->idle_timeout = 0;
1915 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1916 if (xprt_has_timer(xprt))
1917 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1919 timer_setup(&xprt->timer, NULL, 0);
1921 if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1923 return ERR_PTR(-EINVAL);
1925 xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1926 if (xprt->servername == NULL) {
1928 return ERR_PTR(-ENOMEM);
1931 rpc_xprt_debugfs_register(xprt);
1933 trace_xprt_create(xprt);
1938 static void xprt_destroy_cb(struct work_struct *work)
1940 struct rpc_xprt *xprt =
1941 container_of(work, struct rpc_xprt, task_cleanup);
1943 trace_xprt_destroy(xprt);
1945 rpc_xprt_debugfs_unregister(xprt);
1946 rpc_destroy_wait_queue(&xprt->binding);
1947 rpc_destroy_wait_queue(&xprt->pending);
1948 rpc_destroy_wait_queue(&xprt->sending);
1949 rpc_destroy_wait_queue(&xprt->backlog);
1950 kfree(xprt->servername);
1952 * Destroy any existing back channel
1954 xprt_destroy_backchannel(xprt, UINT_MAX);
1957 * Tear down transport state and free the rpc_xprt
1959 xprt->ops->destroy(xprt);
1963 * xprt_destroy - destroy an RPC transport, killing off all requests.
1964 * @xprt: transport to destroy
1967 static void xprt_destroy(struct rpc_xprt *xprt)
1970 * Exclude transport connect/disconnect handlers and autoclose
1972 wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1974 del_timer_sync(&xprt->timer);
1977 * Destroy sockets etc from the system workqueue so they can
1978 * safely flush receive work running on rpciod.
1980 INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1981 schedule_work(&xprt->task_cleanup);
1984 static void xprt_destroy_kref(struct kref *kref)
1986 xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1990 * xprt_get - return a reference to an RPC transport.
1991 * @xprt: pointer to the transport
1994 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1996 if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2000 EXPORT_SYMBOL_GPL(xprt_get);
2003 * xprt_put - release a reference to an RPC transport.
2004 * @xprt: pointer to the transport
2007 void xprt_put(struct rpc_xprt *xprt)
2010 kref_put(&xprt->kref, xprt_destroy_kref);
2012 EXPORT_SYMBOL_GPL(xprt_put);