xref: /openbmc/linux/net/sunrpc/xprt.c (revision 367b8112)
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, it installs a timer that
16  *	is run after the packet's timeout has expired.
17  *  -	When a packet arrives, the data_ready handler walks the list of
18  *	pending requests for that transport. If a matching XID is found, the
19  *	caller is woken up, and the timer removed.
20  *  -	When no reply arrives within the timeout interval, the timer is
21  *	fired by the kernel and runs xprt_timer(). It either adjusts the
22  *	timeout values (minor timeout) or wakes up the caller with a status
23  *	of -ETIMEDOUT.
24  *  -	When the caller receives a notification from RPC that a reply arrived,
25  *	it should release the RPC slot, and process the reply.
26  *	If the call timed out, it may choose to retry the operation by
27  *	adjusting the initial timeout value, and simply calling rpc_call
28  *	again.
29  *
30  *  Support for async RPC is done through a set of RPC-specific scheduling
31  *  primitives that `transparently' work for processes as well as async
32  *  tasks that rely on callbacks.
33  *
34  *  Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
35  *
36  *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
37  */
38 
39 #include <linux/module.h>
40 
41 #include <linux/types.h>
42 #include <linux/interrupt.h>
43 #include <linux/workqueue.h>
44 #include <linux/net.h>
45 
46 #include <linux/sunrpc/clnt.h>
47 #include <linux/sunrpc/metrics.h>
48 
49 /*
50  * Local variables
51  */
52 
53 #ifdef RPC_DEBUG
54 # define RPCDBG_FACILITY	RPCDBG_XPRT
55 #endif
56 
57 /*
58  * Local functions
59  */
60 static void	xprt_request_init(struct rpc_task *, struct rpc_xprt *);
61 static inline void	do_xprt_reserve(struct rpc_task *);
62 static void	xprt_connect_status(struct rpc_task *task);
63 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
64 
65 static DEFINE_SPINLOCK(xprt_list_lock);
66 static LIST_HEAD(xprt_list);
67 
68 /*
69  * The transport code maintains an estimate on the maximum number of out-
70  * standing RPC requests, using a smoothed version of the congestion
71  * avoidance implemented in 44BSD. This is basically the Van Jacobson
72  * congestion algorithm: If a retransmit occurs, the congestion window is
73  * halved; otherwise, it is incremented by 1/cwnd when
74  *
75  *	-	a reply is received and
76  *	-	a full number of requests are outstanding and
77  *	-	the congestion window hasn't been updated recently.
78  */
79 #define RPC_CWNDSHIFT		(8U)
80 #define RPC_CWNDSCALE		(1U << RPC_CWNDSHIFT)
81 #define RPC_INITCWND		RPC_CWNDSCALE
82 #define RPC_MAXCWND(xprt)	((xprt)->max_reqs << RPC_CWNDSHIFT)
83 
84 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
85 
86 /**
87  * xprt_register_transport - register a transport implementation
88  * @transport: transport to register
89  *
90  * If a transport implementation is loaded as a kernel module, it can
91  * call this interface to make itself known to the RPC client.
92  *
93  * Returns:
94  * 0:		transport successfully registered
95  * -EEXIST:	transport already registered
96  * -EINVAL:	transport module being unloaded
97  */
98 int xprt_register_transport(struct xprt_class *transport)
99 {
100 	struct xprt_class *t;
101 	int result;
102 
103 	result = -EEXIST;
104 	spin_lock(&xprt_list_lock);
105 	list_for_each_entry(t, &xprt_list, list) {
106 		/* don't register the same transport class twice */
107 		if (t->ident == transport->ident)
108 			goto out;
109 	}
110 
111 	list_add_tail(&transport->list, &xprt_list);
112 	printk(KERN_INFO "RPC: Registered %s transport module.\n",
113 	       transport->name);
114 	result = 0;
115 
116 out:
117 	spin_unlock(&xprt_list_lock);
118 	return result;
119 }
120 EXPORT_SYMBOL_GPL(xprt_register_transport);
121 
122 /**
123  * xprt_unregister_transport - unregister a transport implementation
124  * @transport: transport to unregister
125  *
126  * Returns:
127  * 0:		transport successfully unregistered
128  * -ENOENT:	transport never registered
129  */
130 int xprt_unregister_transport(struct xprt_class *transport)
131 {
132 	struct xprt_class *t;
133 	int result;
134 
135 	result = 0;
136 	spin_lock(&xprt_list_lock);
137 	list_for_each_entry(t, &xprt_list, list) {
138 		if (t == transport) {
139 			printk(KERN_INFO
140 				"RPC: Unregistered %s transport module.\n",
141 				transport->name);
142 			list_del_init(&transport->list);
143 			goto out;
144 		}
145 	}
146 	result = -ENOENT;
147 
148 out:
149 	spin_unlock(&xprt_list_lock);
150 	return result;
151 }
152 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
153 
154 /**
155  * xprt_reserve_xprt - serialize write access to transports
156  * @task: task that is requesting access to the transport
157  *
158  * This prevents mixing the payload of separate requests, and prevents
159  * transport connects from colliding with writes.  No congestion control
160  * is provided.
161  */
162 int xprt_reserve_xprt(struct rpc_task *task)
163 {
164 	struct rpc_xprt	*xprt = task->tk_xprt;
165 	struct rpc_rqst *req = task->tk_rqstp;
166 
167 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
168 		if (task == xprt->snd_task)
169 			return 1;
170 		if (task == NULL)
171 			return 0;
172 		goto out_sleep;
173 	}
174 	xprt->snd_task = task;
175 	if (req) {
176 		req->rq_bytes_sent = 0;
177 		req->rq_ntrans++;
178 	}
179 	return 1;
180 
181 out_sleep:
182 	dprintk("RPC: %5u failed to lock transport %p\n",
183 			task->tk_pid, xprt);
184 	task->tk_timeout = 0;
185 	task->tk_status = -EAGAIN;
186 	if (req && req->rq_ntrans)
187 		rpc_sleep_on(&xprt->resend, task, NULL);
188 	else
189 		rpc_sleep_on(&xprt->sending, task, NULL);
190 	return 0;
191 }
192 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
193 
194 static void xprt_clear_locked(struct rpc_xprt *xprt)
195 {
196 	xprt->snd_task = NULL;
197 	if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state) || xprt->shutdown) {
198 		smp_mb__before_clear_bit();
199 		clear_bit(XPRT_LOCKED, &xprt->state);
200 		smp_mb__after_clear_bit();
201 	} else
202 		queue_work(rpciod_workqueue, &xprt->task_cleanup);
203 }
204 
205 /*
206  * xprt_reserve_xprt_cong - serialize write access to transports
207  * @task: task that is requesting access to the transport
208  *
209  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
210  * integrated into the decision of whether a request is allowed to be
211  * woken up and given access to the transport.
212  */
213 int xprt_reserve_xprt_cong(struct rpc_task *task)
214 {
215 	struct rpc_xprt	*xprt = task->tk_xprt;
216 	struct rpc_rqst *req = task->tk_rqstp;
217 
218 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
219 		if (task == xprt->snd_task)
220 			return 1;
221 		goto out_sleep;
222 	}
223 	if (__xprt_get_cong(xprt, task)) {
224 		xprt->snd_task = task;
225 		if (req) {
226 			req->rq_bytes_sent = 0;
227 			req->rq_ntrans++;
228 		}
229 		return 1;
230 	}
231 	xprt_clear_locked(xprt);
232 out_sleep:
233 	dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
234 	task->tk_timeout = 0;
235 	task->tk_status = -EAGAIN;
236 	if (req && req->rq_ntrans)
237 		rpc_sleep_on(&xprt->resend, task, NULL);
238 	else
239 		rpc_sleep_on(&xprt->sending, task, NULL);
240 	return 0;
241 }
242 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
243 
244 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
245 {
246 	int retval;
247 
248 	spin_lock_bh(&xprt->transport_lock);
249 	retval = xprt->ops->reserve_xprt(task);
250 	spin_unlock_bh(&xprt->transport_lock);
251 	return retval;
252 }
253 
254 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
255 {
256 	struct rpc_task *task;
257 	struct rpc_rqst *req;
258 
259 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
260 		return;
261 
262 	task = rpc_wake_up_next(&xprt->resend);
263 	if (!task) {
264 		task = rpc_wake_up_next(&xprt->sending);
265 		if (!task)
266 			goto out_unlock;
267 	}
268 
269 	req = task->tk_rqstp;
270 	xprt->snd_task = task;
271 	if (req) {
272 		req->rq_bytes_sent = 0;
273 		req->rq_ntrans++;
274 	}
275 	return;
276 
277 out_unlock:
278 	xprt_clear_locked(xprt);
279 }
280 
281 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
282 {
283 	struct rpc_task *task;
284 
285 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
286 		return;
287 	if (RPCXPRT_CONGESTED(xprt))
288 		goto out_unlock;
289 	task = rpc_wake_up_next(&xprt->resend);
290 	if (!task) {
291 		task = rpc_wake_up_next(&xprt->sending);
292 		if (!task)
293 			goto out_unlock;
294 	}
295 	if (__xprt_get_cong(xprt, task)) {
296 		struct rpc_rqst *req = task->tk_rqstp;
297 		xprt->snd_task = task;
298 		if (req) {
299 			req->rq_bytes_sent = 0;
300 			req->rq_ntrans++;
301 		}
302 		return;
303 	}
304 out_unlock:
305 	xprt_clear_locked(xprt);
306 }
307 
308 /**
309  * xprt_release_xprt - allow other requests to use a transport
310  * @xprt: transport with other tasks potentially waiting
311  * @task: task that is releasing access to the transport
312  *
313  * Note that "task" can be NULL.  No congestion control is provided.
314  */
315 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
316 {
317 	if (xprt->snd_task == task) {
318 		xprt_clear_locked(xprt);
319 		__xprt_lock_write_next(xprt);
320 	}
321 }
322 EXPORT_SYMBOL_GPL(xprt_release_xprt);
323 
324 /**
325  * xprt_release_xprt_cong - allow other requests to use a transport
326  * @xprt: transport with other tasks potentially waiting
327  * @task: task that is releasing access to the transport
328  *
329  * Note that "task" can be NULL.  Another task is awoken to use the
330  * transport if the transport's congestion window allows it.
331  */
332 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
333 {
334 	if (xprt->snd_task == task) {
335 		xprt_clear_locked(xprt);
336 		__xprt_lock_write_next_cong(xprt);
337 	}
338 }
339 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
340 
341 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
342 {
343 	spin_lock_bh(&xprt->transport_lock);
344 	xprt->ops->release_xprt(xprt, task);
345 	spin_unlock_bh(&xprt->transport_lock);
346 }
347 
348 /*
349  * Van Jacobson congestion avoidance. Check if the congestion window
350  * overflowed. Put the task to sleep if this is the case.
351  */
352 static int
353 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
354 {
355 	struct rpc_rqst *req = task->tk_rqstp;
356 
357 	if (req->rq_cong)
358 		return 1;
359 	dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
360 			task->tk_pid, xprt->cong, xprt->cwnd);
361 	if (RPCXPRT_CONGESTED(xprt))
362 		return 0;
363 	req->rq_cong = 1;
364 	xprt->cong += RPC_CWNDSCALE;
365 	return 1;
366 }
367 
368 /*
369  * Adjust the congestion window, and wake up the next task
370  * that has been sleeping due to congestion
371  */
372 static void
373 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
374 {
375 	if (!req->rq_cong)
376 		return;
377 	req->rq_cong = 0;
378 	xprt->cong -= RPC_CWNDSCALE;
379 	__xprt_lock_write_next_cong(xprt);
380 }
381 
382 /**
383  * xprt_release_rqst_cong - housekeeping when request is complete
384  * @task: RPC request that recently completed
385  *
386  * Useful for transports that require congestion control.
387  */
388 void xprt_release_rqst_cong(struct rpc_task *task)
389 {
390 	__xprt_put_cong(task->tk_xprt, task->tk_rqstp);
391 }
392 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
393 
394 /**
395  * xprt_adjust_cwnd - adjust transport congestion window
396  * @task: recently completed RPC request used to adjust window
397  * @result: result code of completed RPC request
398  *
399  * We use a time-smoothed congestion estimator to avoid heavy oscillation.
400  */
401 void xprt_adjust_cwnd(struct rpc_task *task, int result)
402 {
403 	struct rpc_rqst *req = task->tk_rqstp;
404 	struct rpc_xprt *xprt = task->tk_xprt;
405 	unsigned long cwnd = xprt->cwnd;
406 
407 	if (result >= 0 && cwnd <= xprt->cong) {
408 		/* The (cwnd >> 1) term makes sure
409 		 * the result gets rounded properly. */
410 		cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
411 		if (cwnd > RPC_MAXCWND(xprt))
412 			cwnd = RPC_MAXCWND(xprt);
413 		__xprt_lock_write_next_cong(xprt);
414 	} else if (result == -ETIMEDOUT) {
415 		cwnd >>= 1;
416 		if (cwnd < RPC_CWNDSCALE)
417 			cwnd = RPC_CWNDSCALE;
418 	}
419 	dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
420 			xprt->cong, xprt->cwnd, cwnd);
421 	xprt->cwnd = cwnd;
422 	__xprt_put_cong(xprt, req);
423 }
424 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
425 
426 /**
427  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
428  * @xprt: transport with waiting tasks
429  * @status: result code to plant in each task before waking it
430  *
431  */
432 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
433 {
434 	if (status < 0)
435 		rpc_wake_up_status(&xprt->pending, status);
436 	else
437 		rpc_wake_up(&xprt->pending);
438 }
439 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
440 
441 /**
442  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
443  * @task: task to be put to sleep
444  * @action: function pointer to be executed after wait
445  */
446 void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
447 {
448 	struct rpc_rqst *req = task->tk_rqstp;
449 	struct rpc_xprt *xprt = req->rq_xprt;
450 
451 	task->tk_timeout = req->rq_timeout;
452 	rpc_sleep_on(&xprt->pending, task, action);
453 }
454 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
455 
456 /**
457  * xprt_write_space - wake the task waiting for transport output buffer space
458  * @xprt: transport with waiting tasks
459  *
460  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
461  */
462 void xprt_write_space(struct rpc_xprt *xprt)
463 {
464 	if (unlikely(xprt->shutdown))
465 		return;
466 
467 	spin_lock_bh(&xprt->transport_lock);
468 	if (xprt->snd_task) {
469 		dprintk("RPC:       write space: waking waiting task on "
470 				"xprt %p\n", xprt);
471 		rpc_wake_up_queued_task(&xprt->pending, xprt->snd_task);
472 	}
473 	spin_unlock_bh(&xprt->transport_lock);
474 }
475 EXPORT_SYMBOL_GPL(xprt_write_space);
476 
477 /**
478  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
479  * @task: task whose timeout is to be set
480  *
481  * Set a request's retransmit timeout based on the transport's
482  * default timeout parameters.  Used by transports that don't adjust
483  * the retransmit timeout based on round-trip time estimation.
484  */
485 void xprt_set_retrans_timeout_def(struct rpc_task *task)
486 {
487 	task->tk_timeout = task->tk_rqstp->rq_timeout;
488 }
489 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
490 
491 /*
492  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
493  * @task: task whose timeout is to be set
494  *
495  * Set a request's retransmit timeout using the RTT estimator.
496  */
497 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
498 {
499 	int timer = task->tk_msg.rpc_proc->p_timer;
500 	struct rpc_clnt *clnt = task->tk_client;
501 	struct rpc_rtt *rtt = clnt->cl_rtt;
502 	struct rpc_rqst *req = task->tk_rqstp;
503 	unsigned long max_timeout = clnt->cl_timeout->to_maxval;
504 
505 	task->tk_timeout = rpc_calc_rto(rtt, timer);
506 	task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
507 	if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
508 		task->tk_timeout = max_timeout;
509 }
510 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
511 
512 static void xprt_reset_majortimeo(struct rpc_rqst *req)
513 {
514 	const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
515 
516 	req->rq_majortimeo = req->rq_timeout;
517 	if (to->to_exponential)
518 		req->rq_majortimeo <<= to->to_retries;
519 	else
520 		req->rq_majortimeo += to->to_increment * to->to_retries;
521 	if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
522 		req->rq_majortimeo = to->to_maxval;
523 	req->rq_majortimeo += jiffies;
524 }
525 
526 /**
527  * xprt_adjust_timeout - adjust timeout values for next retransmit
528  * @req: RPC request containing parameters to use for the adjustment
529  *
530  */
531 int xprt_adjust_timeout(struct rpc_rqst *req)
532 {
533 	struct rpc_xprt *xprt = req->rq_xprt;
534 	const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
535 	int status = 0;
536 
537 	if (time_before(jiffies, req->rq_majortimeo)) {
538 		if (to->to_exponential)
539 			req->rq_timeout <<= 1;
540 		else
541 			req->rq_timeout += to->to_increment;
542 		if (to->to_maxval && req->rq_timeout >= to->to_maxval)
543 			req->rq_timeout = to->to_maxval;
544 		req->rq_retries++;
545 	} else {
546 		req->rq_timeout = to->to_initval;
547 		req->rq_retries = 0;
548 		xprt_reset_majortimeo(req);
549 		/* Reset the RTT counters == "slow start" */
550 		spin_lock_bh(&xprt->transport_lock);
551 		rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
552 		spin_unlock_bh(&xprt->transport_lock);
553 		status = -ETIMEDOUT;
554 	}
555 
556 	if (req->rq_timeout == 0) {
557 		printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
558 		req->rq_timeout = 5 * HZ;
559 	}
560 	return status;
561 }
562 
563 static void xprt_autoclose(struct work_struct *work)
564 {
565 	struct rpc_xprt *xprt =
566 		container_of(work, struct rpc_xprt, task_cleanup);
567 
568 	xprt->ops->close(xprt);
569 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
570 	xprt_release_write(xprt, NULL);
571 }
572 
573 /**
574  * xprt_disconnect_done - mark a transport as disconnected
575  * @xprt: transport to flag for disconnect
576  *
577  */
578 void xprt_disconnect_done(struct rpc_xprt *xprt)
579 {
580 	dprintk("RPC:       disconnected transport %p\n", xprt);
581 	spin_lock_bh(&xprt->transport_lock);
582 	xprt_clear_connected(xprt);
583 	xprt_wake_pending_tasks(xprt, -ENOTCONN);
584 	spin_unlock_bh(&xprt->transport_lock);
585 }
586 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
587 
588 /**
589  * xprt_force_disconnect - force a transport to disconnect
590  * @xprt: transport to disconnect
591  *
592  */
593 void xprt_force_disconnect(struct rpc_xprt *xprt)
594 {
595 	/* Don't race with the test_bit() in xprt_clear_locked() */
596 	spin_lock_bh(&xprt->transport_lock);
597 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
598 	/* Try to schedule an autoclose RPC call */
599 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
600 		queue_work(rpciod_workqueue, &xprt->task_cleanup);
601 	xprt_wake_pending_tasks(xprt, -ENOTCONN);
602 	spin_unlock_bh(&xprt->transport_lock);
603 }
604 
605 /**
606  * xprt_conditional_disconnect - force a transport to disconnect
607  * @xprt: transport to disconnect
608  * @cookie: 'connection cookie'
609  *
610  * This attempts to break the connection if and only if 'cookie' matches
611  * the current transport 'connection cookie'. It ensures that we don't
612  * try to break the connection more than once when we need to retransmit
613  * a batch of RPC requests.
614  *
615  */
616 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
617 {
618 	/* Don't race with the test_bit() in xprt_clear_locked() */
619 	spin_lock_bh(&xprt->transport_lock);
620 	if (cookie != xprt->connect_cookie)
621 		goto out;
622 	if (test_bit(XPRT_CLOSING, &xprt->state) || !xprt_connected(xprt))
623 		goto out;
624 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
625 	/* Try to schedule an autoclose RPC call */
626 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
627 		queue_work(rpciod_workqueue, &xprt->task_cleanup);
628 	xprt_wake_pending_tasks(xprt, -ENOTCONN);
629 out:
630 	spin_unlock_bh(&xprt->transport_lock);
631 }
632 
633 static void
634 xprt_init_autodisconnect(unsigned long data)
635 {
636 	struct rpc_xprt *xprt = (struct rpc_xprt *)data;
637 
638 	spin_lock(&xprt->transport_lock);
639 	if (!list_empty(&xprt->recv) || xprt->shutdown)
640 		goto out_abort;
641 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
642 		goto out_abort;
643 	spin_unlock(&xprt->transport_lock);
644 	if (xprt_connecting(xprt))
645 		xprt_release_write(xprt, NULL);
646 	else
647 		queue_work(rpciod_workqueue, &xprt->task_cleanup);
648 	return;
649 out_abort:
650 	spin_unlock(&xprt->transport_lock);
651 }
652 
653 /**
654  * xprt_connect - schedule a transport connect operation
655  * @task: RPC task that is requesting the connect
656  *
657  */
658 void xprt_connect(struct rpc_task *task)
659 {
660 	struct rpc_xprt	*xprt = task->tk_xprt;
661 
662 	dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
663 			xprt, (xprt_connected(xprt) ? "is" : "is not"));
664 
665 	if (!xprt_bound(xprt)) {
666 		task->tk_status = -EIO;
667 		return;
668 	}
669 	if (!xprt_lock_write(xprt, task))
670 		return;
671 	if (xprt_connected(xprt))
672 		xprt_release_write(xprt, task);
673 	else {
674 		if (task->tk_rqstp)
675 			task->tk_rqstp->rq_bytes_sent = 0;
676 
677 		task->tk_timeout = xprt->connect_timeout;
678 		rpc_sleep_on(&xprt->pending, task, xprt_connect_status);
679 		xprt->stat.connect_start = jiffies;
680 		xprt->ops->connect(task);
681 	}
682 	return;
683 }
684 
685 static void xprt_connect_status(struct rpc_task *task)
686 {
687 	struct rpc_xprt	*xprt = task->tk_xprt;
688 
689 	if (task->tk_status == 0) {
690 		xprt->stat.connect_count++;
691 		xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
692 		dprintk("RPC: %5u xprt_connect_status: connection established\n",
693 				task->tk_pid);
694 		return;
695 	}
696 
697 	switch (task->tk_status) {
698 	case -ENOTCONN:
699 		dprintk("RPC: %5u xprt_connect_status: connection broken\n",
700 				task->tk_pid);
701 		break;
702 	case -ETIMEDOUT:
703 		dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
704 				"out\n", task->tk_pid);
705 		break;
706 	default:
707 		dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
708 				"server %s\n", task->tk_pid, -task->tk_status,
709 				task->tk_client->cl_server);
710 		xprt_release_write(xprt, task);
711 		task->tk_status = -EIO;
712 	}
713 }
714 
715 /**
716  * xprt_lookup_rqst - find an RPC request corresponding to an XID
717  * @xprt: transport on which the original request was transmitted
718  * @xid: RPC XID of incoming reply
719  *
720  */
721 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
722 {
723 	struct list_head *pos;
724 
725 	list_for_each(pos, &xprt->recv) {
726 		struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
727 		if (entry->rq_xid == xid)
728 			return entry;
729 	}
730 
731 	dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
732 			ntohl(xid));
733 	xprt->stat.bad_xids++;
734 	return NULL;
735 }
736 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
737 
738 /**
739  * xprt_update_rtt - update an RPC client's RTT state after receiving a reply
740  * @task: RPC request that recently completed
741  *
742  */
743 void xprt_update_rtt(struct rpc_task *task)
744 {
745 	struct rpc_rqst *req = task->tk_rqstp;
746 	struct rpc_rtt *rtt = task->tk_client->cl_rtt;
747 	unsigned timer = task->tk_msg.rpc_proc->p_timer;
748 
749 	if (timer) {
750 		if (req->rq_ntrans == 1)
751 			rpc_update_rtt(rtt, timer,
752 					(long)jiffies - req->rq_xtime);
753 		rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
754 	}
755 }
756 EXPORT_SYMBOL_GPL(xprt_update_rtt);
757 
758 /**
759  * xprt_complete_rqst - called when reply processing is complete
760  * @task: RPC request that recently completed
761  * @copied: actual number of bytes received from the transport
762  *
763  * Caller holds transport lock.
764  */
765 void xprt_complete_rqst(struct rpc_task *task, int copied)
766 {
767 	struct rpc_rqst *req = task->tk_rqstp;
768 	struct rpc_xprt *xprt = req->rq_xprt;
769 
770 	dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
771 			task->tk_pid, ntohl(req->rq_xid), copied);
772 
773 	xprt->stat.recvs++;
774 	task->tk_rtt = (long)jiffies - req->rq_xtime;
775 
776 	list_del_init(&req->rq_list);
777 	req->rq_private_buf.len = copied;
778 	/* Ensure all writes are done before we update req->rq_received */
779 	smp_wmb();
780 	req->rq_received = copied;
781 	rpc_wake_up_queued_task(&xprt->pending, task);
782 }
783 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
784 
785 static void xprt_timer(struct rpc_task *task)
786 {
787 	struct rpc_rqst *req = task->tk_rqstp;
788 	struct rpc_xprt *xprt = req->rq_xprt;
789 
790 	if (task->tk_status != -ETIMEDOUT)
791 		return;
792 	dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
793 
794 	spin_lock_bh(&xprt->transport_lock);
795 	if (!req->rq_received) {
796 		if (xprt->ops->timer)
797 			xprt->ops->timer(task);
798 	} else
799 		task->tk_status = 0;
800 	spin_unlock_bh(&xprt->transport_lock);
801 }
802 
803 /**
804  * xprt_prepare_transmit - reserve the transport before sending a request
805  * @task: RPC task about to send a request
806  *
807  */
808 int xprt_prepare_transmit(struct rpc_task *task)
809 {
810 	struct rpc_rqst	*req = task->tk_rqstp;
811 	struct rpc_xprt	*xprt = req->rq_xprt;
812 	int err = 0;
813 
814 	dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
815 
816 	spin_lock_bh(&xprt->transport_lock);
817 	if (req->rq_received && !req->rq_bytes_sent) {
818 		err = req->rq_received;
819 		goto out_unlock;
820 	}
821 	if (!xprt->ops->reserve_xprt(task)) {
822 		err = -EAGAIN;
823 		goto out_unlock;
824 	}
825 
826 	if (!xprt_connected(xprt)) {
827 		err = -ENOTCONN;
828 		goto out_unlock;
829 	}
830 out_unlock:
831 	spin_unlock_bh(&xprt->transport_lock);
832 	return err;
833 }
834 
835 void xprt_end_transmit(struct rpc_task *task)
836 {
837 	xprt_release_write(task->tk_xprt, task);
838 }
839 
840 /**
841  * xprt_transmit - send an RPC request on a transport
842  * @task: controlling RPC task
843  *
844  * We have to copy the iovec because sendmsg fiddles with its contents.
845  */
846 void xprt_transmit(struct rpc_task *task)
847 {
848 	struct rpc_rqst	*req = task->tk_rqstp;
849 	struct rpc_xprt	*xprt = req->rq_xprt;
850 	int status;
851 
852 	dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
853 
854 	if (!req->rq_received) {
855 		if (list_empty(&req->rq_list)) {
856 			spin_lock_bh(&xprt->transport_lock);
857 			/* Update the softirq receive buffer */
858 			memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
859 					sizeof(req->rq_private_buf));
860 			/* Add request to the receive list */
861 			list_add_tail(&req->rq_list, &xprt->recv);
862 			spin_unlock_bh(&xprt->transport_lock);
863 			xprt_reset_majortimeo(req);
864 			/* Turn off autodisconnect */
865 			del_singleshot_timer_sync(&xprt->timer);
866 		}
867 	} else if (!req->rq_bytes_sent)
868 		return;
869 
870 	req->rq_connect_cookie = xprt->connect_cookie;
871 	req->rq_xtime = jiffies;
872 	status = xprt->ops->send_request(task);
873 	if (status == 0) {
874 		dprintk("RPC: %5u xmit complete\n", task->tk_pid);
875 		spin_lock_bh(&xprt->transport_lock);
876 
877 		xprt->ops->set_retrans_timeout(task);
878 
879 		xprt->stat.sends++;
880 		xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
881 		xprt->stat.bklog_u += xprt->backlog.qlen;
882 
883 		/* Don't race with disconnect */
884 		if (!xprt_connected(xprt))
885 			task->tk_status = -ENOTCONN;
886 		else if (!req->rq_received)
887 			rpc_sleep_on(&xprt->pending, task, xprt_timer);
888 		spin_unlock_bh(&xprt->transport_lock);
889 		return;
890 	}
891 
892 	/* Note: at this point, task->tk_sleeping has not yet been set,
893 	 *	 hence there is no danger of the waking up task being put on
894 	 *	 schedq, and being picked up by a parallel run of rpciod().
895 	 */
896 	task->tk_status = status;
897 	if (status == -ECONNREFUSED)
898 		rpc_sleep_on(&xprt->sending, task, NULL);
899 }
900 
901 static inline void do_xprt_reserve(struct rpc_task *task)
902 {
903 	struct rpc_xprt	*xprt = task->tk_xprt;
904 
905 	task->tk_status = 0;
906 	if (task->tk_rqstp)
907 		return;
908 	if (!list_empty(&xprt->free)) {
909 		struct rpc_rqst	*req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
910 		list_del_init(&req->rq_list);
911 		task->tk_rqstp = req;
912 		xprt_request_init(task, xprt);
913 		return;
914 	}
915 	dprintk("RPC:       waiting for request slot\n");
916 	task->tk_status = -EAGAIN;
917 	task->tk_timeout = 0;
918 	rpc_sleep_on(&xprt->backlog, task, NULL);
919 }
920 
921 /**
922  * xprt_reserve - allocate an RPC request slot
923  * @task: RPC task requesting a slot allocation
924  *
925  * If no more slots are available, place the task on the transport's
926  * backlog queue.
927  */
928 void xprt_reserve(struct rpc_task *task)
929 {
930 	struct rpc_xprt	*xprt = task->tk_xprt;
931 
932 	task->tk_status = -EIO;
933 	spin_lock(&xprt->reserve_lock);
934 	do_xprt_reserve(task);
935 	spin_unlock(&xprt->reserve_lock);
936 }
937 
938 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
939 {
940 	return xprt->xid++;
941 }
942 
943 static inline void xprt_init_xid(struct rpc_xprt *xprt)
944 {
945 	xprt->xid = net_random();
946 }
947 
948 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
949 {
950 	struct rpc_rqst	*req = task->tk_rqstp;
951 
952 	req->rq_timeout = task->tk_client->cl_timeout->to_initval;
953 	req->rq_task	= task;
954 	req->rq_xprt    = xprt;
955 	req->rq_buffer  = NULL;
956 	req->rq_xid     = xprt_alloc_xid(xprt);
957 	req->rq_release_snd_buf = NULL;
958 	xprt_reset_majortimeo(req);
959 	dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
960 			req, ntohl(req->rq_xid));
961 }
962 
963 /**
964  * xprt_release - release an RPC request slot
965  * @task: task which is finished with the slot
966  *
967  */
968 void xprt_release(struct rpc_task *task)
969 {
970 	struct rpc_xprt	*xprt = task->tk_xprt;
971 	struct rpc_rqst	*req;
972 
973 	if (!(req = task->tk_rqstp))
974 		return;
975 	rpc_count_iostats(task);
976 	spin_lock_bh(&xprt->transport_lock);
977 	xprt->ops->release_xprt(xprt, task);
978 	if (xprt->ops->release_request)
979 		xprt->ops->release_request(task);
980 	if (!list_empty(&req->rq_list))
981 		list_del(&req->rq_list);
982 	xprt->last_used = jiffies;
983 	if (list_empty(&xprt->recv))
984 		mod_timer(&xprt->timer,
985 				xprt->last_used + xprt->idle_timeout);
986 	spin_unlock_bh(&xprt->transport_lock);
987 	xprt->ops->buf_free(req->rq_buffer);
988 	task->tk_rqstp = NULL;
989 	if (req->rq_release_snd_buf)
990 		req->rq_release_snd_buf(req);
991 	memset(req, 0, sizeof(*req));	/* mark unused */
992 
993 	dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
994 
995 	spin_lock(&xprt->reserve_lock);
996 	list_add(&req->rq_list, &xprt->free);
997 	rpc_wake_up_next(&xprt->backlog);
998 	spin_unlock(&xprt->reserve_lock);
999 }
1000 
1001 /**
1002  * xprt_create_transport - create an RPC transport
1003  * @args: rpc transport creation arguments
1004  *
1005  */
1006 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1007 {
1008 	struct rpc_xprt	*xprt;
1009 	struct rpc_rqst	*req;
1010 	struct xprt_class *t;
1011 
1012 	spin_lock(&xprt_list_lock);
1013 	list_for_each_entry(t, &xprt_list, list) {
1014 		if (t->ident == args->ident) {
1015 			spin_unlock(&xprt_list_lock);
1016 			goto found;
1017 		}
1018 	}
1019 	spin_unlock(&xprt_list_lock);
1020 	printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1021 	return ERR_PTR(-EIO);
1022 
1023 found:
1024 	xprt = t->setup(args);
1025 	if (IS_ERR(xprt)) {
1026 		dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1027 				-PTR_ERR(xprt));
1028 		return xprt;
1029 	}
1030 
1031 	kref_init(&xprt->kref);
1032 	spin_lock_init(&xprt->transport_lock);
1033 	spin_lock_init(&xprt->reserve_lock);
1034 
1035 	INIT_LIST_HEAD(&xprt->free);
1036 	INIT_LIST_HEAD(&xprt->recv);
1037 	INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1038 	setup_timer(&xprt->timer, xprt_init_autodisconnect,
1039 			(unsigned long)xprt);
1040 	xprt->last_used = jiffies;
1041 	xprt->cwnd = RPC_INITCWND;
1042 	xprt->bind_index = 0;
1043 
1044 	rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1045 	rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1046 	rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1047 	rpc_init_wait_queue(&xprt->resend, "xprt_resend");
1048 	rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1049 
1050 	/* initialize free list */
1051 	for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
1052 		list_add(&req->rq_list, &xprt->free);
1053 
1054 	xprt_init_xid(xprt);
1055 
1056 	dprintk("RPC:       created transport %p with %u slots\n", xprt,
1057 			xprt->max_reqs);
1058 
1059 	return xprt;
1060 }
1061 
1062 /**
1063  * xprt_destroy - destroy an RPC transport, killing off all requests.
1064  * @kref: kref for the transport to destroy
1065  *
1066  */
1067 static void xprt_destroy(struct kref *kref)
1068 {
1069 	struct rpc_xprt *xprt = container_of(kref, struct rpc_xprt, kref);
1070 
1071 	dprintk("RPC:       destroying transport %p\n", xprt);
1072 	xprt->shutdown = 1;
1073 	del_timer_sync(&xprt->timer);
1074 
1075 	rpc_destroy_wait_queue(&xprt->binding);
1076 	rpc_destroy_wait_queue(&xprt->pending);
1077 	rpc_destroy_wait_queue(&xprt->sending);
1078 	rpc_destroy_wait_queue(&xprt->resend);
1079 	rpc_destroy_wait_queue(&xprt->backlog);
1080 	/*
1081 	 * Tear down transport state and free the rpc_xprt
1082 	 */
1083 	xprt->ops->destroy(xprt);
1084 }
1085 
1086 /**
1087  * xprt_put - release a reference to an RPC transport.
1088  * @xprt: pointer to the transport
1089  *
1090  */
1091 void xprt_put(struct rpc_xprt *xprt)
1092 {
1093 	kref_put(&xprt->kref, xprt_destroy);
1094 }
1095 
1096 /**
1097  * xprt_get - return a reference to an RPC transport.
1098  * @xprt: pointer to the transport
1099  *
1100  */
1101 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1102 {
1103 	kref_get(&xprt->kref);
1104 	return xprt;
1105 }
1106