xref: /openbmc/linux/net/sunrpc/xprt.c (revision 07c7c6bf)
1 /*
2  *  linux/net/sunrpc/xprt.c
3  *
4  *  This is a generic RPC call interface supporting congestion avoidance,
5  *  and asynchronous calls.
6  *
7  *  The interface works like this:
8  *
9  *  -	When a process places a call, it allocates a request slot if
10  *	one is available. Otherwise, it sleeps on the backlog queue
11  *	(xprt_reserve).
12  *  -	Next, the caller puts together the RPC message, stuffs it into
13  *	the request struct, and calls xprt_transmit().
14  *  -	xprt_transmit sends the message and installs the caller on the
15  *	transport's wait list. At the same time, if a reply is expected,
16  *	it installs a timer that is run after the packet's timeout has
17  *	expired.
18  *  -	When a packet arrives, the data_ready handler walks the list of
19  *	pending requests for that transport. If a matching XID is found, the
20  *	caller is woken up, and the timer removed.
21  *  -	When no reply arrives within the timeout interval, the timer is
22  *	fired by the kernel and runs xprt_timer(). It either adjusts the
23  *	timeout values (minor timeout) or wakes up the caller with a status
24  *	of -ETIMEDOUT.
25  *  -	When the caller receives a notification from RPC that a reply arrived,
26  *	it should release the RPC slot, and process the reply.
27  *	If the call timed out, it may choose to retry the operation by
28  *	adjusting the initial timeout value, and simply calling rpc_call
29  *	again.
30  *
31  *  Support for async RPC is done through a set of RPC-specific scheduling
32  *  primitives that `transparently' work for processes as well as async
33  *  tasks that rely on callbacks.
34  *
35  *  Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
36  *
37  *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
38  */
39 
40 #include <linux/module.h>
41 
42 #include <linux/types.h>
43 #include <linux/interrupt.h>
44 #include <linux/workqueue.h>
45 #include <linux/net.h>
46 #include <linux/ktime.h>
47 
48 #include <linux/sunrpc/clnt.h>
49 #include <linux/sunrpc/metrics.h>
50 #include <linux/sunrpc/bc_xprt.h>
51 #include <linux/rcupdate.h>
52 #include <linux/sched/mm.h>
53 
54 #include <trace/events/sunrpc.h>
55 
56 #include "sunrpc.h"
57 
58 /*
59  * Local variables
60  */
61 
62 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
63 # define RPCDBG_FACILITY	RPCDBG_XPRT
64 #endif
65 
66 /*
67  * Local functions
68  */
69 static void	 xprt_init(struct rpc_xprt *xprt, struct net *net);
70 static __be32	xprt_alloc_xid(struct rpc_xprt *xprt);
71 static void	 xprt_destroy(struct rpc_xprt *xprt);
72 
73 static DEFINE_SPINLOCK(xprt_list_lock);
74 static LIST_HEAD(xprt_list);
75 
76 static unsigned long xprt_request_timeout(const struct rpc_rqst *req)
77 {
78 	unsigned long timeout = jiffies + req->rq_timeout;
79 
80 	if (time_before(timeout, req->rq_majortimeo))
81 		return timeout;
82 	return req->rq_majortimeo;
83 }
84 
85 /**
86  * xprt_register_transport - register a transport implementation
87  * @transport: transport to register
88  *
89  * If a transport implementation is loaded as a kernel module, it can
90  * call this interface to make itself known to the RPC client.
91  *
92  * Returns:
93  * 0:		transport successfully registered
94  * -EEXIST:	transport already registered
95  * -EINVAL:	transport module being unloaded
96  */
97 int xprt_register_transport(struct xprt_class *transport)
98 {
99 	struct xprt_class *t;
100 	int result;
101 
102 	result = -EEXIST;
103 	spin_lock(&xprt_list_lock);
104 	list_for_each_entry(t, &xprt_list, list) {
105 		/* don't register the same transport class twice */
106 		if (t->ident == transport->ident)
107 			goto out;
108 	}
109 
110 	list_add_tail(&transport->list, &xprt_list);
111 	printk(KERN_INFO "RPC: Registered %s transport module.\n",
112 	       transport->name);
113 	result = 0;
114 
115 out:
116 	spin_unlock(&xprt_list_lock);
117 	return result;
118 }
119 EXPORT_SYMBOL_GPL(xprt_register_transport);
120 
121 /**
122  * xprt_unregister_transport - unregister a transport implementation
123  * @transport: transport to unregister
124  *
125  * Returns:
126  * 0:		transport successfully unregistered
127  * -ENOENT:	transport never registered
128  */
129 int xprt_unregister_transport(struct xprt_class *transport)
130 {
131 	struct xprt_class *t;
132 	int result;
133 
134 	result = 0;
135 	spin_lock(&xprt_list_lock);
136 	list_for_each_entry(t, &xprt_list, list) {
137 		if (t == transport) {
138 			printk(KERN_INFO
139 				"RPC: Unregistered %s transport module.\n",
140 				transport->name);
141 			list_del_init(&transport->list);
142 			goto out;
143 		}
144 	}
145 	result = -ENOENT;
146 
147 out:
148 	spin_unlock(&xprt_list_lock);
149 	return result;
150 }
151 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
152 
153 /**
154  * xprt_load_transport - load a transport implementation
155  * @transport_name: transport to load
156  *
157  * Returns:
158  * 0:		transport successfully loaded
159  * -ENOENT:	transport module not available
160  */
161 int xprt_load_transport(const char *transport_name)
162 {
163 	struct xprt_class *t;
164 	int result;
165 
166 	result = 0;
167 	spin_lock(&xprt_list_lock);
168 	list_for_each_entry(t, &xprt_list, list) {
169 		if (strcmp(t->name, transport_name) == 0) {
170 			spin_unlock(&xprt_list_lock);
171 			goto out;
172 		}
173 	}
174 	spin_unlock(&xprt_list_lock);
175 	result = request_module("xprt%s", transport_name);
176 out:
177 	return result;
178 }
179 EXPORT_SYMBOL_GPL(xprt_load_transport);
180 
181 static void xprt_clear_locked(struct rpc_xprt *xprt)
182 {
183 	xprt->snd_task = NULL;
184 	if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
185 		smp_mb__before_atomic();
186 		clear_bit(XPRT_LOCKED, &xprt->state);
187 		smp_mb__after_atomic();
188 	} else
189 		queue_work(xprtiod_workqueue, &xprt->task_cleanup);
190 }
191 
192 /**
193  * xprt_reserve_xprt - serialize write access to transports
194  * @task: task that is requesting access to the transport
195  * @xprt: pointer to the target transport
196  *
197  * This prevents mixing the payload of separate requests, and prevents
198  * transport connects from colliding with writes.  No congestion control
199  * is provided.
200  */
201 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
202 {
203 	struct rpc_rqst *req = task->tk_rqstp;
204 
205 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
206 		if (task == xprt->snd_task)
207 			return 1;
208 		goto out_sleep;
209 	}
210 	if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
211 		goto out_unlock;
212 	xprt->snd_task = task;
213 
214 	return 1;
215 
216 out_unlock:
217 	xprt_clear_locked(xprt);
218 out_sleep:
219 	dprintk("RPC: %5u failed to lock transport %p\n",
220 			task->tk_pid, xprt);
221 	task->tk_status = -EAGAIN;
222 	if  (RPC_IS_SOFT(task))
223 		rpc_sleep_on_timeout(&xprt->sending, task, NULL,
224 				xprt_request_timeout(req));
225 	else
226 		rpc_sleep_on(&xprt->sending, task, NULL);
227 	return 0;
228 }
229 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
230 
231 static bool
232 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
233 {
234 	return test_bit(XPRT_CWND_WAIT, &xprt->state);
235 }
236 
237 static void
238 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
239 {
240 	if (!list_empty(&xprt->xmit_queue)) {
241 		/* Peek at head of queue to see if it can make progress */
242 		if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
243 					rq_xmit)->rq_cong)
244 			return;
245 	}
246 	set_bit(XPRT_CWND_WAIT, &xprt->state);
247 }
248 
249 static void
250 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
251 {
252 	if (!RPCXPRT_CONGESTED(xprt))
253 		clear_bit(XPRT_CWND_WAIT, &xprt->state);
254 }
255 
256 /*
257  * xprt_reserve_xprt_cong - serialize write access to transports
258  * @task: task that is requesting access to the transport
259  *
260  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
261  * integrated into the decision of whether a request is allowed to be
262  * woken up and given access to the transport.
263  * Note that the lock is only granted if we know there are free slots.
264  */
265 int xprt_reserve_xprt_cong(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 			return 1;
272 		goto out_sleep;
273 	}
274 	if (req == NULL) {
275 		xprt->snd_task = task;
276 		return 1;
277 	}
278 	if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
279 		goto out_unlock;
280 	if (!xprt_need_congestion_window_wait(xprt)) {
281 		xprt->snd_task = task;
282 		return 1;
283 	}
284 out_unlock:
285 	xprt_clear_locked(xprt);
286 out_sleep:
287 	dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
288 	task->tk_status = -EAGAIN;
289 	if (RPC_IS_SOFT(task))
290 		rpc_sleep_on_timeout(&xprt->sending, task, NULL,
291 				xprt_request_timeout(req));
292 	else
293 		rpc_sleep_on(&xprt->sending, task, NULL);
294 	return 0;
295 }
296 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
297 
298 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
299 {
300 	int retval;
301 
302 	if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
303 		return 1;
304 	spin_lock_bh(&xprt->transport_lock);
305 	retval = xprt->ops->reserve_xprt(xprt, task);
306 	spin_unlock_bh(&xprt->transport_lock);
307 	return retval;
308 }
309 
310 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
311 {
312 	struct rpc_xprt *xprt = data;
313 
314 	xprt->snd_task = task;
315 	return true;
316 }
317 
318 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
319 {
320 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
321 		return;
322 	if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
323 		goto out_unlock;
324 	if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
325 				__xprt_lock_write_func, xprt))
326 		return;
327 out_unlock:
328 	xprt_clear_locked(xprt);
329 }
330 
331 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
332 {
333 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
334 		return;
335 	if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
336 		goto out_unlock;
337 	if (xprt_need_congestion_window_wait(xprt))
338 		goto out_unlock;
339 	if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
340 				__xprt_lock_write_func, xprt))
341 		return;
342 out_unlock:
343 	xprt_clear_locked(xprt);
344 }
345 
346 /**
347  * xprt_release_xprt - allow other requests to use a transport
348  * @xprt: transport with other tasks potentially waiting
349  * @task: task that is releasing access to the transport
350  *
351  * Note that "task" can be NULL.  No congestion control is provided.
352  */
353 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
354 {
355 	if (xprt->snd_task == task) {
356 		xprt_clear_locked(xprt);
357 		__xprt_lock_write_next(xprt);
358 	}
359 }
360 EXPORT_SYMBOL_GPL(xprt_release_xprt);
361 
362 /**
363  * xprt_release_xprt_cong - allow other requests to use a transport
364  * @xprt: transport with other tasks potentially waiting
365  * @task: task that is releasing access to the transport
366  *
367  * Note that "task" can be NULL.  Another task is awoken to use the
368  * transport if the transport's congestion window allows it.
369  */
370 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
371 {
372 	if (xprt->snd_task == task) {
373 		xprt_clear_locked(xprt);
374 		__xprt_lock_write_next_cong(xprt);
375 	}
376 }
377 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
378 
379 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
380 {
381 	if (xprt->snd_task != task)
382 		return;
383 	spin_lock_bh(&xprt->transport_lock);
384 	xprt->ops->release_xprt(xprt, task);
385 	spin_unlock_bh(&xprt->transport_lock);
386 }
387 
388 /*
389  * Van Jacobson congestion avoidance. Check if the congestion window
390  * overflowed. Put the task to sleep if this is the case.
391  */
392 static int
393 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
394 {
395 	if (req->rq_cong)
396 		return 1;
397 	dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
398 			req->rq_task->tk_pid, xprt->cong, xprt->cwnd);
399 	if (RPCXPRT_CONGESTED(xprt)) {
400 		xprt_set_congestion_window_wait(xprt);
401 		return 0;
402 	}
403 	req->rq_cong = 1;
404 	xprt->cong += RPC_CWNDSCALE;
405 	return 1;
406 }
407 
408 /*
409  * Adjust the congestion window, and wake up the next task
410  * that has been sleeping due to congestion
411  */
412 static void
413 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
414 {
415 	if (!req->rq_cong)
416 		return;
417 	req->rq_cong = 0;
418 	xprt->cong -= RPC_CWNDSCALE;
419 	xprt_test_and_clear_congestion_window_wait(xprt);
420 	__xprt_lock_write_next_cong(xprt);
421 }
422 
423 /**
424  * xprt_request_get_cong - Request congestion control credits
425  * @xprt: pointer to transport
426  * @req: pointer to RPC request
427  *
428  * Useful for transports that require congestion control.
429  */
430 bool
431 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
432 {
433 	bool ret = false;
434 
435 	if (req->rq_cong)
436 		return true;
437 	spin_lock_bh(&xprt->transport_lock);
438 	ret = __xprt_get_cong(xprt, req) != 0;
439 	spin_unlock_bh(&xprt->transport_lock);
440 	return ret;
441 }
442 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
443 
444 /**
445  * xprt_release_rqst_cong - housekeeping when request is complete
446  * @task: RPC request that recently completed
447  *
448  * Useful for transports that require congestion control.
449  */
450 void xprt_release_rqst_cong(struct rpc_task *task)
451 {
452 	struct rpc_rqst *req = task->tk_rqstp;
453 
454 	__xprt_put_cong(req->rq_xprt, req);
455 }
456 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
457 
458 /*
459  * Clear the congestion window wait flag and wake up the next
460  * entry on xprt->sending
461  */
462 static void
463 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
464 {
465 	if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
466 		spin_lock_bh(&xprt->transport_lock);
467 		__xprt_lock_write_next_cong(xprt);
468 		spin_unlock_bh(&xprt->transport_lock);
469 	}
470 }
471 
472 /**
473  * xprt_adjust_cwnd - adjust transport congestion window
474  * @xprt: pointer to xprt
475  * @task: recently completed RPC request used to adjust window
476  * @result: result code of completed RPC request
477  *
478  * The transport code maintains an estimate on the maximum number of out-
479  * standing RPC requests, using a smoothed version of the congestion
480  * avoidance implemented in 44BSD. This is basically the Van Jacobson
481  * congestion algorithm: If a retransmit occurs, the congestion window is
482  * halved; otherwise, it is incremented by 1/cwnd when
483  *
484  *	-	a reply is received and
485  *	-	a full number of requests are outstanding and
486  *	-	the congestion window hasn't been updated recently.
487  */
488 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
489 {
490 	struct rpc_rqst *req = task->tk_rqstp;
491 	unsigned long cwnd = xprt->cwnd;
492 
493 	if (result >= 0 && cwnd <= xprt->cong) {
494 		/* The (cwnd >> 1) term makes sure
495 		 * the result gets rounded properly. */
496 		cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
497 		if (cwnd > RPC_MAXCWND(xprt))
498 			cwnd = RPC_MAXCWND(xprt);
499 		__xprt_lock_write_next_cong(xprt);
500 	} else if (result == -ETIMEDOUT) {
501 		cwnd >>= 1;
502 		if (cwnd < RPC_CWNDSCALE)
503 			cwnd = RPC_CWNDSCALE;
504 	}
505 	dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
506 			xprt->cong, xprt->cwnd, cwnd);
507 	xprt->cwnd = cwnd;
508 	__xprt_put_cong(xprt, req);
509 }
510 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
511 
512 /**
513  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
514  * @xprt: transport with waiting tasks
515  * @status: result code to plant in each task before waking it
516  *
517  */
518 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
519 {
520 	if (status < 0)
521 		rpc_wake_up_status(&xprt->pending, status);
522 	else
523 		rpc_wake_up(&xprt->pending);
524 }
525 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
526 
527 /**
528  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
529  * @xprt: transport
530  *
531  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
532  * we don't in general want to force a socket disconnection due to
533  * an incomplete RPC call transmission.
534  */
535 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
536 {
537 	set_bit(XPRT_WRITE_SPACE, &xprt->state);
538 }
539 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
540 
541 static bool
542 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
543 {
544 	if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
545 		__xprt_lock_write_next(xprt);
546 		dprintk("RPC:       write space: waking waiting task on "
547 				"xprt %p\n", xprt);
548 		return true;
549 	}
550 	return false;
551 }
552 
553 /**
554  * xprt_write_space - wake the task waiting for transport output buffer space
555  * @xprt: transport with waiting tasks
556  *
557  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
558  */
559 bool xprt_write_space(struct rpc_xprt *xprt)
560 {
561 	bool ret;
562 
563 	if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
564 		return false;
565 	spin_lock_bh(&xprt->transport_lock);
566 	ret = xprt_clear_write_space_locked(xprt);
567 	spin_unlock_bh(&xprt->transport_lock);
568 	return ret;
569 }
570 EXPORT_SYMBOL_GPL(xprt_write_space);
571 
572 static unsigned long xprt_abs_ktime_to_jiffies(ktime_t abstime)
573 {
574 	s64 delta = ktime_to_ns(ktime_get() - abstime);
575 	return likely(delta >= 0) ?
576 		jiffies - nsecs_to_jiffies(delta) :
577 		jiffies + nsecs_to_jiffies(-delta);
578 }
579 
580 static unsigned long xprt_calc_majortimeo(struct rpc_rqst *req)
581 {
582 	const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
583 	unsigned long majortimeo = req->rq_timeout;
584 
585 	if (to->to_exponential)
586 		majortimeo <<= to->to_retries;
587 	else
588 		majortimeo += to->to_increment * to->to_retries;
589 	if (majortimeo > to->to_maxval || majortimeo == 0)
590 		majortimeo = to->to_maxval;
591 	return majortimeo;
592 }
593 
594 static void xprt_reset_majortimeo(struct rpc_rqst *req)
595 {
596 	req->rq_majortimeo += xprt_calc_majortimeo(req);
597 }
598 
599 static void xprt_init_majortimeo(struct rpc_task *task, struct rpc_rqst *req)
600 {
601 	unsigned long time_init;
602 	struct rpc_xprt *xprt = req->rq_xprt;
603 
604 	if (likely(xprt && xprt_connected(xprt)))
605 		time_init = jiffies;
606 	else
607 		time_init = xprt_abs_ktime_to_jiffies(task->tk_start);
608 	req->rq_timeout = task->tk_client->cl_timeout->to_initval;
609 	req->rq_majortimeo = time_init + xprt_calc_majortimeo(req);
610 }
611 
612 /**
613  * xprt_adjust_timeout - adjust timeout values for next retransmit
614  * @req: RPC request containing parameters to use for the adjustment
615  *
616  */
617 int xprt_adjust_timeout(struct rpc_rqst *req)
618 {
619 	struct rpc_xprt *xprt = req->rq_xprt;
620 	const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
621 	int status = 0;
622 
623 	if (time_before(jiffies, req->rq_majortimeo)) {
624 		if (to->to_exponential)
625 			req->rq_timeout <<= 1;
626 		else
627 			req->rq_timeout += to->to_increment;
628 		if (to->to_maxval && req->rq_timeout >= to->to_maxval)
629 			req->rq_timeout = to->to_maxval;
630 		req->rq_retries++;
631 	} else {
632 		req->rq_timeout = to->to_initval;
633 		req->rq_retries = 0;
634 		xprt_reset_majortimeo(req);
635 		/* Reset the RTT counters == "slow start" */
636 		spin_lock_bh(&xprt->transport_lock);
637 		rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
638 		spin_unlock_bh(&xprt->transport_lock);
639 		status = -ETIMEDOUT;
640 	}
641 
642 	if (req->rq_timeout == 0) {
643 		printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
644 		req->rq_timeout = 5 * HZ;
645 	}
646 	return status;
647 }
648 
649 static void xprt_autoclose(struct work_struct *work)
650 {
651 	struct rpc_xprt *xprt =
652 		container_of(work, struct rpc_xprt, task_cleanup);
653 	unsigned int pflags = memalloc_nofs_save();
654 
655 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
656 	xprt->ops->close(xprt);
657 	xprt_release_write(xprt, NULL);
658 	wake_up_bit(&xprt->state, XPRT_LOCKED);
659 	memalloc_nofs_restore(pflags);
660 }
661 
662 /**
663  * xprt_disconnect_done - mark a transport as disconnected
664  * @xprt: transport to flag for disconnect
665  *
666  */
667 void xprt_disconnect_done(struct rpc_xprt *xprt)
668 {
669 	dprintk("RPC:       disconnected transport %p\n", xprt);
670 	spin_lock_bh(&xprt->transport_lock);
671 	xprt_clear_connected(xprt);
672 	xprt_clear_write_space_locked(xprt);
673 	xprt_wake_pending_tasks(xprt, -ENOTCONN);
674 	spin_unlock_bh(&xprt->transport_lock);
675 }
676 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
677 
678 /**
679  * xprt_force_disconnect - force a transport to disconnect
680  * @xprt: transport to disconnect
681  *
682  */
683 void xprt_force_disconnect(struct rpc_xprt *xprt)
684 {
685 	/* Don't race with the test_bit() in xprt_clear_locked() */
686 	spin_lock_bh(&xprt->transport_lock);
687 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
688 	/* Try to schedule an autoclose RPC call */
689 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
690 		queue_work(xprtiod_workqueue, &xprt->task_cleanup);
691 	else if (xprt->snd_task)
692 		rpc_wake_up_queued_task_set_status(&xprt->pending,
693 				xprt->snd_task, -ENOTCONN);
694 	spin_unlock_bh(&xprt->transport_lock);
695 }
696 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
697 
698 static unsigned int
699 xprt_connect_cookie(struct rpc_xprt *xprt)
700 {
701 	return READ_ONCE(xprt->connect_cookie);
702 }
703 
704 static bool
705 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
706 {
707 	struct rpc_rqst *req = task->tk_rqstp;
708 	struct rpc_xprt *xprt = req->rq_xprt;
709 
710 	return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
711 		!xprt_connected(xprt);
712 }
713 
714 /**
715  * xprt_conditional_disconnect - force a transport to disconnect
716  * @xprt: transport to disconnect
717  * @cookie: 'connection cookie'
718  *
719  * This attempts to break the connection if and only if 'cookie' matches
720  * the current transport 'connection cookie'. It ensures that we don't
721  * try to break the connection more than once when we need to retransmit
722  * a batch of RPC requests.
723  *
724  */
725 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
726 {
727 	/* Don't race with the test_bit() in xprt_clear_locked() */
728 	spin_lock_bh(&xprt->transport_lock);
729 	if (cookie != xprt->connect_cookie)
730 		goto out;
731 	if (test_bit(XPRT_CLOSING, &xprt->state))
732 		goto out;
733 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
734 	/* Try to schedule an autoclose RPC call */
735 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
736 		queue_work(xprtiod_workqueue, &xprt->task_cleanup);
737 	xprt_wake_pending_tasks(xprt, -EAGAIN);
738 out:
739 	spin_unlock_bh(&xprt->transport_lock);
740 }
741 
742 static bool
743 xprt_has_timer(const struct rpc_xprt *xprt)
744 {
745 	return xprt->idle_timeout != 0;
746 }
747 
748 static void
749 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
750 	__must_hold(&xprt->transport_lock)
751 {
752 	if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
753 		mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
754 }
755 
756 static void
757 xprt_init_autodisconnect(struct timer_list *t)
758 {
759 	struct rpc_xprt *xprt = from_timer(xprt, t, timer);
760 
761 	spin_lock(&xprt->transport_lock);
762 	if (!RB_EMPTY_ROOT(&xprt->recv_queue))
763 		goto out_abort;
764 	/* Reset xprt->last_used to avoid connect/autodisconnect cycling */
765 	xprt->last_used = jiffies;
766 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
767 		goto out_abort;
768 	spin_unlock(&xprt->transport_lock);
769 	queue_work(xprtiod_workqueue, &xprt->task_cleanup);
770 	return;
771 out_abort:
772 	spin_unlock(&xprt->transport_lock);
773 }
774 
775 bool xprt_lock_connect(struct rpc_xprt *xprt,
776 		struct rpc_task *task,
777 		void *cookie)
778 {
779 	bool ret = false;
780 
781 	spin_lock_bh(&xprt->transport_lock);
782 	if (!test_bit(XPRT_LOCKED, &xprt->state))
783 		goto out;
784 	if (xprt->snd_task != task)
785 		goto out;
786 	xprt->snd_task = cookie;
787 	ret = true;
788 out:
789 	spin_unlock_bh(&xprt->transport_lock);
790 	return ret;
791 }
792 
793 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
794 {
795 	spin_lock_bh(&xprt->transport_lock);
796 	if (xprt->snd_task != cookie)
797 		goto out;
798 	if (!test_bit(XPRT_LOCKED, &xprt->state))
799 		goto out;
800 	xprt->snd_task =NULL;
801 	xprt->ops->release_xprt(xprt, NULL);
802 	xprt_schedule_autodisconnect(xprt);
803 out:
804 	spin_unlock_bh(&xprt->transport_lock);
805 	wake_up_bit(&xprt->state, XPRT_LOCKED);
806 }
807 
808 /**
809  * xprt_connect - schedule a transport connect operation
810  * @task: RPC task that is requesting the connect
811  *
812  */
813 void xprt_connect(struct rpc_task *task)
814 {
815 	struct rpc_xprt	*xprt = task->tk_rqstp->rq_xprt;
816 
817 	dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
818 			xprt, (xprt_connected(xprt) ? "is" : "is not"));
819 
820 	if (!xprt_bound(xprt)) {
821 		task->tk_status = -EAGAIN;
822 		return;
823 	}
824 	if (!xprt_lock_write(xprt, task))
825 		return;
826 
827 	if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
828 		xprt->ops->close(xprt);
829 
830 	if (!xprt_connected(xprt)) {
831 		task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
832 		rpc_sleep_on_timeout(&xprt->pending, task, NULL,
833 				xprt_request_timeout(task->tk_rqstp));
834 
835 		if (test_bit(XPRT_CLOSING, &xprt->state))
836 			return;
837 		if (xprt_test_and_set_connecting(xprt))
838 			return;
839 		/* Race breaker */
840 		if (!xprt_connected(xprt)) {
841 			xprt->stat.connect_start = jiffies;
842 			xprt->ops->connect(xprt, task);
843 		} else {
844 			xprt_clear_connecting(xprt);
845 			task->tk_status = 0;
846 			rpc_wake_up_queued_task(&xprt->pending, task);
847 		}
848 	}
849 	xprt_release_write(xprt, task);
850 }
851 
852 enum xprt_xid_rb_cmp {
853 	XID_RB_EQUAL,
854 	XID_RB_LEFT,
855 	XID_RB_RIGHT,
856 };
857 static enum xprt_xid_rb_cmp
858 xprt_xid_cmp(__be32 xid1, __be32 xid2)
859 {
860 	if (xid1 == xid2)
861 		return XID_RB_EQUAL;
862 	if ((__force u32)xid1 < (__force u32)xid2)
863 		return XID_RB_LEFT;
864 	return XID_RB_RIGHT;
865 }
866 
867 static struct rpc_rqst *
868 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
869 {
870 	struct rb_node *n = xprt->recv_queue.rb_node;
871 	struct rpc_rqst *req;
872 
873 	while (n != NULL) {
874 		req = rb_entry(n, struct rpc_rqst, rq_recv);
875 		switch (xprt_xid_cmp(xid, req->rq_xid)) {
876 		case XID_RB_LEFT:
877 			n = n->rb_left;
878 			break;
879 		case XID_RB_RIGHT:
880 			n = n->rb_right;
881 			break;
882 		case XID_RB_EQUAL:
883 			return req;
884 		}
885 	}
886 	return NULL;
887 }
888 
889 static void
890 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
891 {
892 	struct rb_node **p = &xprt->recv_queue.rb_node;
893 	struct rb_node *n = NULL;
894 	struct rpc_rqst *req;
895 
896 	while (*p != NULL) {
897 		n = *p;
898 		req = rb_entry(n, struct rpc_rqst, rq_recv);
899 		switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
900 		case XID_RB_LEFT:
901 			p = &n->rb_left;
902 			break;
903 		case XID_RB_RIGHT:
904 			p = &n->rb_right;
905 			break;
906 		case XID_RB_EQUAL:
907 			WARN_ON_ONCE(new != req);
908 			return;
909 		}
910 	}
911 	rb_link_node(&new->rq_recv, n, p);
912 	rb_insert_color(&new->rq_recv, &xprt->recv_queue);
913 }
914 
915 static void
916 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
917 {
918 	rb_erase(&req->rq_recv, &xprt->recv_queue);
919 }
920 
921 /**
922  * xprt_lookup_rqst - find an RPC request corresponding to an XID
923  * @xprt: transport on which the original request was transmitted
924  * @xid: RPC XID of incoming reply
925  *
926  * Caller holds xprt->queue_lock.
927  */
928 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
929 {
930 	struct rpc_rqst *entry;
931 
932 	entry = xprt_request_rb_find(xprt, xid);
933 	if (entry != NULL) {
934 		trace_xprt_lookup_rqst(xprt, xid, 0);
935 		entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
936 		return entry;
937 	}
938 
939 	dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
940 			ntohl(xid));
941 	trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
942 	xprt->stat.bad_xids++;
943 	return NULL;
944 }
945 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
946 
947 static bool
948 xprt_is_pinned_rqst(struct rpc_rqst *req)
949 {
950 	return atomic_read(&req->rq_pin) != 0;
951 }
952 
953 /**
954  * xprt_pin_rqst - Pin a request on the transport receive list
955  * @req: Request to pin
956  *
957  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
958  * so should be holding xprt->queue_lock.
959  */
960 void xprt_pin_rqst(struct rpc_rqst *req)
961 {
962 	atomic_inc(&req->rq_pin);
963 }
964 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
965 
966 /**
967  * xprt_unpin_rqst - Unpin a request on the transport receive list
968  * @req: Request to pin
969  *
970  * Caller should be holding xprt->queue_lock.
971  */
972 void xprt_unpin_rqst(struct rpc_rqst *req)
973 {
974 	if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
975 		atomic_dec(&req->rq_pin);
976 		return;
977 	}
978 	if (atomic_dec_and_test(&req->rq_pin))
979 		wake_up_var(&req->rq_pin);
980 }
981 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
982 
983 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
984 {
985 	wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
986 }
987 
988 static bool
989 xprt_request_data_received(struct rpc_task *task)
990 {
991 	return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
992 		READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
993 }
994 
995 static bool
996 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
997 {
998 	return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
999 		READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
1000 }
1001 
1002 /**
1003  * xprt_request_enqueue_receive - Add an request to the receive queue
1004  * @task: RPC task
1005  *
1006  */
1007 void
1008 xprt_request_enqueue_receive(struct rpc_task *task)
1009 {
1010 	struct rpc_rqst *req = task->tk_rqstp;
1011 	struct rpc_xprt *xprt = req->rq_xprt;
1012 
1013 	if (!xprt_request_need_enqueue_receive(task, req))
1014 		return;
1015 	spin_lock(&xprt->queue_lock);
1016 
1017 	/* Update the softirq receive buffer */
1018 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1019 			sizeof(req->rq_private_buf));
1020 
1021 	/* Add request to the receive list */
1022 	xprt_request_rb_insert(xprt, req);
1023 	set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1024 	spin_unlock(&xprt->queue_lock);
1025 
1026 	/* Turn off autodisconnect */
1027 	del_singleshot_timer_sync(&xprt->timer);
1028 }
1029 
1030 /**
1031  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1032  * @task: RPC task
1033  *
1034  * Caller must hold xprt->queue_lock.
1035  */
1036 static void
1037 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1038 {
1039 	struct rpc_rqst *req = task->tk_rqstp;
1040 
1041 	if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1042 		xprt_request_rb_remove(req->rq_xprt, req);
1043 }
1044 
1045 /**
1046  * xprt_update_rtt - Update RPC RTT statistics
1047  * @task: RPC request that recently completed
1048  *
1049  * Caller holds xprt->queue_lock.
1050  */
1051 void xprt_update_rtt(struct rpc_task *task)
1052 {
1053 	struct rpc_rqst *req = task->tk_rqstp;
1054 	struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1055 	unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1056 	long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1057 
1058 	if (timer) {
1059 		if (req->rq_ntrans == 1)
1060 			rpc_update_rtt(rtt, timer, m);
1061 		rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1062 	}
1063 }
1064 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1065 
1066 /**
1067  * xprt_complete_rqst - called when reply processing is complete
1068  * @task: RPC request that recently completed
1069  * @copied: actual number of bytes received from the transport
1070  *
1071  * Caller holds xprt->queue_lock.
1072  */
1073 void xprt_complete_rqst(struct rpc_task *task, int copied)
1074 {
1075 	struct rpc_rqst *req = task->tk_rqstp;
1076 	struct rpc_xprt *xprt = req->rq_xprt;
1077 
1078 	dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
1079 			task->tk_pid, ntohl(req->rq_xid), copied);
1080 	trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
1081 
1082 	xprt->stat.recvs++;
1083 
1084 	req->rq_private_buf.len = copied;
1085 	/* Ensure all writes are done before we update */
1086 	/* req->rq_reply_bytes_recvd */
1087 	smp_wmb();
1088 	req->rq_reply_bytes_recvd = copied;
1089 	xprt_request_dequeue_receive_locked(task);
1090 	rpc_wake_up_queued_task(&xprt->pending, task);
1091 }
1092 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1093 
1094 static void xprt_timer(struct rpc_task *task)
1095 {
1096 	struct rpc_rqst *req = task->tk_rqstp;
1097 	struct rpc_xprt *xprt = req->rq_xprt;
1098 
1099 	if (task->tk_status != -ETIMEDOUT)
1100 		return;
1101 
1102 	trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1103 	if (!req->rq_reply_bytes_recvd) {
1104 		if (xprt->ops->timer)
1105 			xprt->ops->timer(xprt, task);
1106 	} else
1107 		task->tk_status = 0;
1108 }
1109 
1110 /**
1111  * xprt_wait_for_reply_request_def - wait for reply
1112  * @task: pointer to rpc_task
1113  *
1114  * Set a request's retransmit timeout based on the transport's
1115  * default timeout parameters.  Used by transports that don't adjust
1116  * the retransmit timeout based on round-trip time estimation,
1117  * and put the task to sleep on the pending queue.
1118  */
1119 void xprt_wait_for_reply_request_def(struct rpc_task *task)
1120 {
1121 	struct rpc_rqst *req = task->tk_rqstp;
1122 
1123 	rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1124 			xprt_request_timeout(req));
1125 }
1126 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_def);
1127 
1128 /**
1129  * xprt_wait_for_reply_request_rtt - wait for reply using RTT estimator
1130  * @task: pointer to rpc_task
1131  *
1132  * Set a request's retransmit timeout using the RTT estimator,
1133  * and put the task to sleep on the pending queue.
1134  */
1135 void xprt_wait_for_reply_request_rtt(struct rpc_task *task)
1136 {
1137 	int timer = task->tk_msg.rpc_proc->p_timer;
1138 	struct rpc_clnt *clnt = task->tk_client;
1139 	struct rpc_rtt *rtt = clnt->cl_rtt;
1140 	struct rpc_rqst *req = task->tk_rqstp;
1141 	unsigned long max_timeout = clnt->cl_timeout->to_maxval;
1142 	unsigned long timeout;
1143 
1144 	timeout = rpc_calc_rto(rtt, timer);
1145 	timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
1146 	if (timeout > max_timeout || timeout == 0)
1147 		timeout = max_timeout;
1148 	rpc_sleep_on_timeout(&req->rq_xprt->pending, task, xprt_timer,
1149 			jiffies + timeout);
1150 }
1151 EXPORT_SYMBOL_GPL(xprt_wait_for_reply_request_rtt);
1152 
1153 /**
1154  * xprt_request_wait_receive - wait for the reply to an RPC request
1155  * @task: RPC task about to send a request
1156  *
1157  */
1158 void xprt_request_wait_receive(struct rpc_task *task)
1159 {
1160 	struct rpc_rqst *req = task->tk_rqstp;
1161 	struct rpc_xprt *xprt = req->rq_xprt;
1162 
1163 	if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1164 		return;
1165 	/*
1166 	 * Sleep on the pending queue if we're expecting a reply.
1167 	 * The spinlock ensures atomicity between the test of
1168 	 * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1169 	 */
1170 	spin_lock(&xprt->queue_lock);
1171 	if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1172 		xprt->ops->wait_for_reply_request(task);
1173 		/*
1174 		 * Send an extra queue wakeup call if the
1175 		 * connection was dropped in case the call to
1176 		 * rpc_sleep_on() raced.
1177 		 */
1178 		if (xprt_request_retransmit_after_disconnect(task))
1179 			rpc_wake_up_queued_task_set_status(&xprt->pending,
1180 					task, -ENOTCONN);
1181 	}
1182 	spin_unlock(&xprt->queue_lock);
1183 }
1184 
1185 static bool
1186 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1187 {
1188 	return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1189 }
1190 
1191 /**
1192  * xprt_request_enqueue_transmit - queue a task for transmission
1193  * @task: pointer to rpc_task
1194  *
1195  * Add a task to the transmission queue.
1196  */
1197 void
1198 xprt_request_enqueue_transmit(struct rpc_task *task)
1199 {
1200 	struct rpc_rqst *pos, *req = task->tk_rqstp;
1201 	struct rpc_xprt *xprt = req->rq_xprt;
1202 
1203 	if (xprt_request_need_enqueue_transmit(task, req)) {
1204 		req->rq_bytes_sent = 0;
1205 		spin_lock(&xprt->queue_lock);
1206 		/*
1207 		 * Requests that carry congestion control credits are added
1208 		 * to the head of the list to avoid starvation issues.
1209 		 */
1210 		if (req->rq_cong) {
1211 			xprt_clear_congestion_window_wait(xprt);
1212 			list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1213 				if (pos->rq_cong)
1214 					continue;
1215 				/* Note: req is added _before_ pos */
1216 				list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1217 				INIT_LIST_HEAD(&req->rq_xmit2);
1218 				trace_xprt_enq_xmit(task, 1);
1219 				goto out;
1220 			}
1221 		} else if (RPC_IS_SWAPPER(task)) {
1222 			list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1223 				if (pos->rq_cong || pos->rq_bytes_sent)
1224 					continue;
1225 				if (RPC_IS_SWAPPER(pos->rq_task))
1226 					continue;
1227 				/* Note: req is added _before_ pos */
1228 				list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1229 				INIT_LIST_HEAD(&req->rq_xmit2);
1230 				trace_xprt_enq_xmit(task, 2);
1231 				goto out;
1232 			}
1233 		} else if (!req->rq_seqno) {
1234 			list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1235 				if (pos->rq_task->tk_owner != task->tk_owner)
1236 					continue;
1237 				list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1238 				INIT_LIST_HEAD(&req->rq_xmit);
1239 				trace_xprt_enq_xmit(task, 3);
1240 				goto out;
1241 			}
1242 		}
1243 		list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1244 		INIT_LIST_HEAD(&req->rq_xmit2);
1245 		trace_xprt_enq_xmit(task, 4);
1246 out:
1247 		set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1248 		spin_unlock(&xprt->queue_lock);
1249 	}
1250 }
1251 
1252 /**
1253  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1254  * @task: pointer to rpc_task
1255  *
1256  * Remove a task from the transmission queue
1257  * Caller must hold xprt->queue_lock
1258  */
1259 static void
1260 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1261 {
1262 	struct rpc_rqst *req = task->tk_rqstp;
1263 
1264 	if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1265 		return;
1266 	if (!list_empty(&req->rq_xmit)) {
1267 		list_del(&req->rq_xmit);
1268 		if (!list_empty(&req->rq_xmit2)) {
1269 			struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1270 					struct rpc_rqst, rq_xmit2);
1271 			list_del(&req->rq_xmit2);
1272 			list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1273 		}
1274 	} else
1275 		list_del(&req->rq_xmit2);
1276 }
1277 
1278 /**
1279  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1280  * @task: pointer to rpc_task
1281  *
1282  * Remove a task from the transmission queue
1283  */
1284 static void
1285 xprt_request_dequeue_transmit(struct rpc_task *task)
1286 {
1287 	struct rpc_rqst *req = task->tk_rqstp;
1288 	struct rpc_xprt *xprt = req->rq_xprt;
1289 
1290 	spin_lock(&xprt->queue_lock);
1291 	xprt_request_dequeue_transmit_locked(task);
1292 	spin_unlock(&xprt->queue_lock);
1293 }
1294 
1295 /**
1296  * xprt_request_prepare - prepare an encoded request for transport
1297  * @req: pointer to rpc_rqst
1298  *
1299  * Calls into the transport layer to do whatever is needed to prepare
1300  * the request for transmission or receive.
1301  */
1302 void
1303 xprt_request_prepare(struct rpc_rqst *req)
1304 {
1305 	struct rpc_xprt *xprt = req->rq_xprt;
1306 
1307 	if (xprt->ops->prepare_request)
1308 		xprt->ops->prepare_request(req);
1309 }
1310 
1311 /**
1312  * xprt_request_need_retransmit - Test if a task needs retransmission
1313  * @task: pointer to rpc_task
1314  *
1315  * Test for whether a connection breakage requires the task to retransmit
1316  */
1317 bool
1318 xprt_request_need_retransmit(struct rpc_task *task)
1319 {
1320 	return xprt_request_retransmit_after_disconnect(task);
1321 }
1322 
1323 /**
1324  * xprt_prepare_transmit - reserve the transport before sending a request
1325  * @task: RPC task about to send a request
1326  *
1327  */
1328 bool xprt_prepare_transmit(struct rpc_task *task)
1329 {
1330 	struct rpc_rqst	*req = task->tk_rqstp;
1331 	struct rpc_xprt	*xprt = req->rq_xprt;
1332 
1333 	dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1334 
1335 	if (!xprt_lock_write(xprt, task)) {
1336 		/* Race breaker: someone may have transmitted us */
1337 		if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1338 			rpc_wake_up_queued_task_set_status(&xprt->sending,
1339 					task, 0);
1340 		return false;
1341 
1342 	}
1343 	return true;
1344 }
1345 
1346 void xprt_end_transmit(struct rpc_task *task)
1347 {
1348 	xprt_release_write(task->tk_rqstp->rq_xprt, task);
1349 }
1350 
1351 /**
1352  * xprt_request_transmit - send an RPC request on a transport
1353  * @req: pointer to request to transmit
1354  * @snd_task: RPC task that owns the transport lock
1355  *
1356  * This performs the transmission of a single request.
1357  * Note that if the request is not the same as snd_task, then it
1358  * does need to be pinned.
1359  * Returns '0' on success.
1360  */
1361 static int
1362 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1363 {
1364 	struct rpc_xprt *xprt = req->rq_xprt;
1365 	struct rpc_task *task = req->rq_task;
1366 	unsigned int connect_cookie;
1367 	int is_retrans = RPC_WAS_SENT(task);
1368 	int status;
1369 
1370 	if (!req->rq_bytes_sent) {
1371 		if (xprt_request_data_received(task)) {
1372 			status = 0;
1373 			goto out_dequeue;
1374 		}
1375 		/* Verify that our message lies in the RPCSEC_GSS window */
1376 		if (rpcauth_xmit_need_reencode(task)) {
1377 			status = -EBADMSG;
1378 			goto out_dequeue;
1379 		}
1380 		if (task->tk_ops->rpc_call_prepare_transmit) {
1381 			task->tk_ops->rpc_call_prepare_transmit(task,
1382 					task->tk_calldata);
1383 			status = task->tk_status;
1384 			if (status < 0)
1385 				goto out_dequeue;
1386 		}
1387 		if (RPC_SIGNALLED(task)) {
1388 			status = -ERESTARTSYS;
1389 			goto out_dequeue;
1390 		}
1391 	}
1392 
1393 	/*
1394 	 * Update req->rq_ntrans before transmitting to avoid races with
1395 	 * xprt_update_rtt(), which needs to know that it is recording a
1396 	 * reply to the first transmission.
1397 	 */
1398 	req->rq_ntrans++;
1399 
1400 	connect_cookie = xprt->connect_cookie;
1401 	status = xprt->ops->send_request(req);
1402 	if (status != 0) {
1403 		req->rq_ntrans--;
1404 		trace_xprt_transmit(req, status);
1405 		return status;
1406 	}
1407 
1408 	if (is_retrans)
1409 		task->tk_client->cl_stats->rpcretrans++;
1410 
1411 	xprt_inject_disconnect(xprt);
1412 
1413 	task->tk_flags |= RPC_TASK_SENT;
1414 	spin_lock_bh(&xprt->transport_lock);
1415 
1416 	xprt->stat.sends++;
1417 	xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1418 	xprt->stat.bklog_u += xprt->backlog.qlen;
1419 	xprt->stat.sending_u += xprt->sending.qlen;
1420 	xprt->stat.pending_u += xprt->pending.qlen;
1421 	spin_unlock_bh(&xprt->transport_lock);
1422 
1423 	req->rq_connect_cookie = connect_cookie;
1424 out_dequeue:
1425 	trace_xprt_transmit(req, status);
1426 	xprt_request_dequeue_transmit(task);
1427 	rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1428 	return status;
1429 }
1430 
1431 /**
1432  * xprt_transmit - send an RPC request on a transport
1433  * @task: controlling RPC task
1434  *
1435  * Attempts to drain the transmit queue. On exit, either the transport
1436  * signalled an error that needs to be handled before transmission can
1437  * resume, or @task finished transmitting, and detected that it already
1438  * received a reply.
1439  */
1440 void
1441 xprt_transmit(struct rpc_task *task)
1442 {
1443 	struct rpc_rqst *next, *req = task->tk_rqstp;
1444 	struct rpc_xprt	*xprt = req->rq_xprt;
1445 	int status;
1446 
1447 	spin_lock(&xprt->queue_lock);
1448 	while (!list_empty(&xprt->xmit_queue)) {
1449 		next = list_first_entry(&xprt->xmit_queue,
1450 				struct rpc_rqst, rq_xmit);
1451 		xprt_pin_rqst(next);
1452 		spin_unlock(&xprt->queue_lock);
1453 		status = xprt_request_transmit(next, task);
1454 		if (status == -EBADMSG && next != req)
1455 			status = 0;
1456 		cond_resched();
1457 		spin_lock(&xprt->queue_lock);
1458 		xprt_unpin_rqst(next);
1459 		if (status == 0) {
1460 			if (!xprt_request_data_received(task) ||
1461 			    test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1462 				continue;
1463 		} else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1464 			task->tk_status = status;
1465 		break;
1466 	}
1467 	spin_unlock(&xprt->queue_lock);
1468 }
1469 
1470 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1471 {
1472 	set_bit(XPRT_CONGESTED, &xprt->state);
1473 	rpc_sleep_on(&xprt->backlog, task, NULL);
1474 }
1475 
1476 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1477 {
1478 	if (rpc_wake_up_next(&xprt->backlog) == NULL)
1479 		clear_bit(XPRT_CONGESTED, &xprt->state);
1480 }
1481 
1482 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1483 {
1484 	bool ret = false;
1485 
1486 	if (!test_bit(XPRT_CONGESTED, &xprt->state))
1487 		goto out;
1488 	spin_lock(&xprt->reserve_lock);
1489 	if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1490 		rpc_sleep_on(&xprt->backlog, task, NULL);
1491 		ret = true;
1492 	}
1493 	spin_unlock(&xprt->reserve_lock);
1494 out:
1495 	return ret;
1496 }
1497 
1498 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1499 {
1500 	struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1501 
1502 	if (xprt->num_reqs >= xprt->max_reqs)
1503 		goto out;
1504 	++xprt->num_reqs;
1505 	spin_unlock(&xprt->reserve_lock);
1506 	req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1507 	spin_lock(&xprt->reserve_lock);
1508 	if (req != NULL)
1509 		goto out;
1510 	--xprt->num_reqs;
1511 	req = ERR_PTR(-ENOMEM);
1512 out:
1513 	return req;
1514 }
1515 
1516 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1517 {
1518 	if (xprt->num_reqs > xprt->min_reqs) {
1519 		--xprt->num_reqs;
1520 		kfree(req);
1521 		return true;
1522 	}
1523 	return false;
1524 }
1525 
1526 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1527 {
1528 	struct rpc_rqst *req;
1529 
1530 	spin_lock(&xprt->reserve_lock);
1531 	if (!list_empty(&xprt->free)) {
1532 		req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1533 		list_del(&req->rq_list);
1534 		goto out_init_req;
1535 	}
1536 	req = xprt_dynamic_alloc_slot(xprt);
1537 	if (!IS_ERR(req))
1538 		goto out_init_req;
1539 	switch (PTR_ERR(req)) {
1540 	case -ENOMEM:
1541 		dprintk("RPC:       dynamic allocation of request slot "
1542 				"failed! Retrying\n");
1543 		task->tk_status = -ENOMEM;
1544 		break;
1545 	case -EAGAIN:
1546 		xprt_add_backlog(xprt, task);
1547 		dprintk("RPC:       waiting for request slot\n");
1548 		/* fall through */
1549 	default:
1550 		task->tk_status = -EAGAIN;
1551 	}
1552 	spin_unlock(&xprt->reserve_lock);
1553 	return;
1554 out_init_req:
1555 	xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1556 				     xprt->num_reqs);
1557 	spin_unlock(&xprt->reserve_lock);
1558 
1559 	task->tk_status = 0;
1560 	task->tk_rqstp = req;
1561 }
1562 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1563 
1564 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1565 {
1566 	spin_lock(&xprt->reserve_lock);
1567 	if (!xprt_dynamic_free_slot(xprt, req)) {
1568 		memset(req, 0, sizeof(*req));	/* mark unused */
1569 		list_add(&req->rq_list, &xprt->free);
1570 	}
1571 	xprt_wake_up_backlog(xprt);
1572 	spin_unlock(&xprt->reserve_lock);
1573 }
1574 EXPORT_SYMBOL_GPL(xprt_free_slot);
1575 
1576 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1577 {
1578 	struct rpc_rqst *req;
1579 	while (!list_empty(&xprt->free)) {
1580 		req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1581 		list_del(&req->rq_list);
1582 		kfree(req);
1583 	}
1584 }
1585 
1586 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1587 		unsigned int num_prealloc,
1588 		unsigned int max_alloc)
1589 {
1590 	struct rpc_xprt *xprt;
1591 	struct rpc_rqst *req;
1592 	int i;
1593 
1594 	xprt = kzalloc(size, GFP_KERNEL);
1595 	if (xprt == NULL)
1596 		goto out;
1597 
1598 	xprt_init(xprt, net);
1599 
1600 	for (i = 0; i < num_prealloc; i++) {
1601 		req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1602 		if (!req)
1603 			goto out_free;
1604 		list_add(&req->rq_list, &xprt->free);
1605 	}
1606 	if (max_alloc > num_prealloc)
1607 		xprt->max_reqs = max_alloc;
1608 	else
1609 		xprt->max_reqs = num_prealloc;
1610 	xprt->min_reqs = num_prealloc;
1611 	xprt->num_reqs = num_prealloc;
1612 
1613 	return xprt;
1614 
1615 out_free:
1616 	xprt_free(xprt);
1617 out:
1618 	return NULL;
1619 }
1620 EXPORT_SYMBOL_GPL(xprt_alloc);
1621 
1622 void xprt_free(struct rpc_xprt *xprt)
1623 {
1624 	put_net(xprt->xprt_net);
1625 	xprt_free_all_slots(xprt);
1626 	kfree_rcu(xprt, rcu);
1627 }
1628 EXPORT_SYMBOL_GPL(xprt_free);
1629 
1630 static void
1631 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1632 {
1633 	req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1634 }
1635 
1636 static __be32
1637 xprt_alloc_xid(struct rpc_xprt *xprt)
1638 {
1639 	__be32 xid;
1640 
1641 	spin_lock(&xprt->reserve_lock);
1642 	xid = (__force __be32)xprt->xid++;
1643 	spin_unlock(&xprt->reserve_lock);
1644 	return xid;
1645 }
1646 
1647 static void
1648 xprt_init_xid(struct rpc_xprt *xprt)
1649 {
1650 	xprt->xid = prandom_u32();
1651 }
1652 
1653 static void
1654 xprt_request_init(struct rpc_task *task)
1655 {
1656 	struct rpc_xprt *xprt = task->tk_xprt;
1657 	struct rpc_rqst	*req = task->tk_rqstp;
1658 
1659 	req->rq_task	= task;
1660 	req->rq_xprt    = xprt;
1661 	req->rq_buffer  = NULL;
1662 	req->rq_xid	= xprt_alloc_xid(xprt);
1663 	xprt_init_connect_cookie(req, xprt);
1664 	req->rq_snd_buf.len = 0;
1665 	req->rq_snd_buf.buflen = 0;
1666 	req->rq_rcv_buf.len = 0;
1667 	req->rq_rcv_buf.buflen = 0;
1668 	req->rq_snd_buf.bvec = NULL;
1669 	req->rq_rcv_buf.bvec = NULL;
1670 	req->rq_release_snd_buf = NULL;
1671 	xprt_init_majortimeo(task, req);
1672 	dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1673 			req, ntohl(req->rq_xid));
1674 }
1675 
1676 static void
1677 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1678 {
1679 	xprt->ops->alloc_slot(xprt, task);
1680 	if (task->tk_rqstp != NULL)
1681 		xprt_request_init(task);
1682 }
1683 
1684 /**
1685  * xprt_reserve - allocate an RPC request slot
1686  * @task: RPC task requesting a slot allocation
1687  *
1688  * If the transport is marked as being congested, or if no more
1689  * slots are available, place the task on the transport's
1690  * backlog queue.
1691  */
1692 void xprt_reserve(struct rpc_task *task)
1693 {
1694 	struct rpc_xprt *xprt = task->tk_xprt;
1695 
1696 	task->tk_status = 0;
1697 	if (task->tk_rqstp != NULL)
1698 		return;
1699 
1700 	task->tk_status = -EAGAIN;
1701 	if (!xprt_throttle_congested(xprt, task))
1702 		xprt_do_reserve(xprt, task);
1703 }
1704 
1705 /**
1706  * xprt_retry_reserve - allocate an RPC request slot
1707  * @task: RPC task requesting a slot allocation
1708  *
1709  * If no more slots are available, place the task on the transport's
1710  * backlog queue.
1711  * Note that the only difference with xprt_reserve is that we now
1712  * ignore the value of the XPRT_CONGESTED flag.
1713  */
1714 void xprt_retry_reserve(struct rpc_task *task)
1715 {
1716 	struct rpc_xprt *xprt = task->tk_xprt;
1717 
1718 	task->tk_status = 0;
1719 	if (task->tk_rqstp != NULL)
1720 		return;
1721 
1722 	task->tk_status = -EAGAIN;
1723 	xprt_do_reserve(xprt, task);
1724 }
1725 
1726 static void
1727 xprt_request_dequeue_all(struct rpc_task *task, struct rpc_rqst *req)
1728 {
1729 	struct rpc_xprt *xprt = req->rq_xprt;
1730 
1731 	if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1732 	    test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1733 	    xprt_is_pinned_rqst(req)) {
1734 		spin_lock(&xprt->queue_lock);
1735 		xprt_request_dequeue_transmit_locked(task);
1736 		xprt_request_dequeue_receive_locked(task);
1737 		while (xprt_is_pinned_rqst(req)) {
1738 			set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1739 			spin_unlock(&xprt->queue_lock);
1740 			xprt_wait_on_pinned_rqst(req);
1741 			spin_lock(&xprt->queue_lock);
1742 			clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1743 		}
1744 		spin_unlock(&xprt->queue_lock);
1745 	}
1746 }
1747 
1748 /**
1749  * xprt_release - release an RPC request slot
1750  * @task: task which is finished with the slot
1751  *
1752  */
1753 void xprt_release(struct rpc_task *task)
1754 {
1755 	struct rpc_xprt	*xprt;
1756 	struct rpc_rqst	*req = task->tk_rqstp;
1757 
1758 	if (req == NULL) {
1759 		if (task->tk_client) {
1760 			xprt = task->tk_xprt;
1761 			xprt_release_write(xprt, task);
1762 		}
1763 		return;
1764 	}
1765 
1766 	xprt = req->rq_xprt;
1767 	if (task->tk_ops->rpc_count_stats != NULL)
1768 		task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1769 	else if (task->tk_client)
1770 		rpc_count_iostats(task, task->tk_client->cl_metrics);
1771 	xprt_request_dequeue_all(task, req);
1772 	spin_lock_bh(&xprt->transport_lock);
1773 	xprt->ops->release_xprt(xprt, task);
1774 	if (xprt->ops->release_request)
1775 		xprt->ops->release_request(task);
1776 	xprt->last_used = jiffies;
1777 	xprt_schedule_autodisconnect(xprt);
1778 	spin_unlock_bh(&xprt->transport_lock);
1779 	if (req->rq_buffer)
1780 		xprt->ops->buf_free(task);
1781 	xprt_inject_disconnect(xprt);
1782 	xdr_free_bvec(&req->rq_rcv_buf);
1783 	xdr_free_bvec(&req->rq_snd_buf);
1784 	if (req->rq_cred != NULL)
1785 		put_rpccred(req->rq_cred);
1786 	task->tk_rqstp = NULL;
1787 	if (req->rq_release_snd_buf)
1788 		req->rq_release_snd_buf(req);
1789 
1790 	dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1791 	if (likely(!bc_prealloc(req)))
1792 		xprt->ops->free_slot(xprt, req);
1793 	else
1794 		xprt_free_bc_request(req);
1795 }
1796 
1797 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1798 void
1799 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1800 {
1801 	struct xdr_buf *xbufp = &req->rq_snd_buf;
1802 
1803 	task->tk_rqstp = req;
1804 	req->rq_task = task;
1805 	xprt_init_connect_cookie(req, req->rq_xprt);
1806 	/*
1807 	 * Set up the xdr_buf length.
1808 	 * This also indicates that the buffer is XDR encoded already.
1809 	 */
1810 	xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1811 		xbufp->tail[0].iov_len;
1812 }
1813 #endif
1814 
1815 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1816 {
1817 	kref_init(&xprt->kref);
1818 
1819 	spin_lock_init(&xprt->transport_lock);
1820 	spin_lock_init(&xprt->reserve_lock);
1821 	spin_lock_init(&xprt->queue_lock);
1822 
1823 	INIT_LIST_HEAD(&xprt->free);
1824 	xprt->recv_queue = RB_ROOT;
1825 	INIT_LIST_HEAD(&xprt->xmit_queue);
1826 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1827 	spin_lock_init(&xprt->bc_pa_lock);
1828 	INIT_LIST_HEAD(&xprt->bc_pa_list);
1829 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1830 	INIT_LIST_HEAD(&xprt->xprt_switch);
1831 
1832 	xprt->last_used = jiffies;
1833 	xprt->cwnd = RPC_INITCWND;
1834 	xprt->bind_index = 0;
1835 
1836 	rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1837 	rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1838 	rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1839 	rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1840 
1841 	xprt_init_xid(xprt);
1842 
1843 	xprt->xprt_net = get_net(net);
1844 }
1845 
1846 /**
1847  * xprt_create_transport - create an RPC transport
1848  * @args: rpc transport creation arguments
1849  *
1850  */
1851 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1852 {
1853 	struct rpc_xprt	*xprt;
1854 	struct xprt_class *t;
1855 
1856 	spin_lock(&xprt_list_lock);
1857 	list_for_each_entry(t, &xprt_list, list) {
1858 		if (t->ident == args->ident) {
1859 			spin_unlock(&xprt_list_lock);
1860 			goto found;
1861 		}
1862 	}
1863 	spin_unlock(&xprt_list_lock);
1864 	dprintk("RPC: transport (%d) not supported\n", args->ident);
1865 	return ERR_PTR(-EIO);
1866 
1867 found:
1868 	xprt = t->setup(args);
1869 	if (IS_ERR(xprt)) {
1870 		dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1871 				-PTR_ERR(xprt));
1872 		goto out;
1873 	}
1874 	if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1875 		xprt->idle_timeout = 0;
1876 	INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1877 	if (xprt_has_timer(xprt))
1878 		timer_setup(&xprt->timer,
1879 				xprt_init_autodisconnect,
1880 				TIMER_DEFERRABLE);
1881 	else
1882 		timer_setup(&xprt->timer, NULL, 0);
1883 
1884 	if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1885 		xprt_destroy(xprt);
1886 		return ERR_PTR(-EINVAL);
1887 	}
1888 	xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1889 	if (xprt->servername == NULL) {
1890 		xprt_destroy(xprt);
1891 		return ERR_PTR(-ENOMEM);
1892 	}
1893 
1894 	rpc_xprt_debugfs_register(xprt);
1895 
1896 	dprintk("RPC:       created transport %p with %u slots\n", xprt,
1897 			xprt->max_reqs);
1898 out:
1899 	return xprt;
1900 }
1901 
1902 static void xprt_destroy_cb(struct work_struct *work)
1903 {
1904 	struct rpc_xprt *xprt =
1905 		container_of(work, struct rpc_xprt, task_cleanup);
1906 
1907 	rpc_xprt_debugfs_unregister(xprt);
1908 	rpc_destroy_wait_queue(&xprt->binding);
1909 	rpc_destroy_wait_queue(&xprt->pending);
1910 	rpc_destroy_wait_queue(&xprt->sending);
1911 	rpc_destroy_wait_queue(&xprt->backlog);
1912 	kfree(xprt->servername);
1913 	/*
1914 	 * Tear down transport state and free the rpc_xprt
1915 	 */
1916 	xprt->ops->destroy(xprt);
1917 }
1918 
1919 /**
1920  * xprt_destroy - destroy an RPC transport, killing off all requests.
1921  * @xprt: transport to destroy
1922  *
1923  */
1924 static void xprt_destroy(struct rpc_xprt *xprt)
1925 {
1926 	dprintk("RPC:       destroying transport %p\n", xprt);
1927 
1928 	/*
1929 	 * Exclude transport connect/disconnect handlers and autoclose
1930 	 */
1931 	wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1932 
1933 	del_timer_sync(&xprt->timer);
1934 
1935 	/*
1936 	 * Destroy sockets etc from the system workqueue so they can
1937 	 * safely flush receive work running on rpciod.
1938 	 */
1939 	INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1940 	schedule_work(&xprt->task_cleanup);
1941 }
1942 
1943 static void xprt_destroy_kref(struct kref *kref)
1944 {
1945 	xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1946 }
1947 
1948 /**
1949  * xprt_get - return a reference to an RPC transport.
1950  * @xprt: pointer to the transport
1951  *
1952  */
1953 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1954 {
1955 	if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1956 		return xprt;
1957 	return NULL;
1958 }
1959 EXPORT_SYMBOL_GPL(xprt_get);
1960 
1961 /**
1962  * xprt_put - release a reference to an RPC transport.
1963  * @xprt: pointer to the transport
1964  *
1965  */
1966 void xprt_put(struct rpc_xprt *xprt)
1967 {
1968 	if (xprt != NULL)
1969 		kref_put(&xprt->kref, xprt_destroy_kref);
1970 }
1971 EXPORT_SYMBOL_GPL(xprt_put);
1972