usb: Introduce DesignWare USB3 DRD Driver
[zen-stable.git] / net / sunrpc / xprt.c
blobf4385e45a5fcb398086c6c8c5c8ab4daff7ee139
1 /*
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
11 * (xprt_reserve).
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
17 * expired.
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
24 * of -ETIMEDOUT.
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
29 * again.
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>
52 #include "sunrpc.h"
55 * Local variables
58 #ifdef RPC_DEBUG
59 # define RPCDBG_FACILITY RPCDBG_XPRT
60 #endif
63 * Local functions
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)
91 /**
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.
98 * Returns:
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;
106 int result;
108 result = -EEXIST;
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)
113 goto out;
116 list_add_tail(&transport->list, &xprt_list);
117 printk(KERN_INFO "RPC: Registered %s transport module.\n",
118 transport->name);
119 result = 0;
121 out:
122 spin_unlock(&xprt_list_lock);
123 return result;
125 EXPORT_SYMBOL_GPL(xprt_register_transport);
128 * xprt_unregister_transport - unregister a transport implementation
129 * @transport: transport to unregister
131 * Returns:
132 * 0: transport successfully unregistered
133 * -ENOENT: transport never registered
135 int xprt_unregister_transport(struct xprt_class *transport)
137 struct xprt_class *t;
138 int result;
140 result = 0;
141 spin_lock(&xprt_list_lock);
142 list_for_each_entry(t, &xprt_list, list) {
143 if (t == transport) {
144 printk(KERN_INFO
145 "RPC: Unregistered %s transport module.\n",
146 transport->name);
147 list_del_init(&transport->list);
148 goto out;
151 result = -ENOENT;
153 out:
154 spin_unlock(&xprt_list_lock);
155 return result;
157 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
160 * xprt_load_transport - load a transport implementation
161 * @transport_name: transport to load
163 * Returns:
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;
170 int result;
172 result = 0;
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);
177 goto out;
180 spin_unlock(&xprt_list_lock);
181 result = request_module("xprt%s", transport_name);
182 out:
183 return result;
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
194 * is provided.
196 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
198 struct rpc_rqst *req = task->tk_rqstp;
199 int priority;
201 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
202 if (task == xprt->snd_task)
203 return 1;
204 goto out_sleep;
206 xprt->snd_task = task;
207 if (req != NULL) {
208 req->rq_bytes_sent = 0;
209 req->rq_ntrans++;
212 return 1;
214 out_sleep:
215 dprintk("RPC: %5u failed to lock transport %p\n",
216 task->tk_pid, xprt);
217 task->tk_timeout = 0;
218 task->tk_status = -EAGAIN;
219 if (req == NULL)
220 priority = RPC_PRIORITY_LOW;
221 else if (!req->rq_ntrans)
222 priority = RPC_PRIORITY_NORMAL;
223 else
224 priority = RPC_PRIORITY_HIGH;
225 rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
226 return 0;
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();
237 } else
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;
252 int priority;
254 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
255 if (task == xprt->snd_task)
256 return 1;
257 goto out_sleep;
259 if (req == NULL) {
260 xprt->snd_task = task;
261 return 1;
263 if (__xprt_get_cong(xprt, task)) {
264 xprt->snd_task = task;
265 req->rq_bytes_sent = 0;
266 req->rq_ntrans++;
267 return 1;
269 xprt_clear_locked(xprt);
270 out_sleep:
271 dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
272 task->tk_timeout = 0;
273 task->tk_status = -EAGAIN;
274 if (req == NULL)
275 priority = RPC_PRIORITY_LOW;
276 else if (!req->rq_ntrans)
277 priority = RPC_PRIORITY_NORMAL;
278 else
279 priority = RPC_PRIORITY_HIGH;
280 rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
281 return 0;
283 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
285 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
287 int retval;
289 spin_lock_bh(&xprt->transport_lock);
290 retval = xprt->ops->reserve_xprt(xprt, task);
291 spin_unlock_bh(&xprt->transport_lock);
292 return retval;
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))
301 return;
303 task = rpc_wake_up_next(&xprt->sending);
304 if (task == NULL)
305 goto out_unlock;
307 req = task->tk_rqstp;
308 xprt->snd_task = task;
309 if (req) {
310 req->rq_bytes_sent = 0;
311 req->rq_ntrans++;
313 return;
315 out_unlock:
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))
325 return;
326 if (RPCXPRT_CONGESTED(xprt))
327 goto out_unlock;
328 task = rpc_wake_up_next(&xprt->sending);
329 if (task == NULL)
330 goto out_unlock;
332 req = task->tk_rqstp;
333 if (req == NULL) {
334 xprt->snd_task = task;
335 return;
337 if (__xprt_get_cong(xprt, task)) {
338 xprt->snd_task = task;
339 req->rq_bytes_sent = 0;
340 req->rq_ntrans++;
341 return;
343 out_unlock:
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.
391 static int
392 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
394 struct rpc_rqst *req = task->tk_rqstp;
396 if (req->rq_cong)
397 return 1;
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))
401 return 0;
402 req->rq_cong = 1;
403 xprt->cong += RPC_CWNDSCALE;
404 return 1;
408 * Adjust the congestion window, and wake up the next task
409 * that has been sleeping due to congestion
411 static void
412 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
414 if (!req->rq_cong)
415 return;
416 req->rq_cong = 0;
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) {
454 cwnd >>= 1;
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);
460 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)
473 if (status < 0)
474 rpc_wake_up_status(&xprt->pending, status);
475 else
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 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
487 struct rpc_rqst *req = task->tk_rqstp;
488 struct rpc_xprt *xprt = req->rq_xprt;
490 task->tk_timeout = req->rq_timeout;
491 rpc_sleep_on(&xprt->pending, task, action);
493 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
496 * xprt_write_space - wake the task waiting for transport output buffer space
497 * @xprt: transport with waiting tasks
499 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
501 void xprt_write_space(struct rpc_xprt *xprt)
503 if (unlikely(xprt->shutdown))
504 return;
506 spin_lock_bh(&xprt->transport_lock);
507 if (xprt->snd_task) {
508 dprintk("RPC: write space: waking waiting task on "
509 "xprt %p\n", xprt);
510 rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
512 spin_unlock_bh(&xprt->transport_lock);
514 EXPORT_SYMBOL_GPL(xprt_write_space);
517 * xprt_set_retrans_timeout_def - set a request's retransmit timeout
518 * @task: task whose timeout is to be set
520 * Set a request's retransmit timeout based on the transport's
521 * default timeout parameters. Used by transports that don't adjust
522 * the retransmit timeout based on round-trip time estimation.
524 void xprt_set_retrans_timeout_def(struct rpc_task *task)
526 task->tk_timeout = task->tk_rqstp->rq_timeout;
528 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
531 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
532 * @task: task whose timeout is to be set
534 * Set a request's retransmit timeout using the RTT estimator.
536 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
538 int timer = task->tk_msg.rpc_proc->p_timer;
539 struct rpc_clnt *clnt = task->tk_client;
540 struct rpc_rtt *rtt = clnt->cl_rtt;
541 struct rpc_rqst *req = task->tk_rqstp;
542 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
544 task->tk_timeout = rpc_calc_rto(rtt, timer);
545 task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
546 if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
547 task->tk_timeout = max_timeout;
549 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
551 static void xprt_reset_majortimeo(struct rpc_rqst *req)
553 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
555 req->rq_majortimeo = req->rq_timeout;
556 if (to->to_exponential)
557 req->rq_majortimeo <<= to->to_retries;
558 else
559 req->rq_majortimeo += to->to_increment * to->to_retries;
560 if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
561 req->rq_majortimeo = to->to_maxval;
562 req->rq_majortimeo += jiffies;
566 * xprt_adjust_timeout - adjust timeout values for next retransmit
567 * @req: RPC request containing parameters to use for the adjustment
570 int xprt_adjust_timeout(struct rpc_rqst *req)
572 struct rpc_xprt *xprt = req->rq_xprt;
573 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
574 int status = 0;
576 if (time_before(jiffies, req->rq_majortimeo)) {
577 if (to->to_exponential)
578 req->rq_timeout <<= 1;
579 else
580 req->rq_timeout += to->to_increment;
581 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
582 req->rq_timeout = to->to_maxval;
583 req->rq_retries++;
584 } else {
585 req->rq_timeout = to->to_initval;
586 req->rq_retries = 0;
587 xprt_reset_majortimeo(req);
588 /* Reset the RTT counters == "slow start" */
589 spin_lock_bh(&xprt->transport_lock);
590 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
591 spin_unlock_bh(&xprt->transport_lock);
592 status = -ETIMEDOUT;
595 if (req->rq_timeout == 0) {
596 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
597 req->rq_timeout = 5 * HZ;
599 return status;
602 static void xprt_autoclose(struct work_struct *work)
604 struct rpc_xprt *xprt =
605 container_of(work, struct rpc_xprt, task_cleanup);
607 xprt->ops->close(xprt);
608 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
609 xprt_release_write(xprt, NULL);
613 * xprt_disconnect_done - mark a transport as disconnected
614 * @xprt: transport to flag for disconnect
617 void xprt_disconnect_done(struct rpc_xprt *xprt)
619 dprintk("RPC: disconnected transport %p\n", xprt);
620 spin_lock_bh(&xprt->transport_lock);
621 xprt_clear_connected(xprt);
622 xprt_wake_pending_tasks(xprt, -EAGAIN);
623 spin_unlock_bh(&xprt->transport_lock);
625 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
628 * xprt_force_disconnect - force a transport to disconnect
629 * @xprt: transport to disconnect
632 void xprt_force_disconnect(struct rpc_xprt *xprt)
634 /* Don't race with the test_bit() in xprt_clear_locked() */
635 spin_lock_bh(&xprt->transport_lock);
636 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
637 /* Try to schedule an autoclose RPC call */
638 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
639 queue_work(rpciod_workqueue, &xprt->task_cleanup);
640 xprt_wake_pending_tasks(xprt, -EAGAIN);
641 spin_unlock_bh(&xprt->transport_lock);
645 * xprt_conditional_disconnect - force a transport to disconnect
646 * @xprt: transport to disconnect
647 * @cookie: 'connection cookie'
649 * This attempts to break the connection if and only if 'cookie' matches
650 * the current transport 'connection cookie'. It ensures that we don't
651 * try to break the connection more than once when we need to retransmit
652 * a batch of RPC requests.
655 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
657 /* Don't race with the test_bit() in xprt_clear_locked() */
658 spin_lock_bh(&xprt->transport_lock);
659 if (cookie != xprt->connect_cookie)
660 goto out;
661 if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
662 goto out;
663 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
664 /* Try to schedule an autoclose RPC call */
665 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
666 queue_work(rpciod_workqueue, &xprt->task_cleanup);
667 xprt_wake_pending_tasks(xprt, -EAGAIN);
668 out:
669 spin_unlock_bh(&xprt->transport_lock);
672 static void
673 xprt_init_autodisconnect(unsigned long data)
675 struct rpc_xprt *xprt = (struct rpc_xprt *)data;
677 spin_lock(&xprt->transport_lock);
678 if (!list_empty(&xprt->recv) || xprt->shutdown)
679 goto out_abort;
680 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
681 goto out_abort;
682 spin_unlock(&xprt->transport_lock);
683 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
684 queue_work(rpciod_workqueue, &xprt->task_cleanup);
685 return;
686 out_abort:
687 spin_unlock(&xprt->transport_lock);
691 * xprt_connect - schedule a transport connect operation
692 * @task: RPC task that is requesting the connect
695 void xprt_connect(struct rpc_task *task)
697 struct rpc_xprt *xprt = task->tk_xprt;
699 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
700 xprt, (xprt_connected(xprt) ? "is" : "is not"));
702 if (!xprt_bound(xprt)) {
703 task->tk_status = -EAGAIN;
704 return;
706 if (!xprt_lock_write(xprt, task))
707 return;
709 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
710 xprt->ops->close(xprt);
712 if (xprt_connected(xprt))
713 xprt_release_write(xprt, task);
714 else {
715 if (task->tk_rqstp)
716 task->tk_rqstp->rq_bytes_sent = 0;
718 task->tk_timeout = task->tk_rqstp->rq_timeout;
719 rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
721 if (test_bit(XPRT_CLOSING, &xprt->state))
722 return;
723 if (xprt_test_and_set_connecting(xprt))
724 return;
725 xprt->stat.connect_start = jiffies;
726 xprt->ops->connect(task);
730 static void xprt_connect_status(struct rpc_task *task)
732 struct rpc_xprt *xprt = task->tk_xprt;
734 if (task->tk_status == 0) {
735 xprt->stat.connect_count++;
736 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
737 dprintk("RPC: %5u xprt_connect_status: connection established\n",
738 task->tk_pid);
739 return;
742 switch (task->tk_status) {
743 case -EAGAIN:
744 dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
745 break;
746 case -ETIMEDOUT:
747 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
748 "out\n", task->tk_pid);
749 break;
750 default:
751 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
752 "server %s\n", task->tk_pid, -task->tk_status,
753 task->tk_client->cl_server);
754 xprt_release_write(xprt, task);
755 task->tk_status = -EIO;
760 * xprt_lookup_rqst - find an RPC request corresponding to an XID
761 * @xprt: transport on which the original request was transmitted
762 * @xid: RPC XID of incoming reply
765 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
767 struct rpc_rqst *entry;
769 list_for_each_entry(entry, &xprt->recv, rq_list)
770 if (entry->rq_xid == xid)
771 return entry;
773 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
774 ntohl(xid));
775 xprt->stat.bad_xids++;
776 return NULL;
778 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
780 static void xprt_update_rtt(struct rpc_task *task)
782 struct rpc_rqst *req = task->tk_rqstp;
783 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
784 unsigned timer = task->tk_msg.rpc_proc->p_timer;
785 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
787 if (timer) {
788 if (req->rq_ntrans == 1)
789 rpc_update_rtt(rtt, timer, m);
790 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
795 * xprt_complete_rqst - called when reply processing is complete
796 * @task: RPC request that recently completed
797 * @copied: actual number of bytes received from the transport
799 * Caller holds transport lock.
801 void xprt_complete_rqst(struct rpc_task *task, int copied)
803 struct rpc_rqst *req = task->tk_rqstp;
804 struct rpc_xprt *xprt = req->rq_xprt;
806 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
807 task->tk_pid, ntohl(req->rq_xid), copied);
809 xprt->stat.recvs++;
810 req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
811 if (xprt->ops->timer != NULL)
812 xprt_update_rtt(task);
814 list_del_init(&req->rq_list);
815 req->rq_private_buf.len = copied;
816 /* Ensure all writes are done before we update */
817 /* req->rq_reply_bytes_recvd */
818 smp_wmb();
819 req->rq_reply_bytes_recvd = copied;
820 rpc_wake_up_queued_task(&xprt->pending, task);
822 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
824 static void xprt_timer(struct rpc_task *task)
826 struct rpc_rqst *req = task->tk_rqstp;
827 struct rpc_xprt *xprt = req->rq_xprt;
829 if (task->tk_status != -ETIMEDOUT)
830 return;
831 dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
833 spin_lock_bh(&xprt->transport_lock);
834 if (!req->rq_reply_bytes_recvd) {
835 if (xprt->ops->timer)
836 xprt->ops->timer(task);
837 } else
838 task->tk_status = 0;
839 spin_unlock_bh(&xprt->transport_lock);
842 static inline int xprt_has_timer(struct rpc_xprt *xprt)
844 return xprt->idle_timeout != 0;
848 * xprt_prepare_transmit - reserve the transport before sending a request
849 * @task: RPC task about to send a request
852 int xprt_prepare_transmit(struct rpc_task *task)
854 struct rpc_rqst *req = task->tk_rqstp;
855 struct rpc_xprt *xprt = req->rq_xprt;
856 int err = 0;
858 dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
860 spin_lock_bh(&xprt->transport_lock);
861 if (req->rq_reply_bytes_recvd && !req->rq_bytes_sent) {
862 err = req->rq_reply_bytes_recvd;
863 goto out_unlock;
865 if (!xprt->ops->reserve_xprt(xprt, task))
866 err = -EAGAIN;
867 out_unlock:
868 spin_unlock_bh(&xprt->transport_lock);
869 return err;
872 void xprt_end_transmit(struct rpc_task *task)
874 xprt_release_write(task->tk_rqstp->rq_xprt, task);
878 * xprt_transmit - send an RPC request on a transport
879 * @task: controlling RPC task
881 * We have to copy the iovec because sendmsg fiddles with its contents.
883 void xprt_transmit(struct rpc_task *task)
885 struct rpc_rqst *req = task->tk_rqstp;
886 struct rpc_xprt *xprt = req->rq_xprt;
887 int status;
889 dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
891 if (!req->rq_reply_bytes_recvd) {
892 if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
894 * Add to the list only if we're expecting a reply
896 spin_lock_bh(&xprt->transport_lock);
897 /* Update the softirq receive buffer */
898 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
899 sizeof(req->rq_private_buf));
900 /* Add request to the receive list */
901 list_add_tail(&req->rq_list, &xprt->recv);
902 spin_unlock_bh(&xprt->transport_lock);
903 xprt_reset_majortimeo(req);
904 /* Turn off autodisconnect */
905 del_singleshot_timer_sync(&xprt->timer);
907 } else if (!req->rq_bytes_sent)
908 return;
910 req->rq_connect_cookie = xprt->connect_cookie;
911 req->rq_xtime = ktime_get();
912 status = xprt->ops->send_request(task);
913 if (status != 0) {
914 task->tk_status = status;
915 return;
918 dprintk("RPC: %5u xmit complete\n", task->tk_pid);
919 task->tk_flags |= RPC_TASK_SENT;
920 spin_lock_bh(&xprt->transport_lock);
922 xprt->ops->set_retrans_timeout(task);
924 xprt->stat.sends++;
925 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
926 xprt->stat.bklog_u += xprt->backlog.qlen;
928 /* Don't race with disconnect */
929 if (!xprt_connected(xprt))
930 task->tk_status = -ENOTCONN;
931 else if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task)) {
933 * Sleep on the pending queue since
934 * we're expecting a reply.
936 rpc_sleep_on(&xprt->pending, task, xprt_timer);
938 spin_unlock_bh(&xprt->transport_lock);
941 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
943 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
945 if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
946 goto out;
947 req = kzalloc(sizeof(struct rpc_rqst), gfp_flags);
948 if (req != NULL)
949 goto out;
950 atomic_dec(&xprt->num_reqs);
951 req = ERR_PTR(-ENOMEM);
952 out:
953 return req;
956 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
958 if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
959 kfree(req);
960 return true;
962 return false;
965 static void xprt_alloc_slot(struct rpc_task *task)
967 struct rpc_xprt *xprt = task->tk_xprt;
968 struct rpc_rqst *req;
970 if (!list_empty(&xprt->free)) {
971 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
972 list_del(&req->rq_list);
973 goto out_init_req;
975 req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT);
976 if (!IS_ERR(req))
977 goto out_init_req;
978 switch (PTR_ERR(req)) {
979 case -ENOMEM:
980 rpc_delay(task, HZ >> 2);
981 dprintk("RPC: dynamic allocation of request slot "
982 "failed! Retrying\n");
983 break;
984 case -EAGAIN:
985 rpc_sleep_on(&xprt->backlog, task, NULL);
986 dprintk("RPC: waiting for request slot\n");
988 task->tk_status = -EAGAIN;
989 return;
990 out_init_req:
991 task->tk_status = 0;
992 task->tk_rqstp = req;
993 xprt_request_init(task, xprt);
996 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
998 if (xprt_dynamic_free_slot(xprt, req))
999 return;
1001 memset(req, 0, sizeof(*req)); /* mark unused */
1003 spin_lock(&xprt->reserve_lock);
1004 list_add(&req->rq_list, &xprt->free);
1005 rpc_wake_up_next(&xprt->backlog);
1006 spin_unlock(&xprt->reserve_lock);
1009 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1011 struct rpc_rqst *req;
1012 while (!list_empty(&xprt->free)) {
1013 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1014 list_del(&req->rq_list);
1015 kfree(req);
1019 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1020 unsigned int num_prealloc,
1021 unsigned int max_alloc)
1023 struct rpc_xprt *xprt;
1024 struct rpc_rqst *req;
1025 int i;
1027 xprt = kzalloc(size, GFP_KERNEL);
1028 if (xprt == NULL)
1029 goto out;
1031 xprt_init(xprt, net);
1033 for (i = 0; i < num_prealloc; i++) {
1034 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1035 if (!req)
1036 break;
1037 list_add(&req->rq_list, &xprt->free);
1039 if (i < num_prealloc)
1040 goto out_free;
1041 if (max_alloc > num_prealloc)
1042 xprt->max_reqs = max_alloc;
1043 else
1044 xprt->max_reqs = num_prealloc;
1045 xprt->min_reqs = num_prealloc;
1046 atomic_set(&xprt->num_reqs, num_prealloc);
1048 return xprt;
1050 out_free:
1051 xprt_free(xprt);
1052 out:
1053 return NULL;
1055 EXPORT_SYMBOL_GPL(xprt_alloc);
1057 void xprt_free(struct rpc_xprt *xprt)
1059 put_net(xprt->xprt_net);
1060 xprt_free_all_slots(xprt);
1061 kfree(xprt);
1063 EXPORT_SYMBOL_GPL(xprt_free);
1066 * xprt_reserve - allocate an RPC request slot
1067 * @task: RPC task requesting a slot allocation
1069 * If no more slots are available, place the task on the transport's
1070 * backlog queue.
1072 void xprt_reserve(struct rpc_task *task)
1074 struct rpc_xprt *xprt = task->tk_xprt;
1076 task->tk_status = 0;
1077 if (task->tk_rqstp != NULL)
1078 return;
1080 /* Note: grabbing the xprt_lock_write() here is not strictly needed,
1081 * but ensures that we throttle new slot allocation if the transport
1082 * is congested (e.g. if reconnecting or if we're out of socket
1083 * write buffer space).
1085 task->tk_timeout = 0;
1086 task->tk_status = -EAGAIN;
1087 if (!xprt_lock_write(xprt, task))
1088 return;
1090 spin_lock(&xprt->reserve_lock);
1091 xprt_alloc_slot(task);
1092 spin_unlock(&xprt->reserve_lock);
1093 xprt_release_write(xprt, task);
1096 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1098 return (__force __be32)xprt->xid++;
1101 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1103 xprt->xid = net_random();
1106 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1108 struct rpc_rqst *req = task->tk_rqstp;
1110 INIT_LIST_HEAD(&req->rq_list);
1111 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1112 req->rq_task = task;
1113 req->rq_xprt = xprt;
1114 req->rq_buffer = NULL;
1115 req->rq_xid = xprt_alloc_xid(xprt);
1116 req->rq_release_snd_buf = NULL;
1117 xprt_reset_majortimeo(req);
1118 dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1119 req, ntohl(req->rq_xid));
1123 * xprt_release - release an RPC request slot
1124 * @task: task which is finished with the slot
1127 void xprt_release(struct rpc_task *task)
1129 struct rpc_xprt *xprt;
1130 struct rpc_rqst *req;
1132 if (!(req = task->tk_rqstp))
1133 return;
1135 xprt = req->rq_xprt;
1136 rpc_count_iostats(task);
1137 spin_lock_bh(&xprt->transport_lock);
1138 xprt->ops->release_xprt(xprt, task);
1139 if (xprt->ops->release_request)
1140 xprt->ops->release_request(task);
1141 if (!list_empty(&req->rq_list))
1142 list_del(&req->rq_list);
1143 xprt->last_used = jiffies;
1144 if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1145 mod_timer(&xprt->timer,
1146 xprt->last_used + xprt->idle_timeout);
1147 spin_unlock_bh(&xprt->transport_lock);
1148 if (req->rq_buffer)
1149 xprt->ops->buf_free(req->rq_buffer);
1150 if (req->rq_cred != NULL)
1151 put_rpccred(req->rq_cred);
1152 task->tk_rqstp = NULL;
1153 if (req->rq_release_snd_buf)
1154 req->rq_release_snd_buf(req);
1156 dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1157 if (likely(!bc_prealloc(req)))
1158 xprt_free_slot(xprt, req);
1159 else
1160 xprt_free_bc_request(req);
1163 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1165 atomic_set(&xprt->count, 1);
1167 spin_lock_init(&xprt->transport_lock);
1168 spin_lock_init(&xprt->reserve_lock);
1170 INIT_LIST_HEAD(&xprt->free);
1171 INIT_LIST_HEAD(&xprt->recv);
1172 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1173 spin_lock_init(&xprt->bc_pa_lock);
1174 INIT_LIST_HEAD(&xprt->bc_pa_list);
1175 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1177 xprt->last_used = jiffies;
1178 xprt->cwnd = RPC_INITCWND;
1179 xprt->bind_index = 0;
1181 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1182 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1183 rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1184 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1186 xprt_init_xid(xprt);
1188 xprt->xprt_net = get_net(net);
1192 * xprt_create_transport - create an RPC transport
1193 * @args: rpc transport creation arguments
1196 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1198 struct rpc_xprt *xprt;
1199 struct xprt_class *t;
1201 spin_lock(&xprt_list_lock);
1202 list_for_each_entry(t, &xprt_list, list) {
1203 if (t->ident == args->ident) {
1204 spin_unlock(&xprt_list_lock);
1205 goto found;
1208 spin_unlock(&xprt_list_lock);
1209 printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1210 return ERR_PTR(-EIO);
1212 found:
1213 xprt = t->setup(args);
1214 if (IS_ERR(xprt)) {
1215 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1216 -PTR_ERR(xprt));
1217 goto out;
1219 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1220 if (xprt_has_timer(xprt))
1221 setup_timer(&xprt->timer, xprt_init_autodisconnect,
1222 (unsigned long)xprt);
1223 else
1224 init_timer(&xprt->timer);
1225 dprintk("RPC: created transport %p with %u slots\n", xprt,
1226 xprt->max_reqs);
1227 out:
1228 return xprt;
1232 * xprt_destroy - destroy an RPC transport, killing off all requests.
1233 * @xprt: transport to destroy
1236 static void xprt_destroy(struct rpc_xprt *xprt)
1238 dprintk("RPC: destroying transport %p\n", xprt);
1239 xprt->shutdown = 1;
1240 del_timer_sync(&xprt->timer);
1242 rpc_destroy_wait_queue(&xprt->binding);
1243 rpc_destroy_wait_queue(&xprt->pending);
1244 rpc_destroy_wait_queue(&xprt->sending);
1245 rpc_destroy_wait_queue(&xprt->backlog);
1246 cancel_work_sync(&xprt->task_cleanup);
1248 * Tear down transport state and free the rpc_xprt
1250 xprt->ops->destroy(xprt);
1254 * xprt_put - release a reference to an RPC transport.
1255 * @xprt: pointer to the transport
1258 void xprt_put(struct rpc_xprt *xprt)
1260 if (atomic_dec_and_test(&xprt->count))
1261 xprt_destroy(xprt);
1265 * xprt_get - return a reference to an RPC transport.
1266 * @xprt: pointer to the transport
1269 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1271 if (atomic_inc_not_zero(&xprt->count))
1272 return xprt;
1273 return NULL;