usr/src/lib/libzpool/common/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  * Copyright (c) 2014 by Delphix. All rights reserved.
  29  */
  30
  31 #include <sys/zfs_context.h>
  32
  33 int taskq_now;
  34 taskq_t *system_taskq;
  35
  36 #define TASKQ_ACTIVE    0x00010000
  37 #define TASKQ_NAMELEN   31
  38
  39 struct taskq {
  40         char            tq_name[TASKQ_NAMELEN + 1];
  41         kmutex_t        tq_lock;
  42         krwlock_t       tq_threadlock;
  43         kcondvar_t      tq_dispatch_cv;
  44         kcondvar_t      tq_wait_cv;
  45         thread_t        *tq_threadlist;
  46         int             tq_flags;
  47         int             tq_active;
  48         int             tq_nthreads;
  49         int             tq_nalloc;
  50         int             tq_minalloc;
  51         int             tq_maxalloc;
  52         kcondvar_t      tq_maxalloc_cv;
  53         int             tq_maxalloc_wait;
  54         taskq_ent_t     *tq_freelist;
  55         taskq_ent_t     tq_task;
  56 };
  57
  58 static taskq_ent_t *
  59 task_alloc(taskq_t *tq, int tqflags)
  60 {
  61         taskq_ent_t *t;
  62         int rv;
  63
  64 again:  if ((t = tq->tq_freelist) != NULL && tq->tq_nalloc >= tq->tq_minalloc) {
  65                 tq->tq_freelist = t->tqent_next;
  66         } else {
  67                 if (tq->tq_nalloc >= tq->tq_maxalloc) {
  68                         if (!(tqflags & KM_SLEEP))
  69                                 return (NULL);
  70
  71                         /*
  72                          * We don't want to exceed tq_maxalloc, but we can't
  73                          * wait for other tasks to complete (and thus free up
  74                          * task structures) without risking deadlock with
  75                          * the caller.  So, we just delay for one second
  76                          * to throttle the allocation rate. If we have tasks
  77                          * complete before one second timeout expires then
  78                          * taskq_ent_free will signal us and we will
  79                          * immediately retry the allocation.
  80                          */
  81                         tq->tq_maxalloc_wait++;
  82                         rv = cv_timedwait(&tq->tq_maxalloc_cv,
  83                             &tq->tq_lock, ddi_get_lbolt() + hz);
  84                         tq->tq_maxalloc_wait--;
  85                         if (rv > 0)
  86                                 goto again;             /* signaled */
  87                 }
  88                 mutex_exit(&tq->tq_lock);
  89
  90                 t = kmem_alloc(sizeof (taskq_ent_t), tqflags);
  91
  92                 mutex_enter(&tq->tq_lock);
  93                 if (t != NULL)
  94                         tq->tq_nalloc++;
  95         }
  96         return (t);
  97 }
  98
  99 static void
 100 task_free(taskq_t *tq, taskq_ent_t *t)
 101 {
 102         if (tq->tq_nalloc <= tq->tq_minalloc) {
 103                 t->tqent_next = tq->tq_freelist;
 104                 tq->tq_freelist = t;
 105         } else {
 106                 tq->tq_nalloc--;
 107                 mutex_exit(&tq->tq_lock);
 108                 kmem_free(t, sizeof (taskq_ent_t));
 109                 mutex_enter(&tq->tq_lock);
 110         }
 111
 112         if (tq->tq_maxalloc_wait)
 113                 cv_signal(&tq->tq_maxalloc_cv);
 114 }
 115
 116 taskqid_t
 117 taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags)
 118 {
 119         taskq_ent_t *t;
 120
 121         if (taskq_now) {
 122                 func(arg);
 123                 return (1);
 124         }
 125
 126         mutex_enter(&tq->tq_lock);
 127         ASSERT(tq->tq_flags & TASKQ_ACTIVE);
 128         if ((t = task_alloc(tq, tqflags)) == NULL) {
 129                 mutex_exit(&tq->tq_lock);
 130                 return (0);
 131         }
 132         if (tqflags & TQ_FRONT) {
 133                 t->tqent_next = tq->tq_task.tqent_next;
 134                 t->tqent_prev = &tq->tq_task;
 135         } else {
 136                 t->tqent_next = &tq->tq_task;
 137                 t->tqent_prev = tq->tq_task.tqent_prev;
 138         }
 139         t->tqent_next->tqent_prev = t;
 140         t->tqent_prev->tqent_next = t;
 141         t->tqent_func = func;
 142         t->tqent_arg = arg;
 143         t->tqent_flags = 0;
 144         cv_signal(&tq->tq_dispatch_cv);
 145         mutex_exit(&tq->tq_lock);
 146         return (1);
 147 }
 148
 149 void
 150 taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags,
 151     taskq_ent_t *t)
 152 {
 153         ASSERT(func != NULL);
 154         ASSERT(!(tq->tq_flags & TASKQ_DYNAMIC));
 155
 156         /*
 157          * Mark it as a prealloc'd task.  This is important
 158          * to ensure that we don't free it later.
 159          */
 160         t->tqent_flags |= TQENT_FLAG_PREALLOC;
 161         /*
 162          * Enqueue the task to the underlying queue.
 163          */
 164         mutex_enter(&tq->tq_lock);
 165
 166         if (flags & TQ_FRONT) {
 167                 t->tqent_next = tq->tq_task.tqent_next;
 168                 t->tqent_prev = &tq->tq_task;
 169         } else {
 170                 t->tqent_next = &tq->tq_task;
 171                 t->tqent_prev = tq->tq_task.tqent_prev;
 172         }
 173         t->tqent_next->tqent_prev = t;
 174         t->tqent_prev->tqent_next = t;
 175         t->tqent_func = func;
 176         t->tqent_arg = arg;
 177         cv_signal(&tq->tq_dispatch_cv);
 178         mutex_exit(&tq->tq_lock);
 179 }
 180
 181 void
 182 taskq_wait(taskq_t *tq)
 183 {
 184         mutex_enter(&tq->tq_lock);
 185         while (tq->tq_task.tqent_next != &tq->tq_task || tq->tq_active != 0)
 186                 cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
 187         mutex_exit(&tq->tq_lock);
 188 }
 189
 190 static void *
 191 taskq_thread(void *arg)
 192 {
 193         taskq_t *tq = arg;
 194         taskq_ent_t *t;
 195         boolean_t prealloc;
 196
 197         mutex_enter(&tq->tq_lock);
 198         while (tq->tq_flags & TASKQ_ACTIVE) {
 199                 if ((t = tq->tq_task.tqent_next) == &tq->tq_task) {
 200                         if (--tq->tq_active == 0)
 201                                 cv_broadcast(&tq->tq_wait_cv);
 202                         cv_wait(&tq->tq_dispatch_cv, &tq->tq_lock);
 203                         tq->tq_active++;
 204                         continue;
 205                 }
 206                 t->tqent_prev->tqent_next = t->tqent_next;
 207                 t->tqent_next->tqent_prev = t->tqent_prev;
 208                 t->tqent_next = NULL;
 209                 t->tqent_prev = NULL;
 210                 prealloc = t->tqent_flags & TQENT_FLAG_PREALLOC;
 211                 mutex_exit(&tq->tq_lock);
 212
 213                 rw_enter(&tq->tq_threadlock, RW_READER);
 214                 t->tqent_func(t->tqent_arg);
 215                 rw_exit(&tq->tq_threadlock);
 216
 217                 mutex_enter(&tq->tq_lock);
 218                 if (!prealloc)
 219                         task_free(tq, t);
 220         }
 221         tq->tq_nthreads--;
 222         cv_broadcast(&tq->tq_wait_cv);
 223         mutex_exit(&tq->tq_lock);
 224         return (NULL);
 225 }
 226
 227 /*ARGSUSED*/
 228 taskq_t *
 229 taskq_create(const char *name, int nthreads, pri_t pri,
 230         int minalloc, int maxalloc, uint_t flags)
 231 {
 232         taskq_t *tq = kmem_zalloc(sizeof (taskq_t), KM_SLEEP);
 233         int t;
 234
 235         if (flags & TASKQ_THREADS_CPU_PCT) {
 236                 int pct;
 237                 ASSERT3S(nthreads, >=, 0);
 238                 ASSERT3S(nthreads, <=, 100);
 239                 pct = MIN(nthreads, 100);
 240                 pct = MAX(pct, 0);
 241
 242                 nthreads = (sysconf(_SC_NPROCESSORS_ONLN) * pct) / 100;
 243                 nthreads = MAX(nthreads, 1);    /* need at least 1 thread */
 244         } else {
 245                 ASSERT3S(nthreads, >=, 1);
 246         }
 247
 248         rw_init(&tq->tq_threadlock, NULL, RW_DEFAULT, NULL);
 249         mutex_init(&tq->tq_lock, NULL, MUTEX_DEFAULT, NULL);
 250         cv_init(&tq->tq_dispatch_cv, NULL, CV_DEFAULT, NULL);
 251         cv_init(&tq->tq_wait_cv, NULL, CV_DEFAULT, NULL);
 252         cv_init(&tq->tq_maxalloc_cv, NULL, CV_DEFAULT, NULL);
 253         (void) strncpy(tq->tq_name, name, TASKQ_NAMELEN + 1);
 254         tq->tq_flags = flags | TASKQ_ACTIVE;
 255         tq->tq_active = nthreads;
 256         tq->tq_nthreads = nthreads;
 257         tq->tq_minalloc = minalloc;
 258         tq->tq_maxalloc = maxalloc;
 259         tq->tq_task.tqent_next = &tq->tq_task;
 260         tq->tq_task.tqent_prev = &tq->tq_task;
 261         tq->tq_threadlist = kmem_alloc(nthreads * sizeof (thread_t), KM_SLEEP);
 262
 263         if (flags & TASKQ_PREPOPULATE) {
 264                 mutex_enter(&tq->tq_lock);
 265                 while (minalloc-- > 0)
 266                         task_free(tq, task_alloc(tq, KM_SLEEP));
 267                 mutex_exit(&tq->tq_lock);
 268         }
 269
 270         for (t = 0; t < nthreads; t++)
 271                 (void) thr_create(0, 0, taskq_thread,
 272                     tq, THR_BOUND, &tq->tq_threadlist[t]);
 273
 274         return (tq);
 275 }
 276
 277 void
 278 taskq_destroy(taskq_t *tq)
 279 {
 280         int t;
 281         int nthreads = tq->tq_nthreads;
 282
 283         taskq_wait(tq);
 284
 285         mutex_enter(&tq->tq_lock);
 286
 287         tq->tq_flags &= ~TASKQ_ACTIVE;
 288         cv_broadcast(&tq->tq_dispatch_cv);
 289
 290         while (tq->tq_nthreads != 0)
 291                 cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
 292
 293         tq->tq_minalloc = 0;
 294         while (tq->tq_nalloc != 0) {
 295                 ASSERT(tq->tq_freelist != NULL);
 296                 task_free(tq, task_alloc(tq, KM_SLEEP));
 297         }
 298
 299         mutex_exit(&tq->tq_lock);
 300
 301         for (t = 0; t < nthreads; t++)
 302                 (void) thr_join(tq->tq_threadlist[t], NULL, NULL);
 303
 304         kmem_free(tq->tq_threadlist, nthreads * sizeof (thread_t));
 305
 306         rw_destroy(&tq->tq_threadlock);
 307         mutex_destroy(&tq->tq_lock);
 308         cv_destroy(&tq->tq_dispatch_cv);
 309         cv_destroy(&tq->tq_wait_cv);
 310         cv_destroy(&tq->tq_maxalloc_cv);
 311
 312         kmem_free(tq, sizeof (taskq_t));
 313 }
 314
 315 int
 316 taskq_member(taskq_t *tq, void *t)
 317 {
 318         int i;
 319
 320         if (taskq_now)
 321                 return (1);
 322
 323         for (i = 0; i < tq->tq_nthreads; i++)
 324                 if (tq->tq_threadlist[i] == (thread_t)(uintptr_t)t)
 325                         return (1);
 326
 327         return (0);
 328 }
 329
 330 void
 331 system_taskq_init(void)
 332 {
 333         system_taskq = taskq_create("system_taskq", 64, minclsyspri, 4, 512,
 334             TASKQ_DYNAMIC | TASKQ_PREPOPULATE);
 335 }
 336
 337 void
 338 system_taskq_fini(void)
 339 {
 340         taskq_destroy(system_taskq);
 341         system_taskq = NULL; /* defensive */
 342 }