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_reset_minortimeo(struct rpc_rqst *req)
612 req->rq_minortimeo += req->rq_timeout;
615 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
617 unsigned long time_init;
618 struct rpc_xprt *xprt = req->rq_xprt;
620 if (likely(xprt && xprt_connected(xprt)))
623 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
624 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
625 req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
626 req->rq_minortimeo = time_init + req->rq_timeout;
630 * xprt_adjust_timeout - adjust timeout values for next retransmit
631 * @req: RPC request containing parameters to use for the adjustment
634 int xprt_adjust_timeout(struct rpc_rqst *req)
636 struct rpc_xprt *xprt = req->rq_xprt;
637 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
640 if (time_before(jiffies, req->rq_minortimeo))
642 if (time_before(jiffies, req->rq_majortimeo)) {
643 if (to->to_exponential)
644 req->rq_timeout <<= 1;
646 req->rq_timeout += to->to_increment;
647 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
648 req->rq_timeout = to->to_maxval;
651 req->rq_timeout = to->to_initval;
653 xprt_reset_majortimeo(req);
654 /* Reset the RTT counters == "slow start" */
655 spin_lock(&xprt->transport_lock);
656 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
657 spin_unlock(&xprt->transport_lock);
660 xprt_reset_minortimeo(req);
662 if (req->rq_timeout == 0) {
663 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
664 req->rq_timeout = 5 * HZ;
669 static void xprt_autoclose(struct work_struct *work)
671 struct rpc_xprt *xprt =
672 container_of(work, struct rpc_xprt, task_cleanup);
673 unsigned int pflags = memalloc_nofs_save();
675 trace_xprt_disconnect_auto(xprt);
676 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
677 xprt->ops->close(xprt);
678 xprt_release_write(xprt, NULL);
679 wake_up_bit(&xprt->state, XPRT_LOCKED);
680 memalloc_nofs_restore(pflags);
684 * xprt_disconnect_done - mark a transport as disconnected
685 * @xprt: transport to flag for disconnect
688 void xprt_disconnect_done(struct rpc_xprt *xprt)
690 trace_xprt_disconnect_done(xprt);
691 spin_lock(&xprt->transport_lock);
692 xprt_clear_connected(xprt);
693 xprt_clear_write_space_locked(xprt);
694 xprt_clear_congestion_window_wait_locked(xprt);
695 xprt_wake_pending_tasks(xprt, -ENOTCONN);
696 spin_unlock(&xprt->transport_lock);
698 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
701 * xprt_force_disconnect - force a transport to disconnect
702 * @xprt: transport to disconnect
705 void xprt_force_disconnect(struct rpc_xprt *xprt)
707 trace_xprt_disconnect_force(xprt);
709 /* Don't race with the test_bit() in xprt_clear_locked() */
710 spin_lock(&xprt->transport_lock);
711 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
712 /* Try to schedule an autoclose RPC call */
713 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
714 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
715 else if (xprt->snd_task)
716 rpc_wake_up_queued_task_set_status(&xprt->pending,
717 xprt->snd_task, -ENOTCONN);
718 spin_unlock(&xprt->transport_lock);
720 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
723 xprt_connect_cookie(struct rpc_xprt *xprt)
725 return READ_ONCE(xprt->connect_cookie);
729 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
731 struct rpc_rqst *req = task->tk_rqstp;
732 struct rpc_xprt *xprt = req->rq_xprt;
734 return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
735 !xprt_connected(xprt);
739 * xprt_conditional_disconnect - force a transport to disconnect
740 * @xprt: transport to disconnect
741 * @cookie: 'connection cookie'
743 * This attempts to break the connection if and only if 'cookie' matches
744 * the current transport 'connection cookie'. It ensures that we don't
745 * try to break the connection more than once when we need to retransmit
746 * a batch of RPC requests.
749 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
751 /* Don't race with the test_bit() in xprt_clear_locked() */
752 spin_lock(&xprt->transport_lock);
753 if (cookie != xprt->connect_cookie)
755 if (test_bit(XPRT_CLOSING, &xprt->state))
757 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
758 /* Try to schedule an autoclose RPC call */
759 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
760 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
761 xprt_wake_pending_tasks(xprt, -EAGAIN);
763 spin_unlock(&xprt->transport_lock);
767 xprt_has_timer(const struct rpc_xprt *xprt)
769 return xprt->idle_timeout != 0;
773 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
774 __must_hold(&xprt->transport_lock)
776 xprt->last_used = jiffies;
777 if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
778 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
782 xprt_init_autodisconnect(struct timer_list *t)
784 struct rpc_xprt *xprt = from_timer(xprt, t, timer);
786 if (!RB_EMPTY_ROOT(&xprt->recv_queue))
788 /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
789 xprt->last_used = jiffies;
790 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
792 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
795 bool xprt_lock_connect(struct rpc_xprt *xprt,
796 struct rpc_task *task,
801 spin_lock(&xprt->transport_lock);
802 if (!test_bit(XPRT_LOCKED, &xprt->state))
804 if (xprt->snd_task != task)
806 xprt->snd_task = cookie;
809 spin_unlock(&xprt->transport_lock);
813 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
815 spin_lock(&xprt->transport_lock);
816 if (xprt->snd_task != cookie)
818 if (!test_bit(XPRT_LOCKED, &xprt->state))
820 xprt->snd_task =NULL;
821 xprt->ops->release_xprt(xprt, NULL);
822 xprt_schedule_autodisconnect(xprt);
824 spin_unlock(&xprt->transport_lock);
825 wake_up_bit(&xprt->state, XPRT_LOCKED);
829 * xprt_connect - schedule a transport connect operation
830 * @task: RPC task that is requesting the connect
833 void xprt_connect(struct rpc_task *task)
835 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
837 trace_xprt_connect(xprt);
839 if (!xprt_bound(xprt)) {
840 task->tk_status = -EAGAIN;
843 if (!xprt_lock_write(xprt, task))
846 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
847 trace_xprt_disconnect_cleanup(xprt);
848 xprt->ops->close(xprt);
851 if (!xprt_connected(xprt)) {
852 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
853 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
854 xprt_request_timeout(task->tk_rqstp));
856 if (test_bit(XPRT_CLOSING, &xprt->state))
858 if (xprt_test_and_set_connecting(xprt))
861 if (!xprt_connected(xprt)) {
862 xprt->stat.connect_start = jiffies;
863 xprt->ops->connect(xprt, task);
865 xprt_clear_connecting(xprt);
867 rpc_wake_up_queued_task(&xprt->pending, task);
870 xprt_release_write(xprt, task);
874 * xprt_reconnect_delay - compute the wait before scheduling a connect
875 * @xprt: transport instance
878 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
880 unsigned long start, now = jiffies;
882 start = xprt->stat.connect_start + xprt->reestablish_timeout;
883 if (time_after(start, now))
887 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
890 * xprt_reconnect_backoff - compute the new re-establish timeout
891 * @xprt: transport instance
892 * @init_to: initial reestablish timeout
895 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
897 xprt->reestablish_timeout <<= 1;
898 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
899 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
900 if (xprt->reestablish_timeout < init_to)
901 xprt->reestablish_timeout = init_to;
903 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
905 enum xprt_xid_rb_cmp {
910 static enum xprt_xid_rb_cmp
911 xprt_xid_cmp(__be32 xid1, __be32 xid2)
915 if ((__force u32)xid1 < (__force u32)xid2)
920 static struct rpc_rqst *
921 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
923 struct rb_node *n = xprt->recv_queue.rb_node;
924 struct rpc_rqst *req;
927 req = rb_entry(n, struct rpc_rqst, rq_recv);
928 switch (xprt_xid_cmp(xid, req->rq_xid)) {
943 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
945 struct rb_node **p = &xprt->recv_queue.rb_node;
946 struct rb_node *n = NULL;
947 struct rpc_rqst *req;
951 req = rb_entry(n, struct rpc_rqst, rq_recv);
952 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
960 WARN_ON_ONCE(new != req);
964 rb_link_node(&new->rq_recv, n, p);
965 rb_insert_color(&new->rq_recv, &xprt->recv_queue);
969 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
971 rb_erase(&req->rq_recv, &xprt->recv_queue);
975 * xprt_lookup_rqst - find an RPC request corresponding to an XID
976 * @xprt: transport on which the original request was transmitted
977 * @xid: RPC XID of incoming reply
979 * Caller holds xprt->queue_lock.
981 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
983 struct rpc_rqst *entry;
985 entry = xprt_request_rb_find(xprt, xid);
987 trace_xprt_lookup_rqst(xprt, xid, 0);
988 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
992 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
994 trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
995 xprt->stat.bad_xids++;
998 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1001 xprt_is_pinned_rqst(struct rpc_rqst *req)
1003 return atomic_read(&req->rq_pin) != 0;
1007 * xprt_pin_rqst - Pin a request on the transport receive list
1008 * @req: Request to pin
1010 * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1011 * so should be holding xprt->queue_lock.
1013 void xprt_pin_rqst(struct rpc_rqst *req)
1015 atomic_inc(&req->rq_pin);
1017 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1020 * xprt_unpin_rqst - Unpin a request on the transport receive list
1021 * @req: Request to pin
1023 * Caller should be holding xprt->queue_lock.
1025 void xprt_unpin_rqst(struct rpc_rqst *req)
1027 if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1028 atomic_dec(&req->rq_pin);
1031 if (atomic_dec_and_test(&req->rq_pin))
1032 wake_up_var(&req->rq_pin);
1034 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1036 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1038 wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1042 xprt_request_data_received(struct rpc_task *task)
1044 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1045 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1049 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1051 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1052 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1056 * xprt_request_enqueue_receive - Add an request to the receive queue
1061 xprt_request_enqueue_receive(struct rpc_task *task)
1063 struct rpc_rqst *req = task->tk_rqstp;
1064 struct rpc_xprt *xprt = req->rq_xprt;
1066 if (!xprt_request_need_enqueue_receive(task, req))
1069 xprt_request_prepare(task->tk_rqstp);
1070 spin_lock(&xprt->queue_lock);
1072 /* Update the softirq receive buffer */
1073 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1074 sizeof(req->rq_private_buf));
1076 /* Add request to the receive list */
1077 xprt_request_rb_insert(xprt, req);
1078 set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1079 spin_unlock(&xprt->queue_lock);
1081 /* Turn off autodisconnect */
1082 del_singleshot_timer_sync(&xprt->timer);
1086 * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1089 * Caller must hold xprt->queue_lock.
1092 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1094 struct rpc_rqst *req = task->tk_rqstp;
1096 if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1097 xprt_request_rb_remove(req->rq_xprt, req);
1101 * xprt_update_rtt - Update RPC RTT statistics
1102 * @task: RPC request that recently completed
1104 * Caller holds xprt->queue_lock.
1106 void xprt_update_rtt(struct rpc_task *task)
1108 struct rpc_rqst *req = task->tk_rqstp;
1109 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1110 unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1111 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1114 if (req->rq_ntrans == 1)
1115 rpc_update_rtt(rtt, timer, m);
1116 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1119 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1122 * xprt_complete_rqst - called when reply processing is complete
1123 * @task: RPC request that recently completed
1124 * @copied: actual number of bytes received from the transport
1126 * Caller holds xprt->queue_lock.
1128 void xprt_complete_rqst(struct rpc_task *task, int copied)
1130 struct rpc_rqst *req = task->tk_rqstp;
1131 struct rpc_xprt *xprt = req->rq_xprt;
1135 req->rq_private_buf.len = copied;
1136 /* Ensure all writes are done before we update */
1137 /* req->rq_reply_bytes_recvd */
1139 req->rq_reply_bytes_recvd = copied;
1140 xprt_request_dequeue_receive_locked(task);
1141 rpc_wake_up_queued_task(&xprt->pending, task);
1143 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1145 static void xprt_timer(struct rpc_task *task)
1147 struct rpc_rqst *req = task->tk_rqstp;
1148 struct rpc_xprt *xprt = req->rq_xprt;
1150 if (task->tk_status != -ETIMEDOUT)
1153 trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1154 if (!req->rq_reply_bytes_recvd) {
1155 if (xprt->ops->timer)
1156 xprt->ops->timer(xprt, task);
1158 task->tk_status = 0;
1162 * xprt_wait_for_reply_request_def - wait for reply
1163 * @task: pointer to rpc_task
1165 * Set a request's retransmit timeout based on the transport's
1166 * default timeout parameters. Used by transports that don't adjust
1167 * the retransmit timeout based on round-trip time estimation,
1168 * and put the task to sleep on the pending queue.
1170 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1172 struct rpc_rqst *req = task->tk_rqstp;
1174 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1175 xprt_request_timeout(req));
1177 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1180 * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1181 * @task: pointer to rpc_task
1183 * Set a request's retransmit timeout using the RTT estimator,
1184 * and put the task to sleep on the pending queue.
1186 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1188 int timer = task->tk_msg.rpc_proc->p_timer;
1189 struct rpc_clnt *clnt = task->tk_client;
1190 struct rpc_rtt *rtt = clnt->cl_rtt;
1191 struct rpc_rqst *req = task->tk_rqstp;
1192 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1193 unsigned long timeout;
1195 timeout = rpc_calc_rto(rtt, timer);
1196 timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1197 if (timeout > max_timeout || timeout == 0)
1198 timeout = max_timeout;
1199 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1202 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1205 * xprt_request_wait_receive - wait for the reply to an RPC request
1206 * @task: RPC task about to send a request
1209 void xprt_request_wait_receive(struct rpc_task *task)
1211 struct rpc_rqst *req = task->tk_rqstp;
1212 struct rpc_xprt *xprt = req->rq_xprt;
1214 if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1217 * Sleep on the pending queue if we're expecting a reply.
1218 * The spinlock ensures atomicity between the test of
1219 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1221 spin_lock(&xprt->queue_lock);
1222 if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1223 xprt->ops->wait_for_reply_request(task);
1225 * Send an extra queue wakeup call if the
1226 * connection was dropped in case the call to
1227 * rpc_sleep_on() raced.
1229 if (xprt_request_retransmit_after_disconnect(task))
1230 rpc_wake_up_queued_task_set_status(&xprt->pending,
1233 spin_unlock(&xprt->queue_lock);
1237 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1239 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1243 * xprt_request_enqueue_transmit - queue a task for transmission
1244 * @task: pointer to rpc_task
1246 * Add a task to the transmission queue.
1249 xprt_request_enqueue_transmit(struct rpc_task *task)
1251 struct rpc_rqst *pos, *req = task->tk_rqstp;
1252 struct rpc_xprt *xprt = req->rq_xprt;
1254 if (xprt_request_need_enqueue_transmit(task, req)) {
1255 req->rq_bytes_sent = 0;
1256 spin_lock(&xprt->queue_lock);
1258 * Requests that carry congestion control credits are added
1259 * to the head of the list to avoid starvation issues.
1262 xprt_clear_congestion_window_wait(xprt);
1263 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1266 /* Note: req is added _before_ pos */
1267 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1268 INIT_LIST_HEAD(&req->rq_xmit2);
1271 } else if (RPC_IS_SWAPPER(task)) {
1272 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1273 if (pos->rq_cong || pos->rq_bytes_sent)
1275 if (RPC_IS_SWAPPER(pos->rq_task))
1277 /* Note: req is added _before_ pos */
1278 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1279 INIT_LIST_HEAD(&req->rq_xmit2);
1282 } else if (!req->rq_seqno) {
1283 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1284 if (pos->rq_task->tk_owner != task->tk_owner)
1286 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1287 INIT_LIST_HEAD(&req->rq_xmit);
1291 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1292 INIT_LIST_HEAD(&req->rq_xmit2);
1294 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1295 spin_unlock(&xprt->queue_lock);
1300 * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1301 * @task: pointer to rpc_task
1303 * Remove a task from the transmission queue
1304 * Caller must hold xprt->queue_lock
1307 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1309 struct rpc_rqst *req = task->tk_rqstp;
1311 if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1313 if (!list_empty(&req->rq_xmit)) {
1314 list_del(&req->rq_xmit);
1315 if (!list_empty(&req->rq_xmit2)) {
1316 struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1317 struct rpc_rqst, rq_xmit2);
1318 list_del(&req->rq_xmit2);
1319 list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1322 list_del(&req->rq_xmit2);
1326 * xprt_request_dequeue_transmit - remove a task from the transmission queue
1327 * @task: pointer to rpc_task
1329 * Remove a task from the transmission queue
1332 xprt_request_dequeue_transmit(struct rpc_task *task)
1334 struct rpc_rqst *req = task->tk_rqstp;
1335 struct rpc_xprt *xprt = req->rq_xprt;
1337 spin_lock(&xprt->queue_lock);
1338 xprt_request_dequeue_transmit_locked(task);
1339 spin_unlock(&xprt->queue_lock);
1343 * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1344 * @task: pointer to rpc_task
1346 * Remove a task from the transmit and receive queues, and ensure that
1347 * it is not pinned by the receive work item.
1350 xprt_request_dequeue_xprt(struct rpc_task *task)
1352 struct rpc_rqst *req = task->tk_rqstp;
1353 struct rpc_xprt *xprt = req->rq_xprt;
1355 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1356 test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1357 xprt_is_pinned_rqst(req)) {
1358 spin_lock(&xprt->queue_lock);
1359 xprt_request_dequeue_transmit_locked(task);
1360 xprt_request_dequeue_receive_locked(task);
1361 while (xprt_is_pinned_rqst(req)) {
1362 set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1363 spin_unlock(&xprt->queue_lock);
1364 xprt_wait_on_pinned_rqst(req);
1365 spin_lock(&xprt->queue_lock);
1366 clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1368 spin_unlock(&xprt->queue_lock);
1373 * xprt_request_prepare - prepare an encoded request for transport
1374 * @req: pointer to rpc_rqst
1376 * Calls into the transport layer to do whatever is needed to prepare
1377 * the request for transmission or receive.
1380 xprt_request_prepare(struct rpc_rqst *req)
1382 struct rpc_xprt *xprt = req->rq_xprt;
1384 if (xprt->ops->prepare_request)
1385 xprt->ops->prepare_request(req);
1389 * xprt_request_need_retransmit - Test if a task needs retransmission
1390 * @task: pointer to rpc_task
1392 * Test for whether a connection breakage requires the task to retransmit
1395 xprt_request_need_retransmit(struct rpc_task *task)
1397 return xprt_request_retransmit_after_disconnect(task);
1401 * xprt_prepare_transmit - reserve the transport before sending a request
1402 * @task: RPC task about to send a request
1405 bool xprt_prepare_transmit(struct rpc_task *task)
1407 struct rpc_rqst *req = task->tk_rqstp;
1408 struct rpc_xprt *xprt = req->rq_xprt;
1410 if (!xprt_lock_write(xprt, task)) {
1411 trace_xprt_transmit_queued(xprt, task);
1413 /* Race breaker: someone may have transmitted us */
1414 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1415 rpc_wake_up_queued_task_set_status(&xprt->sending,
1423 void xprt_end_transmit(struct rpc_task *task)
1425 xprt_release_write(task->tk_rqstp->rq_xprt, task);
1429 * xprt_request_transmit - send an RPC request on a transport
1430 * @req: pointer to request to transmit
1431 * @snd_task: RPC task that owns the transport lock
1433 * This performs the transmission of a single request.
1434 * Note that if the request is not the same as snd_task, then it
1435 * does need to be pinned.
1436 * Returns '0' on success.
1439 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1441 struct rpc_xprt *xprt = req->rq_xprt;
1442 struct rpc_task *task = req->rq_task;
1443 unsigned int connect_cookie;
1444 int is_retrans = RPC_WAS_SENT(task);
1447 if (!req->rq_bytes_sent) {
1448 if (xprt_request_data_received(task)) {
1452 /* Verify that our message lies in the RPCSEC_GSS window */
1453 if (rpcauth_xmit_need_reencode(task)) {
1457 if (RPC_SIGNALLED(task)) {
1458 status = -ERESTARTSYS;
1464 * Update req->rq_ntrans before transmitting to avoid races with
1465 * xprt_update_rtt(), which needs to know that it is recording a
1466 * reply to the first transmission.
1470 trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1471 connect_cookie = xprt->connect_cookie;
1472 status = xprt->ops->send_request(req);
1475 trace_xprt_transmit(req, status);
1480 task->tk_client->cl_stats->rpcretrans++;
1482 xprt_inject_disconnect(xprt);
1484 task->tk_flags |= RPC_TASK_SENT;
1485 spin_lock(&xprt->transport_lock);
1488 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1489 xprt->stat.bklog_u += xprt->backlog.qlen;
1490 xprt->stat.sending_u += xprt->sending.qlen;
1491 xprt->stat.pending_u += xprt->pending.qlen;
1492 spin_unlock(&xprt->transport_lock);
1494 req->rq_connect_cookie = connect_cookie;
1496 trace_xprt_transmit(req, status);
1497 xprt_request_dequeue_transmit(task);
1498 rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1503 * xprt_transmit - send an RPC request on a transport
1504 * @task: controlling RPC task
1506 * Attempts to drain the transmit queue. On exit, either the transport
1507 * signalled an error that needs to be handled before transmission can
1508 * resume, or @task finished transmitting, and detected that it already
1512 xprt_transmit(struct rpc_task *task)
1514 struct rpc_rqst *next, *req = task->tk_rqstp;
1515 struct rpc_xprt *xprt = req->rq_xprt;
1516 int counter, status;
1518 spin_lock(&xprt->queue_lock);
1520 while (!list_empty(&xprt->xmit_queue)) {
1521 if (++counter == 20)
1523 next = list_first_entry(&xprt->xmit_queue,
1524 struct rpc_rqst, rq_xmit);
1525 xprt_pin_rqst(next);
1526 spin_unlock(&xprt->queue_lock);
1527 status = xprt_request_transmit(next, task);
1528 if (status == -EBADMSG && next != req)
1530 spin_lock(&xprt->queue_lock);
1531 xprt_unpin_rqst(next);
1533 if (!xprt_request_data_received(task) ||
1534 test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1536 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1537 task->tk_status = status;
1540 spin_unlock(&xprt->queue_lock);
1543 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1545 set_bit(XPRT_CONGESTED, &xprt->state);
1546 rpc_sleep_on(&xprt->backlog, task, NULL);
1549 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1551 if (rpc_wake_up_next(&xprt->backlog) == NULL)
1552 clear_bit(XPRT_CONGESTED, &xprt->state);
1555 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1559 if (!test_bit(XPRT_CONGESTED, &xprt->state))
1561 spin_lock(&xprt->reserve_lock);
1562 if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1563 rpc_sleep_on(&xprt->backlog, task, NULL);
1566 spin_unlock(&xprt->reserve_lock);
1571 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1573 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1575 if (xprt->num_reqs >= xprt->max_reqs)
1578 spin_unlock(&xprt->reserve_lock);
1579 req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1580 spin_lock(&xprt->reserve_lock);
1584 req = ERR_PTR(-ENOMEM);
1589 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1591 if (xprt->num_reqs > xprt->min_reqs) {
1599 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1601 struct rpc_rqst *req;
1603 spin_lock(&xprt->reserve_lock);
1604 if (!list_empty(&xprt->free)) {
1605 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1606 list_del(&req->rq_list);
1609 req = xprt_dynamic_alloc_slot(xprt);
1612 switch (PTR_ERR(req)) {
1614 dprintk("RPC: dynamic allocation of request slot "
1615 "failed! Retrying\n");
1616 task->tk_status = -ENOMEM;
1619 xprt_add_backlog(xprt, task);
1620 dprintk("RPC: waiting for request slot\n");
1623 task->tk_status = -EAGAIN;
1625 spin_unlock(&xprt->reserve_lock);
1628 xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1630 spin_unlock(&xprt->reserve_lock);
1632 task->tk_status = 0;
1633 task->tk_rqstp = req;
1635 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1637 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1639 spin_lock(&xprt->reserve_lock);
1640 if (!xprt_dynamic_free_slot(xprt, req)) {
1641 memset(req, 0, sizeof(*req)); /* mark unused */
1642 list_add(&req->rq_list, &xprt->free);
1644 xprt_wake_up_backlog(xprt);
1645 spin_unlock(&xprt->reserve_lock);
1647 EXPORT_SYMBOL_GPL(xprt_free_slot);
1649 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1651 struct rpc_rqst *req;
1652 while (!list_empty(&xprt->free)) {
1653 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1654 list_del(&req->rq_list);
1659 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1660 unsigned int num_prealloc,
1661 unsigned int max_alloc)
1663 struct rpc_xprt *xprt;
1664 struct rpc_rqst *req;
1667 xprt = kzalloc(size, GFP_KERNEL);
1671 xprt_init(xprt, net);
1673 for (i = 0; i < num_prealloc; i++) {
1674 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1677 list_add(&req->rq_list, &xprt->free);
1679 if (max_alloc > num_prealloc)
1680 xprt->max_reqs = max_alloc;
1682 xprt->max_reqs = num_prealloc;
1683 xprt->min_reqs = num_prealloc;
1684 xprt->num_reqs = num_prealloc;
1693 EXPORT_SYMBOL_GPL(xprt_alloc);
1695 void xprt_free(struct rpc_xprt *xprt)
1697 put_net(xprt->xprt_net);
1698 xprt_free_all_slots(xprt);
1699 kfree_rcu(xprt, rcu);
1701 EXPORT_SYMBOL_GPL(xprt_free);
1704 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1706 req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1710 xprt_alloc_xid(struct rpc_xprt *xprt)
1714 spin_lock(&xprt->reserve_lock);
1715 xid = (__force __be32)xprt->xid++;
1716 spin_unlock(&xprt->reserve_lock);
1721 xprt_init_xid(struct rpc_xprt *xprt)
1723 xprt->xid = prandom_u32();
1727 xprt_request_init(struct rpc_task *task)
1729 struct rpc_xprt *xprt = task->tk_xprt;
1730 struct rpc_rqst *req = task->tk_rqstp;
1732 req->rq_task = task;
1733 req->rq_xprt = xprt;
1734 req->rq_buffer = NULL;
1735 req->rq_xid = xprt_alloc_xid(xprt);
1736 xprt_init_connect_cookie(req, xprt);
1737 req->rq_snd_buf.len = 0;
1738 req->rq_snd_buf.buflen = 0;
1739 req->rq_rcv_buf.len = 0;
1740 req->rq_rcv_buf.buflen = 0;
1741 req->rq_snd_buf.bvec = NULL;
1742 req->rq_rcv_buf.bvec = NULL;
1743 req->rq_release_snd_buf = NULL;
1744 xprt_init_majortimeo(task, req);
1746 trace_xprt_reserve(req);
1750 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1752 xprt->ops->alloc_slot(xprt, task);
1753 if (task->tk_rqstp != NULL)
1754 xprt_request_init(task);
1758 * xprt_reserve - allocate an RPC request slot
1759 * @task: RPC task requesting a slot allocation
1761 * If the transport is marked as being congested, or if no more
1762 * slots are available, place the task on the transport's
1765 void xprt_reserve(struct rpc_task *task)
1767 struct rpc_xprt *xprt = task->tk_xprt;
1769 task->tk_status = 0;
1770 if (task->tk_rqstp != NULL)
1773 task->tk_status = -EAGAIN;
1774 if (!xprt_throttle_congested(xprt, task))
1775 xprt_do_reserve(xprt, task);
1779 * xprt_retry_reserve - allocate an RPC request slot
1780 * @task: RPC task requesting a slot allocation
1782 * If no more slots are available, place the task on the transport's
1784 * Note that the only difference with xprt_reserve is that we now
1785 * ignore the value of the XPRT_CONGESTED flag.
1787 void xprt_retry_reserve(struct rpc_task *task)
1789 struct rpc_xprt *xprt = task->tk_xprt;
1791 task->tk_status = 0;
1792 if (task->tk_rqstp != NULL)
1795 task->tk_status = -EAGAIN;
1796 xprt_do_reserve(xprt, task);
1800 * xprt_release - release an RPC request slot
1801 * @task: task which is finished with the slot
1804 void xprt_release(struct rpc_task *task)
1806 struct rpc_xprt *xprt;
1807 struct rpc_rqst *req = task->tk_rqstp;
1810 if (task->tk_client) {
1811 xprt = task->tk_xprt;
1812 xprt_release_write(xprt, task);
1817 xprt = req->rq_xprt;
1818 xprt_request_dequeue_xprt(task);
1819 spin_lock(&xprt->transport_lock);
1820 xprt->ops->release_xprt(xprt, task);
1821 if (xprt->ops->release_request)
1822 xprt->ops->release_request(task);
1823 xprt_schedule_autodisconnect(xprt);
1824 spin_unlock(&xprt->transport_lock);
1826 xprt->ops->buf_free(task);
1827 xprt_inject_disconnect(xprt);
1828 xdr_free_bvec(&req->rq_rcv_buf);
1829 xdr_free_bvec(&req->rq_snd_buf);
1830 if (req->rq_cred != NULL)
1831 put_rpccred(req->rq_cred);
1832 task->tk_rqstp = NULL;
1833 if (req->rq_release_snd_buf)
1834 req->rq_release_snd_buf(req);
1836 if (likely(!bc_prealloc(req)))
1837 xprt->ops->free_slot(xprt, req);
1839 xprt_free_bc_request(req);
1842 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1844 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1846 struct xdr_buf *xbufp = &req->rq_snd_buf;
1848 task->tk_rqstp = req;
1849 req->rq_task = task;
1850 xprt_init_connect_cookie(req, req->rq_xprt);
1852 * Set up the xdr_buf length.
1853 * This also indicates that the buffer is XDR encoded already.
1855 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1856 xbufp->tail[0].iov_len;
1860 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1862 kref_init(&xprt->kref);
1864 spin_lock_init(&xprt->transport_lock);
1865 spin_lock_init(&xprt->reserve_lock);
1866 spin_lock_init(&xprt->queue_lock);
1868 INIT_LIST_HEAD(&xprt->free);
1869 xprt->recv_queue = RB_ROOT;
1870 INIT_LIST_HEAD(&xprt->xmit_queue);
1871 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1872 spin_lock_init(&xprt->bc_pa_lock);
1873 INIT_LIST_HEAD(&xprt->bc_pa_list);
1874 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1875 INIT_LIST_HEAD(&xprt->xprt_switch);
1877 xprt->last_used = jiffies;
1878 xprt->cwnd = RPC_INITCWND;
1879 xprt->bind_index = 0;
1881 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1882 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1883 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1884 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1886 xprt_init_xid(xprt);
1888 xprt->xprt_net = get_net(net);
1892 * xprt_create_transport - create an RPC transport
1893 * @args: rpc transport creation arguments
1896 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1898 struct rpc_xprt *xprt;
1899 struct xprt_class *t;
1901 spin_lock(&xprt_list_lock);
1902 list_for_each_entry(t, &xprt_list, list) {
1903 if (t->ident == args->ident) {
1904 spin_unlock(&xprt_list_lock);
1908 spin_unlock(&xprt_list_lock);
1909 dprintk("RPC: transport (%d) not supported\n", args->ident);
1910 return ERR_PTR(-EIO);
1913 xprt = t->setup(args);
1916 if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1917 xprt->idle_timeout = 0;
1918 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1919 if (xprt_has_timer(xprt))
1920 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1922 timer_setup(&xprt->timer, NULL, 0);
1924 if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1926 return ERR_PTR(-EINVAL);
1928 xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1929 if (xprt->servername == NULL) {
1931 return ERR_PTR(-ENOMEM);
1934 rpc_xprt_debugfs_register(xprt);
1936 trace_xprt_create(xprt);
1941 static void xprt_destroy_cb(struct work_struct *work)
1943 struct rpc_xprt *xprt =
1944 container_of(work, struct rpc_xprt, task_cleanup);
1946 trace_xprt_destroy(xprt);
1948 rpc_xprt_debugfs_unregister(xprt);
1949 rpc_destroy_wait_queue(&xprt->binding);
1950 rpc_destroy_wait_queue(&xprt->pending);
1951 rpc_destroy_wait_queue(&xprt->sending);
1952 rpc_destroy_wait_queue(&xprt->backlog);
1953 kfree(xprt->servername);
1955 * Destroy any existing back channel
1957 xprt_destroy_backchannel(xprt, UINT_MAX);
1960 * Tear down transport state and free the rpc_xprt
1962 xprt->ops->destroy(xprt);
1966 * xprt_destroy - destroy an RPC transport, killing off all requests.
1967 * @xprt: transport to destroy
1970 static void xprt_destroy(struct rpc_xprt *xprt)
1973 * Exclude transport connect/disconnect handlers and autoclose
1975 wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1977 del_timer_sync(&xprt->timer);
1980 * Destroy sockets etc from the system workqueue so they can
1981 * safely flush receive work running on rpciod.
1983 INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1984 schedule_work(&xprt->task_cleanup);
1987 static void xprt_destroy_kref(struct kref *kref)
1989 xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1993 * xprt_get - return a reference to an RPC transport.
1994 * @xprt: pointer to the transport
1997 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1999 if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2003 EXPORT_SYMBOL_GPL(xprt_get);
2006 * xprt_put - release a reference to an RPC transport.
2007 * @xprt: pointer to the transport
2010 void xprt_put(struct rpc_xprt *xprt)
2013 kref_put(&xprt->kref, xprt_destroy_kref);
2015 EXPORT_SYMBOL_GPL(xprt_put);