xref: /openbmc/linux/net/sunrpc/xprt.c (revision 79f08d9e)
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 
52 #include "sunrpc.h"
53 
54 /*
55  * Local variables
56  */
57 
58 #ifdef RPC_DEBUG
59 # define RPCDBG_FACILITY	RPCDBG_XPRT
60 #endif
61 
62 /*
63  * Local functions
64  */
65 static void	 xprt_init(struct rpc_xprt *xprt, struct net *net);
66 static void	xprt_request_init(struct rpc_task *, struct rpc_xprt *);
67 static void	xprt_connect_status(struct rpc_task *task);
68 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
69 static void	 xprt_destroy(struct rpc_xprt *xprt);
70 
71 static DEFINE_SPINLOCK(xprt_list_lock);
72 static LIST_HEAD(xprt_list);
73 
74 /*
75  * The transport code maintains an estimate on the maximum number of out-
76  * standing RPC requests, using a smoothed version of the congestion
77  * avoidance implemented in 44BSD. This is basically the Van Jacobson
78  * congestion algorithm: If a retransmit occurs, the congestion window is
79  * halved; otherwise, it is incremented by 1/cwnd when
80  *
81  *	-	a reply is received and
82  *	-	a full number of requests are outstanding and
83  *	-	the congestion window hasn't been updated recently.
84  */
85 #define RPC_CWNDSHIFT		(8U)
86 #define RPC_CWNDSCALE		(1U << RPC_CWNDSHIFT)
87 #define RPC_INITCWND		RPC_CWNDSCALE
88 #define RPC_MAXCWND(xprt)	((xprt)->max_reqs << RPC_CWNDSHIFT)
89 
90 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
91 
92 /**
93  * xprt_register_transport - register a transport implementation
94  * @transport: transport to register
95  *
96  * If a transport implementation is loaded as a kernel module, it can
97  * call this interface to make itself known to the RPC client.
98  *
99  * Returns:
100  * 0:		transport successfully registered
101  * -EEXIST:	transport already registered
102  * -EINVAL:	transport module being unloaded
103  */
104 int xprt_register_transport(struct xprt_class *transport)
105 {
106 	struct xprt_class *t;
107 	int result;
108 
109 	result = -EEXIST;
110 	spin_lock(&xprt_list_lock);
111 	list_for_each_entry(t, &xprt_list, list) {
112 		/* don't register the same transport class twice */
113 		if (t->ident == transport->ident)
114 			goto out;
115 	}
116 
117 	list_add_tail(&transport->list, &xprt_list);
118 	printk(KERN_INFO "RPC: Registered %s transport module.\n",
119 	       transport->name);
120 	result = 0;
121 
122 out:
123 	spin_unlock(&xprt_list_lock);
124 	return result;
125 }
126 EXPORT_SYMBOL_GPL(xprt_register_transport);
127 
128 /**
129  * xprt_unregister_transport - unregister a transport implementation
130  * @transport: transport to unregister
131  *
132  * Returns:
133  * 0:		transport successfully unregistered
134  * -ENOENT:	transport never registered
135  */
136 int xprt_unregister_transport(struct xprt_class *transport)
137 {
138 	struct xprt_class *t;
139 	int result;
140 
141 	result = 0;
142 	spin_lock(&xprt_list_lock);
143 	list_for_each_entry(t, &xprt_list, list) {
144 		if (t == transport) {
145 			printk(KERN_INFO
146 				"RPC: Unregistered %s transport module.\n",
147 				transport->name);
148 			list_del_init(&transport->list);
149 			goto out;
150 		}
151 	}
152 	result = -ENOENT;
153 
154 out:
155 	spin_unlock(&xprt_list_lock);
156 	return result;
157 }
158 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
159 
160 /**
161  * xprt_load_transport - load a transport implementation
162  * @transport_name: transport to load
163  *
164  * Returns:
165  * 0:		transport successfully loaded
166  * -ENOENT:	transport module not available
167  */
168 int xprt_load_transport(const char *transport_name)
169 {
170 	struct xprt_class *t;
171 	int result;
172 
173 	result = 0;
174 	spin_lock(&xprt_list_lock);
175 	list_for_each_entry(t, &xprt_list, list) {
176 		if (strcmp(t->name, transport_name) == 0) {
177 			spin_unlock(&xprt_list_lock);
178 			goto out;
179 		}
180 	}
181 	spin_unlock(&xprt_list_lock);
182 	result = request_module("xprt%s", transport_name);
183 out:
184 	return result;
185 }
186 EXPORT_SYMBOL_GPL(xprt_load_transport);
187 
188 /**
189  * xprt_reserve_xprt - serialize write access to transports
190  * @task: task that is requesting access to the transport
191  * @xprt: pointer to the target transport
192  *
193  * This prevents mixing the payload of separate requests, and prevents
194  * transport connects from colliding with writes.  No congestion control
195  * is provided.
196  */
197 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
198 {
199 	struct rpc_rqst *req = task->tk_rqstp;
200 	int priority;
201 
202 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
203 		if (task == xprt->snd_task)
204 			return 1;
205 		goto out_sleep;
206 	}
207 	xprt->snd_task = task;
208 	if (req != NULL)
209 		req->rq_ntrans++;
210 
211 	return 1;
212 
213 out_sleep:
214 	dprintk("RPC: %5u failed to lock transport %p\n",
215 			task->tk_pid, xprt);
216 	task->tk_timeout = 0;
217 	task->tk_status = -EAGAIN;
218 	if (req == NULL)
219 		priority = RPC_PRIORITY_LOW;
220 	else if (!req->rq_ntrans)
221 		priority = RPC_PRIORITY_NORMAL;
222 	else
223 		priority = RPC_PRIORITY_HIGH;
224 	rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
225 	return 0;
226 }
227 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
228 
229 static void xprt_clear_locked(struct rpc_xprt *xprt)
230 {
231 	xprt->snd_task = NULL;
232 	if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
233 		smp_mb__before_clear_bit();
234 		clear_bit(XPRT_LOCKED, &xprt->state);
235 		smp_mb__after_clear_bit();
236 	} else
237 		queue_work(rpciod_workqueue, &xprt->task_cleanup);
238 }
239 
240 /*
241  * xprt_reserve_xprt_cong - serialize write access to transports
242  * @task: task that is requesting access to the transport
243  *
244  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
245  * integrated into the decision of whether a request is allowed to be
246  * woken up and given access to the transport.
247  */
248 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
249 {
250 	struct rpc_rqst *req = task->tk_rqstp;
251 	int priority;
252 
253 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
254 		if (task == xprt->snd_task)
255 			return 1;
256 		goto out_sleep;
257 	}
258 	if (req == NULL) {
259 		xprt->snd_task = task;
260 		return 1;
261 	}
262 	if (__xprt_get_cong(xprt, task)) {
263 		xprt->snd_task = task;
264 		req->rq_ntrans++;
265 		return 1;
266 	}
267 	xprt_clear_locked(xprt);
268 out_sleep:
269 	dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
270 	task->tk_timeout = 0;
271 	task->tk_status = -EAGAIN;
272 	if (req == NULL)
273 		priority = RPC_PRIORITY_LOW;
274 	else if (!req->rq_ntrans)
275 		priority = RPC_PRIORITY_NORMAL;
276 	else
277 		priority = RPC_PRIORITY_HIGH;
278 	rpc_sleep_on_priority(&xprt->sending, task, NULL, priority);
279 	return 0;
280 }
281 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
282 
283 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
284 {
285 	int retval;
286 
287 	spin_lock_bh(&xprt->transport_lock);
288 	retval = xprt->ops->reserve_xprt(xprt, task);
289 	spin_unlock_bh(&xprt->transport_lock);
290 	return retval;
291 }
292 
293 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
294 {
295 	struct rpc_xprt *xprt = data;
296 	struct rpc_rqst *req;
297 
298 	req = task->tk_rqstp;
299 	xprt->snd_task = task;
300 	if (req)
301 		req->rq_ntrans++;
302 	return true;
303 }
304 
305 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
306 {
307 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
308 		return;
309 
310 	if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_func, xprt))
311 		return;
312 	xprt_clear_locked(xprt);
313 }
314 
315 static bool __xprt_lock_write_cong_func(struct rpc_task *task, void *data)
316 {
317 	struct rpc_xprt *xprt = data;
318 	struct rpc_rqst *req;
319 
320 	req = task->tk_rqstp;
321 	if (req == NULL) {
322 		xprt->snd_task = task;
323 		return true;
324 	}
325 	if (__xprt_get_cong(xprt, task)) {
326 		xprt->snd_task = task;
327 		req->rq_ntrans++;
328 		return true;
329 	}
330 	return false;
331 }
332 
333 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
334 {
335 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
336 		return;
337 	if (RPCXPRT_CONGESTED(xprt))
338 		goto out_unlock;
339 	if (rpc_wake_up_first(&xprt->sending, __xprt_lock_write_cong_func, xprt))
340 		return;
341 out_unlock:
342 	xprt_clear_locked(xprt);
343 }
344 
345 /**
346  * xprt_release_xprt - allow other requests to use a transport
347  * @xprt: transport with other tasks potentially waiting
348  * @task: task that is releasing access to the transport
349  *
350  * Note that "task" can be NULL.  No congestion control is provided.
351  */
352 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
353 {
354 	if (xprt->snd_task == task) {
355 		if (task != NULL) {
356 			struct rpc_rqst *req = task->tk_rqstp;
357 			if (req != NULL)
358 				req->rq_bytes_sent = 0;
359 		}
360 		xprt_clear_locked(xprt);
361 		__xprt_lock_write_next(xprt);
362 	}
363 }
364 EXPORT_SYMBOL_GPL(xprt_release_xprt);
365 
366 /**
367  * xprt_release_xprt_cong - allow other requests to use a transport
368  * @xprt: transport with other tasks potentially waiting
369  * @task: task that is releasing access to the transport
370  *
371  * Note that "task" can be NULL.  Another task is awoken to use the
372  * transport if the transport's congestion window allows it.
373  */
374 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
375 {
376 	if (xprt->snd_task == task) {
377 		if (task != NULL) {
378 			struct rpc_rqst *req = task->tk_rqstp;
379 			if (req != NULL)
380 				req->rq_bytes_sent = 0;
381 		}
382 		xprt_clear_locked(xprt);
383 		__xprt_lock_write_next_cong(xprt);
384 	}
385 }
386 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
387 
388 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
389 {
390 	spin_lock_bh(&xprt->transport_lock);
391 	xprt->ops->release_xprt(xprt, task);
392 	spin_unlock_bh(&xprt->transport_lock);
393 }
394 
395 /*
396  * Van Jacobson congestion avoidance. Check if the congestion window
397  * overflowed. Put the task to sleep if this is the case.
398  */
399 static int
400 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
401 {
402 	struct rpc_rqst *req = task->tk_rqstp;
403 
404 	if (req->rq_cong)
405 		return 1;
406 	dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
407 			task->tk_pid, xprt->cong, xprt->cwnd);
408 	if (RPCXPRT_CONGESTED(xprt))
409 		return 0;
410 	req->rq_cong = 1;
411 	xprt->cong += RPC_CWNDSCALE;
412 	return 1;
413 }
414 
415 /*
416  * Adjust the congestion window, and wake up the next task
417  * that has been sleeping due to congestion
418  */
419 static void
420 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
421 {
422 	if (!req->rq_cong)
423 		return;
424 	req->rq_cong = 0;
425 	xprt->cong -= RPC_CWNDSCALE;
426 	__xprt_lock_write_next_cong(xprt);
427 }
428 
429 /**
430  * xprt_release_rqst_cong - housekeeping when request is complete
431  * @task: RPC request that recently completed
432  *
433  * Useful for transports that require congestion control.
434  */
435 void xprt_release_rqst_cong(struct rpc_task *task)
436 {
437 	struct rpc_rqst *req = task->tk_rqstp;
438 
439 	__xprt_put_cong(req->rq_xprt, req);
440 }
441 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
442 
443 /**
444  * xprt_adjust_cwnd - adjust transport congestion window
445  * @xprt: pointer to xprt
446  * @task: recently completed RPC request used to adjust window
447  * @result: result code of completed RPC request
448  *
449  * We use a time-smoothed congestion estimator to avoid heavy oscillation.
450  */
451 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
452 {
453 	struct rpc_rqst *req = task->tk_rqstp;
454 	unsigned long cwnd = xprt->cwnd;
455 
456 	if (result >= 0 && cwnd <= xprt->cong) {
457 		/* The (cwnd >> 1) term makes sure
458 		 * the result gets rounded properly. */
459 		cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
460 		if (cwnd > RPC_MAXCWND(xprt))
461 			cwnd = RPC_MAXCWND(xprt);
462 		__xprt_lock_write_next_cong(xprt);
463 	} else if (result == -ETIMEDOUT) {
464 		cwnd >>= 1;
465 		if (cwnd < RPC_CWNDSCALE)
466 			cwnd = RPC_CWNDSCALE;
467 	}
468 	dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
469 			xprt->cong, xprt->cwnd, cwnd);
470 	xprt->cwnd = cwnd;
471 	__xprt_put_cong(xprt, req);
472 }
473 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
474 
475 /**
476  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
477  * @xprt: transport with waiting tasks
478  * @status: result code to plant in each task before waking it
479  *
480  */
481 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
482 {
483 	if (status < 0)
484 		rpc_wake_up_status(&xprt->pending, status);
485 	else
486 		rpc_wake_up(&xprt->pending);
487 }
488 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
489 
490 /**
491  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
492  * @task: task to be put to sleep
493  * @action: function pointer to be executed after wait
494  *
495  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
496  * we don't in general want to force a socket disconnection due to
497  * an incomplete RPC call transmission.
498  */
499 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
500 {
501 	struct rpc_rqst *req = task->tk_rqstp;
502 	struct rpc_xprt *xprt = req->rq_xprt;
503 
504 	task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
505 	rpc_sleep_on(&xprt->pending, task, action);
506 }
507 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
508 
509 /**
510  * xprt_write_space - wake the task waiting for transport output buffer space
511  * @xprt: transport with waiting tasks
512  *
513  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
514  */
515 void xprt_write_space(struct rpc_xprt *xprt)
516 {
517 	spin_lock_bh(&xprt->transport_lock);
518 	if (xprt->snd_task) {
519 		dprintk("RPC:       write space: waking waiting task on "
520 				"xprt %p\n", xprt);
521 		rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
522 	}
523 	spin_unlock_bh(&xprt->transport_lock);
524 }
525 EXPORT_SYMBOL_GPL(xprt_write_space);
526 
527 /**
528  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
529  * @task: task whose timeout is to be set
530  *
531  * Set a request's retransmit timeout based on the transport's
532  * default timeout parameters.  Used by transports that don't adjust
533  * the retransmit timeout based on round-trip time estimation.
534  */
535 void xprt_set_retrans_timeout_def(struct rpc_task *task)
536 {
537 	task->tk_timeout = task->tk_rqstp->rq_timeout;
538 }
539 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
540 
541 /**
542  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
543  * @task: task whose timeout is to be set
544  *
545  * Set a request's retransmit timeout using the RTT estimator.
546  */
547 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
548 {
549 	int timer = task->tk_msg.rpc_proc->p_timer;
550 	struct rpc_clnt *clnt = task->tk_client;
551 	struct rpc_rtt *rtt = clnt->cl_rtt;
552 	struct rpc_rqst *req = task->tk_rqstp;
553 	unsigned long max_timeout = clnt->cl_timeout->to_maxval;
554 
555 	task->tk_timeout = rpc_calc_rto(rtt, timer);
556 	task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
557 	if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
558 		task->tk_timeout = max_timeout;
559 }
560 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
561 
562 static void xprt_reset_majortimeo(struct rpc_rqst *req)
563 {
564 	const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
565 
566 	req->rq_majortimeo = req->rq_timeout;
567 	if (to->to_exponential)
568 		req->rq_majortimeo <<= to->to_retries;
569 	else
570 		req->rq_majortimeo += to->to_increment * to->to_retries;
571 	if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
572 		req->rq_majortimeo = to->to_maxval;
573 	req->rq_majortimeo += jiffies;
574 }
575 
576 /**
577  * xprt_adjust_timeout - adjust timeout values for next retransmit
578  * @req: RPC request containing parameters to use for the adjustment
579  *
580  */
581 int xprt_adjust_timeout(struct rpc_rqst *req)
582 {
583 	struct rpc_xprt *xprt = req->rq_xprt;
584 	const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
585 	int status = 0;
586 
587 	if (time_before(jiffies, req->rq_majortimeo)) {
588 		if (to->to_exponential)
589 			req->rq_timeout <<= 1;
590 		else
591 			req->rq_timeout += to->to_increment;
592 		if (to->to_maxval && req->rq_timeout >= to->to_maxval)
593 			req->rq_timeout = to->to_maxval;
594 		req->rq_retries++;
595 	} else {
596 		req->rq_timeout = to->to_initval;
597 		req->rq_retries = 0;
598 		xprt_reset_majortimeo(req);
599 		/* Reset the RTT counters == "slow start" */
600 		spin_lock_bh(&xprt->transport_lock);
601 		rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
602 		spin_unlock_bh(&xprt->transport_lock);
603 		status = -ETIMEDOUT;
604 	}
605 
606 	if (req->rq_timeout == 0) {
607 		printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
608 		req->rq_timeout = 5 * HZ;
609 	}
610 	return status;
611 }
612 
613 static void xprt_autoclose(struct work_struct *work)
614 {
615 	struct rpc_xprt *xprt =
616 		container_of(work, struct rpc_xprt, task_cleanup);
617 
618 	xprt->ops->close(xprt);
619 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
620 	xprt_release_write(xprt, NULL);
621 }
622 
623 /**
624  * xprt_disconnect_done - mark a transport as disconnected
625  * @xprt: transport to flag for disconnect
626  *
627  */
628 void xprt_disconnect_done(struct rpc_xprt *xprt)
629 {
630 	dprintk("RPC:       disconnected transport %p\n", xprt);
631 	spin_lock_bh(&xprt->transport_lock);
632 	xprt_clear_connected(xprt);
633 	xprt_wake_pending_tasks(xprt, -EAGAIN);
634 	spin_unlock_bh(&xprt->transport_lock);
635 }
636 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
637 
638 /**
639  * xprt_force_disconnect - force a transport to disconnect
640  * @xprt: transport to disconnect
641  *
642  */
643 void xprt_force_disconnect(struct rpc_xprt *xprt)
644 {
645 	/* Don't race with the test_bit() in xprt_clear_locked() */
646 	spin_lock_bh(&xprt->transport_lock);
647 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
648 	/* Try to schedule an autoclose RPC call */
649 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
650 		queue_work(rpciod_workqueue, &xprt->task_cleanup);
651 	xprt_wake_pending_tasks(xprt, -EAGAIN);
652 	spin_unlock_bh(&xprt->transport_lock);
653 }
654 
655 /**
656  * xprt_conditional_disconnect - force a transport to disconnect
657  * @xprt: transport to disconnect
658  * @cookie: 'connection cookie'
659  *
660  * This attempts to break the connection if and only if 'cookie' matches
661  * the current transport 'connection cookie'. It ensures that we don't
662  * try to break the connection more than once when we need to retransmit
663  * a batch of RPC requests.
664  *
665  */
666 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
667 {
668 	/* Don't race with the test_bit() in xprt_clear_locked() */
669 	spin_lock_bh(&xprt->transport_lock);
670 	if (cookie != xprt->connect_cookie)
671 		goto out;
672 	if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
673 		goto out;
674 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
675 	/* Try to schedule an autoclose RPC call */
676 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
677 		queue_work(rpciod_workqueue, &xprt->task_cleanup);
678 	xprt_wake_pending_tasks(xprt, -EAGAIN);
679 out:
680 	spin_unlock_bh(&xprt->transport_lock);
681 }
682 
683 static void
684 xprt_init_autodisconnect(unsigned long data)
685 {
686 	struct rpc_xprt *xprt = (struct rpc_xprt *)data;
687 
688 	spin_lock(&xprt->transport_lock);
689 	if (!list_empty(&xprt->recv))
690 		goto out_abort;
691 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
692 		goto out_abort;
693 	spin_unlock(&xprt->transport_lock);
694 	set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
695 	queue_work(rpciod_workqueue, &xprt->task_cleanup);
696 	return;
697 out_abort:
698 	spin_unlock(&xprt->transport_lock);
699 }
700 
701 /**
702  * xprt_connect - schedule a transport connect operation
703  * @task: RPC task that is requesting the connect
704  *
705  */
706 void xprt_connect(struct rpc_task *task)
707 {
708 	struct rpc_xprt	*xprt = task->tk_rqstp->rq_xprt;
709 
710 	dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
711 			xprt, (xprt_connected(xprt) ? "is" : "is not"));
712 
713 	if (!xprt_bound(xprt)) {
714 		task->tk_status = -EAGAIN;
715 		return;
716 	}
717 	if (!xprt_lock_write(xprt, task))
718 		return;
719 
720 	if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
721 		xprt->ops->close(xprt);
722 
723 	if (xprt_connected(xprt))
724 		xprt_release_write(xprt, task);
725 	else {
726 		task->tk_rqstp->rq_bytes_sent = 0;
727 		task->tk_timeout = task->tk_rqstp->rq_timeout;
728 		rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
729 
730 		if (test_bit(XPRT_CLOSING, &xprt->state))
731 			return;
732 		if (xprt_test_and_set_connecting(xprt))
733 			return;
734 		xprt->stat.connect_start = jiffies;
735 		xprt->ops->connect(xprt, task);
736 	}
737 }
738 
739 static void xprt_connect_status(struct rpc_task *task)
740 {
741 	struct rpc_xprt	*xprt = task->tk_rqstp->rq_xprt;
742 
743 	if (task->tk_status == 0) {
744 		xprt->stat.connect_count++;
745 		xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
746 		dprintk("RPC: %5u xprt_connect_status: connection established\n",
747 				task->tk_pid);
748 		return;
749 	}
750 
751 	switch (task->tk_status) {
752 	case -EAGAIN:
753 		dprintk("RPC: %5u xprt_connect_status: retrying\n", task->tk_pid);
754 		break;
755 	case -ETIMEDOUT:
756 		dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
757 				"out\n", task->tk_pid);
758 		break;
759 	default:
760 		dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
761 				"server %s\n", task->tk_pid, -task->tk_status,
762 				xprt->servername);
763 		xprt_release_write(xprt, task);
764 		task->tk_status = -EIO;
765 	}
766 }
767 
768 /**
769  * xprt_lookup_rqst - find an RPC request corresponding to an XID
770  * @xprt: transport on which the original request was transmitted
771  * @xid: RPC XID of incoming reply
772  *
773  */
774 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
775 {
776 	struct rpc_rqst *entry;
777 
778 	list_for_each_entry(entry, &xprt->recv, rq_list)
779 		if (entry->rq_xid == xid)
780 			return entry;
781 
782 	dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
783 			ntohl(xid));
784 	xprt->stat.bad_xids++;
785 	return NULL;
786 }
787 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
788 
789 static void xprt_update_rtt(struct rpc_task *task)
790 {
791 	struct rpc_rqst *req = task->tk_rqstp;
792 	struct rpc_rtt *rtt = task->tk_client->cl_rtt;
793 	unsigned int timer = task->tk_msg.rpc_proc->p_timer;
794 	long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
795 
796 	if (timer) {
797 		if (req->rq_ntrans == 1)
798 			rpc_update_rtt(rtt, timer, m);
799 		rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
800 	}
801 }
802 
803 /**
804  * xprt_complete_rqst - called when reply processing is complete
805  * @task: RPC request that recently completed
806  * @copied: actual number of bytes received from the transport
807  *
808  * Caller holds transport lock.
809  */
810 void xprt_complete_rqst(struct rpc_task *task, int copied)
811 {
812 	struct rpc_rqst *req = task->tk_rqstp;
813 	struct rpc_xprt *xprt = req->rq_xprt;
814 
815 	dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
816 			task->tk_pid, ntohl(req->rq_xid), copied);
817 
818 	xprt->stat.recvs++;
819 	req->rq_rtt = ktime_sub(ktime_get(), req->rq_xtime);
820 	if (xprt->ops->timer != NULL)
821 		xprt_update_rtt(task);
822 
823 	list_del_init(&req->rq_list);
824 	req->rq_private_buf.len = copied;
825 	/* Ensure all writes are done before we update */
826 	/* req->rq_reply_bytes_recvd */
827 	smp_wmb();
828 	req->rq_reply_bytes_recvd = copied;
829 	rpc_wake_up_queued_task(&xprt->pending, task);
830 }
831 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
832 
833 static void xprt_timer(struct rpc_task *task)
834 {
835 	struct rpc_rqst *req = task->tk_rqstp;
836 	struct rpc_xprt *xprt = req->rq_xprt;
837 
838 	if (task->tk_status != -ETIMEDOUT)
839 		return;
840 	dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
841 
842 	spin_lock_bh(&xprt->transport_lock);
843 	if (!req->rq_reply_bytes_recvd) {
844 		if (xprt->ops->timer)
845 			xprt->ops->timer(xprt, task);
846 	} else
847 		task->tk_status = 0;
848 	spin_unlock_bh(&xprt->transport_lock);
849 }
850 
851 static inline int xprt_has_timer(struct rpc_xprt *xprt)
852 {
853 	return xprt->idle_timeout != 0;
854 }
855 
856 /**
857  * xprt_prepare_transmit - reserve the transport before sending a request
858  * @task: RPC task about to send a request
859  *
860  */
861 bool xprt_prepare_transmit(struct rpc_task *task)
862 {
863 	struct rpc_rqst	*req = task->tk_rqstp;
864 	struct rpc_xprt	*xprt = req->rq_xprt;
865 	bool ret = false;
866 
867 	dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
868 
869 	spin_lock_bh(&xprt->transport_lock);
870 	if (!req->rq_bytes_sent) {
871 		if (req->rq_reply_bytes_recvd) {
872 			task->tk_status = req->rq_reply_bytes_recvd;
873 			goto out_unlock;
874 		}
875 		if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
876 		    && xprt_connected(xprt)
877 		    && req->rq_connect_cookie == xprt->connect_cookie) {
878 			xprt->ops->set_retrans_timeout(task);
879 			rpc_sleep_on(&xprt->pending, task, xprt_timer);
880 			goto out_unlock;
881 		}
882 	}
883 	if (!xprt->ops->reserve_xprt(xprt, task)) {
884 		task->tk_status = -EAGAIN;
885 		goto out_unlock;
886 	}
887 	ret = true;
888 out_unlock:
889 	spin_unlock_bh(&xprt->transport_lock);
890 	return ret;
891 }
892 
893 void xprt_end_transmit(struct rpc_task *task)
894 {
895 	xprt_release_write(task->tk_rqstp->rq_xprt, task);
896 }
897 
898 /**
899  * xprt_transmit - send an RPC request on a transport
900  * @task: controlling RPC task
901  *
902  * We have to copy the iovec because sendmsg fiddles with its contents.
903  */
904 void xprt_transmit(struct rpc_task *task)
905 {
906 	struct rpc_rqst	*req = task->tk_rqstp;
907 	struct rpc_xprt	*xprt = req->rq_xprt;
908 	int status, numreqs;
909 
910 	dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
911 
912 	if (!req->rq_reply_bytes_recvd) {
913 		if (list_empty(&req->rq_list) && rpc_reply_expected(task)) {
914 			/*
915 			 * Add to the list only if we're expecting a reply
916 			 */
917 			spin_lock_bh(&xprt->transport_lock);
918 			/* Update the softirq receive buffer */
919 			memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
920 					sizeof(req->rq_private_buf));
921 			/* Add request to the receive list */
922 			list_add_tail(&req->rq_list, &xprt->recv);
923 			spin_unlock_bh(&xprt->transport_lock);
924 			xprt_reset_majortimeo(req);
925 			/* Turn off autodisconnect */
926 			del_singleshot_timer_sync(&xprt->timer);
927 		}
928 	} else if (!req->rq_bytes_sent)
929 		return;
930 
931 	req->rq_xtime = ktime_get();
932 	status = xprt->ops->send_request(task);
933 	if (status != 0) {
934 		task->tk_status = status;
935 		return;
936 	}
937 
938 	dprintk("RPC: %5u xmit complete\n", task->tk_pid);
939 	task->tk_flags |= RPC_TASK_SENT;
940 	spin_lock_bh(&xprt->transport_lock);
941 
942 	xprt->ops->set_retrans_timeout(task);
943 
944 	numreqs = atomic_read(&xprt->num_reqs);
945 	if (numreqs > xprt->stat.max_slots)
946 		xprt->stat.max_slots = numreqs;
947 	xprt->stat.sends++;
948 	xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
949 	xprt->stat.bklog_u += xprt->backlog.qlen;
950 	xprt->stat.sending_u += xprt->sending.qlen;
951 	xprt->stat.pending_u += xprt->pending.qlen;
952 
953 	/* Don't race with disconnect */
954 	if (!xprt_connected(xprt))
955 		task->tk_status = -ENOTCONN;
956 	else {
957 		/*
958 		 * Sleep on the pending queue since
959 		 * we're expecting a reply.
960 		 */
961 		if (!req->rq_reply_bytes_recvd && rpc_reply_expected(task))
962 			rpc_sleep_on(&xprt->pending, task, xprt_timer);
963 		req->rq_connect_cookie = xprt->connect_cookie;
964 	}
965 	spin_unlock_bh(&xprt->transport_lock);
966 }
967 
968 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
969 {
970 	set_bit(XPRT_CONGESTED, &xprt->state);
971 	rpc_sleep_on(&xprt->backlog, task, NULL);
972 }
973 
974 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
975 {
976 	if (rpc_wake_up_next(&xprt->backlog) == NULL)
977 		clear_bit(XPRT_CONGESTED, &xprt->state);
978 }
979 
980 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
981 {
982 	bool ret = false;
983 
984 	if (!test_bit(XPRT_CONGESTED, &xprt->state))
985 		goto out;
986 	spin_lock(&xprt->reserve_lock);
987 	if (test_bit(XPRT_CONGESTED, &xprt->state)) {
988 		rpc_sleep_on(&xprt->backlog, task, NULL);
989 		ret = true;
990 	}
991 	spin_unlock(&xprt->reserve_lock);
992 out:
993 	return ret;
994 }
995 
996 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
997 {
998 	struct rpc_rqst *req = ERR_PTR(-EAGAIN);
999 
1000 	if (!atomic_add_unless(&xprt->num_reqs, 1, xprt->max_reqs))
1001 		goto out;
1002 	req = kzalloc(sizeof(struct rpc_rqst), gfp_flags);
1003 	if (req != NULL)
1004 		goto out;
1005 	atomic_dec(&xprt->num_reqs);
1006 	req = ERR_PTR(-ENOMEM);
1007 out:
1008 	return req;
1009 }
1010 
1011 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1012 {
1013 	if (atomic_add_unless(&xprt->num_reqs, -1, xprt->min_reqs)) {
1014 		kfree(req);
1015 		return true;
1016 	}
1017 	return false;
1018 }
1019 
1020 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1021 {
1022 	struct rpc_rqst *req;
1023 
1024 	spin_lock(&xprt->reserve_lock);
1025 	if (!list_empty(&xprt->free)) {
1026 		req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1027 		list_del(&req->rq_list);
1028 		goto out_init_req;
1029 	}
1030 	req = xprt_dynamic_alloc_slot(xprt, GFP_NOWAIT|__GFP_NOWARN);
1031 	if (!IS_ERR(req))
1032 		goto out_init_req;
1033 	switch (PTR_ERR(req)) {
1034 	case -ENOMEM:
1035 		dprintk("RPC:       dynamic allocation of request slot "
1036 				"failed! Retrying\n");
1037 		task->tk_status = -ENOMEM;
1038 		break;
1039 	case -EAGAIN:
1040 		xprt_add_backlog(xprt, task);
1041 		dprintk("RPC:       waiting for request slot\n");
1042 	default:
1043 		task->tk_status = -EAGAIN;
1044 	}
1045 	spin_unlock(&xprt->reserve_lock);
1046 	return;
1047 out_init_req:
1048 	task->tk_status = 0;
1049 	task->tk_rqstp = req;
1050 	xprt_request_init(task, xprt);
1051 	spin_unlock(&xprt->reserve_lock);
1052 }
1053 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1054 
1055 void xprt_lock_and_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1056 {
1057 	/* Note: grabbing the xprt_lock_write() ensures that we throttle
1058 	 * new slot allocation if the transport is congested (i.e. when
1059 	 * reconnecting a stream transport or when out of socket write
1060 	 * buffer space).
1061 	 */
1062 	if (xprt_lock_write(xprt, task)) {
1063 		xprt_alloc_slot(xprt, task);
1064 		xprt_release_write(xprt, task);
1065 	}
1066 }
1067 EXPORT_SYMBOL_GPL(xprt_lock_and_alloc_slot);
1068 
1069 static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1070 {
1071 	spin_lock(&xprt->reserve_lock);
1072 	if (!xprt_dynamic_free_slot(xprt, req)) {
1073 		memset(req, 0, sizeof(*req));	/* mark unused */
1074 		list_add(&req->rq_list, &xprt->free);
1075 	}
1076 	xprt_wake_up_backlog(xprt);
1077 	spin_unlock(&xprt->reserve_lock);
1078 }
1079 
1080 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1081 {
1082 	struct rpc_rqst *req;
1083 	while (!list_empty(&xprt->free)) {
1084 		req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1085 		list_del(&req->rq_list);
1086 		kfree(req);
1087 	}
1088 }
1089 
1090 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1091 		unsigned int num_prealloc,
1092 		unsigned int max_alloc)
1093 {
1094 	struct rpc_xprt *xprt;
1095 	struct rpc_rqst *req;
1096 	int i;
1097 
1098 	xprt = kzalloc(size, GFP_KERNEL);
1099 	if (xprt == NULL)
1100 		goto out;
1101 
1102 	xprt_init(xprt, net);
1103 
1104 	for (i = 0; i < num_prealloc; i++) {
1105 		req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1106 		if (!req)
1107 			goto out_free;
1108 		list_add(&req->rq_list, &xprt->free);
1109 	}
1110 	if (max_alloc > num_prealloc)
1111 		xprt->max_reqs = max_alloc;
1112 	else
1113 		xprt->max_reqs = num_prealloc;
1114 	xprt->min_reqs = num_prealloc;
1115 	atomic_set(&xprt->num_reqs, num_prealloc);
1116 
1117 	return xprt;
1118 
1119 out_free:
1120 	xprt_free(xprt);
1121 out:
1122 	return NULL;
1123 }
1124 EXPORT_SYMBOL_GPL(xprt_alloc);
1125 
1126 void xprt_free(struct rpc_xprt *xprt)
1127 {
1128 	put_net(xprt->xprt_net);
1129 	xprt_free_all_slots(xprt);
1130 	kfree(xprt);
1131 }
1132 EXPORT_SYMBOL_GPL(xprt_free);
1133 
1134 /**
1135  * xprt_reserve - allocate an RPC request slot
1136  * @task: RPC task requesting a slot allocation
1137  *
1138  * If the transport is marked as being congested, or if no more
1139  * slots are available, place the task on the transport's
1140  * backlog queue.
1141  */
1142 void xprt_reserve(struct rpc_task *task)
1143 {
1144 	struct rpc_xprt	*xprt;
1145 
1146 	task->tk_status = 0;
1147 	if (task->tk_rqstp != NULL)
1148 		return;
1149 
1150 	task->tk_timeout = 0;
1151 	task->tk_status = -EAGAIN;
1152 	rcu_read_lock();
1153 	xprt = rcu_dereference(task->tk_client->cl_xprt);
1154 	if (!xprt_throttle_congested(xprt, task))
1155 		xprt->ops->alloc_slot(xprt, task);
1156 	rcu_read_unlock();
1157 }
1158 
1159 /**
1160  * xprt_retry_reserve - allocate an RPC request slot
1161  * @task: RPC task requesting a slot allocation
1162  *
1163  * If no more slots are available, place the task on the transport's
1164  * backlog queue.
1165  * Note that the only difference with xprt_reserve is that we now
1166  * ignore the value of the XPRT_CONGESTED flag.
1167  */
1168 void xprt_retry_reserve(struct rpc_task *task)
1169 {
1170 	struct rpc_xprt	*xprt;
1171 
1172 	task->tk_status = 0;
1173 	if (task->tk_rqstp != NULL)
1174 		return;
1175 
1176 	task->tk_timeout = 0;
1177 	task->tk_status = -EAGAIN;
1178 	rcu_read_lock();
1179 	xprt = rcu_dereference(task->tk_client->cl_xprt);
1180 	xprt->ops->alloc_slot(xprt, task);
1181 	rcu_read_unlock();
1182 }
1183 
1184 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
1185 {
1186 	return (__force __be32)xprt->xid++;
1187 }
1188 
1189 static inline void xprt_init_xid(struct rpc_xprt *xprt)
1190 {
1191 	xprt->xid = net_random();
1192 }
1193 
1194 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1195 {
1196 	struct rpc_rqst	*req = task->tk_rqstp;
1197 
1198 	INIT_LIST_HEAD(&req->rq_list);
1199 	req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1200 	req->rq_task	= task;
1201 	req->rq_xprt    = xprt;
1202 	req->rq_buffer  = NULL;
1203 	req->rq_xid     = xprt_alloc_xid(xprt);
1204 	req->rq_connect_cookie = xprt->connect_cookie - 1;
1205 	req->rq_bytes_sent = 0;
1206 	req->rq_snd_buf.len = 0;
1207 	req->rq_snd_buf.buflen = 0;
1208 	req->rq_rcv_buf.len = 0;
1209 	req->rq_rcv_buf.buflen = 0;
1210 	req->rq_release_snd_buf = NULL;
1211 	xprt_reset_majortimeo(req);
1212 	dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1213 			req, ntohl(req->rq_xid));
1214 }
1215 
1216 /**
1217  * xprt_release - release an RPC request slot
1218  * @task: task which is finished with the slot
1219  *
1220  */
1221 void xprt_release(struct rpc_task *task)
1222 {
1223 	struct rpc_xprt	*xprt;
1224 	struct rpc_rqst	*req = task->tk_rqstp;
1225 
1226 	if (req == NULL) {
1227 		if (task->tk_client) {
1228 			rcu_read_lock();
1229 			xprt = rcu_dereference(task->tk_client->cl_xprt);
1230 			if (xprt->snd_task == task)
1231 				xprt_release_write(xprt, task);
1232 			rcu_read_unlock();
1233 		}
1234 		return;
1235 	}
1236 
1237 	xprt = req->rq_xprt;
1238 	if (task->tk_ops->rpc_count_stats != NULL)
1239 		task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1240 	else if (task->tk_client)
1241 		rpc_count_iostats(task, task->tk_client->cl_metrics);
1242 	spin_lock_bh(&xprt->transport_lock);
1243 	xprt->ops->release_xprt(xprt, task);
1244 	if (xprt->ops->release_request)
1245 		xprt->ops->release_request(task);
1246 	if (!list_empty(&req->rq_list))
1247 		list_del(&req->rq_list);
1248 	xprt->last_used = jiffies;
1249 	if (list_empty(&xprt->recv) && xprt_has_timer(xprt))
1250 		mod_timer(&xprt->timer,
1251 				xprt->last_used + xprt->idle_timeout);
1252 	spin_unlock_bh(&xprt->transport_lock);
1253 	if (req->rq_buffer)
1254 		xprt->ops->buf_free(req->rq_buffer);
1255 	if (req->rq_cred != NULL)
1256 		put_rpccred(req->rq_cred);
1257 	task->tk_rqstp = NULL;
1258 	if (req->rq_release_snd_buf)
1259 		req->rq_release_snd_buf(req);
1260 
1261 	dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1262 	if (likely(!bc_prealloc(req)))
1263 		xprt_free_slot(xprt, req);
1264 	else
1265 		xprt_free_bc_request(req);
1266 }
1267 
1268 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1269 {
1270 	atomic_set(&xprt->count, 1);
1271 
1272 	spin_lock_init(&xprt->transport_lock);
1273 	spin_lock_init(&xprt->reserve_lock);
1274 
1275 	INIT_LIST_HEAD(&xprt->free);
1276 	INIT_LIST_HEAD(&xprt->recv);
1277 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1278 	spin_lock_init(&xprt->bc_pa_lock);
1279 	INIT_LIST_HEAD(&xprt->bc_pa_list);
1280 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1281 
1282 	xprt->last_used = jiffies;
1283 	xprt->cwnd = RPC_INITCWND;
1284 	xprt->bind_index = 0;
1285 
1286 	rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1287 	rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1288 	rpc_init_priority_wait_queue(&xprt->sending, "xprt_sending");
1289 	rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1290 
1291 	xprt_init_xid(xprt);
1292 
1293 	xprt->xprt_net = get_net(net);
1294 }
1295 
1296 /**
1297  * xprt_create_transport - create an RPC transport
1298  * @args: rpc transport creation arguments
1299  *
1300  */
1301 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1302 {
1303 	struct rpc_xprt	*xprt;
1304 	struct xprt_class *t;
1305 
1306 	spin_lock(&xprt_list_lock);
1307 	list_for_each_entry(t, &xprt_list, list) {
1308 		if (t->ident == args->ident) {
1309 			spin_unlock(&xprt_list_lock);
1310 			goto found;
1311 		}
1312 	}
1313 	spin_unlock(&xprt_list_lock);
1314 	printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1315 	return ERR_PTR(-EIO);
1316 
1317 found:
1318 	xprt = t->setup(args);
1319 	if (IS_ERR(xprt)) {
1320 		dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1321 				-PTR_ERR(xprt));
1322 		goto out;
1323 	}
1324 	if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1325 		xprt->idle_timeout = 0;
1326 	INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1327 	if (xprt_has_timer(xprt))
1328 		setup_timer(&xprt->timer, xprt_init_autodisconnect,
1329 			    (unsigned long)xprt);
1330 	else
1331 		init_timer(&xprt->timer);
1332 
1333 	if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1334 		xprt_destroy(xprt);
1335 		return ERR_PTR(-EINVAL);
1336 	}
1337 	xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1338 	if (xprt->servername == NULL) {
1339 		xprt_destroy(xprt);
1340 		return ERR_PTR(-ENOMEM);
1341 	}
1342 
1343 	dprintk("RPC:       created transport %p with %u slots\n", xprt,
1344 			xprt->max_reqs);
1345 out:
1346 	return xprt;
1347 }
1348 
1349 /**
1350  * xprt_destroy - destroy an RPC transport, killing off all requests.
1351  * @xprt: transport to destroy
1352  *
1353  */
1354 static void xprt_destroy(struct rpc_xprt *xprt)
1355 {
1356 	dprintk("RPC:       destroying transport %p\n", xprt);
1357 	del_timer_sync(&xprt->timer);
1358 
1359 	rpc_destroy_wait_queue(&xprt->binding);
1360 	rpc_destroy_wait_queue(&xprt->pending);
1361 	rpc_destroy_wait_queue(&xprt->sending);
1362 	rpc_destroy_wait_queue(&xprt->backlog);
1363 	cancel_work_sync(&xprt->task_cleanup);
1364 	kfree(xprt->servername);
1365 	/*
1366 	 * Tear down transport state and free the rpc_xprt
1367 	 */
1368 	xprt->ops->destroy(xprt);
1369 }
1370 
1371 /**
1372  * xprt_put - release a reference to an RPC transport.
1373  * @xprt: pointer to the transport
1374  *
1375  */
1376 void xprt_put(struct rpc_xprt *xprt)
1377 {
1378 	if (atomic_dec_and_test(&xprt->count))
1379 		xprt_destroy(xprt);
1380 }
1381 
1382 /**
1383  * xprt_get - return a reference to an RPC transport.
1384  * @xprt: pointer to the transport
1385  *
1386  */
1387 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1388 {
1389 	if (atomic_inc_not_zero(&xprt->count))
1390 		return xprt;
1391 	return NULL;
1392 }
1393