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)
206 xprt->snd_task = task;
208 req->rq_bytes_sent = 0;
214 dprintk("RPC: %5u failed to lock transport %p\n",
216 task->tk_timeout = 0;
217 task->tk_status = -EAGAIN;
218 if (req && req->rq_ntrans)
219 rpc_sleep_on(&xprt->resend, task, NULL);
221 rpc_sleep_on(&xprt->sending, task, NULL);
224 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
226 static void xprt_clear_locked(struct rpc_xprt *xprt)
228 xprt->snd_task = NULL;
229 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state) || xprt->shutdown) {
230 smp_mb__before_clear_bit();
231 clear_bit(XPRT_LOCKED, &xprt->state);
232 smp_mb__after_clear_bit();
234 queue_work(rpciod_workqueue, &xprt->task_cleanup);
238 * xprt_reserve_xprt_cong - serialize write access to transports
239 * @task: task that is requesting access to the transport
241 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
242 * integrated into the decision of whether a request is allowed to be
243 * woken up and given access to the transport.
245 int xprt_reserve_xprt_cong(struct rpc_task *task)
247 struct rpc_xprt *xprt = task->tk_xprt;
248 struct rpc_rqst *req = task->tk_rqstp;
250 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
251 if (task == xprt->snd_task)
255 if (__xprt_get_cong(xprt, task)) {
256 xprt->snd_task = task;
258 req->rq_bytes_sent = 0;
263 xprt_clear_locked(xprt);
265 dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
266 task->tk_timeout = 0;
267 task->tk_status = -EAGAIN;
268 if (req && req->rq_ntrans)
269 rpc_sleep_on(&xprt->resend, task, NULL);
271 rpc_sleep_on(&xprt->sending, task, NULL);
274 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
276 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
280 spin_lock_bh(&xprt->transport_lock);
281 retval = xprt->ops->reserve_xprt(task);
282 spin_unlock_bh(&xprt->transport_lock);
286 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
288 struct rpc_task *task;
289 struct rpc_rqst *req;
291 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
294 task = rpc_wake_up_next(&xprt->resend);
296 task = rpc_wake_up_next(&xprt->sending);
301 req = task->tk_rqstp;
302 xprt->snd_task = task;
304 req->rq_bytes_sent = 0;
310 xprt_clear_locked(xprt);
313 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
315 struct rpc_task *task;
317 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
319 if (RPCXPRT_CONGESTED(xprt))
321 task = rpc_wake_up_next(&xprt->resend);
323 task = rpc_wake_up_next(&xprt->sending);
327 if (__xprt_get_cong(xprt, task)) {
328 struct rpc_rqst *req = task->tk_rqstp;
329 xprt->snd_task = task;
331 req->rq_bytes_sent = 0;
337 xprt_clear_locked(xprt);
341 * xprt_release_xprt - allow other requests to use a transport
342 * @xprt: transport with other tasks potentially waiting
343 * @task: task that is releasing access to the transport
345 * Note that "task" can be NULL. No congestion control is provided.
347 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
349 if (xprt->snd_task == task) {
350 xprt_clear_locked(xprt);
351 __xprt_lock_write_next(xprt);
354 EXPORT_SYMBOL_GPL(xprt_release_xprt);
357 * xprt_release_xprt_cong - allow other requests to use a transport
358 * @xprt: transport with other tasks potentially waiting
359 * @task: task that is releasing access to the transport
361 * Note that "task" can be NULL. Another task is awoken to use the
362 * transport if the transport's congestion window allows it.
364 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
366 if (xprt->snd_task == task) {
367 xprt_clear_locked(xprt);
368 __xprt_lock_write_next_cong(xprt);
371 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
373 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
375 spin_lock_bh(&xprt->transport_lock);
376 xprt->ops->release_xprt(xprt, task);
377 spin_unlock_bh(&xprt->transport_lock);
381 * Van Jacobson congestion avoidance. Check if the congestion window
382 * overflowed. Put the task to sleep if this is the case.
385 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
387 struct rpc_rqst *req = task->tk_rqstp;
391 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
392 task->tk_pid, xprt->cong, xprt->cwnd);
393 if (RPCXPRT_CONGESTED(xprt))
396 xprt->cong += RPC_CWNDSCALE;
401 * Adjust the congestion window, and wake up the next task
402 * that has been sleeping due to congestion
405 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
410 xprt->cong -= RPC_CWNDSCALE;
411 __xprt_lock_write_next_cong(xprt);
415 * xprt_release_rqst_cong - housekeeping when request is complete
416 * @task: RPC request that recently completed
418 * Useful for transports that require congestion control.
420 void xprt_release_rqst_cong(struct rpc_task *task)
422 __xprt_put_cong(task->tk_xprt, task->tk_rqstp);
424 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
427 * xprt_adjust_cwnd - adjust transport congestion window
428 * @task: recently completed RPC request used to adjust window
429 * @result: result code of completed RPC request
431 * We use a time-smoothed congestion estimator to avoid heavy oscillation.
433 void xprt_adjust_cwnd(struct rpc_task *task, int result)
435 struct rpc_rqst *req = task->tk_rqstp;
436 struct rpc_xprt *xprt = task->tk_xprt;
437 unsigned long cwnd = xprt->cwnd;
439 if (result >= 0 && cwnd <= xprt->cong) {
440 /* The (cwnd >> 1) term makes sure
441 * the result gets rounded properly. */
442 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
443 if (cwnd > RPC_MAXCWND(xprt))
444 cwnd = RPC_MAXCWND(xprt);
445 __xprt_lock_write_next_cong(xprt);
446 } else if (result == -ETIMEDOUT) {
448 if (cwnd < RPC_CWNDSCALE)
449 cwnd = RPC_CWNDSCALE;
451 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
452 xprt->cong, xprt->cwnd, cwnd);
454 __xprt_put_cong(xprt, req);
456 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
459 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
460 * @xprt: transport with waiting tasks
461 * @status: result code to plant in each task before waking it
464 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
467 rpc_wake_up_status(&xprt->pending, status);
469 rpc_wake_up(&xprt->pending);
471 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
474 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
475 * @task: task to be put to sleep
476 * @action: function pointer to be executed after wait
478 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
480 struct rpc_rqst *req = task->tk_rqstp;
481 struct rpc_xprt *xprt = req->rq_xprt;
483 task->tk_timeout = req->rq_timeout;
484 rpc_sleep_on(&xprt->pending, task, action);
486 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
489 * xprt_write_space - wake the task waiting for transport output buffer space
490 * @xprt: transport with waiting tasks
492 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
494 void xprt_write_space(struct rpc_xprt *xprt)
496 if (unlikely(xprt->shutdown))
499 spin_lock_bh(&xprt->transport_lock);
500 if (xprt->snd_task) {
501 dprintk("RPC: write space: waking waiting task on "
503 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
505 spin_unlock_bh(&xprt->transport_lock);
507 EXPORT_SYMBOL_GPL(xprt_write_space);
510 * xprt_set_retrans_timeout_def - set a request's retransmit timeout
511 * @task: task whose timeout is to be set
513 * Set a request's retransmit timeout based on the transport's
514 * default timeout parameters. Used by transports that don't adjust
515 * the retransmit timeout based on round-trip time estimation.
517 void xprt_set_retrans_timeout_def(struct rpc_task *task)
519 task->tk_timeout = task->tk_rqstp->rq_timeout;
521 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
524 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
525 * @task: task whose timeout is to be set
527 * Set a request's retransmit timeout using the RTT estimator.
529 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
531 int timer = task->tk_msg.rpc_proc->p_timer;
532 struct rpc_clnt *clnt = task->tk_client;
533 struct rpc_rtt *rtt = clnt->cl_rtt;
534 struct rpc_rqst *req = task->tk_rqstp;
535 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
537 task->tk_timeout = rpc_calc_rto(rtt, timer);
538 task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
539 if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
540 task->tk_timeout = max_timeout;
542 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
544 static void xprt_reset_majortimeo(struct rpc_rqst *req)
546 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
548 req->rq_majortimeo = req->rq_timeout;
549 if (to->to_exponential)
550 req->rq_majortimeo <<= to->to_retries;
552 req->rq_majortimeo += to->to_increment * to->to_retries;
553 if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
554 req->rq_majortimeo = to->to_maxval;
555 req->rq_majortimeo += jiffies;
559 * xprt_adjust_timeout - adjust timeout values for next retransmit
560 * @req: RPC request containing parameters to use for the adjustment
563 int xprt_adjust_timeout(struct rpc_rqst *req)
565 struct rpc_xprt *xprt = req->rq_xprt;
566 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
569 if (time_before(jiffies, req->rq_majortimeo)) {
570 if (to->to_exponential)
571 req->rq_timeout <<= 1;
573 req->rq_timeout += to->to_increment;
574 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
575 req->rq_timeout = to->to_maxval;
578 req->rq_timeout = to->to_initval;
580 xprt_reset_majortimeo(req);
581 /* Reset the RTT counters == "slow start" */
582 spin_lock_bh(&xprt->transport_lock);
583 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
584 spin_unlock_bh(&xprt->transport_lock);
588 if (req->rq_timeout == 0) {
589 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
590 req->rq_timeout = 5 * HZ;
595 static void xprt_autoclose(struct work_struct *work)
597 struct rpc_xprt *xprt =
598 container_of(work, struct rpc_xprt, task_cleanup);
600 xprt->ops->close(xprt);
601 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
602 xprt_release_write(xprt, NULL);
606 * xprt_disconnect_done - mark a transport as disconnected
607 * @xprt: transport to flag for disconnect
610 void xprt_disconnect_done(struct rpc_xprt *xprt)
612 dprintk("RPC: disconnected transport %p\n", xprt);
613 spin_lock_bh(&xprt->transport_lock);
614 xprt_clear_connected(xprt);
615 xprt_wake_pending_tasks(xprt, -EAGAIN);
616 spin_unlock_bh(&xprt->transport_lock);
618 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
621 * xprt_force_disconnect - force a transport to disconnect
622 * @xprt: transport to disconnect
625 void xprt_force_disconnect(struct rpc_xprt *xprt)
627 /* Don't race with the test_bit() in xprt_clear_locked() */
628 spin_lock_bh(&xprt->transport_lock);
629 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
630 /* Try to schedule an autoclose RPC call */
631 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
632 queue_work(rpciod_workqueue, &xprt->task_cleanup);
633 xprt_wake_pending_tasks(xprt, -EAGAIN);
634 spin_unlock_bh(&xprt->transport_lock);
638 * xprt_conditional_disconnect - force a transport to disconnect
639 * @xprt: transport to disconnect
640 * @cookie: 'connection cookie'
642 * This attempts to break the connection if and only if 'cookie' matches
643 * the current transport 'connection cookie'. It ensures that we don't
644 * try to break the connection more than once when we need to retransmit
645 * a batch of RPC requests.
648 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
650 /* Don't race with the test_bit() in xprt_clear_locked() */
651 spin_lock_bh(&xprt->transport_lock);
652 if (cookie != xprt->connect_cookie)
654 if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
656 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
657 /* Try to schedule an autoclose RPC call */
658 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
659 queue_work(rpciod_workqueue, &xprt->task_cleanup);
660 xprt_wake_pending_tasks(xprt, -EAGAIN);
662 spin_unlock_bh(&xprt->transport_lock);
666 xprt_init_autodisconnect(unsigned long data)
668 struct rpc_xprt *xprt = (struct rpc_xprt *)data;
670 spin_lock(&xprt->transport_lock);
671 if (!list_empty(&xprt->recv) || xprt->shutdown)
673 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
675 spin_unlock(&xprt->transport_lock);
676 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
677 queue_work(rpciod_workqueue, &xprt->task_cleanup);
680 spin_unlock(&xprt->transport_lock);
684 * xprt_connect - schedule a transport connect operation
685 * @task: RPC task that is requesting the connect
688 void xprt_connect(struct rpc_task *task)
690 struct rpc_xprt *xprt = task->tk_xprt;
692 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
693 xprt, (xprt_connected(xprt) ? "is" : "is not"));
695 if (!xprt_bound(xprt)) {
696 task->tk_status = -EAGAIN;
699 if (!xprt_lock_write(xprt, task))
702 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
703 xprt->ops->close(xprt);
705 if (xprt_connected(xprt))
706 xprt_release_write(xprt, task);
709 task->tk_rqstp->rq_bytes_sent = 0;
711 task->tk_timeout = task->tk_rqstp->rq_timeout;
712 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
714 if (test_bit(XPRT_CLOSING, &xprt->state))
716 if (xprt_test_and_set_connecting(xprt))
718 xprt->stat.connect_start = jiffies;
719 xprt->ops->connect(task);
723 static void xprt_connect_status(struct rpc_task *task)
725 struct rpc_xprt *xprt = task->tk_xprt;
727 if (task->tk_status == 0) {
728 xprt->stat.connect_count++;
729 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
730 dprintk("RPC: %5u xprt_connect_status: connection established\n",
735 switch (task->tk_status) {
737 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
740 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
741 "out\n", task->tk_pid);
744 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
745 "server %s\n", task->tk_pid, -task->tk_status,
746 task->tk_client->cl_server);
747 xprt_release_write(xprt, task);
748 task->tk_status = -EIO;
753 * xprt_lookup_rqst - find an RPC request corresponding to an XID
754 * @xprt: transport on which the original request was transmitted
755 * @xid: RPC XID of incoming reply
758 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
760 struct list_head *pos;
762 list_for_each(pos, &xprt->recv) {
763 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
764 if (entry->rq_xid == xid)
768 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
770 xprt->stat.bad_xids++;
773 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
775 static void xprt_update_rtt(struct rpc_task *task)
777 struct rpc_rqst *req = task->tk_rqstp;
778 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
779 unsigned timer = task->tk_msg.rpc_proc->p_timer;
780 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
783 if (req->rq_ntrans == 1)
784 rpc_update_rtt(rtt, timer, m);
785 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
790 * xprt_complete_rqst - called when reply processing is complete
791 * @task: RPC request that recently completed
792 * @copied: actual number of bytes received from the transport
794 * Caller holds transport lock.
796 void xprt_complete_rqst(struct rpc_task *task, int copied)
798 struct rpc_rqst *req = task->tk_rqstp;
799 struct rpc_xprt *xprt = req->rq_xprt;
801 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
802 task->tk_pid, ntohl(req->rq_xid), copied);
805 req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
806 if (xprt->ops->timer != NULL)
807 xprt_update_rtt(task);
809 list_del_init(&req->rq_list);
810 req->rq_private_buf.len = copied;
811 /* Ensure all writes are done before we update */
812 /* req->rq_reply_bytes_recvd */
814 req->rq_reply_bytes_recvd = copied;
815 rpc_wake_up_queued_task(&xprt->pending, task);
817 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
819 static void xprt_timer(struct rpc_task *task)
821 struct rpc_rqst *req = task->tk_rqstp;
822 struct rpc_xprt *xprt = req->rq_xprt;
824 if (task->tk_status != -ETIMEDOUT)
826 dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
828 spin_lock_bh(&xprt->transport_lock);
829 if (!req->rq_reply_bytes_recvd) {
830 if (xprt->ops->timer)
831 xprt->ops->timer(task);
834 spin_unlock_bh(&xprt->transport_lock);
837 static inline int xprt_has_timer(struct rpc_xprt *xprt)
839 return xprt->idle_timeout != 0;
843 * xprt_prepare_transmit - reserve the transport before sending a request
844 * @task: RPC task about to send a request
847 int xprt_prepare_transmit(struct rpc_task *task)
849 struct rpc_rqst *req = task->tk_rqstp;
850 struct rpc_xprt *xprt = req->rq_xprt;
853 dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
855 spin_lock_bh(&xprt->transport_lock);
856 if (req->rq_reply_bytes_recvd && !req->rq_bytes_sent) {
857 err = req->rq_reply_bytes_recvd;
860 if (!xprt->ops->reserve_xprt(task))
863 spin_unlock_bh(&xprt->transport_lock);
867 void xprt_end_transmit(struct rpc_task *task)
869 xprt_release_write(task->tk_rqstp->rq_xprt, task);
873 * xprt_transmit - send an RPC request on a transport
874 * @task: controlling RPC task
876 * We have to copy the iovec because sendmsg fiddles with its contents.
878 void xprt_transmit(struct rpc_task *task)
880 struct rpc_rqst *req = task->tk_rqstp;
881 struct rpc_xprt *xprt = req->rq_xprt;
884 dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
886 if (!req->rq_reply_bytes_recvd) {
887 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
889 * Add to the list only if we're expecting a reply
891 spin_lock_bh(&xprt->transport_lock);
892 /* Update the softirq receive buffer */
893 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
894 sizeof(req->rq_private_buf));
895 /* Add request to the receive list */
896 list_add_tail(&req->rq_list, &xprt->recv);
897 spin_unlock_bh(&xprt->transport_lock);
898 xprt_reset_majortimeo(req);
899 /* Turn off autodisconnect */
900 del_singleshot_timer_sync(&xprt->timer);
902 } else if (!req->rq_bytes_sent)
905 req->rq_connect_cookie = xprt->connect_cookie;
906 req->rq_xtime = ktime_get();
907 status = xprt->ops->send_request(task);
909 task->tk_status = status;
913 dprintk("RPC: %5u xmit complete\n", task->tk_pid);
914 spin_lock_bh(&xprt->transport_lock);
916 xprt->ops->set_retrans_timeout(task);
919 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
920 xprt->stat.bklog_u += xprt->backlog.qlen;
922 /* Don't race with disconnect */
923 if (!xprt_connected(xprt))
924 task->tk_status = -ENOTCONN;
925 else if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task)) {
927 * Sleep on the pending queue since
928 * we're expecting a reply.
930 rpc_sleep_on(&xprt->pending, task, xprt_timer);
932 spin_unlock_bh(&xprt->transport_lock);
935 static void xprt_alloc_slot(struct rpc_task *task)
937 struct rpc_xprt *xprt = task->tk_xprt;
942 if (!list_empty(&xprt->free)) {
943 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
944 list_del_init(&req->rq_list);
945 task->tk_rqstp = req;
946 xprt_request_init(task, xprt);
949 dprintk("RPC: waiting for request slot\n");
950 task->tk_status = -EAGAIN;
951 task->tk_timeout = 0;
952 rpc_sleep_on(&xprt->backlog, task, NULL);
955 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
957 memset(req, 0, sizeof(*req)); /* mark unused */
959 spin_lock(&xprt->reserve_lock);
960 list_add(&req->rq_list, &xprt->free);
961 rpc_wake_up_next(&xprt->backlog);
962 spin_unlock(&xprt->reserve_lock);
966 * xprt_reserve - allocate an RPC request slot
967 * @task: RPC task requesting a slot allocation
969 * If no more slots are available, place the task on the transport's
972 void xprt_reserve(struct rpc_task *task)
974 struct rpc_xprt *xprt = task->tk_xprt;
976 task->tk_status = -EIO;
977 spin_lock(&xprt->reserve_lock);
978 xprt_alloc_slot(task);
979 spin_unlock(&xprt->reserve_lock);
982 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
984 return (__force __be32)xprt->xid++;
987 static inline void xprt_init_xid(struct rpc_xprt *xprt)
989 xprt->xid = net_random();
992 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
994 struct rpc_rqst *req = task->tk_rqstp;
996 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
999 req->rq_buffer = NULL;
1000 req->rq_xid = xprt_alloc_xid(xprt);
1001 req->rq_release_snd_buf = NULL;
1002 xprt_reset_majortimeo(req);
1003 dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1004 req, ntohl(req->rq_xid));
1008 * xprt_release - release an RPC request slot
1009 * @task: task which is finished with the slot
1012 void xprt_release(struct rpc_task *task)
1014 struct rpc_xprt *xprt;
1015 struct rpc_rqst *req;
1017 if (!(req = task->tk_rqstp))
1020 xprt = req->rq_xprt;
1021 rpc_count_iostats(task);
1022 spin_lock_bh(&xprt->transport_lock);
1023 xprt->ops->release_xprt(xprt, task);
1024 if (xprt->ops->release_request)
1025 xprt->ops->release_request(task);
1026 if (!list_empty(&req->rq_list))
1027 list_del(&req->rq_list);
1028 xprt->last_used = jiffies;
1029 if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1030 mod_timer(&xprt->timer,
1031 xprt->last_used + xprt->idle_timeout);
1032 spin_unlock_bh(&xprt->transport_lock);
1034 xprt->ops->buf_free(req->rq_buffer);
1035 if (req->rq_cred != NULL)
1036 put_rpccred(req->rq_cred);
1037 task->tk_rqstp = NULL;
1038 if (req->rq_release_snd_buf)
1039 req->rq_release_snd_buf(req);
1041 dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1042 if (likely(!bc_prealloc(req)))
1043 xprt_free_slot(xprt, req);
1045 xprt_free_bc_request(req);
1049 * xprt_create_transport - create an RPC transport
1050 * @args: rpc transport creation arguments
1053 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1055 struct rpc_xprt *xprt;
1056 struct rpc_rqst *req;
1057 struct xprt_class *t;
1059 spin_lock(&xprt_list_lock);
1060 list_for_each_entry(t, &xprt_list, list) {
1061 if (t->ident == args->ident) {
1062 spin_unlock(&xprt_list_lock);
1066 spin_unlock(&xprt_list_lock);
1067 printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1068 return ERR_PTR(-EIO);
1071 xprt = t->setup(args);
1073 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1078 kref_init(&xprt->kref);
1079 spin_lock_init(&xprt->transport_lock);
1080 spin_lock_init(&xprt->reserve_lock);
1082 INIT_LIST_HEAD(&xprt->free);
1083 INIT_LIST_HEAD(&xprt->recv);
1084 #if defined(CONFIG_NFS_V4_1)
1085 spin_lock_init(&xprt->bc_pa_lock);
1086 INIT_LIST_HEAD(&xprt->bc_pa_list);
1087 #endif /* CONFIG_NFS_V4_1 */
1089 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1090 if (xprt_has_timer(xprt))
1091 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1092 (unsigned long)xprt);
1094 init_timer(&xprt->timer);
1095 xprt->last_used = jiffies;
1096 xprt->cwnd = RPC_INITCWND;
1097 xprt->bind_index = 0;
1099 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1100 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1101 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1102 rpc_init_wait_queue(&xprt->resend, "xprt_resend");
1103 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1105 /* initialize free list */
1106 for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
1107 list_add(&req->rq_list, &xprt->free);
1109 xprt_init_xid(xprt);
1111 dprintk("RPC: created transport %p with %u slots\n", xprt,
1117 * xprt_destroy - destroy an RPC transport, killing off all requests.
1118 * @kref: kref for the transport to destroy
1121 static void xprt_destroy(struct kref *kref)
1123 struct rpc_xprt *xprt = container_of(kref, struct rpc_xprt, kref);
1125 dprintk("RPC: destroying transport %p\n", xprt);
1127 del_timer_sync(&xprt->timer);
1129 rpc_destroy_wait_queue(&xprt->binding);
1130 rpc_destroy_wait_queue(&xprt->pending);
1131 rpc_destroy_wait_queue(&xprt->sending);
1132 rpc_destroy_wait_queue(&xprt->resend);
1133 rpc_destroy_wait_queue(&xprt->backlog);
1134 cancel_work_sync(&xprt->task_cleanup);
1136 * Tear down transport state and free the rpc_xprt
1138 xprt->ops->destroy(xprt);
1142 * xprt_put - release a reference to an RPC transport.
1143 * @xprt: pointer to the transport
1146 void xprt_put(struct rpc_xprt *xprt)
1148 kref_put(&xprt->kref, xprt_destroy);
1152 * xprt_get - return a reference to an RPC transport.
1153 * @xprt: pointer to the transport
1156 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1158 kref_get(&xprt->kref);