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_request_init(struct rpc_task *, struct rpc_xprt *);
66 static void xprt_connect_status(struct rpc_task *task);
67 static int __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
69 static DEFINE_SPINLOCK(xprt_list_lock);
70 static LIST_HEAD(xprt_list);
73 * The transport code maintains an estimate on the maximum number of out-
74 * standing RPC requests, using a smoothed version of the congestion
75 * avoidance implemented in 44BSD. This is basically the Van Jacobson
76 * congestion algorithm: If a retransmit occurs, the congestion window is
77 * halved; otherwise, it is incremented by 1/cwnd when
79 * - a reply is received and
80 * - a full number of requests are outstanding and
81 * - the congestion window hasn't been updated recently.
83 #define RPC_CWNDSHIFT (8U)
84 #define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT)
85 #define RPC_INITCWND RPC_CWNDSCALE
86 #define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT)
88 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
91 * xprt_register_transport - register a transport implementation
92 * @transport: transport to register
94 * If a transport implementation is loaded as a kernel module, it can
95 * call this interface to make itself known to the RPC client.
98 * 0: transport successfully registered
99 * -EEXIST: transport already registered
100 * -EINVAL: transport module being unloaded
102 int xprt_register_transport(struct xprt_class *transport)
104 struct xprt_class *t;
108 spin_lock(&xprt_list_lock);
109 list_for_each_entry(t, &xprt_list, list) {
110 /* don't register the same transport class twice */
111 if (t->ident == transport->ident)
115 list_add_tail(&transport->list, &xprt_list);
116 printk(KERN_INFO "RPC: Registered %s transport module.\n",
121 spin_unlock(&xprt_list_lock);
124 EXPORT_SYMBOL_GPL(xprt_register_transport);
127 * xprt_unregister_transport - unregister a transport implementation
128 * @transport: transport to unregister
131 * 0: transport successfully unregistered
132 * -ENOENT: transport never registered
134 int xprt_unregister_transport(struct xprt_class *transport)
136 struct xprt_class *t;
140 spin_lock(&xprt_list_lock);
141 list_for_each_entry(t, &xprt_list, list) {
142 if (t == transport) {
144 "RPC: Unregistered %s transport module.\n",
146 list_del_init(&transport->list);
153 spin_unlock(&xprt_list_lock);
156 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
159 * xprt_load_transport - load a transport implementation
160 * @transport_name: transport to load
163 * 0: transport successfully loaded
164 * -ENOENT: transport module not available
166 int xprt_load_transport(const char *transport_name)
168 struct xprt_class *t;
172 spin_lock(&xprt_list_lock);
173 list_for_each_entry(t, &xprt_list, list) {
174 if (strcmp(t->name, transport_name) == 0) {
175 spin_unlock(&xprt_list_lock);
179 spin_unlock(&xprt_list_lock);
180 result = request_module("xprt%s", transport_name);
184 EXPORT_SYMBOL_GPL(xprt_load_transport);
187 * xprt_reserve_xprt - serialize write access to transports
188 * @task: task that is requesting access to the transport
190 * This prevents mixing the payload of separate requests, and prevents
191 * transport connects from colliding with writes. No congestion control
194 int xprt_reserve_xprt(struct rpc_task *task)
196 struct rpc_rqst *req = task->tk_rqstp;
197 struct rpc_xprt *xprt = req->rq_xprt;
199 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
200 if (task == xprt->snd_task)
204 xprt->snd_task = task;
205 req->rq_bytes_sent = 0;
211 dprintk("RPC: %5u failed to lock transport %p\n",
213 task->tk_timeout = 0;
214 task->tk_status = -EAGAIN;
215 if (req && req->rq_ntrans)
216 rpc_sleep_on(&xprt->resend, task, NULL);
218 rpc_sleep_on(&xprt->sending, task, NULL);
221 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
223 static void xprt_clear_locked(struct rpc_xprt *xprt)
225 xprt->snd_task = NULL;
226 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state) || xprt->shutdown) {
227 smp_mb__before_clear_bit();
228 clear_bit(XPRT_LOCKED, &xprt->state);
229 smp_mb__after_clear_bit();
231 queue_work(rpciod_workqueue, &xprt->task_cleanup);
235 * xprt_reserve_xprt_cong - serialize write access to transports
236 * @task: task that is requesting access to the transport
238 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
239 * integrated into the decision of whether a request is allowed to be
240 * woken up and given access to the transport.
242 int xprt_reserve_xprt_cong(struct rpc_task *task)
244 struct rpc_xprt *xprt = task->tk_xprt;
245 struct rpc_rqst *req = task->tk_rqstp;
247 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
248 if (task == xprt->snd_task)
252 if (__xprt_get_cong(xprt, task)) {
253 xprt->snd_task = task;
255 req->rq_bytes_sent = 0;
260 xprt_clear_locked(xprt);
262 dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
263 task->tk_timeout = 0;
264 task->tk_status = -EAGAIN;
265 if (req && req->rq_ntrans)
266 rpc_sleep_on(&xprt->resend, task, NULL);
268 rpc_sleep_on(&xprt->sending, task, NULL);
271 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
273 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
277 spin_lock_bh(&xprt->transport_lock);
278 retval = xprt->ops->reserve_xprt(task);
279 spin_unlock_bh(&xprt->transport_lock);
283 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
285 struct rpc_task *task;
286 struct rpc_rqst *req;
288 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
291 task = rpc_wake_up_next(&xprt->resend);
293 task = rpc_wake_up_next(&xprt->sending);
298 req = task->tk_rqstp;
299 xprt->snd_task = task;
301 req->rq_bytes_sent = 0;
307 xprt_clear_locked(xprt);
310 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
312 struct rpc_task *task;
314 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
316 if (RPCXPRT_CONGESTED(xprt))
318 task = rpc_wake_up_next(&xprt->resend);
320 task = rpc_wake_up_next(&xprt->sending);
324 if (__xprt_get_cong(xprt, task)) {
325 struct rpc_rqst *req = task->tk_rqstp;
326 xprt->snd_task = task;
328 req->rq_bytes_sent = 0;
334 xprt_clear_locked(xprt);
338 * xprt_release_xprt - allow other requests to use a transport
339 * @xprt: transport with other tasks potentially waiting
340 * @task: task that is releasing access to the transport
342 * Note that "task" can be NULL. No congestion control is provided.
344 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
346 if (xprt->snd_task == task) {
347 xprt_clear_locked(xprt);
348 __xprt_lock_write_next(xprt);
351 EXPORT_SYMBOL_GPL(xprt_release_xprt);
354 * xprt_release_xprt_cong - allow other requests to use a transport
355 * @xprt: transport with other tasks potentially waiting
356 * @task: task that is releasing access to the transport
358 * Note that "task" can be NULL. Another task is awoken to use the
359 * transport if the transport's congestion window allows it.
361 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
363 if (xprt->snd_task == task) {
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 __xprt_put_cong(task->tk_xprt, task->tk_rqstp);
421 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
424 * xprt_adjust_cwnd - adjust transport congestion window
425 * @task: recently completed RPC request used to adjust window
426 * @result: result code of completed RPC request
428 * We use a time-smoothed congestion estimator to avoid heavy oscillation.
430 void xprt_adjust_cwnd(struct rpc_task *task, int result)
432 struct rpc_rqst *req = task->tk_rqstp;
433 struct rpc_xprt *xprt = task->tk_xprt;
434 unsigned long cwnd = xprt->cwnd;
436 if (result >= 0 && cwnd <= xprt->cong) {
437 /* The (cwnd >> 1) term makes sure
438 * the result gets rounded properly. */
439 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
440 if (cwnd > RPC_MAXCWND(xprt))
441 cwnd = RPC_MAXCWND(xprt);
442 __xprt_lock_write_next_cong(xprt);
443 } else if (result == -ETIMEDOUT) {
445 if (cwnd < RPC_CWNDSCALE)
446 cwnd = RPC_CWNDSCALE;
448 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
449 xprt->cong, xprt->cwnd, cwnd);
451 __xprt_put_cong(xprt, req);
453 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
456 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
457 * @xprt: transport with waiting tasks
458 * @status: result code to plant in each task before waking it
461 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
464 rpc_wake_up_status(&xprt->pending, status);
466 rpc_wake_up(&xprt->pending);
468 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
471 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
472 * @task: task to be put to sleep
473 * @action: function pointer to be executed after wait
475 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
477 struct rpc_rqst *req = task->tk_rqstp;
478 struct rpc_xprt *xprt = req->rq_xprt;
480 task->tk_timeout = req->rq_timeout;
481 rpc_sleep_on(&xprt->pending, task, action);
483 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
486 * xprt_write_space - wake the task waiting for transport output buffer space
487 * @xprt: transport with waiting tasks
489 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
491 void xprt_write_space(struct rpc_xprt *xprt)
493 if (unlikely(xprt->shutdown))
496 spin_lock_bh(&xprt->transport_lock);
497 if (xprt->snd_task) {
498 dprintk("RPC: write space: waking waiting task on "
500 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
502 spin_unlock_bh(&xprt->transport_lock);
504 EXPORT_SYMBOL_GPL(xprt_write_space);
507 * xprt_set_retrans_timeout_def - set a request's retransmit timeout
508 * @task: task whose timeout is to be set
510 * Set a request's retransmit timeout based on the transport's
511 * default timeout parameters. Used by transports that don't adjust
512 * the retransmit timeout based on round-trip time estimation.
514 void xprt_set_retrans_timeout_def(struct rpc_task *task)
516 task->tk_timeout = task->tk_rqstp->rq_timeout;
518 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
521 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
522 * @task: task whose timeout is to be set
524 * Set a request's retransmit timeout using the RTT estimator.
526 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
528 int timer = task->tk_msg.rpc_proc->p_timer;
529 struct rpc_clnt *clnt = task->tk_client;
530 struct rpc_rtt *rtt = clnt->cl_rtt;
531 struct rpc_rqst *req = task->tk_rqstp;
532 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
534 task->tk_timeout = rpc_calc_rto(rtt, timer);
535 task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
536 if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
537 task->tk_timeout = max_timeout;
539 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
541 static void xprt_reset_majortimeo(struct rpc_rqst *req)
543 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
545 req->rq_majortimeo = req->rq_timeout;
546 if (to->to_exponential)
547 req->rq_majortimeo <<= to->to_retries;
549 req->rq_majortimeo += to->to_increment * to->to_retries;
550 if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
551 req->rq_majortimeo = to->to_maxval;
552 req->rq_majortimeo += jiffies;
556 * xprt_adjust_timeout - adjust timeout values for next retransmit
557 * @req: RPC request containing parameters to use for the adjustment
560 int xprt_adjust_timeout(struct rpc_rqst *req)
562 struct rpc_xprt *xprt = req->rq_xprt;
563 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
566 if (time_before(jiffies, req->rq_majortimeo)) {
567 if (to->to_exponential)
568 req->rq_timeout <<= 1;
570 req->rq_timeout += to->to_increment;
571 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
572 req->rq_timeout = to->to_maxval;
575 req->rq_timeout = to->to_initval;
577 xprt_reset_majortimeo(req);
578 /* Reset the RTT counters == "slow start" */
579 spin_lock_bh(&xprt->transport_lock);
580 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
581 spin_unlock_bh(&xprt->transport_lock);
585 if (req->rq_timeout == 0) {
586 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
587 req->rq_timeout = 5 * HZ;
592 static void xprt_autoclose(struct work_struct *work)
594 struct rpc_xprt *xprt =
595 container_of(work, struct rpc_xprt, task_cleanup);
597 xprt->ops->close(xprt);
598 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
599 xprt_release_write(xprt, NULL);
603 * xprt_disconnect_done - mark a transport as disconnected
604 * @xprt: transport to flag for disconnect
607 void xprt_disconnect_done(struct rpc_xprt *xprt)
609 dprintk("RPC: disconnected transport %p\n", xprt);
610 spin_lock_bh(&xprt->transport_lock);
611 xprt_clear_connected(xprt);
612 xprt_wake_pending_tasks(xprt, -EAGAIN);
613 spin_unlock_bh(&xprt->transport_lock);
615 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
618 * xprt_force_disconnect - force a transport to disconnect
619 * @xprt: transport to disconnect
622 void xprt_force_disconnect(struct rpc_xprt *xprt)
624 /* Don't race with the test_bit() in xprt_clear_locked() */
625 spin_lock_bh(&xprt->transport_lock);
626 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
627 /* Try to schedule an autoclose RPC call */
628 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
629 queue_work(rpciod_workqueue, &xprt->task_cleanup);
630 xprt_wake_pending_tasks(xprt, -EAGAIN);
631 spin_unlock_bh(&xprt->transport_lock);
635 * xprt_conditional_disconnect - force a transport to disconnect
636 * @xprt: transport to disconnect
637 * @cookie: 'connection cookie'
639 * This attempts to break the connection if and only if 'cookie' matches
640 * the current transport 'connection cookie'. It ensures that we don't
641 * try to break the connection more than once when we need to retransmit
642 * a batch of RPC requests.
645 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
647 /* Don't race with the test_bit() in xprt_clear_locked() */
648 spin_lock_bh(&xprt->transport_lock);
649 if (cookie != xprt->connect_cookie)
651 if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
653 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
654 /* Try to schedule an autoclose RPC call */
655 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
656 queue_work(rpciod_workqueue, &xprt->task_cleanup);
657 xprt_wake_pending_tasks(xprt, -EAGAIN);
659 spin_unlock_bh(&xprt->transport_lock);
663 xprt_init_autodisconnect(unsigned long data)
665 struct rpc_xprt *xprt = (struct rpc_xprt *)data;
667 spin_lock(&xprt->transport_lock);
668 if (!list_empty(&xprt->recv) || xprt->shutdown)
670 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
672 spin_unlock(&xprt->transport_lock);
673 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
674 queue_work(rpciod_workqueue, &xprt->task_cleanup);
677 spin_unlock(&xprt->transport_lock);
681 * xprt_connect - schedule a transport connect operation
682 * @task: RPC task that is requesting the connect
685 void xprt_connect(struct rpc_task *task)
687 struct rpc_xprt *xprt = task->tk_xprt;
689 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
690 xprt, (xprt_connected(xprt) ? "is" : "is not"));
692 if (!xprt_bound(xprt)) {
693 task->tk_status = -EAGAIN;
696 if (!xprt_lock_write(xprt, task))
699 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
700 xprt->ops->close(xprt);
702 if (xprt_connected(xprt))
703 xprt_release_write(xprt, task);
706 task->tk_rqstp->rq_bytes_sent = 0;
708 task->tk_timeout = task->tk_rqstp->rq_timeout;
709 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
711 if (test_bit(XPRT_CLOSING, &xprt->state))
713 if (xprt_test_and_set_connecting(xprt))
715 xprt->stat.connect_start = jiffies;
716 xprt->ops->connect(task);
720 static void xprt_connect_status(struct rpc_task *task)
722 struct rpc_xprt *xprt = task->tk_xprt;
724 if (task->tk_status == 0) {
725 xprt->stat.connect_count++;
726 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
727 dprintk("RPC: %5u xprt_connect_status: connection established\n",
732 switch (task->tk_status) {
734 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
737 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
738 "out\n", task->tk_pid);
741 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
742 "server %s\n", task->tk_pid, -task->tk_status,
743 task->tk_client->cl_server);
744 xprt_release_write(xprt, task);
745 task->tk_status = -EIO;
750 * xprt_lookup_rqst - find an RPC request corresponding to an XID
751 * @xprt: transport on which the original request was transmitted
752 * @xid: RPC XID of incoming reply
755 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
757 struct rpc_rqst *entry;
759 list_for_each_entry(entry, &xprt->recv, rq_list)
760 if (entry->rq_xid == xid)
763 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
765 xprt->stat.bad_xids++;
768 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
770 static void xprt_update_rtt(struct rpc_task *task)
772 struct rpc_rqst *req = task->tk_rqstp;
773 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
774 unsigned timer = task->tk_msg.rpc_proc->p_timer;
775 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
778 if (req->rq_ntrans == 1)
779 rpc_update_rtt(rtt, timer, m);
780 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
785 * xprt_complete_rqst - called when reply processing is complete
786 * @task: RPC request that recently completed
787 * @copied: actual number of bytes received from the transport
789 * Caller holds transport lock.
791 void xprt_complete_rqst(struct rpc_task *task, int copied)
793 struct rpc_rqst *req = task->tk_rqstp;
794 struct rpc_xprt *xprt = req->rq_xprt;
796 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
797 task->tk_pid, ntohl(req->rq_xid), copied);
800 req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
801 if (xprt->ops->timer != NULL)
802 xprt_update_rtt(task);
804 list_del_init(&req->rq_list);
805 req->rq_private_buf.len = copied;
806 /* Ensure all writes are done before we update */
807 /* req->rq_reply_bytes_recvd */
809 req->rq_reply_bytes_recvd = copied;
810 rpc_wake_up_queued_task(&xprt->pending, task);
812 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
814 static void xprt_timer(struct rpc_task *task)
816 struct rpc_rqst *req = task->tk_rqstp;
817 struct rpc_xprt *xprt = req->rq_xprt;
819 if (task->tk_status != -ETIMEDOUT)
821 dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
823 spin_lock_bh(&xprt->transport_lock);
824 if (!req->rq_reply_bytes_recvd) {
825 if (xprt->ops->timer)
826 xprt->ops->timer(task);
829 spin_unlock_bh(&xprt->transport_lock);
832 static inline int xprt_has_timer(struct rpc_xprt *xprt)
834 return xprt->idle_timeout != 0;
838 * xprt_prepare_transmit - reserve the transport before sending a request
839 * @task: RPC task about to send a request
842 int xprt_prepare_transmit(struct rpc_task *task)
844 struct rpc_rqst *req = task->tk_rqstp;
845 struct rpc_xprt *xprt = req->rq_xprt;
848 dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
850 spin_lock_bh(&xprt->transport_lock);
851 if (req->rq_reply_bytes_recvd && !req->rq_bytes_sent) {
852 err = req->rq_reply_bytes_recvd;
855 if (!xprt->ops->reserve_xprt(task))
858 spin_unlock_bh(&xprt->transport_lock);
862 void xprt_end_transmit(struct rpc_task *task)
864 xprt_release_write(task->tk_rqstp->rq_xprt, task);
868 * xprt_transmit - send an RPC request on a transport
869 * @task: controlling RPC task
871 * We have to copy the iovec because sendmsg fiddles with its contents.
873 void xprt_transmit(struct rpc_task *task)
875 struct rpc_rqst *req = task->tk_rqstp;
876 struct rpc_xprt *xprt = req->rq_xprt;
879 dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
881 if (!req->rq_reply_bytes_recvd) {
882 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
884 * Add to the list only if we're expecting a reply
886 spin_lock_bh(&xprt->transport_lock);
887 /* Update the softirq receive buffer */
888 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
889 sizeof(req->rq_private_buf));
890 /* Add request to the receive list */
891 list_add_tail(&req->rq_list, &xprt->recv);
892 spin_unlock_bh(&xprt->transport_lock);
893 xprt_reset_majortimeo(req);
894 /* Turn off autodisconnect */
895 del_singleshot_timer_sync(&xprt->timer);
897 } else if (!req->rq_bytes_sent)
900 req->rq_connect_cookie = xprt->connect_cookie;
901 req->rq_xtime = ktime_get();
902 status = xprt->ops->send_request(task);
904 task->tk_status = status;
908 dprintk("RPC: %5u xmit complete\n", task->tk_pid);
909 spin_lock_bh(&xprt->transport_lock);
911 xprt->ops->set_retrans_timeout(task);
914 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
915 xprt->stat.bklog_u += xprt->backlog.qlen;
917 /* Don't race with disconnect */
918 if (!xprt_connected(xprt))
919 task->tk_status = -ENOTCONN;
920 else if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task)) {
922 * Sleep on the pending queue since
923 * we're expecting a reply.
925 rpc_sleep_on(&xprt->pending, task, xprt_timer);
927 spin_unlock_bh(&xprt->transport_lock);
930 static void xprt_alloc_slot(struct rpc_task *task)
932 struct rpc_xprt *xprt = task->tk_xprt;
937 if (!list_empty(&xprt->free)) {
938 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
939 list_del_init(&req->rq_list);
940 task->tk_rqstp = req;
941 xprt_request_init(task, xprt);
944 dprintk("RPC: waiting for request slot\n");
945 task->tk_status = -EAGAIN;
946 task->tk_timeout = 0;
947 rpc_sleep_on(&xprt->backlog, task, NULL);
950 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
952 memset(req, 0, sizeof(*req)); /* mark unused */
954 spin_lock(&xprt->reserve_lock);
955 list_add(&req->rq_list, &xprt->free);
956 rpc_wake_up_next(&xprt->backlog);
957 spin_unlock(&xprt->reserve_lock);
960 struct rpc_xprt *xprt_alloc(struct net *net, int size, int max_req)
962 struct rpc_xprt *xprt;
964 xprt = kzalloc(size, GFP_KERNEL);
967 kref_init(&xprt->kref);
969 xprt->max_reqs = max_req;
970 xprt->slot = kcalloc(max_req, sizeof(struct rpc_rqst), GFP_KERNEL);
971 if (xprt->slot == NULL)
974 xprt->xprt_net = get_net(net);
982 EXPORT_SYMBOL_GPL(xprt_alloc);
984 void xprt_free(struct rpc_xprt *xprt)
986 put_net(xprt->xprt_net);
990 EXPORT_SYMBOL_GPL(xprt_free);
993 * xprt_reserve - allocate an RPC request slot
994 * @task: RPC task requesting a slot allocation
996 * If no more slots are available, place the task on the transport's
999 void xprt_reserve(struct rpc_task *task)
1001 struct rpc_xprt *xprt = task->tk_xprt;
1003 task->tk_status = -EIO;
1004 spin_lock(&xprt->reserve_lock);
1005 xprt_alloc_slot(task);
1006 spin_unlock(&xprt->reserve_lock);
1009 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1011 return (__force __be32)xprt->xid++;
1014 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1016 xprt->xid = net_random();
1019 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1021 struct rpc_rqst *req = task->tk_rqstp;
1023 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1024 req->rq_task = task;
1025 req->rq_xprt = xprt;
1026 req->rq_buffer = NULL;
1027 req->rq_xid = xprt_alloc_xid(xprt);
1028 req->rq_release_snd_buf = NULL;
1029 xprt_reset_majortimeo(req);
1030 dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1031 req, ntohl(req->rq_xid));
1035 * xprt_release - release an RPC request slot
1036 * @task: task which is finished with the slot
1039 void xprt_release(struct rpc_task *task)
1041 struct rpc_xprt *xprt;
1042 struct rpc_rqst *req;
1044 if (!(req = task->tk_rqstp))
1047 xprt = req->rq_xprt;
1048 rpc_count_iostats(task);
1049 spin_lock_bh(&xprt->transport_lock);
1050 xprt->ops->release_xprt(xprt, task);
1051 if (xprt->ops->release_request)
1052 xprt->ops->release_request(task);
1053 if (!list_empty(&req->rq_list))
1054 list_del(&req->rq_list);
1055 xprt->last_used = jiffies;
1056 if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1057 mod_timer(&xprt->timer,
1058 xprt->last_used + xprt->idle_timeout);
1059 spin_unlock_bh(&xprt->transport_lock);
1061 xprt->ops->buf_free(req->rq_buffer);
1062 if (req->rq_cred != NULL)
1063 put_rpccred(req->rq_cred);
1064 task->tk_rqstp = NULL;
1065 if (req->rq_release_snd_buf)
1066 req->rq_release_snd_buf(req);
1068 dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1069 if (likely(!bc_prealloc(req)))
1070 xprt_free_slot(xprt, req);
1072 xprt_free_bc_request(req);
1076 * xprt_create_transport - create an RPC transport
1077 * @args: rpc transport creation arguments
1080 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1082 struct rpc_xprt *xprt;
1083 struct rpc_rqst *req;
1084 struct xprt_class *t;
1086 spin_lock(&xprt_list_lock);
1087 list_for_each_entry(t, &xprt_list, list) {
1088 if (t->ident == args->ident) {
1089 spin_unlock(&xprt_list_lock);
1093 spin_unlock(&xprt_list_lock);
1094 printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1095 return ERR_PTR(-EIO);
1098 xprt = t->setup(args);
1100 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1104 if (test_and_set_bit(XPRT_INITIALIZED, &xprt->state))
1105 /* ->setup returned a pre-initialized xprt: */
1108 spin_lock_init(&xprt->transport_lock);
1109 spin_lock_init(&xprt->reserve_lock);
1111 INIT_LIST_HEAD(&xprt->free);
1112 INIT_LIST_HEAD(&xprt->recv);
1113 #if defined(CONFIG_NFS_V4_1)
1114 spin_lock_init(&xprt->bc_pa_lock);
1115 INIT_LIST_HEAD(&xprt->bc_pa_list);
1116 #endif /* CONFIG_NFS_V4_1 */
1118 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1119 if (xprt_has_timer(xprt))
1120 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1121 (unsigned long)xprt);
1123 init_timer(&xprt->timer);
1124 xprt->last_used = jiffies;
1125 xprt->cwnd = RPC_INITCWND;
1126 xprt->bind_index = 0;
1128 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1129 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1130 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1131 rpc_init_wait_queue(&xprt->resend, "xprt_resend");
1132 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1134 /* initialize free list */
1135 for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
1136 list_add(&req->rq_list, &xprt->free);
1138 xprt_init_xid(xprt);
1140 dprintk("RPC: created transport %p with %u slots\n", xprt,
1146 * xprt_destroy - destroy an RPC transport, killing off all requests.
1147 * @kref: kref for the transport to destroy
1150 static void xprt_destroy(struct kref *kref)
1152 struct rpc_xprt *xprt = container_of(kref, struct rpc_xprt, kref);
1154 dprintk("RPC: destroying transport %p\n", xprt);
1156 del_timer_sync(&xprt->timer);
1158 rpc_destroy_wait_queue(&xprt->binding);
1159 rpc_destroy_wait_queue(&xprt->pending);
1160 rpc_destroy_wait_queue(&xprt->sending);
1161 rpc_destroy_wait_queue(&xprt->resend);
1162 rpc_destroy_wait_queue(&xprt->backlog);
1163 cancel_work_sync(&xprt->task_cleanup);
1165 * Tear down transport state and free the rpc_xprt
1167 xprt->ops->destroy(xprt);
1171 * xprt_put - release a reference to an RPC transport.
1172 * @xprt: pointer to the transport
1175 void xprt_put(struct rpc_xprt *xprt)
1177 kref_put(&xprt->kref, xprt_destroy);
1181 * xprt_get - return a reference to an RPC transport.
1182 * @xprt: pointer to the transport
1185 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1187 kref_get(&xprt->kref);