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> |
ebac4655 JA |
31 | #include <sys/stat.h> |
32 | #include <sys/wait.h> | |
33 | #include <sys/ipc.h> | |
34 | #include <sys/shm.h> | |
ebac4655 JA |
35 | #include <sys/mman.h> |
36 | ||
37 | #include "fio.h" | |
210f43b3 | 38 | #include "hash.h" |
2e5cdb11 | 39 | #include "smalloc.h" |
4f5af7b2 | 40 | #include "verify.h" |
7c9b1bce | 41 | #include "diskutil.h" |
a696fa2a | 42 | #include "cgroup.h" |
07b3232d | 43 | #include "profile.h" |
d54a144d | 44 | #include "lib/rand.h" |
ebac4655 | 45 | |
cfc99db7 JA |
46 | unsigned long page_mask; |
47 | unsigned long page_size; | |
d529ee19 JA |
48 | |
49 | #define PAGE_ALIGN(buf) \ | |
29d610e1 | 50 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) |
ebac4655 JA |
51 | |
52 | int groupid = 0; | |
53 | int thread_number = 0; | |
9cedf167 JA |
54 | int nr_process = 0; |
55 | int nr_thread = 0; | |
ebac4655 | 56 | int shm_id = 0; |
53cdc686 | 57 | int temp_stall_ts; |
d3eeeabc | 58 | unsigned long done_secs = 0; |
ebac4655 | 59 | |
cdd18ad8 | 60 | static struct fio_mutex *startup_mutex; |
d2bbb8d9 | 61 | static struct fio_mutex *writeout_mutex; |
6ce15a32 | 62 | static volatile int fio_abort; |
437c9b71 | 63 | static int exit_value; |
ccb0fa24 | 64 | static struct itimerval itimer; |
be4ecfdf | 65 | static pthread_t gtod_thread; |
dae5341c | 66 | static struct flist_head *cgroup_list; |
6adb38a1 | 67 | static char *cgroup_mnt; |
ebac4655 | 68 | |
bb3884d8 JA |
69 | struct io_log *agg_io_log[2]; |
70 | ||
ebac4655 | 71 | #define TERMINATE_ALL (-1) |
75154845 | 72 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 73 | |
b29ee5b3 | 74 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 75 | { |
d8fab1b6 JA |
76 | if (td->runstate == runstate) |
77 | return; | |
78 | ||
5921e80c JA |
79 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
80 | td->runstate, runstate); | |
6ce15a32 JA |
81 | td->runstate = runstate; |
82 | } | |
83 | ||
390c40e2 | 84 | static void terminate_threads(int group_id) |
ebac4655 | 85 | { |
34572e28 | 86 | struct thread_data *td; |
ebac4655 JA |
87 | int i; |
88 | ||
6f88bcb0 JA |
89 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
90 | ||
34572e28 | 91 | for_each_td(td, i) { |
ebac4655 | 92 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 93 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 94 | td->o.name, (int) td->pid); |
ad830ed7 JA |
95 | td->terminate = 1; |
96 | td->o.start_delay = 0; | |
97 | ||
7a93ab19 JA |
98 | /* |
99 | * if the thread is running, just let it exit | |
100 | */ | |
101 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 102 | kill(td->pid, SIGQUIT); |
f63f7f3b JA |
103 | else { |
104 | struct ioengine_ops *ops = td->io_ops; | |
105 | ||
106 | if (ops && (ops->flags & FIO_SIGQUIT)) | |
107 | kill(td->pid, SIGQUIT); | |
108 | } | |
ebac4655 JA |
109 | } |
110 | } | |
111 | } | |
112 | ||
ccb0fa24 JA |
113 | static void status_timer_arm(void) |
114 | { | |
115 | itimer.it_value.tv_sec = 0; | |
116 | itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; | |
117 | setitimer(ITIMER_REAL, &itimer, NULL); | |
118 | } | |
119 | ||
02444ad1 | 120 | static void sig_alrm(int fio_unused sig) |
ebac4655 | 121 | { |
697a606c | 122 | if (threads) { |
5ec10eaa | 123 | update_io_ticks(); |
5ec10eaa | 124 | print_thread_status(); |
ccb0fa24 | 125 | status_timer_arm(); |
697a606c JA |
126 | } |
127 | } | |
128 | ||
5f8f9726 JA |
129 | /* |
130 | * Happens on thread runs with ctrl-c, ignore our own SIGQUIT | |
131 | */ | |
132 | static void sig_quit(int sig) | |
133 | { | |
134 | } | |
135 | ||
697a606c JA |
136 | static void sig_int(int sig) |
137 | { | |
138 | if (threads) { | |
4ceb30d4 | 139 | log_info("\nfio: terminating on signal %d\n", sig); |
5ec10eaa JA |
140 | fflush(stdout); |
141 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
142 | } |
143 | } | |
144 | ||
697a606c JA |
145 | static void set_sig_handlers(void) |
146 | { | |
147 | struct sigaction act; | |
148 | ||
149 | memset(&act, 0, sizeof(act)); | |
150 | act.sa_handler = sig_alrm; | |
151 | act.sa_flags = SA_RESTART; | |
152 | sigaction(SIGALRM, &act, NULL); | |
153 | ||
154 | memset(&act, 0, sizeof(act)); | |
155 | act.sa_handler = sig_int; | |
156 | act.sa_flags = SA_RESTART; | |
157 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 158 | |
5f8f9726 JA |
159 | memset(&act, 0, sizeof(act)); |
160 | act.sa_handler = sig_quit; | |
161 | act.sa_flags = SA_RESTART; | |
162 | sigaction(SIGQUIT, &act, NULL); | |
697a606c JA |
163 | } |
164 | ||
906c8d75 JA |
165 | /* |
166 | * Check if we are above the minimum rate given. | |
167 | */ | |
581e7141 JA |
168 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
169 | enum td_ddir ddir) | |
ebac4655 | 170 | { |
0904200b | 171 | unsigned long long bytes = 0; |
4e991c23 | 172 | unsigned long iops = 0; |
ebac4655 JA |
173 | unsigned long spent; |
174 | unsigned long rate; | |
581e7141 JA |
175 | unsigned int ratemin = 0; |
176 | unsigned int rate_iops = 0; | |
177 | unsigned int rate_iops_min = 0; | |
ebac4655 | 178 | |
ff58fced JA |
179 | assert(ddir_rw(ddir)); |
180 | ||
d982d873 JA |
181 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
182 | return 0; | |
183 | ||
ebac4655 JA |
184 | /* |
185 | * allow a 2 second settle period in the beginning | |
186 | */ | |
187 | if (mtime_since(&td->start, now) < 2000) | |
188 | return 0; | |
189 | ||
581e7141 JA |
190 | iops += td->io_blocks[ddir]; |
191 | bytes += td->this_io_bytes[ddir]; | |
192 | ratemin += td->o.ratemin[ddir]; | |
193 | rate_iops += td->o.rate_iops[ddir]; | |
194 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 195 | |
ebac4655 JA |
196 | /* |
197 | * if rate blocks is set, sample is running | |
198 | */ | |
581e7141 JA |
199 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
200 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 201 | if (spent < td->o.ratecycle) |
ebac4655 JA |
202 | return 0; |
203 | ||
581e7141 | 204 | if (td->o.rate[ddir]) { |
4e991c23 JA |
205 | /* |
206 | * check bandwidth specified rate | |
207 | */ | |
581e7141 | 208 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 209 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 210 | ratemin); |
4e991c23 JA |
211 | return 1; |
212 | } else { | |
581e7141 JA |
213 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
214 | if (rate < ratemin || | |
215 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 216 | log_err("%s: min rate %u not met, got" |
b22989b9 | 217 | " %luKB/sec\n", td->o.name, |
581e7141 | 218 | ratemin, rate); |
4e991c23 JA |
219 | return 1; |
220 | } | |
221 | } | |
413dd459 | 222 | } else { |
4e991c23 JA |
223 | /* |
224 | * checks iops specified rate | |
225 | */ | |
581e7141 | 226 | if (iops < rate_iops) { |
5ec10eaa | 227 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 228 | td->o.name, rate_iops); |
413dd459 | 229 | return 1; |
4e991c23 | 230 | } else { |
581e7141 JA |
231 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
232 | if (rate < rate_iops_min || | |
233 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
234 | log_err("%s: min iops rate %u not met," |
235 | " got %lu\n", td->o.name, | |
581e7141 | 236 | rate_iops_min, rate); |
4e991c23 | 237 | } |
413dd459 | 238 | } |
ebac4655 JA |
239 | } |
240 | } | |
241 | ||
581e7141 JA |
242 | td->rate_bytes[ddir] = bytes; |
243 | td->rate_blocks[ddir] = iops; | |
244 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
245 | return 0; |
246 | } | |
247 | ||
581e7141 JA |
248 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
249 | unsigned long *bytes_done) | |
250 | { | |
251 | int ret = 0; | |
252 | ||
253 | if (bytes_done[0]) | |
254 | ret |= __check_min_rate(td, now, 0); | |
255 | if (bytes_done[1]) | |
256 | ret |= __check_min_rate(td, now, 1); | |
257 | ||
258 | return ret; | |
259 | } | |
260 | ||
ebac4655 JA |
261 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
262 | { | |
2dc1bbeb | 263 | if (!td->o.timeout) |
ebac4655 | 264 | return 0; |
2dc1bbeb | 265 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
266 | return 1; |
267 | ||
268 | return 0; | |
269 | } | |
270 | ||
906c8d75 JA |
271 | /* |
272 | * When job exits, we can cancel the in-flight IO if we are using async | |
273 | * io. Attempt to do so. | |
274 | */ | |
ebac4655 JA |
275 | static void cleanup_pending_aio(struct thread_data *td) |
276 | { | |
01743ee1 | 277 | struct flist_head *entry, *n; |
ebac4655 JA |
278 | struct io_u *io_u; |
279 | int r; | |
280 | ||
281 | /* | |
282 | * get immediately available events, if any | |
283 | */ | |
581e7141 | 284 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
285 | if (r < 0) |
286 | return; | |
ebac4655 JA |
287 | |
288 | /* | |
289 | * now cancel remaining active events | |
290 | */ | |
2866c82d | 291 | if (td->io_ops->cancel) { |
01743ee1 JA |
292 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
293 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 294 | |
0c6e7517 JA |
295 | /* |
296 | * if the io_u isn't in flight, then that generally | |
297 | * means someone leaked an io_u. complain but fix | |
298 | * it up, so we don't stall here. | |
299 | */ | |
300 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
301 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 302 | put_io_u(td, io_u); |
0c6e7517 JA |
303 | } else { |
304 | r = td->io_ops->cancel(td, io_u); | |
305 | if (!r) | |
306 | put_io_u(td, io_u); | |
307 | } | |
ebac4655 JA |
308 | } |
309 | } | |
310 | ||
97601024 | 311 | if (td->cur_depth) |
581e7141 | 312 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
313 | } |
314 | ||
858a3d47 JA |
315 | /* |
316 | * Helper to handle the final sync of a file. Works just like the normal | |
317 | * io path, just does everything sync. | |
318 | */ | |
319 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
320 | { | |
321 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
322 | int ret; |
323 | ||
324 | if (!io_u) | |
325 | return 1; | |
326 | ||
327 | io_u->ddir = DDIR_SYNC; | |
328 | io_u->file = f; | |
329 | ||
330 | if (td_io_prep(td, io_u)) { | |
331 | put_io_u(td, io_u); | |
332 | return 1; | |
333 | } | |
334 | ||
755200a3 | 335 | requeue: |
858a3d47 | 336 | ret = td_io_queue(td, io_u); |
36167d82 | 337 | if (ret < 0) { |
e1161c32 | 338 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 339 | put_io_u(td, io_u); |
858a3d47 | 340 | return 1; |
36167d82 | 341 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 342 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 343 | return 1; |
36167d82 JA |
344 | } else if (ret == FIO_Q_COMPLETED) { |
345 | if (io_u->error) { | |
e1161c32 | 346 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
347 | return 1; |
348 | } | |
858a3d47 | 349 | |
581e7141 | 350 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 351 | return 1; |
755200a3 JA |
352 | } else if (ret == FIO_Q_BUSY) { |
353 | if (td_io_commit(td)) | |
354 | return 1; | |
355 | goto requeue; | |
858a3d47 JA |
356 | } |
357 | ||
358 | return 0; | |
359 | } | |
360 | ||
993bf48b JA |
361 | static inline void update_tv_cache(struct thread_data *td) |
362 | { | |
363 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
364 | fio_gettime(&td->tv_cache, NULL); | |
365 | } | |
366 | ||
f2bba182 RR |
367 | static int break_on_this_error(struct thread_data *td, int *retptr) |
368 | { | |
369 | int ret = *retptr; | |
370 | ||
371 | if (ret < 0 || td->error) { | |
372 | int err; | |
373 | ||
1ec99eea JA |
374 | if (!td->o.continue_on_error) |
375 | return 1; | |
f2bba182 RR |
376 | |
377 | if (ret < 0) | |
378 | err = -ret; | |
379 | else | |
380 | err = td->error; | |
381 | ||
f2bba182 RR |
382 | if (td_non_fatal_error(err)) { |
383 | /* | |
384 | * Continue with the I/Os in case of | |
385 | * a non fatal error. | |
386 | */ | |
8a4f8a24 | 387 | update_error_count(td, err); |
f2bba182 RR |
388 | td_clear_error(td); |
389 | *retptr = 0; | |
390 | return 0; | |
8a4f8a24 RR |
391 | } else if (td->o.fill_device && err == ENOSPC) { |
392 | /* | |
393 | * We expect to hit this error if | |
394 | * fill_device option is set. | |
395 | */ | |
396 | td_clear_error(td); | |
397 | td->terminate = 1; | |
398 | return 1; | |
f2bba182 RR |
399 | } else { |
400 | /* | |
401 | * Stop the I/O in case of a fatal | |
402 | * error. | |
403 | */ | |
8a4f8a24 | 404 | update_error_count(td, err); |
f2bba182 RR |
405 | return 1; |
406 | } | |
407 | } | |
408 | ||
409 | return 0; | |
410 | } | |
411 | ||
906c8d75 | 412 | /* |
34403fb1 | 413 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
414 | * reads the blocks back in, and checks the crc/md5 of the data. |
415 | */ | |
1e97cce9 | 416 | static void do_verify(struct thread_data *td) |
ebac4655 | 417 | { |
53cdc686 | 418 | struct fio_file *f; |
36167d82 | 419 | struct io_u *io_u; |
af52b345 JA |
420 | int ret, min_events; |
421 | unsigned int i; | |
e5b401d4 | 422 | |
008618a8 JA |
423 | dprint(FD_VERIFY, "starting loop\n"); |
424 | ||
e5b401d4 JA |
425 | /* |
426 | * sync io first and invalidate cache, to make sure we really | |
427 | * read from disk. | |
428 | */ | |
429 | for_each_file(td, f, i) { | |
d6aed795 | 430 | if (!fio_file_open(f)) |
3d7b485f | 431 | continue; |
b2fdda43 JA |
432 | if (fio_io_sync(td, f)) |
433 | break; | |
434 | if (file_invalidate_cache(td, f)) | |
435 | break; | |
e5b401d4 | 436 | } |
ebac4655 | 437 | |
b2fdda43 JA |
438 | if (td->error) |
439 | return; | |
440 | ||
ebac4655 JA |
441 | td_set_runstate(td, TD_VERIFYING); |
442 | ||
36167d82 JA |
443 | io_u = NULL; |
444 | while (!td->terminate) { | |
4950421a | 445 | int ret2, full; |
5451792e | 446 | |
993bf48b JA |
447 | update_tv_cache(td); |
448 | ||
449 | if (runtime_exceeded(td, &td->tv_cache)) { | |
90a87d4b | 450 | td->terminate = 1; |
02bcaa8c | 451 | break; |
069c2918 | 452 | } |
02bcaa8c | 453 | |
dcb4830d RR |
454 | io_u = __get_io_u(td); |
455 | if (!io_u) | |
456 | break; | |
457 | ||
069c2918 JA |
458 | if (get_next_verify(td, io_u)) { |
459 | put_io_u(td, io_u); | |
ebac4655 | 460 | break; |
069c2918 | 461 | } |
ebac4655 | 462 | |
069c2918 JA |
463 | if (td_io_prep(td, io_u)) { |
464 | put_io_u(td, io_u); | |
53cdc686 | 465 | break; |
069c2918 | 466 | } |
d7762cf8 | 467 | |
e8462bd8 JA |
468 | if (td->o.verify_async) |
469 | io_u->end_io = verify_io_u_async; | |
470 | else | |
471 | io_u->end_io = verify_io_u; | |
53cdc686 | 472 | |
11786802 | 473 | ret = td_io_queue(td, io_u); |
36167d82 JA |
474 | switch (ret) { |
475 | case FIO_Q_COMPLETED: | |
f2bba182 | 476 | if (io_u->error) { |
22819ec2 | 477 | ret = -io_u->error; |
f2bba182 RR |
478 | clear_io_u(td, io_u); |
479 | } else if (io_u->resid) { | |
36167d82 | 480 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 481 | struct fio_file *f = io_u->file; |
ebac4655 | 482 | |
9e9d164e JA |
483 | /* |
484 | * zero read, fail | |
485 | */ | |
486 | if (!bytes) { | |
41d9f0fd | 487 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
488 | put_io_u(td, io_u); |
489 | break; | |
490 | } | |
8400d9b2 | 491 | |
36167d82 JA |
492 | io_u->xfer_buflen = io_u->resid; |
493 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
494 | io_u->offset += bytes; |
495 | ||
ff58fced JA |
496 | if (ddir_rw(io_u->ddir)) |
497 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 498 | |
d460eb31 | 499 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
500 | goto sync_done; |
501 | ||
11786802 JA |
502 | requeue_io_u(td, &io_u); |
503 | } else { | |
8400d9b2 | 504 | sync_done: |
581e7141 | 505 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
506 | if (ret < 0) |
507 | break; | |
36167d82 | 508 | } |
36167d82 JA |
509 | continue; |
510 | case FIO_Q_QUEUED: | |
511 | break; | |
755200a3 JA |
512 | case FIO_Q_BUSY: |
513 | requeue_io_u(td, &io_u); | |
5451792e JA |
514 | ret2 = td_io_commit(td); |
515 | if (ret2 < 0) | |
516 | ret = ret2; | |
755200a3 | 517 | break; |
36167d82 JA |
518 | default: |
519 | assert(ret < 0); | |
e1161c32 | 520 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
521 | break; |
522 | } | |
523 | ||
f2bba182 | 524 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
525 | break; |
526 | ||
ebac4655 | 527 | /* |
3af6ef39 JA |
528 | * if we can queue more, do so. but check if there are |
529 | * completed io_u's first. | |
ebac4655 | 530 | */ |
4950421a JA |
531 | full = queue_full(td) || ret == FIO_Q_BUSY; |
532 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
533 | min_events = min(td->o.iodepth_batch_complete, |
534 | td->cur_depth); | |
4950421a | 535 | if (full && !min_events) |
838bc709 | 536 | min_events = 1; |
e916b390 | 537 | |
4950421a JA |
538 | do { |
539 | /* | |
540 | * Reap required number of io units, if any, | |
541 | * and do the verification on them through | |
542 | * the callback handler | |
543 | */ | |
581e7141 | 544 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
545 | ret = -1; |
546 | break; | |
547 | } | |
548 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
549 | } | |
550 | if (ret < 0) | |
ebac4655 | 551 | break; |
36167d82 | 552 | } |
ebac4655 | 553 | |
c01c0395 JA |
554 | if (!td->error) { |
555 | min_events = td->cur_depth; | |
556 | ||
557 | if (min_events) | |
581e7141 | 558 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 559 | } else |
ebac4655 JA |
560 | cleanup_pending_aio(td); |
561 | ||
562 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
563 | |
564 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
565 | } |
566 | ||
32cd46a0 | 567 | /* |
906c8d75 | 568 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
569 | * and reaps them, checking for rate and errors along the way. |
570 | */ | |
ebac4655 JA |
571 | static void do_io(struct thread_data *td) |
572 | { | |
af52b345 JA |
573 | unsigned int i; |
574 | int ret = 0; | |
ebac4655 | 575 | |
b29ee5b3 JA |
576 | if (in_ramp_time(td)) |
577 | td_set_runstate(td, TD_RAMP); | |
578 | else | |
579 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 580 | |
457bf399 GO |
581 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
582 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { | |
97601024 | 583 | struct timeval comp_time; |
581e7141 | 584 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 585 | int min_evts = 0; |
ebac4655 | 586 | struct io_u *io_u; |
4950421a | 587 | int ret2, full; |
ebac4655 JA |
588 | |
589 | if (td->terminate) | |
590 | break; | |
591 | ||
993bf48b | 592 | update_tv_cache(td); |
97601024 | 593 | |
993bf48b | 594 | if (runtime_exceeded(td, &td->tv_cache)) { |
90a87d4b | 595 | td->terminate = 1; |
97601024 JA |
596 | break; |
597 | } | |
36167d82 | 598 | |
dcb4830d RR |
599 | io_u = get_io_u(td); |
600 | if (!io_u) | |
601 | break; | |
602 | ||
b67740d3 JA |
603 | /* |
604 | * Add verification end_io handler, if asked to verify | |
605 | * a previously written file. | |
606 | */ | |
715d2b3e JA |
607 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
608 | !td_rw(td)) { | |
e8462bd8 JA |
609 | if (td->o.verify_async) |
610 | io_u->end_io = verify_io_u_async; | |
611 | else | |
612 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 613 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
614 | } else if (in_ramp_time(td)) |
615 | td_set_runstate(td, TD_RAMP); | |
616 | else | |
d8fab1b6 | 617 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 618 | |
11786802 | 619 | ret = td_io_queue(td, io_u); |
36167d82 JA |
620 | switch (ret) { |
621 | case FIO_Q_COMPLETED: | |
f2bba182 | 622 | if (io_u->error) { |
5451792e | 623 | ret = -io_u->error; |
f2bba182 RR |
624 | clear_io_u(td, io_u); |
625 | } else if (io_u->resid) { | |
36167d82 | 626 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 627 | struct fio_file *f = io_u->file; |
36167d82 | 628 | |
9e9d164e JA |
629 | /* |
630 | * zero read, fail | |
631 | */ | |
632 | if (!bytes) { | |
41d9f0fd | 633 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
634 | put_io_u(td, io_u); |
635 | break; | |
636 | } | |
637 | ||
cec6b55d | 638 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 639 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
640 | io_u->offset += bytes; |
641 | ||
ff58fced JA |
642 | if (ddir_rw(io_u->ddir)) |
643 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 644 | |
d460eb31 | 645 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
646 | goto sync_done; |
647 | ||
11786802 JA |
648 | requeue_io_u(td, &io_u); |
649 | } else { | |
5a7c5680 | 650 | sync_done: |
581e7141 JA |
651 | if (__should_check_rate(td, 0) || |
652 | __should_check_rate(td, 1)) | |
993bf48b JA |
653 | fio_gettime(&comp_time, NULL); |
654 | ||
581e7141 JA |
655 | ret = io_u_sync_complete(td, io_u, bytes_done); |
656 | if (ret < 0) | |
657 | break; | |
cec6b55d | 658 | } |
36167d82 JA |
659 | break; |
660 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
661 | /* |
662 | * if the engine doesn't have a commit hook, | |
663 | * the io_u is really queued. if it does have such | |
664 | * a hook, it has to call io_u_queued() itself. | |
665 | */ | |
666 | if (td->io_ops->commit == NULL) | |
667 | io_u_queued(td, io_u); | |
36167d82 | 668 | break; |
755200a3 JA |
669 | case FIO_Q_BUSY: |
670 | requeue_io_u(td, &io_u); | |
5451792e JA |
671 | ret2 = td_io_commit(td); |
672 | if (ret2 < 0) | |
673 | ret = ret2; | |
755200a3 | 674 | break; |
36167d82 JA |
675 | default: |
676 | assert(ret < 0); | |
677 | put_io_u(td, io_u); | |
678 | break; | |
ebac4655 JA |
679 | } |
680 | ||
f2bba182 | 681 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
682 | break; |
683 | ||
97601024 JA |
684 | /* |
685 | * See if we need to complete some commands | |
686 | */ | |
4950421a JA |
687 | full = queue_full(td) || ret == FIO_Q_BUSY; |
688 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
689 | min_evts = min(td->o.iodepth_batch_complete, |
690 | td->cur_depth); | |
4950421a | 691 | if (full && !min_evts) |
36167d82 | 692 | min_evts = 1; |
4950421a | 693 | |
581e7141 JA |
694 | if (__should_check_rate(td, 0) || |
695 | __should_check_rate(td, 1)) | |
993bf48b | 696 | fio_gettime(&comp_time, NULL); |
4950421a JA |
697 | |
698 | do { | |
581e7141 JA |
699 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
700 | if (ret < 0) | |
4950421a JA |
701 | break; |
702 | ||
4950421a | 703 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
704 | } |
705 | ||
4950421a JA |
706 | if (ret < 0) |
707 | break; | |
581e7141 | 708 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
709 | continue; |
710 | ||
581e7141 JA |
711 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
712 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
713 | if (exitall_on_terminate) |
714 | terminate_threads(td->groupid); | |
715 | td_verror(td, EIO, "check_min_rate"); | |
716 | break; | |
717 | } | |
ebac4655 JA |
718 | } |
719 | ||
2dc1bbeb | 720 | if (td->o.thinktime) { |
9c1f7434 JA |
721 | unsigned long long b; |
722 | ||
723 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 724 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
725 | int left; |
726 | ||
2dc1bbeb | 727 | if (td->o.thinktime_spin) |
6dd6f2cd | 728 | usec_spin(td->o.thinktime_spin); |
48097d5c | 729 | |
2dc1bbeb | 730 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
731 | if (left) |
732 | usec_sleep(td, left); | |
733 | } | |
9c1f7434 | 734 | } |
ebac4655 JA |
735 | } |
736 | ||
aa31f1f1 SL |
737 | if (td->o.fill_device && td->error == ENOSPC) { |
738 | td->error = 0; | |
739 | td->terminate = 1; | |
740 | } | |
4d2413c6 | 741 | if (!td->error) { |
3d7c391d JA |
742 | struct fio_file *f; |
743 | ||
c01c0395 JA |
744 | i = td->cur_depth; |
745 | if (i) | |
581e7141 | 746 | ret = io_u_queued_complete(td, i, NULL); |
ebac4655 | 747 | |
2dc1bbeb | 748 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 749 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
750 | |
751 | for_each_file(td, f, i) { | |
d6aed795 | 752 | if (!fio_file_open(f)) |
3d7b485f | 753 | continue; |
858a3d47 | 754 | fio_io_sync(td, f); |
3d7b485f | 755 | } |
84585003 | 756 | } |
c01c0395 JA |
757 | } else |
758 | cleanup_pending_aio(td); | |
2ba1c290 JA |
759 | |
760 | /* | |
761 | * stop job if we failed doing any IO | |
762 | */ | |
763 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
764 | td->done = 1; | |
ebac4655 JA |
765 | } |
766 | ||
ebac4655 JA |
767 | static void cleanup_io_u(struct thread_data *td) |
768 | { | |
01743ee1 | 769 | struct flist_head *entry, *n; |
ebac4655 JA |
770 | struct io_u *io_u; |
771 | ||
01743ee1 JA |
772 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
773 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 774 | |
01743ee1 | 775 | flist_del(&io_u->list); |
ebac4655 JA |
776 | free(io_u); |
777 | } | |
778 | ||
2f9ade3c | 779 | free_io_mem(td); |
ebac4655 JA |
780 | } |
781 | ||
782 | static int init_io_u(struct thread_data *td) | |
783 | { | |
784 | struct io_u *io_u; | |
a00735e6 | 785 | unsigned int max_bs; |
eb7ccf38 | 786 | int cl_align, i, max_units; |
ebac4655 JA |
787 | char *p; |
788 | ||
1e3c09af | 789 | max_units = td->o.iodepth; |
2dc1bbeb | 790 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
791 | td->orig_buffer_size = (unsigned long long) max_bs |
792 | * (unsigned long long) max_units; | |
793 | ||
794 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
795 | unsigned long bs; | |
74b025b0 | 796 | |
5ec10eaa JA |
797 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
798 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
799 | } | |
ebac4655 | 800 | |
c3990645 JA |
801 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
802 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
803 | return 1; | |
804 | } | |
805 | ||
2f9ade3c JA |
806 | if (allocate_io_mem(td)) |
807 | return 1; | |
ebac4655 | 808 | |
fc966bef | 809 | if (td->o.odirect || td->o.mem_align) |
d529ee19 | 810 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
811 | else |
812 | p = td->orig_buffer; | |
813 | ||
eb7ccf38 JA |
814 | cl_align = os_cache_line_size(); |
815 | ||
ebac4655 | 816 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
817 | void *ptr; |
818 | ||
db1cd90b JA |
819 | if (td->terminate) |
820 | return 1; | |
eb7ccf38 JA |
821 | |
822 | if (posix_memalign(&ptr, cl_align, sizeof(*io_u))) { | |
823 | log_err("fio: posix_memalign=%s\n", strerror(errno)); | |
824 | break; | |
825 | } | |
826 | ||
827 | io_u = ptr; | |
ebac4655 | 828 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 829 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 830 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 831 | |
b4c5e1ac JA |
832 | if (!(td->io_ops->flags & FIO_NOIO)) { |
833 | io_u->buf = p + max_bs * i; | |
d529ee19 | 834 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 835 | |
5973cafb JA |
836 | if (td_write(td) && !td->o.refill_buffers) |
837 | io_u_fill_buffer(td, io_u, max_bs); | |
cbe8d756 RR |
838 | else if (td_write(td) && td->o.verify_pattern_bytes) { |
839 | /* | |
840 | * Fill the buffer with the pattern if we are | |
841 | * going to be doing writes. | |
842 | */ | |
843 | fill_pattern(td, io_u->buf, max_bs, io_u); | |
844 | } | |
b4c5e1ac | 845 | } |
6b9cea23 | 846 | |
b1ff3403 | 847 | io_u->index = i; |
0c6e7517 | 848 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 849 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
850 | } |
851 | ||
852 | return 0; | |
853 | } | |
854 | ||
da86774e JA |
855 | static int switch_ioscheduler(struct thread_data *td) |
856 | { | |
857 | char tmp[256], tmp2[128]; | |
858 | FILE *f; | |
859 | int ret; | |
860 | ||
ba0fbe10 | 861 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
862 | return 0; |
863 | ||
da86774e JA |
864 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
865 | ||
866 | f = fopen(tmp, "r+"); | |
867 | if (!f) { | |
cbf5c1d7 | 868 | if (errno == ENOENT) { |
5ec10eaa JA |
869 | log_err("fio: os or kernel doesn't support IO scheduler" |
870 | " switching\n"); | |
cbf5c1d7 JA |
871 | return 0; |
872 | } | |
873 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
874 | return 1; |
875 | } | |
876 | ||
877 | /* | |
878 | * Set io scheduler. | |
879 | */ | |
2dc1bbeb | 880 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 881 | if (ferror(f) || ret != 1) { |
e1161c32 | 882 | td_verror(td, errno, "fwrite"); |
da86774e JA |
883 | fclose(f); |
884 | return 1; | |
885 | } | |
886 | ||
887 | rewind(f); | |
888 | ||
889 | /* | |
890 | * Read back and check that the selected scheduler is now the default. | |
891 | */ | |
892 | ret = fread(tmp, 1, sizeof(tmp), f); | |
893 | if (ferror(f) || ret < 0) { | |
e1161c32 | 894 | td_verror(td, errno, "fread"); |
da86774e JA |
895 | fclose(f); |
896 | return 1; | |
897 | } | |
898 | ||
2dc1bbeb | 899 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 900 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 901 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 902 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
903 | fclose(f); |
904 | return 1; | |
905 | } | |
906 | ||
907 | fclose(f); | |
908 | return 0; | |
909 | } | |
910 | ||
492158cf JA |
911 | static int keep_running(struct thread_data *td) |
912 | { | |
913 | unsigned long long io_done; | |
914 | ||
20e354ef JA |
915 | if (td->done) |
916 | return 0; | |
492158cf JA |
917 | if (td->o.time_based) |
918 | return 1; | |
919 | if (td->o.loops) { | |
920 | td->o.loops--; | |
921 | return 1; | |
922 | } | |
923 | ||
5ec10eaa JA |
924 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
925 | + td->io_skip_bytes; | |
492158cf JA |
926 | if (io_done < td->o.size) |
927 | return 1; | |
928 | ||
929 | return 0; | |
930 | } | |
931 | ||
b29ee5b3 | 932 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 933 | { |
756867bd | 934 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 935 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 936 | td->zone_bytes = 0; |
581e7141 JA |
937 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
938 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 939 | |
c1324df1 JA |
940 | td->last_was_sync = 0; |
941 | ||
2ba1c290 JA |
942 | /* |
943 | * reset file done count if we are to start over | |
944 | */ | |
945 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 946 | td->nr_done_files = 0; |
5bfc35d7 JA |
947 | |
948 | /* | |
949 | * Set the same seed to get repeatable runs | |
950 | */ | |
951 | td_fill_rand_seeds(td); | |
b29ee5b3 JA |
952 | } |
953 | ||
954 | void reset_all_stats(struct thread_data *td) | |
955 | { | |
956 | struct timeval tv; | |
957 | int i; | |
958 | ||
959 | reset_io_counters(td); | |
960 | ||
961 | for (i = 0; i < 2; i++) { | |
962 | td->io_bytes[i] = 0; | |
963 | td->io_blocks[i] = 0; | |
964 | td->io_issues[i] = 0; | |
965 | td->ts.total_io_u[i] = 0; | |
966 | } | |
967 | ||
968 | fio_gettime(&tv, NULL); | |
969 | memcpy(&td->epoch, &tv, sizeof(tv)); | |
970 | memcpy(&td->start, &tv, sizeof(tv)); | |
971 | } | |
972 | ||
df9c26b1 | 973 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
974 | { |
975 | struct fio_file *f; | |
976 | unsigned int i; | |
b29ee5b3 JA |
977 | |
978 | reset_io_counters(td); | |
281f9b63 | 979 | |
24ffd2c2 | 980 | close_files(td); |
df9c26b1 | 981 | for_each_file(td, f, i) |
d6aed795 | 982 | fio_file_clear_done(f); |
ebac4655 JA |
983 | } |
984 | ||
398e8c4d JA |
985 | static int exec_string(const char *string) |
986 | { | |
987 | int ret, newlen = strlen(string) + 1 + 8; | |
988 | char *str; | |
989 | ||
990 | str = malloc(newlen); | |
991 | sprintf(str, "sh -c %s", string); | |
992 | ||
993 | ret = system(str); | |
994 | if (ret == -1) | |
995 | log_err("fio: exec of cmd <%s> failed\n", str); | |
996 | ||
997 | free(str); | |
998 | return ret; | |
999 | } | |
1000 | ||
906c8d75 JA |
1001 | /* |
1002 | * Entry point for the thread based jobs. The process based jobs end up | |
1003 | * here as well, after a little setup. | |
1004 | */ | |
ebac4655 JA |
1005 | static void *thread_main(void *data) |
1006 | { | |
10927316 | 1007 | unsigned long long runtime[2], elapsed; |
ebac4655 | 1008 | struct thread_data *td = data; |
e8462bd8 | 1009 | pthread_condattr_t attr; |
a978ba68 | 1010 | int clear_state; |
ebac4655 | 1011 | |
2dc1bbeb | 1012 | if (!td->o.use_thread) |
ebac4655 JA |
1013 | setsid(); |
1014 | ||
1015 | td->pid = getpid(); | |
1016 | ||
5921e80c | 1017 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1018 | |
01743ee1 JA |
1019 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1020 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1021 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1022 | INIT_FLIST_HEAD(&td->io_log_list); | |
1023 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 JA |
1024 | INIT_FLIST_HEAD(&td->verify_list); |
1025 | pthread_mutex_init(&td->io_u_lock, NULL); | |
bb5d7d0b | 1026 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1027 | |
e8462bd8 JA |
1028 | pthread_condattr_init(&attr); |
1029 | pthread_cond_init(&td->verify_cond, &attr); | |
1030 | pthread_cond_init(&td->free_cond, &attr); | |
1031 | ||
db1cd90b | 1032 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1033 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1034 | fio_mutex_up(startup_mutex); |
29adda3c | 1035 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1036 | fio_mutex_down(td->mutex); |
29adda3c | 1037 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1038 | |
1039 | /* | |
cdd18ad8 | 1040 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1041 | * eating a file descriptor |
1042 | */ | |
cdd18ad8 | 1043 | fio_mutex_remove(td->mutex); |
db1cd90b | 1044 | |
e0b0d892 JA |
1045 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1046 | td_verror(td, errno, "setuid"); | |
1047 | goto err; | |
1048 | } | |
1049 | if (td->o.gid != -1U && setgid(td->o.gid)) { | |
1050 | td_verror(td, errno, "setgid"); | |
1051 | goto err; | |
1052 | } | |
1053 | ||
d84f8d49 JA |
1054 | /* |
1055 | * May alter parameters that init_io_u() will use, so we need to | |
1056 | * do this first. | |
1057 | */ | |
1058 | if (init_iolog(td)) | |
1059 | goto err; | |
1060 | ||
ebac4655 | 1061 | if (init_io_u(td)) |
db1cd90b | 1062 | goto err; |
ebac4655 | 1063 | |
e8462bd8 JA |
1064 | if (td->o.verify_async && verify_async_init(td)) |
1065 | goto err; | |
1066 | ||
1067 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
e1161c32 | 1068 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 1069 | goto err; |
ebac4655 JA |
1070 | } |
1071 | ||
f484470b JA |
1072 | /* |
1073 | * If we have a gettimeofday() thread, make sure we exclude that | |
1074 | * thread from this job | |
1075 | */ | |
be4ecfdf | 1076 | if (td->o.gtod_cpu) { |
be4ecfdf | 1077 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); |
e8462bd8 | 1078 | if (fio_setaffinity(td->pid, td->o.cpumask) == -1) { |
be4ecfdf JA |
1079 | td_verror(td, errno, "cpu_set_affinity"); |
1080 | goto err; | |
1081 | } | |
1082 | } | |
1083 | ||
ac684785 | 1084 | if (td->ioprio_set) { |
ebac4655 | 1085 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1086 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1087 | goto err; |
ebac4655 JA |
1088 | } |
1089 | } | |
1090 | ||
6adb38a1 | 1091 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1092 | goto err; |
1093 | ||
2dc1bbeb | 1094 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1095 | td_verror(td, errno, "nice"); |
db1cd90b | 1096 | goto err; |
b6f4d880 JA |
1097 | } |
1098 | ||
2dc1bbeb | 1099 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1100 | goto err; |
37f56873 | 1101 | |
2dc1bbeb | 1102 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1103 | goto err; |
1104 | ||
7d6c5283 JA |
1105 | if (td_io_init(td)) |
1106 | goto err; | |
1107 | ||
68727076 JA |
1108 | if (init_random_map(td)) |
1109 | goto err; | |
1110 | ||
2dc1bbeb | 1111 | if (td->o.exec_prerun) { |
398e8c4d | 1112 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1113 | goto err; |
1114 | } | |
4e0ba8af | 1115 | |
afad68f7 ZY |
1116 | if (td->o.pre_read) { |
1117 | if (pre_read_files(td) < 0) | |
1118 | goto err; | |
1119 | } | |
1120 | ||
69008999 | 1121 | fio_gettime(&td->epoch, NULL); |
756867bd | 1122 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
1123 | |
1124 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 1125 | clear_state = 0; |
492158cf | 1126 | while (keep_running(td)) { |
02bcaa8c | 1127 | fio_gettime(&td->start, NULL); |
c5d3d790 JA |
1128 | memcpy(&td->ts.stat_sample_time[0], &td->start, |
1129 | sizeof(td->start)); | |
1130 | memcpy(&td->ts.stat_sample_time[1], &td->start, | |
1131 | sizeof(td->start)); | |
993bf48b | 1132 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1133 | |
cf2dd8d6 | 1134 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
5ec10eaa JA |
1135 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
1136 | sizeof(td->lastrate)); | |
ebac4655 | 1137 | |
df9c26b1 JA |
1138 | if (clear_state) |
1139 | clear_io_state(td); | |
a978ba68 | 1140 | |
ebac4655 JA |
1141 | prune_io_piece_log(td); |
1142 | ||
ba0fbe10 | 1143 | do_io(td); |
ebac4655 | 1144 | |
a978ba68 JA |
1145 | clear_state = 1; |
1146 | ||
38d77cae | 1147 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1148 | elapsed = utime_since_now(&td->start); |
38d77cae JA |
1149 | runtime[DDIR_READ] += elapsed; |
1150 | } | |
1151 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1152 | elapsed = utime_since_now(&td->start); |
38d77cae JA |
1153 | runtime[DDIR_WRITE] += elapsed; |
1154 | } | |
5ec10eaa | 1155 | |
ebac4655 JA |
1156 | if (td->error || td->terminate) |
1157 | break; | |
1158 | ||
e84c73a8 SL |
1159 | if (!td->o.do_verify || |
1160 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1161 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1162 | continue; |
1163 | ||
df9c26b1 | 1164 | clear_io_state(td); |
a978ba68 | 1165 | |
02bcaa8c | 1166 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1167 | |
1168 | do_verify(td); | |
1169 | ||
69008999 | 1170 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1171 | |
1172 | if (td->error || td->terminate) | |
1173 | break; | |
1174 | } | |
1175 | ||
36dff966 | 1176 | update_rusage_stat(td); |
47f0cc48 YHJT |
1177 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
1178 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
1179 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1180 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1181 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1182 | ||
d2bbb8d9 | 1183 | fio_mutex_down(writeout_mutex); |
e3cedca7 | 1184 | if (td->ts.bw_log) { |
f484470b JA |
1185 | if (td->o.bw_log_file) { |
1186 | finish_log_named(td, td->ts.bw_log, | |
1187 | td->o.bw_log_file, "bw"); | |
1188 | } else | |
e3cedca7 JA |
1189 | finish_log(td, td->ts.bw_log, "bw"); |
1190 | } | |
02af0988 JA |
1191 | if (td->ts.lat_log) { |
1192 | if (td->o.lat_log_file) { | |
1193 | finish_log_named(td, td->ts.lat_log, | |
1194 | td->o.lat_log_file, "lat"); | |
1195 | } else | |
1196 | finish_log(td, td->ts.lat_log, "lat"); | |
1197 | } | |
e3cedca7 | 1198 | if (td->ts.slat_log) { |
f484470b JA |
1199 | if (td->o.lat_log_file) { |
1200 | finish_log_named(td, td->ts.slat_log, | |
1ca95738 | 1201 | td->o.lat_log_file, "slat"); |
f484470b | 1202 | } else |
e3cedca7 JA |
1203 | finish_log(td, td->ts.slat_log, "slat"); |
1204 | } | |
1205 | if (td->ts.clat_log) { | |
f484470b JA |
1206 | if (td->o.lat_log_file) { |
1207 | finish_log_named(td, td->ts.clat_log, | |
1208 | td->o.lat_log_file, "clat"); | |
1209 | } else | |
e3cedca7 JA |
1210 | finish_log(td, td->ts.clat_log, "clat"); |
1211 | } | |
d2bbb8d9 | 1212 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1213 | if (td->o.exec_postrun) |
1214 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1215 | |
1216 | if (exitall_on_terminate) | |
390c40e2 | 1217 | terminate_threads(td->groupid); |
ebac4655 JA |
1218 | |
1219 | err: | |
5bf13a5a | 1220 | if (td->error) |
4ceb30d4 | 1221 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1222 | td->verror); |
b4707ed7 JA |
1223 | |
1224 | if (td->o.verify_async) | |
1225 | verify_async_exit(td); | |
1226 | ||
24ffd2c2 | 1227 | close_and_free_files(td); |
2866c82d | 1228 | close_ioengine(td); |
ebac4655 | 1229 | cleanup_io_u(td); |
6adb38a1 | 1230 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1231 | |
d2ce18b5 JA |
1232 | if (td->o.cpumask_set) { |
1233 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1234 | ||
1235 | td_verror(td, ret, "fio_cpuset_exit"); | |
1236 | } | |
4e78e405 | 1237 | |
f29b25a3 JA |
1238 | /* |
1239 | * do this very late, it will log file closing as well | |
1240 | */ | |
1241 | if (td->o.write_iolog_file) | |
1242 | write_iolog_close(td); | |
1243 | ||
d23bb327 | 1244 | options_mem_free(td); |
ebac4655 | 1245 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1246 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1247 | } |
1248 | ||
906c8d75 JA |
1249 | /* |
1250 | * We cannot pass the td data into a forked process, so attach the td and | |
1251 | * pass it to the thread worker. | |
1252 | */ | |
a6418147 | 1253 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1254 | { |
1255 | struct thread_data *td; | |
a6418147 | 1256 | void *data, *ret; |
ebac4655 JA |
1257 | |
1258 | data = shmat(shmid, NULL, 0); | |
1259 | if (data == (void *) -1) { | |
a6418147 JA |
1260 | int __err = errno; |
1261 | ||
ebac4655 | 1262 | perror("shmat"); |
a6418147 | 1263 | return __err; |
ebac4655 JA |
1264 | } |
1265 | ||
1266 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1267 | ret = thread_main(td); |
ebac4655 | 1268 | shmdt(data); |
43d76807 | 1269 | return (int) (unsigned long) ret; |
ebac4655 JA |
1270 | } |
1271 | ||
906c8d75 JA |
1272 | /* |
1273 | * Run over the job map and reap the threads that have exited, if any. | |
1274 | */ | |
ebac4655 JA |
1275 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1276 | { | |
34572e28 | 1277 | struct thread_data *td; |
1f809d15 | 1278 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1279 | |
1280 | /* | |
1281 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1282 | */ | |
1f809d15 | 1283 | realthreads = pending = cputhreads = 0; |
34572e28 | 1284 | for_each_td(td, i) { |
3707f45b | 1285 | int flags = 0; |
a2f77c9f | 1286 | |
84585003 JA |
1287 | /* |
1288 | * ->io_ops is NULL for a thread that has closed its | |
1289 | * io engine | |
1290 | */ | |
ba0fbe10 | 1291 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1292 | cputhreads++; |
1f809d15 JA |
1293 | else |
1294 | realthreads++; | |
b990b5c0 | 1295 | |
a1dedc1f AC |
1296 | if (!td->pid) { |
1297 | pending++; | |
1298 | continue; | |
1299 | } | |
1300 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1301 | continue; |
2dc1bbeb | 1302 | if (td->o.use_thread) { |
3707f45b JA |
1303 | if (td->runstate == TD_EXITED) { |
1304 | td_set_runstate(td, TD_REAPED); | |
1305 | goto reaped; | |
1306 | } | |
1307 | continue; | |
1308 | } | |
a2f77c9f JA |
1309 | |
1310 | flags = WNOHANG; | |
1311 | if (td->runstate == TD_EXITED) | |
1312 | flags = 0; | |
1313 | ||
1314 | /* | |
1315 | * check if someone quit or got killed in an unusual way | |
1316 | */ | |
1317 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1318 | if (ret < 0) { |
a2f77c9f | 1319 | if (errno == ECHILD) { |
5921e80c JA |
1320 | log_err("fio: pid=%d disappeared %d\n", |
1321 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1322 | td_set_runstate(td, TD_REAPED); |
1323 | goto reaped; | |
1324 | } | |
1325 | perror("waitpid"); | |
1326 | } else if (ret == td->pid) { | |
1327 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1328 | int sig = WTERMSIG(status); |
1329 | ||
d9244941 | 1330 | if (sig != SIGQUIT) |
5ec10eaa | 1331 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1332 | (int) td->pid, sig); |
fab6aa71 JA |
1333 | td_set_runstate(td, TD_REAPED); |
1334 | goto reaped; | |
1335 | } | |
a2f77c9f JA |
1336 | if (WIFEXITED(status)) { |
1337 | if (WEXITSTATUS(status) && !td->error) | |
1338 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1339 | |
a2f77c9f JA |
1340 | td_set_runstate(td, TD_REAPED); |
1341 | goto reaped; | |
a6418147 JA |
1342 | } |
1343 | } | |
ebac4655 | 1344 | |
a2f77c9f JA |
1345 | /* |
1346 | * thread is not dead, continue | |
1347 | */ | |
e48676ba | 1348 | pending++; |
a2f77c9f | 1349 | continue; |
fab6aa71 | 1350 | reaped: |
ebac4655 | 1351 | (*nr_running)--; |
581e7141 JA |
1352 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1353 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1354 | if (!td->pid) |
1355 | pending--; | |
a2f77c9f JA |
1356 | |
1357 | if (td->error) | |
1358 | exit_value++; | |
d3eeeabc JA |
1359 | |
1360 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1361 | } |
b990b5c0 | 1362 | |
1f809d15 | 1363 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1364 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1365 | } |
1366 | ||
be4ecfdf JA |
1367 | static void *gtod_thread_main(void *data) |
1368 | { | |
1369 | fio_mutex_up(startup_mutex); | |
1370 | ||
1371 | /* | |
1372 | * As long as we have jobs around, update the clock. It would be nice | |
1373 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1374 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1375 | * it - we don't want to lose precision. | |
1376 | */ | |
1377 | while (threads) { | |
1378 | fio_gtod_update(); | |
1379 | nop; | |
1380 | } | |
1381 | ||
1382 | return NULL; | |
1383 | } | |
1384 | ||
1385 | static int fio_start_gtod_thread(void) | |
1386 | { | |
304a47c7 | 1387 | pthread_attr_t attr; |
f42c153d JA |
1388 | int ret; |
1389 | ||
304a47c7 VA |
1390 | pthread_attr_init(&attr); |
1391 | pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); | |
1392 | ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); | |
1393 | pthread_attr_destroy(&attr); | |
f42c153d JA |
1394 | if (ret) { |
1395 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1396 | return 1; |
1397 | } | |
f42c153d JA |
1398 | |
1399 | ret = pthread_detach(gtod_thread); | |
1400 | if (ret) { | |
1401 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1402 | return 1; |
1403 | } | |
1404 | ||
29adda3c | 1405 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1406 | fio_mutex_down(startup_mutex); |
29adda3c | 1407 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1408 | return 0; |
1409 | } | |
1410 | ||
906c8d75 JA |
1411 | /* |
1412 | * Main function for kicking off and reaping jobs, as needed. | |
1413 | */ | |
ebac4655 JA |
1414 | static void run_threads(void) |
1415 | { | |
ebac4655 JA |
1416 | struct thread_data *td; |
1417 | unsigned long spent; | |
1418 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1419 | |
2f9ade3c JA |
1420 | if (fio_pin_memory()) |
1421 | return; | |
ebac4655 | 1422 | |
be4ecfdf JA |
1423 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1424 | return; | |
1425 | ||
c6ae0a5b | 1426 | if (!terse_output) { |
4ceb30d4 | 1427 | log_info("Starting "); |
9cedf167 | 1428 | if (nr_thread) |
4ceb30d4 | 1429 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1430 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1431 | if (nr_process) { |
1432 | if (nr_thread) | |
1433 | printf(" and "); | |
4ceb30d4 | 1434 | log_info("%d process%s", nr_process, |
5ec10eaa | 1435 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1436 | } |
4ceb30d4 | 1437 | log_info("\n"); |
c6ae0a5b JA |
1438 | fflush(stdout); |
1439 | } | |
c04f7ec3 | 1440 | |
697a606c | 1441 | set_sig_handlers(); |
4efa970e | 1442 | |
ebac4655 JA |
1443 | todo = thread_number; |
1444 | nr_running = 0; | |
1445 | nr_started = 0; | |
1446 | m_rate = t_rate = 0; | |
1447 | ||
34572e28 | 1448 | for_each_td(td, i) { |
263e529f | 1449 | print_status_init(td->thread_number - 1); |
ebac4655 | 1450 | |
2dc1bbeb | 1451 | if (!td->o.create_serialize) { |
380cf265 | 1452 | init_disk_util(td); |
ebac4655 | 1453 | continue; |
380cf265 | 1454 | } |
ebac4655 JA |
1455 | |
1456 | /* | |
1457 | * do file setup here so it happens sequentially, | |
1458 | * we don't want X number of threads getting their | |
1459 | * client data interspersed on disk | |
1460 | */ | |
53cdc686 | 1461 | if (setup_files(td)) { |
5bf13a5a JA |
1462 | exit_value++; |
1463 | if (td->error) | |
5921e80c JA |
1464 | log_err("fio: pid=%d, err=%d/%s\n", |
1465 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1466 | td_set_runstate(td, TD_REAPED); |
1467 | todo--; | |
c69e8875 JA |
1468 | } else { |
1469 | struct fio_file *f; | |
1470 | unsigned int i; | |
1471 | ||
1472 | /* | |
1473 | * for sharing to work, each job must always open | |
1474 | * its own files. so close them, if we opened them | |
1475 | * for creation | |
1476 | */ | |
22a57ba8 JA |
1477 | for_each_file(td, f, i) { |
1478 | if (fio_file_open(f)) | |
1479 | td_io_close_file(td, f); | |
22a57ba8 | 1480 | } |
5ec10eaa | 1481 | } |
380cf265 JA |
1482 | |
1483 | init_disk_util(td); | |
ebac4655 JA |
1484 | } |
1485 | ||
a2f77c9f JA |
1486 | set_genesis_time(); |
1487 | ||
ebac4655 | 1488 | while (todo) { |
75154845 JA |
1489 | struct thread_data *map[MAX_JOBS]; |
1490 | struct timeval this_start; | |
1491 | int this_jobs = 0, left; | |
1492 | ||
ebac4655 JA |
1493 | /* |
1494 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1495 | */ | |
34572e28 | 1496 | for_each_td(td, i) { |
ebac4655 JA |
1497 | if (td->runstate != TD_NOT_CREATED) |
1498 | continue; | |
1499 | ||
1500 | /* | |
1501 | * never got a chance to start, killed by other | |
1502 | * thread for some reason | |
1503 | */ | |
1504 | if (td->terminate) { | |
1505 | todo--; | |
1506 | continue; | |
1507 | } | |
1508 | ||
2dc1bbeb | 1509 | if (td->o.start_delay) { |
263e529f | 1510 | spent = mtime_since_genesis(); |
ebac4655 | 1511 | |
2dc1bbeb | 1512 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1513 | continue; |
1514 | } | |
1515 | ||
6f88bcb0 JA |
1516 | if (td->o.stonewall && (nr_started || nr_running)) { |
1517 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1518 | td->o.name); | |
ebac4655 | 1519 | break; |
6f88bcb0 | 1520 | } |
ebac4655 | 1521 | |
75154845 JA |
1522 | /* |
1523 | * Set state to created. Thread will transition | |
1524 | * to TD_INITIALIZED when it's done setting up. | |
1525 | */ | |
ebac4655 | 1526 | td_set_runstate(td, TD_CREATED); |
75154845 | 1527 | map[this_jobs++] = td; |
ebac4655 JA |
1528 | nr_started++; |
1529 | ||
2dc1bbeb | 1530 | if (td->o.use_thread) { |
f42c153d JA |
1531 | int ret; |
1532 | ||
ee56ad50 | 1533 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1534 | ret = pthread_create(&td->thread, NULL, |
1535 | thread_main, td); | |
1536 | if (ret) { | |
1537 | log_err("pthread_create: %s\n", | |
1538 | strerror(ret)); | |
ebac4655 | 1539 | nr_started--; |
c8bb6faf | 1540 | break; |
ebac4655 | 1541 | } |
f42c153d JA |
1542 | ret = pthread_detach(td->thread); |
1543 | if (ret) | |
1544 | log_err("pthread_detach: %s", | |
1545 | strerror(ret)); | |
ebac4655 | 1546 | } else { |
5e1d306e | 1547 | pid_t pid; |
ee56ad50 | 1548 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1549 | pid = fork(); |
1550 | if (!pid) { | |
a6418147 JA |
1551 | int ret = fork_main(shm_id, i); |
1552 | ||
5e1d306e JA |
1553 | _exit(ret); |
1554 | } else if (i == fio_debug_jobno) | |
1555 | *fio_debug_jobp = pid; | |
ebac4655 | 1556 | } |
29adda3c | 1557 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1558 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1559 | log_err("fio: job startup hung? exiting.\n"); | |
1560 | terminate_threads(TERMINATE_ALL); | |
1561 | fio_abort = 1; | |
1562 | nr_started--; | |
1563 | break; | |
1564 | } | |
29adda3c | 1565 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1566 | } |
1567 | ||
1568 | /* | |
75154845 JA |
1569 | * Wait for the started threads to transition to |
1570 | * TD_INITIALIZED. | |
ebac4655 | 1571 | */ |
02bcaa8c | 1572 | fio_gettime(&this_start, NULL); |
75154845 | 1573 | left = this_jobs; |
6ce15a32 | 1574 | while (left && !fio_abort) { |
75154845 JA |
1575 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1576 | break; | |
1577 | ||
1578 | usleep(100000); | |
1579 | ||
1580 | for (i = 0; i < this_jobs; i++) { | |
1581 | td = map[i]; | |
1582 | if (!td) | |
1583 | continue; | |
b6f4d880 | 1584 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1585 | map[i] = NULL; |
1586 | left--; | |
b6f4d880 JA |
1587 | } else if (td->runstate >= TD_EXITED) { |
1588 | map[i] = NULL; | |
1589 | left--; | |
1590 | todo--; | |
1591 | nr_running++; /* work-around... */ | |
75154845 JA |
1592 | } |
1593 | } | |
1594 | } | |
1595 | ||
1596 | if (left) { | |
3b70d7e5 | 1597 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1598 | for (i = 0; i < this_jobs; i++) { |
1599 | td = map[i]; | |
1600 | if (!td) | |
1601 | continue; | |
1602 | kill(td->pid, SIGTERM); | |
1603 | } | |
1604 | break; | |
1605 | } | |
1606 | ||
1607 | /* | |
b6f4d880 | 1608 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1609 | */ |
34572e28 | 1610 | for_each_td(td, i) { |
75154845 | 1611 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1612 | continue; |
1613 | ||
b29ee5b3 JA |
1614 | if (in_ramp_time(td)) |
1615 | td_set_runstate(td, TD_RAMP); | |
1616 | else | |
1617 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1618 | nr_running++; |
1619 | nr_started--; | |
581e7141 JA |
1620 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1621 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1622 | todo--; |
cdd18ad8 | 1623 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1624 | } |
1625 | ||
1626 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1627 | ||
1628 | if (todo) | |
1629 | usleep(100000); | |
1630 | } | |
1631 | ||
1632 | while (nr_running) { | |
1633 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1634 | usleep(10000); | |
1635 | } | |
1636 | ||
1637 | update_io_ticks(); | |
2f9ade3c | 1638 | fio_unpin_memory(); |
ebac4655 JA |
1639 | } |
1640 | ||
ebac4655 JA |
1641 | int main(int argc, char *argv[]) |
1642 | { | |
29d610e1 JA |
1643 | long ps; |
1644 | ||
2e5cdb11 | 1645 | sinit(); |
d54a144d | 1646 | init_rand(&__fio_rand_state); |
2e5cdb11 | 1647 | |
dbe1125e JA |
1648 | /* |
1649 | * We need locale for number printing, if it isn't set then just | |
1650 | * go with the US format. | |
1651 | */ | |
1652 | if (!getenv("LC_NUMERIC")) | |
1653 | setlocale(LC_NUMERIC, "en_US"); | |
1654 | ||
29d610e1 JA |
1655 | ps = sysconf(_SC_PAGESIZE); |
1656 | if (ps < 0) { | |
1657 | log_err("Failed to get page size\n"); | |
1658 | return 1; | |
1659 | } | |
1660 | ||
cfc99db7 | 1661 | page_size = ps; |
29d610e1 JA |
1662 | page_mask = ps - 1; |
1663 | ||
74929ac2 JA |
1664 | fio_keywords_init(); |
1665 | ||
1666 | if (parse_options(argc, argv)) | |
1667 | return 1; | |
1668 | ||
07b3232d JA |
1669 | if (exec_profile && load_profile(exec_profile)) |
1670 | return 1; | |
1671 | ||
74929ac2 JA |
1672 | if (!thread_number) |
1673 | return 0; | |
1674 | ||
bb3884d8 JA |
1675 | if (write_bw_log) { |
1676 | setup_log(&agg_io_log[DDIR_READ]); | |
1677 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1678 | } | |
1679 | ||
cdd18ad8 | 1680 | startup_mutex = fio_mutex_init(0); |
d2bbb8d9 | 1681 | writeout_mutex = fio_mutex_init(1); |
07739b57 | 1682 | |
a2f77c9f JA |
1683 | set_genesis_time(); |
1684 | ||
ccb0fa24 | 1685 | status_timer_arm(); |
ebac4655 | 1686 | |
dae5341c JA |
1687 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1688 | INIT_FLIST_HEAD(cgroup_list); | |
1689 | ||
ebac4655 | 1690 | run_threads(); |
6ce15a32 | 1691 | |
bb3884d8 | 1692 | if (!fio_abort) { |
6ce15a32 | 1693 | show_run_stats(); |
bb3884d8 | 1694 | if (write_bw_log) { |
5ec10eaa JA |
1695 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1696 | __finish_log(agg_io_log[DDIR_WRITE], | |
1697 | "agg-write_bw.log"); | |
bb3884d8 JA |
1698 | } |
1699 | } | |
ebac4655 | 1700 | |
dae5341c JA |
1701 | cgroup_kill(cgroup_list); |
1702 | sfree(cgroup_list); | |
61eb313e | 1703 | sfree(cgroup_mnt); |
39f22027 | 1704 | |
cdd18ad8 | 1705 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1706 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1707 | return exit_value; |
ebac4655 | 1708 | } |