xref: /openbmc/linux/net/sunrpc/xprt.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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/random.h>
45 
46 #include <linux/sunrpc/clnt.h>
47 
48 /*
49  * Local variables
50  */
51 
52 #ifdef RPC_DEBUG
53 # undef  RPC_DEBUG_DATA
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 /*
66  * The transport code maintains an estimate on the maximum number of out-
67  * standing RPC requests, using a smoothed version of the congestion
68  * avoidance implemented in 44BSD. This is basically the Van Jacobson
69  * congestion algorithm: If a retransmit occurs, the congestion window is
70  * halved; otherwise, it is incremented by 1/cwnd when
71  *
72  *	-	a reply is received and
73  *	-	a full number of requests are outstanding and
74  *	-	the congestion window hasn't been updated recently.
75  */
76 #define RPC_CWNDSHIFT		(8U)
77 #define RPC_CWNDSCALE		(1U << RPC_CWNDSHIFT)
78 #define RPC_INITCWND		RPC_CWNDSCALE
79 #define RPC_MAXCWND(xprt)	((xprt)->max_reqs << RPC_CWNDSHIFT)
80 
81 #define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
82 
83 /**
84  * xprt_reserve_xprt - serialize write access to transports
85  * @task: task that is requesting access to the transport
86  *
87  * This prevents mixing the payload of separate requests, and prevents
88  * transport connects from colliding with writes.  No congestion control
89  * is provided.
90  */
91 int xprt_reserve_xprt(struct rpc_task *task)
92 {
93 	struct rpc_xprt	*xprt = task->tk_xprt;
94 	struct rpc_rqst *req = task->tk_rqstp;
95 
96 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
97 		if (task == xprt->snd_task)
98 			return 1;
99 		if (task == NULL)
100 			return 0;
101 		goto out_sleep;
102 	}
103 	xprt->snd_task = task;
104 	if (req) {
105 		req->rq_bytes_sent = 0;
106 		req->rq_ntrans++;
107 	}
108 	return 1;
109 
110 out_sleep:
111 	dprintk("RPC: %4d failed to lock transport %p\n",
112 			task->tk_pid, xprt);
113 	task->tk_timeout = 0;
114 	task->tk_status = -EAGAIN;
115 	if (req && req->rq_ntrans)
116 		rpc_sleep_on(&xprt->resend, task, NULL, NULL);
117 	else
118 		rpc_sleep_on(&xprt->sending, task, NULL, NULL);
119 	return 0;
120 }
121 
122 /*
123  * xprt_reserve_xprt_cong - serialize write access to transports
124  * @task: task that is requesting access to the transport
125  *
126  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
127  * integrated into the decision of whether a request is allowed to be
128  * woken up and given access to the transport.
129  */
130 int xprt_reserve_xprt_cong(struct rpc_task *task)
131 {
132 	struct rpc_xprt	*xprt = task->tk_xprt;
133 	struct rpc_rqst *req = task->tk_rqstp;
134 
135 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
136 		if (task == xprt->snd_task)
137 			return 1;
138 		goto out_sleep;
139 	}
140 	if (__xprt_get_cong(xprt, task)) {
141 		xprt->snd_task = task;
142 		if (req) {
143 			req->rq_bytes_sent = 0;
144 			req->rq_ntrans++;
145 		}
146 		return 1;
147 	}
148 	smp_mb__before_clear_bit();
149 	clear_bit(XPRT_LOCKED, &xprt->state);
150 	smp_mb__after_clear_bit();
151 out_sleep:
152 	dprintk("RPC: %4d failed to lock transport %p\n", task->tk_pid, xprt);
153 	task->tk_timeout = 0;
154 	task->tk_status = -EAGAIN;
155 	if (req && req->rq_ntrans)
156 		rpc_sleep_on(&xprt->resend, task, NULL, NULL);
157 	else
158 		rpc_sleep_on(&xprt->sending, task, NULL, NULL);
159 	return 0;
160 }
161 
162 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
163 {
164 	int retval;
165 
166 	spin_lock_bh(&xprt->transport_lock);
167 	retval = xprt->ops->reserve_xprt(task);
168 	spin_unlock_bh(&xprt->transport_lock);
169 	return retval;
170 }
171 
172 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
173 {
174 	struct rpc_task *task;
175 	struct rpc_rqst *req;
176 
177 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
178 		return;
179 
180 	task = rpc_wake_up_next(&xprt->resend);
181 	if (!task) {
182 		task = rpc_wake_up_next(&xprt->sending);
183 		if (!task)
184 			goto out_unlock;
185 	}
186 
187 	req = task->tk_rqstp;
188 	xprt->snd_task = task;
189 	if (req) {
190 		req->rq_bytes_sent = 0;
191 		req->rq_ntrans++;
192 	}
193 	return;
194 
195 out_unlock:
196 	smp_mb__before_clear_bit();
197 	clear_bit(XPRT_LOCKED, &xprt->state);
198 	smp_mb__after_clear_bit();
199 }
200 
201 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
202 {
203 	struct rpc_task *task;
204 
205 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
206 		return;
207 	if (RPCXPRT_CONGESTED(xprt))
208 		goto out_unlock;
209 	task = rpc_wake_up_next(&xprt->resend);
210 	if (!task) {
211 		task = rpc_wake_up_next(&xprt->sending);
212 		if (!task)
213 			goto out_unlock;
214 	}
215 	if (__xprt_get_cong(xprt, task)) {
216 		struct rpc_rqst *req = task->tk_rqstp;
217 		xprt->snd_task = task;
218 		if (req) {
219 			req->rq_bytes_sent = 0;
220 			req->rq_ntrans++;
221 		}
222 		return;
223 	}
224 out_unlock:
225 	smp_mb__before_clear_bit();
226 	clear_bit(XPRT_LOCKED, &xprt->state);
227 	smp_mb__after_clear_bit();
228 }
229 
230 /**
231  * xprt_release_xprt - allow other requests to use a transport
232  * @xprt: transport with other tasks potentially waiting
233  * @task: task that is releasing access to the transport
234  *
235  * Note that "task" can be NULL.  No congestion control is provided.
236  */
237 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
238 {
239 	if (xprt->snd_task == task) {
240 		xprt->snd_task = NULL;
241 		smp_mb__before_clear_bit();
242 		clear_bit(XPRT_LOCKED, &xprt->state);
243 		smp_mb__after_clear_bit();
244 		__xprt_lock_write_next(xprt);
245 	}
246 }
247 
248 /**
249  * xprt_release_xprt_cong - allow other requests to use a transport
250  * @xprt: transport with other tasks potentially waiting
251  * @task: task that is releasing access to the transport
252  *
253  * Note that "task" can be NULL.  Another task is awoken to use the
254  * transport if the transport's congestion window allows it.
255  */
256 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
257 {
258 	if (xprt->snd_task == task) {
259 		xprt->snd_task = NULL;
260 		smp_mb__before_clear_bit();
261 		clear_bit(XPRT_LOCKED, &xprt->state);
262 		smp_mb__after_clear_bit();
263 		__xprt_lock_write_next_cong(xprt);
264 	}
265 }
266 
267 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
268 {
269 	spin_lock_bh(&xprt->transport_lock);
270 	xprt->ops->release_xprt(xprt, task);
271 	spin_unlock_bh(&xprt->transport_lock);
272 }
273 
274 /*
275  * Van Jacobson congestion avoidance. Check if the congestion window
276  * overflowed. Put the task to sleep if this is the case.
277  */
278 static int
279 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
280 {
281 	struct rpc_rqst *req = task->tk_rqstp;
282 
283 	if (req->rq_cong)
284 		return 1;
285 	dprintk("RPC: %4d xprt_cwnd_limited cong = %ld cwnd = %ld\n",
286 			task->tk_pid, xprt->cong, xprt->cwnd);
287 	if (RPCXPRT_CONGESTED(xprt))
288 		return 0;
289 	req->rq_cong = 1;
290 	xprt->cong += RPC_CWNDSCALE;
291 	return 1;
292 }
293 
294 /*
295  * Adjust the congestion window, and wake up the next task
296  * that has been sleeping due to congestion
297  */
298 static void
299 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
300 {
301 	if (!req->rq_cong)
302 		return;
303 	req->rq_cong = 0;
304 	xprt->cong -= RPC_CWNDSCALE;
305 	__xprt_lock_write_next_cong(xprt);
306 }
307 
308 /**
309  * xprt_release_rqst_cong - housekeeping when request is complete
310  * @task: RPC request that recently completed
311  *
312  * Useful for transports that require congestion control.
313  */
314 void xprt_release_rqst_cong(struct rpc_task *task)
315 {
316 	__xprt_put_cong(task->tk_xprt, task->tk_rqstp);
317 }
318 
319 /**
320  * xprt_adjust_cwnd - adjust transport congestion window
321  * @task: recently completed RPC request used to adjust window
322  * @result: result code of completed RPC request
323  *
324  * We use a time-smoothed congestion estimator to avoid heavy oscillation.
325  */
326 void xprt_adjust_cwnd(struct rpc_task *task, int result)
327 {
328 	struct rpc_rqst *req = task->tk_rqstp;
329 	struct rpc_xprt *xprt = task->tk_xprt;
330 	unsigned long cwnd = xprt->cwnd;
331 
332 	if (result >= 0 && cwnd <= xprt->cong) {
333 		/* The (cwnd >> 1) term makes sure
334 		 * the result gets rounded properly. */
335 		cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
336 		if (cwnd > RPC_MAXCWND(xprt))
337 			cwnd = RPC_MAXCWND(xprt);
338 		__xprt_lock_write_next_cong(xprt);
339 	} else if (result == -ETIMEDOUT) {
340 		cwnd >>= 1;
341 		if (cwnd < RPC_CWNDSCALE)
342 			cwnd = RPC_CWNDSCALE;
343 	}
344 	dprintk("RPC:      cong %ld, cwnd was %ld, now %ld\n",
345 			xprt->cong, xprt->cwnd, cwnd);
346 	xprt->cwnd = cwnd;
347 	__xprt_put_cong(xprt, req);
348 }
349 
350 /**
351  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
352  * @xprt: transport with waiting tasks
353  * @status: result code to plant in each task before waking it
354  *
355  */
356 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
357 {
358 	if (status < 0)
359 		rpc_wake_up_status(&xprt->pending, status);
360 	else
361 		rpc_wake_up(&xprt->pending);
362 }
363 
364 /**
365  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
366  * @task: task to be put to sleep
367  *
368  */
369 void xprt_wait_for_buffer_space(struct rpc_task *task)
370 {
371 	struct rpc_rqst *req = task->tk_rqstp;
372 	struct rpc_xprt *xprt = req->rq_xprt;
373 
374 	task->tk_timeout = req->rq_timeout;
375 	rpc_sleep_on(&xprt->pending, task, NULL, NULL);
376 }
377 
378 /**
379  * xprt_write_space - wake the task waiting for transport output buffer space
380  * @xprt: transport with waiting tasks
381  *
382  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
383  */
384 void xprt_write_space(struct rpc_xprt *xprt)
385 {
386 	if (unlikely(xprt->shutdown))
387 		return;
388 
389 	spin_lock_bh(&xprt->transport_lock);
390 	if (xprt->snd_task) {
391 		dprintk("RPC:      write space: waking waiting task on xprt %p\n",
392 				xprt);
393 		rpc_wake_up_task(xprt->snd_task);
394 	}
395 	spin_unlock_bh(&xprt->transport_lock);
396 }
397 
398 /**
399  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
400  * @task: task whose timeout is to be set
401  *
402  * Set a request's retransmit timeout based on the transport's
403  * default timeout parameters.  Used by transports that don't adjust
404  * the retransmit timeout based on round-trip time estimation.
405  */
406 void xprt_set_retrans_timeout_def(struct rpc_task *task)
407 {
408 	task->tk_timeout = task->tk_rqstp->rq_timeout;
409 }
410 
411 /*
412  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
413  * @task: task whose timeout is to be set
414  *
415  * Set a request's retransmit timeout using the RTT estimator.
416  */
417 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
418 {
419 	int timer = task->tk_msg.rpc_proc->p_timer;
420 	struct rpc_rtt *rtt = task->tk_client->cl_rtt;
421 	struct rpc_rqst *req = task->tk_rqstp;
422 	unsigned long max_timeout = req->rq_xprt->timeout.to_maxval;
423 
424 	task->tk_timeout = rpc_calc_rto(rtt, timer);
425 	task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
426 	if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
427 		task->tk_timeout = max_timeout;
428 }
429 
430 static void xprt_reset_majortimeo(struct rpc_rqst *req)
431 {
432 	struct rpc_timeout *to = &req->rq_xprt->timeout;
433 
434 	req->rq_majortimeo = req->rq_timeout;
435 	if (to->to_exponential)
436 		req->rq_majortimeo <<= to->to_retries;
437 	else
438 		req->rq_majortimeo += to->to_increment * to->to_retries;
439 	if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
440 		req->rq_majortimeo = to->to_maxval;
441 	req->rq_majortimeo += jiffies;
442 }
443 
444 /**
445  * xprt_adjust_timeout - adjust timeout values for next retransmit
446  * @req: RPC request containing parameters to use for the adjustment
447  *
448  */
449 int xprt_adjust_timeout(struct rpc_rqst *req)
450 {
451 	struct rpc_xprt *xprt = req->rq_xprt;
452 	struct rpc_timeout *to = &xprt->timeout;
453 	int status = 0;
454 
455 	if (time_before(jiffies, req->rq_majortimeo)) {
456 		if (to->to_exponential)
457 			req->rq_timeout <<= 1;
458 		else
459 			req->rq_timeout += to->to_increment;
460 		if (to->to_maxval && req->rq_timeout >= to->to_maxval)
461 			req->rq_timeout = to->to_maxval;
462 		req->rq_retries++;
463 		pprintk("RPC: %lu retrans\n", jiffies);
464 	} else {
465 		req->rq_timeout = to->to_initval;
466 		req->rq_retries = 0;
467 		xprt_reset_majortimeo(req);
468 		/* Reset the RTT counters == "slow start" */
469 		spin_lock_bh(&xprt->transport_lock);
470 		rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
471 		spin_unlock_bh(&xprt->transport_lock);
472 		pprintk("RPC: %lu timeout\n", jiffies);
473 		status = -ETIMEDOUT;
474 	}
475 
476 	if (req->rq_timeout == 0) {
477 		printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
478 		req->rq_timeout = 5 * HZ;
479 	}
480 	return status;
481 }
482 
483 static void xprt_autoclose(void *args)
484 {
485 	struct rpc_xprt *xprt = (struct rpc_xprt *)args;
486 
487 	xprt_disconnect(xprt);
488 	xprt->ops->close(xprt);
489 	xprt_release_write(xprt, NULL);
490 }
491 
492 /**
493  * xprt_disconnect - mark a transport as disconnected
494  * @xprt: transport to flag for disconnect
495  *
496  */
497 void xprt_disconnect(struct rpc_xprt *xprt)
498 {
499 	dprintk("RPC:      disconnected transport %p\n", xprt);
500 	spin_lock_bh(&xprt->transport_lock);
501 	xprt_clear_connected(xprt);
502 	xprt_wake_pending_tasks(xprt, -ENOTCONN);
503 	spin_unlock_bh(&xprt->transport_lock);
504 }
505 
506 static void
507 xprt_init_autodisconnect(unsigned long data)
508 {
509 	struct rpc_xprt *xprt = (struct rpc_xprt *)data;
510 
511 	spin_lock(&xprt->transport_lock);
512 	if (!list_empty(&xprt->recv) || xprt->shutdown)
513 		goto out_abort;
514 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
515 		goto out_abort;
516 	spin_unlock(&xprt->transport_lock);
517 	if (xprt_connecting(xprt))
518 		xprt_release_write(xprt, NULL);
519 	else
520 		schedule_work(&xprt->task_cleanup);
521 	return;
522 out_abort:
523 	spin_unlock(&xprt->transport_lock);
524 }
525 
526 /**
527  * xprt_connect - schedule a transport connect operation
528  * @task: RPC task that is requesting the connect
529  *
530  */
531 void xprt_connect(struct rpc_task *task)
532 {
533 	struct rpc_xprt	*xprt = task->tk_xprt;
534 
535 	dprintk("RPC: %4d xprt_connect xprt %p %s connected\n", task->tk_pid,
536 			xprt, (xprt_connected(xprt) ? "is" : "is not"));
537 
538 	if (xprt->shutdown) {
539 		task->tk_status = -EIO;
540 		return;
541 	}
542 	if (!xprt->addr.sin_port) {
543 		task->tk_status = -EIO;
544 		return;
545 	}
546 	if (!xprt_lock_write(xprt, task))
547 		return;
548 	if (xprt_connected(xprt))
549 		xprt_release_write(xprt, task);
550 	else {
551 		if (task->tk_rqstp)
552 			task->tk_rqstp->rq_bytes_sent = 0;
553 
554 		task->tk_timeout = xprt->connect_timeout;
555 		rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
556 		xprt->ops->connect(task);
557 	}
558 	return;
559 }
560 
561 static void xprt_connect_status(struct rpc_task *task)
562 {
563 	struct rpc_xprt	*xprt = task->tk_xprt;
564 
565 	if (task->tk_status >= 0) {
566 		dprintk("RPC: %4d xprt_connect_status: connection established\n",
567 				task->tk_pid);
568 		return;
569 	}
570 
571 	switch (task->tk_status) {
572 	case -ECONNREFUSED:
573 	case -ECONNRESET:
574 		dprintk("RPC: %4d xprt_connect_status: server %s refused connection\n",
575 				task->tk_pid, task->tk_client->cl_server);
576 		break;
577 	case -ENOTCONN:
578 		dprintk("RPC: %4d xprt_connect_status: connection broken\n",
579 				task->tk_pid);
580 		break;
581 	case -ETIMEDOUT:
582 		dprintk("RPC: %4d xprt_connect_status: connect attempt timed out\n",
583 				task->tk_pid);
584 		break;
585 	default:
586 		dprintk("RPC: %4d xprt_connect_status: error %d connecting to server %s\n",
587 				task->tk_pid, -task->tk_status, task->tk_client->cl_server);
588 		xprt_release_write(xprt, task);
589 		task->tk_status = -EIO;
590 		return;
591 	}
592 
593 	/* if soft mounted, just cause this RPC to fail */
594 	if (RPC_IS_SOFT(task)) {
595 		xprt_release_write(xprt, task);
596 		task->tk_status = -EIO;
597 	}
598 }
599 
600 /**
601  * xprt_lookup_rqst - find an RPC request corresponding to an XID
602  * @xprt: transport on which the original request was transmitted
603  * @xid: RPC XID of incoming reply
604  *
605  */
606 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
607 {
608 	struct list_head *pos;
609 	struct rpc_rqst	*req = NULL;
610 
611 	list_for_each(pos, &xprt->recv) {
612 		struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
613 		if (entry->rq_xid == xid) {
614 			req = entry;
615 			break;
616 		}
617 	}
618 	return req;
619 }
620 
621 /**
622  * xprt_update_rtt - update an RPC client's RTT state after receiving a reply
623  * @task: RPC request that recently completed
624  *
625  */
626 void xprt_update_rtt(struct rpc_task *task)
627 {
628 	struct rpc_rqst *req = task->tk_rqstp;
629 	struct rpc_rtt *rtt = task->tk_client->cl_rtt;
630 	unsigned timer = task->tk_msg.rpc_proc->p_timer;
631 
632 	if (timer) {
633 		if (req->rq_ntrans == 1)
634 			rpc_update_rtt(rtt, timer,
635 					(long)jiffies - req->rq_xtime);
636 		rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
637 	}
638 }
639 
640 /**
641  * xprt_complete_rqst - called when reply processing is complete
642  * @task: RPC request that recently completed
643  * @copied: actual number of bytes received from the transport
644  *
645  * Caller holds transport lock.
646  */
647 void xprt_complete_rqst(struct rpc_task *task, int copied)
648 {
649 	struct rpc_rqst *req = task->tk_rqstp;
650 
651 	dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
652 			task->tk_pid, ntohl(req->rq_xid), copied);
653 
654 	list_del_init(&req->rq_list);
655 	req->rq_received = req->rq_private_buf.len = copied;
656 	rpc_wake_up_task(task);
657 }
658 
659 static void xprt_timer(struct rpc_task *task)
660 {
661 	struct rpc_rqst *req = task->tk_rqstp;
662 	struct rpc_xprt *xprt = req->rq_xprt;
663 
664 	dprintk("RPC: %4d xprt_timer\n", task->tk_pid);
665 
666 	spin_lock(&xprt->transport_lock);
667 	if (!req->rq_received) {
668 		if (xprt->ops->timer)
669 			xprt->ops->timer(task);
670 		task->tk_status = -ETIMEDOUT;
671 	}
672 	task->tk_timeout = 0;
673 	rpc_wake_up_task(task);
674 	spin_unlock(&xprt->transport_lock);
675 }
676 
677 /**
678  * xprt_prepare_transmit - reserve the transport before sending a request
679  * @task: RPC task about to send a request
680  *
681  */
682 int xprt_prepare_transmit(struct rpc_task *task)
683 {
684 	struct rpc_rqst	*req = task->tk_rqstp;
685 	struct rpc_xprt	*xprt = req->rq_xprt;
686 	int err = 0;
687 
688 	dprintk("RPC: %4d xprt_prepare_transmit\n", task->tk_pid);
689 
690 	if (xprt->shutdown)
691 		return -EIO;
692 
693 	spin_lock_bh(&xprt->transport_lock);
694 	if (req->rq_received && !req->rq_bytes_sent) {
695 		err = req->rq_received;
696 		goto out_unlock;
697 	}
698 	if (!xprt->ops->reserve_xprt(task)) {
699 		err = -EAGAIN;
700 		goto out_unlock;
701 	}
702 
703 	if (!xprt_connected(xprt)) {
704 		err = -ENOTCONN;
705 		goto out_unlock;
706 	}
707 out_unlock:
708 	spin_unlock_bh(&xprt->transport_lock);
709 	return err;
710 }
711 
712 void
713 xprt_abort_transmit(struct rpc_task *task)
714 {
715 	struct rpc_xprt	*xprt = task->tk_xprt;
716 
717 	xprt_release_write(xprt, task);
718 }
719 
720 /**
721  * xprt_transmit - send an RPC request on a transport
722  * @task: controlling RPC task
723  *
724  * We have to copy the iovec because sendmsg fiddles with its contents.
725  */
726 void xprt_transmit(struct rpc_task *task)
727 {
728 	struct rpc_rqst	*req = task->tk_rqstp;
729 	struct rpc_xprt	*xprt = req->rq_xprt;
730 	int status;
731 
732 	dprintk("RPC: %4d xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
733 
734 	smp_rmb();
735 	if (!req->rq_received) {
736 		if (list_empty(&req->rq_list)) {
737 			spin_lock_bh(&xprt->transport_lock);
738 			/* Update the softirq receive buffer */
739 			memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
740 					sizeof(req->rq_private_buf));
741 			/* Add request to the receive list */
742 			list_add_tail(&req->rq_list, &xprt->recv);
743 			spin_unlock_bh(&xprt->transport_lock);
744 			xprt_reset_majortimeo(req);
745 			/* Turn off autodisconnect */
746 			del_singleshot_timer_sync(&xprt->timer);
747 		}
748 	} else if (!req->rq_bytes_sent)
749 		return;
750 
751 	status = xprt->ops->send_request(task);
752 	if (status == 0) {
753 		dprintk("RPC: %4d xmit complete\n", task->tk_pid);
754 		spin_lock_bh(&xprt->transport_lock);
755 		xprt->ops->set_retrans_timeout(task);
756 		/* Don't race with disconnect */
757 		if (!xprt_connected(xprt))
758 			task->tk_status = -ENOTCONN;
759 		else if (!req->rq_received)
760 			rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer);
761 		xprt->ops->release_xprt(xprt, task);
762 		spin_unlock_bh(&xprt->transport_lock);
763 		return;
764 	}
765 
766 	/* Note: at this point, task->tk_sleeping has not yet been set,
767 	 *	 hence there is no danger of the waking up task being put on
768 	 *	 schedq, and being picked up by a parallel run of rpciod().
769 	 */
770 	task->tk_status = status;
771 
772 	switch (status) {
773 	case -ECONNREFUSED:
774 		rpc_sleep_on(&xprt->sending, task, NULL, NULL);
775 	case -EAGAIN:
776 	case -ENOTCONN:
777 		return;
778 	default:
779 		break;
780 	}
781 	xprt_release_write(xprt, task);
782 	return;
783 }
784 
785 static inline void do_xprt_reserve(struct rpc_task *task)
786 {
787 	struct rpc_xprt	*xprt = task->tk_xprt;
788 
789 	task->tk_status = 0;
790 	if (task->tk_rqstp)
791 		return;
792 	if (!list_empty(&xprt->free)) {
793 		struct rpc_rqst	*req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
794 		list_del_init(&req->rq_list);
795 		task->tk_rqstp = req;
796 		xprt_request_init(task, xprt);
797 		return;
798 	}
799 	dprintk("RPC:      waiting for request slot\n");
800 	task->tk_status = -EAGAIN;
801 	task->tk_timeout = 0;
802 	rpc_sleep_on(&xprt->backlog, task, NULL, NULL);
803 }
804 
805 /**
806  * xprt_reserve - allocate an RPC request slot
807  * @task: RPC task requesting a slot allocation
808  *
809  * If no more slots are available, place the task on the transport's
810  * backlog queue.
811  */
812 void xprt_reserve(struct rpc_task *task)
813 {
814 	struct rpc_xprt	*xprt = task->tk_xprt;
815 
816 	task->tk_status = -EIO;
817 	if (!xprt->shutdown) {
818 		spin_lock(&xprt->reserve_lock);
819 		do_xprt_reserve(task);
820 		spin_unlock(&xprt->reserve_lock);
821 	}
822 }
823 
824 static inline u32 xprt_alloc_xid(struct rpc_xprt *xprt)
825 {
826 	return xprt->xid++;
827 }
828 
829 static inline void xprt_init_xid(struct rpc_xprt *xprt)
830 {
831 	get_random_bytes(&xprt->xid, sizeof(xprt->xid));
832 }
833 
834 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
835 {
836 	struct rpc_rqst	*req = task->tk_rqstp;
837 
838 	req->rq_timeout = xprt->timeout.to_initval;
839 	req->rq_task	= task;
840 	req->rq_xprt    = xprt;
841 	req->rq_xid     = xprt_alloc_xid(xprt);
842 	req->rq_release_snd_buf = NULL;
843 	dprintk("RPC: %4d reserved req %p xid %08x\n", task->tk_pid,
844 			req, ntohl(req->rq_xid));
845 }
846 
847 /**
848  * xprt_release - release an RPC request slot
849  * @task: task which is finished with the slot
850  *
851  */
852 void xprt_release(struct rpc_task *task)
853 {
854 	struct rpc_xprt	*xprt = task->tk_xprt;
855 	struct rpc_rqst	*req;
856 
857 	if (!(req = task->tk_rqstp))
858 		return;
859 	spin_lock_bh(&xprt->transport_lock);
860 	xprt->ops->release_xprt(xprt, task);
861 	if (xprt->ops->release_request)
862 		xprt->ops->release_request(task);
863 	if (!list_empty(&req->rq_list))
864 		list_del(&req->rq_list);
865 	xprt->last_used = jiffies;
866 	if (list_empty(&xprt->recv) && !xprt->shutdown)
867 		mod_timer(&xprt->timer,
868 				xprt->last_used + xprt->idle_timeout);
869 	spin_unlock_bh(&xprt->transport_lock);
870 	task->tk_rqstp = NULL;
871 	if (req->rq_release_snd_buf)
872 		req->rq_release_snd_buf(req);
873 	memset(req, 0, sizeof(*req));	/* mark unused */
874 
875 	dprintk("RPC: %4d release request %p\n", task->tk_pid, req);
876 
877 	spin_lock(&xprt->reserve_lock);
878 	list_add(&req->rq_list, &xprt->free);
879 	rpc_wake_up_next(&xprt->backlog);
880 	spin_unlock(&xprt->reserve_lock);
881 }
882 
883 /**
884  * xprt_set_timeout - set constant RPC timeout
885  * @to: RPC timeout parameters to set up
886  * @retr: number of retries
887  * @incr: amount of increase after each retry
888  *
889  */
890 void xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
891 {
892 	to->to_initval   =
893 	to->to_increment = incr;
894 	to->to_maxval    = to->to_initval + (incr * retr);
895 	to->to_retries   = retr;
896 	to->to_exponential = 0;
897 }
898 
899 static struct rpc_xprt *xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
900 {
901 	int result;
902 	struct rpc_xprt	*xprt;
903 	struct rpc_rqst	*req;
904 
905 	if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL)
906 		return ERR_PTR(-ENOMEM);
907 	memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */
908 
909 	xprt->addr = *ap;
910 
911 	switch (proto) {
912 	case IPPROTO_UDP:
913 		result = xs_setup_udp(xprt, to);
914 		break;
915 	case IPPROTO_TCP:
916 		result = xs_setup_tcp(xprt, to);
917 		break;
918 	default:
919 		printk(KERN_ERR "RPC: unrecognized transport protocol: %d\n",
920 				proto);
921 		result = -EIO;
922 		break;
923 	}
924 	if (result) {
925 		kfree(xprt);
926 		return ERR_PTR(result);
927 	}
928 
929 	spin_lock_init(&xprt->transport_lock);
930 	spin_lock_init(&xprt->reserve_lock);
931 
932 	INIT_LIST_HEAD(&xprt->free);
933 	INIT_LIST_HEAD(&xprt->recv);
934 	INIT_WORK(&xprt->task_cleanup, xprt_autoclose, xprt);
935 	init_timer(&xprt->timer);
936 	xprt->timer.function = xprt_init_autodisconnect;
937 	xprt->timer.data = (unsigned long) xprt;
938 	xprt->last_used = jiffies;
939 	xprt->cwnd = RPC_INITCWND;
940 
941 	rpc_init_wait_queue(&xprt->pending, "xprt_pending");
942 	rpc_init_wait_queue(&xprt->sending, "xprt_sending");
943 	rpc_init_wait_queue(&xprt->resend, "xprt_resend");
944 	rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
945 
946 	/* initialize free list */
947 	for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
948 		list_add(&req->rq_list, &xprt->free);
949 
950 	xprt_init_xid(xprt);
951 
952 	dprintk("RPC:      created transport %p with %u slots\n", xprt,
953 			xprt->max_reqs);
954 
955 	return xprt;
956 }
957 
958 /**
959  * xprt_create_proto - create an RPC client transport
960  * @proto: requested transport protocol
961  * @sap: remote peer's address
962  * @to: timeout parameters for new transport
963  *
964  */
965 struct rpc_xprt *xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to)
966 {
967 	struct rpc_xprt	*xprt;
968 
969 	xprt = xprt_setup(proto, sap, to);
970 	if (IS_ERR(xprt))
971 		dprintk("RPC:      xprt_create_proto failed\n");
972 	else
973 		dprintk("RPC:      xprt_create_proto created xprt %p\n", xprt);
974 	return xprt;
975 }
976 
977 static void xprt_shutdown(struct rpc_xprt *xprt)
978 {
979 	xprt->shutdown = 1;
980 	rpc_wake_up(&xprt->sending);
981 	rpc_wake_up(&xprt->resend);
982 	xprt_wake_pending_tasks(xprt, -EIO);
983 	rpc_wake_up(&xprt->backlog);
984 	del_timer_sync(&xprt->timer);
985 }
986 
987 /**
988  * xprt_destroy - destroy an RPC transport, killing off all requests.
989  * @xprt: transport to destroy
990  *
991  */
992 int xprt_destroy(struct rpc_xprt *xprt)
993 {
994 	dprintk("RPC:      destroying transport %p\n", xprt);
995 	xprt_shutdown(xprt);
996 	xprt->ops->destroy(xprt);
997 	kfree(xprt);
998 
999 	return 0;
1000 }
1001