Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/xprt.c
4 *
5 * This is a generic RPC call interface supporting congestion avoidance,
6 * and asynchronous calls.
7 *
8 * The interface works like this:
9 *
10 * - When a process places a call, it allocates a request slot if
11 * one is available. Otherwise, it sleeps on the backlog queue
12 * (xprt_reserve).
13 * - Next, the caller puts together the RPC message, stuffs it into
14 * the request struct, and calls xprt_transmit().
15 * - xprt_transmit sends the message and installs the caller on the
16 * transport's wait list. At the same time, if a reply is expected,
17 * it installs a timer that is run after the packet's timeout has
18 * expired.
19 * - When a packet arrives, the data_ready handler walks the list of
20 * pending requests for that transport. If a matching XID is found, the
21 * caller is woken up, and the timer removed.
22 * - When no reply arrives within the timeout interval, the timer is
23 * fired by the kernel and runs xprt_timer(). It either adjusts the
24 * timeout values (minor timeout) or wakes up the caller with a status
25 * of -ETIMEDOUT.
26 * - When the caller receives a notification from RPC that a reply arrived,
27 * it should release the RPC slot, and process the reply.
28 * If the call timed out, it may choose to retry the operation by
29 * adjusting the initial timeout value, and simply calling rpc_call
30 * again.
31 *
32 * Support for async RPC is done through a set of RPC-specific scheduling
33 * primitives that `transparently' work for processes as well as async
34 * tasks that rely on callbacks.
35 *
36 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
37 *
38 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
39 */
40
41#include <linux/module.h>
42
43#include <linux/types.h>
44#include <linux/interrupt.h>
45#include <linux/workqueue.h>
46#include <linux/net.h>
47#include <linux/ktime.h>
48
49#include <linux/sunrpc/clnt.h>
50#include <linux/sunrpc/metrics.h>
51#include <linux/sunrpc/bc_xprt.h>
52#include <linux/rcupdate.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "sunrpc.h"
58
59/*
60 * Local variables
61 */
62
63#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
64# define RPCDBG_FACILITY RPCDBG_XPRT
65#endif
66
67/*
68 * Local functions
69 */
70static void xprt_init(struct rpc_xprt *xprt, struct net *net);
71static __be32 xprt_alloc_xid(struct rpc_xprt *xprt);
72static void xprt_destroy(struct rpc_xprt *xprt);
73
74static DEFINE_SPINLOCK(xprt_list_lock);
75static LIST_HEAD(xprt_list);
76
77static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
78{
79 unsigned long timeout = jiffies + req->rq_timeout;
80
81 if (time_before(timeout, req->rq_majortimeo))
82 return timeout;
83 return req->rq_majortimeo;
84}
85
86/**
87 * xprt_register_transport - register a transport implementation
88 * @transport: transport to register
89 *
90 * If a transport implementation is loaded as a kernel module, it can
91 * call this interface to make itself known to the RPC client.
92 *
93 * Returns:
94 * 0: transport successfully registered
95 * -EEXIST: transport already registered
96 * -EINVAL: transport module being unloaded
97 */
98int xprt_register_transport(struct xprt_class *transport)
99{
100 struct xprt_class *t;
101 int result;
102
103 result = -EEXIST;
104 spin_lock(&xprt_list_lock);
105 list_for_each_entry(t, &xprt_list, list) {
106 /* don't register the same transport class twice */
107 if (t->ident == transport->ident)
108 goto out;
109 }
110
111 list_add_tail(&transport->list, &xprt_list);
112 printk(KERN_INFO "RPC: Registered %s transport module.\n",
113 transport->name);
114 result = 0;
115
116out:
117 spin_unlock(&xprt_list_lock);
118 return result;
119}
120EXPORT_SYMBOL_GPL(xprt_register_transport);
121
122/**
123 * xprt_unregister_transport - unregister a transport implementation
124 * @transport: transport to unregister
125 *
126 * Returns:
127 * 0: transport successfully unregistered
128 * -ENOENT: transport never registered
129 */
130int xprt_unregister_transport(struct xprt_class *transport)
131{
132 struct xprt_class *t;
133 int result;
134
135 result = 0;
136 spin_lock(&xprt_list_lock);
137 list_for_each_entry(t, &xprt_list, list) {
138 if (t == transport) {
139 printk(KERN_INFO
140 "RPC: Unregistered %s transport module.\n",
141 transport->name);
142 list_del_init(&transport->list);
143 goto out;
144 }
145 }
146 result = -ENOENT;
147
148out:
149 spin_unlock(&xprt_list_lock);
150 return result;
151}
152EXPORT_SYMBOL_GPL(xprt_unregister_transport);
153
154/**
155 * xprt_load_transport - load a transport implementation
156 * @transport_name: transport to load
157 *
158 * Returns:
159 * 0: transport successfully loaded
160 * -ENOENT: transport module not available
161 */
162int xprt_load_transport(const char *transport_name)
163{
164 struct xprt_class *t;
165 int result;
166
167 result = 0;
168 spin_lock(&xprt_list_lock);
169 list_for_each_entry(t, &xprt_list, list) {
170 if (strcmp(t->name, transport_name) == 0) {
171 spin_unlock(&xprt_list_lock);
172 goto out;
173 }
174 }
175 spin_unlock(&xprt_list_lock);
176 result = request_module("xprt%s", transport_name);
177out:
178 return result;
179}
180EXPORT_SYMBOL_GPL(xprt_load_transport);
181
182static void xprt_clear_locked(struct rpc_xprt *xprt)
183{
184 xprt->snd_task = NULL;
185 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
186 smp_mb__before_atomic();
187 clear_bit(XPRT_LOCKED, &xprt->state);
188 smp_mb__after_atomic();
189 } else
190 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
191}
192
193/**
194 * xprt_reserve_xprt - serialize write access to transports
195 * @task: task that is requesting access to the transport
196 * @xprt: pointer to the target transport
197 *
198 * This prevents mixing the payload of separate requests, and prevents
199 * transport connects from colliding with writes. No congestion control
200 * is provided.
201 */
202int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
203{
204 struct rpc_rqst *req = task->tk_rqstp;
205
206 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
207 if (task == xprt->snd_task)
208 return 1;
209 goto out_sleep;
210 }
211 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
212 goto out_unlock;
213 xprt->snd_task = task;
214
215 return 1;
216
217out_unlock:
218 xprt_clear_locked(xprt);
219out_sleep:
220 dprintk("RPC: %5u failed to lock transport %p\n",
221 task->tk_pid, xprt);
222 task->tk_status = -EAGAIN;
223 if (RPC_IS_SOFT(task))
224 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
225 xprt_request_timeout(req));
226 else
227 rpc_sleep_on(&xprt->sending, task, NULL);
228 return 0;
229}
230EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
231
232static bool
233xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
234{
235 return test_bit(XPRT_CWND_WAIT, &xprt->state);
236}
237
238static void
239xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
240{
241 if (!list_empty(&xprt->xmit_queue)) {
242 /* Peek at head of queue to see if it can make progress */
243 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
244 rq_xmit)->rq_cong)
245 return;
246 }
247 set_bit(XPRT_CWND_WAIT, &xprt->state);
248}
249
250static void
251xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
252{
253 if (!RPCXPRT_CONGESTED(xprt))
254 clear_bit(XPRT_CWND_WAIT, &xprt->state);
255}
256
257/*
258 * xprt_reserve_xprt_cong - serialize write access to transports
259 * @task: task that is requesting access to the transport
260 *
261 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
262 * integrated into the decision of whether a request is allowed to be
263 * woken up and given access to the transport.
264 * Note that the lock is only granted if we know there are free slots.
265 */
266int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
267{
268 struct rpc_rqst *req = task->tk_rqstp;
269
270 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
271 if (task == xprt->snd_task)
272 return 1;
273 goto out_sleep;
274 }
275 if (req == NULL) {
276 xprt->snd_task = task;
277 return 1;
278 }
279 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
280 goto out_unlock;
281 if (!xprt_need_congestion_window_wait(xprt)) {
282 xprt->snd_task = task;
283 return 1;
284 }
285out_unlock:
286 xprt_clear_locked(xprt);
287out_sleep:
288 dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
289 task->tk_status = -EAGAIN;
290 if (RPC_IS_SOFT(task))
291 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
292 xprt_request_timeout(req));
293 else
294 rpc_sleep_on(&xprt->sending, task, NULL);
295 return 0;
296}
297EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
298
299static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
300{
301 int retval;
302
303 if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
304 return 1;
305 spin_lock(&xprt->transport_lock);
306 retval = xprt->ops->reserve_xprt(xprt, task);
307 spin_unlock(&xprt->transport_lock);
308 return retval;
309}
310
311static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
312{
313 struct rpc_xprt *xprt = data;
314
315 xprt->snd_task = task;
316 return true;
317}
318
319static void __xprt_lock_write_next(struct rpc_xprt *xprt)
320{
321 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
322 return;
323 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
324 goto out_unlock;
325 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
326 __xprt_lock_write_func, xprt))
327 return;
328out_unlock:
329 xprt_clear_locked(xprt);
330}
331
332static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
333{
334 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
335 return;
336 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
337 goto out_unlock;
338 if (xprt_need_congestion_window_wait(xprt))
339 goto out_unlock;
340 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
341 __xprt_lock_write_func, xprt))
342 return;
343out_unlock:
344 xprt_clear_locked(xprt);
345}
346
347/**
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
351 *
352 * Note that "task" can be NULL. No congestion control is provided.
353 */
354void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
355{
356 if (xprt->snd_task == task) {
357 xprt_clear_locked(xprt);
358 __xprt_lock_write_next(xprt);
359 }
360}
361EXPORT_SYMBOL_GPL(xprt_release_xprt);
362
363/**
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
367 *
368 * Note that "task" can be NULL. Another task is awoken to use the
369 * transport if the transport's congestion window allows it.
370 */
371void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
372{
373 if (xprt->snd_task == task) {
374 xprt_clear_locked(xprt);
375 __xprt_lock_write_next_cong(xprt);
376 }
377}
378EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
379
380static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
381{
382 if (xprt->snd_task != task)
383 return;
384 spin_lock(&xprt->transport_lock);
385 xprt->ops->release_xprt(xprt, task);
386 spin_unlock(&xprt->transport_lock);
387}
388
389/*
390 * Van Jacobson congestion avoidance. Check if the congestion window
391 * overflowed. Put the task to sleep if this is the case.
392 */
393static int
394__xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
395{
396 if (req->rq_cong)
397 return 1;
398 dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
399 req->rq_task->tk_pid, xprt->cong, xprt->cwnd);
400 if (RPCXPRT_CONGESTED(xprt)) {
401 xprt_set_congestion_window_wait(xprt);
402 return 0;
403 }
404 req->rq_cong = 1;
405 xprt->cong += RPC_CWNDSCALE;
406 return 1;
407}
408
409/*
410 * Adjust the congestion window, and wake up the next task
411 * that has been sleeping due to congestion
412 */
413static void
414__xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
415{
416 if (!req->rq_cong)
417 return;
418 req->rq_cong = 0;
419 xprt->cong -= RPC_CWNDSCALE;
420 xprt_test_and_clear_congestion_window_wait(xprt);
421 __xprt_lock_write_next_cong(xprt);
422}
423
424/**
425 * xprt_request_get_cong - Request congestion control credits
426 * @xprt: pointer to transport
427 * @req: pointer to RPC request
428 *
429 * Useful for transports that require congestion control.
430 */
431bool
432xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
433{
434 bool ret = false;
435
436 if (req->rq_cong)
437 return true;
438 spin_lock(&xprt->transport_lock);
439 ret = __xprt_get_cong(xprt, req) != 0;
440 spin_unlock(&xprt->transport_lock);
441 return ret;
442}
443EXPORT_SYMBOL_GPL(xprt_request_get_cong);
444
445/**
446 * xprt_release_rqst_cong - housekeeping when request is complete
447 * @task: RPC request that recently completed
448 *
449 * Useful for transports that require congestion control.
450 */
451void xprt_release_rqst_cong(struct rpc_task *task)
452{
453 struct rpc_rqst *req = task->tk_rqstp;
454
455 __xprt_put_cong(req->rq_xprt, req);
456}
457EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
458
459static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
460{
461 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
462 __xprt_lock_write_next_cong(xprt);
463}
464
465/*
466 * Clear the congestion window wait flag and wake up the next
467 * entry on xprt->sending
468 */
469static void
470xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
471{
472 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
473 spin_lock(&xprt->transport_lock);
474 __xprt_lock_write_next_cong(xprt);
475 spin_unlock(&xprt->transport_lock);
476 }
477}
478
479/**
480 * xprt_adjust_cwnd - adjust transport congestion window
481 * @xprt: pointer to xprt
482 * @task: recently completed RPC request used to adjust window
483 * @result: result code of completed RPC request
484 *
485 * The transport code maintains an estimate on the maximum number of out-
486 * standing RPC requests, using a smoothed version of the congestion
487 * avoidance implemented in 44BSD. This is basically the Van Jacobson
488 * congestion algorithm: If a retransmit occurs, the congestion window is
489 * halved; otherwise, it is incremented by 1/cwnd when
490 *
491 * - a reply is received and
492 * - a full number of requests are outstanding and
493 * - the congestion window hasn't been updated recently.
494 */
495void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
496{
497 struct rpc_rqst *req = task->tk_rqstp;
498 unsigned long cwnd = xprt->cwnd;
499
500 if (result >= 0 && cwnd <= xprt->cong) {
501 /* The (cwnd >> 1) term makes sure
502 * the result gets rounded properly. */
503 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
504 if (cwnd > RPC_MAXCWND(xprt))
505 cwnd = RPC_MAXCWND(xprt);
506 __xprt_lock_write_next_cong(xprt);
507 } else if (result == -ETIMEDOUT) {
508 cwnd >>= 1;
509 if (cwnd < RPC_CWNDSCALE)
510 cwnd = RPC_CWNDSCALE;
511 }
512 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
513 xprt->cong, xprt->cwnd, cwnd);
514 xprt->cwnd = cwnd;
515 __xprt_put_cong(xprt, req);
516}
517EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
518
519/**
520 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
521 * @xprt: transport with waiting tasks
522 * @status: result code to plant in each task before waking it
523 *
524 */
525void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
526{
527 if (status < 0)
528 rpc_wake_up_status(&xprt->pending, status);
529 else
530 rpc_wake_up(&xprt->pending);
531}
532EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
533
534/**
535 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
536 * @xprt: transport
537 *
538 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
539 * we don't in general want to force a socket disconnection due to
540 * an incomplete RPC call transmission.
541 */
542void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
543{
544 set_bit(XPRT_WRITE_SPACE, &xprt->state);
545}
546EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
547
548static bool
549xprt_clear_write_space_locked(struct rpc_xprt *xprt)
550{
551 if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
552 __xprt_lock_write_next(xprt);
553 dprintk("RPC: write space: waking waiting task on "
554 "xprt %p\n", xprt);
555 return true;
556 }
557 return false;
558}
559
560/**
561 * xprt_write_space - wake the task waiting for transport output buffer space
562 * @xprt: transport with waiting tasks
563 *
564 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
565 */
566bool xprt_write_space(struct rpc_xprt *xprt)
567{
568 bool ret;
569
570 if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
571 return false;
572 spin_lock(&xprt->transport_lock);
573 ret = xprt_clear_write_space_locked(xprt);
574 spin_unlock(&xprt->transport_lock);
575 return ret;
576}
577EXPORT_SYMBOL_GPL(xprt_write_space);
578
579static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
580{
581 s64 delta = ktime_to_ns(ktime_get() - abstime);
582 return likely(delta >= 0) ?
583 jiffies - nsecs_to_jiffies(delta) :
584 jiffies + nsecs_to_jiffies(-delta);
585}
586
587static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
588{
589 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
590 unsigned long majortimeo = req->rq_timeout;
591
592 if (to->to_exponential)
593 majortimeo <<= to->to_retries;
594 else
595 majortimeo += to->to_increment * to->to_retries;
596 if (majortimeo > to->to_maxval || majortimeo == 0)
597 majortimeo = to->to_maxval;
598 return majortimeo;
599}
600
601static void xprt_reset_majortimeo(struct rpc_rqst *req)
602{
603 req->rq_majortimeo += xprt_calc_majortimeo(req);
604}
605
606static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
607{
608 unsigned long time_init;
609 struct rpc_xprt *xprt = req->rq_xprt;
610
611 if (likely(xprt && xprt_connected(xprt)))
612 time_init = jiffies;
613 else
614 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
615 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
616 req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
617}
618
619/**
620 * xprt_adjust_timeout - adjust timeout values for next retransmit
621 * @req: RPC request containing parameters to use for the adjustment
622 *
623 */
624int xprt_adjust_timeout(struct rpc_rqst *req)
625{
626 struct rpc_xprt *xprt = req->rq_xprt;
627 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
628 int status = 0;
629
630 if (time_before(jiffies, req->rq_majortimeo)) {
631 if (to->to_exponential)
632 req->rq_timeout <<= 1;
633 else
634 req->rq_timeout += to->to_increment;
635 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
636 req->rq_timeout = to->to_maxval;
637 req->rq_retries++;
638 } else {
639 req->rq_timeout = to->to_initval;
640 req->rq_retries = 0;
641 xprt_reset_majortimeo(req);
642 /* Reset the RTT counters == "slow start" */
643 spin_lock(&xprt->transport_lock);
644 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
645 spin_unlock(&xprt->transport_lock);
646 status = -ETIMEDOUT;
647 }
648
649 if (req->rq_timeout == 0) {
650 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
651 req->rq_timeout = 5 * HZ;
652 }
653 return status;
654}
655
656static void xprt_autoclose(struct work_struct *work)
657{
658 struct rpc_xprt *xprt =
659 container_of(work, struct rpc_xprt, task_cleanup);
660 unsigned int pflags = memalloc_nofs_save();
661
662 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
663 xprt->ops->close(xprt);
664 xprt_release_write(xprt, NULL);
665 wake_up_bit(&xprt->state, XPRT_LOCKED);
666 memalloc_nofs_restore(pflags);
667}
668
669/**
670 * xprt_disconnect_done - mark a transport as disconnected
671 * @xprt: transport to flag for disconnect
672 *
673 */
674void xprt_disconnect_done(struct rpc_xprt *xprt)
675{
676 dprintk("RPC: disconnected transport %p\n", xprt);
677 spin_lock(&xprt->transport_lock);
678 xprt_clear_connected(xprt);
679 xprt_clear_write_space_locked(xprt);
680 xprt_clear_congestion_window_wait_locked(xprt);
681 xprt_wake_pending_tasks(xprt, -ENOTCONN);
682 spin_unlock(&xprt->transport_lock);
683}
684EXPORT_SYMBOL_GPL(xprt_disconnect_done);
685
686/**
687 * xprt_force_disconnect - force a transport to disconnect
688 * @xprt: transport to disconnect
689 *
690 */
691void xprt_force_disconnect(struct rpc_xprt *xprt)
692{
693 /* Don't race with the test_bit() in xprt_clear_locked() */
694 spin_lock(&xprt->transport_lock);
695 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
696 /* Try to schedule an autoclose RPC call */
697 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
698 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
699 else if (xprt->snd_task)
700 rpc_wake_up_queued_task_set_status(&xprt->pending,
701 xprt->snd_task, -ENOTCONN);
702 spin_unlock(&xprt->transport_lock);
703}
704EXPORT_SYMBOL_GPL(xprt_force_disconnect);
705
706static unsigned int
707xprt_connect_cookie(struct rpc_xprt *xprt)
708{
709 return READ_ONCE(xprt->connect_cookie);
710}
711
712static bool
713xprt_request_retransmit_after_disconnect(struct rpc_task *task)
714{
715 struct rpc_rqst *req = task->tk_rqstp;
716 struct rpc_xprt *xprt = req->rq_xprt;
717
718 return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
719 !xprt_connected(xprt);
720}
721
722/**
723 * xprt_conditional_disconnect - force a transport to disconnect
724 * @xprt: transport to disconnect
725 * @cookie: 'connection cookie'
726 *
727 * This attempts to break the connection if and only if 'cookie' matches
728 * the current transport 'connection cookie'. It ensures that we don't
729 * try to break the connection more than once when we need to retransmit
730 * a batch of RPC requests.
731 *
732 */
733void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
734{
735 /* Don't race with the test_bit() in xprt_clear_locked() */
736 spin_lock(&xprt->transport_lock);
737 if (cookie != xprt->connect_cookie)
738 goto out;
739 if (test_bit(XPRT_CLOSING, &xprt->state))
740 goto out;
741 set_bit(XPRT_CLOSE_WAIT, &xprt->state);
742 /* Try to schedule an autoclose RPC call */
743 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
744 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
745 xprt_wake_pending_tasks(xprt, -EAGAIN);
746out:
747 spin_unlock(&xprt->transport_lock);
748}
749
750static bool
751xprt_has_timer(const struct rpc_xprt *xprt)
752{
753 return xprt->idle_timeout != 0;
754}
755
756static void
757xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
758 __must_hold(&xprt->transport_lock)
759{
760 xprt->last_used = jiffies;
761 if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
762 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
763}
764
765static void
766xprt_init_autodisconnect(struct timer_list *t)
767{
768 struct rpc_xprt *xprt = from_timer(xprt, t, timer);
769
770 if (!RB_EMPTY_ROOT(&xprt->recv_queue))
771 return;
772 /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
773 xprt->last_used = jiffies;
774 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
775 return;
776 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
777}
778
779bool xprt_lock_connect(struct rpc_xprt *xprt,
780 struct rpc_task *task,
781 void *cookie)
782{
783 bool ret = false;
784
785 spin_lock(&xprt->transport_lock);
786 if (!test_bit(XPRT_LOCKED, &xprt->state))
787 goto out;
788 if (xprt->snd_task != task)
789 goto out;
790 xprt->snd_task = cookie;
791 ret = true;
792out:
793 spin_unlock(&xprt->transport_lock);
794 return ret;
795}
796
797void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
798{
799 spin_lock(&xprt->transport_lock);
800 if (xprt->snd_task != cookie)
801 goto out;
802 if (!test_bit(XPRT_LOCKED, &xprt->state))
803 goto out;
804 xprt->snd_task =NULL;
805 xprt->ops->release_xprt(xprt, NULL);
806 xprt_schedule_autodisconnect(xprt);
807out:
808 spin_unlock(&xprt->transport_lock);
809 wake_up_bit(&xprt->state, XPRT_LOCKED);
810}
811
812/**
813 * xprt_connect - schedule a transport connect operation
814 * @task: RPC task that is requesting the connect
815 *
816 */
817void xprt_connect(struct rpc_task *task)
818{
819 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
820
821 dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
822 xprt, (xprt_connected(xprt) ? "is" : "is not"));
823
824 if (!xprt_bound(xprt)) {
825 task->tk_status = -EAGAIN;
826 return;
827 }
828 if (!xprt_lock_write(xprt, task))
829 return;
830
831 if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
832 xprt->ops->close(xprt);
833
834 if (!xprt_connected(xprt)) {
835 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
836 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
837 xprt_request_timeout(task->tk_rqstp));
838
839 if (test_bit(XPRT_CLOSING, &xprt->state))
840 return;
841 if (xprt_test_and_set_connecting(xprt))
842 return;
843 /* Race breaker */
844 if (!xprt_connected(xprt)) {
845 xprt->stat.connect_start = jiffies;
846 xprt->ops->connect(xprt, task);
847 } else {
848 xprt_clear_connecting(xprt);
849 task->tk_status = 0;
850 rpc_wake_up_queued_task(&xprt->pending, task);
851 }
852 }
853 xprt_release_write(xprt, task);
854}
855
856/**
857 * xprt_reconnect_delay - compute the wait before scheduling a connect
858 * @xprt: transport instance
859 *
860 */
861unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
862{
863 unsigned long start, now = jiffies;
864
865 start = xprt->stat.connect_start + xprt->reestablish_timeout;
866 if (time_after(start, now))
867 return start - now;
868 return 0;
869}
870EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
871
872/**
873 * xprt_reconnect_backoff - compute the new re-establish timeout
874 * @xprt: transport instance
875 * @init_to: initial reestablish timeout
876 *
877 */
878void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
879{
880 xprt->reestablish_timeout <<= 1;
881 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
882 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
883 if (xprt->reestablish_timeout < init_to)
884 xprt->reestablish_timeout = init_to;
885}
886EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
887
888enum xprt_xid_rb_cmp {
889 XID_RB_EQUAL,
890 XID_RB_LEFT,
891 XID_RB_RIGHT,
892};
893static enum xprt_xid_rb_cmp
894xprt_xid_cmp(__be32 xid1, __be32 xid2)
895{
896 if (xid1 == xid2)
897 return XID_RB_EQUAL;
898 if ((__force u32)xid1 < (__force u32)xid2)
899 return XID_RB_LEFT;
900 return XID_RB_RIGHT;
901}
902
903static struct rpc_rqst *
904xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
905{
906 struct rb_node *n = xprt->recv_queue.rb_node;
907 struct rpc_rqst *req;
908
909 while (n != NULL) {
910 req = rb_entry(n, struct rpc_rqst, rq_recv);
911 switch (xprt_xid_cmp(xid, req->rq_xid)) {
912 case XID_RB_LEFT:
913 n = n->rb_left;
914 break;
915 case XID_RB_RIGHT:
916 n = n->rb_right;
917 break;
918 case XID_RB_EQUAL:
919 return req;
920 }
921 }
922 return NULL;
923}
924
925static void
926xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
927{
928 struct rb_node **p = &xprt->recv_queue.rb_node;
929 struct rb_node *n = NULL;
930 struct rpc_rqst *req;
931
932 while (*p != NULL) {
933 n = *p;
934 req = rb_entry(n, struct rpc_rqst, rq_recv);
935 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
936 case XID_RB_LEFT:
937 p = &n->rb_left;
938 break;
939 case XID_RB_RIGHT:
940 p = &n->rb_right;
941 break;
942 case XID_RB_EQUAL:
943 WARN_ON_ONCE(new != req);
944 return;
945 }
946 }
947 rb_link_node(&new->rq_recv, n, p);
948 rb_insert_color(&new->rq_recv, &xprt->recv_queue);
949}
950
951static void
952xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
953{
954 rb_erase(&req->rq_recv, &xprt->recv_queue);
955}
956
957/**
958 * xprt_lookup_rqst - find an RPC request corresponding to an XID
959 * @xprt: transport on which the original request was transmitted
960 * @xid: RPC XID of incoming reply
961 *
962 * Caller holds xprt->queue_lock.
963 */
964struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
965{
966 struct rpc_rqst *entry;
967
968 entry = xprt_request_rb_find(xprt, xid);
969 if (entry != NULL) {
970 trace_xprt_lookup_rqst(xprt, xid, 0);
971 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
972 return entry;
973 }
974
975 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
976 ntohl(xid));
977 trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
978 xprt->stat.bad_xids++;
979 return NULL;
980}
981EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
982
983static bool
984xprt_is_pinned_rqst(struct rpc_rqst *req)
985{
986 return atomic_read(&req->rq_pin) != 0;
987}
988
989/**
990 * xprt_pin_rqst - Pin a request on the transport receive list
991 * @req: Request to pin
992 *
993 * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
994 * so should be holding xprt->queue_lock.
995 */
996void xprt_pin_rqst(struct rpc_rqst *req)
997{
998 atomic_inc(&req->rq_pin);
999}
1000EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1001
1002/**
1003 * xprt_unpin_rqst - Unpin a request on the transport receive list
1004 * @req: Request to pin
1005 *
1006 * Caller should be holding xprt->queue_lock.
1007 */
1008void xprt_unpin_rqst(struct rpc_rqst *req)
1009{
1010 if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1011 atomic_dec(&req->rq_pin);
1012 return;
1013 }
1014 if (atomic_dec_and_test(&req->rq_pin))
1015 wake_up_var(&req->rq_pin);
1016}
1017EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1018
1019static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1020{
1021 wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1022}
1023
1024static bool
1025xprt_request_data_received(struct rpc_task *task)
1026{
1027 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1028 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1029}
1030
1031static bool
1032xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1033{
1034 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1035 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1036}
1037
1038/**
1039 * xprt_request_enqueue_receive - Add an request to the receive queue
1040 * @task: RPC task
1041 *
1042 */
1043void
1044xprt_request_enqueue_receive(struct rpc_task *task)
1045{
1046 struct rpc_rqst *req = task->tk_rqstp;
1047 struct rpc_xprt *xprt = req->rq_xprt;
1048
1049 if (!xprt_request_need_enqueue_receive(task, req))
1050 return;
1051
1052 xprt_request_prepare(task->tk_rqstp);
1053 spin_lock(&xprt->queue_lock);
1054
1055 /* Update the softirq receive buffer */
1056 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1057 sizeof(req->rq_private_buf));
1058
1059 /* Add request to the receive list */
1060 xprt_request_rb_insert(xprt, req);
1061 set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1062 spin_unlock(&xprt->queue_lock);
1063
1064 /* Turn off autodisconnect */
1065 del_singleshot_timer_sync(&xprt->timer);
1066}
1067
1068/**
1069 * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1070 * @task: RPC task
1071 *
1072 * Caller must hold xprt->queue_lock.
1073 */
1074static void
1075xprt_request_dequeue_receive_locked(struct rpc_task *task)
1076{
1077 struct rpc_rqst *req = task->tk_rqstp;
1078
1079 if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1080 xprt_request_rb_remove(req->rq_xprt, req);
1081}
1082
1083/**
1084 * xprt_update_rtt - Update RPC RTT statistics
1085 * @task: RPC request that recently completed
1086 *
1087 * Caller holds xprt->queue_lock.
1088 */
1089void xprt_update_rtt(struct rpc_task *task)
1090{
1091 struct rpc_rqst *req = task->tk_rqstp;
1092 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1093 unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1094 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1095
1096 if (timer) {
1097 if (req->rq_ntrans == 1)
1098 rpc_update_rtt(rtt, timer, m);
1099 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1100 }
1101}
1102EXPORT_SYMBOL_GPL(xprt_update_rtt);
1103
1104/**
1105 * xprt_complete_rqst - called when reply processing is complete
1106 * @task: RPC request that recently completed
1107 * @copied: actual number of bytes received from the transport
1108 *
1109 * Caller holds xprt->queue_lock.
1110 */
1111void xprt_complete_rqst(struct rpc_task *task, int copied)
1112{
1113 struct rpc_rqst *req = task->tk_rqstp;
1114 struct rpc_xprt *xprt = req->rq_xprt;
1115
1116 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
1117 task->tk_pid, ntohl(req->rq_xid), copied);
1118 trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
1119
1120 xprt->stat.recvs++;
1121
1122 req->rq_private_buf.len = copied;
1123 /* Ensure all writes are done before we update */
1124 /* req->rq_reply_bytes_recvd */
1125 smp_wmb();
1126 req->rq_reply_bytes_recvd = copied;
1127 xprt_request_dequeue_receive_locked(task);
1128 rpc_wake_up_queued_task(&xprt->pending, task);
1129}
1130EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1131
1132static void xprt_timer(struct rpc_task *task)
1133{
1134 struct rpc_rqst *req = task->tk_rqstp;
1135 struct rpc_xprt *xprt = req->rq_xprt;
1136
1137 if (task->tk_status != -ETIMEDOUT)
1138 return;
1139
1140 trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1141 if (!req->rq_reply_bytes_recvd) {
1142 if (xprt->ops->timer)
1143 xprt->ops->timer(xprt, task);
1144 } else
1145 task->tk_status = 0;
1146}
1147
1148/**
1149 * xprt_wait_for_reply_request_def - wait for reply
1150 * @task: pointer to rpc_task
1151 *
1152 * Set a request's retransmit timeout based on the transport's
1153 * default timeout parameters. Used by transports that don't adjust
1154 * the retransmit timeout based on round-trip time estimation,
1155 * and put the task to sleep on the pending queue.
1156 */
1157void xprt_wait_for_reply_request_def(struct rpc_task *task)
1158{
1159 struct rpc_rqst *req = task->tk_rqstp;
1160
1161 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1162 xprt_request_timeout(req));
1163}
1164EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1165
1166/**
1167 * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1168 * @task: pointer to rpc_task
1169 *
1170 * Set a request's retransmit timeout using the RTT estimator,
1171 * and put the task to sleep on the pending queue.
1172 */
1173void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1174{
1175 int timer = task->tk_msg.rpc_proc->p_timer;
1176 struct rpc_clnt *clnt = task->tk_client;
1177 struct rpc_rtt *rtt = clnt->cl_rtt;
1178 struct rpc_rqst *req = task->tk_rqstp;
1179 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1180 unsigned long timeout;
1181
1182 timeout = rpc_calc_rto(rtt, timer);
1183 timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1184 if (timeout > max_timeout || timeout == 0)
1185 timeout = max_timeout;
1186 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1187 jiffies + timeout);
1188}
1189EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1190
1191/**
1192 * xprt_request_wait_receive - wait for the reply to an RPC request
1193 * @task: RPC task about to send a request
1194 *
1195 */
1196void xprt_request_wait_receive(struct rpc_task *task)
1197{
1198 struct rpc_rqst *req = task->tk_rqstp;
1199 struct rpc_xprt *xprt = req->rq_xprt;
1200
1201 if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1202 return;
1203 /*
1204 * Sleep on the pending queue if we're expecting a reply.
1205 * The spinlock ensures atomicity between the test of
1206 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1207 */
1208 spin_lock(&xprt->queue_lock);
1209 if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1210 xprt->ops->wait_for_reply_request(task);
1211 /*
1212 * Send an extra queue wakeup call if the
1213 * connection was dropped in case the call to
1214 * rpc_sleep_on() raced.
1215 */
1216 if (xprt_request_retransmit_after_disconnect(task))
1217 rpc_wake_up_queued_task_set_status(&xprt->pending,
1218 task, -ENOTCONN);
1219 }
1220 spin_unlock(&xprt->queue_lock);
1221}
1222
1223static bool
1224xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1225{
1226 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1227}
1228
1229/**
1230 * xprt_request_enqueue_transmit - queue a task for transmission
1231 * @task: pointer to rpc_task
1232 *
1233 * Add a task to the transmission queue.
1234 */
1235void
1236xprt_request_enqueue_transmit(struct rpc_task *task)
1237{
1238 struct rpc_rqst *pos, *req = task->tk_rqstp;
1239 struct rpc_xprt *xprt = req->rq_xprt;
1240
1241 if (xprt_request_need_enqueue_transmit(task, req)) {
1242 req->rq_bytes_sent = 0;
1243 spin_lock(&xprt->queue_lock);
1244 /*
1245 * Requests that carry congestion control credits are added
1246 * to the head of the list to avoid starvation issues.
1247 */
1248 if (req->rq_cong) {
1249 xprt_clear_congestion_window_wait(xprt);
1250 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1251 if (pos->rq_cong)
1252 continue;
1253 /* Note: req is added _before_ pos */
1254 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1255 INIT_LIST_HEAD(&req->rq_xmit2);
1256 trace_xprt_enq_xmit(task, 1);
1257 goto out;
1258 }
1259 } else if (RPC_IS_SWAPPER(task)) {
1260 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1261 if (pos->rq_cong || pos->rq_bytes_sent)
1262 continue;
1263 if (RPC_IS_SWAPPER(pos->rq_task))
1264 continue;
1265 /* Note: req is added _before_ pos */
1266 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1267 INIT_LIST_HEAD(&req->rq_xmit2);
1268 trace_xprt_enq_xmit(task, 2);
1269 goto out;
1270 }
1271 } else if (!req->rq_seqno) {
1272 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1273 if (pos->rq_task->tk_owner != task->tk_owner)
1274 continue;
1275 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1276 INIT_LIST_HEAD(&req->rq_xmit);
1277 trace_xprt_enq_xmit(task, 3);
1278 goto out;
1279 }
1280 }
1281 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1282 INIT_LIST_HEAD(&req->rq_xmit2);
1283 trace_xprt_enq_xmit(task, 4);
1284out:
1285 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1286 spin_unlock(&xprt->queue_lock);
1287 }
1288}
1289
1290/**
1291 * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1292 * @task: pointer to rpc_task
1293 *
1294 * Remove a task from the transmission queue
1295 * Caller must hold xprt->queue_lock
1296 */
1297static void
1298xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1299{
1300 struct rpc_rqst *req = task->tk_rqstp;
1301
1302 if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1303 return;
1304 if (!list_empty(&req->rq_xmit)) {
1305 list_del(&req->rq_xmit);
1306 if (!list_empty(&req->rq_xmit2)) {
1307 struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1308 struct rpc_rqst, rq_xmit2);
1309 list_del(&req->rq_xmit2);
1310 list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1311 }
1312 } else
1313 list_del(&req->rq_xmit2);
1314}
1315
1316/**
1317 * xprt_request_dequeue_transmit - remove a task from the transmission queue
1318 * @task: pointer to rpc_task
1319 *
1320 * Remove a task from the transmission queue
1321 */
1322static void
1323xprt_request_dequeue_transmit(struct rpc_task *task)
1324{
1325 struct rpc_rqst *req = task->tk_rqstp;
1326 struct rpc_xprt *xprt = req->rq_xprt;
1327
1328 spin_lock(&xprt->queue_lock);
1329 xprt_request_dequeue_transmit_locked(task);
1330 spin_unlock(&xprt->queue_lock);
1331}
1332
1333/**
1334 * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1335 * @task: pointer to rpc_task
1336 *
1337 * Remove a task from the transmit and receive queues, and ensure that
1338 * it is not pinned by the receive work item.
1339 */
1340void
1341xprt_request_dequeue_xprt(struct rpc_task *task)
1342{
1343 struct rpc_rqst *req = task->tk_rqstp;
1344 struct rpc_xprt *xprt = req->rq_xprt;
1345
1346 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1347 test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1348 xprt_is_pinned_rqst(req)) {
1349 spin_lock(&xprt->queue_lock);
1350 xprt_request_dequeue_transmit_locked(task);
1351 xprt_request_dequeue_receive_locked(task);
1352 while (xprt_is_pinned_rqst(req)) {
1353 set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1354 spin_unlock(&xprt->queue_lock);
1355 xprt_wait_on_pinned_rqst(req);
1356 spin_lock(&xprt->queue_lock);
1357 clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1358 }
1359 spin_unlock(&xprt->queue_lock);
1360 }
1361}
1362
1363/**
1364 * xprt_request_prepare - prepare an encoded request for transport
1365 * @req: pointer to rpc_rqst
1366 *
1367 * Calls into the transport layer to do whatever is needed to prepare
1368 * the request for transmission or receive.
1369 */
1370void
1371xprt_request_prepare(struct rpc_rqst *req)
1372{
1373 struct rpc_xprt *xprt = req->rq_xprt;
1374
1375 if (xprt->ops->prepare_request)
1376 xprt->ops->prepare_request(req);
1377}
1378
1379/**
1380 * xprt_request_need_retransmit - Test if a task needs retransmission
1381 * @task: pointer to rpc_task
1382 *
1383 * Test for whether a connection breakage requires the task to retransmit
1384 */
1385bool
1386xprt_request_need_retransmit(struct rpc_task *task)
1387{
1388 return xprt_request_retransmit_after_disconnect(task);
1389}
1390
1391/**
1392 * xprt_prepare_transmit - reserve the transport before sending a request
1393 * @task: RPC task about to send a request
1394 *
1395 */
1396bool xprt_prepare_transmit(struct rpc_task *task)
1397{
1398 struct rpc_rqst *req = task->tk_rqstp;
1399 struct rpc_xprt *xprt = req->rq_xprt;
1400
1401 dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1402
1403 if (!xprt_lock_write(xprt, task)) {
1404 /* Race breaker: someone may have transmitted us */
1405 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1406 rpc_wake_up_queued_task_set_status(&xprt->sending,
1407 task, 0);
1408 return false;
1409
1410 }
1411 return true;
1412}
1413
1414void xprt_end_transmit(struct rpc_task *task)
1415{
1416 xprt_release_write(task->tk_rqstp->rq_xprt, task);
1417}
1418
1419/**
1420 * xprt_request_transmit - send an RPC request on a transport
1421 * @req: pointer to request to transmit
1422 * @snd_task: RPC task that owns the transport lock
1423 *
1424 * This performs the transmission of a single request.
1425 * Note that if the request is not the same as snd_task, then it
1426 * does need to be pinned.
1427 * Returns '0' on success.
1428 */
1429static int
1430xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1431{
1432 struct rpc_xprt *xprt = req->rq_xprt;
1433 struct rpc_task *task = req->rq_task;
1434 unsigned int connect_cookie;
1435 int is_retrans = RPC_WAS_SENT(task);
1436 int status;
1437
1438 if (!req->rq_bytes_sent) {
1439 if (xprt_request_data_received(task)) {
1440 status = 0;
1441 goto out_dequeue;
1442 }
1443 /* Verify that our message lies in the RPCSEC_GSS window */
1444 if (rpcauth_xmit_need_reencode(task)) {
1445 status = -EBADMSG;
1446 goto out_dequeue;
1447 }
1448 if (RPC_SIGNALLED(task)) {
1449 status = -ERESTARTSYS;
1450 goto out_dequeue;
1451 }
1452 }
1453
1454 /*
1455 * Update req->rq_ntrans before transmitting to avoid races with
1456 * xprt_update_rtt(), which needs to know that it is recording a
1457 * reply to the first transmission.
1458 */
1459 req->rq_ntrans++;
1460
1461 connect_cookie = xprt->connect_cookie;
1462 status = xprt->ops->send_request(req);
1463 if (status != 0) {
1464 req->rq_ntrans--;
1465 trace_xprt_transmit(req, status);
1466 return status;
1467 }
1468
1469 if (is_retrans)
1470 task->tk_client->cl_stats->rpcretrans++;
1471
1472 xprt_inject_disconnect(xprt);
1473
1474 task->tk_flags |= RPC_TASK_SENT;
1475 spin_lock(&xprt->transport_lock);
1476
1477 xprt->stat.sends++;
1478 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1479 xprt->stat.bklog_u += xprt->backlog.qlen;
1480 xprt->stat.sending_u += xprt->sending.qlen;
1481 xprt->stat.pending_u += xprt->pending.qlen;
1482 spin_unlock(&xprt->transport_lock);
1483
1484 req->rq_connect_cookie = connect_cookie;
1485out_dequeue:
1486 trace_xprt_transmit(req, status);
1487 xprt_request_dequeue_transmit(task);
1488 rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1489 return status;
1490}
1491
1492/**
1493 * xprt_transmit - send an RPC request on a transport
1494 * @task: controlling RPC task
1495 *
1496 * Attempts to drain the transmit queue. On exit, either the transport
1497 * signalled an error that needs to be handled before transmission can
1498 * resume, or @task finished transmitting, and detected that it already
1499 * received a reply.
1500 */
1501void
1502xprt_transmit(struct rpc_task *task)
1503{
1504 struct rpc_rqst *next, *req = task->tk_rqstp;
1505 struct rpc_xprt *xprt = req->rq_xprt;
1506 int status;
1507
1508 spin_lock(&xprt->queue_lock);
1509 while (!list_empty(&xprt->xmit_queue)) {
1510 next = list_first_entry(&xprt->xmit_queue,
1511 struct rpc_rqst, rq_xmit);
1512 xprt_pin_rqst(next);
1513 spin_unlock(&xprt->queue_lock);
1514 status = xprt_request_transmit(next, task);
1515 if (status == -EBADMSG && next != req)
1516 status = 0;
1517 cond_resched();
1518 spin_lock(&xprt->queue_lock);
1519 xprt_unpin_rqst(next);
1520 if (status == 0) {
1521 if (!xprt_request_data_received(task) ||
1522 test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1523 continue;
1524 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1525 task->tk_status = status;
1526 break;
1527 }
1528 spin_unlock(&xprt->queue_lock);
1529}
1530
1531static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1532{
1533 set_bit(XPRT_CONGESTED, &xprt->state);
1534 rpc_sleep_on(&xprt->backlog, task, NULL);
1535}
1536
1537static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1538{
1539 if (rpc_wake_up_next(&xprt->backlog) == NULL)
1540 clear_bit(XPRT_CONGESTED, &xprt->state);
1541}
1542
1543static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1544{
1545 bool ret = false;
1546
1547 if (!test_bit(XPRT_CONGESTED, &xprt->state))
1548 goto out;
1549 spin_lock(&xprt->reserve_lock);
1550 if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1551 rpc_sleep_on(&xprt->backlog, task, NULL);
1552 ret = true;
1553 }
1554 spin_unlock(&xprt->reserve_lock);
1555out:
1556 return ret;
1557}
1558
1559static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1560{
1561 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1562
1563 if (xprt->num_reqs >= xprt->max_reqs)
1564 goto out;
1565 ++xprt->num_reqs;
1566 spin_unlock(&xprt->reserve_lock);
1567 req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1568 spin_lock(&xprt->reserve_lock);
1569 if (req != NULL)
1570 goto out;
1571 --xprt->num_reqs;
1572 req = ERR_PTR(-ENOMEM);
1573out:
1574 return req;
1575}
1576
1577static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1578{
1579 if (xprt->num_reqs > xprt->min_reqs) {
1580 --xprt->num_reqs;
1581 kfree(req);
1582 return true;
1583 }
1584 return false;
1585}
1586
1587void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1588{
1589 struct rpc_rqst *req;
1590
1591 spin_lock(&xprt->reserve_lock);
1592 if (!list_empty(&xprt->free)) {
1593 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1594 list_del(&req->rq_list);
1595 goto out_init_req;
1596 }
1597 req = xprt_dynamic_alloc_slot(xprt);
1598 if (!IS_ERR(req))
1599 goto out_init_req;
1600 switch (PTR_ERR(req)) {
1601 case -ENOMEM:
1602 dprintk("RPC: dynamic allocation of request slot "
1603 "failed! Retrying\n");
1604 task->tk_status = -ENOMEM;
1605 break;
1606 case -EAGAIN:
1607 xprt_add_backlog(xprt, task);
1608 dprintk("RPC: waiting for request slot\n");
1609 /* fall through */
1610 default:
1611 task->tk_status = -EAGAIN;
1612 }
1613 spin_unlock(&xprt->reserve_lock);
1614 return;
1615out_init_req:
1616 xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1617 xprt->num_reqs);
1618 spin_unlock(&xprt->reserve_lock);
1619
1620 task->tk_status = 0;
1621 task->tk_rqstp = req;
1622}
1623EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1624
1625void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1626{
1627 spin_lock(&xprt->reserve_lock);
1628 if (!xprt_dynamic_free_slot(xprt, req)) {
1629 memset(req, 0, sizeof(*req)); /* mark unused */
1630 list_add(&req->rq_list, &xprt->free);
1631 }
1632 xprt_wake_up_backlog(xprt);
1633 spin_unlock(&xprt->reserve_lock);
1634}
1635EXPORT_SYMBOL_GPL(xprt_free_slot);
1636
1637static void xprt_free_all_slots(struct rpc_xprt *xprt)
1638{
1639 struct rpc_rqst *req;
1640 while (!list_empty(&xprt->free)) {
1641 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1642 list_del(&req->rq_list);
1643 kfree(req);
1644 }
1645}
1646
1647struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1648 unsigned int num_prealloc,
1649 unsigned int max_alloc)
1650{
1651 struct rpc_xprt *xprt;
1652 struct rpc_rqst *req;
1653 int i;
1654
1655 xprt = kzalloc(size, GFP_KERNEL);
1656 if (xprt == NULL)
1657 goto out;
1658
1659 xprt_init(xprt, net);
1660
1661 for (i = 0; i < num_prealloc; i++) {
1662 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1663 if (!req)
1664 goto out_free;
1665 list_add(&req->rq_list, &xprt->free);
1666 }
1667 if (max_alloc > num_prealloc)
1668 xprt->max_reqs = max_alloc;
1669 else
1670 xprt->max_reqs = num_prealloc;
1671 xprt->min_reqs = num_prealloc;
1672 xprt->num_reqs = num_prealloc;
1673
1674 return xprt;
1675
1676out_free:
1677 xprt_free(xprt);
1678out:
1679 return NULL;
1680}
1681EXPORT_SYMBOL_GPL(xprt_alloc);
1682
1683void xprt_free(struct rpc_xprt *xprt)
1684{
1685 put_net(xprt->xprt_net);
1686 xprt_free_all_slots(xprt);
1687 kfree_rcu(xprt, rcu);
1688}
1689EXPORT_SYMBOL_GPL(xprt_free);
1690
1691static void
1692xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1693{
1694 req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1695}
1696
1697static __be32
1698xprt_alloc_xid(struct rpc_xprt *xprt)
1699{
1700 __be32 xid;
1701
1702 spin_lock(&xprt->reserve_lock);
1703 xid = (__force __be32)xprt->xid++;
1704 spin_unlock(&xprt->reserve_lock);
1705 return xid;
1706}
1707
1708static void
1709xprt_init_xid(struct rpc_xprt *xprt)
1710{
1711 xprt->xid = prandom_u32();
1712}
1713
1714static void
1715xprt_request_init(struct rpc_task *task)
1716{
1717 struct rpc_xprt *xprt = task->tk_xprt;
1718 struct rpc_rqst *req = task->tk_rqstp;
1719
1720 req->rq_task = task;
1721 req->rq_xprt = xprt;
1722 req->rq_buffer = NULL;
1723 req->rq_xid = xprt_alloc_xid(xprt);
1724 xprt_init_connect_cookie(req, xprt);
1725 req->rq_snd_buf.len = 0;
1726 req->rq_snd_buf.buflen = 0;
1727 req->rq_rcv_buf.len = 0;
1728 req->rq_rcv_buf.buflen = 0;
1729 req->rq_snd_buf.bvec = NULL;
1730 req->rq_rcv_buf.bvec = NULL;
1731 req->rq_release_snd_buf = NULL;
1732 xprt_init_majortimeo(task, req);
1733 dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1734 req, ntohl(req->rq_xid));
1735}
1736
1737static void
1738xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1739{
1740 xprt->ops->alloc_slot(xprt, task);
1741 if (task->tk_rqstp != NULL)
1742 xprt_request_init(task);
1743}
1744
1745/**
1746 * xprt_reserve - allocate an RPC request slot
1747 * @task: RPC task requesting a slot allocation
1748 *
1749 * If the transport is marked as being congested, or if no more
1750 * slots are available, place the task on the transport's
1751 * backlog queue.
1752 */
1753void xprt_reserve(struct rpc_task *task)
1754{
1755 struct rpc_xprt *xprt = task->tk_xprt;
1756
1757 task->tk_status = 0;
1758 if (task->tk_rqstp != NULL)
1759 return;
1760
1761 task->tk_status = -EAGAIN;
1762 if (!xprt_throttle_congested(xprt, task))
1763 xprt_do_reserve(xprt, task);
1764}
1765
1766/**
1767 * xprt_retry_reserve - allocate an RPC request slot
1768 * @task: RPC task requesting a slot allocation
1769 *
1770 * If no more slots are available, place the task on the transport's
1771 * backlog queue.
1772 * Note that the only difference with xprt_reserve is that we now
1773 * ignore the value of the XPRT_CONGESTED flag.
1774 */
1775void xprt_retry_reserve(struct rpc_task *task)
1776{
1777 struct rpc_xprt *xprt = task->tk_xprt;
1778
1779 task->tk_status = 0;
1780 if (task->tk_rqstp != NULL)
1781 return;
1782
1783 task->tk_status = -EAGAIN;
1784 xprt_do_reserve(xprt, task);
1785}
1786
1787/**
1788 * xprt_release - release an RPC request slot
1789 * @task: task which is finished with the slot
1790 *
1791 */
1792void xprt_release(struct rpc_task *task)
1793{
1794 struct rpc_xprt *xprt;
1795 struct rpc_rqst *req = task->tk_rqstp;
1796
1797 if (req == NULL) {
1798 if (task->tk_client) {
1799 xprt = task->tk_xprt;
1800 xprt_release_write(xprt, task);
1801 }
1802 return;
1803 }
1804
1805 xprt = req->rq_xprt;
1806 xprt_request_dequeue_xprt(task);
1807 spin_lock(&xprt->transport_lock);
1808 xprt->ops->release_xprt(xprt, task);
1809 if (xprt->ops->release_request)
1810 xprt->ops->release_request(task);
1811 xprt_schedule_autodisconnect(xprt);
1812 spin_unlock(&xprt->transport_lock);
1813 if (req->rq_buffer)
1814 xprt->ops->buf_free(task);
1815 xprt_inject_disconnect(xprt);
1816 xdr_free_bvec(&req->rq_rcv_buf);
1817 xdr_free_bvec(&req->rq_snd_buf);
1818 if (req->rq_cred != NULL)
1819 put_rpccred(req->rq_cred);
1820 task->tk_rqstp = NULL;
1821 if (req->rq_release_snd_buf)
1822 req->rq_release_snd_buf(req);
1823
1824 dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1825 if (likely(!bc_prealloc(req)))
1826 xprt->ops->free_slot(xprt, req);
1827 else
1828 xprt_free_bc_request(req);
1829}
1830
1831#ifdef CONFIG_SUNRPC_BACKCHANNEL
1832void
1833xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1834{
1835 struct xdr_buf *xbufp = &req->rq_snd_buf;
1836
1837 task->tk_rqstp = req;
1838 req->rq_task = task;
1839 xprt_init_connect_cookie(req, req->rq_xprt);
1840 /*
1841 * Set up the xdr_buf length.
1842 * This also indicates that the buffer is XDR encoded already.
1843 */
1844 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1845 xbufp->tail[0].iov_len;
1846}
1847#endif
1848
1849static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1850{
1851 kref_init(&xprt->kref);
1852
1853 spin_lock_init(&xprt->transport_lock);
1854 spin_lock_init(&xprt->reserve_lock);
1855 spin_lock_init(&xprt->queue_lock);
1856
1857 INIT_LIST_HEAD(&xprt->free);
1858 xprt->recv_queue = RB_ROOT;
1859 INIT_LIST_HEAD(&xprt->xmit_queue);
1860#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1861 spin_lock_init(&xprt->bc_pa_lock);
1862 INIT_LIST_HEAD(&xprt->bc_pa_list);
1863#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1864 INIT_LIST_HEAD(&xprt->xprt_switch);
1865
1866 xprt->last_used = jiffies;
1867 xprt->cwnd = RPC_INITCWND;
1868 xprt->bind_index = 0;
1869
1870 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1871 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1872 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1873 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1874
1875 xprt_init_xid(xprt);
1876
1877 xprt->xprt_net = get_net(net);
1878}
1879
1880/**
1881 * xprt_create_transport - create an RPC transport
1882 * @args: rpc transport creation arguments
1883 *
1884 */
1885struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1886{
1887 struct rpc_xprt *xprt;
1888 struct xprt_class *t;
1889
1890 spin_lock(&xprt_list_lock);
1891 list_for_each_entry(t, &xprt_list, list) {
1892 if (t->ident == args->ident) {
1893 spin_unlock(&xprt_list_lock);
1894 goto found;
1895 }
1896 }
1897 spin_unlock(&xprt_list_lock);
1898 dprintk("RPC: transport (%d) not supported\n", args->ident);
1899 return ERR_PTR(-EIO);
1900
1901found:
1902 xprt = t->setup(args);
1903 if (IS_ERR(xprt)) {
1904 dprintk("RPC: xprt_create_transport: failed, %ld\n",
1905 -PTR_ERR(xprt));
1906 goto out;
1907 }
1908 if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1909 xprt->idle_timeout = 0;
1910 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1911 if (xprt_has_timer(xprt))
1912 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1913 else
1914 timer_setup(&xprt->timer, NULL, 0);
1915
1916 if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1917 xprt_destroy(xprt);
1918 return ERR_PTR(-EINVAL);
1919 }
1920 xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1921 if (xprt->servername == NULL) {
1922 xprt_destroy(xprt);
1923 return ERR_PTR(-ENOMEM);
1924 }
1925
1926 rpc_xprt_debugfs_register(xprt);
1927
1928 dprintk("RPC: created transport %p with %u slots\n", xprt,
1929 xprt->max_reqs);
1930out:
1931 return xprt;
1932}
1933
1934static void xprt_destroy_cb(struct work_struct *work)
1935{
1936 struct rpc_xprt *xprt =
1937 container_of(work, struct rpc_xprt, task_cleanup);
1938
1939 rpc_xprt_debugfs_unregister(xprt);
1940 rpc_destroy_wait_queue(&xprt->binding);
1941 rpc_destroy_wait_queue(&xprt->pending);
1942 rpc_destroy_wait_queue(&xprt->sending);
1943 rpc_destroy_wait_queue(&xprt->backlog);
1944 kfree(xprt->servername);
1945 /*
1946 * Destroy any existing back channel
1947 */
1948 xprt_destroy_backchannel(xprt, UINT_MAX);
1949
1950 /*
1951 * Tear down transport state and free the rpc_xprt
1952 */
1953 xprt->ops->destroy(xprt);
1954}
1955
1956/**
1957 * xprt_destroy - destroy an RPC transport, killing off all requests.
1958 * @xprt: transport to destroy
1959 *
1960 */
1961static void xprt_destroy(struct rpc_xprt *xprt)
1962{
1963 dprintk("RPC: destroying transport %p\n", xprt);
1964
1965 /*
1966 * Exclude transport connect/disconnect handlers and autoclose
1967 */
1968 wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1969
1970 del_timer_sync(&xprt->timer);
1971
1972 /*
1973 * Destroy sockets etc from the system workqueue so they can
1974 * safely flush receive work running on rpciod.
1975 */
1976 INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1977 schedule_work(&xprt->task_cleanup);
1978}
1979
1980static void xprt_destroy_kref(struct kref *kref)
1981{
1982 xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1983}
1984
1985/**
1986 * xprt_get - return a reference to an RPC transport.
1987 * @xprt: pointer to the transport
1988 *
1989 */
1990struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1991{
1992 if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1993 return xprt;
1994 return NULL;
1995}
1996EXPORT_SYMBOL_GPL(xprt_get);
1997
1998/**
1999 * xprt_put - release a reference to an RPC transport.
2000 * @xprt: pointer to the transport
2001 *
2002 */
2003void xprt_put(struct rpc_xprt *xprt)
2004{
2005 if (xprt != NULL)
2006 kref_put(&xprt->kref, xprt_destroy_kref);
2007}
2008EXPORT_SYMBOL_GPL(xprt_put);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/xprt.c
4 *
5 * This is a generic RPC call interface supporting congestion avoidance,
6 * and asynchronous calls.
7 *
8 * The interface works like this:
9 *
10 * - When a process places a call, it allocates a request slot if
11 * one is available. Otherwise, it sleeps on the backlog queue
12 * (xprt_reserve).
13 * - Next, the caller puts together the RPC message, stuffs it into
14 * the request struct, and calls xprt_transmit().
15 * - xprt_transmit sends the message and installs the caller on the
16 * transport's wait list. At the same time, if a reply is expected,
17 * it installs a timer that is run after the packet's timeout has
18 * expired.
19 * - When a packet arrives, the data_ready handler walks the list of
20 * pending requests for that transport. If a matching XID is found, the
21 * caller is woken up, and the timer removed.
22 * - When no reply arrives within the timeout interval, the timer is
23 * fired by the kernel and runs xprt_timer(). It either adjusts the
24 * timeout values (minor timeout) or wakes up the caller with a status
25 * of -ETIMEDOUT.
26 * - When the caller receives a notification from RPC that a reply arrived,
27 * it should release the RPC slot, and process the reply.
28 * If the call timed out, it may choose to retry the operation by
29 * adjusting the initial timeout value, and simply calling rpc_call
30 * again.
31 *
32 * Support for async RPC is done through a set of RPC-specific scheduling
33 * primitives that `transparently' work for processes as well as async
34 * tasks that rely on callbacks.
35 *
36 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
37 *
38 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
39 */
40
41#include <linux/module.h>
42
43#include <linux/types.h>
44#include <linux/interrupt.h>
45#include <linux/workqueue.h>
46#include <linux/net.h>
47#include <linux/ktime.h>
48
49#include <linux/sunrpc/clnt.h>
50#include <linux/sunrpc/metrics.h>
51#include <linux/sunrpc/bc_xprt.h>
52#include <linux/rcupdate.h>
53#include <linux/sched/mm.h>
54
55#include <trace/events/sunrpc.h>
56
57#include "sunrpc.h"
58#include "sysfs.h"
59#include "fail.h"
60
61/*
62 * Local variables
63 */
64
65#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
66# define RPCDBG_FACILITY RPCDBG_XPRT
67#endif
68
69/*
70 * Local functions
71 */
72static void xprt_init(struct rpc_xprt *xprt, struct net *net);
73static __be32 xprt_alloc_xid(struct rpc_xprt *xprt);
74static void xprt_destroy(struct rpc_xprt *xprt);
75static void xprt_request_init(struct rpc_task *task);
76static int xprt_request_prepare(struct rpc_rqst *req, struct xdr_buf *buf);
77
78static DEFINE_SPINLOCK(xprt_list_lock);
79static LIST_HEAD(xprt_list);
80
81static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
82{
83 unsigned long timeout = jiffies + req->rq_timeout;
84
85 if (time_before(timeout, req->rq_majortimeo))
86 return timeout;
87 return req->rq_majortimeo;
88}
89
90/**
91 * xprt_register_transport - register a transport implementation
92 * @transport: transport to register
93 *
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.
96 *
97 * Returns:
98 * 0: transport successfully registered
99 * -EEXIST: transport already registered
100 * -EINVAL: transport module being unloaded
101 */
102int xprt_register_transport(struct xprt_class *transport)
103{
104 struct xprt_class *t;
105 int result;
106
107 result = -EEXIST;
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)
112 goto out;
113 }
114
115 list_add_tail(&transport->list, &xprt_list);
116 printk(KERN_INFO "RPC: Registered %s transport module.\n",
117 transport->name);
118 result = 0;
119
120out:
121 spin_unlock(&xprt_list_lock);
122 return result;
123}
124EXPORT_SYMBOL_GPL(xprt_register_transport);
125
126/**
127 * xprt_unregister_transport - unregister a transport implementation
128 * @transport: transport to unregister
129 *
130 * Returns:
131 * 0: transport successfully unregistered
132 * -ENOENT: transport never registered
133 */
134int xprt_unregister_transport(struct xprt_class *transport)
135{
136 struct xprt_class *t;
137 int result;
138
139 result = 0;
140 spin_lock(&xprt_list_lock);
141 list_for_each_entry(t, &xprt_list, list) {
142 if (t == transport) {
143 printk(KERN_INFO
144 "RPC: Unregistered %s transport module.\n",
145 transport->name);
146 list_del_init(&transport->list);
147 goto out;
148 }
149 }
150 result = -ENOENT;
151
152out:
153 spin_unlock(&xprt_list_lock);
154 return result;
155}
156EXPORT_SYMBOL_GPL(xprt_unregister_transport);
157
158static void
159xprt_class_release(const struct xprt_class *t)
160{
161 module_put(t->owner);
162}
163
164static const struct xprt_class *
165xprt_class_find_by_ident_locked(int ident)
166{
167 const struct xprt_class *t;
168
169 list_for_each_entry(t, &xprt_list, list) {
170 if (t->ident != ident)
171 continue;
172 if (!try_module_get(t->owner))
173 continue;
174 return t;
175 }
176 return NULL;
177}
178
179static const struct xprt_class *
180xprt_class_find_by_ident(int ident)
181{
182 const struct xprt_class *t;
183
184 spin_lock(&xprt_list_lock);
185 t = xprt_class_find_by_ident_locked(ident);
186 spin_unlock(&xprt_list_lock);
187 return t;
188}
189
190static const struct xprt_class *
191xprt_class_find_by_netid_locked(const char *netid)
192{
193 const struct xprt_class *t;
194 unsigned int i;
195
196 list_for_each_entry(t, &xprt_list, list) {
197 for (i = 0; t->netid[i][0] != '\0'; i++) {
198 if (strcmp(t->netid[i], netid) != 0)
199 continue;
200 if (!try_module_get(t->owner))
201 continue;
202 return t;
203 }
204 }
205 return NULL;
206}
207
208static const struct xprt_class *
209xprt_class_find_by_netid(const char *netid)
210{
211 const struct xprt_class *t;
212
213 spin_lock(&xprt_list_lock);
214 t = xprt_class_find_by_netid_locked(netid);
215 if (!t) {
216 spin_unlock(&xprt_list_lock);
217 request_module("rpc%s", netid);
218 spin_lock(&xprt_list_lock);
219 t = xprt_class_find_by_netid_locked(netid);
220 }
221 spin_unlock(&xprt_list_lock);
222 return t;
223}
224
225/**
226 * xprt_find_transport_ident - convert a netid into a transport identifier
227 * @netid: transport to load
228 *
229 * Returns:
230 * > 0: transport identifier
231 * -ENOENT: transport module not available
232 */
233int xprt_find_transport_ident(const char *netid)
234{
235 const struct xprt_class *t;
236 int ret;
237
238 t = xprt_class_find_by_netid(netid);
239 if (!t)
240 return -ENOENT;
241 ret = t->ident;
242 xprt_class_release(t);
243 return ret;
244}
245EXPORT_SYMBOL_GPL(xprt_find_transport_ident);
246
247static void xprt_clear_locked(struct rpc_xprt *xprt)
248{
249 xprt->snd_task = NULL;
250 if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state))
251 clear_bit_unlock(XPRT_LOCKED, &xprt->state);
252 else
253 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
254}
255
256/**
257 * xprt_reserve_xprt - serialize write access to transports
258 * @task: task that is requesting access to the transport
259 * @xprt: pointer to the target transport
260 *
261 * This prevents mixing the payload of separate requests, and prevents
262 * transport connects from colliding with writes. No congestion control
263 * is provided.
264 */
265int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
266{
267 struct rpc_rqst *req = task->tk_rqstp;
268
269 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
270 if (task == xprt->snd_task)
271 goto out_locked;
272 goto out_sleep;
273 }
274 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
275 goto out_unlock;
276 xprt->snd_task = task;
277
278out_locked:
279 trace_xprt_reserve_xprt(xprt, task);
280 return 1;
281
282out_unlock:
283 xprt_clear_locked(xprt);
284out_sleep:
285 task->tk_status = -EAGAIN;
286 if (RPC_IS_SOFT(task))
287 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
288 xprt_request_timeout(req));
289 else
290 rpc_sleep_on(&xprt->sending, task, NULL);
291 return 0;
292}
293EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
294
295static bool
296xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
297{
298 return test_bit(XPRT_CWND_WAIT, &xprt->state);
299}
300
301static void
302xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
303{
304 if (!list_empty(&xprt->xmit_queue)) {
305 /* Peek at head of queue to see if it can make progress */
306 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
307 rq_xmit)->rq_cong)
308 return;
309 }
310 set_bit(XPRT_CWND_WAIT, &xprt->state);
311}
312
313static void
314xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
315{
316 if (!RPCXPRT_CONGESTED(xprt))
317 clear_bit(XPRT_CWND_WAIT, &xprt->state);
318}
319
320/*
321 * xprt_reserve_xprt_cong - serialize write access to transports
322 * @task: task that is requesting access to the transport
323 *
324 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
325 * integrated into the decision of whether a request is allowed to be
326 * woken up and given access to the transport.
327 * Note that the lock is only granted if we know there are free slots.
328 */
329int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
330{
331 struct rpc_rqst *req = task->tk_rqstp;
332
333 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
334 if (task == xprt->snd_task)
335 goto out_locked;
336 goto out_sleep;
337 }
338 if (req == NULL) {
339 xprt->snd_task = task;
340 goto out_locked;
341 }
342 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
343 goto out_unlock;
344 if (!xprt_need_congestion_window_wait(xprt)) {
345 xprt->snd_task = task;
346 goto out_locked;
347 }
348out_unlock:
349 xprt_clear_locked(xprt);
350out_sleep:
351 task->tk_status = -EAGAIN;
352 if (RPC_IS_SOFT(task))
353 rpc_sleep_on_timeout(&xprt->sending, task, NULL,
354 xprt_request_timeout(req));
355 else
356 rpc_sleep_on(&xprt->sending, task, NULL);
357 return 0;
358out_locked:
359 trace_xprt_reserve_cong(xprt, task);
360 return 1;
361}
362EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
363
364static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
365{
366 int retval;
367
368 if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
369 return 1;
370 spin_lock(&xprt->transport_lock);
371 retval = xprt->ops->reserve_xprt(xprt, task);
372 spin_unlock(&xprt->transport_lock);
373 return retval;
374}
375
376static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
377{
378 struct rpc_xprt *xprt = data;
379
380 xprt->snd_task = task;
381 return true;
382}
383
384static void __xprt_lock_write_next(struct rpc_xprt *xprt)
385{
386 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
387 return;
388 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
389 goto out_unlock;
390 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
391 __xprt_lock_write_func, xprt))
392 return;
393out_unlock:
394 xprt_clear_locked(xprt);
395}
396
397static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
398{
399 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
400 return;
401 if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
402 goto out_unlock;
403 if (xprt_need_congestion_window_wait(xprt))
404 goto out_unlock;
405 if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
406 __xprt_lock_write_func, xprt))
407 return;
408out_unlock:
409 xprt_clear_locked(xprt);
410}
411
412/**
413 * xprt_release_xprt - allow other requests to use a transport
414 * @xprt: transport with other tasks potentially waiting
415 * @task: task that is releasing access to the transport
416 *
417 * Note that "task" can be NULL. No congestion control is provided.
418 */
419void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
420{
421 if (xprt->snd_task == task) {
422 xprt_clear_locked(xprt);
423 __xprt_lock_write_next(xprt);
424 }
425 trace_xprt_release_xprt(xprt, task);
426}
427EXPORT_SYMBOL_GPL(xprt_release_xprt);
428
429/**
430 * xprt_release_xprt_cong - allow other requests to use a transport
431 * @xprt: transport with other tasks potentially waiting
432 * @task: task that is releasing access to the transport
433 *
434 * Note that "task" can be NULL. Another task is awoken to use the
435 * transport if the transport's congestion window allows it.
436 */
437void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
438{
439 if (xprt->snd_task == task) {
440 xprt_clear_locked(xprt);
441 __xprt_lock_write_next_cong(xprt);
442 }
443 trace_xprt_release_cong(xprt, task);
444}
445EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
446
447void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
448{
449 if (xprt->snd_task != task)
450 return;
451 spin_lock(&xprt->transport_lock);
452 xprt->ops->release_xprt(xprt, task);
453 spin_unlock(&xprt->transport_lock);
454}
455
456/*
457 * Van Jacobson congestion avoidance. Check if the congestion window
458 * overflowed. Put the task to sleep if this is the case.
459 */
460static int
461__xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
462{
463 if (req->rq_cong)
464 return 1;
465 trace_xprt_get_cong(xprt, req->rq_task);
466 if (RPCXPRT_CONGESTED(xprt)) {
467 xprt_set_congestion_window_wait(xprt);
468 return 0;
469 }
470 req->rq_cong = 1;
471 xprt->cong += RPC_CWNDSCALE;
472 return 1;
473}
474
475/*
476 * Adjust the congestion window, and wake up the next task
477 * that has been sleeping due to congestion
478 */
479static void
480__xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
481{
482 if (!req->rq_cong)
483 return;
484 req->rq_cong = 0;
485 xprt->cong -= RPC_CWNDSCALE;
486 xprt_test_and_clear_congestion_window_wait(xprt);
487 trace_xprt_put_cong(xprt, req->rq_task);
488 __xprt_lock_write_next_cong(xprt);
489}
490
491/**
492 * xprt_request_get_cong - Request congestion control credits
493 * @xprt: pointer to transport
494 * @req: pointer to RPC request
495 *
496 * Useful for transports that require congestion control.
497 */
498bool
499xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
500{
501 bool ret = false;
502
503 if (req->rq_cong)
504 return true;
505 spin_lock(&xprt->transport_lock);
506 ret = __xprt_get_cong(xprt, req) != 0;
507 spin_unlock(&xprt->transport_lock);
508 return ret;
509}
510EXPORT_SYMBOL_GPL(xprt_request_get_cong);
511
512/**
513 * xprt_release_rqst_cong - housekeeping when request is complete
514 * @task: RPC request that recently completed
515 *
516 * Useful for transports that require congestion control.
517 */
518void xprt_release_rqst_cong(struct rpc_task *task)
519{
520 struct rpc_rqst *req = task->tk_rqstp;
521
522 __xprt_put_cong(req->rq_xprt, req);
523}
524EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
525
526static void xprt_clear_congestion_window_wait_locked(struct rpc_xprt *xprt)
527{
528 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state))
529 __xprt_lock_write_next_cong(xprt);
530}
531
532/*
533 * Clear the congestion window wait flag and wake up the next
534 * entry on xprt->sending
535 */
536static void
537xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
538{
539 if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
540 spin_lock(&xprt->transport_lock);
541 __xprt_lock_write_next_cong(xprt);
542 spin_unlock(&xprt->transport_lock);
543 }
544}
545
546/**
547 * xprt_adjust_cwnd - adjust transport congestion window
548 * @xprt: pointer to xprt
549 * @task: recently completed RPC request used to adjust window
550 * @result: result code of completed RPC request
551 *
552 * The transport code maintains an estimate on the maximum number of out-
553 * standing RPC requests, using a smoothed version of the congestion
554 * avoidance implemented in 44BSD. This is basically the Van Jacobson
555 * congestion algorithm: If a retransmit occurs, the congestion window is
556 * halved; otherwise, it is incremented by 1/cwnd when
557 *
558 * - a reply is received and
559 * - a full number of requests are outstanding and
560 * - the congestion window hasn't been updated recently.
561 */
562void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
563{
564 struct rpc_rqst *req = task->tk_rqstp;
565 unsigned long cwnd = xprt->cwnd;
566
567 if (result >= 0 && cwnd <= xprt->cong) {
568 /* The (cwnd >> 1) term makes sure
569 * the result gets rounded properly. */
570 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
571 if (cwnd > RPC_MAXCWND(xprt))
572 cwnd = RPC_MAXCWND(xprt);
573 __xprt_lock_write_next_cong(xprt);
574 } else if (result == -ETIMEDOUT) {
575 cwnd >>= 1;
576 if (cwnd < RPC_CWNDSCALE)
577 cwnd = RPC_CWNDSCALE;
578 }
579 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
580 xprt->cong, xprt->cwnd, cwnd);
581 xprt->cwnd = cwnd;
582 __xprt_put_cong(xprt, req);
583}
584EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
585
586/**
587 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
588 * @xprt: transport with waiting tasks
589 * @status: result code to plant in each task before waking it
590 *
591 */
592void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
593{
594 if (status < 0)
595 rpc_wake_up_status(&xprt->pending, status);
596 else
597 rpc_wake_up(&xprt->pending);
598}
599EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
600
601/**
602 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
603 * @xprt: transport
604 *
605 * Note that we only set the timer for the case of RPC_IS_SOFT(), since
606 * we don't in general want to force a socket disconnection due to
607 * an incomplete RPC call transmission.
608 */
609void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
610{
611 set_bit(XPRT_WRITE_SPACE, &xprt->state);
612}
613EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
614
615static bool
616xprt_clear_write_space_locked(struct rpc_xprt *xprt)
617{
618 if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
619 __xprt_lock_write_next(xprt);
620 dprintk("RPC: write space: waking waiting task on "
621 "xprt %p\n", xprt);
622 return true;
623 }
624 return false;
625}
626
627/**
628 * xprt_write_space - wake the task waiting for transport output buffer space
629 * @xprt: transport with waiting tasks
630 *
631 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
632 */
633bool xprt_write_space(struct rpc_xprt *xprt)
634{
635 bool ret;
636
637 if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
638 return false;
639 spin_lock(&xprt->transport_lock);
640 ret = xprt_clear_write_space_locked(xprt);
641 spin_unlock(&xprt->transport_lock);
642 return ret;
643}
644EXPORT_SYMBOL_GPL(xprt_write_space);
645
646static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
647{
648 s64 delta = ktime_to_ns(ktime_get() - abstime);
649 return likely(delta >= 0) ?
650 jiffies - nsecs_to_jiffies(delta) :
651 jiffies + nsecs_to_jiffies(-delta);
652}
653
654static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
655{
656 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
657 unsigned long majortimeo = req->rq_timeout;
658
659 if (to->to_exponential)
660 majortimeo <<= to->to_retries;
661 else
662 majortimeo += to->to_increment * to->to_retries;
663 if (majortimeo > to->to_maxval || majortimeo == 0)
664 majortimeo = to->to_maxval;
665 return majortimeo;
666}
667
668static void xprt_reset_majortimeo(struct rpc_rqst *req)
669{
670 req->rq_majortimeo += xprt_calc_majortimeo(req);
671}
672
673static void xprt_reset_minortimeo(struct rpc_rqst *req)
674{
675 req->rq_minortimeo += req->rq_timeout;
676}
677
678static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
679{
680 unsigned long time_init;
681 struct rpc_xprt *xprt = req->rq_xprt;
682
683 if (likely(xprt && xprt_connected(xprt)))
684 time_init = jiffies;
685 else
686 time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
687 req->rq_timeout = task->tk_client->cl_timeout->to_initval;
688 req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
689 req->rq_minortimeo = time_init + req->rq_timeout;
690}
691
692/**
693 * xprt_adjust_timeout - adjust timeout values for next retransmit
694 * @req: RPC request containing parameters to use for the adjustment
695 *
696 */
697int xprt_adjust_timeout(struct rpc_rqst *req)
698{
699 struct rpc_xprt *xprt = req->rq_xprt;
700 const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
701 int status = 0;
702
703 if (time_before(jiffies, req->rq_majortimeo)) {
704 if (time_before(jiffies, req->rq_minortimeo))
705 return status;
706 if (to->to_exponential)
707 req->rq_timeout <<= 1;
708 else
709 req->rq_timeout += to->to_increment;
710 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
711 req->rq_timeout = to->to_maxval;
712 req->rq_retries++;
713 } else {
714 req->rq_timeout = to->to_initval;
715 req->rq_retries = 0;
716 xprt_reset_majortimeo(req);
717 /* Reset the RTT counters == "slow start" */
718 spin_lock(&xprt->transport_lock);
719 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
720 spin_unlock(&xprt->transport_lock);
721 status = -ETIMEDOUT;
722 }
723 xprt_reset_minortimeo(req);
724
725 if (req->rq_timeout == 0) {
726 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
727 req->rq_timeout = 5 * HZ;
728 }
729 return status;
730}
731
732static void xprt_autoclose(struct work_struct *work)
733{
734 struct rpc_xprt *xprt =
735 container_of(work, struct rpc_xprt, task_cleanup);
736 unsigned int pflags = memalloc_nofs_save();
737
738 trace_xprt_disconnect_auto(xprt);
739 xprt->connect_cookie++;
740 smp_mb__before_atomic();
741 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
742 xprt->ops->close(xprt);
743 xprt_release_write(xprt, NULL);
744 wake_up_bit(&xprt->state, XPRT_LOCKED);
745 memalloc_nofs_restore(pflags);
746}
747
748/**
749 * xprt_disconnect_done - mark a transport as disconnected
750 * @xprt: transport to flag for disconnect
751 *
752 */
753void xprt_disconnect_done(struct rpc_xprt *xprt)
754{
755 trace_xprt_disconnect_done(xprt);
756 spin_lock(&xprt->transport_lock);
757 xprt_clear_connected(xprt);
758 xprt_clear_write_space_locked(xprt);
759 xprt_clear_congestion_window_wait_locked(xprt);
760 xprt_wake_pending_tasks(xprt, -ENOTCONN);
761 spin_unlock(&xprt->transport_lock);
762}
763EXPORT_SYMBOL_GPL(xprt_disconnect_done);
764
765/**
766 * xprt_schedule_autoclose_locked - Try to schedule an autoclose RPC call
767 * @xprt: transport to disconnect
768 */
769static void xprt_schedule_autoclose_locked(struct rpc_xprt *xprt)
770{
771 if (test_and_set_bit(XPRT_CLOSE_WAIT, &xprt->state))
772 return;
773 if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
774 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
775 else if (xprt->snd_task && !test_bit(XPRT_SND_IS_COOKIE, &xprt->state))
776 rpc_wake_up_queued_task_set_status(&xprt->pending,
777 xprt->snd_task, -ENOTCONN);
778}
779
780/**
781 * xprt_force_disconnect - force a transport to disconnect
782 * @xprt: transport to disconnect
783 *
784 */
785void xprt_force_disconnect(struct rpc_xprt *xprt)
786{
787 trace_xprt_disconnect_force(xprt);
788
789 /* Don't race with the test_bit() in xprt_clear_locked() */
790 spin_lock(&xprt->transport_lock);
791 xprt_schedule_autoclose_locked(xprt);
792 spin_unlock(&xprt->transport_lock);
793}
794EXPORT_SYMBOL_GPL(xprt_force_disconnect);
795
796static unsigned int
797xprt_connect_cookie(struct rpc_xprt *xprt)
798{
799 return READ_ONCE(xprt->connect_cookie);
800}
801
802static bool
803xprt_request_retransmit_after_disconnect(struct rpc_task *task)
804{
805 struct rpc_rqst *req = task->tk_rqstp;
806 struct rpc_xprt *xprt = req->rq_xprt;
807
808 return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
809 !xprt_connected(xprt);
810}
811
812/**
813 * xprt_conditional_disconnect - force a transport to disconnect
814 * @xprt: transport to disconnect
815 * @cookie: 'connection cookie'
816 *
817 * This attempts to break the connection if and only if 'cookie' matches
818 * the current transport 'connection cookie'. It ensures that we don't
819 * try to break the connection more than once when we need to retransmit
820 * a batch of RPC requests.
821 *
822 */
823void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
824{
825 /* Don't race with the test_bit() in xprt_clear_locked() */
826 spin_lock(&xprt->transport_lock);
827 if (cookie != xprt->connect_cookie)
828 goto out;
829 if (test_bit(XPRT_CLOSING, &xprt->state))
830 goto out;
831 xprt_schedule_autoclose_locked(xprt);
832out:
833 spin_unlock(&xprt->transport_lock);
834}
835
836static bool
837xprt_has_timer(const struct rpc_xprt *xprt)
838{
839 return xprt->idle_timeout != 0;
840}
841
842static void
843xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
844 __must_hold(&xprt->transport_lock)
845{
846 xprt->last_used = jiffies;
847 if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
848 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
849}
850
851static void
852xprt_init_autodisconnect(struct timer_list *t)
853{
854 struct rpc_xprt *xprt = from_timer(xprt, t, timer);
855
856 if (!RB_EMPTY_ROOT(&xprt->recv_queue))
857 return;
858 /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
859 xprt->last_used = jiffies;
860 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
861 return;
862 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
863}
864
865#if IS_ENABLED(CONFIG_FAIL_SUNRPC)
866static void xprt_inject_disconnect(struct rpc_xprt *xprt)
867{
868 if (!fail_sunrpc.ignore_client_disconnect &&
869 should_fail(&fail_sunrpc.attr, 1))
870 xprt->ops->inject_disconnect(xprt);
871}
872#else
873static inline void xprt_inject_disconnect(struct rpc_xprt *xprt)
874{
875}
876#endif
877
878bool xprt_lock_connect(struct rpc_xprt *xprt,
879 struct rpc_task *task,
880 void *cookie)
881{
882 bool ret = false;
883
884 spin_lock(&xprt->transport_lock);
885 if (!test_bit(XPRT_LOCKED, &xprt->state))
886 goto out;
887 if (xprt->snd_task != task)
888 goto out;
889 set_bit(XPRT_SND_IS_COOKIE, &xprt->state);
890 xprt->snd_task = cookie;
891 ret = true;
892out:
893 spin_unlock(&xprt->transport_lock);
894 return ret;
895}
896EXPORT_SYMBOL_GPL(xprt_lock_connect);
897
898void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
899{
900 spin_lock(&xprt->transport_lock);
901 if (xprt->snd_task != cookie)
902 goto out;
903 if (!test_bit(XPRT_LOCKED, &xprt->state))
904 goto out;
905 xprt->snd_task =NULL;
906 clear_bit(XPRT_SND_IS_COOKIE, &xprt->state);
907 xprt->ops->release_xprt(xprt, NULL);
908 xprt_schedule_autodisconnect(xprt);
909out:
910 spin_unlock(&xprt->transport_lock);
911 wake_up_bit(&xprt->state, XPRT_LOCKED);
912}
913EXPORT_SYMBOL_GPL(xprt_unlock_connect);
914
915/**
916 * xprt_connect - schedule a transport connect operation
917 * @task: RPC task that is requesting the connect
918 *
919 */
920void xprt_connect(struct rpc_task *task)
921{
922 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
923
924 trace_xprt_connect(xprt);
925
926 if (!xprt_bound(xprt)) {
927 task->tk_status = -EAGAIN;
928 return;
929 }
930 if (!xprt_lock_write(xprt, task))
931 return;
932
933 if (!xprt_connected(xprt) && !test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
934 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
935 rpc_sleep_on_timeout(&xprt->pending, task, NULL,
936 xprt_request_timeout(task->tk_rqstp));
937
938 if (test_bit(XPRT_CLOSING, &xprt->state))
939 return;
940 if (xprt_test_and_set_connecting(xprt))
941 return;
942 /* Race breaker */
943 if (!xprt_connected(xprt)) {
944 xprt->stat.connect_start = jiffies;
945 xprt->ops->connect(xprt, task);
946 } else {
947 xprt_clear_connecting(xprt);
948 task->tk_status = 0;
949 rpc_wake_up_queued_task(&xprt->pending, task);
950 }
951 }
952 xprt_release_write(xprt, task);
953}
954
955/**
956 * xprt_reconnect_delay - compute the wait before scheduling a connect
957 * @xprt: transport instance
958 *
959 */
960unsigned long xprt_reconnect_delay(const struct rpc_xprt *xprt)
961{
962 unsigned long start, now = jiffies;
963
964 start = xprt->stat.connect_start + xprt->reestablish_timeout;
965 if (time_after(start, now))
966 return start - now;
967 return 0;
968}
969EXPORT_SYMBOL_GPL(xprt_reconnect_delay);
970
971/**
972 * xprt_reconnect_backoff - compute the new re-establish timeout
973 * @xprt: transport instance
974 * @init_to: initial reestablish timeout
975 *
976 */
977void xprt_reconnect_backoff(struct rpc_xprt *xprt, unsigned long init_to)
978{
979 xprt->reestablish_timeout <<= 1;
980 if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
981 xprt->reestablish_timeout = xprt->max_reconnect_timeout;
982 if (xprt->reestablish_timeout < init_to)
983 xprt->reestablish_timeout = init_to;
984}
985EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
986
987enum xprt_xid_rb_cmp {
988 XID_RB_EQUAL,
989 XID_RB_LEFT,
990 XID_RB_RIGHT,
991};
992static enum xprt_xid_rb_cmp
993xprt_xid_cmp(__be32 xid1, __be32 xid2)
994{
995 if (xid1 == xid2)
996 return XID_RB_EQUAL;
997 if ((__force u32)xid1 < (__force u32)xid2)
998 return XID_RB_LEFT;
999 return XID_RB_RIGHT;
1000}
1001
1002static struct rpc_rqst *
1003xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
1004{
1005 struct rb_node *n = xprt->recv_queue.rb_node;
1006 struct rpc_rqst *req;
1007
1008 while (n != NULL) {
1009 req = rb_entry(n, struct rpc_rqst, rq_recv);
1010 switch (xprt_xid_cmp(xid, req->rq_xid)) {
1011 case XID_RB_LEFT:
1012 n = n->rb_left;
1013 break;
1014 case XID_RB_RIGHT:
1015 n = n->rb_right;
1016 break;
1017 case XID_RB_EQUAL:
1018 return req;
1019 }
1020 }
1021 return NULL;
1022}
1023
1024static void
1025xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
1026{
1027 struct rb_node **p = &xprt->recv_queue.rb_node;
1028 struct rb_node *n = NULL;
1029 struct rpc_rqst *req;
1030
1031 while (*p != NULL) {
1032 n = *p;
1033 req = rb_entry(n, struct rpc_rqst, rq_recv);
1034 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
1035 case XID_RB_LEFT:
1036 p = &n->rb_left;
1037 break;
1038 case XID_RB_RIGHT:
1039 p = &n->rb_right;
1040 break;
1041 case XID_RB_EQUAL:
1042 WARN_ON_ONCE(new != req);
1043 return;
1044 }
1045 }
1046 rb_link_node(&new->rq_recv, n, p);
1047 rb_insert_color(&new->rq_recv, &xprt->recv_queue);
1048}
1049
1050static void
1051xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
1052{
1053 rb_erase(&req->rq_recv, &xprt->recv_queue);
1054}
1055
1056/**
1057 * xprt_lookup_rqst - find an RPC request corresponding to an XID
1058 * @xprt: transport on which the original request was transmitted
1059 * @xid: RPC XID of incoming reply
1060 *
1061 * Caller holds xprt->queue_lock.
1062 */
1063struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
1064{
1065 struct rpc_rqst *entry;
1066
1067 entry = xprt_request_rb_find(xprt, xid);
1068 if (entry != NULL) {
1069 trace_xprt_lookup_rqst(xprt, xid, 0);
1070 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
1071 return entry;
1072 }
1073
1074 dprintk("RPC: xprt_lookup_rqst did not find xid %08x\n",
1075 ntohl(xid));
1076 trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
1077 xprt->stat.bad_xids++;
1078 return NULL;
1079}
1080EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
1081
1082static bool
1083xprt_is_pinned_rqst(struct rpc_rqst *req)
1084{
1085 return atomic_read(&req->rq_pin) != 0;
1086}
1087
1088/**
1089 * xprt_pin_rqst - Pin a request on the transport receive list
1090 * @req: Request to pin
1091 *
1092 * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
1093 * so should be holding xprt->queue_lock.
1094 */
1095void xprt_pin_rqst(struct rpc_rqst *req)
1096{
1097 atomic_inc(&req->rq_pin);
1098}
1099EXPORT_SYMBOL_GPL(xprt_pin_rqst);
1100
1101/**
1102 * xprt_unpin_rqst - Unpin a request on the transport receive list
1103 * @req: Request to pin
1104 *
1105 * Caller should be holding xprt->queue_lock.
1106 */
1107void xprt_unpin_rqst(struct rpc_rqst *req)
1108{
1109 if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
1110 atomic_dec(&req->rq_pin);
1111 return;
1112 }
1113 if (atomic_dec_and_test(&req->rq_pin))
1114 wake_up_var(&req->rq_pin);
1115}
1116EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1117
1118static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
1119{
1120 wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
1121}
1122
1123static bool
1124xprt_request_data_received(struct rpc_task *task)
1125{
1126 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1127 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
1128}
1129
1130static bool
1131xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
1132{
1133 return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
1134 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1135}
1136
1137/**
1138 * xprt_request_enqueue_receive - Add an request to the receive queue
1139 * @task: RPC task
1140 *
1141 */
1142int
1143xprt_request_enqueue_receive(struct rpc_task *task)
1144{
1145 struct rpc_rqst *req = task->tk_rqstp;
1146 struct rpc_xprt *xprt = req->rq_xprt;
1147 int ret;
1148
1149 if (!xprt_request_need_enqueue_receive(task, req))
1150 return 0;
1151
1152 ret = xprt_request_prepare(task->tk_rqstp, &req->rq_rcv_buf);
1153 if (ret)
1154 return ret;
1155 spin_lock(&xprt->queue_lock);
1156
1157 /* Update the softirq receive buffer */
1158 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1159 sizeof(req->rq_private_buf));
1160
1161 /* Add request to the receive list */
1162 xprt_request_rb_insert(xprt, req);
1163 set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1164 spin_unlock(&xprt->queue_lock);
1165
1166 /* Turn off autodisconnect */
1167 del_timer_sync(&xprt->timer);
1168 return 0;
1169}
1170
1171/**
1172 * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1173 * @task: RPC task
1174 *
1175 * Caller must hold xprt->queue_lock.
1176 */
1177static void
1178xprt_request_dequeue_receive_locked(struct rpc_task *task)
1179{
1180 struct rpc_rqst *req = task->tk_rqstp;
1181
1182 if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1183 xprt_request_rb_remove(req->rq_xprt, req);
1184}
1185
1186/**
1187 * xprt_update_rtt - Update RPC RTT statistics
1188 * @task: RPC request that recently completed
1189 *
1190 * Caller holds xprt->queue_lock.
1191 */
1192void xprt_update_rtt(struct rpc_task *task)
1193{
1194 struct rpc_rqst *req = task->tk_rqstp;
1195 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1196 unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1197 long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1198
1199 if (timer) {
1200 if (req->rq_ntrans == 1)
1201 rpc_update_rtt(rtt, timer, m);
1202 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1203 }
1204}
1205EXPORT_SYMBOL_GPL(xprt_update_rtt);
1206
1207/**
1208 * xprt_complete_rqst - called when reply processing is complete
1209 * @task: RPC request that recently completed
1210 * @copied: actual number of bytes received from the transport
1211 *
1212 * Caller holds xprt->queue_lock.
1213 */
1214void xprt_complete_rqst(struct rpc_task *task, int copied)
1215{
1216 struct rpc_rqst *req = task->tk_rqstp;
1217 struct rpc_xprt *xprt = req->rq_xprt;
1218
1219 xprt->stat.recvs++;
1220
1221 xdr_free_bvec(&req->rq_rcv_buf);
1222 req->rq_private_buf.bvec = NULL;
1223 req->rq_private_buf.len = copied;
1224 /* Ensure all writes are done before we update */
1225 /* req->rq_reply_bytes_recvd */
1226 smp_wmb();
1227 req->rq_reply_bytes_recvd = copied;
1228 xprt_request_dequeue_receive_locked(task);
1229 rpc_wake_up_queued_task(&xprt->pending, task);
1230}
1231EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1232
1233static void xprt_timer(struct rpc_task *task)
1234{
1235 struct rpc_rqst *req = task->tk_rqstp;
1236 struct rpc_xprt *xprt = req->rq_xprt;
1237
1238 if (task->tk_status != -ETIMEDOUT)
1239 return;
1240
1241 trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1242 if (!req->rq_reply_bytes_recvd) {
1243 if (xprt->ops->timer)
1244 xprt->ops->timer(xprt, task);
1245 } else
1246 task->tk_status = 0;
1247}
1248
1249/**
1250 * xprt_wait_for_reply_request_def - wait for reply
1251 * @task: pointer to rpc_task
1252 *
1253 * Set a request's retransmit timeout based on the transport's
1254 * default timeout parameters. Used by transports that don't adjust
1255 * the retransmit timeout based on round-trip time estimation,
1256 * and put the task to sleep on the pending queue.
1257 */
1258void xprt_wait_for_reply_request_def(struct rpc_task *task)
1259{
1260 struct rpc_rqst *req = task->tk_rqstp;
1261
1262 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1263 xprt_request_timeout(req));
1264}
1265EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1266
1267/**
1268 * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1269 * @task: pointer to rpc_task
1270 *
1271 * Set a request's retransmit timeout using the RTT estimator,
1272 * and put the task to sleep on the pending queue.
1273 */
1274void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1275{
1276 int timer = task->tk_msg.rpc_proc->p_timer;
1277 struct rpc_clnt *clnt = task->tk_client;
1278 struct rpc_rtt *rtt = clnt->cl_rtt;
1279 struct rpc_rqst *req = task->tk_rqstp;
1280 unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1281 unsigned long timeout;
1282
1283 timeout = rpc_calc_rto(rtt, timer);
1284 timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1285 if (timeout > max_timeout || timeout == 0)
1286 timeout = max_timeout;
1287 rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1288 jiffies + timeout);
1289}
1290EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1291
1292/**
1293 * xprt_request_wait_receive - wait for the reply to an RPC request
1294 * @task: RPC task about to send a request
1295 *
1296 */
1297void xprt_request_wait_receive(struct rpc_task *task)
1298{
1299 struct rpc_rqst *req = task->tk_rqstp;
1300 struct rpc_xprt *xprt = req->rq_xprt;
1301
1302 if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1303 return;
1304 /*
1305 * Sleep on the pending queue if we're expecting a reply.
1306 * The spinlock ensures atomicity between the test of
1307 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1308 */
1309 spin_lock(&xprt->queue_lock);
1310 if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1311 xprt->ops->wait_for_reply_request(task);
1312 /*
1313 * Send an extra queue wakeup call if the
1314 * connection was dropped in case the call to
1315 * rpc_sleep_on() raced.
1316 */
1317 if (xprt_request_retransmit_after_disconnect(task))
1318 rpc_wake_up_queued_task_set_status(&xprt->pending,
1319 task, -ENOTCONN);
1320 }
1321 spin_unlock(&xprt->queue_lock);
1322}
1323
1324static bool
1325xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1326{
1327 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1328}
1329
1330/**
1331 * xprt_request_enqueue_transmit - queue a task for transmission
1332 * @task: pointer to rpc_task
1333 *
1334 * Add a task to the transmission queue.
1335 */
1336void
1337xprt_request_enqueue_transmit(struct rpc_task *task)
1338{
1339 struct rpc_rqst *pos, *req = task->tk_rqstp;
1340 struct rpc_xprt *xprt = req->rq_xprt;
1341 int ret;
1342
1343 if (xprt_request_need_enqueue_transmit(task, req)) {
1344 ret = xprt_request_prepare(task->tk_rqstp, &req->rq_snd_buf);
1345 if (ret) {
1346 task->tk_status = ret;
1347 return;
1348 }
1349 req->rq_bytes_sent = 0;
1350 spin_lock(&xprt->queue_lock);
1351 /*
1352 * Requests that carry congestion control credits are added
1353 * to the head of the list to avoid starvation issues.
1354 */
1355 if (req->rq_cong) {
1356 xprt_clear_congestion_window_wait(xprt);
1357 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1358 if (pos->rq_cong)
1359 continue;
1360 /* Note: req is added _before_ pos */
1361 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1362 INIT_LIST_HEAD(&req->rq_xmit2);
1363 goto out;
1364 }
1365 } else if (!req->rq_seqno) {
1366 list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1367 if (pos->rq_task->tk_owner != task->tk_owner)
1368 continue;
1369 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1370 INIT_LIST_HEAD(&req->rq_xmit);
1371 goto out;
1372 }
1373 }
1374 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1375 INIT_LIST_HEAD(&req->rq_xmit2);
1376out:
1377 atomic_long_inc(&xprt->xmit_queuelen);
1378 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1379 spin_unlock(&xprt->queue_lock);
1380 }
1381}
1382
1383/**
1384 * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1385 * @task: pointer to rpc_task
1386 *
1387 * Remove a task from the transmission queue
1388 * Caller must hold xprt->queue_lock
1389 */
1390static void
1391xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1392{
1393 struct rpc_rqst *req = task->tk_rqstp;
1394
1395 if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1396 return;
1397 if (!list_empty(&req->rq_xmit)) {
1398 list_del(&req->rq_xmit);
1399 if (!list_empty(&req->rq_xmit2)) {
1400 struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1401 struct rpc_rqst, rq_xmit2);
1402 list_del(&req->rq_xmit2);
1403 list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1404 }
1405 } else
1406 list_del(&req->rq_xmit2);
1407 atomic_long_dec(&req->rq_xprt->xmit_queuelen);
1408 xdr_free_bvec(&req->rq_snd_buf);
1409}
1410
1411/**
1412 * xprt_request_dequeue_transmit - remove a task from the transmission queue
1413 * @task: pointer to rpc_task
1414 *
1415 * Remove a task from the transmission queue
1416 */
1417static void
1418xprt_request_dequeue_transmit(struct rpc_task *task)
1419{
1420 struct rpc_rqst *req = task->tk_rqstp;
1421 struct rpc_xprt *xprt = req->rq_xprt;
1422
1423 spin_lock(&xprt->queue_lock);
1424 xprt_request_dequeue_transmit_locked(task);
1425 spin_unlock(&xprt->queue_lock);
1426}
1427
1428/**
1429 * xprt_request_dequeue_xprt - remove a task from the transmit+receive queue
1430 * @task: pointer to rpc_task
1431 *
1432 * Remove a task from the transmit and receive queues, and ensure that
1433 * it is not pinned by the receive work item.
1434 */
1435void
1436xprt_request_dequeue_xprt(struct rpc_task *task)
1437{
1438 struct rpc_rqst *req = task->tk_rqstp;
1439 struct rpc_xprt *xprt = req->rq_xprt;
1440
1441 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1442 test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1443 xprt_is_pinned_rqst(req)) {
1444 spin_lock(&xprt->queue_lock);
1445 while (xprt_is_pinned_rqst(req)) {
1446 set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1447 spin_unlock(&xprt->queue_lock);
1448 xprt_wait_on_pinned_rqst(req);
1449 spin_lock(&xprt->queue_lock);
1450 clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1451 }
1452 xprt_request_dequeue_transmit_locked(task);
1453 xprt_request_dequeue_receive_locked(task);
1454 spin_unlock(&xprt->queue_lock);
1455 xdr_free_bvec(&req->rq_rcv_buf);
1456 }
1457}
1458
1459/**
1460 * xprt_request_prepare - prepare an encoded request for transport
1461 * @req: pointer to rpc_rqst
1462 * @buf: pointer to send/rcv xdr_buf
1463 *
1464 * Calls into the transport layer to do whatever is needed to prepare
1465 * the request for transmission or receive.
1466 * Returns error, or zero.
1467 */
1468static int
1469xprt_request_prepare(struct rpc_rqst *req, struct xdr_buf *buf)
1470{
1471 struct rpc_xprt *xprt = req->rq_xprt;
1472
1473 if (xprt->ops->prepare_request)
1474 return xprt->ops->prepare_request(req, buf);
1475 return 0;
1476}
1477
1478/**
1479 * xprt_request_need_retransmit - Test if a task needs retransmission
1480 * @task: pointer to rpc_task
1481 *
1482 * Test for whether a connection breakage requires the task to retransmit
1483 */
1484bool
1485xprt_request_need_retransmit(struct rpc_task *task)
1486{
1487 return xprt_request_retransmit_after_disconnect(task);
1488}
1489
1490/**
1491 * xprt_prepare_transmit - reserve the transport before sending a request
1492 * @task: RPC task about to send a request
1493 *
1494 */
1495bool xprt_prepare_transmit(struct rpc_task *task)
1496{
1497 struct rpc_rqst *req = task->tk_rqstp;
1498 struct rpc_xprt *xprt = req->rq_xprt;
1499
1500 if (!xprt_lock_write(xprt, task)) {
1501 /* Race breaker: someone may have transmitted us */
1502 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1503 rpc_wake_up_queued_task_set_status(&xprt->sending,
1504 task, 0);
1505 return false;
1506
1507 }
1508 if (atomic_read(&xprt->swapper))
1509 /* This will be clear in __rpc_execute */
1510 current->flags |= PF_MEMALLOC;
1511 return true;
1512}
1513
1514void xprt_end_transmit(struct rpc_task *task)
1515{
1516 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1517
1518 xprt_inject_disconnect(xprt);
1519 xprt_release_write(xprt, task);
1520}
1521
1522/**
1523 * xprt_request_transmit - send an RPC request on a transport
1524 * @req: pointer to request to transmit
1525 * @snd_task: RPC task that owns the transport lock
1526 *
1527 * This performs the transmission of a single request.
1528 * Note that if the request is not the same as snd_task, then it
1529 * does need to be pinned.
1530 * Returns '0' on success.
1531 */
1532static int
1533xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1534{
1535 struct rpc_xprt *xprt = req->rq_xprt;
1536 struct rpc_task *task = req->rq_task;
1537 unsigned int connect_cookie;
1538 int is_retrans = RPC_WAS_SENT(task);
1539 int status;
1540
1541 if (!req->rq_bytes_sent) {
1542 if (xprt_request_data_received(task)) {
1543 status = 0;
1544 goto out_dequeue;
1545 }
1546 /* Verify that our message lies in the RPCSEC_GSS window */
1547 if (rpcauth_xmit_need_reencode(task)) {
1548 status = -EBADMSG;
1549 goto out_dequeue;
1550 }
1551 if (RPC_SIGNALLED(task)) {
1552 status = -ERESTARTSYS;
1553 goto out_dequeue;
1554 }
1555 }
1556
1557 /*
1558 * Update req->rq_ntrans before transmitting to avoid races with
1559 * xprt_update_rtt(), which needs to know that it is recording a
1560 * reply to the first transmission.
1561 */
1562 req->rq_ntrans++;
1563
1564 trace_rpc_xdr_sendto(task, &req->rq_snd_buf);
1565 connect_cookie = xprt->connect_cookie;
1566 status = xprt->ops->send_request(req);
1567 if (status != 0) {
1568 req->rq_ntrans--;
1569 trace_xprt_transmit(req, status);
1570 return status;
1571 }
1572
1573 if (is_retrans) {
1574 task->tk_client->cl_stats->rpcretrans++;
1575 trace_xprt_retransmit(req);
1576 }
1577
1578 xprt_inject_disconnect(xprt);
1579
1580 task->tk_flags |= RPC_TASK_SENT;
1581 spin_lock(&xprt->transport_lock);
1582
1583 xprt->stat.sends++;
1584 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1585 xprt->stat.bklog_u += xprt->backlog.qlen;
1586 xprt->stat.sending_u += xprt->sending.qlen;
1587 xprt->stat.pending_u += xprt->pending.qlen;
1588 spin_unlock(&xprt->transport_lock);
1589
1590 req->rq_connect_cookie = connect_cookie;
1591out_dequeue:
1592 trace_xprt_transmit(req, status);
1593 xprt_request_dequeue_transmit(task);
1594 rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1595 return status;
1596}
1597
1598/**
1599 * xprt_transmit - send an RPC request on a transport
1600 * @task: controlling RPC task
1601 *
1602 * Attempts to drain the transmit queue. On exit, either the transport
1603 * signalled an error that needs to be handled before transmission can
1604 * resume, or @task finished transmitting, and detected that it already
1605 * received a reply.
1606 */
1607void
1608xprt_transmit(struct rpc_task *task)
1609{
1610 struct rpc_rqst *next, *req = task->tk_rqstp;
1611 struct rpc_xprt *xprt = req->rq_xprt;
1612 int status;
1613
1614 spin_lock(&xprt->queue_lock);
1615 for (;;) {
1616 next = list_first_entry_or_null(&xprt->xmit_queue,
1617 struct rpc_rqst, rq_xmit);
1618 if (!next)
1619 break;
1620 xprt_pin_rqst(next);
1621 spin_unlock(&xprt->queue_lock);
1622 status = xprt_request_transmit(next, task);
1623 if (status == -EBADMSG && next != req)
1624 status = 0;
1625 spin_lock(&xprt->queue_lock);
1626 xprt_unpin_rqst(next);
1627 if (status < 0) {
1628 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1629 task->tk_status = status;
1630 break;
1631 }
1632 /* Was @task transmitted, and has it received a reply? */
1633 if (xprt_request_data_received(task) &&
1634 !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1635 break;
1636 cond_resched_lock(&xprt->queue_lock);
1637 }
1638 spin_unlock(&xprt->queue_lock);
1639}
1640
1641static void xprt_complete_request_init(struct rpc_task *task)
1642{
1643 if (task->tk_rqstp)
1644 xprt_request_init(task);
1645}
1646
1647void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1648{
1649 set_bit(XPRT_CONGESTED, &xprt->state);
1650 rpc_sleep_on(&xprt->backlog, task, xprt_complete_request_init);
1651}
1652EXPORT_SYMBOL_GPL(xprt_add_backlog);
1653
1654static bool __xprt_set_rq(struct rpc_task *task, void *data)
1655{
1656 struct rpc_rqst *req = data;
1657
1658 if (task->tk_rqstp == NULL) {
1659 memset(req, 0, sizeof(*req)); /* mark unused */
1660 task->tk_rqstp = req;
1661 return true;
1662 }
1663 return false;
1664}
1665
1666bool xprt_wake_up_backlog(struct rpc_xprt *xprt, struct rpc_rqst *req)
1667{
1668 if (rpc_wake_up_first(&xprt->backlog, __xprt_set_rq, req) == NULL) {
1669 clear_bit(XPRT_CONGESTED, &xprt->state);
1670 return false;
1671 }
1672 return true;
1673}
1674EXPORT_SYMBOL_GPL(xprt_wake_up_backlog);
1675
1676static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1677{
1678 bool ret = false;
1679
1680 if (!test_bit(XPRT_CONGESTED, &xprt->state))
1681 goto out;
1682 spin_lock(&xprt->reserve_lock);
1683 if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1684 xprt_add_backlog(xprt, task);
1685 ret = true;
1686 }
1687 spin_unlock(&xprt->reserve_lock);
1688out:
1689 return ret;
1690}
1691
1692static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1693{
1694 struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1695
1696 if (xprt->num_reqs >= xprt->max_reqs)
1697 goto out;
1698 ++xprt->num_reqs;
1699 spin_unlock(&xprt->reserve_lock);
1700 req = kzalloc(sizeof(*req), rpc_task_gfp_mask());
1701 spin_lock(&xprt->reserve_lock);
1702 if (req != NULL)
1703 goto out;
1704 --xprt->num_reqs;
1705 req = ERR_PTR(-ENOMEM);
1706out:
1707 return req;
1708}
1709
1710static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1711{
1712 if (xprt->num_reqs > xprt->min_reqs) {
1713 --xprt->num_reqs;
1714 kfree(req);
1715 return true;
1716 }
1717 return false;
1718}
1719
1720void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1721{
1722 struct rpc_rqst *req;
1723
1724 spin_lock(&xprt->reserve_lock);
1725 if (!list_empty(&xprt->free)) {
1726 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1727 list_del(&req->rq_list);
1728 goto out_init_req;
1729 }
1730 req = xprt_dynamic_alloc_slot(xprt);
1731 if (!IS_ERR(req))
1732 goto out_init_req;
1733 switch (PTR_ERR(req)) {
1734 case -ENOMEM:
1735 dprintk("RPC: dynamic allocation of request slot "
1736 "failed! Retrying\n");
1737 task->tk_status = -ENOMEM;
1738 break;
1739 case -EAGAIN:
1740 xprt_add_backlog(xprt, task);
1741 dprintk("RPC: waiting for request slot\n");
1742 fallthrough;
1743 default:
1744 task->tk_status = -EAGAIN;
1745 }
1746 spin_unlock(&xprt->reserve_lock);
1747 return;
1748out_init_req:
1749 xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1750 xprt->num_reqs);
1751 spin_unlock(&xprt->reserve_lock);
1752
1753 task->tk_status = 0;
1754 task->tk_rqstp = req;
1755}
1756EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1757
1758void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1759{
1760 spin_lock(&xprt->reserve_lock);
1761 if (!xprt_wake_up_backlog(xprt, req) &&
1762 !xprt_dynamic_free_slot(xprt, req)) {
1763 memset(req, 0, sizeof(*req)); /* mark unused */
1764 list_add(&req->rq_list, &xprt->free);
1765 }
1766 spin_unlock(&xprt->reserve_lock);
1767}
1768EXPORT_SYMBOL_GPL(xprt_free_slot);
1769
1770static void xprt_free_all_slots(struct rpc_xprt *xprt)
1771{
1772 struct rpc_rqst *req;
1773 while (!list_empty(&xprt->free)) {
1774 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1775 list_del(&req->rq_list);
1776 kfree(req);
1777 }
1778}
1779
1780static DEFINE_IDA(rpc_xprt_ids);
1781
1782void xprt_cleanup_ids(void)
1783{
1784 ida_destroy(&rpc_xprt_ids);
1785}
1786
1787static int xprt_alloc_id(struct rpc_xprt *xprt)
1788{
1789 int id;
1790
1791 id = ida_alloc(&rpc_xprt_ids, GFP_KERNEL);
1792 if (id < 0)
1793 return id;
1794
1795 xprt->id = id;
1796 return 0;
1797}
1798
1799static void xprt_free_id(struct rpc_xprt *xprt)
1800{
1801 ida_free(&rpc_xprt_ids, xprt->id);
1802}
1803
1804struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1805 unsigned int num_prealloc,
1806 unsigned int max_alloc)
1807{
1808 struct rpc_xprt *xprt;
1809 struct rpc_rqst *req;
1810 int i;
1811
1812 xprt = kzalloc(size, GFP_KERNEL);
1813 if (xprt == NULL)
1814 goto out;
1815
1816 xprt_alloc_id(xprt);
1817 xprt_init(xprt, net);
1818
1819 for (i = 0; i < num_prealloc; i++) {
1820 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1821 if (!req)
1822 goto out_free;
1823 list_add(&req->rq_list, &xprt->free);
1824 }
1825 xprt->max_reqs = max_t(unsigned int, max_alloc, num_prealloc);
1826 xprt->min_reqs = num_prealloc;
1827 xprt->num_reqs = num_prealloc;
1828
1829 return xprt;
1830
1831out_free:
1832 xprt_free(xprt);
1833out:
1834 return NULL;
1835}
1836EXPORT_SYMBOL_GPL(xprt_alloc);
1837
1838void xprt_free(struct rpc_xprt *xprt)
1839{
1840 put_net_track(xprt->xprt_net, &xprt->ns_tracker);
1841 xprt_free_all_slots(xprt);
1842 xprt_free_id(xprt);
1843 rpc_sysfs_xprt_destroy(xprt);
1844 kfree_rcu(xprt, rcu);
1845}
1846EXPORT_SYMBOL_GPL(xprt_free);
1847
1848static void
1849xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1850{
1851 req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1852}
1853
1854static __be32
1855xprt_alloc_xid(struct rpc_xprt *xprt)
1856{
1857 __be32 xid;
1858
1859 spin_lock(&xprt->reserve_lock);
1860 xid = (__force __be32)xprt->xid++;
1861 spin_unlock(&xprt->reserve_lock);
1862 return xid;
1863}
1864
1865static void
1866xprt_init_xid(struct rpc_xprt *xprt)
1867{
1868 xprt->xid = get_random_u32();
1869}
1870
1871static void
1872xprt_request_init(struct rpc_task *task)
1873{
1874 struct rpc_xprt *xprt = task->tk_xprt;
1875 struct rpc_rqst *req = task->tk_rqstp;
1876
1877 req->rq_task = task;
1878 req->rq_xprt = xprt;
1879 req->rq_buffer = NULL;
1880 req->rq_xid = xprt_alloc_xid(xprt);
1881 xprt_init_connect_cookie(req, xprt);
1882 req->rq_snd_buf.len = 0;
1883 req->rq_snd_buf.buflen = 0;
1884 req->rq_rcv_buf.len = 0;
1885 req->rq_rcv_buf.buflen = 0;
1886 req->rq_snd_buf.bvec = NULL;
1887 req->rq_rcv_buf.bvec = NULL;
1888 req->rq_release_snd_buf = NULL;
1889 xprt_init_majortimeo(task, req);
1890
1891 trace_xprt_reserve(req);
1892}
1893
1894static void
1895xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1896{
1897 xprt->ops->alloc_slot(xprt, task);
1898 if (task->tk_rqstp != NULL)
1899 xprt_request_init(task);
1900}
1901
1902/**
1903 * xprt_reserve - allocate an RPC request slot
1904 * @task: RPC task requesting a slot allocation
1905 *
1906 * If the transport is marked as being congested, or if no more
1907 * slots are available, place the task on the transport's
1908 * backlog queue.
1909 */
1910void xprt_reserve(struct rpc_task *task)
1911{
1912 struct rpc_xprt *xprt = task->tk_xprt;
1913
1914 task->tk_status = 0;
1915 if (task->tk_rqstp != NULL)
1916 return;
1917
1918 task->tk_status = -EAGAIN;
1919 if (!xprt_throttle_congested(xprt, task))
1920 xprt_do_reserve(xprt, task);
1921}
1922
1923/**
1924 * xprt_retry_reserve - allocate an RPC request slot
1925 * @task: RPC task requesting a slot allocation
1926 *
1927 * If no more slots are available, place the task on the transport's
1928 * backlog queue.
1929 * Note that the only difference with xprt_reserve is that we now
1930 * ignore the value of the XPRT_CONGESTED flag.
1931 */
1932void xprt_retry_reserve(struct rpc_task *task)
1933{
1934 struct rpc_xprt *xprt = task->tk_xprt;
1935
1936 task->tk_status = 0;
1937 if (task->tk_rqstp != NULL)
1938 return;
1939
1940 task->tk_status = -EAGAIN;
1941 xprt_do_reserve(xprt, task);
1942}
1943
1944/**
1945 * xprt_release - release an RPC request slot
1946 * @task: task which is finished with the slot
1947 *
1948 */
1949void xprt_release(struct rpc_task *task)
1950{
1951 struct rpc_xprt *xprt;
1952 struct rpc_rqst *req = task->tk_rqstp;
1953
1954 if (req == NULL) {
1955 if (task->tk_client) {
1956 xprt = task->tk_xprt;
1957 xprt_release_write(xprt, task);
1958 }
1959 return;
1960 }
1961
1962 xprt = req->rq_xprt;
1963 xprt_request_dequeue_xprt(task);
1964 spin_lock(&xprt->transport_lock);
1965 xprt->ops->release_xprt(xprt, task);
1966 if (xprt->ops->release_request)
1967 xprt->ops->release_request(task);
1968 xprt_schedule_autodisconnect(xprt);
1969 spin_unlock(&xprt->transport_lock);
1970 if (req->rq_buffer)
1971 xprt->ops->buf_free(task);
1972 if (req->rq_cred != NULL)
1973 put_rpccred(req->rq_cred);
1974 if (req->rq_release_snd_buf)
1975 req->rq_release_snd_buf(req);
1976
1977 task->tk_rqstp = NULL;
1978 if (likely(!bc_prealloc(req)))
1979 xprt->ops->free_slot(xprt, req);
1980 else
1981 xprt_free_bc_request(req);
1982}
1983
1984#ifdef CONFIG_SUNRPC_BACKCHANNEL
1985void
1986xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1987{
1988 struct xdr_buf *xbufp = &req->rq_snd_buf;
1989
1990 task->tk_rqstp = req;
1991 req->rq_task = task;
1992 xprt_init_connect_cookie(req, req->rq_xprt);
1993 /*
1994 * Set up the xdr_buf length.
1995 * This also indicates that the buffer is XDR encoded already.
1996 */
1997 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1998 xbufp->tail[0].iov_len;
1999}
2000#endif
2001
2002static void xprt_init(struct rpc_xprt *xprt, struct net *net)
2003{
2004 kref_init(&xprt->kref);
2005
2006 spin_lock_init(&xprt->transport_lock);
2007 spin_lock_init(&xprt->reserve_lock);
2008 spin_lock_init(&xprt->queue_lock);
2009
2010 INIT_LIST_HEAD(&xprt->free);
2011 xprt->recv_queue = RB_ROOT;
2012 INIT_LIST_HEAD(&xprt->xmit_queue);
2013#if defined(CONFIG_SUNRPC_BACKCHANNEL)
2014 spin_lock_init(&xprt->bc_pa_lock);
2015 INIT_LIST_HEAD(&xprt->bc_pa_list);
2016#endif /* CONFIG_SUNRPC_BACKCHANNEL */
2017 INIT_LIST_HEAD(&xprt->xprt_switch);
2018
2019 xprt->last_used = jiffies;
2020 xprt->cwnd = RPC_INITCWND;
2021 xprt->bind_index = 0;
2022
2023 rpc_init_wait_queue(&xprt->binding, "xprt_binding");
2024 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
2025 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
2026 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
2027
2028 xprt_init_xid(xprt);
2029
2030 xprt->xprt_net = get_net_track(net, &xprt->ns_tracker, GFP_KERNEL);
2031}
2032
2033/**
2034 * xprt_create_transport - create an RPC transport
2035 * @args: rpc transport creation arguments
2036 *
2037 */
2038struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
2039{
2040 struct rpc_xprt *xprt;
2041 const struct xprt_class *t;
2042
2043 t = xprt_class_find_by_ident(args->ident);
2044 if (!t) {
2045 dprintk("RPC: transport (%d) not supported\n", args->ident);
2046 return ERR_PTR(-EIO);
2047 }
2048
2049 xprt = t->setup(args);
2050 xprt_class_release(t);
2051
2052 if (IS_ERR(xprt))
2053 goto out;
2054 if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
2055 xprt->idle_timeout = 0;
2056 INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
2057 if (xprt_has_timer(xprt))
2058 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
2059 else
2060 timer_setup(&xprt->timer, NULL, 0);
2061
2062 if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
2063 xprt_destroy(xprt);
2064 return ERR_PTR(-EINVAL);
2065 }
2066 xprt->servername = kstrdup(args->servername, GFP_KERNEL);
2067 if (xprt->servername == NULL) {
2068 xprt_destroy(xprt);
2069 return ERR_PTR(-ENOMEM);
2070 }
2071
2072 rpc_xprt_debugfs_register(xprt);
2073
2074 trace_xprt_create(xprt);
2075out:
2076 return xprt;
2077}
2078
2079static void xprt_destroy_cb(struct work_struct *work)
2080{
2081 struct rpc_xprt *xprt =
2082 container_of(work, struct rpc_xprt, task_cleanup);
2083
2084 trace_xprt_destroy(xprt);
2085
2086 rpc_xprt_debugfs_unregister(xprt);
2087 rpc_destroy_wait_queue(&xprt->binding);
2088 rpc_destroy_wait_queue(&xprt->pending);
2089 rpc_destroy_wait_queue(&xprt->sending);
2090 rpc_destroy_wait_queue(&xprt->backlog);
2091 kfree(xprt->servername);
2092 /*
2093 * Destroy any existing back channel
2094 */
2095 xprt_destroy_backchannel(xprt, UINT_MAX);
2096
2097 /*
2098 * Tear down transport state and free the rpc_xprt
2099 */
2100 xprt->ops->destroy(xprt);
2101}
2102
2103/**
2104 * xprt_destroy - destroy an RPC transport, killing off all requests.
2105 * @xprt: transport to destroy
2106 *
2107 */
2108static void xprt_destroy(struct rpc_xprt *xprt)
2109{
2110 /*
2111 * Exclude transport connect/disconnect handlers and autoclose
2112 */
2113 wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
2114
2115 /*
2116 * xprt_schedule_autodisconnect() can run after XPRT_LOCKED
2117 * is cleared. We use ->transport_lock to ensure the mod_timer()
2118 * can only run *before* del_time_sync(), never after.
2119 */
2120 spin_lock(&xprt->transport_lock);
2121 del_timer_sync(&xprt->timer);
2122 spin_unlock(&xprt->transport_lock);
2123
2124 /*
2125 * Destroy sockets etc from the system workqueue so they can
2126 * safely flush receive work running on rpciod.
2127 */
2128 INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
2129 schedule_work(&xprt->task_cleanup);
2130}
2131
2132static void xprt_destroy_kref(struct kref *kref)
2133{
2134 xprt_destroy(container_of(kref, struct rpc_xprt, kref));
2135}
2136
2137/**
2138 * xprt_get - return a reference to an RPC transport.
2139 * @xprt: pointer to the transport
2140 *
2141 */
2142struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
2143{
2144 if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
2145 return xprt;
2146 return NULL;
2147}
2148EXPORT_SYMBOL_GPL(xprt_get);
2149
2150/**
2151 * xprt_put - release a reference to an RPC transport.
2152 * @xprt: pointer to the transport
2153 *
2154 */
2155void xprt_put(struct rpc_xprt *xprt)
2156{
2157 if (xprt != NULL)
2158 kref_put(&xprt->kref, xprt_destroy_kref);
2159}
2160EXPORT_SYMBOL_GPL(xprt_put);
2161
2162void xprt_set_offline_locked(struct rpc_xprt *xprt, struct rpc_xprt_switch *xps)
2163{
2164 if (!test_and_set_bit(XPRT_OFFLINE, &xprt->state)) {
2165 spin_lock(&xps->xps_lock);
2166 xps->xps_nactive--;
2167 spin_unlock(&xps->xps_lock);
2168 }
2169}
2170
2171void xprt_set_online_locked(struct rpc_xprt *xprt, struct rpc_xprt_switch *xps)
2172{
2173 if (test_and_clear_bit(XPRT_OFFLINE, &xprt->state)) {
2174 spin_lock(&xps->xps_lock);
2175 xps->xps_nactive++;
2176 spin_unlock(&xps->xps_lock);
2177 }
2178}
2179
2180void xprt_delete_locked(struct rpc_xprt *xprt, struct rpc_xprt_switch *xps)
2181{
2182 if (test_and_set_bit(XPRT_REMOVE, &xprt->state))
2183 return;
2184
2185 xprt_force_disconnect(xprt);
2186 if (!test_bit(XPRT_CONNECTED, &xprt->state))
2187 return;
2188
2189 if (!xprt->sending.qlen && !xprt->pending.qlen &&
2190 !xprt->backlog.qlen && !atomic_long_read(&xprt->queuelen))
2191 rpc_xprt_switch_remove_xprt(xps, xprt, true);
2192}