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 *);
70 static DEFINE_SPINLOCK(xprt_list_lock);
71 static LIST_HEAD(xprt_list);
74 * The transport code maintains an estimate on the maximum number of out-
75 * standing RPC requests, using a smoothed version of the congestion
76 * avoidance implemented in 44BSD. This is basically the Van Jacobson
77 * congestion algorithm: If a retransmit occurs, the congestion window is
78 * halved; otherwise, it is incremented by 1/cwnd when
80 * - a reply is received and
81 * - a full number of requests are outstanding and
82 * - the congestion window hasn't been updated recently.
84 #define RPC_CWNDSHIFT (8U)
85 #define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT)
86 #define RPC_INITCWND RPC_CWNDSCALE
87 #define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT)
89 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
92 * xprt_register_transport - register a transport implementation
93 * @transport: transport to register
95 * If a transport implementation is loaded as a kernel module, it can
96 * call this interface to make itself known to the RPC client.
99 * 0: transport successfully registered
100 * -EEXIST: transport already registered
101 * -EINVAL: transport module being unloaded
103 int xprt_register_transport(struct xprt_class *transport)
105 struct xprt_class *t;
109 spin_lock(&xprt_list_lock);
110 list_for_each_entry(t, &xprt_list, list) {
111 /* don't register the same transport class twice */
112 if (t->ident == transport->ident)
116 list_add_tail(&transport->list, &xprt_list);
117 printk(KERN_INFO "RPC: Registered %s transport module.\n",
122 spin_unlock(&xprt_list_lock);
125 EXPORT_SYMBOL_GPL(xprt_register_transport);
128 * xprt_unregister_transport - unregister a transport implementation
129 * @transport: transport to unregister
132 * 0: transport successfully unregistered
133 * -ENOENT: transport never registered
135 int xprt_unregister_transport(struct xprt_class *transport)
137 struct xprt_class *t;
141 spin_lock(&xprt_list_lock);
142 list_for_each_entry(t, &xprt_list, list) {
143 if (t == transport) {
145 "RPC: Unregistered %s transport module.\n",
147 list_del_init(&transport->list);
154 spin_unlock(&xprt_list_lock);
157 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
160 * xprt_load_transport - load a transport implementation
161 * @transport_name: transport to load
164 * 0: transport successfully loaded
165 * -ENOENT: transport module not available
167 int xprt_load_transport(const char *transport_name)
169 struct xprt_class *t;
173 spin_lock(&xprt_list_lock);
174 list_for_each_entry(t, &xprt_list, list) {
175 if (strcmp(t->name, transport_name) == 0) {
176 spin_unlock(&xprt_list_lock);
180 spin_unlock(&xprt_list_lock);
181 result = request_module("xprt%s", transport_name);
185 EXPORT_SYMBOL_GPL(xprt_load_transport);
188 * xprt_reserve_xprt - serialize write access to transports
189 * @task: task that is requesting access to the transport
190 * @xprt: pointer to the target transport
192 * This prevents mixing the payload of separate requests, and prevents
193 * transport connects from colliding with writes. No congestion control
196 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
198 struct rpc_rqst *req = task->tk_rqstp;
201 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
202 if (task == xprt->snd_task)
206 xprt->snd_task = task;
208 req->rq_bytes_sent = 0;
215 dprintk("RPC: %5u failed to lock transport %p\n",
217 task->tk_timeout = 0;
218 task->tk_status = -EAGAIN;
220 priority = RPC_PRIORITY_LOW;
221 else if (!req->rq_ntrans)
222 priority = RPC_PRIORITY_NORMAL;
224 priority = RPC_PRIORITY_HIGH;
225 rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
228 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
230 static void xprt_clear_locked(struct rpc_xprt *xprt)
232 xprt->snd_task = NULL;
233 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state) || xprt->shutdown) {
234 smp_mb__before_clear_bit();
235 clear_bit(XPRT_LOCKED, &xprt->state);
236 smp_mb__after_clear_bit();
238 queue_work(rpciod_workqueue, &xprt->task_cleanup);
242 * xprt_reserve_xprt_cong - serialize write access to transports
243 * @task: task that is requesting access to the transport
245 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
246 * integrated into the decision of whether a request is allowed to be
247 * woken up and given access to the transport.
249 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
251 struct rpc_rqst *req = task->tk_rqstp;
254 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
255 if (task == xprt->snd_task)
260 xprt->snd_task = task;
263 if (__xprt_get_cong(xprt, task)) {
264 xprt->snd_task = task;
265 req->rq_bytes_sent = 0;
269 xprt_clear_locked(xprt);
271 dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
272 task->tk_timeout = 0;
273 task->tk_status = -EAGAIN;
275 priority = RPC_PRIORITY_LOW;
276 else if (!req->rq_ntrans)
277 priority = RPC_PRIORITY_NORMAL;
279 priority = RPC_PRIORITY_HIGH;
280 rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
283 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
285 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
289 spin_lock_bh(&xprt->transport_lock);
290 retval = xprt->ops->reserve_xprt(xprt, task);
291 spin_unlock_bh(&xprt->transport_lock);
295 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
297 struct rpc_task *task;
298 struct rpc_rqst *req;
300 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
303 task = rpc_wake_up_next(&xprt->sending);
307 req = task->tk_rqstp;
308 xprt->snd_task = task;
310 req->rq_bytes_sent = 0;
316 xprt_clear_locked(xprt);
319 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
321 struct rpc_task *task;
322 struct rpc_rqst *req;
324 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
326 if (RPCXPRT_CONGESTED(xprt))
328 task = rpc_wake_up_next(&xprt->sending);
332 req = task->tk_rqstp;
334 xprt->snd_task = task;
337 if (__xprt_get_cong(xprt, task)) {
338 xprt->snd_task = task;
339 req->rq_bytes_sent = 0;
344 xprt_clear_locked(xprt);
348 * xprt_release_xprt - allow other requests to use a transport
349 * @xprt: transport with other tasks potentially waiting
350 * @task: task that is releasing access to the transport
352 * Note that "task" can be NULL. No congestion control is provided.
354 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
356 if (xprt->snd_task == task) {
357 xprt_clear_locked(xprt);
358 __xprt_lock_write_next(xprt);
361 EXPORT_SYMBOL_GPL(xprt_release_xprt);
364 * xprt_release_xprt_cong - allow other requests to use a transport
365 * @xprt: transport with other tasks potentially waiting
366 * @task: task that is releasing access to the transport
368 * Note that "task" can be NULL. Another task is awoken to use the
369 * transport if the transport's congestion window allows it.
371 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
373 if (xprt->snd_task == task) {
374 xprt_clear_locked(xprt);
375 __xprt_lock_write_next_cong(xprt);
378 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
380 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
382 spin_lock_bh(&xprt->transport_lock);
383 xprt->ops->release_xprt(xprt, task);
384 spin_unlock_bh(&xprt->transport_lock);
388 * Van Jacobson congestion avoidance. Check if the congestion window
389 * overflowed. Put the task to sleep if this is the case.
392 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
394 struct rpc_rqst *req = task->tk_rqstp;
398 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
399 task->tk_pid, xprt->cong, xprt->cwnd);
400 if (RPCXPRT_CONGESTED(xprt))
403 xprt->cong += RPC_CWNDSCALE;
408 * Adjust the congestion window, and wake up the next task
409 * that has been sleeping due to congestion
412 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
417 xprt->cong -= RPC_CWNDSCALE;
418 __xprt_lock_write_next_cong(xprt);
422 * xprt_release_rqst_cong - housekeeping when request is complete
423 * @task: RPC request that recently completed
425 * Useful for transports that require congestion control.
427 void xprt_release_rqst_cong(struct rpc_task *task)
429 __xprt_put_cong(task->tk_xprt, task->tk_rqstp);
431 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
434 * xprt_adjust_cwnd - adjust transport congestion window
435 * @task: recently completed RPC request used to adjust window
436 * @result: result code of completed RPC request
438 * We use a time-smoothed congestion estimator to avoid heavy oscillation.
440 void xprt_adjust_cwnd(struct rpc_task *task, int result)
442 struct rpc_rqst *req = task->tk_rqstp;
443 struct rpc_xprt *xprt = task->tk_xprt;
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 if (unlikely(xprt->shutdown))
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) || xprt->shutdown)
684 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
686 spin_unlock(&xprt->transport_lock);
687 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
688 queue_work(rpciod_workqueue, &xprt->task_cleanup);
691 spin_unlock(&xprt->transport_lock);
695 * xprt_connect - schedule a transport connect operation
696 * @task: RPC task that is requesting the connect
699 void xprt_connect(struct rpc_task *task)
701 struct rpc_xprt *xprt = task->tk_xprt;
703 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
704 xprt, (xprt_connected(xprt) ? "is" : "is not"));
706 if (!xprt_bound(xprt)) {
707 task->tk_status = -EAGAIN;
710 if (!xprt_lock_write(xprt, task))
713 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
714 xprt->ops->close(xprt);
716 if (xprt_connected(xprt))
717 xprt_release_write(xprt, task);
720 task->tk_rqstp->rq_bytes_sent = 0;
722 task->tk_timeout = task->tk_rqstp->rq_timeout;
723 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
725 if (test_bit(XPRT_CLOSING, &xprt->state))
727 if (xprt_test_and_set_connecting(xprt))
729 xprt->stat.connect_start = jiffies;
730 xprt->ops->connect(task);
734 static void xprt_connect_status(struct rpc_task *task)
736 struct rpc_xprt *xprt = task->tk_xprt;
738 if (task->tk_status == 0) {
739 xprt->stat.connect_count++;
740 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
741 dprintk("RPC: %5u xprt_connect_status: connection established\n",
746 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,
757 task->tk_client->cl_server);
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 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(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 int 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_reply_bytes_recvd && !req->rq_bytes_sent) {
866 err = req->rq_reply_bytes_recvd;
869 if (!xprt->ops->reserve_xprt(xprt, task))
872 spin_unlock_bh(&xprt->transport_lock);
876 void xprt_end_transmit(struct rpc_task *task)
878 xprt_release_write(task->tk_rqstp->rq_xprt, task);
882 * xprt_transmit - send an RPC request on a transport
883 * @task: controlling RPC task
885 * We have to copy the iovec because sendmsg fiddles with its contents.
887 void xprt_transmit(struct rpc_task *task)
889 struct rpc_rqst *req = task->tk_rqstp;
890 struct rpc_xprt *xprt = req->rq_xprt;
893 dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
895 if (!req->rq_reply_bytes_recvd) {
896 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
898 * Add to the list only if we're expecting a reply
900 spin_lock_bh(&xprt->transport_lock);
901 /* Update the softirq receive buffer */
902 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
903 sizeof(req->rq_private_buf));
904 /* Add request to the receive list */
905 list_add_tail(&req->rq_list, &xprt->recv);
906 spin_unlock_bh(&xprt->transport_lock);
907 xprt_reset_majortimeo(req);
908 /* Turn off autodisconnect */
909 del_singleshot_timer_sync(&xprt->timer);
911 } else if (!req->rq_bytes_sent)
914 req->rq_connect_cookie = xprt->connect_cookie;
915 req->rq_xtime = ktime_get();
916 status = xprt->ops->send_request(task);
918 task->tk_status = status;
922 dprintk("RPC: %5u xmit complete\n", task->tk_pid);
923 task->tk_flags |= RPC_TASK_SENT;
924 spin_lock_bh(&xprt->transport_lock);
926 xprt->ops->set_retrans_timeout(task);
929 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
930 xprt->stat.bklog_u += xprt->backlog.qlen;
932 /* Don't race with disconnect */
933 if (!xprt_connected(xprt))
934 task->tk_status = -ENOTCONN;
935 else if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task)) {
937 * Sleep on the pending queue since
938 * we're expecting a reply.
940 rpc_sleep_on(&xprt->pending, task, xprt_timer);
942 spin_unlock_bh(&xprt->transport_lock);
945 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
947 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
949 if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
951 req = kzalloc(sizeof(struct rpc_rqst), gfp_flags);
954 atomic_dec(&xprt->num_reqs);
955 req = ERR_PTR(-ENOMEM);
960 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
962 if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
969 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
971 struct rpc_rqst *req;
973 spin_lock(&xprt->reserve_lock);
974 if (!list_empty(&xprt->free)) {
975 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
976 list_del(&req->rq_list);
979 req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT);
982 switch (PTR_ERR(req)) {
984 dprintk("RPC: dynamic allocation of request slot "
985 "failed! Retrying\n");
986 task->tk_status = -ENOMEM;
989 rpc_sleep_on(&xprt->backlog, task, NULL);
990 dprintk("RPC: waiting for request slot\n");
992 task->tk_status = -EAGAIN;
994 spin_unlock(&xprt->reserve_lock);
998 task->tk_rqstp = req;
999 xprt_request_init(task, xprt);
1000 spin_unlock(&xprt->reserve_lock);
1002 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1004 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1006 /* Note: grabbing the xprt_lock_write() ensures that we throttle
1007 * new slot allocation if the transport is congested (i.e. when
1008 * reconnecting a stream transport or when out of socket write
1011 if (xprt_lock_write(xprt, task)) {
1012 xprt_alloc_slot(xprt, task);
1013 xprt_release_write(xprt, task);
1016 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1018 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1020 spin_lock(&xprt->reserve_lock);
1021 if (!xprt_dynamic_free_slot(xprt, req)) {
1022 memset(req, 0, sizeof(*req)); /* mark unused */
1023 list_add(&req->rq_list, &xprt->free);
1025 rpc_wake_up_next(&xprt->backlog);
1026 spin_unlock(&xprt->reserve_lock);
1029 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1031 struct rpc_rqst *req;
1032 while (!list_empty(&xprt->free)) {
1033 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1034 list_del(&req->rq_list);
1039 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1040 unsigned int num_prealloc,
1041 unsigned int max_alloc)
1043 struct rpc_xprt *xprt;
1044 struct rpc_rqst *req;
1047 xprt = kzalloc(size, GFP_KERNEL);
1051 xprt_init(xprt, net);
1053 for (i = 0; i < num_prealloc; i++) {
1054 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1057 list_add(&req->rq_list, &xprt->free);
1059 if (i < num_prealloc)
1061 if (max_alloc > num_prealloc)
1062 xprt->max_reqs = max_alloc;
1064 xprt->max_reqs = num_prealloc;
1065 xprt->min_reqs = num_prealloc;
1066 atomic_set(&xprt->num_reqs, num_prealloc);
1075 EXPORT_SYMBOL_GPL(xprt_alloc);
1077 void xprt_free(struct rpc_xprt *xprt)
1079 put_net(xprt->xprt_net);
1080 xprt_free_all_slots(xprt);
1083 EXPORT_SYMBOL_GPL(xprt_free);
1086 * xprt_reserve - allocate an RPC request slot
1087 * @task: RPC task requesting a slot allocation
1089 * If no more slots are available, place the task on the transport's
1092 void xprt_reserve(struct rpc_task *task)
1094 struct rpc_xprt *xprt = task->tk_xprt;
1096 task->tk_status = 0;
1097 if (task->tk_rqstp != NULL)
1100 task->tk_timeout = 0;
1101 task->tk_status = -EAGAIN;
1102 xprt->ops->alloc_slot(xprt, task);
1105 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1107 return (__force __be32)xprt->xid++;
1110 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1112 xprt->xid = net_random();
1115 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1117 struct rpc_rqst *req = task->tk_rqstp;
1119 INIT_LIST_HEAD(&req->rq_list);
1120 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1121 req->rq_task = task;
1122 req->rq_xprt = xprt;
1123 req->rq_buffer = NULL;
1124 req->rq_xid = xprt_alloc_xid(xprt);
1125 req->rq_release_snd_buf = NULL;
1126 xprt_reset_majortimeo(req);
1127 dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1128 req, ntohl(req->rq_xid));
1132 * xprt_release - release an RPC request slot
1133 * @task: task which is finished with the slot
1136 void xprt_release(struct rpc_task *task)
1138 struct rpc_xprt *xprt;
1139 struct rpc_rqst *req = task->tk_rqstp;
1142 if (task->tk_client) {
1144 xprt = rcu_dereference(task->tk_client->cl_xprt);
1145 if (xprt->snd_task == task)
1146 xprt_release_write(xprt, task);
1152 xprt = req->rq_xprt;
1153 rpc_count_iostats(task);
1154 spin_lock_bh(&xprt->transport_lock);
1155 xprt->ops->release_xprt(xprt, task);
1156 if (xprt->ops->release_request)
1157 xprt->ops->release_request(task);
1158 if (!list_empty(&req->rq_list))
1159 list_del(&req->rq_list);
1160 xprt->last_used = jiffies;
1161 if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1162 mod_timer(&xprt->timer,
1163 xprt->last_used + xprt->idle_timeout);
1164 spin_unlock_bh(&xprt->transport_lock);
1166 xprt->ops->buf_free(req->rq_buffer);
1167 if (req->rq_cred != NULL)
1168 put_rpccred(req->rq_cred);
1169 task->tk_rqstp = NULL;
1170 if (req->rq_release_snd_buf)
1171 req->rq_release_snd_buf(req);
1173 dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1174 if (likely(!bc_prealloc(req)))
1175 xprt_free_slot(xprt, req);
1177 xprt_free_bc_request(req);
1180 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1182 atomic_set(&xprt->count, 1);
1184 spin_lock_init(&xprt->transport_lock);
1185 spin_lock_init(&xprt->reserve_lock);
1187 INIT_LIST_HEAD(&xprt->free);
1188 INIT_LIST_HEAD(&xprt->recv);
1189 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1190 spin_lock_init(&xprt->bc_pa_lock);
1191 INIT_LIST_HEAD(&xprt->bc_pa_list);
1192 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1194 xprt->last_used = jiffies;
1195 xprt->cwnd = RPC_INITCWND;
1196 xprt->bind_index = 0;
1198 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1199 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1200 rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1201 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1203 xprt_init_xid(xprt);
1205 xprt->xprt_net = get_net(net);
1209 * xprt_create_transport - create an RPC transport
1210 * @args: rpc transport creation arguments
1213 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1215 struct rpc_xprt *xprt;
1216 struct xprt_class *t;
1218 spin_lock(&xprt_list_lock);
1219 list_for_each_entry(t, &xprt_list, list) {
1220 if (t->ident == args->ident) {
1221 spin_unlock(&xprt_list_lock);
1225 spin_unlock(&xprt_list_lock);
1226 printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1227 return ERR_PTR(-EIO);
1230 xprt = t->setup(args);
1232 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1236 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1237 if (xprt_has_timer(xprt))
1238 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1239 (unsigned long)xprt);
1241 init_timer(&xprt->timer);
1242 dprintk("RPC: created transport %p with %u slots\n", xprt,
1249 * xprt_destroy - destroy an RPC transport, killing off all requests.
1250 * @xprt: transport to destroy
1253 static void xprt_destroy(struct rpc_xprt *xprt)
1255 dprintk("RPC: destroying transport %p\n", xprt);
1257 del_timer_sync(&xprt->timer);
1259 rpc_destroy_wait_queue(&xprt->binding);
1260 rpc_destroy_wait_queue(&xprt->pending);
1261 rpc_destroy_wait_queue(&xprt->sending);
1262 rpc_destroy_wait_queue(&xprt->backlog);
1263 cancel_work_sync(&xprt->task_cleanup);
1265 * Tear down transport state and free the rpc_xprt
1267 xprt->ops->destroy(xprt);
1271 * xprt_put - release a reference to an RPC transport.
1272 * @xprt: pointer to the transport
1275 void xprt_put(struct rpc_xprt *xprt)
1277 if (atomic_dec_and_test(&xprt->count))
1282 * xprt_get - return a reference to an RPC transport.
1283 * @xprt: pointer to the transport
1286 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1288 if (atomic_inc_not_zero(&xprt->count))