2 * linux/net/sunrpc/xprt.c
4 * This is a generic RPC call interface supporting congestion avoidance,
5 * and asynchronous calls.
7 * The interface works like this:
9 * - When a process places a call, it allocates a request slot if
10 * one is available. Otherwise, it sleeps on the backlog queue
12 * - Next, the caller puts together the RPC message, stuffs it into
13 * the request struct, and calls xprt_transmit().
14 * - xprt_transmit sends the message and installs the caller on the
15 * transport's wait list. At the same time, if a reply is expected,
16 * it installs a timer that is run after the packet's timeout has
18 * - When a packet arrives, the data_ready handler walks the list of
19 * pending requests for that transport. If a matching XID is found, the
20 * caller is woken up, and the timer removed.
21 * - When no reply arrives within the timeout interval, the timer is
22 * fired by the kernel and runs xprt_timer(). It either adjusts the
23 * timeout values (minor timeout) or wakes up the caller with a status
25 * - When the caller receives a notification from RPC that a reply arrived,
26 * it should release the RPC slot, and process the reply.
27 * If the call timed out, it may choose to retry the operation by
28 * adjusting the initial timeout value, and simply calling rpc_call
31 * Support for async RPC is done through a set of RPC-specific scheduling
32 * primitives that `transparently' work for processes as well as async
33 * tasks that rely on callbacks.
35 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
37 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
40 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/interrupt.h>
44 #include <linux/workqueue.h>
45 #include <linux/net.h>
46 #include <linux/ktime.h>
48 #include <linux/sunrpc/clnt.h>
49 #include <linux/sunrpc/metrics.h>
50 #include <linux/sunrpc/bc_xprt.h>
52 #include <trace/events/sunrpc.h>
60 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
61 # define RPCDBG_FACILITY RPCDBG_XPRT
67 static void xprt_init(struct rpc_xprt
*xprt
, struct net
*net
);
68 static void xprt_request_init(struct rpc_task
*, struct rpc_xprt
*);
69 static void xprt_connect_status(struct rpc_task
*task
);
70 static int __xprt_get_cong(struct rpc_xprt
*, struct rpc_task
*);
71 static void xprt_destroy(struct rpc_xprt
*xprt
);
73 static DEFINE_SPINLOCK(xprt_list_lock
);
74 static LIST_HEAD(xprt_list
);
77 * xprt_register_transport - register a transport implementation
78 * @transport: transport to register
80 * If a transport implementation is loaded as a kernel module, it can
81 * call this interface to make itself known to the RPC client.
84 * 0: transport successfully registered
85 * -EEXIST: transport already registered
86 * -EINVAL: transport module being unloaded
88 int xprt_register_transport(struct xprt_class
*transport
)
94 spin_lock(&xprt_list_lock
);
95 list_for_each_entry(t
, &xprt_list
, list
) {
96 /* don't register the same transport class twice */
97 if (t
->ident
== transport
->ident
)
101 list_add_tail(&transport
->list
, &xprt_list
);
102 printk(KERN_INFO
"RPC: Registered %s transport module.\n",
107 spin_unlock(&xprt_list_lock
);
110 EXPORT_SYMBOL_GPL(xprt_register_transport
);
113 * xprt_unregister_transport - unregister a transport implementation
114 * @transport: transport to unregister
117 * 0: transport successfully unregistered
118 * -ENOENT: transport never registered
120 int xprt_unregister_transport(struct xprt_class
*transport
)
122 struct xprt_class
*t
;
126 spin_lock(&xprt_list_lock
);
127 list_for_each_entry(t
, &xprt_list
, list
) {
128 if (t
== transport
) {
130 "RPC: Unregistered %s transport module.\n",
132 list_del_init(&transport
->list
);
139 spin_unlock(&xprt_list_lock
);
142 EXPORT_SYMBOL_GPL(xprt_unregister_transport
);
145 * xprt_load_transport - load a transport implementation
146 * @transport_name: transport to load
149 * 0: transport successfully loaded
150 * -ENOENT: transport module not available
152 int xprt_load_transport(const char *transport_name
)
154 struct xprt_class
*t
;
158 spin_lock(&xprt_list_lock
);
159 list_for_each_entry(t
, &xprt_list
, list
) {
160 if (strcmp(t
->name
, transport_name
) == 0) {
161 spin_unlock(&xprt_list_lock
);
165 spin_unlock(&xprt_list_lock
);
166 result
= request_module("xprt%s", transport_name
);
170 EXPORT_SYMBOL_GPL(xprt_load_transport
);
173 * xprt_reserve_xprt - serialize write access to transports
174 * @task: task that is requesting access to the transport
175 * @xprt: pointer to the target transport
177 * This prevents mixing the payload of separate requests, and prevents
178 * transport connects from colliding with writes. No congestion control
181 int xprt_reserve_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
183 struct rpc_rqst
*req
= task
->tk_rqstp
;
186 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
)) {
187 if (task
== xprt
->snd_task
)
191 xprt
->snd_task
= task
;
198 dprintk("RPC: %5u failed to lock transport %p\n",
200 task
->tk_timeout
= 0;
201 task
->tk_status
= -EAGAIN
;
203 priority
= RPC_PRIORITY_LOW
;
204 else if (!req
->rq_ntrans
)
205 priority
= RPC_PRIORITY_NORMAL
;
207 priority
= RPC_PRIORITY_HIGH
;
208 rpc_sleep_on_priority(&xprt
->sending
, task
, NULL
, priority
);
211 EXPORT_SYMBOL_GPL(xprt_reserve_xprt
);
213 static void xprt_clear_locked(struct rpc_xprt
*xprt
)
215 xprt
->snd_task
= NULL
;
216 if (!test_bit(XPRT_CLOSE_WAIT
, &xprt
->state
)) {
217 smp_mb__before_atomic();
218 clear_bit(XPRT_LOCKED
, &xprt
->state
);
219 smp_mb__after_atomic();
221 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
225 * xprt_reserve_xprt_cong - serialize write access to transports
226 * @task: task that is requesting access to the transport
228 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
229 * integrated into the decision of whether a request is allowed to be
230 * woken up and given access to the transport.
232 int xprt_reserve_xprt_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
234 struct rpc_rqst
*req
= task
->tk_rqstp
;
237 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
)) {
238 if (task
== xprt
->snd_task
)
243 xprt
->snd_task
= task
;
246 if (__xprt_get_cong(xprt
, task
)) {
247 xprt
->snd_task
= task
;
251 xprt_clear_locked(xprt
);
253 dprintk("RPC: %5u failed to lock transport %p\n", task
->tk_pid
, xprt
);
254 task
->tk_timeout
= 0;
255 task
->tk_status
= -EAGAIN
;
257 priority
= RPC_PRIORITY_LOW
;
258 else if (!req
->rq_ntrans
)
259 priority
= RPC_PRIORITY_NORMAL
;
261 priority
= RPC_PRIORITY_HIGH
;
262 rpc_sleep_on_priority(&xprt
->sending
, task
, NULL
, priority
);
265 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong
);
267 static inline int xprt_lock_write(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
271 spin_lock_bh(&xprt
->transport_lock
);
272 retval
= xprt
->ops
->reserve_xprt(xprt
, task
);
273 spin_unlock_bh(&xprt
->transport_lock
);
277 static bool __xprt_lock_write_func(struct rpc_task
*task
, void *data
)
279 struct rpc_xprt
*xprt
= data
;
280 struct rpc_rqst
*req
;
282 req
= task
->tk_rqstp
;
283 xprt
->snd_task
= task
;
289 static void __xprt_lock_write_next(struct rpc_xprt
*xprt
)
291 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
294 if (rpc_wake_up_first(&xprt
->sending
, __xprt_lock_write_func
, xprt
))
296 xprt_clear_locked(xprt
);
299 static bool __xprt_lock_write_cong_func(struct rpc_task
*task
, void *data
)
301 struct rpc_xprt
*xprt
= data
;
302 struct rpc_rqst
*req
;
304 req
= task
->tk_rqstp
;
306 xprt
->snd_task
= task
;
309 if (__xprt_get_cong(xprt
, task
)) {
310 xprt
->snd_task
= task
;
317 static void __xprt_lock_write_next_cong(struct rpc_xprt
*xprt
)
319 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
321 if (RPCXPRT_CONGESTED(xprt
))
323 if (rpc_wake_up_first(&xprt
->sending
, __xprt_lock_write_cong_func
, xprt
))
326 xprt_clear_locked(xprt
);
329 static void xprt_task_clear_bytes_sent(struct rpc_task
*task
)
332 struct rpc_rqst
*req
= task
->tk_rqstp
;
334 req
->rq_bytes_sent
= 0;
339 * xprt_release_xprt - allow other requests to use a transport
340 * @xprt: transport with other tasks potentially waiting
341 * @task: task that is releasing access to the transport
343 * Note that "task" can be NULL. No congestion control is provided.
345 void xprt_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
347 if (xprt
->snd_task
== task
) {
348 xprt_task_clear_bytes_sent(task
);
349 xprt_clear_locked(xprt
);
350 __xprt_lock_write_next(xprt
);
353 EXPORT_SYMBOL_GPL(xprt_release_xprt
);
356 * xprt_release_xprt_cong - allow other requests to use a transport
357 * @xprt: transport with other tasks potentially waiting
358 * @task: task that is releasing access to the transport
360 * Note that "task" can be NULL. Another task is awoken to use the
361 * transport if the transport's congestion window allows it.
363 void xprt_release_xprt_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
365 if (xprt
->snd_task
== task
) {
366 xprt_task_clear_bytes_sent(task
);
367 xprt_clear_locked(xprt
);
368 __xprt_lock_write_next_cong(xprt
);
371 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong
);
373 static inline void xprt_release_write(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
375 spin_lock_bh(&xprt
->transport_lock
);
376 xprt
->ops
->release_xprt(xprt
, task
);
377 spin_unlock_bh(&xprt
->transport_lock
);
381 * Van Jacobson congestion avoidance. Check if the congestion window
382 * overflowed. Put the task to sleep if this is the case.
385 __xprt_get_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
387 struct rpc_rqst
*req
= task
->tk_rqstp
;
391 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
392 task
->tk_pid
, xprt
->cong
, xprt
->cwnd
);
393 if (RPCXPRT_CONGESTED(xprt
))
396 xprt
->cong
+= RPC_CWNDSCALE
;
401 * Adjust the congestion window, and wake up the next task
402 * that has been sleeping due to congestion
405 __xprt_put_cong(struct rpc_xprt
*xprt
, struct rpc_rqst
*req
)
410 xprt
->cong
-= RPC_CWNDSCALE
;
411 __xprt_lock_write_next_cong(xprt
);
415 * xprt_release_rqst_cong - housekeeping when request is complete
416 * @task: RPC request that recently completed
418 * Useful for transports that require congestion control.
420 void xprt_release_rqst_cong(struct rpc_task
*task
)
422 struct rpc_rqst
*req
= task
->tk_rqstp
;
424 __xprt_put_cong(req
->rq_xprt
, req
);
426 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong
);
429 * xprt_adjust_cwnd - adjust transport congestion window
430 * @xprt: pointer to xprt
431 * @task: recently completed RPC request used to adjust window
432 * @result: result code of completed RPC request
434 * The transport code maintains an estimate on the maximum number of out-
435 * standing RPC requests, using a smoothed version of the congestion
436 * avoidance implemented in 44BSD. This is basically the Van Jacobson
437 * congestion algorithm: If a retransmit occurs, the congestion window is
438 * halved; otherwise, it is incremented by 1/cwnd when
440 * - a reply is received and
441 * - a full number of requests are outstanding and
442 * - the congestion window hasn't been updated recently.
444 void xprt_adjust_cwnd(struct rpc_xprt
*xprt
, struct rpc_task
*task
, int result
)
446 struct rpc_rqst
*req
= task
->tk_rqstp
;
447 unsigned long cwnd
= xprt
->cwnd
;
449 if (result
>= 0 && cwnd
<= xprt
->cong
) {
450 /* The (cwnd >> 1) term makes sure
451 * the result gets rounded properly. */
452 cwnd
+= (RPC_CWNDSCALE
* RPC_CWNDSCALE
+ (cwnd
>> 1)) / cwnd
;
453 if (cwnd
> RPC_MAXCWND(xprt
))
454 cwnd
= RPC_MAXCWND(xprt
);
455 __xprt_lock_write_next_cong(xprt
);
456 } else if (result
== -ETIMEDOUT
) {
458 if (cwnd
< RPC_CWNDSCALE
)
459 cwnd
= RPC_CWNDSCALE
;
461 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
462 xprt
->cong
, xprt
->cwnd
, cwnd
);
464 __xprt_put_cong(xprt
, req
);
466 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd
);
469 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
470 * @xprt: transport with waiting tasks
471 * @status: result code to plant in each task before waking it
474 void xprt_wake_pending_tasks(struct rpc_xprt
*xprt
, int status
)
477 rpc_wake_up_status(&xprt
->pending
, status
);
479 rpc_wake_up(&xprt
->pending
);
481 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks
);
484 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
485 * @task: task to be put to sleep
486 * @action: function pointer to be executed after wait
488 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
489 * we don't in general want to force a socket disconnection due to
490 * an incomplete RPC call transmission.
492 void xprt_wait_for_buffer_space(struct rpc_task
*task
, rpc_action action
)
494 struct rpc_rqst
*req
= task
->tk_rqstp
;
495 struct rpc_xprt
*xprt
= req
->rq_xprt
;
497 task
->tk_timeout
= RPC_IS_SOFT(task
) ? req
->rq_timeout
: 0;
498 rpc_sleep_on(&xprt
->pending
, task
, action
);
500 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space
);
503 * xprt_write_space - wake the task waiting for transport output buffer space
504 * @xprt: transport with waiting tasks
506 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
508 void xprt_write_space(struct rpc_xprt
*xprt
)
510 spin_lock_bh(&xprt
->transport_lock
);
511 if (xprt
->snd_task
) {
512 dprintk("RPC: write space: waking waiting task on "
514 rpc_wake_up_queued_task(&xprt
->pending
, xprt
->snd_task
);
516 spin_unlock_bh(&xprt
->transport_lock
);
518 EXPORT_SYMBOL_GPL(xprt_write_space
);
521 * xprt_set_retrans_timeout_def - set a request's retransmit timeout
522 * @task: task whose timeout is to be set
524 * Set a request's retransmit timeout based on the transport's
525 * default timeout parameters. Used by transports that don't adjust
526 * the retransmit timeout based on round-trip time estimation.
528 void xprt_set_retrans_timeout_def(struct rpc_task
*task
)
530 task
->tk_timeout
= task
->tk_rqstp
->rq_timeout
;
532 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def
);
535 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
536 * @task: task whose timeout is to be set
538 * Set a request's retransmit timeout using the RTT estimator.
540 void xprt_set_retrans_timeout_rtt(struct rpc_task
*task
)
542 int timer
= task
->tk_msg
.rpc_proc
->p_timer
;
543 struct rpc_clnt
*clnt
= task
->tk_client
;
544 struct rpc_rtt
*rtt
= clnt
->cl_rtt
;
545 struct rpc_rqst
*req
= task
->tk_rqstp
;
546 unsigned long max_timeout
= clnt
->cl_timeout
->to_maxval
;
548 task
->tk_timeout
= rpc_calc_rto(rtt
, timer
);
549 task
->tk_timeout
<<= rpc_ntimeo(rtt
, timer
) + req
->rq_retries
;
550 if (task
->tk_timeout
> max_timeout
|| task
->tk_timeout
== 0)
551 task
->tk_timeout
= max_timeout
;
553 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt
);
555 static void xprt_reset_majortimeo(struct rpc_rqst
*req
)
557 const struct rpc_timeout
*to
= req
->rq_task
->tk_client
->cl_timeout
;
559 req
->rq_majortimeo
= req
->rq_timeout
;
560 if (to
->to_exponential
)
561 req
->rq_majortimeo
<<= to
->to_retries
;
563 req
->rq_majortimeo
+= to
->to_increment
* to
->to_retries
;
564 if (req
->rq_majortimeo
> to
->to_maxval
|| req
->rq_majortimeo
== 0)
565 req
->rq_majortimeo
= to
->to_maxval
;
566 req
->rq_majortimeo
+= jiffies
;
570 * xprt_adjust_timeout - adjust timeout values for next retransmit
571 * @req: RPC request containing parameters to use for the adjustment
574 int xprt_adjust_timeout(struct rpc_rqst
*req
)
576 struct rpc_xprt
*xprt
= req
->rq_xprt
;
577 const struct rpc_timeout
*to
= req
->rq_task
->tk_client
->cl_timeout
;
580 if (time_before(jiffies
, req
->rq_majortimeo
)) {
581 if (to
->to_exponential
)
582 req
->rq_timeout
<<= 1;
584 req
->rq_timeout
+= to
->to_increment
;
585 if (to
->to_maxval
&& req
->rq_timeout
>= to
->to_maxval
)
586 req
->rq_timeout
= to
->to_maxval
;
589 req
->rq_timeout
= to
->to_initval
;
591 xprt_reset_majortimeo(req
);
592 /* Reset the RTT counters == "slow start" */
593 spin_lock_bh(&xprt
->transport_lock
);
594 rpc_init_rtt(req
->rq_task
->tk_client
->cl_rtt
, to
->to_initval
);
595 spin_unlock_bh(&xprt
->transport_lock
);
599 if (req
->rq_timeout
== 0) {
600 printk(KERN_WARNING
"xprt_adjust_timeout: rq_timeout = 0!\n");
601 req
->rq_timeout
= 5 * HZ
;
606 static void xprt_autoclose(struct work_struct
*work
)
608 struct rpc_xprt
*xprt
=
609 container_of(work
, struct rpc_xprt
, task_cleanup
);
611 xprt
->ops
->close(xprt
);
612 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
613 xprt_release_write(xprt
, NULL
);
617 * xprt_disconnect_done - mark a transport as disconnected
618 * @xprt: transport to flag for disconnect
621 void xprt_disconnect_done(struct rpc_xprt
*xprt
)
623 dprintk("RPC: disconnected transport %p\n", xprt
);
624 spin_lock_bh(&xprt
->transport_lock
);
625 xprt_clear_connected(xprt
);
626 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
627 spin_unlock_bh(&xprt
->transport_lock
);
629 EXPORT_SYMBOL_GPL(xprt_disconnect_done
);
632 * xprt_force_disconnect - force a transport to disconnect
633 * @xprt: transport to disconnect
636 void xprt_force_disconnect(struct rpc_xprt
*xprt
)
638 /* Don't race with the test_bit() in xprt_clear_locked() */
639 spin_lock_bh(&xprt
->transport_lock
);
640 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
641 /* Try to schedule an autoclose RPC call */
642 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
643 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
644 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
645 spin_unlock_bh(&xprt
->transport_lock
);
649 * xprt_conditional_disconnect - force a transport to disconnect
650 * @xprt: transport to disconnect
651 * @cookie: 'connection cookie'
653 * This attempts to break the connection if and only if 'cookie' matches
654 * the current transport 'connection cookie'. It ensures that we don't
655 * try to break the connection more than once when we need to retransmit
656 * a batch of RPC requests.
659 void xprt_conditional_disconnect(struct rpc_xprt
*xprt
, unsigned int cookie
)
661 /* Don't race with the test_bit() in xprt_clear_locked() */
662 spin_lock_bh(&xprt
->transport_lock
);
663 if (cookie
!= xprt
->connect_cookie
)
665 if (test_bit(XPRT_CLOSING
, &xprt
->state
) || !xprt_connected(xprt
))
667 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
668 /* Try to schedule an autoclose RPC call */
669 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
670 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
671 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
673 spin_unlock_bh(&xprt
->transport_lock
);
677 xprt_init_autodisconnect(unsigned long data
)
679 struct rpc_xprt
*xprt
= (struct rpc_xprt
*)data
;
681 spin_lock(&xprt
->transport_lock
);
682 if (!list_empty(&xprt
->recv
))
684 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
686 spin_unlock(&xprt
->transport_lock
);
687 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
690 spin_unlock(&xprt
->transport_lock
);
693 bool xprt_lock_connect(struct rpc_xprt
*xprt
,
694 struct rpc_task
*task
,
699 spin_lock_bh(&xprt
->transport_lock
);
700 if (!test_bit(XPRT_LOCKED
, &xprt
->state
))
702 if (xprt
->snd_task
!= task
)
704 xprt_task_clear_bytes_sent(task
);
705 xprt
->snd_task
= cookie
;
708 spin_unlock_bh(&xprt
->transport_lock
);
712 void xprt_unlock_connect(struct rpc_xprt
*xprt
, void *cookie
)
714 spin_lock_bh(&xprt
->transport_lock
);
715 if (xprt
->snd_task
!= cookie
)
717 if (!test_bit(XPRT_LOCKED
, &xprt
->state
))
719 xprt
->snd_task
=NULL
;
720 xprt
->ops
->release_xprt(xprt
, NULL
);
722 spin_unlock_bh(&xprt
->transport_lock
);
726 * xprt_connect - schedule a transport connect operation
727 * @task: RPC task that is requesting the connect
730 void xprt_connect(struct rpc_task
*task
)
732 struct rpc_xprt
*xprt
= task
->tk_rqstp
->rq_xprt
;
734 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task
->tk_pid
,
735 xprt
, (xprt_connected(xprt
) ? "is" : "is not"));
737 if (!xprt_bound(xprt
)) {
738 task
->tk_status
= -EAGAIN
;
741 if (!xprt_lock_write(xprt
, task
))
744 if (test_and_clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
))
745 xprt
->ops
->close(xprt
);
747 if (!xprt_connected(xprt
)) {
748 task
->tk_rqstp
->rq_bytes_sent
= 0;
749 task
->tk_timeout
= task
->tk_rqstp
->rq_timeout
;
750 rpc_sleep_on(&xprt
->pending
, task
, xprt_connect_status
);
752 if (test_bit(XPRT_CLOSING
, &xprt
->state
))
754 if (xprt_test_and_set_connecting(xprt
))
756 xprt
->stat
.connect_start
= jiffies
;
757 xprt
->ops
->connect(xprt
, task
);
759 xprt_release_write(xprt
, task
);
762 static void xprt_connect_status(struct rpc_task
*task
)
764 struct rpc_xprt
*xprt
= task
->tk_rqstp
->rq_xprt
;
766 if (task
->tk_status
== 0) {
767 xprt
->stat
.connect_count
++;
768 xprt
->stat
.connect_time
+= (long)jiffies
- xprt
->stat
.connect_start
;
769 dprintk("RPC: %5u xprt_connect_status: connection established\n",
774 switch (task
->tk_status
) {
782 dprintk("RPC: %5u xprt_connect_status: retrying\n", task
->tk_pid
);
785 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
786 "out\n", task
->tk_pid
);
789 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
790 "server %s\n", task
->tk_pid
, -task
->tk_status
,
792 task
->tk_status
= -EIO
;
797 * xprt_lookup_rqst - find an RPC request corresponding to an XID
798 * @xprt: transport on which the original request was transmitted
799 * @xid: RPC XID of incoming reply
802 struct rpc_rqst
*xprt_lookup_rqst(struct rpc_xprt
*xprt
, __be32 xid
)
804 struct rpc_rqst
*entry
;
806 list_for_each_entry(entry
, &xprt
->recv
, rq_list
)
807 if (entry
->rq_xid
== xid
) {
808 trace_xprt_lookup_rqst(xprt
, xid
, 0);
812 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
814 trace_xprt_lookup_rqst(xprt
, xid
, -ENOENT
);
815 xprt
->stat
.bad_xids
++;
818 EXPORT_SYMBOL_GPL(xprt_lookup_rqst
);
820 static void xprt_update_rtt(struct rpc_task
*task
)
822 struct rpc_rqst
*req
= task
->tk_rqstp
;
823 struct rpc_rtt
*rtt
= task
->tk_client
->cl_rtt
;
824 unsigned int timer
= task
->tk_msg
.rpc_proc
->p_timer
;
825 long m
= usecs_to_jiffies(ktime_to_us(req
->rq_rtt
));
828 if (req
->rq_ntrans
== 1)
829 rpc_update_rtt(rtt
, timer
, m
);
830 rpc_set_timeo(rtt
, timer
, req
->rq_ntrans
- 1);
835 * xprt_complete_rqst - called when reply processing is complete
836 * @task: RPC request that recently completed
837 * @copied: actual number of bytes received from the transport
839 * Caller holds transport lock.
841 void xprt_complete_rqst(struct rpc_task
*task
, int copied
)
843 struct rpc_rqst
*req
= task
->tk_rqstp
;
844 struct rpc_xprt
*xprt
= req
->rq_xprt
;
846 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
847 task
->tk_pid
, ntohl(req
->rq_xid
), copied
);
848 trace_xprt_complete_rqst(xprt
, req
->rq_xid
, copied
);
851 req
->rq_rtt
= ktime_sub(ktime_get(), req
->rq_xtime
);
852 if (xprt
->ops
->timer
!= NULL
)
853 xprt_update_rtt(task
);
855 list_del_init(&req
->rq_list
);
856 req
->rq_private_buf
.len
= copied
;
857 /* Ensure all writes are done before we update */
858 /* req->rq_reply_bytes_recvd */
860 req
->rq_reply_bytes_recvd
= copied
;
861 rpc_wake_up_queued_task(&xprt
->pending
, task
);
863 EXPORT_SYMBOL_GPL(xprt_complete_rqst
);
865 static void xprt_timer(struct rpc_task
*task
)
867 struct rpc_rqst
*req
= task
->tk_rqstp
;
868 struct rpc_xprt
*xprt
= req
->rq_xprt
;
870 if (task
->tk_status
!= -ETIMEDOUT
)
872 dprintk("RPC: %5u xprt_timer\n", task
->tk_pid
);
874 spin_lock_bh(&xprt
->transport_lock
);
875 if (!req
->rq_reply_bytes_recvd
) {
876 if (xprt
->ops
->timer
)
877 xprt
->ops
->timer(xprt
, task
);
880 spin_unlock_bh(&xprt
->transport_lock
);
883 static inline int xprt_has_timer(struct rpc_xprt
*xprt
)
885 return xprt
->idle_timeout
!= 0;
889 * xprt_prepare_transmit - reserve the transport before sending a request
890 * @task: RPC task about to send a request
893 bool xprt_prepare_transmit(struct rpc_task
*task
)
895 struct rpc_rqst
*req
= task
->tk_rqstp
;
896 struct rpc_xprt
*xprt
= req
->rq_xprt
;
899 dprintk("RPC: %5u xprt_prepare_transmit\n", task
->tk_pid
);
901 spin_lock_bh(&xprt
->transport_lock
);
902 if (!req
->rq_bytes_sent
) {
903 if (req
->rq_reply_bytes_recvd
) {
904 task
->tk_status
= req
->rq_reply_bytes_recvd
;
907 if ((task
->tk_flags
& RPC_TASK_NO_RETRANS_TIMEOUT
)
908 && xprt_connected(xprt
)
909 && req
->rq_connect_cookie
== xprt
->connect_cookie
) {
910 xprt
->ops
->set_retrans_timeout(task
);
911 rpc_sleep_on(&xprt
->pending
, task
, xprt_timer
);
915 if (!xprt
->ops
->reserve_xprt(xprt
, task
)) {
916 task
->tk_status
= -EAGAIN
;
921 spin_unlock_bh(&xprt
->transport_lock
);
925 void xprt_end_transmit(struct rpc_task
*task
)
927 xprt_release_write(task
->tk_rqstp
->rq_xprt
, task
);
931 * xprt_transmit - send an RPC request on a transport
932 * @task: controlling RPC task
934 * We have to copy the iovec because sendmsg fiddles with its contents.
936 void xprt_transmit(struct rpc_task
*task
)
938 struct rpc_rqst
*req
= task
->tk_rqstp
;
939 struct rpc_xprt
*xprt
= req
->rq_xprt
;
942 dprintk("RPC: %5u xprt_transmit(%u)\n", task
->tk_pid
, req
->rq_slen
);
944 if (!req
->rq_reply_bytes_recvd
) {
945 if (list_empty(&req
->rq_list
) && rpc_reply_expected(task
)) {
947 * Add to the list only if we're expecting a reply
949 spin_lock_bh(&xprt
->transport_lock
);
950 /* Update the softirq receive buffer */
951 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
,
952 sizeof(req
->rq_private_buf
));
953 /* Add request to the receive list */
954 list_add_tail(&req
->rq_list
, &xprt
->recv
);
955 spin_unlock_bh(&xprt
->transport_lock
);
956 xprt_reset_majortimeo(req
);
957 /* Turn off autodisconnect */
958 del_singleshot_timer_sync(&xprt
->timer
);
960 } else if (!req
->rq_bytes_sent
)
963 req
->rq_xtime
= ktime_get();
964 status
= xprt
->ops
->send_request(task
);
965 trace_xprt_transmit(xprt
, req
->rq_xid
, status
);
967 task
->tk_status
= status
;
971 dprintk("RPC: %5u xmit complete\n", task
->tk_pid
);
972 task
->tk_flags
|= RPC_TASK_SENT
;
973 spin_lock_bh(&xprt
->transport_lock
);
975 xprt
->ops
->set_retrans_timeout(task
);
977 numreqs
= atomic_read(&xprt
->num_reqs
);
978 if (numreqs
> xprt
->stat
.max_slots
)
979 xprt
->stat
.max_slots
= numreqs
;
981 xprt
->stat
.req_u
+= xprt
->stat
.sends
- xprt
->stat
.recvs
;
982 xprt
->stat
.bklog_u
+= xprt
->backlog
.qlen
;
983 xprt
->stat
.sending_u
+= xprt
->sending
.qlen
;
984 xprt
->stat
.pending_u
+= xprt
->pending
.qlen
;
986 /* Don't race with disconnect */
987 if (!xprt_connected(xprt
))
988 task
->tk_status
= -ENOTCONN
;
991 * Sleep on the pending queue since
992 * we're expecting a reply.
994 if (!req
->rq_reply_bytes_recvd
&& rpc_reply_expected(task
))
995 rpc_sleep_on(&xprt
->pending
, task
, xprt_timer
);
996 req
->rq_connect_cookie
= xprt
->connect_cookie
;
998 spin_unlock_bh(&xprt
->transport_lock
);
1001 static void xprt_add_backlog(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1003 set_bit(XPRT_CONGESTED
, &xprt
->state
);
1004 rpc_sleep_on(&xprt
->backlog
, task
, NULL
);
1007 static void xprt_wake_up_backlog(struct rpc_xprt
*xprt
)
1009 if (rpc_wake_up_next(&xprt
->backlog
) == NULL
)
1010 clear_bit(XPRT_CONGESTED
, &xprt
->state
);
1013 static bool xprt_throttle_congested(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1017 if (!test_bit(XPRT_CONGESTED
, &xprt
->state
))
1019 spin_lock(&xprt
->reserve_lock
);
1020 if (test_bit(XPRT_CONGESTED
, &xprt
->state
)) {
1021 rpc_sleep_on(&xprt
->backlog
, task
, NULL
);
1024 spin_unlock(&xprt
->reserve_lock
);
1029 static struct rpc_rqst
*xprt_dynamic_alloc_slot(struct rpc_xprt
*xprt
, gfp_t gfp_flags
)
1031 struct rpc_rqst
*req
= ERR_PTR(-EAGAIN
);
1033 if (!atomic_add_unless(&xprt
->num_reqs
, 1, xprt
->max_reqs
))
1035 req
= kzalloc(sizeof(struct rpc_rqst
), gfp_flags
);
1038 atomic_dec(&xprt
->num_reqs
);
1039 req
= ERR_PTR(-ENOMEM
);
1044 static bool xprt_dynamic_free_slot(struct rpc_xprt
*xprt
, struct rpc_rqst
*req
)
1046 if (atomic_add_unless(&xprt
->num_reqs
, -1, xprt
->min_reqs
)) {
1053 void xprt_alloc_slot(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1055 struct rpc_rqst
*req
;
1057 spin_lock(&xprt
->reserve_lock
);
1058 if (!list_empty(&xprt
->free
)) {
1059 req
= list_entry(xprt
->free
.next
, struct rpc_rqst
, rq_list
);
1060 list_del(&req
->rq_list
);
1063 req
= xprt_dynamic_alloc_slot(xprt
, GFP_NOWAIT
|__GFP_NOWARN
);
1066 switch (PTR_ERR(req
)) {
1068 dprintk("RPC: dynamic allocation of request slot "
1069 "failed! Retrying\n");
1070 task
->tk_status
= -ENOMEM
;
1073 xprt_add_backlog(xprt
, task
);
1074 dprintk("RPC: waiting for request slot\n");
1076 task
->tk_status
= -EAGAIN
;
1078 spin_unlock(&xprt
->reserve_lock
);
1081 task
->tk_status
= 0;
1082 task
->tk_rqstp
= req
;
1083 xprt_request_init(task
, xprt
);
1084 spin_unlock(&xprt
->reserve_lock
);
1086 EXPORT_SYMBOL_GPL(xprt_alloc_slot
);
1088 void xprt_lock_and_alloc_slot(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1090 /* Note: grabbing the xprt_lock_write() ensures that we throttle
1091 * new slot allocation if the transport is congested (i.e. when
1092 * reconnecting a stream transport or when out of socket write
1095 if (xprt_lock_write(xprt
, task
)) {
1096 xprt_alloc_slot(xprt
, task
);
1097 xprt_release_write(xprt
, task
);
1100 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot
);
1102 static void xprt_free_slot(struct rpc_xprt
*xprt
, struct rpc_rqst
*req
)
1104 spin_lock(&xprt
->reserve_lock
);
1105 if (!xprt_dynamic_free_slot(xprt
, req
)) {
1106 memset(req
, 0, sizeof(*req
)); /* mark unused */
1107 list_add(&req
->rq_list
, &xprt
->free
);
1109 xprt_wake_up_backlog(xprt
);
1110 spin_unlock(&xprt
->reserve_lock
);
1113 static void xprt_free_all_slots(struct rpc_xprt
*xprt
)
1115 struct rpc_rqst
*req
;
1116 while (!list_empty(&xprt
->free
)) {
1117 req
= list_first_entry(&xprt
->free
, struct rpc_rqst
, rq_list
);
1118 list_del(&req
->rq_list
);
1123 struct rpc_xprt
*xprt_alloc(struct net
*net
, size_t size
,
1124 unsigned int num_prealloc
,
1125 unsigned int max_alloc
)
1127 struct rpc_xprt
*xprt
;
1128 struct rpc_rqst
*req
;
1131 xprt
= kzalloc(size
, GFP_KERNEL
);
1135 xprt_init(xprt
, net
);
1137 for (i
= 0; i
< num_prealloc
; i
++) {
1138 req
= kzalloc(sizeof(struct rpc_rqst
), GFP_KERNEL
);
1141 list_add(&req
->rq_list
, &xprt
->free
);
1143 if (max_alloc
> num_prealloc
)
1144 xprt
->max_reqs
= max_alloc
;
1146 xprt
->max_reqs
= num_prealloc
;
1147 xprt
->min_reqs
= num_prealloc
;
1148 atomic_set(&xprt
->num_reqs
, num_prealloc
);
1157 EXPORT_SYMBOL_GPL(xprt_alloc
);
1159 void xprt_free(struct rpc_xprt
*xprt
)
1161 put_net(xprt
->xprt_net
);
1162 xprt_free_all_slots(xprt
);
1165 EXPORT_SYMBOL_GPL(xprt_free
);
1168 * xprt_reserve - allocate an RPC request slot
1169 * @task: RPC task requesting a slot allocation
1171 * If the transport is marked as being congested, or if no more
1172 * slots are available, place the task on the transport's
1175 void xprt_reserve(struct rpc_task
*task
)
1177 struct rpc_xprt
*xprt
;
1179 task
->tk_status
= 0;
1180 if (task
->tk_rqstp
!= NULL
)
1183 task
->tk_timeout
= 0;
1184 task
->tk_status
= -EAGAIN
;
1186 xprt
= rcu_dereference(task
->tk_client
->cl_xprt
);
1187 if (!xprt_throttle_congested(xprt
, task
))
1188 xprt
->ops
->alloc_slot(xprt
, task
);
1193 * xprt_retry_reserve - allocate an RPC request slot
1194 * @task: RPC task requesting a slot allocation
1196 * If no more slots are available, place the task on the transport's
1198 * Note that the only difference with xprt_reserve is that we now
1199 * ignore the value of the XPRT_CONGESTED flag.
1201 void xprt_retry_reserve(struct rpc_task
*task
)
1203 struct rpc_xprt
*xprt
;
1205 task
->tk_status
= 0;
1206 if (task
->tk_rqstp
!= NULL
)
1209 task
->tk_timeout
= 0;
1210 task
->tk_status
= -EAGAIN
;
1212 xprt
= rcu_dereference(task
->tk_client
->cl_xprt
);
1213 xprt
->ops
->alloc_slot(xprt
, task
);
1217 static inline __be32
xprt_alloc_xid(struct rpc_xprt
*xprt
)
1219 return (__force __be32
)xprt
->xid
++;
1222 static inline void xprt_init_xid(struct rpc_xprt
*xprt
)
1224 xprt
->xid
= prandom_u32();
1227 static void xprt_request_init(struct rpc_task
*task
, struct rpc_xprt
*xprt
)
1229 struct rpc_rqst
*req
= task
->tk_rqstp
;
1231 INIT_LIST_HEAD(&req
->rq_list
);
1232 req
->rq_timeout
= task
->tk_client
->cl_timeout
->to_initval
;
1233 req
->rq_task
= task
;
1234 req
->rq_xprt
= xprt
;
1235 req
->rq_buffer
= NULL
;
1236 req
->rq_xid
= xprt_alloc_xid(xprt
);
1237 req
->rq_connect_cookie
= xprt
->connect_cookie
- 1;
1238 req
->rq_bytes_sent
= 0;
1239 req
->rq_snd_buf
.len
= 0;
1240 req
->rq_snd_buf
.buflen
= 0;
1241 req
->rq_rcv_buf
.len
= 0;
1242 req
->rq_rcv_buf
.buflen
= 0;
1243 req
->rq_release_snd_buf
= NULL
;
1244 xprt_reset_majortimeo(req
);
1245 dprintk("RPC: %5u reserved req %p xid %08x\n", task
->tk_pid
,
1246 req
, ntohl(req
->rq_xid
));
1250 * xprt_release - release an RPC request slot
1251 * @task: task which is finished with the slot
1254 void xprt_release(struct rpc_task
*task
)
1256 struct rpc_xprt
*xprt
;
1257 struct rpc_rqst
*req
= task
->tk_rqstp
;
1260 if (task
->tk_client
) {
1262 xprt
= rcu_dereference(task
->tk_client
->cl_xprt
);
1263 if (xprt
->snd_task
== task
)
1264 xprt_release_write(xprt
, task
);
1270 xprt
= req
->rq_xprt
;
1271 if (task
->tk_ops
->rpc_count_stats
!= NULL
)
1272 task
->tk_ops
->rpc_count_stats(task
, task
->tk_calldata
);
1273 else if (task
->tk_client
)
1274 rpc_count_iostats(task
, task
->tk_client
->cl_metrics
);
1275 spin_lock_bh(&xprt
->transport_lock
);
1276 xprt
->ops
->release_xprt(xprt
, task
);
1277 if (xprt
->ops
->release_request
)
1278 xprt
->ops
->release_request(task
);
1279 if (!list_empty(&req
->rq_list
))
1280 list_del(&req
->rq_list
);
1281 xprt
->last_used
= jiffies
;
1282 if (list_empty(&xprt
->recv
) && xprt_has_timer(xprt
))
1283 mod_timer(&xprt
->timer
,
1284 xprt
->last_used
+ xprt
->idle_timeout
);
1285 spin_unlock_bh(&xprt
->transport_lock
);
1287 xprt
->ops
->buf_free(req
->rq_buffer
);
1288 if (req
->rq_cred
!= NULL
)
1289 put_rpccred(req
->rq_cred
);
1290 task
->tk_rqstp
= NULL
;
1291 if (req
->rq_release_snd_buf
)
1292 req
->rq_release_snd_buf(req
);
1294 dprintk("RPC: %5u release request %p\n", task
->tk_pid
, req
);
1295 if (likely(!bc_prealloc(req
)))
1296 xprt_free_slot(xprt
, req
);
1298 xprt_free_bc_request(req
);
1301 static void xprt_init(struct rpc_xprt
*xprt
, struct net
*net
)
1303 atomic_set(&xprt
->count
, 1);
1305 spin_lock_init(&xprt
->transport_lock
);
1306 spin_lock_init(&xprt
->reserve_lock
);
1308 INIT_LIST_HEAD(&xprt
->free
);
1309 INIT_LIST_HEAD(&xprt
->recv
);
1310 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1311 spin_lock_init(&xprt
->bc_pa_lock
);
1312 INIT_LIST_HEAD(&xprt
->bc_pa_list
);
1313 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1315 xprt
->last_used
= jiffies
;
1316 xprt
->cwnd
= RPC_INITCWND
;
1317 xprt
->bind_index
= 0;
1319 rpc_init_wait_queue(&xprt
->binding
, "xprt_binding");
1320 rpc_init_wait_queue(&xprt
->pending
, "xprt_pending");
1321 rpc_init_priority_wait_queue(&xprt
->sending
, "xprt_sending");
1322 rpc_init_priority_wait_queue(&xprt
->backlog
, "xprt_backlog");
1324 xprt_init_xid(xprt
);
1326 xprt
->xprt_net
= get_net(net
);
1330 * xprt_create_transport - create an RPC transport
1331 * @args: rpc transport creation arguments
1334 struct rpc_xprt
*xprt_create_transport(struct xprt_create
*args
)
1336 struct rpc_xprt
*xprt
;
1337 struct xprt_class
*t
;
1339 spin_lock(&xprt_list_lock
);
1340 list_for_each_entry(t
, &xprt_list
, list
) {
1341 if (t
->ident
== args
->ident
) {
1342 spin_unlock(&xprt_list_lock
);
1346 spin_unlock(&xprt_list_lock
);
1347 dprintk("RPC: transport (%d) not supported\n", args
->ident
);
1348 return ERR_PTR(-EIO
);
1351 xprt
= t
->setup(args
);
1353 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1357 if (args
->flags
& XPRT_CREATE_NO_IDLE_TIMEOUT
)
1358 xprt
->idle_timeout
= 0;
1359 INIT_WORK(&xprt
->task_cleanup
, xprt_autoclose
);
1360 if (xprt_has_timer(xprt
))
1361 setup_timer(&xprt
->timer
, xprt_init_autodisconnect
,
1362 (unsigned long)xprt
);
1364 init_timer(&xprt
->timer
);
1366 if (strlen(args
->servername
) > RPC_MAXNETNAMELEN
) {
1368 return ERR_PTR(-EINVAL
);
1370 xprt
->servername
= kstrdup(args
->servername
, GFP_KERNEL
);
1371 if (xprt
->servername
== NULL
) {
1373 return ERR_PTR(-ENOMEM
);
1376 rpc_xprt_debugfs_register(xprt
);
1378 dprintk("RPC: created transport %p with %u slots\n", xprt
,
1385 * xprt_destroy - destroy an RPC transport, killing off all requests.
1386 * @xprt: transport to destroy
1389 static void xprt_destroy(struct rpc_xprt
*xprt
)
1391 dprintk("RPC: destroying transport %p\n", xprt
);
1392 del_timer_sync(&xprt
->timer
);
1394 rpc_xprt_debugfs_unregister(xprt
);
1395 rpc_destroy_wait_queue(&xprt
->binding
);
1396 rpc_destroy_wait_queue(&xprt
->pending
);
1397 rpc_destroy_wait_queue(&xprt
->sending
);
1398 rpc_destroy_wait_queue(&xprt
->backlog
);
1399 cancel_work_sync(&xprt
->task_cleanup
);
1400 kfree(xprt
->servername
);
1402 * Tear down transport state and free the rpc_xprt
1404 xprt
->ops
->destroy(xprt
);
1408 * xprt_put - release a reference to an RPC transport.
1409 * @xprt: pointer to the transport
1412 void xprt_put(struct rpc_xprt
*xprt
)
1414 if (atomic_dec_and_test(&xprt
->count
))