blob: f14f24ee998344f5c356ef49e3b3fc52d581a39c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/net/sunrpc/sched.c
3 *
4 * Scheduling for synchronous and asynchronous RPC requests.
5 *
6 * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
YOSHIFUJI Hideakicca51722007-02-09 15:38:13 -08007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 */
11
12#include <linux/module.h>
13
14#include <linux/sched.h>
15#include <linux/interrupt.h>
16#include <linux/slab.h>
17#include <linux/mempool.h>
18#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/spinlock.h>
Arjan van de Ven4a3e2f72006-03-20 22:33:17 -080020#include <linux/mutex.h>
Jeff Laytond3103102011-12-01 22:44:39 +010021#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
23#include <linux/sunrpc/clnt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Benny Halevy69518672009-09-10 12:25:04 +030025#include "sunrpc.h"
26
Jeff Laytonf895b252014-11-17 16:58:04 -050027#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#define RPCDBG_FACILITY RPCDBG_SCHED
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#endif
30
Trond Myklebust82b0a4c2012-01-20 14:52:23 -050031#define CREATE_TRACE_POINTS
32#include <trace/events/sunrpc.h>
33
Linus Torvalds1da177e2005-04-16 15:20:36 -070034/*
35 * RPC slabs and memory pools
36 */
37#define RPC_BUFFER_MAXSIZE (2048)
38#define RPC_BUFFER_POOLSIZE (8)
39#define RPC_TASK_POOLSIZE (8)
Christoph Lametere18b8902006-12-06 20:33:20 -080040static struct kmem_cache *rpc_task_slabp __read_mostly;
41static struct kmem_cache *rpc_buffer_slabp __read_mostly;
Eric Dumazetba899662005-08-26 12:05:31 -070042static mempool_t *rpc_task_mempool __read_mostly;
43static mempool_t *rpc_buffer_mempool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
David Howells65f27f32006-11-22 14:55:48 +000045static void rpc_async_schedule(struct work_struct *);
Trond Myklebustbde8f002007-01-24 11:54:53 -080046static void rpc_release_task(struct rpc_task *task);
Trond Myklebust36df9aa2007-07-18 16:18:52 -040047static void __rpc_queue_timer_fn(unsigned long ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 * RPC tasks sit here while waiting for conditions to improve.
51 */
Trond Myklebusta4a87492007-07-18 13:24:19 -040052static struct rpc_wait_queue delay_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 * rpciod-related stuff
56 */
Trond Myklebust24c5d9d2006-03-20 13:44:08 -050057struct workqueue_struct *rpciod_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * Disable the timer for a given RPC task. Should be called with
61 * queue->lock and bh_disabled in order to avoid races within
62 * rpc_run_timer().
63 */
Trond Myklebust5d008372008-02-22 16:34:17 -050064static void
Trond Myklebusteb276c02008-02-22 17:27:59 -050065__rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066{
Trond Myklebust36df9aa2007-07-18 16:18:52 -040067 if (task->tk_timeout == 0)
68 return;
Chuck Lever46121cf2007-01-31 12:14:08 -050069 dprintk("RPC: %5u disabling timer\n", task->tk_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 task->tk_timeout = 0;
Trond Myklebust36df9aa2007-07-18 16:18:52 -040071 list_del(&task->u.tk_wait.timer_list);
Trond Myklebusteb276c02008-02-22 17:27:59 -050072 if (list_empty(&queue->timer_list.list))
73 del_timer(&queue->timer_list.timer);
Trond Myklebust36df9aa2007-07-18 16:18:52 -040074}
75
76static void
77rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
78{
79 queue->timer_list.expires = expires;
80 mod_timer(&queue->timer_list.timer, expires);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081}
82
83/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * Set up a timer for the current task.
85 */
Trond Myklebust5d008372008-02-22 16:34:17 -050086static void
Trond Myklebusteb276c02008-02-22 17:27:59 -050087__rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
89 if (!task->tk_timeout)
90 return;
91
Nicholas Mc Guire55cc1d72015-03-13 04:25:34 -040092 dprintk("RPC: %5u setting alarm for %u ms\n",
93 task->tk_pid, jiffies_to_msecs(task->tk_timeout));
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Trond Myklebusteb276c02008-02-22 17:27:59 -050095 task->u.tk_wait.expires = jiffies + task->tk_timeout;
96 if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
97 rpc_set_queue_timer(queue, task->u.tk_wait.expires);
98 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099}
100
Trond Myklebustedd2e362013-01-27 14:20:49 -0500101static void rpc_rotate_queue_owner(struct rpc_wait_queue *queue)
102{
103 struct list_head *q = &queue->tasks[queue->priority];
104 struct rpc_task *task;
105
106 if (!list_empty(q)) {
107 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
108 if (task->tk_owner == queue->owner)
109 list_move_tail(&task->u.tk_wait.list, q);
110 }
111}
112
Trond Myklebustc05eecf2012-11-30 23:59:29 -0500113static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
114{
Trond Myklebustedd2e362013-01-27 14:20:49 -0500115 if (queue->priority != priority) {
116 /* Fairness: rotate the list when changing priority */
117 rpc_rotate_queue_owner(queue);
118 queue->priority = priority;
119 }
Trond Myklebustc05eecf2012-11-30 23:59:29 -0500120}
121
122static void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
123{
124 queue->owner = pid;
125 queue->nr = RPC_BATCH_COUNT;
126}
127
128static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
129{
130 rpc_set_waitqueue_priority(queue, queue->maxpriority);
131 rpc_set_waitqueue_owner(queue, 0);
132}
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134/*
135 * Add new request to a priority queue.
136 */
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400137static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
138 struct rpc_task *task,
139 unsigned char queue_priority)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
141 struct list_head *q;
142 struct rpc_task *t;
143
144 INIT_LIST_HEAD(&task->u.tk_wait.links);
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400145 if (unlikely(queue_priority > queue->maxpriority))
Trond Myklebustc05eecf2012-11-30 23:59:29 -0500146 queue_priority = queue->maxpriority;
147 if (queue_priority > queue->priority)
148 rpc_set_waitqueue_priority(queue, queue_priority);
149 q = &queue->tasks[queue_priority];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 list_for_each_entry(t, q, u.tk_wait.list) {
Trond Myklebust3ff75762007-07-14 15:40:00 -0400151 if (t->tk_owner == task->tk_owner) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
153 return;
154 }
155 }
156 list_add_tail(&task->u.tk_wait.list, q);
157}
158
159/*
160 * Add new request to wait queue.
161 *
162 * Swapper tasks always get inserted at the head of the queue.
163 * This should avoid many nasty memory deadlocks and hopefully
164 * improve overall performance.
165 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
166 */
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400167static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
168 struct rpc_task *task,
169 unsigned char queue_priority)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Weston Andros Adamson2bd4eef2012-10-23 10:43:47 -0400171 WARN_ON_ONCE(RPC_IS_QUEUED(task));
172 if (RPC_IS_QUEUED(task))
173 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175 if (RPC_IS_PRIORITY(queue))
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400176 __rpc_add_wait_queue_priority(queue, task, queue_priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 else if (RPC_IS_SWAPPER(task))
178 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
179 else
180 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500181 task->tk_waitqueue = queue;
Chuck Levere19b63d2006-03-20 13:44:15 -0500182 queue->qlen++;
Trond Myklebust1166fde62013-03-25 11:23:40 -0400183 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
184 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 rpc_set_queued(task);
186
Chuck Lever46121cf2007-01-31 12:14:08 -0500187 dprintk("RPC: %5u added to queue %p \"%s\"\n",
188 task->tk_pid, queue, rpc_qname(queue));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
190
191/*
192 * Remove request from a priority queue.
193 */
194static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
195{
196 struct rpc_task *t;
197
198 if (!list_empty(&task->u.tk_wait.links)) {
199 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
200 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
201 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
205/*
206 * Remove request from queue.
207 * Note: must be called with spin lock held.
208 */
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500209static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210{
Trond Myklebusteb276c02008-02-22 17:27:59 -0500211 __rpc_disable_timer(queue, task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 if (RPC_IS_PRIORITY(queue))
213 __rpc_remove_wait_queue_priority(task);
Trond Myklebust36df9aa2007-07-18 16:18:52 -0400214 list_del(&task->u.tk_wait.list);
Chuck Levere19b63d2006-03-20 13:44:15 -0500215 queue->qlen--;
Chuck Lever46121cf2007-01-31 12:14:08 -0500216 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
217 task->tk_pid, queue, rpc_qname(queue));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
219
Trond Myklebust3ff75762007-07-14 15:40:00 -0400220static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
222 int i;
223
224 spin_lock_init(&queue->lock);
225 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
226 INIT_LIST_HEAD(&queue->tasks[i]);
Trond Myklebust3ff75762007-07-14 15:40:00 -0400227 queue->maxpriority = nr_queues - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 rpc_reset_waitqueue_priority(queue);
Trond Myklebust36df9aa2007-07-18 16:18:52 -0400229 queue->qlen = 0;
230 setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
231 INIT_LIST_HEAD(&queue->timer_list.list);
Trond Myklebust2f09c242012-02-08 22:01:15 -0500232 rpc_assign_waitqueue_name(queue, qname);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
235void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
236{
Trond Myklebust3ff75762007-07-14 15:40:00 -0400237 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
Alexandros Batsakis689cf5c2009-12-14 21:27:56 -0800239EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
241void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
242{
Trond Myklebust3ff75762007-07-14 15:40:00 -0400243 __rpc_init_priority_wait_queue(queue, qname, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
Trond Myklebuste8914c62007-07-14 15:39:59 -0400245EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Trond Myklebustf6a1cc82008-02-22 17:06:55 -0500247void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
248{
Trond Myklebust36df9aa2007-07-18 16:18:52 -0400249 del_timer_sync(&queue->timer_list.timer);
Trond Myklebustf6a1cc82008-02-22 17:06:55 -0500250}
251EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
252
NeilBrownc1221322014-07-07 15:16:04 +1000253static int rpc_wait_bit_killable(struct wait_bit_key *key)
Trond Myklebust44c28872006-01-03 09:55:06 +0100254{
Matthew Wilcox150030b2007-12-06 16:24:39 -0500255 if (fatal_signal_pending(current))
Trond Myklebust44c28872006-01-03 09:55:06 +0100256 return -ERESTARTSYS;
Colin Cross416ad3c2013-05-06 23:50:06 +0000257 freezable_schedule_unsafe();
Trond Myklebust44c28872006-01-03 09:55:06 +0100258 return 0;
259}
260
Jeff Layton13067292014-11-17 16:58:05 -0500261#if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
Trond Myklebustc44fe702007-06-16 14:17:01 -0400262static void rpc_task_set_debuginfo(struct rpc_task *task)
263{
264 static atomic_t rpc_pid;
265
Trond Myklebustc44fe702007-06-16 14:17:01 -0400266 task->tk_pid = atomic_inc_return(&rpc_pid);
267}
268#else
269static inline void rpc_task_set_debuginfo(struct rpc_task *task)
270{
271}
272#endif
273
Trond Myklebuste6b3c4d2006-11-11 22:18:03 -0500274static void rpc_set_active(struct rpc_task *task)
275{
Trond Myklebust82b0a4c2012-01-20 14:52:23 -0500276 trace_rpc_task_begin(task->tk_client, task, NULL);
277
Trond Myklebustc44fe702007-06-16 14:17:01 -0400278 rpc_task_set_debuginfo(task);
Trond Myklebust58f96122010-07-31 14:29:08 -0400279 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
Trond Myklebuste6b3c4d2006-11-11 22:18:03 -0500280}
281
Trond Myklebust44c28872006-01-03 09:55:06 +0100282/*
283 * Mark an RPC call as having completed by clearing the 'active' bit
Trond Myklebustbf294b42011-02-21 11:05:41 -0800284 * and then waking up all tasks that were sleeping.
Trond Myklebust44c28872006-01-03 09:55:06 +0100285 */
Trond Myklebustbf294b42011-02-21 11:05:41 -0800286static int rpc_complete_task(struct rpc_task *task)
Trond Myklebust44c28872006-01-03 09:55:06 +0100287{
Trond Myklebustbf294b42011-02-21 11:05:41 -0800288 void *m = &task->tk_runstate;
289 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
290 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
291 unsigned long flags;
292 int ret;
293
Trond Myklebust82b0a4c2012-01-20 14:52:23 -0500294 trace_rpc_task_complete(task->tk_client, task, NULL);
295
Trond Myklebustbf294b42011-02-21 11:05:41 -0800296 spin_lock_irqsave(&wq->lock, flags);
Trond Myklebuste6b3c4d2006-11-11 22:18:03 -0500297 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
Trond Myklebustbf294b42011-02-21 11:05:41 -0800298 ret = atomic_dec_and_test(&task->tk_count);
299 if (waitqueue_active(wq))
Andrea Arcangeliac5be6b2015-09-22 14:58:49 -0700300 __wake_up_locked_key(wq, TASK_NORMAL, &k);
Trond Myklebustbf294b42011-02-21 11:05:41 -0800301 spin_unlock_irqrestore(&wq->lock, flags);
302 return ret;
Trond Myklebust44c28872006-01-03 09:55:06 +0100303}
304
305/*
306 * Allow callers to wait for completion of an RPC call
Trond Myklebustbf294b42011-02-21 11:05:41 -0800307 *
308 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
309 * to enforce taking of the wq->lock and hence avoid races with
310 * rpc_complete_task().
Trond Myklebust44c28872006-01-03 09:55:06 +0100311 */
NeilBrownc1221322014-07-07 15:16:04 +1000312int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
Trond Myklebust44c28872006-01-03 09:55:06 +0100313{
314 if (action == NULL)
Matthew Wilcox150030b2007-12-06 16:24:39 -0500315 action = rpc_wait_bit_killable;
Trond Myklebustbf294b42011-02-21 11:05:41 -0800316 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
Matthew Wilcox150030b2007-12-06 16:24:39 -0500317 action, TASK_KILLABLE);
Trond Myklebust44c28872006-01-03 09:55:06 +0100318}
Trond Myklebuste8914c62007-07-14 15:39:59 -0400319EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
Trond Myklebust44c28872006-01-03 09:55:06 +0100320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321/*
322 * Make an RPC task runnable.
323 *
Jeff Layton506026c2012-07-23 15:51:55 -0400324 * Note: If the task is ASYNC, and is being made runnable after sitting on an
325 * rpc_wait_queue, this must be called with the queue spinlock held to protect
326 * the wait queue operation.
Trond Myklebusta3c3cac2013-05-22 12:57:24 -0400327 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
328 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
329 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
330 * the RPC_TASK_RUNNING flag.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 */
332static void rpc_make_runnable(struct rpc_task *task)
333{
Trond Myklebusta3c3cac2013-05-22 12:57:24 -0400334 bool need_wakeup = !rpc_test_and_set_running(task);
335
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 rpc_clear_queued(task);
Trond Myklebusta3c3cac2013-05-22 12:57:24 -0400337 if (!need_wakeup)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 return;
339 if (RPC_IS_ASYNC(task)) {
David Howells65f27f32006-11-22 14:55:48 +0000340 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
Fred Isamancee6a532011-02-11 15:42:35 +0000341 queue_work(rpciod_workqueue, &task->u.tk_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 } else
Trond Myklebust96651ab2005-06-22 17:16:21 +0000343 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
346/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 * Prepare for sleeping on a wait queue.
348 * By always appending tasks to the list we ensure FIFO behavior.
349 * NB: An RPC task will only receive interrupt-driven events as long
350 * as it's on a wait queue.
351 */
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400352static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
353 struct rpc_task *task,
354 rpc_action action,
355 unsigned char queue_priority)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356{
Chuck Lever46121cf2007-01-31 12:14:08 -0500357 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
358 task->tk_pid, rpc_qname(q), jiffies);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
Trond Myklebust82b0a4c2012-01-20 14:52:23 -0500360 trace_rpc_task_sleep(task->tk_client, task, q);
361
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400362 __rpc_add_wait_queue(q, task, queue_priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
Weston Andros Adamsonf50ad422012-10-23 10:43:46 -0400364 WARN_ON_ONCE(task->tk_callback != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 task->tk_callback = action;
Trond Myklebusteb276c02008-02-22 17:27:59 -0500366 __rpc_add_timer(q, task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
369void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
Trond Myklebust5d008372008-02-22 16:34:17 -0500370 rpc_action action)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Trond Myklebust58f96122010-07-31 14:29:08 -0400372 /* We shouldn't ever put an inactive task to sleep */
Weston Andros Adamsone454a7a2012-10-23 10:43:32 -0400373 WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
374 if (!RPC_IS_ACTIVATED(task)) {
375 task->tk_status = -EIO;
376 rpc_put_task_async(task);
377 return;
378 }
Trond Myklebuste6b3c4d2006-11-11 22:18:03 -0500379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 /*
381 * Protect the queue operations.
382 */
383 spin_lock_bh(&q->lock);
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400384 __rpc_sleep_on_priority(q, task, action, task->tk_priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 spin_unlock_bh(&q->lock);
386}
Trond Myklebuste8914c62007-07-14 15:39:59 -0400387EXPORT_SYMBOL_GPL(rpc_sleep_on);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400389void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
390 rpc_action action, int priority)
391{
392 /* We shouldn't ever put an inactive task to sleep */
Weston Andros Adamsone454a7a2012-10-23 10:43:32 -0400393 WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
394 if (!RPC_IS_ACTIVATED(task)) {
395 task->tk_status = -EIO;
396 rpc_put_task_async(task);
397 return;
398 }
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400399
400 /*
401 * Protect the queue operations.
402 */
403 spin_lock_bh(&q->lock);
404 __rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
405 spin_unlock_bh(&q->lock);
406}
Trond Myklebust1e1093c2012-11-01 16:44:05 -0400407EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
Trond Myklebust3b27bad2011-07-17 18:11:34 -0400408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409/**
410 * __rpc_do_wake_up_task - wake up a single rpc_task
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500411 * @queue: wait queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 * @task: task to be woken up
413 *
414 * Caller must hold queue->lock, and have cleared the task queued flag.
415 */
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500416static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417{
Chuck Lever46121cf2007-01-31 12:14:08 -0500418 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
419 task->tk_pid, jiffies);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 /* Has the task been executed yet? If not, we cannot wake it up! */
422 if (!RPC_IS_ACTIVATED(task)) {
423 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
424 return;
425 }
426
Trond Myklebust82b0a4c2012-01-20 14:52:23 -0500427 trace_rpc_task_wakeup(task->tk_client, task, queue);
428
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500429 __rpc_remove_wait_queue(queue, task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431 rpc_make_runnable(task);
432
Chuck Lever46121cf2007-01-31 12:14:08 -0500433 dprintk("RPC: __rpc_wake_up_task done\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
436/*
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500437 * Wake up a queued task while the queue lock is being held
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 */
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500439static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
Trond Myklebust1166fde62013-03-25 11:23:40 -0400441 if (RPC_IS_QUEUED(task)) {
442 smp_rmb();
443 if (task->tk_waitqueue == queue)
444 __rpc_do_wake_up_task(queue, task);
445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
447
448/*
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500449 * Wake up a task on a specific queue
450 */
451void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
452{
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800453 spin_lock_bh(&queue->lock);
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500454 rpc_wake_up_task_queue_locked(queue, task);
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800455 spin_unlock_bh(&queue->lock);
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500456}
457EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
458
459/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 * Wake up the next task on a priority queue.
461 */
Trond Myklebust961a8282012-01-17 22:57:37 -0500462static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
464 struct list_head *q;
465 struct rpc_task *task;
466
467 /*
Trond Myklebust3ff75762007-07-14 15:40:00 -0400468 * Service a batch of tasks from a single owner.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 */
470 q = &queue->tasks[queue->priority];
471 if (!list_empty(q)) {
472 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
Trond Myklebust3ff75762007-07-14 15:40:00 -0400473 if (queue->owner == task->tk_owner) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 if (--queue->nr)
475 goto out;
476 list_move_tail(&task->u.tk_wait.list, q);
477 }
478 /*
479 * Check if we need to switch queues.
480 */
Trond Myklebustc05eecf2012-11-30 23:59:29 -0500481 goto new_owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 }
483
484 /*
485 * Service the next queue.
486 */
487 do {
488 if (q == &queue->tasks[0])
489 q = &queue->tasks[queue->maxpriority];
490 else
491 q = q - 1;
492 if (!list_empty(q)) {
493 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
494 goto new_queue;
495 }
496 } while (q != &queue->tasks[queue->priority]);
497
498 rpc_reset_waitqueue_priority(queue);
499 return NULL;
500
501new_queue:
502 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
Trond Myklebust3ff75762007-07-14 15:40:00 -0400503new_owner:
504 rpc_set_waitqueue_owner(queue, task->tk_owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 return task;
507}
508
Trond Myklebust961a8282012-01-17 22:57:37 -0500509static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
510{
511 if (RPC_IS_PRIORITY(queue))
512 return __rpc_find_next_queued_priority(queue);
513 if (!list_empty(&queue->tasks[0]))
514 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
515 return NULL;
516}
517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518/*
Trond Myklebust961a8282012-01-17 22:57:37 -0500519 * Wake up the first task on the wait queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 */
Trond Myklebust961a8282012-01-17 22:57:37 -0500521struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
522 bool (*func)(struct rpc_task *, void *), void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
524 struct rpc_task *task = NULL;
525
Trond Myklebust961a8282012-01-17 22:57:37 -0500526 dprintk("RPC: wake_up_first(%p \"%s\")\n",
Chuck Lever46121cf2007-01-31 12:14:08 -0500527 queue, rpc_qname(queue));
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800528 spin_lock_bh(&queue->lock);
Trond Myklebust961a8282012-01-17 22:57:37 -0500529 task = __rpc_find_next_queued(queue);
530 if (task != NULL) {
531 if (func(task, data))
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500532 rpc_wake_up_task_queue_locked(queue, task);
Trond Myklebust961a8282012-01-17 22:57:37 -0500533 else
534 task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800536 spin_unlock_bh(&queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
538 return task;
539}
Trond Myklebust961a8282012-01-17 22:57:37 -0500540EXPORT_SYMBOL_GPL(rpc_wake_up_first);
541
542static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
543{
544 return true;
545}
546
547/*
548 * Wake up the next task on the wait queue.
549*/
550struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
551{
552 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
553}
Trond Myklebuste8914c62007-07-14 15:39:59 -0400554EXPORT_SYMBOL_GPL(rpc_wake_up_next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
556/**
557 * rpc_wake_up - wake up all rpc_tasks
558 * @queue: rpc_wait_queue on which the tasks are sleeping
559 *
560 * Grabs queue->lock
561 */
562void rpc_wake_up(struct rpc_wait_queue *queue)
563{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 struct list_head *head;
Trond Myklebuste6d83d52006-03-13 21:20:48 -0800565
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800566 spin_lock_bh(&queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 head = &queue->tasks[queue->maxpriority];
568 for (;;) {
Trond Myklebust540a0f72012-03-19 13:39:35 -0400569 while (!list_empty(head)) {
570 struct rpc_task *task;
571 task = list_first_entry(head,
572 struct rpc_task,
573 u.tk_wait.list);
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500574 rpc_wake_up_task_queue_locked(queue, task);
Trond Myklebust540a0f72012-03-19 13:39:35 -0400575 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 if (head == &queue->tasks[0])
577 break;
578 head--;
579 }
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800580 spin_unlock_bh(&queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581}
Trond Myklebuste8914c62007-07-14 15:39:59 -0400582EXPORT_SYMBOL_GPL(rpc_wake_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584/**
585 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
586 * @queue: rpc_wait_queue on which the tasks are sleeping
587 * @status: status value to set
588 *
589 * Grabs queue->lock
590 */
591void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
592{
593 struct list_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800595 spin_lock_bh(&queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 head = &queue->tasks[queue->maxpriority];
597 for (;;) {
Trond Myklebust540a0f72012-03-19 13:39:35 -0400598 while (!list_empty(head)) {
599 struct rpc_task *task;
600 task = list_first_entry(head,
601 struct rpc_task,
602 u.tk_wait.list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 task->tk_status = status;
Trond Myklebust96ef13b2008-02-22 15:46:41 -0500604 rpc_wake_up_task_queue_locked(queue, task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 }
606 if (head == &queue->tasks[0])
607 break;
608 head--;
609 }
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800610 spin_unlock_bh(&queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611}
Trond Myklebuste8914c62007-07-14 15:39:59 -0400612EXPORT_SYMBOL_GPL(rpc_wake_up_status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Trond Myklebust36df9aa2007-07-18 16:18:52 -0400614static void __rpc_queue_timer_fn(unsigned long ptr)
615{
616 struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
617 struct rpc_task *task, *n;
618 unsigned long expires, now, timeo;
619
620 spin_lock(&queue->lock);
621 expires = now = jiffies;
622 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
623 timeo = task->u.tk_wait.expires;
624 if (time_after_eq(now, timeo)) {
Trond Myklebust36df9aa2007-07-18 16:18:52 -0400625 dprintk("RPC: %5u timeout\n", task->tk_pid);
626 task->tk_status = -ETIMEDOUT;
627 rpc_wake_up_task_queue_locked(queue, task);
628 continue;
629 }
630 if (expires == now || time_after(expires, timeo))
631 expires = timeo;
632 }
633 if (!list_empty(&queue->timer_list.list))
634 rpc_set_queue_timer(queue, expires);
635 spin_unlock(&queue->lock);
636}
637
Trond Myklebust80147932006-08-31 18:24:08 -0400638static void __rpc_atrun(struct rpc_task *task)
639{
Trond Myklebust6bd14412014-03-20 12:53:54 -0400640 if (task->tk_status == -ETIMEDOUT)
641 task->tk_status = 0;
Trond Myklebust80147932006-08-31 18:24:08 -0400642}
643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644/*
645 * Run a task at a later time
646 */
Trond Myklebust80147932006-08-31 18:24:08 -0400647void rpc_delay(struct rpc_task *task, unsigned long delay)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
649 task->tk_timeout = delay;
Trond Myklebust5d008372008-02-22 16:34:17 -0500650 rpc_sleep_on(&delay_queue, task, __rpc_atrun);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651}
Trond Myklebuste8914c62007-07-14 15:39:59 -0400652EXPORT_SYMBOL_GPL(rpc_delay);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654/*
Trond Myklebust4ce70ad2006-01-03 09:55:05 +0100655 * Helper to call task->tk_ops->rpc_call_prepare
656 */
Andy Adamsonaae20062009-04-01 09:22:40 -0400657void rpc_prepare_task(struct rpc_task *task)
Trond Myklebust4ce70ad2006-01-03 09:55:05 +0100658{
659 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
660}
661
Trond Myklebust7fdcf132011-12-01 14:00:15 -0500662static void
663rpc_init_task_statistics(struct rpc_task *task)
664{
665 /* Initialize retry counters */
666 task->tk_garb_retry = 2;
667 task->tk_cred_retry = 2;
668 task->tk_rebind_retry = 2;
669
670 /* starting timestamp */
671 task->tk_start = ktime_get();
672}
673
674static void
675rpc_reset_task_statistics(struct rpc_task *task)
676{
677 task->tk_timeouts = 0;
678 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT);
679
680 rpc_init_task_statistics(task);
681}
682
Trond Myklebust4ce70ad2006-01-03 09:55:05 +0100683/*
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100684 * Helper that calls task->tk_ops->rpc_call_done if it exists
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000685 */
Trond Myklebustabbcf282006-01-03 09:55:03 +0100686void rpc_exit_task(struct rpc_task *task)
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000687{
Trond Myklebustabbcf282006-01-03 09:55:03 +0100688 task->tk_action = NULL;
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100689 if (task->tk_ops->rpc_call_done != NULL) {
690 task->tk_ops->rpc_call_done(task, task->tk_calldata);
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000691 if (task->tk_action != NULL) {
Trond Myklebustabbcf282006-01-03 09:55:03 +0100692 WARN_ON(RPC_ASSASSINATED(task));
693 /* Always release the RPC slot and buffer memory */
694 xprt_release(task);
Trond Myklebust7fdcf132011-12-01 14:00:15 -0500695 rpc_reset_task_statistics(task);
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000696 }
697 }
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000698}
Trond Myklebustd9b6cd92010-07-31 14:29:08 -0400699
700void rpc_exit(struct rpc_task *task, int status)
701{
702 task->tk_status = status;
703 task->tk_action = rpc_exit_task;
704 if (RPC_IS_QUEUED(task))
705 rpc_wake_up_queued_task(task->tk_waitqueue, task);
706}
707EXPORT_SYMBOL_GPL(rpc_exit);
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000708
Trond Myklebustbbd5a1f2006-10-18 16:01:05 -0400709void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
710{
Trond Myklebusta86dc492008-06-11 13:37:09 -0400711 if (ops->rpc_release != NULL)
Trond Myklebustbbd5a1f2006-10-18 16:01:05 -0400712 ops->rpc_release(calldata);
Trond Myklebustbbd5a1f2006-10-18 16:01:05 -0400713}
714
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000715/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 * This is the RPC `scheduler' (or rather, the finite state machine).
717 */
Trond Myklebust2efef832007-02-03 13:38:41 -0800718static void __rpc_execute(struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
Trond Myklebusteb9b55a2009-03-10 20:33:16 -0400720 struct rpc_wait_queue *queue;
721 int task_is_async = RPC_IS_ASYNC(task);
722 int status = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Chuck Lever46121cf2007-01-31 12:14:08 -0500724 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
725 task->tk_pid, task->tk_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726
Weston Andros Adamson2bd4eef2012-10-23 10:43:47 -0400727 WARN_ON_ONCE(RPC_IS_QUEUED(task));
728 if (RPC_IS_QUEUED(task))
729 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000731 for (;;) {
Trond Myklebustb55c5982011-07-06 19:58:23 -0400732 void (*do_action)(struct rpc_task *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 /*
Trond Myklebustb55c5982011-07-06 19:58:23 -0400735 * Execute any pending callback first.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 */
Trond Myklebustb55c5982011-07-06 19:58:23 -0400737 do_action = task->tk_callback;
738 task->tk_callback = NULL;
739 if (do_action == NULL) {
Trond Myklebuste020c682011-03-15 19:56:30 -0400740 /*
741 * Perform the next FSM step.
Trond Myklebustb55c5982011-07-06 19:58:23 -0400742 * tk_action may be NULL if the task has been killed.
743 * In particular, note that rpc_killall_tasks may
744 * do this at any time, so beware when dereferencing.
Trond Myklebuste020c682011-03-15 19:56:30 -0400745 */
Trond Myklebustb55c5982011-07-06 19:58:23 -0400746 do_action = task->tk_action;
747 if (do_action == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 }
Trond Myklebust82b0a4c2012-01-20 14:52:23 -0500750 trace_rpc_task_run_action(task->tk_client, task, task->tk_action);
Trond Myklebustb55c5982011-07-06 19:58:23 -0400751 do_action(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753 /*
754 * Lockless check for whether task is sleeping or not.
755 */
756 if (!RPC_IS_QUEUED(task))
757 continue;
Trond Myklebusteb9b55a2009-03-10 20:33:16 -0400758 /*
759 * The queue->lock protects against races with
760 * rpc_make_runnable().
761 *
762 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
763 * rpc_task, rpc_make_runnable() can assign it to a
764 * different workqueue. We therefore cannot assume that the
765 * rpc_task pointer may still be dereferenced.
766 */
767 queue = task->tk_waitqueue;
768 spin_lock_bh(&queue->lock);
769 if (!RPC_IS_QUEUED(task)) {
770 spin_unlock_bh(&queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 continue;
772 }
Trond Myklebusteb9b55a2009-03-10 20:33:16 -0400773 rpc_clear_running(task);
774 spin_unlock_bh(&queue->lock);
775 if (task_is_async)
776 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778 /* sync task: sleep here */
Chuck Lever46121cf2007-01-31 12:14:08 -0500779 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
Trond Myklebust96651ab2005-06-22 17:16:21 +0000780 status = out_of_line_wait_on_bit(&task->tk_runstate,
Matthew Wilcox150030b2007-12-06 16:24:39 -0500781 RPC_TASK_QUEUED, rpc_wait_bit_killable,
782 TASK_KILLABLE);
Trond Myklebust96651ab2005-06-22 17:16:21 +0000783 if (status == -ERESTARTSYS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 /*
785 * When a sync task receives a signal, it exits with
786 * -ERESTARTSYS. In order to catch any callbacks that
787 * clean up after sleeping on some queue, we don't
788 * break the loop here, but go around once more.
789 */
Chuck Lever46121cf2007-01-31 12:14:08 -0500790 dprintk("RPC: %5u got signal\n", task->tk_pid);
Trond Myklebust96651ab2005-06-22 17:16:21 +0000791 task->tk_flags |= RPC_TASK_KILLED;
792 rpc_exit(task, -ERESTARTSYS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 }
Chuck Lever46121cf2007-01-31 12:14:08 -0500794 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 }
796
Chuck Lever46121cf2007-01-31 12:14:08 -0500797 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
798 task->tk_status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 /* Release all resources associated with the task */
800 rpc_release_task(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801}
802
803/*
804 * User-visible entry point to the scheduler.
805 *
806 * This may be called recursively if e.g. an async NFS task updates
807 * the attributes and finds that dirty pages must be flushed.
808 * NOTE: Upon exit of this function the task is guaranteed to be
809 * released. In particular note that tk_release() will have
810 * been called, so your task memory may have been freed.
811 */
Trond Myklebust2efef832007-02-03 13:38:41 -0800812void rpc_execute(struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
Trond Myklebusta76580f2013-05-20 23:00:18 -0400814 bool is_async = RPC_IS_ASYNC(task);
815
Trond Myklebust44c28872006-01-03 09:55:06 +0100816 rpc_set_active(task);
Trond Myklebustd6a1ed02010-07-31 14:29:08 -0400817 rpc_make_runnable(task);
Trond Myklebusta76580f2013-05-20 23:00:18 -0400818 if (!is_async)
Trond Myklebustd6a1ed02010-07-31 14:29:08 -0400819 __rpc_execute(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820}
821
David Howells65f27f32006-11-22 14:55:48 +0000822static void rpc_async_schedule(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
David Howells65f27f32006-11-22 14:55:48 +0000824 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825}
826
Chuck Lever02107142006-01-03 09:55:49 +0100827/**
828 * rpc_malloc - allocate an RPC buffer
829 * @task: RPC task that will use this buffer
830 * @size: requested byte size
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 *
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400832 * To prevent rpciod from hanging, this allocator never sleeps,
David Rientjesc6c8fe72014-05-07 13:03:41 -0700833 * returning NULL and suppressing warning if the request cannot be serviced
834 * immediately.
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400835 * The caller can arrange to sleep in a way that is safe for rpciod.
836 *
837 * Most requests are 'small' (under 2KiB) and can be serviced from a
838 * mempool, ensuring that NFS reads and writes can always proceed,
839 * and that there is good locality of reference for these buffers.
840 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 * In order to avoid memory starvation triggering more writebacks of
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400842 * NFS requests, we avoid using GFP_KERNEL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 */
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400844void *rpc_malloc(struct rpc_task *task, size_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
Chuck Leveraa3d1fa2007-05-08 18:23:28 -0400846 struct rpc_buffer *buf;
Trond Myklebustc4a7ca72015-01-23 14:50:56 -0500847 gfp_t gfp = GFP_NOIO | __GFP_NOWARN;
Mel Gormana564b8f2012-07-31 16:45:12 -0700848
849 if (RPC_IS_SWAPPER(task))
Trond Myklebustc4a7ca72015-01-23 14:50:56 -0500850 gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
Chuck Leveraa3d1fa2007-05-08 18:23:28 -0400852 size += sizeof(struct rpc_buffer);
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400853 if (size <= RPC_BUFFER_MAXSIZE)
854 buf = mempool_alloc(rpc_buffer_mempool, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 else
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400856 buf = kmalloc(size, gfp);
Peter Zijlstraddce40d2007-05-09 08:30:11 +0200857
858 if (!buf)
859 return NULL;
860
Chuck Leveraa3d1fa2007-05-08 18:23:28 -0400861 buf->len = size;
Geert Uytterhoeven215d0672007-05-08 11:37:26 +0200862 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400863 task->tk_pid, size, buf);
Chuck Leveraa3d1fa2007-05-08 18:23:28 -0400864 return &buf->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865}
\"Talpey, Thomas\12444802007-09-10 13:45:36 -0400866EXPORT_SYMBOL_GPL(rpc_malloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Chuck Lever02107142006-01-03 09:55:49 +0100868/**
869 * rpc_free - free buffer allocated via rpc_malloc
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400870 * @buffer: buffer to free
Chuck Lever02107142006-01-03 09:55:49 +0100871 *
872 */
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400873void rpc_free(void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874{
Chuck Leveraa3d1fa2007-05-08 18:23:28 -0400875 size_t size;
876 struct rpc_buffer *buf;
Chuck Lever02107142006-01-03 09:55:49 +0100877
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400878 if (!buffer)
879 return;
Chuck Leveraa3d1fa2007-05-08 18:23:28 -0400880
881 buf = container_of(buffer, struct rpc_buffer, data);
882 size = buf->len;
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400883
Geert Uytterhoeven215d0672007-05-08 11:37:26 +0200884 dprintk("RPC: freeing buffer of size %zu at %p\n",
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400885 size, buf);
Chuck Leveraa3d1fa2007-05-08 18:23:28 -0400886
Chuck Leverc5a4dd82007-03-29 16:47:58 -0400887 if (size <= RPC_BUFFER_MAXSIZE)
888 mempool_free(buf, rpc_buffer_mempool);
889 else
890 kfree(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891}
\"Talpey, Thomas\12444802007-09-10 13:45:36 -0400892EXPORT_SYMBOL_GPL(rpc_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
894/*
895 * Creation and deletion of RPC task structures
896 */
Trond Myklebust47fe0642007-10-25 18:42:55 -0400897static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898{
899 memset(task, 0, sizeof(*task));
Trond Myklebust44c28872006-01-03 09:55:06 +0100900 atomic_set(&task->tk_count, 1);
Trond Myklebust84115e12007-07-14 15:39:59 -0400901 task->tk_flags = task_setup_data->flags;
902 task->tk_ops = task_setup_data->callback_ops;
903 task->tk_calldata = task_setup_data->callback_data;
Trond Myklebust6529eba2007-06-14 16:40:14 -0400904 INIT_LIST_HEAD(&task->tk_task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Trond Myklebust3ff75762007-07-14 15:40:00 -0400906 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
907 task->tk_owner = current->tgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 /* Initialize workqueue for async tasks */
Trond Myklebust32bfb5c2008-02-19 20:04:21 -0500910 task->tk_workqueue = task_setup_data->workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Trond Myklebust84115e12007-07-14 15:39:59 -0400912 if (task->tk_ops->rpc_call_prepare != NULL)
913 task->tk_action = rpc_prepare_task;
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100914
Trond Myklebust7fdcf132011-12-01 14:00:15 -0500915 rpc_init_task_statistics(task);
Chuck Leveref759a22006-03-20 13:44:17 -0500916
Chuck Lever46121cf2007-01-31 12:14:08 -0500917 dprintk("RPC: new task initialized, procpid %u\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700918 task_pid_nr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919}
920
921static struct rpc_task *
922rpc_alloc_task(void)
923{
Mel Gormana564b8f2012-07-31 16:45:12 -0700924 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925}
926
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927/*
Trond Myklebust90c57552007-06-09 19:49:36 -0400928 * Create a new task for the specified client.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 */
Trond Myklebust84115e12007-07-14 15:39:59 -0400930struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
Trond Myklebuste8f5d772007-10-25 18:42:53 -0400932 struct rpc_task *task = setup_data->task;
933 unsigned short flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Trond Myklebuste8f5d772007-10-25 18:42:53 -0400935 if (task == NULL) {
936 task = rpc_alloc_task();
Trond Myklebust19445b92010-04-16 16:41:10 -0400937 if (task == NULL) {
938 rpc_release_calldata(setup_data->callback_ops,
939 setup_data->callback_data);
940 return ERR_PTR(-ENOMEM);
941 }
Trond Myklebuste8f5d772007-10-25 18:42:53 -0400942 flags = RPC_TASK_DYNAMIC;
943 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Trond Myklebust84115e12007-07-14 15:39:59 -0400945 rpc_init_task(task, setup_data);
Trond Myklebuste8f5d772007-10-25 18:42:53 -0400946 task->tk_flags |= flags;
Chuck Lever46121cf2007-01-31 12:14:08 -0500947 dprintk("RPC: allocated task %p\n", task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 return task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949}
950
Trond Myklebustc6567ed2013-01-04 12:23:21 -0500951/*
952 * rpc_free_task - release rpc task and perform cleanups
953 *
954 * Note that we free up the rpc_task _after_ rpc_release_calldata()
955 * in order to work around a workqueue dependency issue.
956 *
957 * Tejun Heo states:
958 * "Workqueue currently considers two work items to be the same if they're
959 * on the same address and won't execute them concurrently - ie. it
960 * makes a work item which is queued again while being executed wait
961 * for the previous execution to complete.
962 *
963 * If a work function frees the work item, and then waits for an event
964 * which should be performed by another work item and *that* work item
965 * recycles the freed work item, it can create a false dependency loop.
966 * There really is no reliable way to detect this short of verifying
967 * every memory free."
968 *
969 */
Trond Myklebust32bfb5c2008-02-19 20:04:21 -0500970static void rpc_free_task(struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971{
Trond Myklebustc6567ed2013-01-04 12:23:21 -0500972 unsigned short tk_flags = task->tk_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Trond Myklebustc6567ed2013-01-04 12:23:21 -0500974 rpc_release_calldata(task->tk_ops, task->tk_calldata);
975
976 if (tk_flags & RPC_TASK_DYNAMIC) {
Trond Myklebust5e4424a2008-02-25 21:53:49 -0800977 dprintk("RPC: %5u freeing task\n", task->tk_pid);
978 mempool_free(task, rpc_task_mempool);
979 }
Trond Myklebust32bfb5c2008-02-19 20:04:21 -0500980}
981
982static void rpc_async_release(struct work_struct *work)
983{
984 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
985}
986
Trond Myklebustbf294b42011-02-21 11:05:41 -0800987static void rpc_release_resources_task(struct rpc_task *task)
Trond Myklebust32bfb5c2008-02-19 20:04:21 -0500988{
Trond Myklebust87ed5002013-01-07 14:30:46 -0500989 xprt_release(task);
OGAWA Hirofumia271c5a2011-03-27 17:48:57 +0200990 if (task->tk_msg.rpc_cred) {
Trond Myklebusta17c2152010-07-31 14:29:08 -0400991 put_rpccred(task->tk_msg.rpc_cred);
OGAWA Hirofumia271c5a2011-03-27 17:48:57 +0200992 task->tk_msg.rpc_cred = NULL;
993 }
Trond Myklebust58f96122010-07-31 14:29:08 -0400994 rpc_task_release_client(task);
Trond Myklebustbf294b42011-02-21 11:05:41 -0800995}
996
997static void rpc_final_put_task(struct rpc_task *task,
998 struct workqueue_struct *q)
999{
1000 if (q != NULL) {
Trond Myklebust32bfb5c2008-02-19 20:04:21 -05001001 INIT_WORK(&task->u.tk_work, rpc_async_release);
Trond Myklebustbf294b42011-02-21 11:05:41 -08001002 queue_work(q, &task->u.tk_work);
Trond Myklebust32bfb5c2008-02-19 20:04:21 -05001003 } else
1004 rpc_free_task(task);
Trond Myklebuste6b3c4d2006-11-11 22:18:03 -05001005}
Trond Myklebustbf294b42011-02-21 11:05:41 -08001006
1007static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1008{
1009 if (atomic_dec_and_test(&task->tk_count)) {
1010 rpc_release_resources_task(task);
1011 rpc_final_put_task(task, q);
1012 }
1013}
1014
1015void rpc_put_task(struct rpc_task *task)
1016{
1017 rpc_do_put_task(task, NULL);
1018}
Trond Myklebuste8914c62007-07-14 15:39:59 -04001019EXPORT_SYMBOL_GPL(rpc_put_task);
Trond Myklebuste6b3c4d2006-11-11 22:18:03 -05001020
Trond Myklebustbf294b42011-02-21 11:05:41 -08001021void rpc_put_task_async(struct rpc_task *task)
1022{
1023 rpc_do_put_task(task, task->tk_workqueue);
1024}
1025EXPORT_SYMBOL_GPL(rpc_put_task_async);
1026
Trond Myklebustbde8f002007-01-24 11:54:53 -08001027static void rpc_release_task(struct rpc_task *task)
Trond Myklebuste6b3c4d2006-11-11 22:18:03 -05001028{
Chuck Lever46121cf2007-01-31 12:14:08 -05001029 dprintk("RPC: %5u release task\n", task->tk_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Weston Andros Adamson0a0c2a52012-10-23 10:43:49 -04001031 WARN_ON_ONCE(RPC_IS_QUEUED(task));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Trond Myklebustbf294b42011-02-21 11:05:41 -08001033 rpc_release_resources_task(task);
Trond Myklebuste6b3c4d2006-11-11 22:18:03 -05001034
Trond Myklebustbf294b42011-02-21 11:05:41 -08001035 /*
1036 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1037 * so it should be safe to use task->tk_count as a test for whether
1038 * or not any other processes still hold references to our rpc_task.
1039 */
1040 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1041 /* Wake up anyone who may be waiting for task completion */
1042 if (!rpc_complete_task(task))
1043 return;
1044 } else {
1045 if (!atomic_dec_and_test(&task->tk_count))
1046 return;
1047 }
1048 rpc_final_put_task(task, task->tk_workqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049}
1050
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001051int rpciod_up(void)
1052{
1053 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1054}
1055
1056void rpciod_down(void)
1057{
1058 module_put(THIS_MODULE);
1059}
1060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061/*
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001062 * Start up the rpciod workqueue.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 */
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001064static int rpciod_start(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065{
1066 struct workqueue_struct *wq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 /*
1069 * Create the rpciod thread and wait for it to start.
1070 */
Trond Myklebustab418d72007-06-14 17:08:36 -04001071 dprintk("RPC: creating workqueue rpciod\n");
Trond Myklebust127b21b2015-01-23 13:19:17 -05001072 /* Note: highpri because network receive is latency sensitive */
1073 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 rpciod_workqueue = wq;
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001075 return rpciod_workqueue != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076}
1077
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001078static void rpciod_stop(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079{
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001080 struct workqueue_struct *wq = NULL;
Trond Myklebustab418d72007-06-14 17:08:36 -04001081
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001082 if (rpciod_workqueue == NULL)
1083 return;
Trond Myklebustab418d72007-06-14 17:08:36 -04001084 dprintk("RPC: destroying workqueue rpciod\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001086 wq = rpciod_workqueue;
1087 rpciod_workqueue = NULL;
1088 destroy_workqueue(wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089}
1090
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091void
1092rpc_destroy_mempool(void)
1093{
Trond Myklebustb247bbf2007-07-19 16:32:20 -04001094 rpciod_stop();
Julia Lawall17a96182015-09-13 14:15:07 +02001095 mempool_destroy(rpc_buffer_mempool);
1096 mempool_destroy(rpc_task_mempool);
1097 kmem_cache_destroy(rpc_task_slabp);
1098 kmem_cache_destroy(rpc_buffer_slabp);
Trond Myklebustf6a1cc82008-02-22 17:06:55 -05001099 rpc_destroy_wait_queue(&delay_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100}
1101
1102int
1103rpc_init_mempool(void)
1104{
Trond Myklebustf6a1cc82008-02-22 17:06:55 -05001105 /*
1106 * The following is not strictly a mempool initialisation,
1107 * but there is no harm in doing it here
1108 */
1109 rpc_init_wait_queue(&delay_queue, "delayq");
1110 if (!rpciod_start())
1111 goto err_nomem;
1112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1114 sizeof(struct rpc_task),
1115 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001116 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 if (!rpc_task_slabp)
1118 goto err_nomem;
1119 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1120 RPC_BUFFER_MAXSIZE,
1121 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001122 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 if (!rpc_buffer_slabp)
1124 goto err_nomem;
Matthew Dobson93d23412006-03-26 01:37:50 -08001125 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1126 rpc_task_slabp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (!rpc_task_mempool)
1128 goto err_nomem;
Matthew Dobson93d23412006-03-26 01:37:50 -08001129 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1130 rpc_buffer_slabp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 if (!rpc_buffer_mempool)
1132 goto err_nomem;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return 0;
1134err_nomem:
1135 rpc_destroy_mempool();
1136 return -ENOMEM;
1137}