]> git.proxmox.com Git - mirror_zfs.git/blob - lib/libzpool/taskq.c
3246 ZFS I/O deadman thread
[mirror_zfs.git] / lib / libzpool / taskq.c
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25 /*
26 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
27 * Copyright 2012 Garrett D'Amore <garrett@damore.org>. All rights reserved.
28 */
29
30 #include <sys/zfs_context.h>
31
32 int taskq_now;
33 taskq_t *system_taskq;
34
35 #define TASKQ_ACTIVE 0x00010000
36
37 struct taskq {
38 kmutex_t tq_lock;
39 krwlock_t tq_threadlock;
40 kcondvar_t tq_dispatch_cv;
41 kcondvar_t tq_wait_cv;
42 kthread_t **tq_threadlist;
43 int tq_flags;
44 int tq_active;
45 int tq_nthreads;
46 int tq_nalloc;
47 int tq_minalloc;
48 int tq_maxalloc;
49 kcondvar_t tq_maxalloc_cv;
50 int tq_maxalloc_wait;
51 taskq_ent_t *tq_freelist;
52 taskq_ent_t tq_task;
53 };
54
55 static taskq_ent_t *
56 task_alloc(taskq_t *tq, int tqflags)
57 {
58 taskq_ent_t *t;
59 int rv;
60
61 again: if ((t = tq->tq_freelist) != NULL && tq->tq_nalloc >= tq->tq_minalloc) {
62 ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC));
63 tq->tq_freelist = t->tqent_next;
64 } else {
65 if (tq->tq_nalloc >= tq->tq_maxalloc) {
66 if (!(tqflags & KM_SLEEP))
67 return (NULL);
68
69 /*
70 * We don't want to exceed tq_maxalloc, but we can't
71 * wait for other tasks to complete (and thus free up
72 * task structures) without risking deadlock with
73 * the caller. So, we just delay for one second
74 * to throttle the allocation rate. If we have tasks
75 * complete before one second timeout expires then
76 * taskq_ent_free will signal us and we will
77 * immediately retry the allocation.
78 */
79 tq->tq_maxalloc_wait++;
80 rv = cv_timedwait(&tq->tq_maxalloc_cv,
81 &tq->tq_lock, ddi_get_lbolt() + hz);
82 tq->tq_maxalloc_wait--;
83 if (rv > 0)
84 goto again; /* signaled */
85 }
86 mutex_exit(&tq->tq_lock);
87
88 t = kmem_alloc(sizeof (taskq_ent_t), tqflags);
89
90 mutex_enter(&tq->tq_lock);
91 if (t != NULL) {
92 /* Make sure we start without any flags */
93 t->tqent_flags = 0;
94 tq->tq_nalloc++;
95 }
96 }
97 return (t);
98 }
99
100 static void
101 task_free(taskq_t *tq, taskq_ent_t *t)
102 {
103 if (tq->tq_nalloc <= tq->tq_minalloc) {
104 t->tqent_next = tq->tq_freelist;
105 tq->tq_freelist = t;
106 } else {
107 tq->tq_nalloc--;
108 mutex_exit(&tq->tq_lock);
109 kmem_free(t, sizeof (taskq_ent_t));
110 mutex_enter(&tq->tq_lock);
111 }
112
113 if (tq->tq_maxalloc_wait)
114 cv_signal(&tq->tq_maxalloc_cv);
115 }
116
117 taskqid_t
118 taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags)
119 {
120 taskq_ent_t *t;
121
122 if (taskq_now) {
123 func(arg);
124 return (1);
125 }
126
127 mutex_enter(&tq->tq_lock);
128 ASSERT(tq->tq_flags & TASKQ_ACTIVE);
129 if ((t = task_alloc(tq, tqflags)) == NULL) {
130 mutex_exit(&tq->tq_lock);
131 return (0);
132 }
133 if (tqflags & TQ_FRONT) {
134 t->tqent_next = tq->tq_task.tqent_next;
135 t->tqent_prev = &tq->tq_task;
136 } else {
137 t->tqent_next = &tq->tq_task;
138 t->tqent_prev = tq->tq_task.tqent_prev;
139 }
140 t->tqent_next->tqent_prev = t;
141 t->tqent_prev->tqent_next = t;
142 t->tqent_func = func;
143 t->tqent_arg = arg;
144 t->tqent_flags = 0;
145 cv_signal(&tq->tq_dispatch_cv);
146 mutex_exit(&tq->tq_lock);
147 return (1);
148 }
149
150 taskqid_t
151 taskq_dispatch_delay(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags,
152 clock_t expire_time)
153 {
154 return (0);
155 }
156
157 int
158 taskq_empty_ent(taskq_ent_t *t)
159 {
160 return t->tqent_next == NULL;
161 }
162
163 void
164 taskq_init_ent(taskq_ent_t *t)
165 {
166 t->tqent_next = NULL;
167 t->tqent_prev = NULL;
168 t->tqent_func = NULL;
169 t->tqent_arg = NULL;
170 t->tqent_flags = 0;
171 }
172
173 void
174 taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags,
175 taskq_ent_t *t)
176 {
177 ASSERT(func != NULL);
178 ASSERT(!(tq->tq_flags & TASKQ_DYNAMIC));
179
180 /*
181 * Mark it as a prealloc'd task. This is important
182 * to ensure that we don't free it later.
183 */
184 t->tqent_flags |= TQENT_FLAG_PREALLOC;
185 /*
186 * Enqueue the task to the underlying queue.
187 */
188 mutex_enter(&tq->tq_lock);
189
190 if (flags & TQ_FRONT) {
191 t->tqent_next = tq->tq_task.tqent_next;
192 t->tqent_prev = &tq->tq_task;
193 } else {
194 t->tqent_next = &tq->tq_task;
195 t->tqent_prev = tq->tq_task.tqent_prev;
196 }
197 t->tqent_next->tqent_prev = t;
198 t->tqent_prev->tqent_next = t;
199 t->tqent_func = func;
200 t->tqent_arg = arg;
201 cv_signal(&tq->tq_dispatch_cv);
202 mutex_exit(&tq->tq_lock);
203 }
204
205 void
206 taskq_wait(taskq_t *tq)
207 {
208 mutex_enter(&tq->tq_lock);
209 while (tq->tq_task.tqent_next != &tq->tq_task || tq->tq_active != 0)
210 cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
211 mutex_exit(&tq->tq_lock);
212 }
213
214 static void
215 taskq_thread(void *arg)
216 {
217 taskq_t *tq = arg;
218 taskq_ent_t *t;
219 boolean_t prealloc;
220
221 mutex_enter(&tq->tq_lock);
222 while (tq->tq_flags & TASKQ_ACTIVE) {
223 if ((t = tq->tq_task.tqent_next) == &tq->tq_task) {
224 if (--tq->tq_active == 0)
225 cv_broadcast(&tq->tq_wait_cv);
226 cv_wait(&tq->tq_dispatch_cv, &tq->tq_lock);
227 tq->tq_active++;
228 continue;
229 }
230 t->tqent_prev->tqent_next = t->tqent_next;
231 t->tqent_next->tqent_prev = t->tqent_prev;
232 t->tqent_next = NULL;
233 t->tqent_prev = NULL;
234 prealloc = t->tqent_flags & TQENT_FLAG_PREALLOC;
235 mutex_exit(&tq->tq_lock);
236
237 rw_enter(&tq->tq_threadlock, RW_READER);
238 t->tqent_func(t->tqent_arg);
239 rw_exit(&tq->tq_threadlock);
240
241 mutex_enter(&tq->tq_lock);
242 if (!prealloc)
243 task_free(tq, t);
244 }
245 tq->tq_nthreads--;
246 cv_broadcast(&tq->tq_wait_cv);
247 mutex_exit(&tq->tq_lock);
248 thread_exit();
249 }
250
251 /*ARGSUSED*/
252 taskq_t *
253 taskq_create(const char *name, int nthreads, pri_t pri,
254 int minalloc, int maxalloc, uint_t flags)
255 {
256 taskq_t *tq = kmem_zalloc(sizeof (taskq_t), KM_SLEEP);
257 int t;
258
259 if (flags & TASKQ_THREADS_CPU_PCT) {
260 int pct;
261 ASSERT3S(nthreads, >=, 0);
262 ASSERT3S(nthreads, <=, 100);
263 pct = MIN(nthreads, 100);
264 pct = MAX(pct, 0);
265
266 nthreads = (sysconf(_SC_NPROCESSORS_ONLN) * pct) / 100;
267 nthreads = MAX(nthreads, 1); /* need at least 1 thread */
268 } else {
269 ASSERT3S(nthreads, >=, 1);
270 }
271
272 rw_init(&tq->tq_threadlock, NULL, RW_DEFAULT, NULL);
273 mutex_init(&tq->tq_lock, NULL, MUTEX_DEFAULT, NULL);
274 cv_init(&tq->tq_dispatch_cv, NULL, CV_DEFAULT, NULL);
275 cv_init(&tq->tq_wait_cv, NULL, CV_DEFAULT, NULL);
276 cv_init(&tq->tq_maxalloc_cv, NULL, CV_DEFAULT, NULL);
277 tq->tq_flags = flags | TASKQ_ACTIVE;
278 tq->tq_active = nthreads;
279 tq->tq_nthreads = nthreads;
280 tq->tq_minalloc = minalloc;
281 tq->tq_maxalloc = maxalloc;
282 tq->tq_task.tqent_next = &tq->tq_task;
283 tq->tq_task.tqent_prev = &tq->tq_task;
284 tq->tq_threadlist = kmem_alloc(nthreads*sizeof(kthread_t *), KM_SLEEP);
285
286 if (flags & TASKQ_PREPOPULATE) {
287 mutex_enter(&tq->tq_lock);
288 while (minalloc-- > 0)
289 task_free(tq, task_alloc(tq, KM_SLEEP));
290 mutex_exit(&tq->tq_lock);
291 }
292
293 for (t = 0; t < nthreads; t++)
294 VERIFY((tq->tq_threadlist[t] = thread_create(NULL, 0,
295 taskq_thread, tq, TS_RUN, NULL, 0, 0)) != NULL);
296
297 return (tq);
298 }
299
300 void
301 taskq_destroy(taskq_t *tq)
302 {
303 int nthreads = tq->tq_nthreads;
304
305 taskq_wait(tq);
306
307 mutex_enter(&tq->tq_lock);
308
309 tq->tq_flags &= ~TASKQ_ACTIVE;
310 cv_broadcast(&tq->tq_dispatch_cv);
311
312 while (tq->tq_nthreads != 0)
313 cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
314
315 tq->tq_minalloc = 0;
316 while (tq->tq_nalloc != 0) {
317 ASSERT(tq->tq_freelist != NULL);
318 task_free(tq, task_alloc(tq, KM_SLEEP));
319 }
320
321 mutex_exit(&tq->tq_lock);
322
323 kmem_free(tq->tq_threadlist, nthreads * sizeof (kthread_t *));
324
325 rw_destroy(&tq->tq_threadlock);
326 mutex_destroy(&tq->tq_lock);
327 cv_destroy(&tq->tq_dispatch_cv);
328 cv_destroy(&tq->tq_wait_cv);
329 cv_destroy(&tq->tq_maxalloc_cv);
330
331 kmem_free(tq, sizeof (taskq_t));
332 }
333
334 int
335 taskq_member(taskq_t *tq, kthread_t *t)
336 {
337 int i;
338
339 if (taskq_now)
340 return (1);
341
342 for (i = 0; i < tq->tq_nthreads; i++)
343 if (tq->tq_threadlist[i] == t)
344 return (1);
345
346 return (0);
347 }
348
349 int
350 taskq_cancel_id(taskq_t *tq, taskqid_t id)
351 {
352 return (ENOENT);
353 }
354
355 void
356 system_taskq_init(void)
357 {
358 system_taskq = taskq_create("system_taskq", 64, minclsyspri, 4, 512,
359 TASKQ_DYNAMIC | TASKQ_PREPOPULATE);
360 }
361
362 void
363 system_taskq_fini(void)
364 {
365 taskq_destroy(system_taskq);
366 system_taskq = NULL; /* defensive */
367 }