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, it installs a timer that
16 * is run after the packet's timeout has expired.
17 * - When a packet arrives, the data_ready handler walks the list of
18 * pending requests for that transport. If a matching XID is found, the
19 * caller is woken up, and the timer removed.
20 * - When no reply arrives within the timeout interval, the timer is
21 * fired by the kernel and runs xprt_timer(). It either adjusts the
22 * timeout values (minor timeout) or wakes up the caller with a status
24 * - When the caller receives a notification from RPC that a reply arrived,
25 * it should release the RPC slot, and process the reply.
26 * If the call timed out, it may choose to retry the operation by
27 * adjusting the initial timeout value, and simply calling rpc_call
30 * Support for async RPC is done through a set of RPC-specific scheduling
31 * primitives that `transparently' work for processes as well as async
32 * tasks that rely on callbacks.
34 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
36 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
39 #include <linux/module.h>
41 #include <linux/types.h>
42 #include <linux/interrupt.h>
43 #include <linux/workqueue.h>
44 #include <linux/net.h>
46 #include <linux/sunrpc/clnt.h>
47 #include <linux/sunrpc/metrics.h>
54 # define RPCDBG_FACILITY RPCDBG_XPRT
60 static void xprt_request_init(struct rpc_task
*, struct rpc_xprt
*);
61 static inline void do_xprt_reserve(struct rpc_task
*);
62 static void xprt_connect_status(struct rpc_task
*task
);
63 static int __xprt_get_cong(struct rpc_xprt
*, struct rpc_task
*);
65 static DEFINE_SPINLOCK(xprt_list_lock
);
66 static LIST_HEAD(xprt_list
);
69 * The transport code maintains an estimate on the maximum number of out-
70 * standing RPC requests, using a smoothed version of the congestion
71 * avoidance implemented in 44BSD. This is basically the Van Jacobson
72 * congestion algorithm: If a retransmit occurs, the congestion window is
73 * halved; otherwise, it is incremented by 1/cwnd when
75 * - a reply is received and
76 * - a full number of requests are outstanding and
77 * - the congestion window hasn't been updated recently.
79 #define RPC_CWNDSHIFT (8U)
80 #define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT)
81 #define RPC_INITCWND RPC_CWNDSCALE
82 #define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT)
84 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
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
)
112 if (try_module_get(THIS_MODULE
)) {
113 list_add_tail(&transport
->list
, &xprt_list
);
114 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
);
146 module_put(THIS_MODULE
);
153 spin_unlock(&xprt_list_lock
);
156 EXPORT_SYMBOL_GPL(xprt_unregister_transport
);
159 * xprt_reserve_xprt - serialize write access to transports
160 * @task: task that is requesting access to the transport
162 * This prevents mixing the payload of separate requests, and prevents
163 * transport connects from colliding with writes. No congestion control
166 int xprt_reserve_xprt(struct rpc_task
*task
)
168 struct rpc_xprt
*xprt
= task
->tk_xprt
;
169 struct rpc_rqst
*req
= task
->tk_rqstp
;
171 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
)) {
172 if (task
== xprt
->snd_task
)
178 xprt
->snd_task
= task
;
180 req
->rq_bytes_sent
= 0;
186 dprintk("RPC: %5u failed to lock transport %p\n",
188 task
->tk_timeout
= 0;
189 task
->tk_status
= -EAGAIN
;
190 if (req
&& req
->rq_ntrans
)
191 rpc_sleep_on(&xprt
->resend
, task
, NULL
);
193 rpc_sleep_on(&xprt
->sending
, task
, NULL
);
196 EXPORT_SYMBOL_GPL(xprt_reserve_xprt
);
198 static void xprt_clear_locked(struct rpc_xprt
*xprt
)
200 xprt
->snd_task
= NULL
;
201 if (!test_bit(XPRT_CLOSE_WAIT
, &xprt
->state
) || xprt
->shutdown
) {
202 smp_mb__before_clear_bit();
203 clear_bit(XPRT_LOCKED
, &xprt
->state
);
204 smp_mb__after_clear_bit();
206 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
210 * xprt_reserve_xprt_cong - serialize write access to transports
211 * @task: task that is requesting access to the transport
213 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
214 * integrated into the decision of whether a request is allowed to be
215 * woken up and given access to the transport.
217 int xprt_reserve_xprt_cong(struct rpc_task
*task
)
219 struct rpc_xprt
*xprt
= task
->tk_xprt
;
220 struct rpc_rqst
*req
= task
->tk_rqstp
;
222 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
)) {
223 if (task
== xprt
->snd_task
)
227 if (__xprt_get_cong(xprt
, task
)) {
228 xprt
->snd_task
= task
;
230 req
->rq_bytes_sent
= 0;
235 xprt_clear_locked(xprt
);
237 dprintk("RPC: %5u failed to lock transport %p\n", task
->tk_pid
, xprt
);
238 task
->tk_timeout
= 0;
239 task
->tk_status
= -EAGAIN
;
240 if (req
&& req
->rq_ntrans
)
241 rpc_sleep_on(&xprt
->resend
, task
, NULL
);
243 rpc_sleep_on(&xprt
->sending
, task
, NULL
);
246 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong
);
248 static inline int xprt_lock_write(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
252 spin_lock_bh(&xprt
->transport_lock
);
253 retval
= xprt
->ops
->reserve_xprt(task
);
254 spin_unlock_bh(&xprt
->transport_lock
);
258 static void __xprt_lock_write_next(struct rpc_xprt
*xprt
)
260 struct rpc_task
*task
;
261 struct rpc_rqst
*req
;
263 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
266 task
= rpc_wake_up_next(&xprt
->resend
);
268 task
= rpc_wake_up_next(&xprt
->sending
);
273 req
= task
->tk_rqstp
;
274 xprt
->snd_task
= task
;
276 req
->rq_bytes_sent
= 0;
282 xprt_clear_locked(xprt
);
285 static void __xprt_lock_write_next_cong(struct rpc_xprt
*xprt
)
287 struct rpc_task
*task
;
289 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
291 if (RPCXPRT_CONGESTED(xprt
))
293 task
= rpc_wake_up_next(&xprt
->resend
);
295 task
= rpc_wake_up_next(&xprt
->sending
);
299 if (__xprt_get_cong(xprt
, task
)) {
300 struct rpc_rqst
*req
= task
->tk_rqstp
;
301 xprt
->snd_task
= task
;
303 req
->rq_bytes_sent
= 0;
309 xprt_clear_locked(xprt
);
313 * xprt_release_xprt - allow other requests to use a transport
314 * @xprt: transport with other tasks potentially waiting
315 * @task: task that is releasing access to the transport
317 * Note that "task" can be NULL. No congestion control is provided.
319 void xprt_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
321 if (xprt
->snd_task
== task
) {
322 xprt_clear_locked(xprt
);
323 __xprt_lock_write_next(xprt
);
326 EXPORT_SYMBOL_GPL(xprt_release_xprt
);
329 * xprt_release_xprt_cong - allow other requests to use a transport
330 * @xprt: transport with other tasks potentially waiting
331 * @task: task that is releasing access to the transport
333 * Note that "task" can be NULL. Another task is awoken to use the
334 * transport if the transport's congestion window allows it.
336 void xprt_release_xprt_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
338 if (xprt
->snd_task
== task
) {
339 xprt_clear_locked(xprt
);
340 __xprt_lock_write_next_cong(xprt
);
343 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong
);
345 static inline void xprt_release_write(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
347 spin_lock_bh(&xprt
->transport_lock
);
348 xprt
->ops
->release_xprt(xprt
, task
);
349 spin_unlock_bh(&xprt
->transport_lock
);
353 * Van Jacobson congestion avoidance. Check if the congestion window
354 * overflowed. Put the task to sleep if this is the case.
357 __xprt_get_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
359 struct rpc_rqst
*req
= task
->tk_rqstp
;
363 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
364 task
->tk_pid
, xprt
->cong
, xprt
->cwnd
);
365 if (RPCXPRT_CONGESTED(xprt
))
368 xprt
->cong
+= RPC_CWNDSCALE
;
373 * Adjust the congestion window, and wake up the next task
374 * that has been sleeping due to congestion
377 __xprt_put_cong(struct rpc_xprt
*xprt
, struct rpc_rqst
*req
)
382 xprt
->cong
-= RPC_CWNDSCALE
;
383 __xprt_lock_write_next_cong(xprt
);
387 * xprt_release_rqst_cong - housekeeping when request is complete
388 * @task: RPC request that recently completed
390 * Useful for transports that require congestion control.
392 void xprt_release_rqst_cong(struct rpc_task
*task
)
394 __xprt_put_cong(task
->tk_xprt
, task
->tk_rqstp
);
396 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong
);
399 * xprt_adjust_cwnd - adjust transport congestion window
400 * @task: recently completed RPC request used to adjust window
401 * @result: result code of completed RPC request
403 * We use a time-smoothed congestion estimator to avoid heavy oscillation.
405 void xprt_adjust_cwnd(struct rpc_task
*task
, int result
)
407 struct rpc_rqst
*req
= task
->tk_rqstp
;
408 struct rpc_xprt
*xprt
= task
->tk_xprt
;
409 unsigned long cwnd
= xprt
->cwnd
;
411 if (result
>= 0 && cwnd
<= xprt
->cong
) {
412 /* The (cwnd >> 1) term makes sure
413 * the result gets rounded properly. */
414 cwnd
+= (RPC_CWNDSCALE
* RPC_CWNDSCALE
+ (cwnd
>> 1)) / cwnd
;
415 if (cwnd
> RPC_MAXCWND(xprt
))
416 cwnd
= RPC_MAXCWND(xprt
);
417 __xprt_lock_write_next_cong(xprt
);
418 } else if (result
== -ETIMEDOUT
) {
420 if (cwnd
< RPC_CWNDSCALE
)
421 cwnd
= RPC_CWNDSCALE
;
423 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
424 xprt
->cong
, xprt
->cwnd
, cwnd
);
426 __xprt_put_cong(xprt
, req
);
428 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd
);
431 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
432 * @xprt: transport with waiting tasks
433 * @status: result code to plant in each task before waking it
436 void xprt_wake_pending_tasks(struct rpc_xprt
*xprt
, int status
)
439 rpc_wake_up_status(&xprt
->pending
, status
);
441 rpc_wake_up(&xprt
->pending
);
443 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks
);
446 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
447 * @task: task to be put to sleep
448 * @action: function pointer to be executed after wait
450 void xprt_wait_for_buffer_space(struct rpc_task
*task
, rpc_action action
)
452 struct rpc_rqst
*req
= task
->tk_rqstp
;
453 struct rpc_xprt
*xprt
= req
->rq_xprt
;
455 task
->tk_timeout
= req
->rq_timeout
;
456 rpc_sleep_on(&xprt
->pending
, task
, action
);
458 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space
);
461 * xprt_write_space - wake the task waiting for transport output buffer space
462 * @xprt: transport with waiting tasks
464 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
466 void xprt_write_space(struct rpc_xprt
*xprt
)
468 if (unlikely(xprt
->shutdown
))
471 spin_lock_bh(&xprt
->transport_lock
);
472 if (xprt
->snd_task
) {
473 dprintk("RPC: write space: waking waiting task on "
475 rpc_wake_up_queued_task(&xprt
->pending
, xprt
->snd_task
);
477 spin_unlock_bh(&xprt
->transport_lock
);
479 EXPORT_SYMBOL_GPL(xprt_write_space
);
482 * xprt_set_retrans_timeout_def - set a request's retransmit timeout
483 * @task: task whose timeout is to be set
485 * Set a request's retransmit timeout based on the transport's
486 * default timeout parameters. Used by transports that don't adjust
487 * the retransmit timeout based on round-trip time estimation.
489 void xprt_set_retrans_timeout_def(struct rpc_task
*task
)
491 task
->tk_timeout
= task
->tk_rqstp
->rq_timeout
;
493 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def
);
496 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
497 * @task: task whose timeout is to be set
499 * Set a request's retransmit timeout using the RTT estimator.
501 void xprt_set_retrans_timeout_rtt(struct rpc_task
*task
)
503 int timer
= task
->tk_msg
.rpc_proc
->p_timer
;
504 struct rpc_clnt
*clnt
= task
->tk_client
;
505 struct rpc_rtt
*rtt
= clnt
->cl_rtt
;
506 struct rpc_rqst
*req
= task
->tk_rqstp
;
507 unsigned long max_timeout
= clnt
->cl_timeout
->to_maxval
;
509 task
->tk_timeout
= rpc_calc_rto(rtt
, timer
);
510 task
->tk_timeout
<<= rpc_ntimeo(rtt
, timer
) + req
->rq_retries
;
511 if (task
->tk_timeout
> max_timeout
|| task
->tk_timeout
== 0)
512 task
->tk_timeout
= max_timeout
;
514 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt
);
516 static void xprt_reset_majortimeo(struct rpc_rqst
*req
)
518 const struct rpc_timeout
*to
= req
->rq_task
->tk_client
->cl_timeout
;
520 req
->rq_majortimeo
= req
->rq_timeout
;
521 if (to
->to_exponential
)
522 req
->rq_majortimeo
<<= to
->to_retries
;
524 req
->rq_majortimeo
+= to
->to_increment
* to
->to_retries
;
525 if (req
->rq_majortimeo
> to
->to_maxval
|| req
->rq_majortimeo
== 0)
526 req
->rq_majortimeo
= to
->to_maxval
;
527 req
->rq_majortimeo
+= jiffies
;
531 * xprt_adjust_timeout - adjust timeout values for next retransmit
532 * @req: RPC request containing parameters to use for the adjustment
535 int xprt_adjust_timeout(struct rpc_rqst
*req
)
537 struct rpc_xprt
*xprt
= req
->rq_xprt
;
538 const struct rpc_timeout
*to
= req
->rq_task
->tk_client
->cl_timeout
;
541 if (time_before(jiffies
, req
->rq_majortimeo
)) {
542 if (to
->to_exponential
)
543 req
->rq_timeout
<<= 1;
545 req
->rq_timeout
+= to
->to_increment
;
546 if (to
->to_maxval
&& req
->rq_timeout
>= to
->to_maxval
)
547 req
->rq_timeout
= to
->to_maxval
;
550 req
->rq_timeout
= to
->to_initval
;
552 xprt_reset_majortimeo(req
);
553 /* Reset the RTT counters == "slow start" */
554 spin_lock_bh(&xprt
->transport_lock
);
555 rpc_init_rtt(req
->rq_task
->tk_client
->cl_rtt
, to
->to_initval
);
556 spin_unlock_bh(&xprt
->transport_lock
);
560 if (req
->rq_timeout
== 0) {
561 printk(KERN_WARNING
"xprt_adjust_timeout: rq_timeout = 0!\n");
562 req
->rq_timeout
= 5 * HZ
;
567 static void xprt_autoclose(struct work_struct
*work
)
569 struct rpc_xprt
*xprt
=
570 container_of(work
, struct rpc_xprt
, task_cleanup
);
572 xprt
->ops
->close(xprt
);
573 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
574 xprt_release_write(xprt
, NULL
);
578 * xprt_disconnect_done - mark a transport as disconnected
579 * @xprt: transport to flag for disconnect
582 void xprt_disconnect_done(struct rpc_xprt
*xprt
)
584 dprintk("RPC: disconnected transport %p\n", xprt
);
585 spin_lock_bh(&xprt
->transport_lock
);
586 xprt_clear_connected(xprt
);
587 xprt_wake_pending_tasks(xprt
, -ENOTCONN
);
588 spin_unlock_bh(&xprt
->transport_lock
);
590 EXPORT_SYMBOL_GPL(xprt_disconnect_done
);
593 * xprt_force_disconnect - force a transport to disconnect
594 * @xprt: transport to disconnect
597 void xprt_force_disconnect(struct rpc_xprt
*xprt
)
599 /* Don't race with the test_bit() in xprt_clear_locked() */
600 spin_lock_bh(&xprt
->transport_lock
);
601 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
602 /* Try to schedule an autoclose RPC call */
603 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
604 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
605 xprt_wake_pending_tasks(xprt
, -ENOTCONN
);
606 spin_unlock_bh(&xprt
->transport_lock
);
610 * xprt_conditional_disconnect - force a transport to disconnect
611 * @xprt: transport to disconnect
612 * @cookie: 'connection cookie'
614 * This attempts to break the connection if and only if 'cookie' matches
615 * the current transport 'connection cookie'. It ensures that we don't
616 * try to break the connection more than once when we need to retransmit
617 * a batch of RPC requests.
620 void xprt_conditional_disconnect(struct rpc_xprt
*xprt
, unsigned int cookie
)
622 /* Don't race with the test_bit() in xprt_clear_locked() */
623 spin_lock_bh(&xprt
->transport_lock
);
624 if (cookie
!= xprt
->connect_cookie
)
626 if (test_bit(XPRT_CLOSING
, &xprt
->state
) || !xprt_connected(xprt
))
628 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
629 /* Try to schedule an autoclose RPC call */
630 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
631 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
632 xprt_wake_pending_tasks(xprt
, -ENOTCONN
);
634 spin_unlock_bh(&xprt
->transport_lock
);
638 xprt_init_autodisconnect(unsigned long data
)
640 struct rpc_xprt
*xprt
= (struct rpc_xprt
*)data
;
642 spin_lock(&xprt
->transport_lock
);
643 if (!list_empty(&xprt
->recv
) || xprt
->shutdown
)
645 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
647 spin_unlock(&xprt
->transport_lock
);
648 set_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
);
649 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
652 spin_unlock(&xprt
->transport_lock
);
656 * xprt_connect - schedule a transport connect operation
657 * @task: RPC task that is requesting the connect
660 void xprt_connect(struct rpc_task
*task
)
662 struct rpc_xprt
*xprt
= task
->tk_xprt
;
664 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task
->tk_pid
,
665 xprt
, (xprt_connected(xprt
) ? "is" : "is not"));
667 if (!xprt_bound(xprt
)) {
668 task
->tk_status
= -EIO
;
671 if (!xprt_lock_write(xprt
, task
))
673 if (xprt_connected(xprt
))
674 xprt_release_write(xprt
, task
);
677 task
->tk_rqstp
->rq_bytes_sent
= 0;
679 task
->tk_timeout
= xprt
->connect_timeout
;
680 rpc_sleep_on(&xprt
->pending
, task
, xprt_connect_status
);
681 xprt
->stat
.connect_start
= jiffies
;
682 xprt
->ops
->connect(task
);
687 static void xprt_connect_status(struct rpc_task
*task
)
689 struct rpc_xprt
*xprt
= task
->tk_xprt
;
691 if (task
->tk_status
== 0) {
692 xprt
->stat
.connect_count
++;
693 xprt
->stat
.connect_time
+= (long)jiffies
- xprt
->stat
.connect_start
;
694 dprintk("RPC: %5u xprt_connect_status: connection established\n",
699 switch (task
->tk_status
) {
701 dprintk("RPC: %5u xprt_connect_status: connection broken\n",
705 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
706 "out\n", task
->tk_pid
);
709 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
710 "server %s\n", task
->tk_pid
, -task
->tk_status
,
711 task
->tk_client
->cl_server
);
712 xprt_release_write(xprt
, task
);
713 task
->tk_status
= -EIO
;
718 * xprt_lookup_rqst - find an RPC request corresponding to an XID
719 * @xprt: transport on which the original request was transmitted
720 * @xid: RPC XID of incoming reply
723 struct rpc_rqst
*xprt_lookup_rqst(struct rpc_xprt
*xprt
, __be32 xid
)
725 struct list_head
*pos
;
727 list_for_each(pos
, &xprt
->recv
) {
728 struct rpc_rqst
*entry
= list_entry(pos
, struct rpc_rqst
, rq_list
);
729 if (entry
->rq_xid
== xid
)
733 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
735 xprt
->stat
.bad_xids
++;
738 EXPORT_SYMBOL_GPL(xprt_lookup_rqst
);
741 * xprt_update_rtt - update an RPC client's RTT state after receiving a reply
742 * @task: RPC request that recently completed
745 void xprt_update_rtt(struct rpc_task
*task
)
747 struct rpc_rqst
*req
= task
->tk_rqstp
;
748 struct rpc_rtt
*rtt
= task
->tk_client
->cl_rtt
;
749 unsigned timer
= task
->tk_msg
.rpc_proc
->p_timer
;
752 if (req
->rq_ntrans
== 1)
753 rpc_update_rtt(rtt
, timer
,
754 (long)jiffies
- req
->rq_xtime
);
755 rpc_set_timeo(rtt
, timer
, req
->rq_ntrans
- 1);
758 EXPORT_SYMBOL_GPL(xprt_update_rtt
);
761 * xprt_complete_rqst - called when reply processing is complete
762 * @task: RPC request that recently completed
763 * @copied: actual number of bytes received from the transport
765 * Caller holds transport lock.
767 void xprt_complete_rqst(struct rpc_task
*task
, int copied
)
769 struct rpc_rqst
*req
= task
->tk_rqstp
;
770 struct rpc_xprt
*xprt
= req
->rq_xprt
;
772 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
773 task
->tk_pid
, ntohl(req
->rq_xid
), copied
);
776 task
->tk_rtt
= (long)jiffies
- req
->rq_xtime
;
778 list_del_init(&req
->rq_list
);
779 req
->rq_private_buf
.len
= copied
;
780 /* Ensure all writes are done before we update req->rq_received */
782 req
->rq_received
= copied
;
783 rpc_wake_up_queued_task(&xprt
->pending
, task
);
785 EXPORT_SYMBOL_GPL(xprt_complete_rqst
);
787 static void xprt_timer(struct rpc_task
*task
)
789 struct rpc_rqst
*req
= task
->tk_rqstp
;
790 struct rpc_xprt
*xprt
= req
->rq_xprt
;
792 if (task
->tk_status
!= -ETIMEDOUT
)
794 dprintk("RPC: %5u xprt_timer\n", task
->tk_pid
);
796 spin_lock_bh(&xprt
->transport_lock
);
797 if (!req
->rq_received
) {
798 if (xprt
->ops
->timer
)
799 xprt
->ops
->timer(task
);
802 spin_unlock_bh(&xprt
->transport_lock
);
806 * xprt_prepare_transmit - reserve the transport before sending a request
807 * @task: RPC task about to send a request
810 int xprt_prepare_transmit(struct rpc_task
*task
)
812 struct rpc_rqst
*req
= task
->tk_rqstp
;
813 struct rpc_xprt
*xprt
= req
->rq_xprt
;
816 dprintk("RPC: %5u xprt_prepare_transmit\n", task
->tk_pid
);
818 spin_lock_bh(&xprt
->transport_lock
);
819 if (req
->rq_received
&& !req
->rq_bytes_sent
) {
820 err
= req
->rq_received
;
823 if (!xprt
->ops
->reserve_xprt(task
)) {
828 if (!xprt_connected(xprt
)) {
833 spin_unlock_bh(&xprt
->transport_lock
);
837 void xprt_end_transmit(struct rpc_task
*task
)
839 xprt_release_write(task
->tk_xprt
, task
);
843 * xprt_transmit - send an RPC request on a transport
844 * @task: controlling RPC task
846 * We have to copy the iovec because sendmsg fiddles with its contents.
848 void xprt_transmit(struct rpc_task
*task
)
850 struct rpc_rqst
*req
= task
->tk_rqstp
;
851 struct rpc_xprt
*xprt
= req
->rq_xprt
;
854 dprintk("RPC: %5u xprt_transmit(%u)\n", task
->tk_pid
, req
->rq_slen
);
856 if (!req
->rq_received
) {
857 if (list_empty(&req
->rq_list
)) {
858 spin_lock_bh(&xprt
->transport_lock
);
859 /* Update the softirq receive buffer */
860 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
,
861 sizeof(req
->rq_private_buf
));
862 /* Add request to the receive list */
863 list_add_tail(&req
->rq_list
, &xprt
->recv
);
864 spin_unlock_bh(&xprt
->transport_lock
);
865 xprt_reset_majortimeo(req
);
866 /* Turn off autodisconnect */
867 del_singleshot_timer_sync(&xprt
->timer
);
869 } else if (!req
->rq_bytes_sent
)
872 req
->rq_connect_cookie
= xprt
->connect_cookie
;
873 req
->rq_xtime
= jiffies
;
874 status
= xprt
->ops
->send_request(task
);
876 dprintk("RPC: %5u xmit complete\n", task
->tk_pid
);
877 spin_lock_bh(&xprt
->transport_lock
);
879 xprt
->ops
->set_retrans_timeout(task
);
882 xprt
->stat
.req_u
+= xprt
->stat
.sends
- xprt
->stat
.recvs
;
883 xprt
->stat
.bklog_u
+= xprt
->backlog
.qlen
;
885 /* Don't race with disconnect */
886 if (!xprt_connected(xprt
))
887 task
->tk_status
= -ENOTCONN
;
888 else if (!req
->rq_received
)
889 rpc_sleep_on(&xprt
->pending
, task
, xprt_timer
);
890 spin_unlock_bh(&xprt
->transport_lock
);
894 /* Note: at this point, task->tk_sleeping has not yet been set,
895 * hence there is no danger of the waking up task being put on
896 * schedq, and being picked up by a parallel run of rpciod().
898 task
->tk_status
= status
;
899 if (status
== -ECONNREFUSED
)
900 rpc_sleep_on(&xprt
->sending
, task
, NULL
);
903 static inline void do_xprt_reserve(struct rpc_task
*task
)
905 struct rpc_xprt
*xprt
= task
->tk_xprt
;
910 if (!list_empty(&xprt
->free
)) {
911 struct rpc_rqst
*req
= list_entry(xprt
->free
.next
, struct rpc_rqst
, rq_list
);
912 list_del_init(&req
->rq_list
);
913 task
->tk_rqstp
= req
;
914 xprt_request_init(task
, xprt
);
917 dprintk("RPC: waiting for request slot\n");
918 task
->tk_status
= -EAGAIN
;
919 task
->tk_timeout
= 0;
920 rpc_sleep_on(&xprt
->backlog
, task
, NULL
);
924 * xprt_reserve - allocate an RPC request slot
925 * @task: RPC task requesting a slot allocation
927 * If no more slots are available, place the task on the transport's
930 void xprt_reserve(struct rpc_task
*task
)
932 struct rpc_xprt
*xprt
= task
->tk_xprt
;
934 task
->tk_status
= -EIO
;
935 spin_lock(&xprt
->reserve_lock
);
936 do_xprt_reserve(task
);
937 spin_unlock(&xprt
->reserve_lock
);
940 static inline __be32
xprt_alloc_xid(struct rpc_xprt
*xprt
)
945 static inline void xprt_init_xid(struct rpc_xprt
*xprt
)
947 xprt
->xid
= net_random();
950 static void xprt_request_init(struct rpc_task
*task
, struct rpc_xprt
*xprt
)
952 struct rpc_rqst
*req
= task
->tk_rqstp
;
954 req
->rq_timeout
= task
->tk_client
->cl_timeout
->to_initval
;
957 req
->rq_buffer
= NULL
;
958 req
->rq_xid
= xprt_alloc_xid(xprt
);
959 req
->rq_release_snd_buf
= NULL
;
960 xprt_reset_majortimeo(req
);
961 dprintk("RPC: %5u reserved req %p xid %08x\n", task
->tk_pid
,
962 req
, ntohl(req
->rq_xid
));
966 * xprt_release - release an RPC request slot
967 * @task: task which is finished with the slot
970 void xprt_release(struct rpc_task
*task
)
972 struct rpc_xprt
*xprt
= task
->tk_xprt
;
973 struct rpc_rqst
*req
;
975 if (!(req
= task
->tk_rqstp
))
977 rpc_count_iostats(task
);
978 spin_lock_bh(&xprt
->transport_lock
);
979 xprt
->ops
->release_xprt(xprt
, task
);
980 if (xprt
->ops
->release_request
)
981 xprt
->ops
->release_request(task
);
982 if (!list_empty(&req
->rq_list
))
983 list_del(&req
->rq_list
);
984 xprt
->last_used
= jiffies
;
985 if (list_empty(&xprt
->recv
))
986 mod_timer(&xprt
->timer
,
987 xprt
->last_used
+ xprt
->idle_timeout
);
988 spin_unlock_bh(&xprt
->transport_lock
);
989 xprt
->ops
->buf_free(req
->rq_buffer
);
990 task
->tk_rqstp
= NULL
;
991 if (req
->rq_release_snd_buf
)
992 req
->rq_release_snd_buf(req
);
993 memset(req
, 0, sizeof(*req
)); /* mark unused */
995 dprintk("RPC: %5u release request %p\n", task
->tk_pid
, req
);
997 spin_lock(&xprt
->reserve_lock
);
998 list_add(&req
->rq_list
, &xprt
->free
);
999 rpc_wake_up_next(&xprt
->backlog
);
1000 spin_unlock(&xprt
->reserve_lock
);
1004 * xprt_create_transport - create an RPC transport
1005 * @args: rpc transport creation arguments
1008 struct rpc_xprt
*xprt_create_transport(struct xprt_create
*args
)
1010 struct rpc_xprt
*xprt
;
1011 struct rpc_rqst
*req
;
1012 struct xprt_class
*t
;
1014 spin_lock(&xprt_list_lock
);
1015 list_for_each_entry(t
, &xprt_list
, list
) {
1016 if (t
->ident
== args
->ident
) {
1017 spin_unlock(&xprt_list_lock
);
1021 spin_unlock(&xprt_list_lock
);
1022 printk(KERN_ERR
"RPC: transport (%d) not supported\n", args
->ident
);
1023 return ERR_PTR(-EIO
);
1026 xprt
= t
->setup(args
);
1028 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1033 kref_init(&xprt
->kref
);
1034 spin_lock_init(&xprt
->transport_lock
);
1035 spin_lock_init(&xprt
->reserve_lock
);
1037 INIT_LIST_HEAD(&xprt
->free
);
1038 INIT_LIST_HEAD(&xprt
->recv
);
1039 INIT_WORK(&xprt
->task_cleanup
, xprt_autoclose
);
1040 setup_timer(&xprt
->timer
, xprt_init_autodisconnect
,
1041 (unsigned long)xprt
);
1042 xprt
->last_used
= jiffies
;
1043 xprt
->cwnd
= RPC_INITCWND
;
1044 xprt
->bind_index
= 0;
1046 rpc_init_wait_queue(&xprt
->binding
, "xprt_binding");
1047 rpc_init_wait_queue(&xprt
->pending
, "xprt_pending");
1048 rpc_init_wait_queue(&xprt
->sending
, "xprt_sending");
1049 rpc_init_wait_queue(&xprt
->resend
, "xprt_resend");
1050 rpc_init_priority_wait_queue(&xprt
->backlog
, "xprt_backlog");
1052 /* initialize free list */
1053 for (req
= &xprt
->slot
[xprt
->max_reqs
-1]; req
>= &xprt
->slot
[0]; req
--)
1054 list_add(&req
->rq_list
, &xprt
->free
);
1056 xprt_init_xid(xprt
);
1058 dprintk("RPC: created transport %p with %u slots\n", xprt
,
1065 * xprt_destroy - destroy an RPC transport, killing off all requests.
1066 * @kref: kref for the transport to destroy
1069 static void xprt_destroy(struct kref
*kref
)
1071 struct rpc_xprt
*xprt
= container_of(kref
, struct rpc_xprt
, kref
);
1073 dprintk("RPC: destroying transport %p\n", xprt
);
1075 del_timer_sync(&xprt
->timer
);
1077 rpc_destroy_wait_queue(&xprt
->binding
);
1078 rpc_destroy_wait_queue(&xprt
->pending
);
1079 rpc_destroy_wait_queue(&xprt
->sending
);
1080 rpc_destroy_wait_queue(&xprt
->resend
);
1081 rpc_destroy_wait_queue(&xprt
->backlog
);
1083 * Tear down transport state and free the rpc_xprt
1085 xprt
->ops
->destroy(xprt
);
1089 * xprt_put - release a reference to an RPC transport.
1090 * @xprt: pointer to the transport
1093 void xprt_put(struct rpc_xprt
*xprt
)
1095 kref_put(&xprt
->kref
, xprt_destroy
);
1099 * xprt_get - return a reference to an RPC transport.
1100 * @xprt: pointer to the transport
1103 struct rpc_xprt
*xprt_get(struct rpc_xprt
*xprt
)
1105 kref_get(&xprt
->kref
);