SUNRPC: Move exported declarations to the function declarations
[linux-2.6] / net / sunrpc / xprt.c
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         result = -EINVAL;
112         if (try_module_get(THIS_MODULE)) {
113                 list_add_tail(&transport->list, &xprt_list);
114                 printk(KERN_INFO "RPC: Registered %s transport module.\n",
115                         transport->name);
116                 result = 0;
117         }
118
119 out:
120         spin_unlock(&xprt_list_lock);
121         return result;
122 }
123 EXPORT_SYMBOL_GPL(xprt_register_transport);
124
125 /**
126  * xprt_unregister_transport - unregister a transport implementation
127  * transport: transport to unregister
128  *
129  * Returns:
130  * 0:           transport successfully unregistered
131  * -ENOENT:     transport never registered
132  */
133 int xprt_unregister_transport(struct xprt_class *transport)
134 {
135         struct xprt_class *t;
136         int result;
137
138         result = 0;
139         spin_lock(&xprt_list_lock);
140         list_for_each_entry(t, &xprt_list, list) {
141                 if (t == transport) {
142                         printk(KERN_INFO
143                                 "RPC: Unregistered %s transport module.\n",
144                                 transport->name);
145                         list_del_init(&transport->list);
146                         module_put(THIS_MODULE);
147                         goto out;
148                 }
149         }
150         result = -ENOENT;
151
152 out:
153         spin_unlock(&xprt_list_lock);
154         return result;
155 }
156 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
157
158 /**
159  * xprt_reserve_xprt - serialize write access to transports
160  * @task: task that is requesting access to the transport
161  *
162  * This prevents mixing the payload of separate requests, and prevents
163  * transport connects from colliding with writes.  No congestion control
164  * is provided.
165  */
166 int xprt_reserve_xprt(struct rpc_task *task)
167 {
168         struct rpc_xprt *xprt = task->tk_xprt;
169         struct rpc_rqst *req = task->tk_rqstp;
170
171         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
172                 if (task == xprt->snd_task)
173                         return 1;
174                 if (task == NULL)
175                         return 0;
176                 goto out_sleep;
177         }
178         xprt->snd_task = task;
179         if (req) {
180                 req->rq_bytes_sent = 0;
181                 req->rq_ntrans++;
182         }
183         return 1;
184
185 out_sleep:
186         dprintk("RPC: %5u failed to lock transport %p\n",
187                         task->tk_pid, xprt);
188         task->tk_timeout = 0;
189         task->tk_status = -EAGAIN;
190         if (req && req->rq_ntrans)
191                 rpc_sleep_on(&xprt->resend, task, NULL, NULL);
192         else
193                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
194         return 0;
195 }
196 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
197
198 static void xprt_clear_locked(struct rpc_xprt *xprt)
199 {
200         xprt->snd_task = NULL;
201         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state) || xprt->shutdown) {
202                 smp_mb__before_clear_bit();
203                 clear_bit(XPRT_LOCKED, &xprt->state);
204                 smp_mb__after_clear_bit();
205         } else
206                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
207 }
208
209 /*
210  * xprt_reserve_xprt_cong - serialize write access to transports
211  * @task: task that is requesting access to the transport
212  *
213  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
214  * integrated into the decision of whether a request is allowed to be
215  * woken up and given access to the transport.
216  */
217 int xprt_reserve_xprt_cong(struct rpc_task *task)
218 {
219         struct rpc_xprt *xprt = task->tk_xprt;
220         struct rpc_rqst *req = task->tk_rqstp;
221
222         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
223                 if (task == xprt->snd_task)
224                         return 1;
225                 goto out_sleep;
226         }
227         if (__xprt_get_cong(xprt, task)) {
228                 xprt->snd_task = task;
229                 if (req) {
230                         req->rq_bytes_sent = 0;
231                         req->rq_ntrans++;
232                 }
233                 return 1;
234         }
235         xprt_clear_locked(xprt);
236 out_sleep:
237         dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
238         task->tk_timeout = 0;
239         task->tk_status = -EAGAIN;
240         if (req && req->rq_ntrans)
241                 rpc_sleep_on(&xprt->resend, task, NULL, NULL);
242         else
243                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
244         return 0;
245 }
246 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
247
248 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
249 {
250         int retval;
251
252         spin_lock_bh(&xprt->transport_lock);
253         retval = xprt->ops->reserve_xprt(task);
254         spin_unlock_bh(&xprt->transport_lock);
255         return retval;
256 }
257
258 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
259 {
260         struct rpc_task *task;
261         struct rpc_rqst *req;
262
263         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
264                 return;
265
266         task = rpc_wake_up_next(&xprt->resend);
267         if (!task) {
268                 task = rpc_wake_up_next(&xprt->sending);
269                 if (!task)
270                         goto out_unlock;
271         }
272
273         req = task->tk_rqstp;
274         xprt->snd_task = task;
275         if (req) {
276                 req->rq_bytes_sent = 0;
277                 req->rq_ntrans++;
278         }
279         return;
280
281 out_unlock:
282         xprt_clear_locked(xprt);
283 }
284
285 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
286 {
287         struct rpc_task *task;
288
289         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
290                 return;
291         if (RPCXPRT_CONGESTED(xprt))
292                 goto out_unlock;
293         task = rpc_wake_up_next(&xprt->resend);
294         if (!task) {
295                 task = rpc_wake_up_next(&xprt->sending);
296                 if (!task)
297                         goto out_unlock;
298         }
299         if (__xprt_get_cong(xprt, task)) {
300                 struct rpc_rqst *req = task->tk_rqstp;
301                 xprt->snd_task = task;
302                 if (req) {
303                         req->rq_bytes_sent = 0;
304                         req->rq_ntrans++;
305                 }
306                 return;
307         }
308 out_unlock:
309         xprt_clear_locked(xprt);
310 }
311
312 /**
313  * xprt_release_xprt - allow other requests to use a transport
314  * @xprt: transport with other tasks potentially waiting
315  * @task: task that is releasing access to the transport
316  *
317  * Note that "task" can be NULL.  No congestion control is provided.
318  */
319 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
320 {
321         if (xprt->snd_task == task) {
322                 xprt_clear_locked(xprt);
323                 __xprt_lock_write_next(xprt);
324         }
325 }
326 EXPORT_SYMBOL_GPL(xprt_release_xprt);
327
328 /**
329  * xprt_release_xprt_cong - allow other requests to use a transport
330  * @xprt: transport with other tasks potentially waiting
331  * @task: task that is releasing access to the transport
332  *
333  * Note that "task" can be NULL.  Another task is awoken to use the
334  * transport if the transport's congestion window allows it.
335  */
336 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
337 {
338         if (xprt->snd_task == task) {
339                 xprt_clear_locked(xprt);
340                 __xprt_lock_write_next_cong(xprt);
341         }
342 }
343 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
344
345 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
346 {
347         spin_lock_bh(&xprt->transport_lock);
348         xprt->ops->release_xprt(xprt, task);
349         spin_unlock_bh(&xprt->transport_lock);
350 }
351
352 /*
353  * Van Jacobson congestion avoidance. Check if the congestion window
354  * overflowed. Put the task to sleep if this is the case.
355  */
356 static int
357 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_task *task)
358 {
359         struct rpc_rqst *req = task->tk_rqstp;
360
361         if (req->rq_cong)
362                 return 1;
363         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
364                         task->tk_pid, xprt->cong, xprt->cwnd);
365         if (RPCXPRT_CONGESTED(xprt))
366                 return 0;
367         req->rq_cong = 1;
368         xprt->cong += RPC_CWNDSCALE;
369         return 1;
370 }
371
372 /*
373  * Adjust the congestion window, and wake up the next task
374  * that has been sleeping due to congestion
375  */
376 static void
377 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
378 {
379         if (!req->rq_cong)
380                 return;
381         req->rq_cong = 0;
382         xprt->cong -= RPC_CWNDSCALE;
383         __xprt_lock_write_next_cong(xprt);
384 }
385
386 /**
387  * xprt_release_rqst_cong - housekeeping when request is complete
388  * @task: RPC request that recently completed
389  *
390  * Useful for transports that require congestion control.
391  */
392 void xprt_release_rqst_cong(struct rpc_task *task)
393 {
394         __xprt_put_cong(task->tk_xprt, task->tk_rqstp);
395 }
396 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
397
398 /**
399  * xprt_adjust_cwnd - adjust transport congestion window
400  * @task: recently completed RPC request used to adjust window
401  * @result: result code of completed RPC request
402  *
403  * We use a time-smoothed congestion estimator to avoid heavy oscillation.
404  */
405 void xprt_adjust_cwnd(struct rpc_task *task, int result)
406 {
407         struct rpc_rqst *req = task->tk_rqstp;
408         struct rpc_xprt *xprt = task->tk_xprt;
409         unsigned long cwnd = xprt->cwnd;
410
411         if (result >= 0 && cwnd <= xprt->cong) {
412                 /* The (cwnd >> 1) term makes sure
413                  * the result gets rounded properly. */
414                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
415                 if (cwnd > RPC_MAXCWND(xprt))
416                         cwnd = RPC_MAXCWND(xprt);
417                 __xprt_lock_write_next_cong(xprt);
418         } else if (result == -ETIMEDOUT) {
419                 cwnd >>= 1;
420                 if (cwnd < RPC_CWNDSCALE)
421                         cwnd = RPC_CWNDSCALE;
422         }
423         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
424                         xprt->cong, xprt->cwnd, cwnd);
425         xprt->cwnd = cwnd;
426         __xprt_put_cong(xprt, req);
427 }
428 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
429
430 /**
431  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
432  * @xprt: transport with waiting tasks
433  * @status: result code to plant in each task before waking it
434  *
435  */
436 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
437 {
438         if (status < 0)
439                 rpc_wake_up_status(&xprt->pending, status);
440         else
441                 rpc_wake_up(&xprt->pending);
442 }
443 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
444
445 /**
446  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
447  * @task: task to be put to sleep
448  *
449  */
450 void xprt_wait_for_buffer_space(struct rpc_task *task)
451 {
452         struct rpc_rqst *req = task->tk_rqstp;
453         struct rpc_xprt *xprt = req->rq_xprt;
454
455         task->tk_timeout = req->rq_timeout;
456         rpc_sleep_on(&xprt->pending, task, NULL, NULL);
457 }
458 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
459
460 /**
461  * xprt_write_space - wake the task waiting for transport output buffer space
462  * @xprt: transport with waiting tasks
463  *
464  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
465  */
466 void xprt_write_space(struct rpc_xprt *xprt)
467 {
468         if (unlikely(xprt->shutdown))
469                 return;
470
471         spin_lock_bh(&xprt->transport_lock);
472         if (xprt->snd_task) {
473                 dprintk("RPC:       write space: waking waiting task on "
474                                 "xprt %p\n", xprt);
475                 rpc_wake_up_task(xprt->snd_task);
476         }
477         spin_unlock_bh(&xprt->transport_lock);
478 }
479 EXPORT_SYMBOL_GPL(xprt_write_space);
480
481 /**
482  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
483  * @task: task whose timeout is to be set
484  *
485  * Set a request's retransmit timeout based on the transport's
486  * default timeout parameters.  Used by transports that don't adjust
487  * the retransmit timeout based on round-trip time estimation.
488  */
489 void xprt_set_retrans_timeout_def(struct rpc_task *task)
490 {
491         task->tk_timeout = task->tk_rqstp->rq_timeout;
492 }
493 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
494
495 /*
496  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
497  * @task: task whose timeout is to be set
498  *
499  * Set a request's retransmit timeout using the RTT estimator.
500  */
501 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
502 {
503         int timer = task->tk_msg.rpc_proc->p_timer;
504         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
505         struct rpc_rqst *req = task->tk_rqstp;
506         unsigned long max_timeout = req->rq_xprt->timeout.to_maxval;
507
508         task->tk_timeout = rpc_calc_rto(rtt, timer);
509         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
510         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
511                 task->tk_timeout = max_timeout;
512 }
513 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
514
515 static void xprt_reset_majortimeo(struct rpc_rqst *req)
516 {
517         struct rpc_timeout *to = &req->rq_xprt->timeout;
518
519         req->rq_majortimeo = req->rq_timeout;
520         if (to->to_exponential)
521                 req->rq_majortimeo <<= to->to_retries;
522         else
523                 req->rq_majortimeo += to->to_increment * to->to_retries;
524         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
525                 req->rq_majortimeo = to->to_maxval;
526         req->rq_majortimeo += jiffies;
527 }
528
529 /**
530  * xprt_adjust_timeout - adjust timeout values for next retransmit
531  * @req: RPC request containing parameters to use for the adjustment
532  *
533  */
534 int xprt_adjust_timeout(struct rpc_rqst *req)
535 {
536         struct rpc_xprt *xprt = req->rq_xprt;
537         struct rpc_timeout *to = &xprt->timeout;
538         int status = 0;
539
540         if (time_before(jiffies, req->rq_majortimeo)) {
541                 if (to->to_exponential)
542                         req->rq_timeout <<= 1;
543                 else
544                         req->rq_timeout += to->to_increment;
545                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
546                         req->rq_timeout = to->to_maxval;
547                 req->rq_retries++;
548         } else {
549                 req->rq_timeout = to->to_initval;
550                 req->rq_retries = 0;
551                 xprt_reset_majortimeo(req);
552                 /* Reset the RTT counters == "slow start" */
553                 spin_lock_bh(&xprt->transport_lock);
554                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
555                 spin_unlock_bh(&xprt->transport_lock);
556                 status = -ETIMEDOUT;
557         }
558
559         if (req->rq_timeout == 0) {
560                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
561                 req->rq_timeout = 5 * HZ;
562         }
563         return status;
564 }
565
566 static void xprt_autoclose(struct work_struct *work)
567 {
568         struct rpc_xprt *xprt =
569                 container_of(work, struct rpc_xprt, task_cleanup);
570
571         xprt->ops->close(xprt);
572         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
573         xprt_release_write(xprt, NULL);
574 }
575
576 /**
577  * xprt_disconnect_done - mark a transport as disconnected
578  * @xprt: transport to flag for disconnect
579  *
580  */
581 void xprt_disconnect_done(struct rpc_xprt *xprt)
582 {
583         dprintk("RPC:       disconnected transport %p\n", xprt);
584         spin_lock_bh(&xprt->transport_lock);
585         xprt_clear_connected(xprt);
586         xprt_wake_pending_tasks(xprt, -ENOTCONN);
587         spin_unlock_bh(&xprt->transport_lock);
588 }
589 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
590
591 /**
592  * xprt_force_disconnect - force a transport to disconnect
593  * @xprt: transport to disconnect
594  *
595  */
596 void xprt_force_disconnect(struct rpc_xprt *xprt)
597 {
598         /* Don't race with the test_bit() in xprt_clear_locked() */
599         spin_lock_bh(&xprt->transport_lock);
600         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
601         /* Try to schedule an autoclose RPC call */
602         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
603                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
604         else if (xprt->snd_task != NULL)
605                 rpc_wake_up_task(xprt->snd_task);
606         spin_unlock_bh(&xprt->transport_lock);
607 }
608 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
609
610 static void
611 xprt_init_autodisconnect(unsigned long data)
612 {
613         struct rpc_xprt *xprt = (struct rpc_xprt *)data;
614
615         spin_lock(&xprt->transport_lock);
616         if (!list_empty(&xprt->recv) || xprt->shutdown)
617                 goto out_abort;
618         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
619                 goto out_abort;
620         spin_unlock(&xprt->transport_lock);
621         if (xprt_connecting(xprt))
622                 xprt_release_write(xprt, NULL);
623         else
624                 queue_work(rpciod_workqueue, &xprt->task_cleanup);
625         return;
626 out_abort:
627         spin_unlock(&xprt->transport_lock);
628 }
629
630 /**
631  * xprt_connect - schedule a transport connect operation
632  * @task: RPC task that is requesting the connect
633  *
634  */
635 void xprt_connect(struct rpc_task *task)
636 {
637         struct rpc_xprt *xprt = task->tk_xprt;
638
639         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
640                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
641
642         if (!xprt_bound(xprt)) {
643                 task->tk_status = -EIO;
644                 return;
645         }
646         if (!xprt_lock_write(xprt, task))
647                 return;
648         if (xprt_connected(xprt))
649                 xprt_release_write(xprt, task);
650         else {
651                 if (task->tk_rqstp)
652                         task->tk_rqstp->rq_bytes_sent = 0;
653
654                 task->tk_timeout = xprt->connect_timeout;
655                 rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
656                 xprt->stat.connect_start = jiffies;
657                 xprt->ops->connect(task);
658         }
659         return;
660 }
661
662 static void xprt_connect_status(struct rpc_task *task)
663 {
664         struct rpc_xprt *xprt = task->tk_xprt;
665
666         if (task->tk_status >= 0) {
667                 xprt->stat.connect_count++;
668                 xprt->stat.connect_time += (long)jiffies - xprt->stat.connect_start;
669                 dprintk("RPC: %5u xprt_connect_status: connection established\n",
670                                 task->tk_pid);
671                 return;
672         }
673
674         switch (task->tk_status) {
675         case -ECONNREFUSED:
676         case -ECONNRESET:
677                 dprintk("RPC: %5u xprt_connect_status: server %s refused "
678                                 "connection\n", task->tk_pid,
679                                 task->tk_client->cl_server);
680                 break;
681         case -ENOTCONN:
682                 dprintk("RPC: %5u xprt_connect_status: connection broken\n",
683                                 task->tk_pid);
684                 break;
685         case -ETIMEDOUT:
686                 dprintk("RPC: %5u xprt_connect_status: connect attempt timed "
687                                 "out\n", task->tk_pid);
688                 break;
689         default:
690                 dprintk("RPC: %5u xprt_connect_status: error %d connecting to "
691                                 "server %s\n", task->tk_pid, -task->tk_status,
692                                 task->tk_client->cl_server);
693                 xprt_release_write(xprt, task);
694                 task->tk_status = -EIO;
695         }
696 }
697
698 /**
699  * xprt_lookup_rqst - find an RPC request corresponding to an XID
700  * @xprt: transport on which the original request was transmitted
701  * @xid: RPC XID of incoming reply
702  *
703  */
704 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
705 {
706         struct list_head *pos;
707
708         list_for_each(pos, &xprt->recv) {
709                 struct rpc_rqst *entry = list_entry(pos, struct rpc_rqst, rq_list);
710                 if (entry->rq_xid == xid)
711                         return entry;
712         }
713
714         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
715                         ntohl(xid));
716         xprt->stat.bad_xids++;
717         return NULL;
718 }
719 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
720
721 /**
722  * xprt_update_rtt - update an RPC client's RTT state after receiving a reply
723  * @task: RPC request that recently completed
724  *
725  */
726 void xprt_update_rtt(struct rpc_task *task)
727 {
728         struct rpc_rqst *req = task->tk_rqstp;
729         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
730         unsigned timer = task->tk_msg.rpc_proc->p_timer;
731
732         if (timer) {
733                 if (req->rq_ntrans == 1)
734                         rpc_update_rtt(rtt, timer,
735                                         (long)jiffies - req->rq_xtime);
736                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
737         }
738 }
739 EXPORT_SYMBOL_GPL(xprt_update_rtt);
740
741 /**
742  * xprt_complete_rqst - called when reply processing is complete
743  * @task: RPC request that recently completed
744  * @copied: actual number of bytes received from the transport
745  *
746  * Caller holds transport lock.
747  */
748 void xprt_complete_rqst(struct rpc_task *task, int copied)
749 {
750         struct rpc_rqst *req = task->tk_rqstp;
751
752         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
753                         task->tk_pid, ntohl(req->rq_xid), copied);
754
755         task->tk_xprt->stat.recvs++;
756         task->tk_rtt = (long)jiffies - req->rq_xtime;
757
758         list_del_init(&req->rq_list);
759         /* Ensure all writes are done before we update req->rq_received */
760         smp_wmb();
761         req->rq_received = req->rq_private_buf.len = copied;
762         rpc_wake_up_task(task);
763 }
764 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
765
766 static void xprt_timer(struct rpc_task *task)
767 {
768         struct rpc_rqst *req = task->tk_rqstp;
769         struct rpc_xprt *xprt = req->rq_xprt;
770
771         dprintk("RPC: %5u xprt_timer\n", task->tk_pid);
772
773         spin_lock(&xprt->transport_lock);
774         if (!req->rq_received) {
775                 if (xprt->ops->timer)
776                         xprt->ops->timer(task);
777                 task->tk_status = -ETIMEDOUT;
778         }
779         task->tk_timeout = 0;
780         rpc_wake_up_task(task);
781         spin_unlock(&xprt->transport_lock);
782 }
783
784 /**
785  * xprt_prepare_transmit - reserve the transport before sending a request
786  * @task: RPC task about to send a request
787  *
788  */
789 int xprt_prepare_transmit(struct rpc_task *task)
790 {
791         struct rpc_rqst *req = task->tk_rqstp;
792         struct rpc_xprt *xprt = req->rq_xprt;
793         int err = 0;
794
795         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
796
797         spin_lock_bh(&xprt->transport_lock);
798         if (req->rq_received && !req->rq_bytes_sent) {
799                 err = req->rq_received;
800                 goto out_unlock;
801         }
802         if (!xprt->ops->reserve_xprt(task)) {
803                 err = -EAGAIN;
804                 goto out_unlock;
805         }
806
807         if (!xprt_connected(xprt)) {
808                 err = -ENOTCONN;
809                 goto out_unlock;
810         }
811 out_unlock:
812         spin_unlock_bh(&xprt->transport_lock);
813         return err;
814 }
815
816 void xprt_end_transmit(struct rpc_task *task)
817 {
818         xprt_release_write(task->tk_xprt, task);
819 }
820
821 /**
822  * xprt_transmit - send an RPC request on a transport
823  * @task: controlling RPC task
824  *
825  * We have to copy the iovec because sendmsg fiddles with its contents.
826  */
827 void xprt_transmit(struct rpc_task *task)
828 {
829         struct rpc_rqst *req = task->tk_rqstp;
830         struct rpc_xprt *xprt = req->rq_xprt;
831         int status;
832
833         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
834
835         if (!req->rq_received) {
836                 if (list_empty(&req->rq_list)) {
837                         spin_lock_bh(&xprt->transport_lock);
838                         /* Update the softirq receive buffer */
839                         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
840                                         sizeof(req->rq_private_buf));
841                         /* Add request to the receive list */
842                         list_add_tail(&req->rq_list, &xprt->recv);
843                         spin_unlock_bh(&xprt->transport_lock);
844                         xprt_reset_majortimeo(req);
845                         /* Turn off autodisconnect */
846                         del_singleshot_timer_sync(&xprt->timer);
847                 }
848         } else if (!req->rq_bytes_sent)
849                 return;
850
851         status = xprt->ops->send_request(task);
852         if (status == 0) {
853                 dprintk("RPC: %5u xmit complete\n", task->tk_pid);
854                 spin_lock_bh(&xprt->transport_lock);
855
856                 xprt->ops->set_retrans_timeout(task);
857
858                 xprt->stat.sends++;
859                 xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
860                 xprt->stat.bklog_u += xprt->backlog.qlen;
861
862                 /* Don't race with disconnect */
863                 if (!xprt_connected(xprt))
864                         task->tk_status = -ENOTCONN;
865                 else if (!req->rq_received)
866                         rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer);
867                 spin_unlock_bh(&xprt->transport_lock);
868                 return;
869         }
870
871         /* Note: at this point, task->tk_sleeping has not yet been set,
872          *       hence there is no danger of the waking up task being put on
873          *       schedq, and being picked up by a parallel run of rpciod().
874          */
875         task->tk_status = status;
876         if (status == -ECONNREFUSED)
877                 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
878 }
879
880 static inline void do_xprt_reserve(struct rpc_task *task)
881 {
882         struct rpc_xprt *xprt = task->tk_xprt;
883
884         task->tk_status = 0;
885         if (task->tk_rqstp)
886                 return;
887         if (!list_empty(&xprt->free)) {
888                 struct rpc_rqst *req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
889                 list_del_init(&req->rq_list);
890                 task->tk_rqstp = req;
891                 xprt_request_init(task, xprt);
892                 return;
893         }
894         dprintk("RPC:       waiting for request slot\n");
895         task->tk_status = -EAGAIN;
896         task->tk_timeout = 0;
897         rpc_sleep_on(&xprt->backlog, task, NULL, NULL);
898 }
899
900 /**
901  * xprt_reserve - allocate an RPC request slot
902  * @task: RPC task requesting a slot allocation
903  *
904  * If no more slots are available, place the task on the transport's
905  * backlog queue.
906  */
907 void xprt_reserve(struct rpc_task *task)
908 {
909         struct rpc_xprt *xprt = task->tk_xprt;
910
911         task->tk_status = -EIO;
912         spin_lock(&xprt->reserve_lock);
913         do_xprt_reserve(task);
914         spin_unlock(&xprt->reserve_lock);
915 }
916
917 static inline __be32 xprt_alloc_xid(struct rpc_xprt *xprt)
918 {
919         return xprt->xid++;
920 }
921
922 static inline void xprt_init_xid(struct rpc_xprt *xprt)
923 {
924         xprt->xid = net_random();
925 }
926
927 static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
928 {
929         struct rpc_rqst *req = task->tk_rqstp;
930
931         req->rq_timeout = xprt->timeout.to_initval;
932         req->rq_task    = task;
933         req->rq_xprt    = xprt;
934         req->rq_buffer  = NULL;
935         req->rq_xid     = xprt_alloc_xid(xprt);
936         req->rq_release_snd_buf = NULL;
937         xprt_reset_majortimeo(req);
938         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
939                         req, ntohl(req->rq_xid));
940 }
941
942 /**
943  * xprt_release - release an RPC request slot
944  * @task: task which is finished with the slot
945  *
946  */
947 void xprt_release(struct rpc_task *task)
948 {
949         struct rpc_xprt *xprt = task->tk_xprt;
950         struct rpc_rqst *req;
951
952         if (!(req = task->tk_rqstp))
953                 return;
954         rpc_count_iostats(task);
955         spin_lock_bh(&xprt->transport_lock);
956         xprt->ops->release_xprt(xprt, task);
957         if (xprt->ops->release_request)
958                 xprt->ops->release_request(task);
959         if (!list_empty(&req->rq_list))
960                 list_del(&req->rq_list);
961         xprt->last_used = jiffies;
962         if (list_empty(&xprt->recv))
963                 mod_timer(&xprt->timer,
964                                 xprt->last_used + xprt->idle_timeout);
965         spin_unlock_bh(&xprt->transport_lock);
966         xprt->ops->buf_free(req->rq_buffer);
967         task->tk_rqstp = NULL;
968         if (req->rq_release_snd_buf)
969                 req->rq_release_snd_buf(req);
970         memset(req, 0, sizeof(*req));   /* mark unused */
971
972         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
973
974         spin_lock(&xprt->reserve_lock);
975         list_add(&req->rq_list, &xprt->free);
976         rpc_wake_up_next(&xprt->backlog);
977         spin_unlock(&xprt->reserve_lock);
978 }
979
980 /**
981  * xprt_set_timeout - set constant RPC timeout
982  * @to: RPC timeout parameters to set up
983  * @retr: number of retries
984  * @incr: amount of increase after each retry
985  *
986  */
987 void xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
988 {
989         to->to_initval   =
990         to->to_increment = incr;
991         to->to_maxval    = to->to_initval + (incr * retr);
992         to->to_retries   = retr;
993         to->to_exponential = 0;
994 }
995 EXPORT_SYMBOL(xprt_set_timeout);
996
997 /**
998  * xprt_create_transport - create an RPC transport
999  * @args: rpc transport creation arguments
1000  *
1001  */
1002 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1003 {
1004         struct rpc_xprt *xprt;
1005         struct rpc_rqst *req;
1006         struct xprt_class *t;
1007
1008         spin_lock(&xprt_list_lock);
1009         list_for_each_entry(t, &xprt_list, list) {
1010                 if (t->ident == args->ident) {
1011                         spin_unlock(&xprt_list_lock);
1012                         goto found;
1013                 }
1014         }
1015         spin_unlock(&xprt_list_lock);
1016         printk(KERN_ERR "RPC: transport (%d) not supported\n", args->ident);
1017         return ERR_PTR(-EIO);
1018
1019 found:
1020         xprt = t->setup(args);
1021         if (IS_ERR(xprt)) {
1022                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1023                                 -PTR_ERR(xprt));
1024                 return xprt;
1025         }
1026
1027         kref_init(&xprt->kref);
1028         spin_lock_init(&xprt->transport_lock);
1029         spin_lock_init(&xprt->reserve_lock);
1030
1031         INIT_LIST_HEAD(&xprt->free);
1032         INIT_LIST_HEAD(&xprt->recv);
1033         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1034         setup_timer(&xprt->timer, xprt_init_autodisconnect,
1035                         (unsigned long)xprt);
1036         xprt->last_used = jiffies;
1037         xprt->cwnd = RPC_INITCWND;
1038         xprt->bind_index = 0;
1039
1040         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1041         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1042         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1043         rpc_init_wait_queue(&xprt->resend, "xprt_resend");
1044         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1045
1046         /* initialize free list */
1047         for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
1048                 list_add(&req->rq_list, &xprt->free);
1049
1050         xprt_init_xid(xprt);
1051
1052         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1053                         xprt->max_reqs);
1054
1055         return xprt;
1056 }
1057
1058 /**
1059  * xprt_destroy - destroy an RPC transport, killing off all requests.
1060  * @kref: kref for the transport to destroy
1061  *
1062  */
1063 static void xprt_destroy(struct kref *kref)
1064 {
1065         struct rpc_xprt *xprt = container_of(kref, struct rpc_xprt, kref);
1066
1067         dprintk("RPC:       destroying transport %p\n", xprt);
1068         xprt->shutdown = 1;
1069         del_timer_sync(&xprt->timer);
1070
1071         /*
1072          * Tear down transport state and free the rpc_xprt
1073          */
1074         xprt->ops->destroy(xprt);
1075 }
1076
1077 /**
1078  * xprt_put - release a reference to an RPC transport.
1079  * @xprt: pointer to the transport
1080  *
1081  */
1082 void xprt_put(struct rpc_xprt *xprt)
1083 {
1084         kref_put(&xprt->kref, xprt_destroy);
1085 }
1086
1087 /**
1088  * xprt_get - return a reference to an RPC transport.
1089  * @xprt: pointer to the transport
1090  *
1091  */
1092 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1093 {
1094         kref_get(&xprt->kref);
1095         return xprt;
1096 }