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>
59 # define RPCDBG_FACILITY RPCDBG_XPRT
65 static void xprt_init(struct rpc_xprt
*xprt
, struct net
*net
);
66 static void xprt_request_init(struct rpc_task
*, struct rpc_xprt
*);
67 static void xprt_connect_status(struct rpc_task
*task
);
68 static int __xprt_get_cong(struct rpc_xprt
*, struct rpc_task
*);
69 static void xprt_destroy(struct rpc_xprt
*xprt
);
71 static DEFINE_SPINLOCK(xprt_list_lock
);
72 static LIST_HEAD(xprt_list
);
75 * xprt_register_transport - register a transport implementation
76 * @transport: transport to register
78 * If a transport implementation is loaded as a kernel module, it can
79 * call this interface to make itself known to the RPC client.
82 * 0: transport successfully registered
83 * -EEXIST: transport already registered
84 * -EINVAL: transport module being unloaded
86 int xprt_register_transport(struct xprt_class
*transport
)
92 spin_lock(&xprt_list_lock
);
93 list_for_each_entry(t
, &xprt_list
, list
) {
94 /* don't register the same transport class twice */
95 if (t
->ident
== transport
->ident
)
99 list_add_tail(&transport
->list
, &xprt_list
);
100 printk(KERN_INFO
"RPC: Registered %s transport module.\n",
105 spin_unlock(&xprt_list_lock
);
108 EXPORT_SYMBOL_GPL(xprt_register_transport
);
111 * xprt_unregister_transport - unregister a transport implementation
112 * @transport: transport to unregister
115 * 0: transport successfully unregistered
116 * -ENOENT: transport never registered
118 int xprt_unregister_transport(struct xprt_class
*transport
)
120 struct xprt_class
*t
;
124 spin_lock(&xprt_list_lock
);
125 list_for_each_entry(t
, &xprt_list
, list
) {
126 if (t
== transport
) {
128 "RPC: Unregistered %s transport module.\n",
130 list_del_init(&transport
->list
);
137 spin_unlock(&xprt_list_lock
);
140 EXPORT_SYMBOL_GPL(xprt_unregister_transport
);
143 * xprt_load_transport - load a transport implementation
144 * @transport_name: transport to load
147 * 0: transport successfully loaded
148 * -ENOENT: transport module not available
150 int xprt_load_transport(const char *transport_name
)
152 struct xprt_class
*t
;
156 spin_lock(&xprt_list_lock
);
157 list_for_each_entry(t
, &xprt_list
, list
) {
158 if (strcmp(t
->name
, transport_name
) == 0) {
159 spin_unlock(&xprt_list_lock
);
163 spin_unlock(&xprt_list_lock
);
164 result
= request_module("xprt%s", transport_name
);
168 EXPORT_SYMBOL_GPL(xprt_load_transport
);
171 * xprt_reserve_xprt - serialize write access to transports
172 * @task: task that is requesting access to the transport
173 * @xprt: pointer to the target transport
175 * This prevents mixing the payload of separate requests, and prevents
176 * transport connects from colliding with writes. No congestion control
179 int xprt_reserve_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
181 struct rpc_rqst
*req
= task
->tk_rqstp
;
184 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
)) {
185 if (task
== xprt
->snd_task
)
189 xprt
->snd_task
= task
;
196 dprintk("RPC: %5u failed to lock transport %p\n",
198 task
->tk_timeout
= 0;
199 task
->tk_status
= -EAGAIN
;
201 priority
= RPC_PRIORITY_LOW
;
202 else if (!req
->rq_ntrans
)
203 priority
= RPC_PRIORITY_NORMAL
;
205 priority
= RPC_PRIORITY_HIGH
;
206 rpc_sleep_on_priority(&xprt
->sending
, task
, NULL
, priority
);
209 EXPORT_SYMBOL_GPL(xprt_reserve_xprt
);
211 static void xprt_clear_locked(struct rpc_xprt
*xprt
)
213 xprt
->snd_task
= NULL
;
214 if (!test_bit(XPRT_CLOSE_WAIT
, &xprt
->state
)) {
215 smp_mb__before_atomic();
216 clear_bit(XPRT_LOCKED
, &xprt
->state
);
217 smp_mb__after_atomic();
219 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
223 * xprt_reserve_xprt_cong - serialize write access to transports
224 * @task: task that is requesting access to the transport
226 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
227 * integrated into the decision of whether a request is allowed to be
228 * woken up and given access to the transport.
230 int xprt_reserve_xprt_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
232 struct rpc_rqst
*req
= task
->tk_rqstp
;
235 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
)) {
236 if (task
== xprt
->snd_task
)
241 xprt
->snd_task
= task
;
244 if (__xprt_get_cong(xprt
, task
)) {
245 xprt
->snd_task
= task
;
249 xprt_clear_locked(xprt
);
251 dprintk("RPC: %5u failed to lock transport %p\n", task
->tk_pid
, xprt
);
252 task
->tk_timeout
= 0;
253 task
->tk_status
= -EAGAIN
;
255 priority
= RPC_PRIORITY_LOW
;
256 else if (!req
->rq_ntrans
)
257 priority
= RPC_PRIORITY_NORMAL
;
259 priority
= RPC_PRIORITY_HIGH
;
260 rpc_sleep_on_priority(&xprt
->sending
, task
, NULL
, priority
);
263 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong
);
265 static inline int xprt_lock_write(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
269 spin_lock_bh(&xprt
->transport_lock
);
270 retval
= xprt
->ops
->reserve_xprt(xprt
, task
);
271 spin_unlock_bh(&xprt
->transport_lock
);
275 static bool __xprt_lock_write_func(struct rpc_task
*task
, void *data
)
277 struct rpc_xprt
*xprt
= data
;
278 struct rpc_rqst
*req
;
280 req
= task
->tk_rqstp
;
281 xprt
->snd_task
= task
;
287 static void __xprt_lock_write_next(struct rpc_xprt
*xprt
)
289 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
292 if (rpc_wake_up_first(&xprt
->sending
, __xprt_lock_write_func
, xprt
))
294 xprt_clear_locked(xprt
);
297 static bool __xprt_lock_write_cong_func(struct rpc_task
*task
, void *data
)
299 struct rpc_xprt
*xprt
= data
;
300 struct rpc_rqst
*req
;
302 req
= task
->tk_rqstp
;
304 xprt
->snd_task
= task
;
307 if (__xprt_get_cong(xprt
, task
)) {
308 xprt
->snd_task
= task
;
315 static void __xprt_lock_write_next_cong(struct rpc_xprt
*xprt
)
317 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
319 if (RPCXPRT_CONGESTED(xprt
))
321 if (rpc_wake_up_first(&xprt
->sending
, __xprt_lock_write_cong_func
, xprt
))
324 xprt_clear_locked(xprt
);
328 * xprt_release_xprt - allow other requests to use a transport
329 * @xprt: transport with other tasks potentially waiting
330 * @task: task that is releasing access to the transport
332 * Note that "task" can be NULL. No congestion control is provided.
334 void xprt_release_xprt(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
336 if (xprt
->snd_task
== task
) {
338 struct rpc_rqst
*req
= task
->tk_rqstp
;
340 req
->rq_bytes_sent
= 0;
342 xprt_clear_locked(xprt
);
343 __xprt_lock_write_next(xprt
);
346 EXPORT_SYMBOL_GPL(xprt_release_xprt
);
349 * xprt_release_xprt_cong - allow other requests to use a transport
350 * @xprt: transport with other tasks potentially waiting
351 * @task: task that is releasing access to the transport
353 * Note that "task" can be NULL. Another task is awoken to use the
354 * transport if the transport's congestion window allows it.
356 void xprt_release_xprt_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
358 if (xprt
->snd_task
== task
) {
360 struct rpc_rqst
*req
= task
->tk_rqstp
;
362 req
->rq_bytes_sent
= 0;
364 xprt_clear_locked(xprt
);
365 __xprt_lock_write_next_cong(xprt
);
368 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong
);
370 static inline void xprt_release_write(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
372 spin_lock_bh(&xprt
->transport_lock
);
373 xprt
->ops
->release_xprt(xprt
, task
);
374 spin_unlock_bh(&xprt
->transport_lock
);
378 * Van Jacobson congestion avoidance. Check if the congestion window
379 * overflowed. Put the task to sleep if this is the case.
382 __xprt_get_cong(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
384 struct rpc_rqst
*req
= task
->tk_rqstp
;
388 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
389 task
->tk_pid
, xprt
->cong
, xprt
->cwnd
);
390 if (RPCXPRT_CONGESTED(xprt
))
393 xprt
->cong
+= RPC_CWNDSCALE
;
398 * Adjust the congestion window, and wake up the next task
399 * that has been sleeping due to congestion
402 __xprt_put_cong(struct rpc_xprt
*xprt
, struct rpc_rqst
*req
)
407 xprt
->cong
-= RPC_CWNDSCALE
;
408 __xprt_lock_write_next_cong(xprt
);
412 * xprt_release_rqst_cong - housekeeping when request is complete
413 * @task: RPC request that recently completed
415 * Useful for transports that require congestion control.
417 void xprt_release_rqst_cong(struct rpc_task
*task
)
419 struct rpc_rqst
*req
= task
->tk_rqstp
;
421 __xprt_put_cong(req
->rq_xprt
, req
);
423 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong
);
426 * xprt_adjust_cwnd - adjust transport congestion window
427 * @xprt: pointer to xprt
428 * @task: recently completed RPC request used to adjust window
429 * @result: result code of completed RPC request
431 * The transport code maintains an estimate on the maximum number of out-
432 * standing RPC requests, using a smoothed version of the congestion
433 * avoidance implemented in 44BSD. This is basically the Van Jacobson
434 * congestion algorithm: If a retransmit occurs, the congestion window is
435 * halved; otherwise, it is incremented by 1/cwnd when
437 * - a reply is received and
438 * - a full number of requests are outstanding and
439 * - the congestion window hasn't been updated recently.
441 void xprt_adjust_cwnd(struct rpc_xprt
*xprt
, struct rpc_task
*task
, int result
)
443 struct rpc_rqst
*req
= task
->tk_rqstp
;
444 unsigned long cwnd
= xprt
->cwnd
;
446 if (result
>= 0 && cwnd
<= xprt
->cong
) {
447 /* The (cwnd >> 1) term makes sure
448 * the result gets rounded properly. */
449 cwnd
+= (RPC_CWNDSCALE
* RPC_CWNDSCALE
+ (cwnd
>> 1)) / cwnd
;
450 if (cwnd
> RPC_MAXCWND(xprt
))
451 cwnd
= RPC_MAXCWND(xprt
);
452 __xprt_lock_write_next_cong(xprt
);
453 } else if (result
== -ETIMEDOUT
) {
455 if (cwnd
< RPC_CWNDSCALE
)
456 cwnd
= RPC_CWNDSCALE
;
458 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
459 xprt
->cong
, xprt
->cwnd
, cwnd
);
461 __xprt_put_cong(xprt
, req
);
463 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd
);
466 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
467 * @xprt: transport with waiting tasks
468 * @status: result code to plant in each task before waking it
471 void xprt_wake_pending_tasks(struct rpc_xprt
*xprt
, int status
)
474 rpc_wake_up_status(&xprt
->pending
, status
);
476 rpc_wake_up(&xprt
->pending
);
478 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks
);
481 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
482 * @task: task to be put to sleep
483 * @action: function pointer to be executed after wait
485 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
486 * we don't in general want to force a socket disconnection due to
487 * an incomplete RPC call transmission.
489 void xprt_wait_for_buffer_space(struct rpc_task
*task
, rpc_action action
)
491 struct rpc_rqst
*req
= task
->tk_rqstp
;
492 struct rpc_xprt
*xprt
= req
->rq_xprt
;
494 task
->tk_timeout
= RPC_IS_SOFT(task
) ? req
->rq_timeout
: 0;
495 rpc_sleep_on(&xprt
->pending
, task
, action
);
497 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space
);
500 * xprt_write_space - wake the task waiting for transport output buffer space
501 * @xprt: transport with waiting tasks
503 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
505 void xprt_write_space(struct rpc_xprt
*xprt
)
507 spin_lock_bh(&xprt
->transport_lock
);
508 if (xprt
->snd_task
) {
509 dprintk("RPC: write space: waking waiting task on "
511 rpc_wake_up_queued_task(&xprt
->pending
, xprt
->snd_task
);
513 spin_unlock_bh(&xprt
->transport_lock
);
515 EXPORT_SYMBOL_GPL(xprt_write_space
);
518 * xprt_set_retrans_timeout_def - set a request's retransmit timeout
519 * @task: task whose timeout is to be set
521 * Set a request's retransmit timeout based on the transport's
522 * default timeout parameters. Used by transports that don't adjust
523 * the retransmit timeout based on round-trip time estimation.
525 void xprt_set_retrans_timeout_def(struct rpc_task
*task
)
527 task
->tk_timeout
= task
->tk_rqstp
->rq_timeout
;
529 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def
);
532 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
533 * @task: task whose timeout is to be set
535 * Set a request's retransmit timeout using the RTT estimator.
537 void xprt_set_retrans_timeout_rtt(struct rpc_task
*task
)
539 int timer
= task
->tk_msg
.rpc_proc
->p_timer
;
540 struct rpc_clnt
*clnt
= task
->tk_client
;
541 struct rpc_rtt
*rtt
= clnt
->cl_rtt
;
542 struct rpc_rqst
*req
= task
->tk_rqstp
;
543 unsigned long max_timeout
= clnt
->cl_timeout
->to_maxval
;
545 task
->tk_timeout
= rpc_calc_rto(rtt
, timer
);
546 task
->tk_timeout
<<= rpc_ntimeo(rtt
, timer
) + req
->rq_retries
;
547 if (task
->tk_timeout
> max_timeout
|| task
->tk_timeout
== 0)
548 task
->tk_timeout
= max_timeout
;
550 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt
);
552 static void xprt_reset_majortimeo(struct rpc_rqst
*req
)
554 const struct rpc_timeout
*to
= req
->rq_task
->tk_client
->cl_timeout
;
556 req
->rq_majortimeo
= req
->rq_timeout
;
557 if (to
->to_exponential
)
558 req
->rq_majortimeo
<<= to
->to_retries
;
560 req
->rq_majortimeo
+= to
->to_increment
* to
->to_retries
;
561 if (req
->rq_majortimeo
> to
->to_maxval
|| req
->rq_majortimeo
== 0)
562 req
->rq_majortimeo
= to
->to_maxval
;
563 req
->rq_majortimeo
+= jiffies
;
567 * xprt_adjust_timeout - adjust timeout values for next retransmit
568 * @req: RPC request containing parameters to use for the adjustment
571 int xprt_adjust_timeout(struct rpc_rqst
*req
)
573 struct rpc_xprt
*xprt
= req
->rq_xprt
;
574 const struct rpc_timeout
*to
= req
->rq_task
->tk_client
->cl_timeout
;
577 if (time_before(jiffies
, req
->rq_majortimeo
)) {
578 if (to
->to_exponential
)
579 req
->rq_timeout
<<= 1;
581 req
->rq_timeout
+= to
->to_increment
;
582 if (to
->to_maxval
&& req
->rq_timeout
>= to
->to_maxval
)
583 req
->rq_timeout
= to
->to_maxval
;
586 req
->rq_timeout
= to
->to_initval
;
588 xprt_reset_majortimeo(req
);
589 /* Reset the RTT counters == "slow start" */
590 spin_lock_bh(&xprt
->transport_lock
);
591 rpc_init_rtt(req
->rq_task
->tk_client
->cl_rtt
, to
->to_initval
);
592 spin_unlock_bh(&xprt
->transport_lock
);
596 if (req
->rq_timeout
== 0) {
597 printk(KERN_WARNING
"xprt_adjust_timeout: rq_timeout = 0!\n");
598 req
->rq_timeout
= 5 * HZ
;
603 static void xprt_autoclose(struct work_struct
*work
)
605 struct rpc_xprt
*xprt
=
606 container_of(work
, struct rpc_xprt
, task_cleanup
);
608 xprt
->ops
->close(xprt
);
609 clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
610 xprt_release_write(xprt
, NULL
);
614 * xprt_disconnect_done - mark a transport as disconnected
615 * @xprt: transport to flag for disconnect
618 void xprt_disconnect_done(struct rpc_xprt
*xprt
)
620 dprintk("RPC: disconnected transport %p\n", xprt
);
621 spin_lock_bh(&xprt
->transport_lock
);
622 xprt_clear_connected(xprt
);
623 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
624 spin_unlock_bh(&xprt
->transport_lock
);
626 EXPORT_SYMBOL_GPL(xprt_disconnect_done
);
629 * xprt_force_disconnect - force a transport to disconnect
630 * @xprt: transport to disconnect
633 void xprt_force_disconnect(struct rpc_xprt
*xprt
)
635 /* Don't race with the test_bit() in xprt_clear_locked() */
636 spin_lock_bh(&xprt
->transport_lock
);
637 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
638 /* Try to schedule an autoclose RPC call */
639 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
640 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
641 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
642 spin_unlock_bh(&xprt
->transport_lock
);
646 * xprt_conditional_disconnect - force a transport to disconnect
647 * @xprt: transport to disconnect
648 * @cookie: 'connection cookie'
650 * This attempts to break the connection if and only if 'cookie' matches
651 * the current transport 'connection cookie'. It ensures that we don't
652 * try to break the connection more than once when we need to retransmit
653 * a batch of RPC requests.
656 void xprt_conditional_disconnect(struct rpc_xprt
*xprt
, unsigned int cookie
)
658 /* Don't race with the test_bit() in xprt_clear_locked() */
659 spin_lock_bh(&xprt
->transport_lock
);
660 if (cookie
!= xprt
->connect_cookie
)
662 if (test_bit(XPRT_CLOSING
, &xprt
->state
) || !xprt_connected(xprt
))
664 set_bit(XPRT_CLOSE_WAIT
, &xprt
->state
);
665 /* Try to schedule an autoclose RPC call */
666 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
) == 0)
667 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
668 xprt_wake_pending_tasks(xprt
, -EAGAIN
);
670 spin_unlock_bh(&xprt
->transport_lock
);
674 xprt_init_autodisconnect(unsigned long data
)
676 struct rpc_xprt
*xprt
= (struct rpc_xprt
*)data
;
678 spin_lock(&xprt
->transport_lock
);
679 if (!list_empty(&xprt
->recv
))
681 if (test_and_set_bit(XPRT_LOCKED
, &xprt
->state
))
683 spin_unlock(&xprt
->transport_lock
);
684 set_bit(XPRT_CONNECTION_CLOSE
, &xprt
->state
);
685 queue_work(rpciod_workqueue
, &xprt
->task_cleanup
);
688 spin_unlock(&xprt
->transport_lock
);
692 * xprt_connect - schedule a transport connect operation
693 * @task: RPC task that is requesting the connect
696 void xprt_connect(struct rpc_task
*task
)
698 struct rpc_xprt
*xprt
= task
->tk_rqstp
->rq_xprt
;
700 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task
->tk_pid
,
701 xprt
, (xprt_connected(xprt
) ? "is" : "is not"));
703 if (!xprt_bound(xprt
)) {
704 task
->tk_status
= -EAGAIN
;
707 if (!xprt_lock_write(xprt
, task
))
710 if (test_and_clear_bit(XPRT_CLOSE_WAIT
, &xprt
->state
))
711 xprt
->ops
->close(xprt
);
713 if (xprt_connected(xprt
))
714 xprt_release_write(xprt
, task
);
716 task
->tk_rqstp
->rq_bytes_sent
= 0;
717 task
->tk_timeout
= task
->tk_rqstp
->rq_timeout
;
718 rpc_sleep_on(&xprt
->pending
, task
, xprt_connect_status
);
720 if (test_bit(XPRT_CLOSING
, &xprt
->state
))
722 if (xprt_test_and_set_connecting(xprt
))
724 xprt
->stat
.connect_start
= jiffies
;
725 xprt
->ops
->connect(xprt
, task
);
729 static void xprt_connect_status(struct rpc_task
*task
)
731 struct rpc_xprt
*xprt
= task
->tk_rqstp
->rq_xprt
;
733 if (task
->tk_status
== 0) {
734 xprt
->stat
.connect_count
++;
735 xprt
->stat
.connect_time
+= (long)jiffies
- xprt
->stat
.connect_start
;
736 dprintk("RPC: %5u xprt_connect_status: connection established\n",
741 switch (task
->tk_status
) {
748 dprintk("RPC: %5u xprt_connect_status: retrying\n", task
->tk_pid
);
751 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
752 "out\n", task
->tk_pid
);
755 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
756 "server %s\n", task
->tk_pid
, -task
->tk_status
,
758 xprt_release_write(xprt
, task
);
759 task
->tk_status
= -EIO
;
764 * xprt_lookup_rqst - find an RPC request corresponding to an XID
765 * @xprt: transport on which the original request was transmitted
766 * @xid: RPC XID of incoming reply
769 struct rpc_rqst
*xprt_lookup_rqst(struct rpc_xprt
*xprt
, __be32 xid
)
771 struct rpc_rqst
*entry
;
773 list_for_each_entry(entry
, &xprt
->recv
, rq_list
)
774 if (entry
->rq_xid
== xid
)
777 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
779 xprt
->stat
.bad_xids
++;
782 EXPORT_SYMBOL_GPL(xprt_lookup_rqst
);
784 static void xprt_update_rtt(struct rpc_task
*task
)
786 struct rpc_rqst
*req
= task
->tk_rqstp
;
787 struct rpc_rtt
*rtt
= task
->tk_client
->cl_rtt
;
788 unsigned int timer
= task
->tk_msg
.rpc_proc
->p_timer
;
789 long m
= usecs_to_jiffies(ktime_to_us(req
->rq_rtt
));
792 if (req
->rq_ntrans
== 1)
793 rpc_update_rtt(rtt
, timer
, m
);
794 rpc_set_timeo(rtt
, timer
, req
->rq_ntrans
- 1);
799 * xprt_complete_rqst - called when reply processing is complete
800 * @task: RPC request that recently completed
801 * @copied: actual number of bytes received from the transport
803 * Caller holds transport lock.
805 void xprt_complete_rqst(struct rpc_task
*task
, int copied
)
807 struct rpc_rqst
*req
= task
->tk_rqstp
;
808 struct rpc_xprt
*xprt
= req
->rq_xprt
;
810 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
811 task
->tk_pid
, ntohl(req
->rq_xid
), copied
);
814 req
->rq_rtt
= ktime_sub(ktime_get(), req
->rq_xtime
);
815 if (xprt
->ops
->timer
!= NULL
)
816 xprt_update_rtt(task
);
818 list_del_init(&req
->rq_list
);
819 req
->rq_private_buf
.len
= copied
;
820 /* Ensure all writes are done before we update */
821 /* req->rq_reply_bytes_recvd */
823 req
->rq_reply_bytes_recvd
= copied
;
824 rpc_wake_up_queued_task(&xprt
->pending
, task
);
826 EXPORT_SYMBOL_GPL(xprt_complete_rqst
);
828 static void xprt_timer(struct rpc_task
*task
)
830 struct rpc_rqst
*req
= task
->tk_rqstp
;
831 struct rpc_xprt
*xprt
= req
->rq_xprt
;
833 if (task
->tk_status
!= -ETIMEDOUT
)
835 dprintk("RPC: %5u xprt_timer\n", task
->tk_pid
);
837 spin_lock_bh(&xprt
->transport_lock
);
838 if (!req
->rq_reply_bytes_recvd
) {
839 if (xprt
->ops
->timer
)
840 xprt
->ops
->timer(xprt
, task
);
843 spin_unlock_bh(&xprt
->transport_lock
);
846 static inline int xprt_has_timer(struct rpc_xprt
*xprt
)
848 return xprt
->idle_timeout
!= 0;
852 * xprt_prepare_transmit - reserve the transport before sending a request
853 * @task: RPC task about to send a request
856 bool xprt_prepare_transmit(struct rpc_task
*task
)
858 struct rpc_rqst
*req
= task
->tk_rqstp
;
859 struct rpc_xprt
*xprt
= req
->rq_xprt
;
862 dprintk("RPC: %5u xprt_prepare_transmit\n", task
->tk_pid
);
864 spin_lock_bh(&xprt
->transport_lock
);
865 if (!req
->rq_bytes_sent
) {
866 if (req
->rq_reply_bytes_recvd
) {
867 task
->tk_status
= req
->rq_reply_bytes_recvd
;
870 if ((task
->tk_flags
& RPC_TASK_NO_RETRANS_TIMEOUT
)
871 && xprt_connected(xprt
)
872 && req
->rq_connect_cookie
== xprt
->connect_cookie
) {
873 xprt
->ops
->set_retrans_timeout(task
);
874 rpc_sleep_on(&xprt
->pending
, task
, xprt_timer
);
878 if (!xprt
->ops
->reserve_xprt(xprt
, task
)) {
879 task
->tk_status
= -EAGAIN
;
884 spin_unlock_bh(&xprt
->transport_lock
);
888 void xprt_end_transmit(struct rpc_task
*task
)
890 xprt_release_write(task
->tk_rqstp
->rq_xprt
, task
);
894 * xprt_transmit - send an RPC request on a transport
895 * @task: controlling RPC task
897 * We have to copy the iovec because sendmsg fiddles with its contents.
899 void xprt_transmit(struct rpc_task
*task
)
901 struct rpc_rqst
*req
= task
->tk_rqstp
;
902 struct rpc_xprt
*xprt
= req
->rq_xprt
;
905 dprintk("RPC: %5u xprt_transmit(%u)\n", task
->tk_pid
, req
->rq_slen
);
907 if (!req
->rq_reply_bytes_recvd
) {
908 if (list_empty(&req
->rq_list
) && rpc_reply_expected(task
)) {
910 * Add to the list only if we're expecting a reply
912 spin_lock_bh(&xprt
->transport_lock
);
913 /* Update the softirq receive buffer */
914 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
,
915 sizeof(req
->rq_private_buf
));
916 /* Add request to the receive list */
917 list_add_tail(&req
->rq_list
, &xprt
->recv
);
918 spin_unlock_bh(&xprt
->transport_lock
);
919 xprt_reset_majortimeo(req
);
920 /* Turn off autodisconnect */
921 del_singleshot_timer_sync(&xprt
->timer
);
923 } else if (!req
->rq_bytes_sent
)
926 req
->rq_xtime
= ktime_get();
927 status
= xprt
->ops
->send_request(task
);
929 task
->tk_status
= status
;
933 dprintk("RPC: %5u xmit complete\n", task
->tk_pid
);
934 task
->tk_flags
|= RPC_TASK_SENT
;
935 spin_lock_bh(&xprt
->transport_lock
);
937 xprt
->ops
->set_retrans_timeout(task
);
939 numreqs
= atomic_read(&xprt
->num_reqs
);
940 if (numreqs
> xprt
->stat
.max_slots
)
941 xprt
->stat
.max_slots
= numreqs
;
943 xprt
->stat
.req_u
+= xprt
->stat
.sends
- xprt
->stat
.recvs
;
944 xprt
->stat
.bklog_u
+= xprt
->backlog
.qlen
;
945 xprt
->stat
.sending_u
+= xprt
->sending
.qlen
;
946 xprt
->stat
.pending_u
+= xprt
->pending
.qlen
;
948 /* Don't race with disconnect */
949 if (!xprt_connected(xprt
))
950 task
->tk_status
= -ENOTCONN
;
953 * Sleep on the pending queue since
954 * we're expecting a reply.
956 if (!req
->rq_reply_bytes_recvd
&& rpc_reply_expected(task
))
957 rpc_sleep_on(&xprt
->pending
, task
, xprt_timer
);
958 req
->rq_connect_cookie
= xprt
->connect_cookie
;
960 spin_unlock_bh(&xprt
->transport_lock
);
963 static void xprt_add_backlog(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
965 set_bit(XPRT_CONGESTED
, &xprt
->state
);
966 rpc_sleep_on(&xprt
->backlog
, task
, NULL
);
969 static void xprt_wake_up_backlog(struct rpc_xprt
*xprt
)
971 if (rpc_wake_up_next(&xprt
->backlog
) == NULL
)
972 clear_bit(XPRT_CONGESTED
, &xprt
->state
);
975 static bool xprt_throttle_congested(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
979 if (!test_bit(XPRT_CONGESTED
, &xprt
->state
))
981 spin_lock(&xprt
->reserve_lock
);
982 if (test_bit(XPRT_CONGESTED
, &xprt
->state
)) {
983 rpc_sleep_on(&xprt
->backlog
, task
, NULL
);
986 spin_unlock(&xprt
->reserve_lock
);
991 static struct rpc_rqst
*xprt_dynamic_alloc_slot(struct rpc_xprt
*xprt
, gfp_t gfp_flags
)
993 struct rpc_rqst
*req
= ERR_PTR(-EAGAIN
);
995 if (!atomic_add_unless(&xprt
->num_reqs
, 1, xprt
->max_reqs
))
997 req
= kzalloc(sizeof(struct rpc_rqst
), gfp_flags
);
1000 atomic_dec(&xprt
->num_reqs
);
1001 req
= ERR_PTR(-ENOMEM
);
1006 static bool xprt_dynamic_free_slot(struct rpc_xprt
*xprt
, struct rpc_rqst
*req
)
1008 if (atomic_add_unless(&xprt
->num_reqs
, -1, xprt
->min_reqs
)) {
1015 void xprt_alloc_slot(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1017 struct rpc_rqst
*req
;
1019 spin_lock(&xprt
->reserve_lock
);
1020 if (!list_empty(&xprt
->free
)) {
1021 req
= list_entry(xprt
->free
.next
, struct rpc_rqst
, rq_list
);
1022 list_del(&req
->rq_list
);
1025 req
= xprt_dynamic_alloc_slot(xprt
, GFP_NOWAIT
|__GFP_NOWARN
);
1028 switch (PTR_ERR(req
)) {
1030 dprintk("RPC: dynamic allocation of request slot "
1031 "failed! Retrying\n");
1032 task
->tk_status
= -ENOMEM
;
1035 xprt_add_backlog(xprt
, task
);
1036 dprintk("RPC: waiting for request slot\n");
1038 task
->tk_status
= -EAGAIN
;
1040 spin_unlock(&xprt
->reserve_lock
);
1043 task
->tk_status
= 0;
1044 task
->tk_rqstp
= req
;
1045 xprt_request_init(task
, xprt
);
1046 spin_unlock(&xprt
->reserve_lock
);
1048 EXPORT_SYMBOL_GPL(xprt_alloc_slot
);
1050 void xprt_lock_and_alloc_slot(struct rpc_xprt
*xprt
, struct rpc_task
*task
)
1052 /* Note: grabbing the xprt_lock_write() ensures that we throttle
1053 * new slot allocation if the transport is congested (i.e. when
1054 * reconnecting a stream transport or when out of socket write
1057 if (xprt_lock_write(xprt
, task
)) {
1058 xprt_alloc_slot(xprt
, task
);
1059 xprt_release_write(xprt
, task
);
1062 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot
);
1064 static void xprt_free_slot(struct rpc_xprt
*xprt
, struct rpc_rqst
*req
)
1066 spin_lock(&xprt
->reserve_lock
);
1067 if (!xprt_dynamic_free_slot(xprt
, req
)) {
1068 memset(req
, 0, sizeof(*req
)); /* mark unused */
1069 list_add(&req
->rq_list
, &xprt
->free
);
1071 xprt_wake_up_backlog(xprt
);
1072 spin_unlock(&xprt
->reserve_lock
);
1075 static void xprt_free_all_slots(struct rpc_xprt
*xprt
)
1077 struct rpc_rqst
*req
;
1078 while (!list_empty(&xprt
->free
)) {
1079 req
= list_first_entry(&xprt
->free
, struct rpc_rqst
, rq_list
);
1080 list_del(&req
->rq_list
);
1085 struct rpc_xprt
*xprt_alloc(struct net
*net
, size_t size
,
1086 unsigned int num_prealloc
,
1087 unsigned int max_alloc
)
1089 struct rpc_xprt
*xprt
;
1090 struct rpc_rqst
*req
;
1093 xprt
= kzalloc(size
, GFP_KERNEL
);
1097 xprt_init(xprt
, net
);
1099 for (i
= 0; i
< num_prealloc
; i
++) {
1100 req
= kzalloc(sizeof(struct rpc_rqst
), GFP_KERNEL
);
1103 list_add(&req
->rq_list
, &xprt
->free
);
1105 if (max_alloc
> num_prealloc
)
1106 xprt
->max_reqs
= max_alloc
;
1108 xprt
->max_reqs
= num_prealloc
;
1109 xprt
->min_reqs
= num_prealloc
;
1110 atomic_set(&xprt
->num_reqs
, num_prealloc
);
1119 EXPORT_SYMBOL_GPL(xprt_alloc
);
1121 void xprt_free(struct rpc_xprt
*xprt
)
1123 put_net(xprt
->xprt_net
);
1124 xprt_free_all_slots(xprt
);
1127 EXPORT_SYMBOL_GPL(xprt_free
);
1130 * xprt_reserve - allocate an RPC request slot
1131 * @task: RPC task requesting a slot allocation
1133 * If the transport is marked as being congested, or if no more
1134 * slots are available, place the task on the transport's
1137 void xprt_reserve(struct rpc_task
*task
)
1139 struct rpc_xprt
*xprt
;
1141 task
->tk_status
= 0;
1142 if (task
->tk_rqstp
!= NULL
)
1145 task
->tk_timeout
= 0;
1146 task
->tk_status
= -EAGAIN
;
1148 xprt
= rcu_dereference(task
->tk_client
->cl_xprt
);
1149 if (!xprt_throttle_congested(xprt
, task
))
1150 xprt
->ops
->alloc_slot(xprt
, task
);
1155 * xprt_retry_reserve - allocate an RPC request slot
1156 * @task: RPC task requesting a slot allocation
1158 * If no more slots are available, place the task on the transport's
1160 * Note that the only difference with xprt_reserve is that we now
1161 * ignore the value of the XPRT_CONGESTED flag.
1163 void xprt_retry_reserve(struct rpc_task
*task
)
1165 struct rpc_xprt
*xprt
;
1167 task
->tk_status
= 0;
1168 if (task
->tk_rqstp
!= NULL
)
1171 task
->tk_timeout
= 0;
1172 task
->tk_status
= -EAGAIN
;
1174 xprt
= rcu_dereference(task
->tk_client
->cl_xprt
);
1175 xprt
->ops
->alloc_slot(xprt
, task
);
1179 static inline __be32
xprt_alloc_xid(struct rpc_xprt
*xprt
)
1181 return (__force __be32
)xprt
->xid
++;
1184 static inline void xprt_init_xid(struct rpc_xprt
*xprt
)
1186 xprt
->xid
= prandom_u32();
1189 static void xprt_request_init(struct rpc_task
*task
, struct rpc_xprt
*xprt
)
1191 struct rpc_rqst
*req
= task
->tk_rqstp
;
1193 INIT_LIST_HEAD(&req
->rq_list
);
1194 req
->rq_timeout
= task
->tk_client
->cl_timeout
->to_initval
;
1195 req
->rq_task
= task
;
1196 req
->rq_xprt
= xprt
;
1197 req
->rq_buffer
= NULL
;
1198 req
->rq_xid
= xprt_alloc_xid(xprt
);
1199 req
->rq_connect_cookie
= xprt
->connect_cookie
- 1;
1200 req
->rq_bytes_sent
= 0;
1201 req
->rq_snd_buf
.len
= 0;
1202 req
->rq_snd_buf
.buflen
= 0;
1203 req
->rq_rcv_buf
.len
= 0;
1204 req
->rq_rcv_buf
.buflen
= 0;
1205 req
->rq_release_snd_buf
= NULL
;
1206 xprt_reset_majortimeo(req
);
1207 dprintk("RPC: %5u reserved req %p xid %08x\n", task
->tk_pid
,
1208 req
, ntohl(req
->rq_xid
));
1212 * xprt_release - release an RPC request slot
1213 * @task: task which is finished with the slot
1216 void xprt_release(struct rpc_task
*task
)
1218 struct rpc_xprt
*xprt
;
1219 struct rpc_rqst
*req
= task
->tk_rqstp
;
1222 if (task
->tk_client
) {
1224 xprt
= rcu_dereference(task
->tk_client
->cl_xprt
);
1225 if (xprt
->snd_task
== task
)
1226 xprt_release_write(xprt
, task
);
1232 xprt
= req
->rq_xprt
;
1233 if (task
->tk_ops
->rpc_count_stats
!= NULL
)
1234 task
->tk_ops
->rpc_count_stats(task
, task
->tk_calldata
);
1235 else if (task
->tk_client
)
1236 rpc_count_iostats(task
, task
->tk_client
->cl_metrics
);
1237 spin_lock_bh(&xprt
->transport_lock
);
1238 xprt
->ops
->release_xprt(xprt
, task
);
1239 if (xprt
->ops
->release_request
)
1240 xprt
->ops
->release_request(task
);
1241 if (!list_empty(&req
->rq_list
))
1242 list_del(&req
->rq_list
);
1243 xprt
->last_used
= jiffies
;
1244 if (list_empty(&xprt
->recv
) && xprt_has_timer(xprt
))
1245 mod_timer(&xprt
->timer
,
1246 xprt
->last_used
+ xprt
->idle_timeout
);
1247 spin_unlock_bh(&xprt
->transport_lock
);
1249 xprt
->ops
->buf_free(req
->rq_buffer
);
1250 if (req
->rq_cred
!= NULL
)
1251 put_rpccred(req
->rq_cred
);
1252 task
->tk_rqstp
= NULL
;
1253 if (req
->rq_release_snd_buf
)
1254 req
->rq_release_snd_buf(req
);
1256 dprintk("RPC: %5u release request %p\n", task
->tk_pid
, req
);
1257 if (likely(!bc_prealloc(req
)))
1258 xprt_free_slot(xprt
, req
);
1260 xprt_free_bc_request(req
);
1263 static void xprt_init(struct rpc_xprt
*xprt
, struct net
*net
)
1265 atomic_set(&xprt
->count
, 1);
1267 spin_lock_init(&xprt
->transport_lock
);
1268 spin_lock_init(&xprt
->reserve_lock
);
1270 INIT_LIST_HEAD(&xprt
->free
);
1271 INIT_LIST_HEAD(&xprt
->recv
);
1272 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1273 spin_lock_init(&xprt
->bc_pa_lock
);
1274 INIT_LIST_HEAD(&xprt
->bc_pa_list
);
1275 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1277 xprt
->last_used
= jiffies
;
1278 xprt
->cwnd
= RPC_INITCWND
;
1279 xprt
->bind_index
= 0;
1281 rpc_init_wait_queue(&xprt
->binding
, "xprt_binding");
1282 rpc_init_wait_queue(&xprt
->pending
, "xprt_pending");
1283 rpc_init_priority_wait_queue(&xprt
->sending
, "xprt_sending");
1284 rpc_init_priority_wait_queue(&xprt
->backlog
, "xprt_backlog");
1286 xprt_init_xid(xprt
);
1288 xprt
->xprt_net
= get_net(net
);
1292 * xprt_create_transport - create an RPC transport
1293 * @args: rpc transport creation arguments
1296 struct rpc_xprt
*xprt_create_transport(struct xprt_create
*args
)
1298 struct rpc_xprt
*xprt
;
1299 struct xprt_class
*t
;
1301 spin_lock(&xprt_list_lock
);
1302 list_for_each_entry(t
, &xprt_list
, list
) {
1303 if (t
->ident
== args
->ident
) {
1304 spin_unlock(&xprt_list_lock
);
1308 spin_unlock(&xprt_list_lock
);
1309 dprintk("RPC: transport (%d) not supported\n", args
->ident
);
1310 return ERR_PTR(-EIO
);
1313 xprt
= t
->setup(args
);
1315 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1319 if (args
->flags
& XPRT_CREATE_NO_IDLE_TIMEOUT
)
1320 xprt
->idle_timeout
= 0;
1321 INIT_WORK(&xprt
->task_cleanup
, xprt_autoclose
);
1322 if (xprt_has_timer(xprt
))
1323 setup_timer(&xprt
->timer
, xprt_init_autodisconnect
,
1324 (unsigned long)xprt
);
1326 init_timer(&xprt
->timer
);
1328 if (strlen(args
->servername
) > RPC_MAXNETNAMELEN
) {
1330 return ERR_PTR(-EINVAL
);
1332 xprt
->servername
= kstrdup(args
->servername
, GFP_KERNEL
);
1333 if (xprt
->servername
== NULL
) {
1335 return ERR_PTR(-ENOMEM
);
1338 dprintk("RPC: created transport %p with %u slots\n", xprt
,
1345 * xprt_destroy - destroy an RPC transport, killing off all requests.
1346 * @xprt: transport to destroy
1349 static void xprt_destroy(struct rpc_xprt
*xprt
)
1351 dprintk("RPC: destroying transport %p\n", xprt
);
1352 del_timer_sync(&xprt
->timer
);
1354 rpc_destroy_wait_queue(&xprt
->binding
);
1355 rpc_destroy_wait_queue(&xprt
->pending
);
1356 rpc_destroy_wait_queue(&xprt
->sending
);
1357 rpc_destroy_wait_queue(&xprt
->backlog
);
1358 cancel_work_sync(&xprt
->task_cleanup
);
1359 kfree(xprt
->servername
);
1361 * Tear down transport state and free the rpc_xprt
1363 xprt
->ops
->destroy(xprt
);
1367 * xprt_put - release a reference to an RPC transport.
1368 * @xprt: pointer to the transport
1371 void xprt_put(struct rpc_xprt
*xprt
)
1373 if (atomic_dec_and_test(&xprt
->count
))