Merge tag 'iommu-updates-v4.21' of git://git.kernel.org/pub/scm/linux/kernel/git...
[sfrench/cifs-2.6.git] / 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, if a reply is expected,
16  *      it installs a timer that is run after the packet's timeout has
17  *      expired.
18  *  -   When a packet arrives, the data_ready handler walks the list of
19  *      pending requests for that transport. If a matching XID is found, the
20  *      caller is woken up, and the timer removed.
21  *  -   When no reply arrives within the timeout interval, the timer is
22  *      fired by the kernel and runs xprt_timer(). It either adjusts the
23  *      timeout values (minor timeout) or wakes up the caller with a status
24  *      of -ETIMEDOUT.
25  *  -   When the caller receives a notification from RPC that a reply arrived,
26  *      it should release the RPC slot, and process the reply.
27  *      If the call timed out, it may choose to retry the operation by
28  *      adjusting the initial timeout value, and simply calling rpc_call
29  *      again.
30  *
31  *  Support for async RPC is done through a set of RPC-specific scheduling
32  *  primitives that `transparently' work for processes as well as async
33  *  tasks that rely on callbacks.
34  *
35  *  Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
36  *
37  *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
38  */
39
40 #include <linux/module.h>
41
42 #include <linux/types.h>
43 #include <linux/interrupt.h>
44 #include <linux/workqueue.h>
45 #include <linux/net.h>
46 #include <linux/ktime.h>
47
48 #include <linux/sunrpc/clnt.h>
49 #include <linux/sunrpc/metrics.h>
50 #include <linux/sunrpc/bc_xprt.h>
51 #include <linux/rcupdate.h>
52
53 #include <trace/events/sunrpc.h>
54
55 #include "sunrpc.h"
56
57 /*
58  * Local variables
59  */
60
61 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
62 # define RPCDBG_FACILITY        RPCDBG_XPRT
63 #endif
64
65 /*
66  * Local functions
67  */
68 static void      xprt_init(struct rpc_xprt *xprt, struct net *net);
69 static __be32   xprt_alloc_xid(struct rpc_xprt *xprt);
70 static void      xprt_destroy(struct rpc_xprt *xprt);
71
72 static DEFINE_SPINLOCK(xprt_list_lock);
73 static LIST_HEAD(xprt_list);
74
75 /**
76  * xprt_register_transport - register a transport implementation
77  * @transport: transport to register
78  *
79  * If a transport implementation is loaded as a kernel module, it can
80  * call this interface to make itself known to the RPC client.
81  *
82  * Returns:
83  * 0:           transport successfully registered
84  * -EEXIST:     transport already registered
85  * -EINVAL:     transport module being unloaded
86  */
87 int xprt_register_transport(struct xprt_class *transport)
88 {
89         struct xprt_class *t;
90         int result;
91
92         result = -EEXIST;
93         spin_lock(&xprt_list_lock);
94         list_for_each_entry(t, &xprt_list, list) {
95                 /* don't register the same transport class twice */
96                 if (t->ident == transport->ident)
97                         goto out;
98         }
99
100         list_add_tail(&transport->list, &xprt_list);
101         printk(KERN_INFO "RPC: Registered %s transport module.\n",
102                transport->name);
103         result = 0;
104
105 out:
106         spin_unlock(&xprt_list_lock);
107         return result;
108 }
109 EXPORT_SYMBOL_GPL(xprt_register_transport);
110
111 /**
112  * xprt_unregister_transport - unregister a transport implementation
113  * @transport: transport to unregister
114  *
115  * Returns:
116  * 0:           transport successfully unregistered
117  * -ENOENT:     transport never registered
118  */
119 int xprt_unregister_transport(struct xprt_class *transport)
120 {
121         struct xprt_class *t;
122         int result;
123
124         result = 0;
125         spin_lock(&xprt_list_lock);
126         list_for_each_entry(t, &xprt_list, list) {
127                 if (t == transport) {
128                         printk(KERN_INFO
129                                 "RPC: Unregistered %s transport module.\n",
130                                 transport->name);
131                         list_del_init(&transport->list);
132                         goto out;
133                 }
134         }
135         result = -ENOENT;
136
137 out:
138         spin_unlock(&xprt_list_lock);
139         return result;
140 }
141 EXPORT_SYMBOL_GPL(xprt_unregister_transport);
142
143 /**
144  * xprt_load_transport - load a transport implementation
145  * @transport_name: transport to load
146  *
147  * Returns:
148  * 0:           transport successfully loaded
149  * -ENOENT:     transport module not available
150  */
151 int xprt_load_transport(const char *transport_name)
152 {
153         struct xprt_class *t;
154         int result;
155
156         result = 0;
157         spin_lock(&xprt_list_lock);
158         list_for_each_entry(t, &xprt_list, list) {
159                 if (strcmp(t->name, transport_name) == 0) {
160                         spin_unlock(&xprt_list_lock);
161                         goto out;
162                 }
163         }
164         spin_unlock(&xprt_list_lock);
165         result = request_module("xprt%s", transport_name);
166 out:
167         return result;
168 }
169 EXPORT_SYMBOL_GPL(xprt_load_transport);
170
171 static void xprt_clear_locked(struct rpc_xprt *xprt)
172 {
173         xprt->snd_task = NULL;
174         if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
175                 smp_mb__before_atomic();
176                 clear_bit(XPRT_LOCKED, &xprt->state);
177                 smp_mb__after_atomic();
178         } else
179                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
180 }
181
182 /**
183  * xprt_reserve_xprt - serialize write access to transports
184  * @task: task that is requesting access to the transport
185  * @xprt: pointer to the target transport
186  *
187  * This prevents mixing the payload of separate requests, and prevents
188  * transport connects from colliding with writes.  No congestion control
189  * is provided.
190  */
191 int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
192 {
193         struct rpc_rqst *req = task->tk_rqstp;
194
195         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
196                 if (task == xprt->snd_task)
197                         return 1;
198                 goto out_sleep;
199         }
200         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
201                 goto out_unlock;
202         xprt->snd_task = task;
203
204         return 1;
205
206 out_unlock:
207         xprt_clear_locked(xprt);
208 out_sleep:
209         dprintk("RPC: %5u failed to lock transport %p\n",
210                         task->tk_pid, xprt);
211         task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
212         task->tk_status = -EAGAIN;
213         rpc_sleep_on(&xprt->sending, task, NULL);
214         return 0;
215 }
216 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
217
218 static bool
219 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
220 {
221         return test_bit(XPRT_CWND_WAIT, &xprt->state);
222 }
223
224 static void
225 xprt_set_congestion_window_wait(struct rpc_xprt *xprt)
226 {
227         if (!list_empty(&xprt->xmit_queue)) {
228                 /* Peek at head of queue to see if it can make progress */
229                 if (list_first_entry(&xprt->xmit_queue, struct rpc_rqst,
230                                         rq_xmit)->rq_cong)
231                         return;
232         }
233         set_bit(XPRT_CWND_WAIT, &xprt->state);
234 }
235
236 static void
237 xprt_test_and_clear_congestion_window_wait(struct rpc_xprt *xprt)
238 {
239         if (!RPCXPRT_CONGESTED(xprt))
240                 clear_bit(XPRT_CWND_WAIT, &xprt->state);
241 }
242
243 /*
244  * xprt_reserve_xprt_cong - serialize write access to transports
245  * @task: task that is requesting access to the transport
246  *
247  * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
248  * integrated into the decision of whether a request is allowed to be
249  * woken up and given access to the transport.
250  * Note that the lock is only granted if we know there are free slots.
251  */
252 int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
253 {
254         struct rpc_rqst *req = task->tk_rqstp;
255
256         if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
257                 if (task == xprt->snd_task)
258                         return 1;
259                 goto out_sleep;
260         }
261         if (req == NULL) {
262                 xprt->snd_task = task;
263                 return 1;
264         }
265         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
266                 goto out_unlock;
267         if (!xprt_need_congestion_window_wait(xprt)) {
268                 xprt->snd_task = task;
269                 return 1;
270         }
271 out_unlock:
272         xprt_clear_locked(xprt);
273 out_sleep:
274         dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
275         task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
276         task->tk_status = -EAGAIN;
277         rpc_sleep_on(&xprt->sending, task, NULL);
278         return 0;
279 }
280 EXPORT_SYMBOL_GPL(xprt_reserve_xprt_cong);
281
282 static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
283 {
284         int retval;
285
286         if (test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == task)
287                 return 1;
288         spin_lock_bh(&xprt->transport_lock);
289         retval = xprt->ops->reserve_xprt(xprt, task);
290         spin_unlock_bh(&xprt->transport_lock);
291         return retval;
292 }
293
294 static bool __xprt_lock_write_func(struct rpc_task *task, void *data)
295 {
296         struct rpc_xprt *xprt = data;
297
298         xprt->snd_task = task;
299         return true;
300 }
301
302 static void __xprt_lock_write_next(struct rpc_xprt *xprt)
303 {
304         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
305                 return;
306         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
307                 goto out_unlock;
308         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
309                                 __xprt_lock_write_func, xprt))
310                 return;
311 out_unlock:
312         xprt_clear_locked(xprt);
313 }
314
315 static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
316 {
317         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
318                 return;
319         if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
320                 goto out_unlock;
321         if (xprt_need_congestion_window_wait(xprt))
322                 goto out_unlock;
323         if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
324                                 __xprt_lock_write_func, xprt))
325                 return;
326 out_unlock:
327         xprt_clear_locked(xprt);
328 }
329
330 /**
331  * xprt_release_xprt - allow other requests to use a transport
332  * @xprt: transport with other tasks potentially waiting
333  * @task: task that is releasing access to the transport
334  *
335  * Note that "task" can be NULL.  No congestion control is provided.
336  */
337 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
338 {
339         if (xprt->snd_task == task) {
340                 xprt_clear_locked(xprt);
341                 __xprt_lock_write_next(xprt);
342         }
343 }
344 EXPORT_SYMBOL_GPL(xprt_release_xprt);
345
346 /**
347  * xprt_release_xprt_cong - allow other requests to use a transport
348  * @xprt: transport with other tasks potentially waiting
349  * @task: task that is releasing access to the transport
350  *
351  * Note that "task" can be NULL.  Another task is awoken to use the
352  * transport if the transport's congestion window allows it.
353  */
354 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
355 {
356         if (xprt->snd_task == task) {
357                 xprt_clear_locked(xprt);
358                 __xprt_lock_write_next_cong(xprt);
359         }
360 }
361 EXPORT_SYMBOL_GPL(xprt_release_xprt_cong);
362
363 static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
364 {
365         if (xprt->snd_task != task)
366                 return;
367         spin_lock_bh(&xprt->transport_lock);
368         xprt->ops->release_xprt(xprt, task);
369         spin_unlock_bh(&xprt->transport_lock);
370 }
371
372 /*
373  * Van Jacobson congestion avoidance. Check if the congestion window
374  * overflowed. Put the task to sleep if this is the case.
375  */
376 static int
377 __xprt_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
378 {
379         if (req->rq_cong)
380                 return 1;
381         dprintk("RPC: %5u xprt_cwnd_limited cong = %lu cwnd = %lu\n",
382                         req->rq_task->tk_pid, xprt->cong, xprt->cwnd);
383         if (RPCXPRT_CONGESTED(xprt)) {
384                 xprt_set_congestion_window_wait(xprt);
385                 return 0;
386         }
387         req->rq_cong = 1;
388         xprt->cong += RPC_CWNDSCALE;
389         return 1;
390 }
391
392 /*
393  * Adjust the congestion window, and wake up the next task
394  * that has been sleeping due to congestion
395  */
396 static void
397 __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
398 {
399         if (!req->rq_cong)
400                 return;
401         req->rq_cong = 0;
402         xprt->cong -= RPC_CWNDSCALE;
403         xprt_test_and_clear_congestion_window_wait(xprt);
404         __xprt_lock_write_next_cong(xprt);
405 }
406
407 /**
408  * xprt_request_get_cong - Request congestion control credits
409  * @xprt: pointer to transport
410  * @req: pointer to RPC request
411  *
412  * Useful for transports that require congestion control.
413  */
414 bool
415 xprt_request_get_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
416 {
417         bool ret = false;
418
419         if (req->rq_cong)
420                 return true;
421         spin_lock_bh(&xprt->transport_lock);
422         ret = __xprt_get_cong(xprt, req) != 0;
423         spin_unlock_bh(&xprt->transport_lock);
424         return ret;
425 }
426 EXPORT_SYMBOL_GPL(xprt_request_get_cong);
427
428 /**
429  * xprt_release_rqst_cong - housekeeping when request is complete
430  * @task: RPC request that recently completed
431  *
432  * Useful for transports that require congestion control.
433  */
434 void xprt_release_rqst_cong(struct rpc_task *task)
435 {
436         struct rpc_rqst *req = task->tk_rqstp;
437
438         __xprt_put_cong(req->rq_xprt, req);
439 }
440 EXPORT_SYMBOL_GPL(xprt_release_rqst_cong);
441
442 /*
443  * Clear the congestion window wait flag and wake up the next
444  * entry on xprt->sending
445  */
446 static void
447 xprt_clear_congestion_window_wait(struct rpc_xprt *xprt)
448 {
449         if (test_and_clear_bit(XPRT_CWND_WAIT, &xprt->state)) {
450                 spin_lock_bh(&xprt->transport_lock);
451                 __xprt_lock_write_next_cong(xprt);
452                 spin_unlock_bh(&xprt->transport_lock);
453         }
454 }
455
456 /**
457  * xprt_adjust_cwnd - adjust transport congestion window
458  * @xprt: pointer to xprt
459  * @task: recently completed RPC request used to adjust window
460  * @result: result code of completed RPC request
461  *
462  * The transport code maintains an estimate on the maximum number of out-
463  * standing RPC requests, using a smoothed version of the congestion
464  * avoidance implemented in 44BSD. This is basically the Van Jacobson
465  * congestion algorithm: If a retransmit occurs, the congestion window is
466  * halved; otherwise, it is incremented by 1/cwnd when
467  *
468  *      -       a reply is received and
469  *      -       a full number of requests are outstanding and
470  *      -       the congestion window hasn't been updated recently.
471  */
472 void xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result)
473 {
474         struct rpc_rqst *req = task->tk_rqstp;
475         unsigned long cwnd = xprt->cwnd;
476
477         if (result >= 0 && cwnd <= xprt->cong) {
478                 /* The (cwnd >> 1) term makes sure
479                  * the result gets rounded properly. */
480                 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
481                 if (cwnd > RPC_MAXCWND(xprt))
482                         cwnd = RPC_MAXCWND(xprt);
483                 __xprt_lock_write_next_cong(xprt);
484         } else if (result == -ETIMEDOUT) {
485                 cwnd >>= 1;
486                 if (cwnd < RPC_CWNDSCALE)
487                         cwnd = RPC_CWNDSCALE;
488         }
489         dprintk("RPC:       cong %ld, cwnd was %ld, now %ld\n",
490                         xprt->cong, xprt->cwnd, cwnd);
491         xprt->cwnd = cwnd;
492         __xprt_put_cong(xprt, req);
493 }
494 EXPORT_SYMBOL_GPL(xprt_adjust_cwnd);
495
496 /**
497  * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
498  * @xprt: transport with waiting tasks
499  * @status: result code to plant in each task before waking it
500  *
501  */
502 void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
503 {
504         if (status < 0)
505                 rpc_wake_up_status(&xprt->pending, status);
506         else
507                 rpc_wake_up(&xprt->pending);
508 }
509 EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
510
511 /**
512  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
513  * @xprt: transport
514  *
515  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
516  * we don't in general want to force a socket disconnection due to
517  * an incomplete RPC call transmission.
518  */
519 void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
520 {
521         set_bit(XPRT_WRITE_SPACE, &xprt->state);
522 }
523 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
524
525 static bool
526 xprt_clear_write_space_locked(struct rpc_xprt *xprt)
527 {
528         if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
529                 __xprt_lock_write_next(xprt);
530                 dprintk("RPC:       write space: waking waiting task on "
531                                 "xprt %p\n", xprt);
532                 return true;
533         }
534         return false;
535 }
536
537 /**
538  * xprt_write_space - wake the task waiting for transport output buffer space
539  * @xprt: transport with waiting tasks
540  *
541  * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
542  */
543 bool xprt_write_space(struct rpc_xprt *xprt)
544 {
545         bool ret;
546
547         if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
548                 return false;
549         spin_lock_bh(&xprt->transport_lock);
550         ret = xprt_clear_write_space_locked(xprt);
551         spin_unlock_bh(&xprt->transport_lock);
552         return ret;
553 }
554 EXPORT_SYMBOL_GPL(xprt_write_space);
555
556 /**
557  * xprt_set_retrans_timeout_def - set a request's retransmit timeout
558  * @task: task whose timeout is to be set
559  *
560  * Set a request's retransmit timeout based on the transport's
561  * default timeout parameters.  Used by transports that don't adjust
562  * the retransmit timeout based on round-trip time estimation.
563  */
564 void xprt_set_retrans_timeout_def(struct rpc_task *task)
565 {
566         task->tk_timeout = task->tk_rqstp->rq_timeout;
567 }
568 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_def);
569
570 /**
571  * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
572  * @task: task whose timeout is to be set
573  *
574  * Set a request's retransmit timeout using the RTT estimator.
575  */
576 void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
577 {
578         int timer = task->tk_msg.rpc_proc->p_timer;
579         struct rpc_clnt *clnt = task->tk_client;
580         struct rpc_rtt *rtt = clnt->cl_rtt;
581         struct rpc_rqst *req = task->tk_rqstp;
582         unsigned long max_timeout = clnt->cl_timeout->to_maxval;
583
584         task->tk_timeout = rpc_calc_rto(rtt, timer);
585         task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
586         if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
587                 task->tk_timeout = max_timeout;
588 }
589 EXPORT_SYMBOL_GPL(xprt_set_retrans_timeout_rtt);
590
591 static void xprt_reset_majortimeo(struct rpc_rqst *req)
592 {
593         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
594
595         req->rq_majortimeo = req->rq_timeout;
596         if (to->to_exponential)
597                 req->rq_majortimeo <<= to->to_retries;
598         else
599                 req->rq_majortimeo += to->to_increment * to->to_retries;
600         if (req->rq_majortimeo > to->to_maxval || req->rq_majortimeo == 0)
601                 req->rq_majortimeo = to->to_maxval;
602         req->rq_majortimeo += jiffies;
603 }
604
605 /**
606  * xprt_adjust_timeout - adjust timeout values for next retransmit
607  * @req: RPC request containing parameters to use for the adjustment
608  *
609  */
610 int xprt_adjust_timeout(struct rpc_rqst *req)
611 {
612         struct rpc_xprt *xprt = req->rq_xprt;
613         const struct rpc_timeout *to = req->rq_task->tk_client->cl_timeout;
614         int status = 0;
615
616         if (time_before(jiffies, req->rq_majortimeo)) {
617                 if (to->to_exponential)
618                         req->rq_timeout <<= 1;
619                 else
620                         req->rq_timeout += to->to_increment;
621                 if (to->to_maxval && req->rq_timeout >= to->to_maxval)
622                         req->rq_timeout = to->to_maxval;
623                 req->rq_retries++;
624         } else {
625                 req->rq_timeout = to->to_initval;
626                 req->rq_retries = 0;
627                 xprt_reset_majortimeo(req);
628                 /* Reset the RTT counters == "slow start" */
629                 spin_lock_bh(&xprt->transport_lock);
630                 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
631                 spin_unlock_bh(&xprt->transport_lock);
632                 status = -ETIMEDOUT;
633         }
634
635         if (req->rq_timeout == 0) {
636                 printk(KERN_WARNING "xprt_adjust_timeout: rq_timeout = 0!\n");
637                 req->rq_timeout = 5 * HZ;
638         }
639         return status;
640 }
641
642 static void xprt_autoclose(struct work_struct *work)
643 {
644         struct rpc_xprt *xprt =
645                 container_of(work, struct rpc_xprt, task_cleanup);
646
647         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
648         xprt->ops->close(xprt);
649         xprt_release_write(xprt, NULL);
650         wake_up_bit(&xprt->state, XPRT_LOCKED);
651 }
652
653 /**
654  * xprt_disconnect_done - mark a transport as disconnected
655  * @xprt: transport to flag for disconnect
656  *
657  */
658 void xprt_disconnect_done(struct rpc_xprt *xprt)
659 {
660         dprintk("RPC:       disconnected transport %p\n", xprt);
661         spin_lock_bh(&xprt->transport_lock);
662         xprt_clear_connected(xprt);
663         xprt_clear_write_space_locked(xprt);
664         xprt_wake_pending_tasks(xprt, -EAGAIN);
665         spin_unlock_bh(&xprt->transport_lock);
666 }
667 EXPORT_SYMBOL_GPL(xprt_disconnect_done);
668
669 /**
670  * xprt_force_disconnect - force a transport to disconnect
671  * @xprt: transport to disconnect
672  *
673  */
674 void xprt_force_disconnect(struct rpc_xprt *xprt)
675 {
676         /* Don't race with the test_bit() in xprt_clear_locked() */
677         spin_lock_bh(&xprt->transport_lock);
678         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
679         /* Try to schedule an autoclose RPC call */
680         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
681                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
682         else if (xprt->snd_task)
683                 rpc_wake_up_queued_task_set_status(&xprt->pending,
684                                 xprt->snd_task, -ENOTCONN);
685         spin_unlock_bh(&xprt->transport_lock);
686 }
687 EXPORT_SYMBOL_GPL(xprt_force_disconnect);
688
689 static unsigned int
690 xprt_connect_cookie(struct rpc_xprt *xprt)
691 {
692         return READ_ONCE(xprt->connect_cookie);
693 }
694
695 static bool
696 xprt_request_retransmit_after_disconnect(struct rpc_task *task)
697 {
698         struct rpc_rqst *req = task->tk_rqstp;
699         struct rpc_xprt *xprt = req->rq_xprt;
700
701         return req->rq_connect_cookie != xprt_connect_cookie(xprt) ||
702                 !xprt_connected(xprt);
703 }
704
705 /**
706  * xprt_conditional_disconnect - force a transport to disconnect
707  * @xprt: transport to disconnect
708  * @cookie: 'connection cookie'
709  *
710  * This attempts to break the connection if and only if 'cookie' matches
711  * the current transport 'connection cookie'. It ensures that we don't
712  * try to break the connection more than once when we need to retransmit
713  * a batch of RPC requests.
714  *
715  */
716 void xprt_conditional_disconnect(struct rpc_xprt *xprt, unsigned int cookie)
717 {
718         /* Don't race with the test_bit() in xprt_clear_locked() */
719         spin_lock_bh(&xprt->transport_lock);
720         if (cookie != xprt->connect_cookie)
721                 goto out;
722         if (test_bit(XPRT_CLOSING, &xprt->state))
723                 goto out;
724         set_bit(XPRT_CLOSE_WAIT, &xprt->state);
725         /* Try to schedule an autoclose RPC call */
726         if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
727                 queue_work(xprtiod_workqueue, &xprt->task_cleanup);
728         xprt_wake_pending_tasks(xprt, -EAGAIN);
729 out:
730         spin_unlock_bh(&xprt->transport_lock);
731 }
732
733 static bool
734 xprt_has_timer(const struct rpc_xprt *xprt)
735 {
736         return xprt->idle_timeout != 0;
737 }
738
739 static void
740 xprt_schedule_autodisconnect(struct rpc_xprt *xprt)
741         __must_hold(&xprt->transport_lock)
742 {
743         if (RB_EMPTY_ROOT(&xprt->recv_queue) && xprt_has_timer(xprt))
744                 mod_timer(&xprt->timer, xprt->last_used + xprt->idle_timeout);
745 }
746
747 static void
748 xprt_init_autodisconnect(struct timer_list *t)
749 {
750         struct rpc_xprt *xprt = from_timer(xprt, t, timer);
751
752         spin_lock(&xprt->transport_lock);
753         if (!RB_EMPTY_ROOT(&xprt->recv_queue))
754                 goto out_abort;
755         /* Reset xprt->last_used to avoid connect/autodisconnect cycling */
756         xprt->last_used = jiffies;
757         if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
758                 goto out_abort;
759         spin_unlock(&xprt->transport_lock);
760         queue_work(xprtiod_workqueue, &xprt->task_cleanup);
761         return;
762 out_abort:
763         spin_unlock(&xprt->transport_lock);
764 }
765
766 bool xprt_lock_connect(struct rpc_xprt *xprt,
767                 struct rpc_task *task,
768                 void *cookie)
769 {
770         bool ret = false;
771
772         spin_lock_bh(&xprt->transport_lock);
773         if (!test_bit(XPRT_LOCKED, &xprt->state))
774                 goto out;
775         if (xprt->snd_task != task)
776                 goto out;
777         xprt->snd_task = cookie;
778         ret = true;
779 out:
780         spin_unlock_bh(&xprt->transport_lock);
781         return ret;
782 }
783
784 void xprt_unlock_connect(struct rpc_xprt *xprt, void *cookie)
785 {
786         spin_lock_bh(&xprt->transport_lock);
787         if (xprt->snd_task != cookie)
788                 goto out;
789         if (!test_bit(XPRT_LOCKED, &xprt->state))
790                 goto out;
791         xprt->snd_task =NULL;
792         xprt->ops->release_xprt(xprt, NULL);
793         xprt_schedule_autodisconnect(xprt);
794 out:
795         spin_unlock_bh(&xprt->transport_lock);
796         wake_up_bit(&xprt->state, XPRT_LOCKED);
797 }
798
799 /**
800  * xprt_connect - schedule a transport connect operation
801  * @task: RPC task that is requesting the connect
802  *
803  */
804 void xprt_connect(struct rpc_task *task)
805 {
806         struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
807
808         dprintk("RPC: %5u xprt_connect xprt %p %s connected\n", task->tk_pid,
809                         xprt, (xprt_connected(xprt) ? "is" : "is not"));
810
811         if (!xprt_bound(xprt)) {
812                 task->tk_status = -EAGAIN;
813                 return;
814         }
815         if (!xprt_lock_write(xprt, task))
816                 return;
817
818         if (test_and_clear_bit(XPRT_CLOSE_WAIT, &xprt->state))
819                 xprt->ops->close(xprt);
820
821         if (!xprt_connected(xprt)) {
822                 task->tk_timeout = task->tk_rqstp->rq_timeout;
823                 task->tk_rqstp->rq_connect_cookie = xprt->connect_cookie;
824                 rpc_sleep_on(&xprt->pending, task, NULL);
825
826                 if (test_bit(XPRT_CLOSING, &xprt->state))
827                         return;
828                 if (xprt_test_and_set_connecting(xprt))
829                         return;
830                 /* Race breaker */
831                 if (!xprt_connected(xprt)) {
832                         xprt->stat.connect_start = jiffies;
833                         xprt->ops->connect(xprt, task);
834                 } else {
835                         xprt_clear_connecting(xprt);
836                         task->tk_status = 0;
837                         rpc_wake_up_queued_task(&xprt->pending, task);
838                 }
839         }
840         xprt_release_write(xprt, task);
841 }
842
843 enum xprt_xid_rb_cmp {
844         XID_RB_EQUAL,
845         XID_RB_LEFT,
846         XID_RB_RIGHT,
847 };
848 static enum xprt_xid_rb_cmp
849 xprt_xid_cmp(__be32 xid1, __be32 xid2)
850 {
851         if (xid1 == xid2)
852                 return XID_RB_EQUAL;
853         if ((__force u32)xid1 < (__force u32)xid2)
854                 return XID_RB_LEFT;
855         return XID_RB_RIGHT;
856 }
857
858 static struct rpc_rqst *
859 xprt_request_rb_find(struct rpc_xprt *xprt, __be32 xid)
860 {
861         struct rb_node *n = xprt->recv_queue.rb_node;
862         struct rpc_rqst *req;
863
864         while (n != NULL) {
865                 req = rb_entry(n, struct rpc_rqst, rq_recv);
866                 switch (xprt_xid_cmp(xid, req->rq_xid)) {
867                 case XID_RB_LEFT:
868                         n = n->rb_left;
869                         break;
870                 case XID_RB_RIGHT:
871                         n = n->rb_right;
872                         break;
873                 case XID_RB_EQUAL:
874                         return req;
875                 }
876         }
877         return NULL;
878 }
879
880 static void
881 xprt_request_rb_insert(struct rpc_xprt *xprt, struct rpc_rqst *new)
882 {
883         struct rb_node **p = &xprt->recv_queue.rb_node;
884         struct rb_node *n = NULL;
885         struct rpc_rqst *req;
886
887         while (*p != NULL) {
888                 n = *p;
889                 req = rb_entry(n, struct rpc_rqst, rq_recv);
890                 switch(xprt_xid_cmp(new->rq_xid, req->rq_xid)) {
891                 case XID_RB_LEFT:
892                         p = &n->rb_left;
893                         break;
894                 case XID_RB_RIGHT:
895                         p = &n->rb_right;
896                         break;
897                 case XID_RB_EQUAL:
898                         WARN_ON_ONCE(new != req);
899                         return;
900                 }
901         }
902         rb_link_node(&new->rq_recv, n, p);
903         rb_insert_color(&new->rq_recv, &xprt->recv_queue);
904 }
905
906 static void
907 xprt_request_rb_remove(struct rpc_xprt *xprt, struct rpc_rqst *req)
908 {
909         rb_erase(&req->rq_recv, &xprt->recv_queue);
910 }
911
912 /**
913  * xprt_lookup_rqst - find an RPC request corresponding to an XID
914  * @xprt: transport on which the original request was transmitted
915  * @xid: RPC XID of incoming reply
916  *
917  * Caller holds xprt->queue_lock.
918  */
919 struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid)
920 {
921         struct rpc_rqst *entry;
922
923         entry = xprt_request_rb_find(xprt, xid);
924         if (entry != NULL) {
925                 trace_xprt_lookup_rqst(xprt, xid, 0);
926                 entry->rq_rtt = ktime_sub(ktime_get(), entry->rq_xtime);
927                 return entry;
928         }
929
930         dprintk("RPC:       xprt_lookup_rqst did not find xid %08x\n",
931                         ntohl(xid));
932         trace_xprt_lookup_rqst(xprt, xid, -ENOENT);
933         xprt->stat.bad_xids++;
934         return NULL;
935 }
936 EXPORT_SYMBOL_GPL(xprt_lookup_rqst);
937
938 static bool
939 xprt_is_pinned_rqst(struct rpc_rqst *req)
940 {
941         return atomic_read(&req->rq_pin) != 0;
942 }
943
944 /**
945  * xprt_pin_rqst - Pin a request on the transport receive list
946  * @req: Request to pin
947  *
948  * Caller must ensure this is atomic with the call to xprt_lookup_rqst()
949  * so should be holding the xprt receive lock.
950  */
951 void xprt_pin_rqst(struct rpc_rqst *req)
952 {
953         atomic_inc(&req->rq_pin);
954 }
955 EXPORT_SYMBOL_GPL(xprt_pin_rqst);
956
957 /**
958  * xprt_unpin_rqst - Unpin a request on the transport receive list
959  * @req: Request to pin
960  *
961  * Caller should be holding the xprt receive lock.
962  */
963 void xprt_unpin_rqst(struct rpc_rqst *req)
964 {
965         if (!test_bit(RPC_TASK_MSG_PIN_WAIT, &req->rq_task->tk_runstate)) {
966                 atomic_dec(&req->rq_pin);
967                 return;
968         }
969         if (atomic_dec_and_test(&req->rq_pin))
970                 wake_up_var(&req->rq_pin);
971 }
972 EXPORT_SYMBOL_GPL(xprt_unpin_rqst);
973
974 static void xprt_wait_on_pinned_rqst(struct rpc_rqst *req)
975 {
976         wait_var_event(&req->rq_pin, !xprt_is_pinned_rqst(req));
977 }
978
979 static bool
980 xprt_request_data_received(struct rpc_task *task)
981 {
982         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
983                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) != 0;
984 }
985
986 static bool
987 xprt_request_need_enqueue_receive(struct rpc_task *task, struct rpc_rqst *req)
988 {
989         return !test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) &&
990                 READ_ONCE(task->tk_rqstp->rq_reply_bytes_recvd) == 0;
991 }
992
993 /**
994  * xprt_request_enqueue_receive - Add an request to the receive queue
995  * @task: RPC task
996  *
997  */
998 void
999 xprt_request_enqueue_receive(struct rpc_task *task)
1000 {
1001         struct rpc_rqst *req = task->tk_rqstp;
1002         struct rpc_xprt *xprt = req->rq_xprt;
1003
1004         if (!xprt_request_need_enqueue_receive(task, req))
1005                 return;
1006         spin_lock(&xprt->queue_lock);
1007
1008         /* Update the softirq receive buffer */
1009         memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1010                         sizeof(req->rq_private_buf));
1011
1012         /* Add request to the receive list */
1013         xprt_request_rb_insert(xprt, req);
1014         set_bit(RPC_TASK_NEED_RECV, &task->tk_runstate);
1015         spin_unlock(&xprt->queue_lock);
1016
1017         xprt_reset_majortimeo(req);
1018         /* Turn off autodisconnect */
1019         del_singleshot_timer_sync(&xprt->timer);
1020 }
1021
1022 /**
1023  * xprt_request_dequeue_receive_locked - Remove a request from the receive queue
1024  * @task: RPC task
1025  *
1026  * Caller must hold xprt->queue_lock.
1027  */
1028 static void
1029 xprt_request_dequeue_receive_locked(struct rpc_task *task)
1030 {
1031         struct rpc_rqst *req = task->tk_rqstp;
1032
1033         if (test_and_clear_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1034                 xprt_request_rb_remove(req->rq_xprt, req);
1035 }
1036
1037 /**
1038  * xprt_update_rtt - Update RPC RTT statistics
1039  * @task: RPC request that recently completed
1040  *
1041  * Caller holds xprt->queue_lock.
1042  */
1043 void xprt_update_rtt(struct rpc_task *task)
1044 {
1045         struct rpc_rqst *req = task->tk_rqstp;
1046         struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1047         unsigned int timer = task->tk_msg.rpc_proc->p_timer;
1048         long m = usecs_to_jiffies(ktime_to_us(req->rq_rtt));
1049
1050         if (timer) {
1051                 if (req->rq_ntrans == 1)
1052                         rpc_update_rtt(rtt, timer, m);
1053                 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1054         }
1055 }
1056 EXPORT_SYMBOL_GPL(xprt_update_rtt);
1057
1058 /**
1059  * xprt_complete_rqst - called when reply processing is complete
1060  * @task: RPC request that recently completed
1061  * @copied: actual number of bytes received from the transport
1062  *
1063  * Caller holds xprt->queue_lock.
1064  */
1065 void xprt_complete_rqst(struct rpc_task *task, int copied)
1066 {
1067         struct rpc_rqst *req = task->tk_rqstp;
1068         struct rpc_xprt *xprt = req->rq_xprt;
1069
1070         dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
1071                         task->tk_pid, ntohl(req->rq_xid), copied);
1072         trace_xprt_complete_rqst(xprt, req->rq_xid, copied);
1073
1074         xprt->stat.recvs++;
1075
1076         req->rq_private_buf.len = copied;
1077         /* Ensure all writes are done before we update */
1078         /* req->rq_reply_bytes_recvd */
1079         smp_wmb();
1080         req->rq_reply_bytes_recvd = copied;
1081         xprt_request_dequeue_receive_locked(task);
1082         rpc_wake_up_queued_task(&xprt->pending, task);
1083 }
1084 EXPORT_SYMBOL_GPL(xprt_complete_rqst);
1085
1086 static void xprt_timer(struct rpc_task *task)
1087 {
1088         struct rpc_rqst *req = task->tk_rqstp;
1089         struct rpc_xprt *xprt = req->rq_xprt;
1090
1091         if (task->tk_status != -ETIMEDOUT)
1092                 return;
1093
1094         trace_xprt_timer(xprt, req->rq_xid, task->tk_status);
1095         if (!req->rq_reply_bytes_recvd) {
1096                 if (xprt->ops->timer)
1097                         xprt->ops->timer(xprt, task);
1098         } else
1099                 task->tk_status = 0;
1100 }
1101
1102 /**
1103  * xprt_request_wait_receive - wait for the reply to an RPC request
1104  * @task: RPC task about to send a request
1105  *
1106  */
1107 void xprt_request_wait_receive(struct rpc_task *task)
1108 {
1109         struct rpc_rqst *req = task->tk_rqstp;
1110         struct rpc_xprt *xprt = req->rq_xprt;
1111
1112         if (!test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate))
1113                 return;
1114         /*
1115          * Sleep on the pending queue if we're expecting a reply.
1116          * The spinlock ensures atomicity between the test of
1117          * req->rq_reply_bytes_recvd, and the call to rpc_sleep_on().
1118          */
1119         spin_lock(&xprt->queue_lock);
1120         if (test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate)) {
1121                 xprt->ops->set_retrans_timeout(task);
1122                 rpc_sleep_on(&xprt->pending, task, xprt_timer);
1123                 /*
1124                  * Send an extra queue wakeup call if the
1125                  * connection was dropped in case the call to
1126                  * rpc_sleep_on() raced.
1127                  */
1128                 if (xprt_request_retransmit_after_disconnect(task))
1129                         rpc_wake_up_queued_task_set_status(&xprt->pending,
1130                                         task, -ENOTCONN);
1131         }
1132         spin_unlock(&xprt->queue_lock);
1133 }
1134
1135 static bool
1136 xprt_request_need_enqueue_transmit(struct rpc_task *task, struct rpc_rqst *req)
1137 {
1138         return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1139 }
1140
1141 /**
1142  * xprt_request_enqueue_transmit - queue a task for transmission
1143  * @task: pointer to rpc_task
1144  *
1145  * Add a task to the transmission queue.
1146  */
1147 void
1148 xprt_request_enqueue_transmit(struct rpc_task *task)
1149 {
1150         struct rpc_rqst *pos, *req = task->tk_rqstp;
1151         struct rpc_xprt *xprt = req->rq_xprt;
1152
1153         if (xprt_request_need_enqueue_transmit(task, req)) {
1154                 spin_lock(&xprt->queue_lock);
1155                 /*
1156                  * Requests that carry congestion control credits are added
1157                  * to the head of the list to avoid starvation issues.
1158                  */
1159                 if (req->rq_cong) {
1160                         xprt_clear_congestion_window_wait(xprt);
1161                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1162                                 if (pos->rq_cong)
1163                                         continue;
1164                                 /* Note: req is added _before_ pos */
1165                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1166                                 INIT_LIST_HEAD(&req->rq_xmit2);
1167                                 goto out;
1168                         }
1169                 } else if (RPC_IS_SWAPPER(task)) {
1170                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1171                                 if (pos->rq_cong || pos->rq_bytes_sent)
1172                                         continue;
1173                                 if (RPC_IS_SWAPPER(pos->rq_task))
1174                                         continue;
1175                                 /* Note: req is added _before_ pos */
1176                                 list_add_tail(&req->rq_xmit, &pos->rq_xmit);
1177                                 INIT_LIST_HEAD(&req->rq_xmit2);
1178                                 goto out;
1179                         }
1180                 } else {
1181                         list_for_each_entry(pos, &xprt->xmit_queue, rq_xmit) {
1182                                 if (pos->rq_task->tk_owner != task->tk_owner)
1183                                         continue;
1184                                 list_add_tail(&req->rq_xmit2, &pos->rq_xmit2);
1185                                 INIT_LIST_HEAD(&req->rq_xmit);
1186                                 goto out;
1187                         }
1188                 }
1189                 list_add_tail(&req->rq_xmit, &xprt->xmit_queue);
1190                 INIT_LIST_HEAD(&req->rq_xmit2);
1191 out:
1192                 set_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate);
1193                 spin_unlock(&xprt->queue_lock);
1194         }
1195 }
1196
1197 /**
1198  * xprt_request_dequeue_transmit_locked - remove a task from the transmission queue
1199  * @task: pointer to rpc_task
1200  *
1201  * Remove a task from the transmission queue
1202  * Caller must hold xprt->queue_lock
1203  */
1204 static void
1205 xprt_request_dequeue_transmit_locked(struct rpc_task *task)
1206 {
1207         struct rpc_rqst *req = task->tk_rqstp;
1208
1209         if (!test_and_clear_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1210                 return;
1211         if (!list_empty(&req->rq_xmit)) {
1212                 list_del(&req->rq_xmit);
1213                 if (!list_empty(&req->rq_xmit2)) {
1214                         struct rpc_rqst *next = list_first_entry(&req->rq_xmit2,
1215                                         struct rpc_rqst, rq_xmit2);
1216                         list_del(&req->rq_xmit2);
1217                         list_add_tail(&next->rq_xmit, &next->rq_xprt->xmit_queue);
1218                 }
1219         } else
1220                 list_del(&req->rq_xmit2);
1221 }
1222
1223 /**
1224  * xprt_request_dequeue_transmit - remove a task from the transmission queue
1225  * @task: pointer to rpc_task
1226  *
1227  * Remove a task from the transmission queue
1228  */
1229 static void
1230 xprt_request_dequeue_transmit(struct rpc_task *task)
1231 {
1232         struct rpc_rqst *req = task->tk_rqstp;
1233         struct rpc_xprt *xprt = req->rq_xprt;
1234
1235         spin_lock(&xprt->queue_lock);
1236         xprt_request_dequeue_transmit_locked(task);
1237         spin_unlock(&xprt->queue_lock);
1238 }
1239
1240 /**
1241  * xprt_request_prepare - prepare an encoded request for transport
1242  * @req: pointer to rpc_rqst
1243  *
1244  * Calls into the transport layer to do whatever is needed to prepare
1245  * the request for transmission or receive.
1246  */
1247 void
1248 xprt_request_prepare(struct rpc_rqst *req)
1249 {
1250         struct rpc_xprt *xprt = req->rq_xprt;
1251
1252         if (xprt->ops->prepare_request)
1253                 xprt->ops->prepare_request(req);
1254 }
1255
1256 /**
1257  * xprt_request_need_retransmit - Test if a task needs retransmission
1258  * @task: pointer to rpc_task
1259  *
1260  * Test for whether a connection breakage requires the task to retransmit
1261  */
1262 bool
1263 xprt_request_need_retransmit(struct rpc_task *task)
1264 {
1265         return xprt_request_retransmit_after_disconnect(task);
1266 }
1267
1268 /**
1269  * xprt_prepare_transmit - reserve the transport before sending a request
1270  * @task: RPC task about to send a request
1271  *
1272  */
1273 bool xprt_prepare_transmit(struct rpc_task *task)
1274 {
1275         struct rpc_rqst *req = task->tk_rqstp;
1276         struct rpc_xprt *xprt = req->rq_xprt;
1277
1278         dprintk("RPC: %5u xprt_prepare_transmit\n", task->tk_pid);
1279
1280         if (!xprt_lock_write(xprt, task)) {
1281                 /* Race breaker: someone may have transmitted us */
1282                 if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1283                         rpc_wake_up_queued_task_set_status(&xprt->sending,
1284                                         task, 0);
1285                 return false;
1286
1287         }
1288         return true;
1289 }
1290
1291 void xprt_end_transmit(struct rpc_task *task)
1292 {
1293         xprt_release_write(task->tk_rqstp->rq_xprt, task);
1294 }
1295
1296 /**
1297  * xprt_request_transmit - send an RPC request on a transport
1298  * @req: pointer to request to transmit
1299  * @snd_task: RPC task that owns the transport lock
1300  *
1301  * This performs the transmission of a single request.
1302  * Note that if the request is not the same as snd_task, then it
1303  * does need to be pinned.
1304  * Returns '0' on success.
1305  */
1306 static int
1307 xprt_request_transmit(struct rpc_rqst *req, struct rpc_task *snd_task)
1308 {
1309         struct rpc_xprt *xprt = req->rq_xprt;
1310         struct rpc_task *task = req->rq_task;
1311         unsigned int connect_cookie;
1312         int is_retrans = RPC_WAS_SENT(task);
1313         int status;
1314
1315         dprintk("RPC: %5u xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
1316
1317         if (!req->rq_bytes_sent) {
1318                 if (xprt_request_data_received(task)) {
1319                         status = 0;
1320                         goto out_dequeue;
1321                 }
1322                 /* Verify that our message lies in the RPCSEC_GSS window */
1323                 if (rpcauth_xmit_need_reencode(task)) {
1324                         status = -EBADMSG;
1325                         goto out_dequeue;
1326                 }
1327         }
1328
1329         /*
1330          * Update req->rq_ntrans before transmitting to avoid races with
1331          * xprt_update_rtt(), which needs to know that it is recording a
1332          * reply to the first transmission.
1333          */
1334         req->rq_ntrans++;
1335
1336         connect_cookie = xprt->connect_cookie;
1337         status = xprt->ops->send_request(req);
1338         trace_xprt_transmit(xprt, req->rq_xid, status);
1339         if (status != 0) {
1340                 req->rq_ntrans--;
1341                 return status;
1342         }
1343
1344         if (is_retrans)
1345                 task->tk_client->cl_stats->rpcretrans++;
1346
1347         xprt_inject_disconnect(xprt);
1348
1349         dprintk("RPC: %5u xmit complete\n", task->tk_pid);
1350         task->tk_flags |= RPC_TASK_SENT;
1351         spin_lock_bh(&xprt->transport_lock);
1352
1353         xprt->stat.sends++;
1354         xprt->stat.req_u += xprt->stat.sends - xprt->stat.recvs;
1355         xprt->stat.bklog_u += xprt->backlog.qlen;
1356         xprt->stat.sending_u += xprt->sending.qlen;
1357         xprt->stat.pending_u += xprt->pending.qlen;
1358         spin_unlock_bh(&xprt->transport_lock);
1359
1360         req->rq_connect_cookie = connect_cookie;
1361 out_dequeue:
1362         xprt_request_dequeue_transmit(task);
1363         rpc_wake_up_queued_task_set_status(&xprt->sending, task, status);
1364         return status;
1365 }
1366
1367 /**
1368  * xprt_transmit - send an RPC request on a transport
1369  * @task: controlling RPC task
1370  *
1371  * Attempts to drain the transmit queue. On exit, either the transport
1372  * signalled an error that needs to be handled before transmission can
1373  * resume, or @task finished transmitting, and detected that it already
1374  * received a reply.
1375  */
1376 void
1377 xprt_transmit(struct rpc_task *task)
1378 {
1379         struct rpc_rqst *next, *req = task->tk_rqstp;
1380         struct rpc_xprt *xprt = req->rq_xprt;
1381         int status;
1382
1383         spin_lock(&xprt->queue_lock);
1384         while (!list_empty(&xprt->xmit_queue)) {
1385                 next = list_first_entry(&xprt->xmit_queue,
1386                                 struct rpc_rqst, rq_xmit);
1387                 xprt_pin_rqst(next);
1388                 spin_unlock(&xprt->queue_lock);
1389                 status = xprt_request_transmit(next, task);
1390                 if (status == -EBADMSG && next != req)
1391                         status = 0;
1392                 cond_resched();
1393                 spin_lock(&xprt->queue_lock);
1394                 xprt_unpin_rqst(next);
1395                 if (status == 0) {
1396                         if (!xprt_request_data_received(task) ||
1397                             test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1398                                 continue;
1399                 } else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
1400                         task->tk_status = status;
1401                 break;
1402         }
1403         spin_unlock(&xprt->queue_lock);
1404 }
1405
1406 static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
1407 {
1408         set_bit(XPRT_CONGESTED, &xprt->state);
1409         rpc_sleep_on(&xprt->backlog, task, NULL);
1410 }
1411
1412 static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
1413 {
1414         if (rpc_wake_up_next(&xprt->backlog) == NULL)
1415                 clear_bit(XPRT_CONGESTED, &xprt->state);
1416 }
1417
1418 static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
1419 {
1420         bool ret = false;
1421
1422         if (!test_bit(XPRT_CONGESTED, &xprt->state))
1423                 goto out;
1424         spin_lock(&xprt->reserve_lock);
1425         if (test_bit(XPRT_CONGESTED, &xprt->state)) {
1426                 rpc_sleep_on(&xprt->backlog, task, NULL);
1427                 ret = true;
1428         }
1429         spin_unlock(&xprt->reserve_lock);
1430 out:
1431         return ret;
1432 }
1433
1434 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt)
1435 {
1436         struct rpc_rqst *req = ERR_PTR(-EAGAIN);
1437
1438         if (xprt->num_reqs >= xprt->max_reqs)
1439                 goto out;
1440         ++xprt->num_reqs;
1441         spin_unlock(&xprt->reserve_lock);
1442         req = kzalloc(sizeof(struct rpc_rqst), GFP_NOFS);
1443         spin_lock(&xprt->reserve_lock);
1444         if (req != NULL)
1445                 goto out;
1446         --xprt->num_reqs;
1447         req = ERR_PTR(-ENOMEM);
1448 out:
1449         return req;
1450 }
1451
1452 static bool xprt_dynamic_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1453 {
1454         if (xprt->num_reqs > xprt->min_reqs) {
1455                 --xprt->num_reqs;
1456                 kfree(req);
1457                 return true;
1458         }
1459         return false;
1460 }
1461
1462 void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
1463 {
1464         struct rpc_rqst *req;
1465
1466         spin_lock(&xprt->reserve_lock);
1467         if (!list_empty(&xprt->free)) {
1468                 req = list_entry(xprt->free.next, struct rpc_rqst, rq_list);
1469                 list_del(&req->rq_list);
1470                 goto out_init_req;
1471         }
1472         req = xprt_dynamic_alloc_slot(xprt);
1473         if (!IS_ERR(req))
1474                 goto out_init_req;
1475         switch (PTR_ERR(req)) {
1476         case -ENOMEM:
1477                 dprintk("RPC:       dynamic allocation of request slot "
1478                                 "failed! Retrying\n");
1479                 task->tk_status = -ENOMEM;
1480                 break;
1481         case -EAGAIN:
1482                 xprt_add_backlog(xprt, task);
1483                 dprintk("RPC:       waiting for request slot\n");
1484                 /* fall through */
1485         default:
1486                 task->tk_status = -EAGAIN;
1487         }
1488         spin_unlock(&xprt->reserve_lock);
1489         return;
1490 out_init_req:
1491         xprt->stat.max_slots = max_t(unsigned int, xprt->stat.max_slots,
1492                                      xprt->num_reqs);
1493         spin_unlock(&xprt->reserve_lock);
1494
1495         task->tk_status = 0;
1496         task->tk_rqstp = req;
1497 }
1498 EXPORT_SYMBOL_GPL(xprt_alloc_slot);
1499
1500 void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
1501 {
1502         spin_lock(&xprt->reserve_lock);
1503         if (!xprt_dynamic_free_slot(xprt, req)) {
1504                 memset(req, 0, sizeof(*req));   /* mark unused */
1505                 list_add(&req->rq_list, &xprt->free);
1506         }
1507         xprt_wake_up_backlog(xprt);
1508         spin_unlock(&xprt->reserve_lock);
1509 }
1510 EXPORT_SYMBOL_GPL(xprt_free_slot);
1511
1512 static void xprt_free_all_slots(struct rpc_xprt *xprt)
1513 {
1514         struct rpc_rqst *req;
1515         while (!list_empty(&xprt->free)) {
1516                 req = list_first_entry(&xprt->free, struct rpc_rqst, rq_list);
1517                 list_del(&req->rq_list);
1518                 kfree(req);
1519         }
1520 }
1521
1522 struct rpc_xprt *xprt_alloc(struct net *net, size_t size,
1523                 unsigned int num_prealloc,
1524                 unsigned int max_alloc)
1525 {
1526         struct rpc_xprt *xprt;
1527         struct rpc_rqst *req;
1528         int i;
1529
1530         xprt = kzalloc(size, GFP_KERNEL);
1531         if (xprt == NULL)
1532                 goto out;
1533
1534         xprt_init(xprt, net);
1535
1536         for (i = 0; i < num_prealloc; i++) {
1537                 req = kzalloc(sizeof(struct rpc_rqst), GFP_KERNEL);
1538                 if (!req)
1539                         goto out_free;
1540                 list_add(&req->rq_list, &xprt->free);
1541         }
1542         if (max_alloc > num_prealloc)
1543                 xprt->max_reqs = max_alloc;
1544         else
1545                 xprt->max_reqs = num_prealloc;
1546         xprt->min_reqs = num_prealloc;
1547         xprt->num_reqs = num_prealloc;
1548
1549         return xprt;
1550
1551 out_free:
1552         xprt_free(xprt);
1553 out:
1554         return NULL;
1555 }
1556 EXPORT_SYMBOL_GPL(xprt_alloc);
1557
1558 void xprt_free(struct rpc_xprt *xprt)
1559 {
1560         put_net(xprt->xprt_net);
1561         xprt_free_all_slots(xprt);
1562         kfree_rcu(xprt, rcu);
1563 }
1564 EXPORT_SYMBOL_GPL(xprt_free);
1565
1566 static void
1567 xprt_init_connect_cookie(struct rpc_rqst *req, struct rpc_xprt *xprt)
1568 {
1569         req->rq_connect_cookie = xprt_connect_cookie(xprt) - 1;
1570 }
1571
1572 static __be32
1573 xprt_alloc_xid(struct rpc_xprt *xprt)
1574 {
1575         __be32 xid;
1576
1577         spin_lock(&xprt->reserve_lock);
1578         xid = (__force __be32)xprt->xid++;
1579         spin_unlock(&xprt->reserve_lock);
1580         return xid;
1581 }
1582
1583 static void
1584 xprt_init_xid(struct rpc_xprt *xprt)
1585 {
1586         xprt->xid = prandom_u32();
1587 }
1588
1589 static void
1590 xprt_request_init(struct rpc_task *task)
1591 {
1592         struct rpc_xprt *xprt = task->tk_xprt;
1593         struct rpc_rqst *req = task->tk_rqstp;
1594
1595         req->rq_timeout = task->tk_client->cl_timeout->to_initval;
1596         req->rq_task    = task;
1597         req->rq_xprt    = xprt;
1598         req->rq_buffer  = NULL;
1599         req->rq_xid     = xprt_alloc_xid(xprt);
1600         xprt_init_connect_cookie(req, xprt);
1601         req->rq_bytes_sent = 0;
1602         req->rq_snd_buf.len = 0;
1603         req->rq_snd_buf.buflen = 0;
1604         req->rq_rcv_buf.len = 0;
1605         req->rq_rcv_buf.buflen = 0;
1606         req->rq_snd_buf.bvec = NULL;
1607         req->rq_rcv_buf.bvec = NULL;
1608         req->rq_release_snd_buf = NULL;
1609         xprt_reset_majortimeo(req);
1610         dprintk("RPC: %5u reserved req %p xid %08x\n", task->tk_pid,
1611                         req, ntohl(req->rq_xid));
1612 }
1613
1614 static void
1615 xprt_do_reserve(struct rpc_xprt *xprt, struct rpc_task *task)
1616 {
1617         xprt->ops->alloc_slot(xprt, task);
1618         if (task->tk_rqstp != NULL)
1619                 xprt_request_init(task);
1620 }
1621
1622 /**
1623  * xprt_reserve - allocate an RPC request slot
1624  * @task: RPC task requesting a slot allocation
1625  *
1626  * If the transport is marked as being congested, or if no more
1627  * slots are available, place the task on the transport's
1628  * backlog queue.
1629  */
1630 void xprt_reserve(struct rpc_task *task)
1631 {
1632         struct rpc_xprt *xprt = task->tk_xprt;
1633
1634         task->tk_status = 0;
1635         if (task->tk_rqstp != NULL)
1636                 return;
1637
1638         task->tk_timeout = 0;
1639         task->tk_status = -EAGAIN;
1640         if (!xprt_throttle_congested(xprt, task))
1641                 xprt_do_reserve(xprt, task);
1642 }
1643
1644 /**
1645  * xprt_retry_reserve - allocate an RPC request slot
1646  * @task: RPC task requesting a slot allocation
1647  *
1648  * If no more slots are available, place the task on the transport's
1649  * backlog queue.
1650  * Note that the only difference with xprt_reserve is that we now
1651  * ignore the value of the XPRT_CONGESTED flag.
1652  */
1653 void xprt_retry_reserve(struct rpc_task *task)
1654 {
1655         struct rpc_xprt *xprt = task->tk_xprt;
1656
1657         task->tk_status = 0;
1658         if (task->tk_rqstp != NULL)
1659                 return;
1660
1661         task->tk_timeout = 0;
1662         task->tk_status = -EAGAIN;
1663         xprt_do_reserve(xprt, task);
1664 }
1665
1666 static void
1667 xprt_request_dequeue_all(struct rpc_task *task, struct rpc_rqst *req)
1668 {
1669         struct rpc_xprt *xprt = req->rq_xprt;
1670
1671         if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) ||
1672             test_bit(RPC_TASK_NEED_RECV, &task->tk_runstate) ||
1673             xprt_is_pinned_rqst(req)) {
1674                 spin_lock(&xprt->queue_lock);
1675                 xprt_request_dequeue_transmit_locked(task);
1676                 xprt_request_dequeue_receive_locked(task);
1677                 while (xprt_is_pinned_rqst(req)) {
1678                         set_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1679                         spin_unlock(&xprt->queue_lock);
1680                         xprt_wait_on_pinned_rqst(req);
1681                         spin_lock(&xprt->queue_lock);
1682                         clear_bit(RPC_TASK_MSG_PIN_WAIT, &task->tk_runstate);
1683                 }
1684                 spin_unlock(&xprt->queue_lock);
1685         }
1686 }
1687
1688 /**
1689  * xprt_release - release an RPC request slot
1690  * @task: task which is finished with the slot
1691  *
1692  */
1693 void xprt_release(struct rpc_task *task)
1694 {
1695         struct rpc_xprt *xprt;
1696         struct rpc_rqst *req = task->tk_rqstp;
1697
1698         if (req == NULL) {
1699                 if (task->tk_client) {
1700                         xprt = task->tk_xprt;
1701                         xprt_release_write(xprt, task);
1702                 }
1703                 return;
1704         }
1705
1706         xprt = req->rq_xprt;
1707         if (task->tk_ops->rpc_count_stats != NULL)
1708                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
1709         else if (task->tk_client)
1710                 rpc_count_iostats(task, task->tk_client->cl_metrics);
1711         xprt_request_dequeue_all(task, req);
1712         spin_lock_bh(&xprt->transport_lock);
1713         xprt->ops->release_xprt(xprt, task);
1714         if (xprt->ops->release_request)
1715                 xprt->ops->release_request(task);
1716         xprt->last_used = jiffies;
1717         xprt_schedule_autodisconnect(xprt);
1718         spin_unlock_bh(&xprt->transport_lock);
1719         if (req->rq_buffer)
1720                 xprt->ops->buf_free(task);
1721         xprt_inject_disconnect(xprt);
1722         xdr_free_bvec(&req->rq_rcv_buf);
1723         if (req->rq_cred != NULL)
1724                 put_rpccred(req->rq_cred);
1725         task->tk_rqstp = NULL;
1726         if (req->rq_release_snd_buf)
1727                 req->rq_release_snd_buf(req);
1728
1729         dprintk("RPC: %5u release request %p\n", task->tk_pid, req);
1730         if (likely(!bc_prealloc(req)))
1731                 xprt->ops->free_slot(xprt, req);
1732         else
1733                 xprt_free_bc_request(req);
1734 }
1735
1736 #ifdef CONFIG_SUNRPC_BACKCHANNEL
1737 void
1738 xprt_init_bc_request(struct rpc_rqst *req, struct rpc_task *task)
1739 {
1740         struct xdr_buf *xbufp = &req->rq_snd_buf;
1741
1742         task->tk_rqstp = req;
1743         req->rq_task = task;
1744         xprt_init_connect_cookie(req, req->rq_xprt);
1745         /*
1746          * Set up the xdr_buf length.
1747          * This also indicates that the buffer is XDR encoded already.
1748          */
1749         xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1750                 xbufp->tail[0].iov_len;
1751         req->rq_bytes_sent = 0;
1752 }
1753 #endif
1754
1755 static void xprt_init(struct rpc_xprt *xprt, struct net *net)
1756 {
1757         kref_init(&xprt->kref);
1758
1759         spin_lock_init(&xprt->transport_lock);
1760         spin_lock_init(&xprt->reserve_lock);
1761         spin_lock_init(&xprt->queue_lock);
1762
1763         INIT_LIST_HEAD(&xprt->free);
1764         xprt->recv_queue = RB_ROOT;
1765         INIT_LIST_HEAD(&xprt->xmit_queue);
1766 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1767         spin_lock_init(&xprt->bc_pa_lock);
1768         INIT_LIST_HEAD(&xprt->bc_pa_list);
1769 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1770         INIT_LIST_HEAD(&xprt->xprt_switch);
1771
1772         xprt->last_used = jiffies;
1773         xprt->cwnd = RPC_INITCWND;
1774         xprt->bind_index = 0;
1775
1776         rpc_init_wait_queue(&xprt->binding, "xprt_binding");
1777         rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1778         rpc_init_wait_queue(&xprt->sending, "xprt_sending");
1779         rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1780
1781         xprt_init_xid(xprt);
1782
1783         xprt->xprt_net = get_net(net);
1784 }
1785
1786 /**
1787  * xprt_create_transport - create an RPC transport
1788  * @args: rpc transport creation arguments
1789  *
1790  */
1791 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
1792 {
1793         struct rpc_xprt *xprt;
1794         struct xprt_class *t;
1795
1796         spin_lock(&xprt_list_lock);
1797         list_for_each_entry(t, &xprt_list, list) {
1798                 if (t->ident == args->ident) {
1799                         spin_unlock(&xprt_list_lock);
1800                         goto found;
1801                 }
1802         }
1803         spin_unlock(&xprt_list_lock);
1804         dprintk("RPC: transport (%d) not supported\n", args->ident);
1805         return ERR_PTR(-EIO);
1806
1807 found:
1808         xprt = t->setup(args);
1809         if (IS_ERR(xprt)) {
1810                 dprintk("RPC:       xprt_create_transport: failed, %ld\n",
1811                                 -PTR_ERR(xprt));
1812                 goto out;
1813         }
1814         if (args->flags & XPRT_CREATE_NO_IDLE_TIMEOUT)
1815                 xprt->idle_timeout = 0;
1816         INIT_WORK(&xprt->task_cleanup, xprt_autoclose);
1817         if (xprt_has_timer(xprt))
1818                 timer_setup(&xprt->timer, xprt_init_autodisconnect, 0);
1819         else
1820                 timer_setup(&xprt->timer, NULL, 0);
1821
1822         if (strlen(args->servername) > RPC_MAXNETNAMELEN) {
1823                 xprt_destroy(xprt);
1824                 return ERR_PTR(-EINVAL);
1825         }
1826         xprt->servername = kstrdup(args->servername, GFP_KERNEL);
1827         if (xprt->servername == NULL) {
1828                 xprt_destroy(xprt);
1829                 return ERR_PTR(-ENOMEM);
1830         }
1831
1832         rpc_xprt_debugfs_register(xprt);
1833
1834         dprintk("RPC:       created transport %p with %u slots\n", xprt,
1835                         xprt->max_reqs);
1836 out:
1837         return xprt;
1838 }
1839
1840 static void xprt_destroy_cb(struct work_struct *work)
1841 {
1842         struct rpc_xprt *xprt =
1843                 container_of(work, struct rpc_xprt, task_cleanup);
1844
1845         rpc_xprt_debugfs_unregister(xprt);
1846         rpc_destroy_wait_queue(&xprt->binding);
1847         rpc_destroy_wait_queue(&xprt->pending);
1848         rpc_destroy_wait_queue(&xprt->sending);
1849         rpc_destroy_wait_queue(&xprt->backlog);
1850         kfree(xprt->servername);
1851         /*
1852          * Tear down transport state and free the rpc_xprt
1853          */
1854         xprt->ops->destroy(xprt);
1855 }
1856
1857 /**
1858  * xprt_destroy - destroy an RPC transport, killing off all requests.
1859  * @xprt: transport to destroy
1860  *
1861  */
1862 static void xprt_destroy(struct rpc_xprt *xprt)
1863 {
1864         dprintk("RPC:       destroying transport %p\n", xprt);
1865
1866         /*
1867          * Exclude transport connect/disconnect handlers and autoclose
1868          */
1869         wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_UNINTERRUPTIBLE);
1870
1871         del_timer_sync(&xprt->timer);
1872
1873         /*
1874          * Destroy sockets etc from the system workqueue so they can
1875          * safely flush receive work running on rpciod.
1876          */
1877         INIT_WORK(&xprt->task_cleanup, xprt_destroy_cb);
1878         schedule_work(&xprt->task_cleanup);
1879 }
1880
1881 static void xprt_destroy_kref(struct kref *kref)
1882 {
1883         xprt_destroy(container_of(kref, struct rpc_xprt, kref));
1884 }
1885
1886 /**
1887  * xprt_get - return a reference to an RPC transport.
1888  * @xprt: pointer to the transport
1889  *
1890  */
1891 struct rpc_xprt *xprt_get(struct rpc_xprt *xprt)
1892 {
1893         if (xprt != NULL && kref_get_unless_zero(&xprt->kref))
1894                 return xprt;
1895         return NULL;
1896 }
1897 EXPORT_SYMBOL_GPL(xprt_get);
1898
1899 /**
1900  * xprt_put - release a reference to an RPC transport.
1901  * @xprt: pointer to the transport
1902  *
1903  */
1904 void xprt_put(struct rpc_xprt *xprt)
1905 {
1906         if (xprt != NULL)
1907                 kref_put(&xprt->kref, xprt_destroy_kref);
1908 }
1909 EXPORT_SYMBOL_GPL(xprt_put);