xref: /openbmc/linux/net/sunrpc/xprt.c (revision 2b77dcc5)
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  */
70 static void	 xprt_init(struct rpc_xprt *xprt, struct net *net);
71 static __be32	xprt_alloc_xid(struct rpc_xprt *xprt);
72 static void	 xprt_destroy(struct rpc_xprt *xprt);
73 
74 static DEFINE_SPINLOCK(xprt_list_lock);
75 static LIST_HEAD(xprt_list);
76 
77 static 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  */
98 int 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 
116 out:
117 	spin_unlock(&xprt_list_lock);
118 	return result;
119 }
120 EXPORT_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  */
130 int 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 
148 out:
149 	spin_unlock(&xprt_list_lock);
150 	return result;
151 }
152 EXPORT_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  */
162 int 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);
177 out:
178 	return result;
179 }
180 EXPORT_SYMBOL_GPL(xprt_load_transport);
181 
182 static 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  */
202 int 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 
217 out_unlock:
218 	xprt_clear_locked(xprt);
219 out_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 }
230 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
231 
232 static bool
233 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
234 {
235 	return test_bit(XPRT_CWND_WAIT, &xprt->state);
236 }
237 
238 static void
239 xprt_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 
250 static void
251 xprt_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  */
266 int 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 	}
285 out_unlock:
286 	xprt_clear_locked(xprt);
287 out_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 }
297 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
298 
299 static 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 
311 static 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 
319 static 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;
328 out_unlock:
329 	xprt_clear_locked(xprt);
330 }
331 
332 static 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;
343 out_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  */
354 void 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 }
361 EXPORT_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  */
371 void 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 }
378 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
379 
380 static 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  */
393 static 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  */
413 static 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  */
431 bool
432 xprt_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 }
443 EXPORT_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  */
451 void 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 }
457 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
458 
459 static 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  */
469 static void
470 xprt_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  */
495 void 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 }
517 EXPORT_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  */
525 void 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 }
532 EXPORT_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  */
542 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
543 {
544 	set_bit(XPRT_WRITE_SPACE, &xprt->state);
545 }
546 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
547 
548 static bool
549 xprt_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  */
566 bool 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 }
577 EXPORT_SYMBOL_GPL(xprt_write_space);
578 
579 static 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 
587 static 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 
601 static void xprt_reset_majortimeo(struct rpc_rqst *req)
602 {
603 	req->rq_majortimeo += xprt_calc_majortimeo(req);
604 }
605 
606 static 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  */
624 int 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 
656 static 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  */
674 void 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 }
684 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
685 
686 /**
687  * xprt_force_disconnect - force a transport to disconnect
688  * @xprt: transport to disconnect
689  *
690  */
691 void 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 }
704 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
705 
706 static unsigned int
707 xprt_connect_cookie(struct rpc_xprt *xprt)
708 {
709 	return READ_ONCE(xprt->connect_cookie);
710 }
711 
712 static bool
713 xprt_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  */
733 void 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);
746 out:
747 	spin_unlock(&xprt->transport_lock);
748 }
749 
750 static bool
751 xprt_has_timer(const struct rpc_xprt *xprt)
752 {
753 	return xprt->idle_timeout != 0;
754 }
755 
756 static void
757 xprt_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 
765 static void
766 xprt_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 
779 bool 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;
792 out:
793 	spin_unlock(&xprt->transport_lock);
794 	return ret;
795 }
796 
797 void 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);
807 out:
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  */
817 void 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  */
861 unsigned 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 }
870 EXPORT_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  */
878 void 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 }
886 EXPORT_SYMBOL_GPL(xprt_reconnect_backoff);
887 
888 enum xprt_xid_rb_cmp {
889 	XID_RB_EQUAL,
890 	XID_RB_LEFT,
891 	XID_RB_RIGHT,
892 };
893 static enum xprt_xid_rb_cmp
894 xprt_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 
903 static struct rpc_rqst *
904 xprt_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 
925 static void
926 xprt_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 
951 static void
952 xprt_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  */
964 struct 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 }
981 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
982 
983 static bool
984 xprt_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  */
996 void xprt_pin_rqst(struct rpc_rqst *req)
997 {
998 	atomic_inc(&req->rq_pin);
999 }
1000 EXPORT_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  */
1008 void 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 }
1017 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
1018 
1019 static 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 
1024 static bool
1025 xprt_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 
1031 static bool
1032 xprt_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  */
1043 void
1044 xprt_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  */
1074 static void
1075 xprt_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  */
1089 void 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 }
1102 EXPORT_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  */
1111 void 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 }
1130 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1131 
1132 static 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  */
1157 void 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 }
1164 EXPORT_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  */
1173 void 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 }
1189 EXPORT_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  */
1196 void 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 
1223 static bool
1224 xprt_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  */
1235 void
1236 xprt_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);
1284 out:
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  */
1297 static void
1298 xprt_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  */
1322 static void
1323 xprt_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  */
1340 void
1341 xprt_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  */
1370 void
1371 xprt_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  */
1385 bool
1386 xprt_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  */
1396 bool 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 
1414 void 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  */
1429 static int
1430 xprt_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;
1485 out_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  */
1501 void
1502 xprt_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 
1531 static 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 
1537 static 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 
1543 static 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);
1555 out:
1556 	return ret;
1557 }
1558 
1559 static 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);
1573 out:
1574 	return req;
1575 }
1576 
1577 static 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 
1587 void 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;
1615 out_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 }
1623 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1624 
1625 void 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 }
1635 EXPORT_SYMBOL_GPL(xprt_free_slot);
1636 
1637 static 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 
1647 struct 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 
1676 out_free:
1677 	xprt_free(xprt);
1678 out:
1679 	return NULL;
1680 }
1681 EXPORT_SYMBOL_GPL(xprt_alloc);
1682 
1683 void 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 }
1689 EXPORT_SYMBOL_GPL(xprt_free);
1690 
1691 static void
1692 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1693 {
1694 	req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1695 }
1696 
1697 static __be32
1698 xprt_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 
1708 static void
1709 xprt_init_xid(struct rpc_xprt *xprt)
1710 {
1711 	xprt->xid = prandom_u32();
1712 }
1713 
1714 static void
1715 xprt_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 
1737 static void
1738 xprt_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  */
1753 void 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  */
1775 void 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  */
1792 void 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
1832 void
1833 xprt_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 
1849 static 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  */
1885 struct 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 
1901 found:
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);
1930 out:
1931 	return xprt;
1932 }
1933 
1934 static 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 	 * Tear down transport state and free the rpc_xprt
1947 	 */
1948 	xprt->ops->destroy(xprt);
1949 }
1950 
1951 /**
1952  * xprt_destroy - destroy an RPC transport, killing off all requests.
1953  * @xprt: transport to destroy
1954  *
1955  */
1956 static void xprt_destroy(struct rpc_xprt *xprt)
1957 {
1958 	dprintk("RPC:       destroying transport %p\n", xprt);
1959 
1960 	/*
1961 	 * Exclude transport connect/disconnect handlers and autoclose
1962 	 */
1963 	wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1964 
1965 	del_timer_sync(&xprt->timer);
1966 
1967 	/*
1968 	 * Destroy sockets etc from the system workqueue so they can
1969 	 * safely flush receive work running on rpciod.
1970 	 */
1971 	INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1972 	schedule_work(&xprt->task_cleanup);
1973 }
1974 
1975 static void xprt_destroy_kref(struct kref *kref)
1976 {
1977 	xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1978 }
1979 
1980 /**
1981  * xprt_get - return a reference to an RPC transport.
1982  * @xprt: pointer to the transport
1983  *
1984  */
1985 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1986 {
1987 	if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1988 		return xprt;
1989 	return NULL;
1990 }
1991 EXPORT_SYMBOL_GPL(xprt_get);
1992 
1993 /**
1994  * xprt_put - release a reference to an RPC transport.
1995  * @xprt: pointer to the transport
1996  *
1997  */
1998 void xprt_put(struct rpc_xprt *xprt)
1999 {
2000 	if (xprt != NULL)
2001 		kref_put(&xprt->kref, xprt_destroy_kref);
2002 }
2003 EXPORT_SYMBOL_GPL(xprt_put);
2004