Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
aae22ca7 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
ebac4655 | 6 | * |
8e9fe637 JA |
7 | * The license below covers all files distributed with fio unless otherwise |
8 | * noted in the file itself. | |
9 | * | |
ebac4655 | 10 | * This program is free software; you can redistribute it and/or modify |
8e9fe637 JA |
11 | * it under the terms of the GNU General Public License version 2 as |
12 | * published by the Free Software Foundation. | |
ebac4655 JA |
13 | * |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
21 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
22 | * | |
23 | */ | |
ebac4655 JA |
24 | #include <unistd.h> |
25 | #include <fcntl.h> | |
26 | #include <string.h> | |
ebac4655 JA |
27 | #include <signal.h> |
28 | #include <time.h> | |
dbe1125e | 29 | #include <locale.h> |
36167d82 | 30 | #include <assert.h> |
a5b01f1b | 31 | #include <time.h> |
ebac4655 JA |
32 | #include <sys/stat.h> |
33 | #include <sys/wait.h> | |
34 | #include <sys/ipc.h> | |
35 | #include <sys/shm.h> | |
ebac4655 JA |
36 | #include <sys/mman.h> |
37 | ||
38 | #include "fio.h" | |
210f43b3 | 39 | #include "hash.h" |
2e5cdb11 | 40 | #include "smalloc.h" |
4f5af7b2 | 41 | #include "verify.h" |
a917a8b3 | 42 | #include "trim.h" |
7c9b1bce | 43 | #include "diskutil.h" |
a696fa2a | 44 | #include "cgroup.h" |
07b3232d | 45 | #include "profile.h" |
d54a144d | 46 | #include "lib/rand.h" |
ebac4655 | 47 | |
cfc99db7 JA |
48 | unsigned long page_mask; |
49 | unsigned long page_size; | |
d529ee19 JA |
50 | |
51 | #define PAGE_ALIGN(buf) \ | |
29d610e1 | 52 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) |
ebac4655 JA |
53 | |
54 | int groupid = 0; | |
55 | int thread_number = 0; | |
9cedf167 JA |
56 | int nr_process = 0; |
57 | int nr_thread = 0; | |
ebac4655 | 58 | int shm_id = 0; |
53cdc686 | 59 | int temp_stall_ts; |
d3eeeabc | 60 | unsigned long done_secs = 0; |
ebac4655 | 61 | |
cdd18ad8 | 62 | static struct fio_mutex *startup_mutex; |
d2bbb8d9 | 63 | static struct fio_mutex *writeout_mutex; |
6ce15a32 | 64 | static volatile int fio_abort; |
437c9b71 | 65 | static int exit_value; |
a5b01f1b | 66 | static timer_t ival_timer; |
be4ecfdf | 67 | static pthread_t gtod_thread; |
dae5341c | 68 | static struct flist_head *cgroup_list; |
6adb38a1 | 69 | static char *cgroup_mnt; |
ebac4655 | 70 | |
bb3884d8 JA |
71 | struct io_log *agg_io_log[2]; |
72 | ||
ebac4655 | 73 | #define TERMINATE_ALL (-1) |
75154845 | 74 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 75 | |
b29ee5b3 | 76 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 77 | { |
d8fab1b6 JA |
78 | if (td->runstate == runstate) |
79 | return; | |
80 | ||
5921e80c JA |
81 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
82 | td->runstate, runstate); | |
6ce15a32 JA |
83 | td->runstate = runstate; |
84 | } | |
85 | ||
390c40e2 | 86 | static void terminate_threads(int group_id) |
ebac4655 | 87 | { |
34572e28 | 88 | struct thread_data *td; |
ebac4655 JA |
89 | int i; |
90 | ||
6f88bcb0 JA |
91 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
92 | ||
34572e28 | 93 | for_each_td(td, i) { |
ebac4655 | 94 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 95 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 96 | td->o.name, (int) td->pid); |
ad830ed7 JA |
97 | td->terminate = 1; |
98 | td->o.start_delay = 0; | |
99 | ||
7a93ab19 JA |
100 | /* |
101 | * if the thread is running, just let it exit | |
102 | */ | |
103 | if (td->runstate < TD_RUNNING) | |
03e20d68 | 104 | kill(td->pid, SIGTERM); |
f63f7f3b JA |
105 | else { |
106 | struct ioengine_ops *ops = td->io_ops; | |
107 | ||
03e20d68 BC |
108 | if (ops && (ops->flags & FIO_SIGTERM)) |
109 | kill(td->pid, SIGTERM); | |
f63f7f3b | 110 | } |
ebac4655 JA |
111 | } |
112 | } | |
113 | } | |
114 | ||
ccb0fa24 JA |
115 | static void status_timer_arm(void) |
116 | { | |
a5b01f1b JA |
117 | struct itimerspec value; |
118 | ||
119 | value.it_value.tv_sec = 0; | |
120 | value.it_value.tv_nsec = DISK_UTIL_MSEC * 1000000; | |
121 | value.it_interval.tv_sec = 0; | |
122 | value.it_interval.tv_nsec = DISK_UTIL_MSEC * 1000000; | |
123 | ||
124 | timer_settime(ival_timer, 0, &value, NULL); | |
ccb0fa24 JA |
125 | } |
126 | ||
a5b01f1b | 127 | static void ival_fn(union sigval sig) |
ebac4655 | 128 | { |
697a606c | 129 | if (threads) { |
5ec10eaa | 130 | update_io_ticks(); |
5ec10eaa | 131 | print_thread_status(); |
697a606c JA |
132 | } |
133 | } | |
134 | ||
5f8f9726 JA |
135 | /* |
136 | * Happens on thread runs with ctrl-c, ignore our own SIGQUIT | |
137 | */ | |
138 | static void sig_quit(int sig) | |
139 | { | |
140 | } | |
141 | ||
697a606c JA |
142 | static void sig_int(int sig) |
143 | { | |
144 | if (threads) { | |
4ceb30d4 | 145 | log_info("\nfio: terminating on signal %d\n", sig); |
5ec10eaa JA |
146 | fflush(stdout); |
147 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
148 | } |
149 | } | |
150 | ||
a5b01f1b JA |
151 | static void posix_timer_teardown(void) |
152 | { | |
153 | timer_delete(ival_timer); | |
154 | } | |
155 | ||
156 | static void posix_timer_setup(void) | |
157 | { | |
158 | struct sigevent evt; | |
159 | ||
160 | memset(&evt, 0, sizeof(evt)); | |
161 | evt.sigev_notify = SIGEV_THREAD; | |
162 | evt.sigev_notify_function = ival_fn; | |
163 | ||
03e20d68 | 164 | #ifndef __CYGWIN__ |
a5b01f1b JA |
165 | if (timer_create(CLOCK_MONOTONIC, &evt, &ival_timer) < 0) |
166 | perror("timer_create"); | |
03e20d68 BC |
167 | #else /* Windows (and thus Cygwin) doesn't have a monotonic clock */ |
168 | if (timer_create(CLOCK_REALTIME, &evt, &ival_timer) < 0) | |
169 | perror("timer_create"); | |
170 | #endif | |
171 | ||
a5b01f1b JA |
172 | } |
173 | ||
697a606c JA |
174 | static void set_sig_handlers(void) |
175 | { | |
176 | struct sigaction act; | |
177 | ||
697a606c JA |
178 | memset(&act, 0, sizeof(act)); |
179 | act.sa_handler = sig_int; | |
180 | act.sa_flags = SA_RESTART; | |
181 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 182 | |
5f8f9726 JA |
183 | memset(&act, 0, sizeof(act)); |
184 | act.sa_handler = sig_quit; | |
185 | act.sa_flags = SA_RESTART; | |
03e20d68 | 186 | sigaction(SIGTERM, &act, NULL); |
697a606c JA |
187 | } |
188 | ||
906c8d75 JA |
189 | /* |
190 | * Check if we are above the minimum rate given. | |
191 | */ | |
581e7141 JA |
192 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
193 | enum td_ddir ddir) | |
ebac4655 | 194 | { |
0904200b | 195 | unsigned long long bytes = 0; |
4e991c23 | 196 | unsigned long iops = 0; |
ebac4655 JA |
197 | unsigned long spent; |
198 | unsigned long rate; | |
581e7141 JA |
199 | unsigned int ratemin = 0; |
200 | unsigned int rate_iops = 0; | |
201 | unsigned int rate_iops_min = 0; | |
ebac4655 | 202 | |
ff58fced JA |
203 | assert(ddir_rw(ddir)); |
204 | ||
d982d873 JA |
205 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
206 | return 0; | |
207 | ||
ebac4655 JA |
208 | /* |
209 | * allow a 2 second settle period in the beginning | |
210 | */ | |
211 | if (mtime_since(&td->start, now) < 2000) | |
212 | return 0; | |
213 | ||
581e7141 JA |
214 | iops += td->io_blocks[ddir]; |
215 | bytes += td->this_io_bytes[ddir]; | |
216 | ratemin += td->o.ratemin[ddir]; | |
217 | rate_iops += td->o.rate_iops[ddir]; | |
218 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 219 | |
ebac4655 JA |
220 | /* |
221 | * if rate blocks is set, sample is running | |
222 | */ | |
581e7141 JA |
223 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
224 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 225 | if (spent < td->o.ratecycle) |
ebac4655 JA |
226 | return 0; |
227 | ||
581e7141 | 228 | if (td->o.rate[ddir]) { |
4e991c23 JA |
229 | /* |
230 | * check bandwidth specified rate | |
231 | */ | |
581e7141 | 232 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 233 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 234 | ratemin); |
4e991c23 JA |
235 | return 1; |
236 | } else { | |
581e7141 JA |
237 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
238 | if (rate < ratemin || | |
239 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 240 | log_err("%s: min rate %u not met, got" |
b22989b9 | 241 | " %luKB/sec\n", td->o.name, |
581e7141 | 242 | ratemin, rate); |
4e991c23 JA |
243 | return 1; |
244 | } | |
245 | } | |
413dd459 | 246 | } else { |
4e991c23 JA |
247 | /* |
248 | * checks iops specified rate | |
249 | */ | |
581e7141 | 250 | if (iops < rate_iops) { |
5ec10eaa | 251 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 252 | td->o.name, rate_iops); |
413dd459 | 253 | return 1; |
4e991c23 | 254 | } else { |
581e7141 JA |
255 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
256 | if (rate < rate_iops_min || | |
257 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
258 | log_err("%s: min iops rate %u not met," |
259 | " got %lu\n", td->o.name, | |
581e7141 | 260 | rate_iops_min, rate); |
4e991c23 | 261 | } |
413dd459 | 262 | } |
ebac4655 JA |
263 | } |
264 | } | |
265 | ||
581e7141 JA |
266 | td->rate_bytes[ddir] = bytes; |
267 | td->rate_blocks[ddir] = iops; | |
268 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
269 | return 0; |
270 | } | |
271 | ||
581e7141 JA |
272 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
273 | unsigned long *bytes_done) | |
274 | { | |
275 | int ret = 0; | |
276 | ||
277 | if (bytes_done[0]) | |
278 | ret |= __check_min_rate(td, now, 0); | |
279 | if (bytes_done[1]) | |
280 | ret |= __check_min_rate(td, now, 1); | |
281 | ||
282 | return ret; | |
283 | } | |
284 | ||
ebac4655 JA |
285 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
286 | { | |
2dc1bbeb | 287 | if (!td->o.timeout) |
ebac4655 | 288 | return 0; |
2dc1bbeb | 289 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
290 | return 1; |
291 | ||
292 | return 0; | |
293 | } | |
294 | ||
906c8d75 JA |
295 | /* |
296 | * When job exits, we can cancel the in-flight IO if we are using async | |
297 | * io. Attempt to do so. | |
298 | */ | |
ebac4655 JA |
299 | static void cleanup_pending_aio(struct thread_data *td) |
300 | { | |
01743ee1 | 301 | struct flist_head *entry, *n; |
ebac4655 JA |
302 | struct io_u *io_u; |
303 | int r; | |
304 | ||
305 | /* | |
306 | * get immediately available events, if any | |
307 | */ | |
581e7141 | 308 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
309 | if (r < 0) |
310 | return; | |
ebac4655 JA |
311 | |
312 | /* | |
313 | * now cancel remaining active events | |
314 | */ | |
2866c82d | 315 | if (td->io_ops->cancel) { |
01743ee1 JA |
316 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
317 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 318 | |
0c6e7517 JA |
319 | /* |
320 | * if the io_u isn't in flight, then that generally | |
321 | * means someone leaked an io_u. complain but fix | |
322 | * it up, so we don't stall here. | |
323 | */ | |
324 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
325 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 326 | put_io_u(td, io_u); |
0c6e7517 JA |
327 | } else { |
328 | r = td->io_ops->cancel(td, io_u); | |
329 | if (!r) | |
330 | put_io_u(td, io_u); | |
331 | } | |
ebac4655 JA |
332 | } |
333 | } | |
334 | ||
97601024 | 335 | if (td->cur_depth) |
581e7141 | 336 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
337 | } |
338 | ||
858a3d47 JA |
339 | /* |
340 | * Helper to handle the final sync of a file. Works just like the normal | |
341 | * io path, just does everything sync. | |
342 | */ | |
343 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
344 | { | |
345 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
346 | int ret; |
347 | ||
348 | if (!io_u) | |
349 | return 1; | |
350 | ||
351 | io_u->ddir = DDIR_SYNC; | |
352 | io_u->file = f; | |
353 | ||
354 | if (td_io_prep(td, io_u)) { | |
355 | put_io_u(td, io_u); | |
356 | return 1; | |
357 | } | |
358 | ||
755200a3 | 359 | requeue: |
858a3d47 | 360 | ret = td_io_queue(td, io_u); |
36167d82 | 361 | if (ret < 0) { |
e1161c32 | 362 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 363 | put_io_u(td, io_u); |
858a3d47 | 364 | return 1; |
36167d82 | 365 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 366 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 367 | return 1; |
36167d82 JA |
368 | } else if (ret == FIO_Q_COMPLETED) { |
369 | if (io_u->error) { | |
e1161c32 | 370 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
371 | return 1; |
372 | } | |
858a3d47 | 373 | |
581e7141 | 374 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 375 | return 1; |
755200a3 JA |
376 | } else if (ret == FIO_Q_BUSY) { |
377 | if (td_io_commit(td)) | |
378 | return 1; | |
379 | goto requeue; | |
858a3d47 JA |
380 | } |
381 | ||
382 | return 0; | |
383 | } | |
384 | ||
993bf48b JA |
385 | static inline void update_tv_cache(struct thread_data *td) |
386 | { | |
387 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
388 | fio_gettime(&td->tv_cache, NULL); | |
389 | } | |
390 | ||
f2bba182 RR |
391 | static int break_on_this_error(struct thread_data *td, int *retptr) |
392 | { | |
393 | int ret = *retptr; | |
394 | ||
395 | if (ret < 0 || td->error) { | |
396 | int err; | |
397 | ||
1ec99eea JA |
398 | if (!td->o.continue_on_error) |
399 | return 1; | |
f2bba182 RR |
400 | |
401 | if (ret < 0) | |
402 | err = -ret; | |
403 | else | |
404 | err = td->error; | |
405 | ||
f2bba182 RR |
406 | if (td_non_fatal_error(err)) { |
407 | /* | |
408 | * Continue with the I/Os in case of | |
409 | * a non fatal error. | |
410 | */ | |
8a4f8a24 | 411 | update_error_count(td, err); |
f2bba182 RR |
412 | td_clear_error(td); |
413 | *retptr = 0; | |
414 | return 0; | |
8a4f8a24 RR |
415 | } else if (td->o.fill_device && err == ENOSPC) { |
416 | /* | |
417 | * We expect to hit this error if | |
418 | * fill_device option is set. | |
419 | */ | |
420 | td_clear_error(td); | |
421 | td->terminate = 1; | |
422 | return 1; | |
f2bba182 RR |
423 | } else { |
424 | /* | |
425 | * Stop the I/O in case of a fatal | |
426 | * error. | |
427 | */ | |
8a4f8a24 | 428 | update_error_count(td, err); |
f2bba182 RR |
429 | return 1; |
430 | } | |
431 | } | |
432 | ||
433 | return 0; | |
434 | } | |
435 | ||
906c8d75 | 436 | /* |
34403fb1 | 437 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
438 | * reads the blocks back in, and checks the crc/md5 of the data. |
439 | */ | |
1e97cce9 | 440 | static void do_verify(struct thread_data *td) |
ebac4655 | 441 | { |
53cdc686 | 442 | struct fio_file *f; |
36167d82 | 443 | struct io_u *io_u; |
af52b345 JA |
444 | int ret, min_events; |
445 | unsigned int i; | |
e5b401d4 | 446 | |
008618a8 JA |
447 | dprint(FD_VERIFY, "starting loop\n"); |
448 | ||
e5b401d4 JA |
449 | /* |
450 | * sync io first and invalidate cache, to make sure we really | |
451 | * read from disk. | |
452 | */ | |
453 | for_each_file(td, f, i) { | |
d6aed795 | 454 | if (!fio_file_open(f)) |
3d7b485f | 455 | continue; |
b2fdda43 JA |
456 | if (fio_io_sync(td, f)) |
457 | break; | |
458 | if (file_invalidate_cache(td, f)) | |
459 | break; | |
e5b401d4 | 460 | } |
ebac4655 | 461 | |
b2fdda43 JA |
462 | if (td->error) |
463 | return; | |
464 | ||
ebac4655 JA |
465 | td_set_runstate(td, TD_VERIFYING); |
466 | ||
36167d82 JA |
467 | io_u = NULL; |
468 | while (!td->terminate) { | |
4950421a | 469 | int ret2, full; |
5451792e | 470 | |
993bf48b JA |
471 | update_tv_cache(td); |
472 | ||
473 | if (runtime_exceeded(td, &td->tv_cache)) { | |
90a87d4b | 474 | td->terminate = 1; |
02bcaa8c | 475 | break; |
069c2918 | 476 | } |
02bcaa8c | 477 | |
dcb4830d RR |
478 | io_u = __get_io_u(td); |
479 | if (!io_u) | |
480 | break; | |
481 | ||
069c2918 JA |
482 | if (get_next_verify(td, io_u)) { |
483 | put_io_u(td, io_u); | |
ebac4655 | 484 | break; |
069c2918 | 485 | } |
ebac4655 | 486 | |
069c2918 JA |
487 | if (td_io_prep(td, io_u)) { |
488 | put_io_u(td, io_u); | |
53cdc686 | 489 | break; |
069c2918 | 490 | } |
d7762cf8 | 491 | |
e8462bd8 JA |
492 | if (td->o.verify_async) |
493 | io_u->end_io = verify_io_u_async; | |
494 | else | |
495 | io_u->end_io = verify_io_u; | |
53cdc686 | 496 | |
11786802 | 497 | ret = td_io_queue(td, io_u); |
36167d82 JA |
498 | switch (ret) { |
499 | case FIO_Q_COMPLETED: | |
f2bba182 | 500 | if (io_u->error) { |
22819ec2 | 501 | ret = -io_u->error; |
f2bba182 RR |
502 | clear_io_u(td, io_u); |
503 | } else if (io_u->resid) { | |
36167d82 | 504 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 505 | struct fio_file *f = io_u->file; |
ebac4655 | 506 | |
9e9d164e JA |
507 | /* |
508 | * zero read, fail | |
509 | */ | |
510 | if (!bytes) { | |
41d9f0fd | 511 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
512 | put_io_u(td, io_u); |
513 | break; | |
514 | } | |
8400d9b2 | 515 | |
36167d82 JA |
516 | io_u->xfer_buflen = io_u->resid; |
517 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
518 | io_u->offset += bytes; |
519 | ||
ff58fced JA |
520 | if (ddir_rw(io_u->ddir)) |
521 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 522 | |
d460eb31 | 523 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
524 | goto sync_done; |
525 | ||
11786802 JA |
526 | requeue_io_u(td, &io_u); |
527 | } else { | |
8400d9b2 | 528 | sync_done: |
581e7141 | 529 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
530 | if (ret < 0) |
531 | break; | |
36167d82 | 532 | } |
36167d82 JA |
533 | continue; |
534 | case FIO_Q_QUEUED: | |
535 | break; | |
755200a3 JA |
536 | case FIO_Q_BUSY: |
537 | requeue_io_u(td, &io_u); | |
5451792e JA |
538 | ret2 = td_io_commit(td); |
539 | if (ret2 < 0) | |
540 | ret = ret2; | |
755200a3 | 541 | break; |
36167d82 JA |
542 | default: |
543 | assert(ret < 0); | |
e1161c32 | 544 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
545 | break; |
546 | } | |
547 | ||
f2bba182 | 548 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
549 | break; |
550 | ||
ebac4655 | 551 | /* |
3af6ef39 JA |
552 | * if we can queue more, do so. but check if there are |
553 | * completed io_u's first. | |
ebac4655 | 554 | */ |
4950421a JA |
555 | full = queue_full(td) || ret == FIO_Q_BUSY; |
556 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
557 | min_events = min(td->o.iodepth_batch_complete, |
558 | td->cur_depth); | |
4950421a | 559 | if (full && !min_events) |
838bc709 | 560 | min_events = 1; |
e916b390 | 561 | |
4950421a JA |
562 | do { |
563 | /* | |
564 | * Reap required number of io units, if any, | |
565 | * and do the verification on them through | |
566 | * the callback handler | |
567 | */ | |
581e7141 | 568 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
569 | ret = -1; |
570 | break; | |
571 | } | |
572 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
573 | } | |
574 | if (ret < 0) | |
ebac4655 | 575 | break; |
36167d82 | 576 | } |
ebac4655 | 577 | |
c01c0395 JA |
578 | if (!td->error) { |
579 | min_events = td->cur_depth; | |
580 | ||
581 | if (min_events) | |
581e7141 | 582 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 583 | } else |
ebac4655 JA |
584 | cleanup_pending_aio(td); |
585 | ||
586 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
587 | |
588 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
589 | } |
590 | ||
32cd46a0 | 591 | /* |
906c8d75 | 592 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
593 | * and reaps them, checking for rate and errors along the way. |
594 | */ | |
ebac4655 JA |
595 | static void do_io(struct thread_data *td) |
596 | { | |
af52b345 JA |
597 | unsigned int i; |
598 | int ret = 0; | |
ebac4655 | 599 | |
b29ee5b3 JA |
600 | if (in_ramp_time(td)) |
601 | td_set_runstate(td, TD_RAMP); | |
602 | else | |
603 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 604 | |
457bf399 | 605 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
a917a8b3 | 606 | (!flist_empty(&td->trim_list)) || |
457bf399 | 607 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { |
97601024 | 608 | struct timeval comp_time; |
581e7141 | 609 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 610 | int min_evts = 0; |
ebac4655 | 611 | struct io_u *io_u; |
4950421a | 612 | int ret2, full; |
ebac4655 JA |
613 | |
614 | if (td->terminate) | |
615 | break; | |
616 | ||
993bf48b | 617 | update_tv_cache(td); |
97601024 | 618 | |
993bf48b | 619 | if (runtime_exceeded(td, &td->tv_cache)) { |
90a87d4b | 620 | td->terminate = 1; |
97601024 JA |
621 | break; |
622 | } | |
36167d82 | 623 | |
dcb4830d RR |
624 | io_u = get_io_u(td); |
625 | if (!io_u) | |
626 | break; | |
627 | ||
b67740d3 JA |
628 | /* |
629 | * Add verification end_io handler, if asked to verify | |
630 | * a previously written file. | |
631 | */ | |
715d2b3e JA |
632 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
633 | !td_rw(td)) { | |
e8462bd8 JA |
634 | if (td->o.verify_async) |
635 | io_u->end_io = verify_io_u_async; | |
636 | else | |
637 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 638 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
639 | } else if (in_ramp_time(td)) |
640 | td_set_runstate(td, TD_RAMP); | |
641 | else | |
d8fab1b6 | 642 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 643 | |
11786802 | 644 | ret = td_io_queue(td, io_u); |
36167d82 JA |
645 | switch (ret) { |
646 | case FIO_Q_COMPLETED: | |
f2bba182 | 647 | if (io_u->error) { |
5451792e | 648 | ret = -io_u->error; |
f2bba182 RR |
649 | clear_io_u(td, io_u); |
650 | } else if (io_u->resid) { | |
36167d82 | 651 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 652 | struct fio_file *f = io_u->file; |
36167d82 | 653 | |
9e9d164e JA |
654 | /* |
655 | * zero read, fail | |
656 | */ | |
657 | if (!bytes) { | |
41d9f0fd | 658 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
659 | put_io_u(td, io_u); |
660 | break; | |
661 | } | |
662 | ||
cec6b55d | 663 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 664 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
665 | io_u->offset += bytes; |
666 | ||
ff58fced JA |
667 | if (ddir_rw(io_u->ddir)) |
668 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 669 | |
d460eb31 | 670 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
671 | goto sync_done; |
672 | ||
11786802 JA |
673 | requeue_io_u(td, &io_u); |
674 | } else { | |
5a7c5680 | 675 | sync_done: |
581e7141 JA |
676 | if (__should_check_rate(td, 0) || |
677 | __should_check_rate(td, 1)) | |
993bf48b JA |
678 | fio_gettime(&comp_time, NULL); |
679 | ||
581e7141 JA |
680 | ret = io_u_sync_complete(td, io_u, bytes_done); |
681 | if (ret < 0) | |
682 | break; | |
cec6b55d | 683 | } |
36167d82 JA |
684 | break; |
685 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
686 | /* |
687 | * if the engine doesn't have a commit hook, | |
688 | * the io_u is really queued. if it does have such | |
689 | * a hook, it has to call io_u_queued() itself. | |
690 | */ | |
691 | if (td->io_ops->commit == NULL) | |
692 | io_u_queued(td, io_u); | |
36167d82 | 693 | break; |
755200a3 JA |
694 | case FIO_Q_BUSY: |
695 | requeue_io_u(td, &io_u); | |
5451792e JA |
696 | ret2 = td_io_commit(td); |
697 | if (ret2 < 0) | |
698 | ret = ret2; | |
755200a3 | 699 | break; |
36167d82 JA |
700 | default: |
701 | assert(ret < 0); | |
702 | put_io_u(td, io_u); | |
703 | break; | |
ebac4655 JA |
704 | } |
705 | ||
f2bba182 | 706 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
707 | break; |
708 | ||
97601024 JA |
709 | /* |
710 | * See if we need to complete some commands | |
711 | */ | |
4950421a JA |
712 | full = queue_full(td) || ret == FIO_Q_BUSY; |
713 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
714 | min_evts = min(td->o.iodepth_batch_complete, |
715 | td->cur_depth); | |
4950421a | 716 | if (full && !min_evts) |
36167d82 | 717 | min_evts = 1; |
4950421a | 718 | |
581e7141 JA |
719 | if (__should_check_rate(td, 0) || |
720 | __should_check_rate(td, 1)) | |
993bf48b | 721 | fio_gettime(&comp_time, NULL); |
4950421a JA |
722 | |
723 | do { | |
581e7141 JA |
724 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
725 | if (ret < 0) | |
4950421a JA |
726 | break; |
727 | ||
4950421a | 728 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
729 | } |
730 | ||
4950421a JA |
731 | if (ret < 0) |
732 | break; | |
581e7141 | 733 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
734 | continue; |
735 | ||
581e7141 JA |
736 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
737 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
738 | if (exitall_on_terminate) |
739 | terminate_threads(td->groupid); | |
740 | td_verror(td, EIO, "check_min_rate"); | |
741 | break; | |
742 | } | |
ebac4655 JA |
743 | } |
744 | ||
2dc1bbeb | 745 | if (td->o.thinktime) { |
9c1f7434 JA |
746 | unsigned long long b; |
747 | ||
748 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 749 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
750 | int left; |
751 | ||
2dc1bbeb | 752 | if (td->o.thinktime_spin) |
6dd6f2cd | 753 | usec_spin(td->o.thinktime_spin); |
48097d5c | 754 | |
2dc1bbeb | 755 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
756 | if (left) |
757 | usec_sleep(td, left); | |
758 | } | |
9c1f7434 | 759 | } |
ebac4655 JA |
760 | } |
761 | ||
d708ac5e JA |
762 | if (td->trim_entries) |
763 | printf("trim entries %ld\n", td->trim_entries); | |
a917a8b3 | 764 | |
aa31f1f1 SL |
765 | if (td->o.fill_device && td->error == ENOSPC) { |
766 | td->error = 0; | |
767 | td->terminate = 1; | |
768 | } | |
4d2413c6 | 769 | if (!td->error) { |
3d7c391d JA |
770 | struct fio_file *f; |
771 | ||
c01c0395 JA |
772 | i = td->cur_depth; |
773 | if (i) | |
581e7141 | 774 | ret = io_u_queued_complete(td, i, NULL); |
ebac4655 | 775 | |
2dc1bbeb | 776 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 777 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
778 | |
779 | for_each_file(td, f, i) { | |
d6aed795 | 780 | if (!fio_file_open(f)) |
3d7b485f | 781 | continue; |
858a3d47 | 782 | fio_io_sync(td, f); |
3d7b485f | 783 | } |
84585003 | 784 | } |
c01c0395 JA |
785 | } else |
786 | cleanup_pending_aio(td); | |
2ba1c290 JA |
787 | |
788 | /* | |
789 | * stop job if we failed doing any IO | |
790 | */ | |
791 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
792 | td->done = 1; | |
ebac4655 JA |
793 | } |
794 | ||
ebac4655 JA |
795 | static void cleanup_io_u(struct thread_data *td) |
796 | { | |
01743ee1 | 797 | struct flist_head *entry, *n; |
ebac4655 JA |
798 | struct io_u *io_u; |
799 | ||
01743ee1 JA |
800 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
801 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 802 | |
01743ee1 | 803 | flist_del(&io_u->list); |
ebac4655 JA |
804 | free(io_u); |
805 | } | |
806 | ||
2f9ade3c | 807 | free_io_mem(td); |
ebac4655 JA |
808 | } |
809 | ||
810 | static int init_io_u(struct thread_data *td) | |
811 | { | |
812 | struct io_u *io_u; | |
a00735e6 | 813 | unsigned int max_bs; |
eb7ccf38 | 814 | int cl_align, i, max_units; |
ebac4655 JA |
815 | char *p; |
816 | ||
1e3c09af | 817 | max_units = td->o.iodepth; |
2dc1bbeb | 818 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
819 | td->orig_buffer_size = (unsigned long long) max_bs |
820 | * (unsigned long long) max_units; | |
821 | ||
822 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
823 | unsigned long bs; | |
74b025b0 | 824 | |
5ec10eaa JA |
825 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
826 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
827 | } | |
ebac4655 | 828 | |
c3990645 JA |
829 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
830 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
831 | return 1; | |
832 | } | |
833 | ||
2f9ade3c JA |
834 | if (allocate_io_mem(td)) |
835 | return 1; | |
ebac4655 | 836 | |
1c04bf78 JA |
837 | if (td->o.odirect || td->o.mem_align || |
838 | (td->io_ops->flags & FIO_RAWIO)) | |
d529ee19 | 839 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
840 | else |
841 | p = td->orig_buffer; | |
842 | ||
eb7ccf38 JA |
843 | cl_align = os_cache_line_size(); |
844 | ||
ebac4655 | 845 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
846 | void *ptr; |
847 | ||
db1cd90b JA |
848 | if (td->terminate) |
849 | return 1; | |
eb7ccf38 JA |
850 | |
851 | if (posix_memalign(&ptr, cl_align, sizeof(*io_u))) { | |
852 | log_err("fio: posix_memalign=%s\n", strerror(errno)); | |
853 | break; | |
854 | } | |
855 | ||
856 | io_u = ptr; | |
ebac4655 | 857 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 858 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 859 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 860 | |
b4c5e1ac JA |
861 | if (!(td->io_ops->flags & FIO_NOIO)) { |
862 | io_u->buf = p + max_bs * i; | |
d529ee19 | 863 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 864 | |
5973cafb JA |
865 | if (td_write(td) && !td->o.refill_buffers) |
866 | io_u_fill_buffer(td, io_u, max_bs); | |
cbe8d756 RR |
867 | else if (td_write(td) && td->o.verify_pattern_bytes) { |
868 | /* | |
869 | * Fill the buffer with the pattern if we are | |
870 | * going to be doing writes. | |
871 | */ | |
872 | fill_pattern(td, io_u->buf, max_bs, io_u); | |
873 | } | |
b4c5e1ac | 874 | } |
6b9cea23 | 875 | |
b1ff3403 | 876 | io_u->index = i; |
0c6e7517 | 877 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 878 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
879 | } |
880 | ||
881 | return 0; | |
882 | } | |
883 | ||
da86774e JA |
884 | static int switch_ioscheduler(struct thread_data *td) |
885 | { | |
886 | char tmp[256], tmp2[128]; | |
887 | FILE *f; | |
888 | int ret; | |
889 | ||
ba0fbe10 | 890 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
891 | return 0; |
892 | ||
da86774e JA |
893 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
894 | ||
895 | f = fopen(tmp, "r+"); | |
896 | if (!f) { | |
cbf5c1d7 | 897 | if (errno == ENOENT) { |
5ec10eaa JA |
898 | log_err("fio: os or kernel doesn't support IO scheduler" |
899 | " switching\n"); | |
cbf5c1d7 JA |
900 | return 0; |
901 | } | |
902 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
903 | return 1; |
904 | } | |
905 | ||
906 | /* | |
907 | * Set io scheduler. | |
908 | */ | |
2dc1bbeb | 909 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 910 | if (ferror(f) || ret != 1) { |
e1161c32 | 911 | td_verror(td, errno, "fwrite"); |
da86774e JA |
912 | fclose(f); |
913 | return 1; | |
914 | } | |
915 | ||
916 | rewind(f); | |
917 | ||
918 | /* | |
919 | * Read back and check that the selected scheduler is now the default. | |
920 | */ | |
921 | ret = fread(tmp, 1, sizeof(tmp), f); | |
922 | if (ferror(f) || ret < 0) { | |
e1161c32 | 923 | td_verror(td, errno, "fread"); |
da86774e JA |
924 | fclose(f); |
925 | return 1; | |
926 | } | |
927 | ||
2dc1bbeb | 928 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 929 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 930 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 931 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
932 | fclose(f); |
933 | return 1; | |
934 | } | |
935 | ||
936 | fclose(f); | |
937 | return 0; | |
938 | } | |
939 | ||
492158cf JA |
940 | static int keep_running(struct thread_data *td) |
941 | { | |
942 | unsigned long long io_done; | |
943 | ||
20e354ef JA |
944 | if (td->done) |
945 | return 0; | |
492158cf JA |
946 | if (td->o.time_based) |
947 | return 1; | |
948 | if (td->o.loops) { | |
949 | td->o.loops--; | |
950 | return 1; | |
951 | } | |
952 | ||
5ec10eaa JA |
953 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
954 | + td->io_skip_bytes; | |
492158cf JA |
955 | if (io_done < td->o.size) |
956 | return 1; | |
957 | ||
958 | return 0; | |
959 | } | |
960 | ||
b29ee5b3 | 961 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 962 | { |
756867bd | 963 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 964 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 965 | td->zone_bytes = 0; |
581e7141 JA |
966 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
967 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 968 | |
c1324df1 JA |
969 | td->last_was_sync = 0; |
970 | ||
2ba1c290 JA |
971 | /* |
972 | * reset file done count if we are to start over | |
973 | */ | |
974 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 975 | td->nr_done_files = 0; |
5bfc35d7 JA |
976 | |
977 | /* | |
978 | * Set the same seed to get repeatable runs | |
979 | */ | |
980 | td_fill_rand_seeds(td); | |
b29ee5b3 JA |
981 | } |
982 | ||
983 | void reset_all_stats(struct thread_data *td) | |
984 | { | |
985 | struct timeval tv; | |
986 | int i; | |
987 | ||
988 | reset_io_counters(td); | |
989 | ||
990 | for (i = 0; i < 2; i++) { | |
991 | td->io_bytes[i] = 0; | |
992 | td->io_blocks[i] = 0; | |
993 | td->io_issues[i] = 0; | |
994 | td->ts.total_io_u[i] = 0; | |
995 | } | |
996 | ||
997 | fio_gettime(&tv, NULL); | |
cda99fa0 SSP |
998 | td->ts.runtime[0] = 0; |
999 | td->ts.runtime[1] = 0; | |
b29ee5b3 JA |
1000 | memcpy(&td->epoch, &tv, sizeof(tv)); |
1001 | memcpy(&td->start, &tv, sizeof(tv)); | |
1002 | } | |
1003 | ||
df9c26b1 | 1004 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
1005 | { |
1006 | struct fio_file *f; | |
1007 | unsigned int i; | |
b29ee5b3 JA |
1008 | |
1009 | reset_io_counters(td); | |
281f9b63 | 1010 | |
24ffd2c2 | 1011 | close_files(td); |
df9c26b1 | 1012 | for_each_file(td, f, i) |
d6aed795 | 1013 | fio_file_clear_done(f); |
ebac4655 JA |
1014 | } |
1015 | ||
398e8c4d JA |
1016 | static int exec_string(const char *string) |
1017 | { | |
1018 | int ret, newlen = strlen(string) + 1 + 8; | |
1019 | char *str; | |
1020 | ||
1021 | str = malloc(newlen); | |
1022 | sprintf(str, "sh -c %s", string); | |
1023 | ||
1024 | ret = system(str); | |
1025 | if (ret == -1) | |
1026 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1027 | ||
1028 | free(str); | |
1029 | return ret; | |
1030 | } | |
1031 | ||
906c8d75 JA |
1032 | /* |
1033 | * Entry point for the thread based jobs. The process based jobs end up | |
1034 | * here as well, after a little setup. | |
1035 | */ | |
ebac4655 JA |
1036 | static void *thread_main(void *data) |
1037 | { | |
cda99fa0 | 1038 | unsigned long long elapsed; |
ebac4655 | 1039 | struct thread_data *td = data; |
e8462bd8 | 1040 | pthread_condattr_t attr; |
a978ba68 | 1041 | int clear_state; |
ebac4655 | 1042 | |
2dc1bbeb | 1043 | if (!td->o.use_thread) |
ebac4655 JA |
1044 | setsid(); |
1045 | ||
1046 | td->pid = getpid(); | |
1047 | ||
5921e80c | 1048 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1049 | |
01743ee1 JA |
1050 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1051 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1052 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1053 | INIT_FLIST_HEAD(&td->io_log_list); | |
1054 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 | 1055 | INIT_FLIST_HEAD(&td->verify_list); |
0d29de83 | 1056 | INIT_FLIST_HEAD(&td->trim_list); |
e8462bd8 | 1057 | pthread_mutex_init(&td->io_u_lock, NULL); |
bb5d7d0b | 1058 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1059 | |
e8462bd8 JA |
1060 | pthread_condattr_init(&attr); |
1061 | pthread_cond_init(&td->verify_cond, &attr); | |
1062 | pthread_cond_init(&td->free_cond, &attr); | |
1063 | ||
db1cd90b | 1064 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1065 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1066 | fio_mutex_up(startup_mutex); |
29adda3c | 1067 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1068 | fio_mutex_down(td->mutex); |
29adda3c | 1069 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1070 | |
1071 | /* | |
cdd18ad8 | 1072 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1073 | * eating a file descriptor |
1074 | */ | |
cdd18ad8 | 1075 | fio_mutex_remove(td->mutex); |
db1cd90b | 1076 | |
e0b0d892 JA |
1077 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1078 | td_verror(td, errno, "setuid"); | |
1079 | goto err; | |
1080 | } | |
1081 | if (td->o.gid != -1U && setgid(td->o.gid)) { | |
1082 | td_verror(td, errno, "setgid"); | |
1083 | goto err; | |
1084 | } | |
1085 | ||
d84f8d49 JA |
1086 | /* |
1087 | * May alter parameters that init_io_u() will use, so we need to | |
1088 | * do this first. | |
1089 | */ | |
1090 | if (init_iolog(td)) | |
1091 | goto err; | |
1092 | ||
ebac4655 | 1093 | if (init_io_u(td)) |
db1cd90b | 1094 | goto err; |
ebac4655 | 1095 | |
e8462bd8 JA |
1096 | if (td->o.verify_async && verify_async_init(td)) |
1097 | goto err; | |
1098 | ||
1099 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
e1161c32 | 1100 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 1101 | goto err; |
ebac4655 JA |
1102 | } |
1103 | ||
f484470b JA |
1104 | /* |
1105 | * If we have a gettimeofday() thread, make sure we exclude that | |
1106 | * thread from this job | |
1107 | */ | |
be4ecfdf | 1108 | if (td->o.gtod_cpu) { |
be4ecfdf | 1109 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); |
e8462bd8 | 1110 | if (fio_setaffinity(td->pid, td->o.cpumask) == -1) { |
be4ecfdf JA |
1111 | td_verror(td, errno, "cpu_set_affinity"); |
1112 | goto err; | |
1113 | } | |
1114 | } | |
1115 | ||
ac684785 | 1116 | if (td->ioprio_set) { |
ebac4655 | 1117 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1118 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1119 | goto err; |
ebac4655 JA |
1120 | } |
1121 | } | |
1122 | ||
6adb38a1 | 1123 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1124 | goto err; |
1125 | ||
2dc1bbeb | 1126 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1127 | td_verror(td, errno, "nice"); |
db1cd90b | 1128 | goto err; |
b6f4d880 JA |
1129 | } |
1130 | ||
2dc1bbeb | 1131 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1132 | goto err; |
37f56873 | 1133 | |
2dc1bbeb | 1134 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1135 | goto err; |
1136 | ||
7d6c5283 JA |
1137 | if (td_io_init(td)) |
1138 | goto err; | |
1139 | ||
68727076 JA |
1140 | if (init_random_map(td)) |
1141 | goto err; | |
1142 | ||
2dc1bbeb | 1143 | if (td->o.exec_prerun) { |
398e8c4d | 1144 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1145 | goto err; |
1146 | } | |
4e0ba8af | 1147 | |
afad68f7 ZY |
1148 | if (td->o.pre_read) { |
1149 | if (pre_read_files(td) < 0) | |
1150 | goto err; | |
1151 | } | |
1152 | ||
69008999 | 1153 | fio_gettime(&td->epoch, NULL); |
756867bd | 1154 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 | 1155 | |
a978ba68 | 1156 | clear_state = 0; |
492158cf | 1157 | while (keep_running(td)) { |
02bcaa8c | 1158 | fio_gettime(&td->start, NULL); |
c5d3d790 JA |
1159 | memcpy(&td->ts.stat_sample_time[0], &td->start, |
1160 | sizeof(td->start)); | |
1161 | memcpy(&td->ts.stat_sample_time[1], &td->start, | |
1162 | sizeof(td->start)); | |
993bf48b | 1163 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1164 | |
cf2dd8d6 | 1165 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
5ec10eaa JA |
1166 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
1167 | sizeof(td->lastrate)); | |
ebac4655 | 1168 | |
df9c26b1 JA |
1169 | if (clear_state) |
1170 | clear_io_state(td); | |
a978ba68 | 1171 | |
ebac4655 JA |
1172 | prune_io_piece_log(td); |
1173 | ||
ba0fbe10 | 1174 | do_io(td); |
ebac4655 | 1175 | |
a978ba68 JA |
1176 | clear_state = 1; |
1177 | ||
38d77cae | 1178 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1179 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1180 | td->ts.runtime[DDIR_READ] += elapsed; |
38d77cae JA |
1181 | } |
1182 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1183 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1184 | td->ts.runtime[DDIR_WRITE] += elapsed; |
38d77cae | 1185 | } |
5ec10eaa | 1186 | |
ebac4655 JA |
1187 | if (td->error || td->terminate) |
1188 | break; | |
1189 | ||
e84c73a8 SL |
1190 | if (!td->o.do_verify || |
1191 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1192 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1193 | continue; |
1194 | ||
df9c26b1 | 1195 | clear_io_state(td); |
a978ba68 | 1196 | |
02bcaa8c | 1197 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1198 | |
1199 | do_verify(td); | |
1200 | ||
cda99fa0 | 1201 | td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1202 | |
1203 | if (td->error || td->terminate) | |
1204 | break; | |
1205 | } | |
1206 | ||
36dff966 | 1207 | update_rusage_stat(td); |
cda99fa0 SSP |
1208 | td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000; |
1209 | td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000; | |
756867bd JA |
1210 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1211 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1212 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1213 | ||
d2bbb8d9 | 1214 | fio_mutex_down(writeout_mutex); |
e3cedca7 | 1215 | if (td->ts.bw_log) { |
f484470b JA |
1216 | if (td->o.bw_log_file) { |
1217 | finish_log_named(td, td->ts.bw_log, | |
1218 | td->o.bw_log_file, "bw"); | |
1219 | } else | |
e3cedca7 JA |
1220 | finish_log(td, td->ts.bw_log, "bw"); |
1221 | } | |
02af0988 JA |
1222 | if (td->ts.lat_log) { |
1223 | if (td->o.lat_log_file) { | |
1224 | finish_log_named(td, td->ts.lat_log, | |
1225 | td->o.lat_log_file, "lat"); | |
1226 | } else | |
1227 | finish_log(td, td->ts.lat_log, "lat"); | |
1228 | } | |
e3cedca7 | 1229 | if (td->ts.slat_log) { |
f484470b JA |
1230 | if (td->o.lat_log_file) { |
1231 | finish_log_named(td, td->ts.slat_log, | |
1ca95738 | 1232 | td->o.lat_log_file, "slat"); |
f484470b | 1233 | } else |
e3cedca7 JA |
1234 | finish_log(td, td->ts.slat_log, "slat"); |
1235 | } | |
1236 | if (td->ts.clat_log) { | |
f484470b JA |
1237 | if (td->o.lat_log_file) { |
1238 | finish_log_named(td, td->ts.clat_log, | |
1239 | td->o.lat_log_file, "clat"); | |
1240 | } else | |
e3cedca7 JA |
1241 | finish_log(td, td->ts.clat_log, "clat"); |
1242 | } | |
d2bbb8d9 | 1243 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1244 | if (td->o.exec_postrun) |
1245 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1246 | |
1247 | if (exitall_on_terminate) | |
390c40e2 | 1248 | terminate_threads(td->groupid); |
ebac4655 JA |
1249 | |
1250 | err: | |
5bf13a5a | 1251 | if (td->error) |
4ceb30d4 | 1252 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1253 | td->verror); |
b4707ed7 JA |
1254 | |
1255 | if (td->o.verify_async) | |
1256 | verify_async_exit(td); | |
1257 | ||
24ffd2c2 | 1258 | close_and_free_files(td); |
2866c82d | 1259 | close_ioengine(td); |
ebac4655 | 1260 | cleanup_io_u(td); |
6adb38a1 | 1261 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1262 | |
d2ce18b5 JA |
1263 | if (td->o.cpumask_set) { |
1264 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1265 | ||
1266 | td_verror(td, ret, "fio_cpuset_exit"); | |
1267 | } | |
4e78e405 | 1268 | |
f29b25a3 JA |
1269 | /* |
1270 | * do this very late, it will log file closing as well | |
1271 | */ | |
1272 | if (td->o.write_iolog_file) | |
1273 | write_iolog_close(td); | |
1274 | ||
d23bb327 | 1275 | options_mem_free(td); |
ebac4655 | 1276 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1277 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1278 | } |
1279 | ||
906c8d75 JA |
1280 | /* |
1281 | * We cannot pass the td data into a forked process, so attach the td and | |
1282 | * pass it to the thread worker. | |
1283 | */ | |
a6418147 | 1284 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1285 | { |
1286 | struct thread_data *td; | |
a6418147 | 1287 | void *data, *ret; |
ebac4655 JA |
1288 | |
1289 | data = shmat(shmid, NULL, 0); | |
1290 | if (data == (void *) -1) { | |
a6418147 JA |
1291 | int __err = errno; |
1292 | ||
ebac4655 | 1293 | perror("shmat"); |
a6418147 | 1294 | return __err; |
ebac4655 JA |
1295 | } |
1296 | ||
1297 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1298 | ret = thread_main(td); |
ebac4655 | 1299 | shmdt(data); |
43d76807 | 1300 | return (int) (unsigned long) ret; |
ebac4655 JA |
1301 | } |
1302 | ||
906c8d75 JA |
1303 | /* |
1304 | * Run over the job map and reap the threads that have exited, if any. | |
1305 | */ | |
ebac4655 JA |
1306 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1307 | { | |
34572e28 | 1308 | struct thread_data *td; |
1f809d15 | 1309 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1310 | |
1311 | /* | |
1312 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1313 | */ | |
1f809d15 | 1314 | realthreads = pending = cputhreads = 0; |
34572e28 | 1315 | for_each_td(td, i) { |
3707f45b | 1316 | int flags = 0; |
a2f77c9f | 1317 | |
84585003 JA |
1318 | /* |
1319 | * ->io_ops is NULL for a thread that has closed its | |
1320 | * io engine | |
1321 | */ | |
ba0fbe10 | 1322 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1323 | cputhreads++; |
1f809d15 JA |
1324 | else |
1325 | realthreads++; | |
b990b5c0 | 1326 | |
a1dedc1f AC |
1327 | if (!td->pid) { |
1328 | pending++; | |
1329 | continue; | |
1330 | } | |
1331 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1332 | continue; |
2dc1bbeb | 1333 | if (td->o.use_thread) { |
3707f45b JA |
1334 | if (td->runstate == TD_EXITED) { |
1335 | td_set_runstate(td, TD_REAPED); | |
1336 | goto reaped; | |
1337 | } | |
1338 | continue; | |
1339 | } | |
a2f77c9f JA |
1340 | |
1341 | flags = WNOHANG; | |
1342 | if (td->runstate == TD_EXITED) | |
1343 | flags = 0; | |
1344 | ||
1345 | /* | |
1346 | * check if someone quit or got killed in an unusual way | |
1347 | */ | |
1348 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1349 | if (ret < 0) { |
a2f77c9f | 1350 | if (errno == ECHILD) { |
5921e80c JA |
1351 | log_err("fio: pid=%d disappeared %d\n", |
1352 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1353 | td_set_runstate(td, TD_REAPED); |
1354 | goto reaped; | |
1355 | } | |
1356 | perror("waitpid"); | |
1357 | } else if (ret == td->pid) { | |
1358 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1359 | int sig = WTERMSIG(status); |
1360 | ||
03e20d68 | 1361 | if (sig != SIGTERM) |
5ec10eaa | 1362 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1363 | (int) td->pid, sig); |
fab6aa71 JA |
1364 | td_set_runstate(td, TD_REAPED); |
1365 | goto reaped; | |
1366 | } | |
a2f77c9f JA |
1367 | if (WIFEXITED(status)) { |
1368 | if (WEXITSTATUS(status) && !td->error) | |
1369 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1370 | |
a2f77c9f JA |
1371 | td_set_runstate(td, TD_REAPED); |
1372 | goto reaped; | |
a6418147 JA |
1373 | } |
1374 | } | |
ebac4655 | 1375 | |
a2f77c9f JA |
1376 | /* |
1377 | * thread is not dead, continue | |
1378 | */ | |
e48676ba | 1379 | pending++; |
a2f77c9f | 1380 | continue; |
fab6aa71 | 1381 | reaped: |
ebac4655 | 1382 | (*nr_running)--; |
581e7141 JA |
1383 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1384 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1385 | if (!td->pid) |
1386 | pending--; | |
a2f77c9f JA |
1387 | |
1388 | if (td->error) | |
1389 | exit_value++; | |
d3eeeabc JA |
1390 | |
1391 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1392 | } |
b990b5c0 | 1393 | |
1f809d15 | 1394 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1395 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1396 | } |
1397 | ||
be4ecfdf JA |
1398 | static void *gtod_thread_main(void *data) |
1399 | { | |
1400 | fio_mutex_up(startup_mutex); | |
1401 | ||
1402 | /* | |
1403 | * As long as we have jobs around, update the clock. It would be nice | |
1404 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1405 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1406 | * it - we don't want to lose precision. | |
1407 | */ | |
1408 | while (threads) { | |
1409 | fio_gtod_update(); | |
1410 | nop; | |
1411 | } | |
1412 | ||
1413 | return NULL; | |
1414 | } | |
1415 | ||
1416 | static int fio_start_gtod_thread(void) | |
1417 | { | |
304a47c7 | 1418 | pthread_attr_t attr; |
f42c153d JA |
1419 | int ret; |
1420 | ||
304a47c7 VA |
1421 | pthread_attr_init(&attr); |
1422 | pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); | |
1423 | ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); | |
1424 | pthread_attr_destroy(&attr); | |
f42c153d JA |
1425 | if (ret) { |
1426 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1427 | return 1; |
1428 | } | |
f42c153d JA |
1429 | |
1430 | ret = pthread_detach(gtod_thread); | |
1431 | if (ret) { | |
1432 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1433 | return 1; |
1434 | } | |
1435 | ||
29adda3c | 1436 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1437 | fio_mutex_down(startup_mutex); |
29adda3c | 1438 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1439 | return 0; |
1440 | } | |
1441 | ||
906c8d75 JA |
1442 | /* |
1443 | * Main function for kicking off and reaping jobs, as needed. | |
1444 | */ | |
ebac4655 JA |
1445 | static void run_threads(void) |
1446 | { | |
ebac4655 JA |
1447 | struct thread_data *td; |
1448 | unsigned long spent; | |
1449 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1450 | |
2f9ade3c JA |
1451 | if (fio_pin_memory()) |
1452 | return; | |
ebac4655 | 1453 | |
be4ecfdf JA |
1454 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1455 | return; | |
1456 | ||
c6ae0a5b | 1457 | if (!terse_output) { |
4ceb30d4 | 1458 | log_info("Starting "); |
9cedf167 | 1459 | if (nr_thread) |
4ceb30d4 | 1460 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1461 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1462 | if (nr_process) { |
1463 | if (nr_thread) | |
1464 | printf(" and "); | |
4ceb30d4 | 1465 | log_info("%d process%s", nr_process, |
5ec10eaa | 1466 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1467 | } |
4ceb30d4 | 1468 | log_info("\n"); |
c6ae0a5b JA |
1469 | fflush(stdout); |
1470 | } | |
c04f7ec3 | 1471 | |
697a606c | 1472 | set_sig_handlers(); |
4efa970e | 1473 | |
ebac4655 JA |
1474 | todo = thread_number; |
1475 | nr_running = 0; | |
1476 | nr_started = 0; | |
1477 | m_rate = t_rate = 0; | |
1478 | ||
34572e28 | 1479 | for_each_td(td, i) { |
263e529f | 1480 | print_status_init(td->thread_number - 1); |
ebac4655 | 1481 | |
2dc1bbeb | 1482 | if (!td->o.create_serialize) { |
380cf265 | 1483 | init_disk_util(td); |
ebac4655 | 1484 | continue; |
380cf265 | 1485 | } |
ebac4655 JA |
1486 | |
1487 | /* | |
1488 | * do file setup here so it happens sequentially, | |
1489 | * we don't want X number of threads getting their | |
1490 | * client data interspersed on disk | |
1491 | */ | |
53cdc686 | 1492 | if (setup_files(td)) { |
5bf13a5a JA |
1493 | exit_value++; |
1494 | if (td->error) | |
5921e80c JA |
1495 | log_err("fio: pid=%d, err=%d/%s\n", |
1496 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1497 | td_set_runstate(td, TD_REAPED); |
1498 | todo--; | |
c69e8875 JA |
1499 | } else { |
1500 | struct fio_file *f; | |
1501 | unsigned int i; | |
1502 | ||
1503 | /* | |
1504 | * for sharing to work, each job must always open | |
1505 | * its own files. so close them, if we opened them | |
1506 | * for creation | |
1507 | */ | |
22a57ba8 JA |
1508 | for_each_file(td, f, i) { |
1509 | if (fio_file_open(f)) | |
1510 | td_io_close_file(td, f); | |
22a57ba8 | 1511 | } |
5ec10eaa | 1512 | } |
380cf265 JA |
1513 | |
1514 | init_disk_util(td); | |
ebac4655 JA |
1515 | } |
1516 | ||
a2f77c9f JA |
1517 | set_genesis_time(); |
1518 | ||
ebac4655 | 1519 | while (todo) { |
75154845 JA |
1520 | struct thread_data *map[MAX_JOBS]; |
1521 | struct timeval this_start; | |
1522 | int this_jobs = 0, left; | |
1523 | ||
ebac4655 JA |
1524 | /* |
1525 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1526 | */ | |
34572e28 | 1527 | for_each_td(td, i) { |
ebac4655 JA |
1528 | if (td->runstate != TD_NOT_CREATED) |
1529 | continue; | |
1530 | ||
1531 | /* | |
1532 | * never got a chance to start, killed by other | |
1533 | * thread for some reason | |
1534 | */ | |
1535 | if (td->terminate) { | |
1536 | todo--; | |
1537 | continue; | |
1538 | } | |
1539 | ||
2dc1bbeb | 1540 | if (td->o.start_delay) { |
263e529f | 1541 | spent = mtime_since_genesis(); |
ebac4655 | 1542 | |
2dc1bbeb | 1543 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1544 | continue; |
1545 | } | |
1546 | ||
6f88bcb0 JA |
1547 | if (td->o.stonewall && (nr_started || nr_running)) { |
1548 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1549 | td->o.name); | |
ebac4655 | 1550 | break; |
6f88bcb0 | 1551 | } |
ebac4655 | 1552 | |
75154845 JA |
1553 | /* |
1554 | * Set state to created. Thread will transition | |
1555 | * to TD_INITIALIZED when it's done setting up. | |
1556 | */ | |
ebac4655 | 1557 | td_set_runstate(td, TD_CREATED); |
75154845 | 1558 | map[this_jobs++] = td; |
ebac4655 JA |
1559 | nr_started++; |
1560 | ||
2dc1bbeb | 1561 | if (td->o.use_thread) { |
f42c153d JA |
1562 | int ret; |
1563 | ||
ee56ad50 | 1564 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1565 | ret = pthread_create(&td->thread, NULL, |
1566 | thread_main, td); | |
1567 | if (ret) { | |
1568 | log_err("pthread_create: %s\n", | |
1569 | strerror(ret)); | |
ebac4655 | 1570 | nr_started--; |
c8bb6faf | 1571 | break; |
ebac4655 | 1572 | } |
f42c153d JA |
1573 | ret = pthread_detach(td->thread); |
1574 | if (ret) | |
1575 | log_err("pthread_detach: %s", | |
1576 | strerror(ret)); | |
ebac4655 | 1577 | } else { |
5e1d306e | 1578 | pid_t pid; |
ee56ad50 | 1579 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1580 | pid = fork(); |
1581 | if (!pid) { | |
a6418147 JA |
1582 | int ret = fork_main(shm_id, i); |
1583 | ||
5e1d306e JA |
1584 | _exit(ret); |
1585 | } else if (i == fio_debug_jobno) | |
1586 | *fio_debug_jobp = pid; | |
ebac4655 | 1587 | } |
29adda3c | 1588 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1589 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1590 | log_err("fio: job startup hung? exiting.\n"); | |
1591 | terminate_threads(TERMINATE_ALL); | |
1592 | fio_abort = 1; | |
1593 | nr_started--; | |
1594 | break; | |
1595 | } | |
29adda3c | 1596 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1597 | } |
1598 | ||
1599 | /* | |
75154845 JA |
1600 | * Wait for the started threads to transition to |
1601 | * TD_INITIALIZED. | |
ebac4655 | 1602 | */ |
02bcaa8c | 1603 | fio_gettime(&this_start, NULL); |
75154845 | 1604 | left = this_jobs; |
6ce15a32 | 1605 | while (left && !fio_abort) { |
75154845 JA |
1606 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1607 | break; | |
1608 | ||
1609 | usleep(100000); | |
1610 | ||
1611 | for (i = 0; i < this_jobs; i++) { | |
1612 | td = map[i]; | |
1613 | if (!td) | |
1614 | continue; | |
b6f4d880 | 1615 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1616 | map[i] = NULL; |
1617 | left--; | |
b6f4d880 JA |
1618 | } else if (td->runstate >= TD_EXITED) { |
1619 | map[i] = NULL; | |
1620 | left--; | |
1621 | todo--; | |
1622 | nr_running++; /* work-around... */ | |
75154845 JA |
1623 | } |
1624 | } | |
1625 | } | |
1626 | ||
1627 | if (left) { | |
3b70d7e5 | 1628 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1629 | for (i = 0; i < this_jobs; i++) { |
1630 | td = map[i]; | |
1631 | if (!td) | |
1632 | continue; | |
1633 | kill(td->pid, SIGTERM); | |
1634 | } | |
1635 | break; | |
1636 | } | |
1637 | ||
1638 | /* | |
b6f4d880 | 1639 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1640 | */ |
34572e28 | 1641 | for_each_td(td, i) { |
75154845 | 1642 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1643 | continue; |
1644 | ||
b29ee5b3 JA |
1645 | if (in_ramp_time(td)) |
1646 | td_set_runstate(td, TD_RAMP); | |
1647 | else | |
1648 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1649 | nr_running++; |
1650 | nr_started--; | |
581e7141 JA |
1651 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1652 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1653 | todo--; |
cdd18ad8 | 1654 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1655 | } |
1656 | ||
1657 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1658 | ||
1659 | if (todo) | |
1660 | usleep(100000); | |
1661 | } | |
1662 | ||
1663 | while (nr_running) { | |
1664 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1665 | usleep(10000); | |
1666 | } | |
1667 | ||
1668 | update_io_ticks(); | |
2f9ade3c | 1669 | fio_unpin_memory(); |
ebac4655 JA |
1670 | } |
1671 | ||
ebac4655 JA |
1672 | int main(int argc, char *argv[]) |
1673 | { | |
29d610e1 JA |
1674 | long ps; |
1675 | ||
2e5cdb11 | 1676 | sinit(); |
d54a144d | 1677 | init_rand(&__fio_rand_state); |
2e5cdb11 | 1678 | |
dbe1125e JA |
1679 | /* |
1680 | * We need locale for number printing, if it isn't set then just | |
1681 | * go with the US format. | |
1682 | */ | |
1683 | if (!getenv("LC_NUMERIC")) | |
1684 | setlocale(LC_NUMERIC, "en_US"); | |
1685 | ||
29d610e1 JA |
1686 | ps = sysconf(_SC_PAGESIZE); |
1687 | if (ps < 0) { | |
1688 | log_err("Failed to get page size\n"); | |
1689 | return 1; | |
1690 | } | |
1691 | ||
cfc99db7 | 1692 | page_size = ps; |
29d610e1 JA |
1693 | page_mask = ps - 1; |
1694 | ||
74929ac2 JA |
1695 | fio_keywords_init(); |
1696 | ||
1697 | if (parse_options(argc, argv)) | |
1698 | return 1; | |
1699 | ||
07b3232d JA |
1700 | if (exec_profile && load_profile(exec_profile)) |
1701 | return 1; | |
1702 | ||
74929ac2 JA |
1703 | if (!thread_number) |
1704 | return 0; | |
1705 | ||
bb3884d8 JA |
1706 | if (write_bw_log) { |
1707 | setup_log(&agg_io_log[DDIR_READ]); | |
1708 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1709 | } | |
1710 | ||
cdd18ad8 | 1711 | startup_mutex = fio_mutex_init(0); |
03e20d68 BC |
1712 | if (startup_mutex == NULL) |
1713 | return 1; | |
d2bbb8d9 | 1714 | writeout_mutex = fio_mutex_init(1); |
03e20d68 BC |
1715 | if (writeout_mutex == NULL) |
1716 | return 1; | |
07739b57 | 1717 | |
a2f77c9f JA |
1718 | set_genesis_time(); |
1719 | ||
a5b01f1b | 1720 | posix_timer_setup(); |
ccb0fa24 | 1721 | status_timer_arm(); |
ebac4655 | 1722 | |
dae5341c JA |
1723 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1724 | INIT_FLIST_HEAD(cgroup_list); | |
1725 | ||
ebac4655 | 1726 | run_threads(); |
6ce15a32 | 1727 | |
bb3884d8 | 1728 | if (!fio_abort) { |
6ce15a32 | 1729 | show_run_stats(); |
bb3884d8 | 1730 | if (write_bw_log) { |
5ec10eaa JA |
1731 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1732 | __finish_log(agg_io_log[DDIR_WRITE], | |
1733 | "agg-write_bw.log"); | |
bb3884d8 JA |
1734 | } |
1735 | } | |
ebac4655 | 1736 | |
dae5341c JA |
1737 | cgroup_kill(cgroup_list); |
1738 | sfree(cgroup_list); | |
61eb313e | 1739 | sfree(cgroup_mnt); |
39f22027 | 1740 | |
a5b01f1b | 1741 | posix_timer_teardown(); |
cdd18ad8 | 1742 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1743 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1744 | return exit_value; |
ebac4655 | 1745 | } |