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>
65 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
66 # define RPCDBG_FACILITY RPCDBG_XPRT
72 static void xprt_init(struct rpc_xprt *xprt, struct net *net);
73 static __be32 xprt_alloc_xid(struct rpc_xprt *xprt);
74 static void xprt_destroy(struct rpc_xprt *xprt);
75 static void xprt_request_init(struct rpc_task *task);
77 static DEFINE_SPINLOCK(xprt_list_lock);
78 static LIST_HEAD(xprt_list);
80 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
82 unsigned long timeout = jiffies + req->rq_timeout;
84 if (time_before(timeout, req->rq_majortimeo))
86 return req->rq_majortimeo;
90 * xprt_register_transport - register a transport implementation
91 * @transport: transport to register
93 * If a transport implementation is loaded as a kernel module, it can
94 * call this interface to make itself known to the RPC client.
97 * 0: transport successfully registered
98 * -EEXIST: transport already registered
99 * -EINVAL: transport module being unloaded
101 int xprt_register_transport(struct xprt_class *transport)
103 struct xprt_class *t;
107 spin_lock(&xprt_list_lock);
108 list_for_each_entry(t, &xprt_list, list) {
109 /* don't register the same transport class twice */
110 if (t->ident == transport->ident)
114 list_add_tail(&transport->list, &xprt_list);
115 printk(KERN_INFO "RPC: Registered %s transport module.\n",
120 spin_unlock(&xprt_list_lock);
123 EXPORT_SYMBOL_GPL(xprt_register_transport);
126 * xprt_unregister_transport - unregister a transport implementation
127 * @transport: transport to unregister
130 * 0: transport successfully unregistered
131 * -ENOENT: transport never registered
133 int xprt_unregister_transport(struct xprt_class *transport)
135 struct xprt_class *t;
139 spin_lock(&xprt_list_lock);
140 list_for_each_entry(t, &xprt_list, list) {
141 if (t == transport) {
143 "RPC: Unregistered %s transport module.\n",
145 list_del_init(&transport->list);
152 spin_unlock(&xprt_list_lock);
155 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
158 xprt_class_release(const struct xprt_class *t)
160 module_put(t->owner);
163 static const struct xprt_class *
164 xprt_class_find_by_ident_locked(int ident)
166 const struct xprt_class *t;
168 list_for_each_entry(t, &xprt_list, list) {
169 if (t->ident != ident)
171 if (!try_module_get(t->owner))
178 static const struct xprt_class *
179 xprt_class_find_by_ident(int ident)
181 const struct xprt_class *t;
183 spin_lock(&xprt_list_lock);
184 t = xprt_class_find_by_ident_locked(ident);
185 spin_unlock(&xprt_list_lock);
189 static const struct xprt_class *
190 xprt_class_find_by_netid_locked(const char *netid)
192 const struct xprt_class *t;
195 list_for_each_entry(t, &xprt_list, list) {
196 for (i = 0; t->netid[i][0] != '\0'; i++) {
197 if (strcmp(t->netid[i], netid) != 0)
199 if (!try_module_get(t->owner))
207 static const struct xprt_class *
208 xprt_class_find_by_netid(const char *netid)
210 const struct xprt_class *t;
212 spin_lock(&xprt_list_lock);
213 t = xprt_class_find_by_netid_locked(netid);
215 spin_unlock(&xprt_list_lock);
216 request_module("rpc%s", netid);
217 spin_lock(&xprt_list_lock);
218 t = xprt_class_find_by_netid_locked(netid);
220 spin_unlock(&xprt_list_lock);
225 * xprt_find_transport_ident - convert a netid into a transport identifier
226 * @netid: transport to load
229 * > 0: transport identifier
230 * -ENOENT: transport module not available
232 int xprt_find_transport_ident(const char *netid)
234 const struct xprt_class *t;
237 t = xprt_class_find_by_netid(netid);
241 xprt_class_release(t);
244 EXPORT_SYMBOL_GPL(xprt_find_transport_ident);
246 static void xprt_clear_locked(struct rpc_xprt *xprt)
248 xprt->snd_task = NULL;
249 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state))
250 clear_bit_unlock(XPRT_LOCKED, &xprt->state);
252 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
256 * xprt_reserve_xprt - serialize write access to transports
257 * @task: task that is requesting access to the transport
258 * @xprt: pointer to the target transport
260 * This prevents mixing the payload of separate requests, and prevents
261 * transport connects from colliding with writes. No congestion control
264 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
266 struct rpc_rqst *req = task->tk_rqstp;
268 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
269 if (task == xprt->snd_task)
273 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
275 xprt->snd_task = task;
278 trace_xprt_reserve_xprt(xprt, task);
282 xprt_clear_locked(xprt);
284 task->tk_status = -EAGAIN;
285 if (RPC_IS_SOFT(task))
286 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
287 xprt_request_timeout(req));
289 rpc_sleep_on(&xprt->sending, task, NULL);
292 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
295 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
297 return test_bit(XPRT_CWND_WAIT, &xprt->state);
301 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
303 if (!list_empty(&xprt->xmit_queue)) {
304 /* Peek at head of queue to see if it can make progress */
305 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
309 set_bit(XPRT_CWND_WAIT, &xprt->state);
313 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
315 if (!RPCXPRT_CONGESTED(xprt))
316 clear_bit(XPRT_CWND_WAIT, &xprt->state);
320 * xprt_reserve_xprt_cong - serialize write access to transports
321 * @task: task that is requesting access to the transport
323 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
324 * integrated into the decision of whether a request is allowed to be
325 * woken up and given access to the transport.
326 * Note that the lock is only granted if we know there are free slots.
328 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
330 struct rpc_rqst *req = task->tk_rqstp;
332 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
333 if (task == xprt->snd_task)
338 xprt->snd_task = task;
341 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
343 if (!xprt_need_congestion_window_wait(xprt)) {
344 xprt->snd_task = task;
348 xprt_clear_locked(xprt);
350 task->tk_status = -EAGAIN;
351 if (RPC_IS_SOFT(task))
352 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
353 xprt_request_timeout(req));
355 rpc_sleep_on(&xprt->sending, task, NULL);
358 trace_xprt_reserve_cong(xprt, task);
361 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
363 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
367 if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
369 spin_lock(&xprt->transport_lock);
370 retval = xprt->ops->reserve_xprt(xprt, task);
371 spin_unlock(&xprt->transport_lock);
375 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
377 struct rpc_xprt *xprt = data;
379 xprt->snd_task = task;
383 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
385 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
387 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
389 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
390 __xprt_lock_write_func, xprt))
393 xprt_clear_locked(xprt);
396 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
398 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
400 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
402 if (xprt_need_congestion_window_wait(xprt))
404 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
405 __xprt_lock_write_func, xprt))
408 xprt_clear_locked(xprt);
412 * xprt_release_xprt - allow other requests to use a transport
413 * @xprt: transport with other tasks potentially waiting
414 * @task: task that is releasing access to the transport
416 * Note that "task" can be NULL. No congestion control is provided.
418 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
420 if (xprt->snd_task == task) {
421 xprt_clear_locked(xprt);
422 __xprt_lock_write_next(xprt);
424 trace_xprt_release_xprt(xprt, task);
426 EXPORT_SYMBOL_GPL(xprt_release_xprt);
429 * xprt_release_xprt_cong - allow other requests to use a transport
430 * @xprt: transport with other tasks potentially waiting
431 * @task: task that is releasing access to the transport
433 * Note that "task" can be NULL. Another task is awoken to use the
434 * transport if the transport's congestion window allows it.
436 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
438 if (xprt->snd_task == task) {
439 xprt_clear_locked(xprt);
440 __xprt_lock_write_next_cong(xprt);
442 trace_xprt_release_cong(xprt, task);
444 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
446 void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
448 if (xprt->snd_task != task)
450 spin_lock(&xprt->transport_lock);
451 xprt->ops->release_xprt(xprt, task);
452 spin_unlock(&xprt->transport_lock);
456 * Van Jacobson congestion avoidance. Check if the congestion window
457 * overflowed. Put the task to sleep if this is the case.
460 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
464 trace_xprt_get_cong(xprt, req->rq_task);
465 if (RPCXPRT_CONGESTED(xprt)) {
466 xprt_set_congestion_window_wait(xprt);
470 xprt->cong += RPC_CWNDSCALE;
475 * Adjust the congestion window, and wake up the next task
476 * that has been sleeping due to congestion
479 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
484 xprt->cong -= RPC_CWNDSCALE;
485 xprt_test_and_clear_congestion_window_wait(xprt);
486 trace_xprt_put_cong(xprt, req->rq_task);
487 __xprt_lock_write_next_cong(xprt);
491 * xprt_request_get_cong - Request congestion control credits
492 * @xprt: pointer to transport
493 * @req: pointer to RPC request
495 * Useful for transports that require congestion control.
498 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
504 spin_lock(&xprt->transport_lock);
505 ret = __xprt_get_cong(xprt, req) != 0;
506 spin_unlock(&xprt->transport_lock);
509 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
512 * xprt_release_rqst_cong - housekeeping when request is complete
513 * @task: RPC request that recently completed
515 * Useful for transports that require congestion control.
517 void xprt_release_rqst_cong(struct rpc_task *task)
519 struct rpc_rqst *req = task->tk_rqstp;
521 __xprt_put_cong(req->rq_xprt, req);
523 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
525 static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
527 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
528 __xprt_lock_write_next_cong(xprt);
532 * Clear the congestion window wait flag and wake up the next
533 * entry on xprt->sending
536 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
538 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
539 spin_lock(&xprt->transport_lock);
540 __xprt_lock_write_next_cong(xprt);
541 spin_unlock(&xprt->transport_lock);
546 * xprt_adjust_cwnd - adjust transport congestion window
547 * @xprt: pointer to xprt
548 * @task: recently completed RPC request used to adjust window
549 * @result: result code of completed RPC request
551 * The transport code maintains an estimate on the maximum number of out-
552 * standing RPC requests, using a smoothed version of the congestion
553 * avoidance implemented in 44BSD. This is basically the Van Jacobson
554 * congestion algorithm: If a retransmit occurs, the congestion window is
555 * halved; otherwise, it is incremented by 1/cwnd when
557 * - a reply is received and
558 * - a full number of requests are outstanding and
559 * - the congestion window hasn't been updated recently.
561 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
563 struct rpc_rqst *req = task->tk_rqstp;
564 unsigned long cwnd = xprt->cwnd;
566 if (result >= 0 && cwnd <= xprt->cong) {
567 /* The (cwnd >> 1) term makes sure
568 * the result gets rounded properly. */
569 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
570 if (cwnd > RPC_MAXCWND(xprt))
571 cwnd = RPC_MAXCWND(xprt);
572 __xprt_lock_write_next_cong(xprt);
573 } else if (result == -ETIMEDOUT) {
575 if (cwnd < RPC_CWNDSCALE)
576 cwnd = RPC_CWNDSCALE;
578 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
579 xprt->cong, xprt->cwnd, cwnd);
581 __xprt_put_cong(xprt, req);
583 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
586 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
587 * @xprt: transport with waiting tasks
588 * @status: result code to plant in each task before waking it
591 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
594 rpc_wake_up_status(&xprt->pending, status);
596 rpc_wake_up(&xprt->pending);
598 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
601 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
604 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
605 * we don't in general want to force a socket disconnection due to
606 * an incomplete RPC call transmission.
608 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
610 set_bit(XPRT_WRITE_SPACE, &xprt->state);
612 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
615 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
617 if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
618 __xprt_lock_write_next(xprt);
619 dprintk("RPC: write space: waking waiting task on "
627 * xprt_write_space - wake the task waiting for transport output buffer space
628 * @xprt: transport with waiting tasks
630 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
632 bool xprt_write_space(struct rpc_xprt *xprt)
636 if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
638 spin_lock(&xprt->transport_lock);
639 ret = xprt_clear_write_space_locked(xprt);
640 spin_unlock(&xprt->transport_lock);
643 EXPORT_SYMBOL_GPL(xprt_write_space);
645 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
647 s64 delta = ktime_to_ns(ktime_get() - abstime);
648 return likely(delta >= 0) ?
649 jiffies - nsecs_to_jiffies(delta) :
650 jiffies + nsecs_to_jiffies(-delta);
653 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
655 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
656 unsigned long majortimeo = req->rq_timeout;
658 if (to->to_exponential)
659 majortimeo <<= to->to_retries;
661 majortimeo += to->to_increment * to->to_retries;
662 if (majortimeo > to->to_maxval || majortimeo == 0)
663 majortimeo = to->to_maxval;
667 static void xprt_reset_majortimeo(struct rpc_rqst *req)
669 req->rq_majortimeo += xprt_calc_majortimeo(req);
672 static void xprt_reset_minortimeo(struct rpc_rqst *req)
674 req->rq_minortimeo += req->rq_timeout;
677 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
679 unsigned long time_init;
680 struct rpc_xprt *xprt = req->rq_xprt;
682 if (likely(xprt && xprt_connected(xprt)))
685 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
686 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
687 req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
688 req->rq_minortimeo = time_init + req->rq_timeout;
692 * xprt_adjust_timeout - adjust timeout values for next retransmit
693 * @req: RPC request containing parameters to use for the adjustment
696 int xprt_adjust_timeout(struct rpc_rqst *req)
698 struct rpc_xprt *xprt = req->rq_xprt;
699 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
702 if (time_before(jiffies, req->rq_majortimeo)) {
703 if (time_before(jiffies, req->rq_minortimeo))
705 if (to->to_exponential)
706 req->rq_timeout <<= 1;
708 req->rq_timeout += to->to_increment;
709 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
710 req->rq_timeout = to->to_maxval;
713 req->rq_timeout = to->to_initval;
715 xprt_reset_majortimeo(req);
716 /* Reset the RTT counters == "slow start" */
717 spin_lock(&xprt->transport_lock);
718 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
719 spin_unlock(&xprt->transport_lock);
722 xprt_reset_minortimeo(req);
724 if (req->rq_timeout == 0) {
725 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
726 req->rq_timeout = 5 * HZ;
731 static void xprt_autoclose(struct work_struct *work)
733 struct rpc_xprt *xprt =
734 container_of(work, struct rpc_xprt, task_cleanup);
735 unsigned int pflags = memalloc_nofs_save();
737 trace_xprt_disconnect_auto(xprt);
738 xprt->connect_cookie++;
739 smp_mb__before_atomic();
740 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
741 xprt->ops->close(xprt);
742 xprt_release_write(xprt, NULL);
743 wake_up_bit(&xprt->state, XPRT_LOCKED);
744 memalloc_nofs_restore(pflags);
748 * xprt_disconnect_done - mark a transport as disconnected
749 * @xprt: transport to flag for disconnect
752 void xprt_disconnect_done(struct rpc_xprt *xprt)
754 trace_xprt_disconnect_done(xprt);
755 spin_lock(&xprt->transport_lock);
756 xprt_clear_connected(xprt);
757 xprt_clear_write_space_locked(xprt);
758 xprt_clear_congestion_window_wait_locked(xprt);
759 xprt_wake_pending_tasks(xprt, -ENOTCONN);
760 spin_unlock(&xprt->transport_lock);
762 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
765 * xprt_schedule_autoclose_locked - Try to schedule an autoclose RPC call
766 * @xprt: transport to disconnect
768 static void xprt_schedule_autoclose_locked(struct rpc_xprt *xprt)
770 if (test_and_set_bit(XPRT_CLOSE_WAIT, &xprt->state))
772 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
773 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
774 else if (xprt->snd_task && !test_bit(XPRT_SND_IS_COOKIE, &xprt->state))
775 rpc_wake_up_queued_task_set_status(&xprt->pending,
776 xprt->snd_task, -ENOTCONN);
780 * xprt_force_disconnect - force a transport to disconnect
781 * @xprt: transport to disconnect
784 void xprt_force_disconnect(struct rpc_xprt *xprt)
786 trace_xprt_disconnect_force(xprt);
788 /* Don't race with the test_bit() in xprt_clear_locked() */
789 spin_lock(&xprt->transport_lock);
790 xprt_schedule_autoclose_locked(xprt);
791 spin_unlock(&xprt->transport_lock);
793 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
796 xprt_connect_cookie(struct rpc_xprt *xprt)
798 return READ_ONCE(xprt->connect_cookie);
802 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
804 struct rpc_rqst *req = task->tk_rqstp;
805 struct rpc_xprt *xprt = req->rq_xprt;
807 return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
808 !xprt_connected(xprt);
812 * xprt_conditional_disconnect - force a transport to disconnect
813 * @xprt: transport to disconnect
814 * @cookie: 'connection cookie'
816 * This attempts to break the connection if and only if 'cookie' matches
817 * the current transport 'connection cookie'. It ensures that we don't
818 * try to break the connection more than once when we need to retransmit
819 * a batch of RPC requests.
822 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
824 /* Don't race with the test_bit() in xprt_clear_locked() */
825 spin_lock(&xprt->transport_lock);
826 if (cookie != xprt->connect_cookie)
828 if (test_bit(XPRT_CLOSING, &xprt->state))
830 xprt_schedule_autoclose_locked(xprt);
832 spin_unlock(&xprt->transport_lock);
836 xprt_has_timer(const struct rpc_xprt *xprt)
838 return xprt->idle_timeout != 0;
842 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
843 __must_hold(&xprt->transport_lock)
845 xprt->last_used = jiffies;
846 if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
847 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
851 xprt_init_autodisconnect(struct timer_list *t)
853 struct rpc_xprt *xprt = from_timer(xprt, t, timer);
855 if (!RB_EMPTY_ROOT(&xprt->recv_queue))
857 /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
858 xprt->last_used = jiffies;
859 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
861 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
864 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
865 static void xprt_inject_disconnect(struct rpc_xprt *xprt)
867 if (!fail_sunrpc.ignore_client_disconnect &&
868 should_fail(&fail_sunrpc.attr, 1))
869 xprt->ops->inject_disconnect(xprt);
872 static inline void xprt_inject_disconnect(struct rpc_xprt *xprt)
877 bool xprt_lock_connect(struct rpc_xprt *xprt,
878 struct rpc_task *task,
883 spin_lock(&xprt->transport_lock);
884 if (!test_bit(XPRT_LOCKED, &xprt->state))
886 if (xprt->snd_task != task)
888 set_bit(XPRT_SND_IS_COOKIE, &xprt->state);
889 xprt->snd_task = cookie;
892 spin_unlock(&xprt->transport_lock);
895 EXPORT_SYMBOL_GPL(xprt_lock_connect);
897 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
899 spin_lock(&xprt->transport_lock);
900 if (xprt->snd_task != cookie)
902 if (!test_bit(XPRT_LOCKED, &xprt->state))
904 xprt->snd_task =NULL;
905 clear_bit(XPRT_SND_IS_COOKIE, &xprt->state);
906 xprt->ops->release_xprt(xprt, NULL);
907 xprt_schedule_autodisconnect(xprt);
909 spin_unlock(&xprt->transport_lock);
910 wake_up_bit(&xprt->state, XPRT_LOCKED);
912 EXPORT_SYMBOL_GPL(xprt_unlock_connect);
915 * xprt_connect - schedule a transport connect operation
916 * @task: RPC task that is requesting the connect
919 void xprt_connect(struct rpc_task *task)
921 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
923 trace_xprt_connect(xprt);
925 if (!xprt_bound(xprt)) {
926 task->tk_status = -EAGAIN;
929 if (!xprt_lock_write(xprt, task))
932 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
933 trace_xprt_disconnect_cleanup(xprt);
934 xprt->ops->close(xprt);
937 if (!xprt_connected(xprt)) {
938 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
939 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
940 xprt_request_timeout(task->tk_rqstp));
942 if (test_bit(XPRT_CLOSING, &xprt->state))
944 if (xprt_test_and_set_connecting(xprt))
947 if (!xprt_connected(xprt)) {
948 xprt->stat.connect_start = jiffies;
949 xprt->ops->connect(xprt, task);
951 xprt_clear_connecting(xprt);
953 rpc_wake_up_queued_task(&xprt->pending, task);
956 xprt_release_write(xprt, task);
960 * xprt_reconnect_delay - compute the wait before scheduling a connect
961 * @xprt: transport instance
964 unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
966 unsigned long start, now = jiffies;
968 start = xprt->stat.connect_start + xprt->reestablish_timeout;
969 if (time_after(start, now))
973 EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
976 * xprt_reconnect_backoff - compute the new re-establish timeout
977 * @xprt: transport instance
978 * @init_to: initial reestablish timeout
981 void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
983 xprt->reestablish_timeout <<= 1;
984 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
985 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
986 if (xprt->reestablish_timeout < init_to)
987 xprt->reestablish_timeout = init_to;
989 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
991 enum xprt_xid_rb_cmp {
996 static enum xprt_xid_rb_cmp
997 xprt_xid_cmp(__be32 xid1, __be32 xid2)
1000 return XID_RB_EQUAL;
1001 if ((__force u32)xid1 < (__force u32)xid2)
1003 return XID_RB_RIGHT;
1006 static struct rpc_rqst *
1007 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
1009 struct rb_node *n = xprt->recv_queue.rb_node;
1010 struct rpc_rqst *req;
1013 req = rb_entry(n, struct rpc_rqst, rq_recv);
1014 switch (xprt_xid_cmp(xid, req->rq_xid)) {
1029 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
1031 struct rb_node **p = &xprt->recv_queue.rb_node;
1032 struct rb_node *n = NULL;
1033 struct rpc_rqst *req;
1035 while (*p != NULL) {
1037 req = rb_entry(n, struct rpc_rqst, rq_recv);
1038 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
1046 WARN_ON_ONCE(new != req);
1050 rb_link_node(&new->rq_recv, n, p);
1051 rb_insert_color(&new->rq_recv, &xprt->recv_queue);
1055 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
1057 rb_erase(&req->rq_recv, &xprt->recv_queue);
1061 * xprt_lookup_rqst - find an RPC request corresponding to an XID
1062 * @xprt: transport on which the original request was transmitted
1063 * @xid: RPC XID of incoming reply
1065 * Caller holds xprt->queue_lock.
1067 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
1069 struct rpc_rqst *entry;
1071 entry = xprt_request_rb_find(xprt, xid);
1072 if (entry != NULL) {
1073 trace_xprt_lookup_rqst(xprt, xid, 0);
1074 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
1078 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
1080 trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
1081 xprt->stat.bad_xids++;
1084 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1087 xprt_is_pinned_rqst(struct rpc_rqst *req)
1089 return atomic_read(&req->rq_pin) != 0;
1093 * xprt_pin_rqst - Pin a request on the transport receive list
1094 * @req: Request to pin
1096 * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1097 * so should be holding xprt->queue_lock.
1099 void xprt_pin_rqst(struct rpc_rqst *req)
1101 atomic_inc(&req->rq_pin);
1103 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1106 * xprt_unpin_rqst - Unpin a request on the transport receive list
1107 * @req: Request to pin
1109 * Caller should be holding xprt->queue_lock.
1111 void xprt_unpin_rqst(struct rpc_rqst *req)
1113 if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1114 atomic_dec(&req->rq_pin);
1117 if (atomic_dec_and_test(&req->rq_pin))
1118 wake_up_var(&req->rq_pin);
1120 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1122 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1124 wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1128 xprt_request_data_received(struct rpc_task *task)
1130 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1131 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1135 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1137 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1138 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1142 * xprt_request_enqueue_receive - Add an request to the receive queue
1147 xprt_request_enqueue_receive(struct rpc_task *task)
1149 struct rpc_rqst *req = task->tk_rqstp;
1150 struct rpc_xprt *xprt = req->rq_xprt;
1152 if (!xprt_request_need_enqueue_receive(task, req))
1155 xprt_request_prepare(task->tk_rqstp);
1156 spin_lock(&xprt->queue_lock);
1158 /* Update the softirq receive buffer */
1159 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1160 sizeof(req->rq_private_buf));
1162 /* Add request to the receive list */
1163 xprt_request_rb_insert(xprt, req);
1164 set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1165 spin_unlock(&xprt->queue_lock);
1167 /* Turn off autodisconnect */
1168 del_singleshot_timer_sync(&xprt->timer);
1172 * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1175 * Caller must hold xprt->queue_lock.
1178 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1180 struct rpc_rqst *req = task->tk_rqstp;
1182 if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1183 xprt_request_rb_remove(req->rq_xprt, req);
1187 * xprt_update_rtt - Update RPC RTT statistics
1188 * @task: RPC request that recently completed
1190 * Caller holds xprt->queue_lock.
1192 void xprt_update_rtt(struct rpc_task *task)
1194 struct rpc_rqst *req = task->tk_rqstp;
1195 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1196 unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1197 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1200 if (req->rq_ntrans == 1)
1201 rpc_update_rtt(rtt, timer, m);
1202 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1205 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1208 * xprt_complete_rqst - called when reply processing is complete
1209 * @task: RPC request that recently completed
1210 * @copied: actual number of bytes received from the transport
1212 * Caller holds xprt->queue_lock.
1214 void xprt_complete_rqst(struct rpc_task *task, int copied)
1216 struct rpc_rqst *req = task->tk_rqstp;
1217 struct rpc_xprt *xprt = req->rq_xprt;
1221 req->rq_private_buf.len = copied;
1222 /* Ensure all writes are done before we update */
1223 /* req->rq_reply_bytes_recvd */
1225 req->rq_reply_bytes_recvd = copied;
1226 xprt_request_dequeue_receive_locked(task);
1227 rpc_wake_up_queued_task(&xprt->pending, task);
1229 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1231 static void xprt_timer(struct rpc_task *task)
1233 struct rpc_rqst *req = task->tk_rqstp;
1234 struct rpc_xprt *xprt = req->rq_xprt;
1236 if (task->tk_status != -ETIMEDOUT)
1239 trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1240 if (!req->rq_reply_bytes_recvd) {
1241 if (xprt->ops->timer)
1242 xprt->ops->timer(xprt, task);
1244 task->tk_status = 0;
1248 * xprt_wait_for_reply_request_def - wait for reply
1249 * @task: pointer to rpc_task
1251 * Set a request's retransmit timeout based on the transport's
1252 * default timeout parameters. Used by transports that don't adjust
1253 * the retransmit timeout based on round-trip time estimation,
1254 * and put the task to sleep on the pending queue.
1256 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1258 struct rpc_rqst *req = task->tk_rqstp;
1260 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1261 xprt_request_timeout(req));
1263 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1266 * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1267 * @task: pointer to rpc_task
1269 * Set a request's retransmit timeout using the RTT estimator,
1270 * and put the task to sleep on the pending queue.
1272 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1274 int timer = task->tk_msg.rpc_proc->p_timer;
1275 struct rpc_clnt *clnt = task->tk_client;
1276 struct rpc_rtt *rtt = clnt->cl_rtt;
1277 struct rpc_rqst *req = task->tk_rqstp;
1278 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1279 unsigned long timeout;
1281 timeout = rpc_calc_rto(rtt, timer);
1282 timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1283 if (timeout > max_timeout || timeout == 0)
1284 timeout = max_timeout;
1285 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1288 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1291 * xprt_request_wait_receive - wait for the reply to an RPC request
1292 * @task: RPC task about to send a request
1295 void xprt_request_wait_receive(struct rpc_task *task)
1297 struct rpc_rqst *req = task->tk_rqstp;
1298 struct rpc_xprt *xprt = req->rq_xprt;
1300 if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1303 * Sleep on the pending queue if we're expecting a reply.
1304 * The spinlock ensures atomicity between the test of
1305 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1307 spin_lock(&xprt->queue_lock);
1308 if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1309 xprt->ops->wait_for_reply_request(task);
1311 * Send an extra queue wakeup call if the
1312 * connection was dropped in case the call to
1313 * rpc_sleep_on() raced.
1315 if (xprt_request_retransmit_after_disconnect(task))
1316 rpc_wake_up_queued_task_set_status(&xprt->pending,
1319 spin_unlock(&xprt->queue_lock);
1323 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1325 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1329 * xprt_request_enqueue_transmit - queue a task for transmission
1330 * @task: pointer to rpc_task
1332 * Add a task to the transmission queue.
1335 xprt_request_enqueue_transmit(struct rpc_task *task)
1337 struct rpc_rqst *pos, *req = task->tk_rqstp;
1338 struct rpc_xprt *xprt = req->rq_xprt;
1340 if (xprt_request_need_enqueue_transmit(task, req)) {
1341 req->rq_bytes_sent = 0;
1342 spin_lock(&xprt->queue_lock);
1344 * Requests that carry congestion control credits are added
1345 * to the head of the list to avoid starvation issues.
1348 xprt_clear_congestion_window_wait(xprt);
1349 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1352 /* Note: req is added _before_ pos */
1353 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1354 INIT_LIST_HEAD(&req->rq_xmit2);
1357 } else if (RPC_IS_SWAPPER(task)) {
1358 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1359 if (pos->rq_cong || pos->rq_bytes_sent)
1361 if (RPC_IS_SWAPPER(pos->rq_task))
1363 /* Note: req is added _before_ pos */
1364 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1365 INIT_LIST_HEAD(&req->rq_xmit2);
1368 } else if (!req->rq_seqno) {
1369 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1370 if (pos->rq_task->tk_owner != task->tk_owner)
1372 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1373 INIT_LIST_HEAD(&req->rq_xmit);
1377 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1378 INIT_LIST_HEAD(&req->rq_xmit2);
1380 atomic_long_inc(&xprt->xmit_queuelen);
1381 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1382 spin_unlock(&xprt->queue_lock);
1387 * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1388 * @task: pointer to rpc_task
1390 * Remove a task from the transmission queue
1391 * Caller must hold xprt->queue_lock
1394 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1396 struct rpc_rqst *req = task->tk_rqstp;
1398 if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1400 if (!list_empty(&req->rq_xmit)) {
1401 list_del(&req->rq_xmit);
1402 if (!list_empty(&req->rq_xmit2)) {
1403 struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1404 struct rpc_rqst, rq_xmit2);
1405 list_del(&req->rq_xmit2);
1406 list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1409 list_del(&req->rq_xmit2);
1410 atomic_long_dec(&req->rq_xprt->xmit_queuelen);
1414 * xprt_request_dequeue_transmit - remove a task from the transmission queue
1415 * @task: pointer to rpc_task
1417 * Remove a task from the transmission queue
1420 xprt_request_dequeue_transmit(struct rpc_task *task)
1422 struct rpc_rqst *req = task->tk_rqstp;
1423 struct rpc_xprt *xprt = req->rq_xprt;
1425 spin_lock(&xprt->queue_lock);
1426 xprt_request_dequeue_transmit_locked(task);
1427 spin_unlock(&xprt->queue_lock);
1431 * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1432 * @task: pointer to rpc_task
1434 * Remove a task from the transmit and receive queues, and ensure that
1435 * it is not pinned by the receive work item.
1438 xprt_request_dequeue_xprt(struct rpc_task *task)
1440 struct rpc_rqst *req = task->tk_rqstp;
1441 struct rpc_xprt *xprt = req->rq_xprt;
1443 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1444 test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1445 xprt_is_pinned_rqst(req)) {
1446 spin_lock(&xprt->queue_lock);
1447 xprt_request_dequeue_transmit_locked(task);
1448 xprt_request_dequeue_receive_locked(task);
1449 while (xprt_is_pinned_rqst(req)) {
1450 set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1451 spin_unlock(&xprt->queue_lock);
1452 xprt_wait_on_pinned_rqst(req);
1453 spin_lock(&xprt->queue_lock);
1454 clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1456 spin_unlock(&xprt->queue_lock);
1461 * xprt_request_prepare - prepare an encoded request for transport
1462 * @req: pointer to rpc_rqst
1464 * Calls into the transport layer to do whatever is needed to prepare
1465 * the request for transmission or receive.
1468 xprt_request_prepare(struct rpc_rqst *req)
1470 struct rpc_xprt *xprt = req->rq_xprt;
1472 if (xprt->ops->prepare_request)
1473 xprt->ops->prepare_request(req);
1477 * xprt_request_need_retransmit - Test if a task needs retransmission
1478 * @task: pointer to rpc_task
1480 * Test for whether a connection breakage requires the task to retransmit
1483 xprt_request_need_retransmit(struct rpc_task *task)
1485 return xprt_request_retransmit_after_disconnect(task);
1489 * xprt_prepare_transmit - reserve the transport before sending a request
1490 * @task: RPC task about to send a request
1493 bool xprt_prepare_transmit(struct rpc_task *task)
1495 struct rpc_rqst *req = task->tk_rqstp;
1496 struct rpc_xprt *xprt = req->rq_xprt;
1498 if (!xprt_lock_write(xprt, task)) {
1499 /* Race breaker: someone may have transmitted us */
1500 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1501 rpc_wake_up_queued_task_set_status(&xprt->sending,
1509 void xprt_end_transmit(struct rpc_task *task)
1511 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1513 xprt_inject_disconnect(xprt);
1514 xprt_release_write(xprt, task);
1518 * xprt_request_transmit - send an RPC request on a transport
1519 * @req: pointer to request to transmit
1520 * @snd_task: RPC task that owns the transport lock
1522 * This performs the transmission of a single request.
1523 * Note that if the request is not the same as snd_task, then it
1524 * does need to be pinned.
1525 * Returns '0' on success.
1528 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1530 struct rpc_xprt *xprt = req->rq_xprt;
1531 struct rpc_task *task = req->rq_task;
1532 unsigned int connect_cookie;
1533 int is_retrans = RPC_WAS_SENT(task);
1536 if (!req->rq_bytes_sent) {
1537 if (xprt_request_data_received(task)) {
1541 /* Verify that our message lies in the RPCSEC_GSS window */
1542 if (rpcauth_xmit_need_reencode(task)) {
1546 if (RPC_SIGNALLED(task)) {
1547 status = -ERESTARTSYS;
1553 * Update req->rq_ntrans before transmitting to avoid races with
1554 * xprt_update_rtt(), which needs to know that it is recording a
1555 * reply to the first transmission.
1559 trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1560 connect_cookie = xprt->connect_cookie;
1561 status = xprt->ops->send_request(req);
1564 trace_xprt_transmit(req, status);
1569 task->tk_client->cl_stats->rpcretrans++;
1570 trace_xprt_retransmit(req);
1573 xprt_inject_disconnect(xprt);
1575 task->tk_flags |= RPC_TASK_SENT;
1576 spin_lock(&xprt->transport_lock);
1579 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1580 xprt->stat.bklog_u += xprt->backlog.qlen;
1581 xprt->stat.sending_u += xprt->sending.qlen;
1582 xprt->stat.pending_u += xprt->pending.qlen;
1583 spin_unlock(&xprt->transport_lock);
1585 req->rq_connect_cookie = connect_cookie;
1587 trace_xprt_transmit(req, status);
1588 xprt_request_dequeue_transmit(task);
1589 rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1594 * xprt_transmit - send an RPC request on a transport
1595 * @task: controlling RPC task
1597 * Attempts to drain the transmit queue. On exit, either the transport
1598 * signalled an error that needs to be handled before transmission can
1599 * resume, or @task finished transmitting, and detected that it already
1603 xprt_transmit(struct rpc_task *task)
1605 struct rpc_rqst *next, *req = task->tk_rqstp;
1606 struct rpc_xprt *xprt = req->rq_xprt;
1609 spin_lock(&xprt->queue_lock);
1611 next = list_first_entry_or_null(&xprt->xmit_queue,
1612 struct rpc_rqst, rq_xmit);
1615 xprt_pin_rqst(next);
1616 spin_unlock(&xprt->queue_lock);
1617 status = xprt_request_transmit(next, task);
1618 if (status == -EBADMSG && next != req)
1620 spin_lock(&xprt->queue_lock);
1621 xprt_unpin_rqst(next);
1623 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1624 task->tk_status = status;
1627 /* Was @task transmitted, and has it received a reply? */
1628 if (xprt_request_data_received(task) &&
1629 !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1631 cond_resched_lock(&xprt->queue_lock);
1633 spin_unlock(&xprt->queue_lock);
1636 static void xprt_complete_request_init(struct rpc_task *task)
1639 xprt_request_init(task);
1642 void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1644 set_bit(XPRT_CONGESTED, &xprt->state);
1645 rpc_sleep_on(&xprt->backlog, task, xprt_complete_request_init);
1647 EXPORT_SYMBOL_GPL(xprt_add_backlog);
1649 static bool __xprt_set_rq(struct rpc_task *task, void *data)
1651 struct rpc_rqst *req = data;
1653 if (task->tk_rqstp == NULL) {
1654 memset(req, 0, sizeof(*req)); /* mark unused */
1655 task->tk_rqstp = req;
1661 bool xprt_wake_up_backlog(struct rpc_xprt *xprt, struct rpc_rqst *req)
1663 if (rpc_wake_up_first(&xprt->backlog, __xprt_set_rq, req) == NULL) {
1664 clear_bit(XPRT_CONGESTED, &xprt->state);
1669 EXPORT_SYMBOL_GPL(xprt_wake_up_backlog);
1671 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1675 if (!test_bit(XPRT_CONGESTED, &xprt->state))
1677 spin_lock(&xprt->reserve_lock);
1678 if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1679 xprt_add_backlog(xprt, task);
1682 spin_unlock(&xprt->reserve_lock);
1687 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1689 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1691 if (xprt->num_reqs >= xprt->max_reqs)
1694 spin_unlock(&xprt->reserve_lock);
1695 req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1696 spin_lock(&xprt->reserve_lock);
1700 req = ERR_PTR(-ENOMEM);
1705 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1707 if (xprt->num_reqs > xprt->min_reqs) {
1715 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1717 struct rpc_rqst *req;
1719 spin_lock(&xprt->reserve_lock);
1720 if (!list_empty(&xprt->free)) {
1721 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1722 list_del(&req->rq_list);
1725 req = xprt_dynamic_alloc_slot(xprt);
1728 switch (PTR_ERR(req)) {
1730 dprintk("RPC: dynamic allocation of request slot "
1731 "failed! Retrying\n");
1732 task->tk_status = -ENOMEM;
1735 xprt_add_backlog(xprt, task);
1736 dprintk("RPC: waiting for request slot\n");
1739 task->tk_status = -EAGAIN;
1741 spin_unlock(&xprt->reserve_lock);
1744 xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1746 spin_unlock(&xprt->reserve_lock);
1748 task->tk_status = 0;
1749 task->tk_rqstp = req;
1751 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1753 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1755 spin_lock(&xprt->reserve_lock);
1756 if (!xprt_wake_up_backlog(xprt, req) &&
1757 !xprt_dynamic_free_slot(xprt, req)) {
1758 memset(req, 0, sizeof(*req)); /* mark unused */
1759 list_add(&req->rq_list, &xprt->free);
1761 spin_unlock(&xprt->reserve_lock);
1763 EXPORT_SYMBOL_GPL(xprt_free_slot);
1765 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1767 struct rpc_rqst *req;
1768 while (!list_empty(&xprt->free)) {
1769 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1770 list_del(&req->rq_list);
1775 static DEFINE_IDA(rpc_xprt_ids);
1777 void xprt_cleanup_ids(void)
1779 ida_destroy(&rpc_xprt_ids);
1782 static int xprt_alloc_id(struct rpc_xprt *xprt)
1786 id = ida_simple_get(&rpc_xprt_ids, 0, 0, GFP_KERNEL);
1794 static void xprt_free_id(struct rpc_xprt *xprt)
1796 ida_simple_remove(&rpc_xprt_ids, xprt->id);
1799 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1800 unsigned int num_prealloc,
1801 unsigned int max_alloc)
1803 struct rpc_xprt *xprt;
1804 struct rpc_rqst *req;
1807 xprt = kzalloc(size, GFP_KERNEL);
1811 xprt_alloc_id(xprt);
1812 xprt_init(xprt, net);
1814 for (i = 0; i < num_prealloc; i++) {
1815 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1818 list_add(&req->rq_list, &xprt->free);
1820 if (max_alloc > num_prealloc)
1821 xprt->max_reqs = max_alloc;
1823 xprt->max_reqs = num_prealloc;
1824 xprt->min_reqs = num_prealloc;
1825 xprt->num_reqs = num_prealloc;
1834 EXPORT_SYMBOL_GPL(xprt_alloc);
1836 void xprt_free(struct rpc_xprt *xprt)
1838 put_net(xprt->xprt_net);
1839 xprt_free_all_slots(xprt);
1841 rpc_sysfs_xprt_destroy(xprt);
1842 kfree_rcu(xprt, rcu);
1844 EXPORT_SYMBOL_GPL(xprt_free);
1847 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1849 req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1853 xprt_alloc_xid(struct rpc_xprt *xprt)
1857 spin_lock(&xprt->reserve_lock);
1858 xid = (__force __be32)xprt->xid++;
1859 spin_unlock(&xprt->reserve_lock);
1864 xprt_init_xid(struct rpc_xprt *xprt)
1866 xprt->xid = prandom_u32();
1870 xprt_request_init(struct rpc_task *task)
1872 struct rpc_xprt *xprt = task->tk_xprt;
1873 struct rpc_rqst *req = task->tk_rqstp;
1875 req->rq_task = task;
1876 req->rq_xprt = xprt;
1877 req->rq_buffer = NULL;
1878 req->rq_xid = xprt_alloc_xid(xprt);
1879 xprt_init_connect_cookie(req, xprt);
1880 req->rq_snd_buf.len = 0;
1881 req->rq_snd_buf.buflen = 0;
1882 req->rq_rcv_buf.len = 0;
1883 req->rq_rcv_buf.buflen = 0;
1884 req->rq_snd_buf.bvec = NULL;
1885 req->rq_rcv_buf.bvec = NULL;
1886 req->rq_release_snd_buf = NULL;
1887 xprt_init_majortimeo(task, req);
1889 trace_xprt_reserve(req);
1893 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1895 xprt->ops->alloc_slot(xprt, task);
1896 if (task->tk_rqstp != NULL)
1897 xprt_request_init(task);
1901 * xprt_reserve - allocate an RPC request slot
1902 * @task: RPC task requesting a slot allocation
1904 * If the transport is marked as being congested, or if no more
1905 * slots are available, place the task on the transport's
1908 void xprt_reserve(struct rpc_task *task)
1910 struct rpc_xprt *xprt = task->tk_xprt;
1912 task->tk_status = 0;
1913 if (task->tk_rqstp != NULL)
1916 task->tk_status = -EAGAIN;
1917 if (!xprt_throttle_congested(xprt, task))
1918 xprt_do_reserve(xprt, task);
1922 * xprt_retry_reserve - allocate an RPC request slot
1923 * @task: RPC task requesting a slot allocation
1925 * If no more slots are available, place the task on the transport's
1927 * Note that the only difference with xprt_reserve is that we now
1928 * ignore the value of the XPRT_CONGESTED flag.
1930 void xprt_retry_reserve(struct rpc_task *task)
1932 struct rpc_xprt *xprt = task->tk_xprt;
1934 task->tk_status = 0;
1935 if (task->tk_rqstp != NULL)
1938 task->tk_status = -EAGAIN;
1939 xprt_do_reserve(xprt, task);
1943 * xprt_release - release an RPC request slot
1944 * @task: task which is finished with the slot
1947 void xprt_release(struct rpc_task *task)
1949 struct rpc_xprt *xprt;
1950 struct rpc_rqst *req = task->tk_rqstp;
1953 if (task->tk_client) {
1954 xprt = task->tk_xprt;
1955 xprt_release_write(xprt, task);
1960 xprt = req->rq_xprt;
1961 xprt_request_dequeue_xprt(task);
1962 spin_lock(&xprt->transport_lock);
1963 xprt->ops->release_xprt(xprt, task);
1964 if (xprt->ops->release_request)
1965 xprt->ops->release_request(task);
1966 xprt_schedule_autodisconnect(xprt);
1967 spin_unlock(&xprt->transport_lock);
1969 xprt->ops->buf_free(task);
1970 xdr_free_bvec(&req->rq_rcv_buf);
1971 xdr_free_bvec(&req->rq_snd_buf);
1972 if (req->rq_cred != NULL)
1973 put_rpccred(req->rq_cred);
1974 if (req->rq_release_snd_buf)
1975 req->rq_release_snd_buf(req);
1977 task->tk_rqstp = NULL;
1978 if (likely(!bc_prealloc(req)))
1979 xprt->ops->free_slot(xprt, req);
1981 xprt_free_bc_request(req);
1984 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1986 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1988 struct xdr_buf *xbufp = &req->rq_snd_buf;
1990 task->tk_rqstp = req;
1991 req->rq_task = task;
1992 xprt_init_connect_cookie(req, req->rq_xprt);
1994 * Set up the xdr_buf length.
1995 * This also indicates that the buffer is XDR encoded already.
1997 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1998 xbufp->tail[0].iov_len;
2002 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
2004 kref_init(&xprt->kref);
2006 spin_lock_init(&xprt->transport_lock);
2007 spin_lock_init(&xprt->reserve_lock);
2008 spin_lock_init(&xprt->queue_lock);
2010 INIT_LIST_HEAD(&xprt->free);
2011 xprt->recv_queue = RB_ROOT;
2012 INIT_LIST_HEAD(&xprt->xmit_queue);
2013 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
2014 spin_lock_init(&xprt->bc_pa_lock);
2015 INIT_LIST_HEAD(&xprt->bc_pa_list);
2016 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
2017 INIT_LIST_HEAD(&xprt->xprt_switch);
2019 xprt->last_used = jiffies;
2020 xprt->cwnd = RPC_INITCWND;
2021 xprt->bind_index = 0;
2023 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
2024 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
2025 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
2026 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
2028 xprt_init_xid(xprt);
2030 xprt->xprt_net = get_net(net);
2034 * xprt_create_transport - create an RPC transport
2035 * @args: rpc transport creation arguments
2038 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
2040 struct rpc_xprt *xprt;
2041 const struct xprt_class *t;
2043 t = xprt_class_find_by_ident(args->ident);
2045 dprintk("RPC: transport (%d) not supported\n", args->ident);
2046 return ERR_PTR(-EIO);
2049 xprt = t->setup(args);
2050 xprt_class_release(t);
2054 if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
2055 xprt->idle_timeout = 0;
2056 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
2057 if (xprt_has_timer(xprt))
2058 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
2060 timer_setup(&xprt->timer, NULL, 0);
2062 if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
2064 return ERR_PTR(-EINVAL);
2066 xprt->servername = kstrdup(args->servername, GFP_KERNEL);
2067 if (xprt->servername == NULL) {
2069 return ERR_PTR(-ENOMEM);
2072 rpc_xprt_debugfs_register(xprt);
2074 trace_xprt_create(xprt);
2079 static void xprt_destroy_cb(struct work_struct *work)
2081 struct rpc_xprt *xprt =
2082 container_of(work, struct rpc_xprt, task_cleanup);
2084 trace_xprt_destroy(xprt);
2086 rpc_xprt_debugfs_unregister(xprt);
2087 rpc_destroy_wait_queue(&xprt->binding);
2088 rpc_destroy_wait_queue(&xprt->pending);
2089 rpc_destroy_wait_queue(&xprt->sending);
2090 rpc_destroy_wait_queue(&xprt->backlog);
2091 kfree(xprt->servername);
2093 * Destroy any existing back channel
2095 xprt_destroy_backchannel(xprt, UINT_MAX);
2098 * Tear down transport state and free the rpc_xprt
2100 xprt->ops->destroy(xprt);
2104 * xprt_destroy - destroy an RPC transport, killing off all requests.
2105 * @xprt: transport to destroy
2108 static void xprt_destroy(struct rpc_xprt *xprt)
2111 * Exclude transport connect/disconnect handlers and autoclose
2113 wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
2115 del_timer_sync(&xprt->timer);
2118 * Destroy sockets etc from the system workqueue so they can
2119 * safely flush receive work running on rpciod.
2121 INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
2122 schedule_work(&xprt->task_cleanup);
2125 static void xprt_destroy_kref(struct kref *kref)
2127 xprt_destroy(container_of(kref, struct rpc_xprt, kref));
2131 * xprt_get - return a reference to an RPC transport.
2132 * @xprt: pointer to the transport
2135 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2137 if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2141 EXPORT_SYMBOL_GPL(xprt_get);
2144 * xprt_put - release a reference to an RPC transport.
2145 * @xprt: pointer to the transport
2148 void xprt_put(struct rpc_xprt *xprt)
2151 kref_put(&xprt->kref, xprt_destroy_kref);
2153 EXPORT_SYMBOL_GPL(xprt_put);