Commit | Line | Data |
---|---|---|
2e1df07d JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
5 | * Copyright (C) 2006-2012 Jens Axboe <axboe@kernel.dk> | |
6 | * | |
7 | * The license below covers all files distributed with fio unless otherwise | |
8 | * noted in the file itself. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or modify | |
11 | * it under the terms of the GNU General Public License version 2 as | |
12 | * published by the Free Software Foundation. | |
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 | */ | |
24 | #include <unistd.h> | |
25 | #include <fcntl.h> | |
26 | #include <string.h> | |
27 | #include <limits.h> | |
28 | #include <signal.h> | |
29 | #include <time.h> | |
30 | #include <locale.h> | |
31 | #include <assert.h> | |
32 | #include <time.h> | |
e43606c2 | 33 | #include <inttypes.h> |
2e1df07d JA |
34 | #include <sys/stat.h> |
35 | #include <sys/wait.h> | |
36 | #include <sys/ipc.h> | |
2e1df07d JA |
37 | #include <sys/mman.h> |
38 | ||
39 | #include "fio.h" | |
a5e0ee11 O |
40 | #ifndef FIO_NO_HAVE_SHM_H |
41 | #include <sys/shm.h> | |
42 | #endif | |
2e1df07d JA |
43 | #include "hash.h" |
44 | #include "smalloc.h" | |
45 | #include "verify.h" | |
46 | #include "trim.h" | |
47 | #include "diskutil.h" | |
48 | #include "cgroup.h" | |
49 | #include "profile.h" | |
50 | #include "lib/rand.h" | |
51 | #include "memalign.h" | |
52 | #include "server.h" | |
44404c5a | 53 | #include "lib/getrusage.h" |
f2a2ce0e | 54 | #include "idletime.h" |
002fe734 | 55 | #include "err.h" |
78d55e72 | 56 | #include "lib/tp.h" |
a9da8ab2 | 57 | #include "workqueue.h" |
e81ecca3 | 58 | #include "lib/mountcheck.h" |
2e1df07d | 59 | |
5ddc6707 JA |
60 | static pthread_t helper_thread; |
61 | static pthread_mutex_t helper_lock; | |
62 | pthread_cond_t helper_cond; | |
63 | int helper_do_stat = 0; | |
8aab824f | 64 | |
2e1df07d | 65 | static struct fio_mutex *startup_mutex; |
2e1df07d JA |
66 | static struct flist_head *cgroup_list; |
67 | static char *cgroup_mnt; | |
68 | static int exit_value; | |
69 | static volatile int fio_abort; | |
3a5f6bde JA |
70 | static unsigned int nr_process = 0; |
71 | static unsigned int nr_thread = 0; | |
2e1df07d | 72 | |
6eaf09d6 | 73 | struct io_log *agg_io_log[DDIR_RWDIR_CNT]; |
2e1df07d | 74 | |
a3efc919 JA |
75 | int groupid = 0; |
76 | unsigned int thread_number = 0; | |
108fea77 | 77 | unsigned int stat_number = 0; |
a3efc919 JA |
78 | int shm_id = 0; |
79 | int temp_stall_ts; | |
80 | unsigned long done_secs = 0; | |
5ddc6707 | 81 | volatile int helper_exit = 0; |
a3efc919 | 82 | |
2e1df07d | 83 | #define PAGE_ALIGN(buf) \ |
e43606c2 | 84 | (char *) (((uintptr_t) (buf) + page_mask) & ~page_mask) |
2e1df07d JA |
85 | |
86 | #define JOB_START_TIMEOUT (5 * 1000) | |
87 | ||
88 | static void sig_int(int sig) | |
89 | { | |
90 | if (threads) { | |
91 | if (is_backend) | |
92 | fio_server_got_signal(sig); | |
93 | else { | |
94 | log_info("\nfio: terminating on signal %d\n", sig); | |
e411c301 | 95 | log_info_flush(); |
2e1df07d JA |
96 | exit_value = 128; |
97 | } | |
98 | ||
99 | fio_terminate_threads(TERMINATE_ALL); | |
100 | } | |
101 | } | |
102 | ||
4c6d91e8 JA |
103 | static void sig_show_status(int sig) |
104 | { | |
105 | show_running_run_stats(); | |
106 | } | |
107 | ||
2e1df07d JA |
108 | static void set_sig_handlers(void) |
109 | { | |
110 | struct sigaction act; | |
111 | ||
112 | memset(&act, 0, sizeof(act)); | |
113 | act.sa_handler = sig_int; | |
114 | act.sa_flags = SA_RESTART; | |
115 | sigaction(SIGINT, &act, NULL); | |
116 | ||
117 | memset(&act, 0, sizeof(act)); | |
118 | act.sa_handler = sig_int; | |
119 | act.sa_flags = SA_RESTART; | |
120 | sigaction(SIGTERM, &act, NULL); | |
121 | ||
2f694507 BC |
122 | /* Windows uses SIGBREAK as a quit signal from other applications */ |
123 | #ifdef WIN32 | |
124 | memset(&act, 0, sizeof(act)); | |
125 | act.sa_handler = sig_int; | |
126 | act.sa_flags = SA_RESTART; | |
127 | sigaction(SIGBREAK, &act, NULL); | |
128 | #endif | |
129 | ||
4c6d91e8 JA |
130 | memset(&act, 0, sizeof(act)); |
131 | act.sa_handler = sig_show_status; | |
132 | act.sa_flags = SA_RESTART; | |
133 | sigaction(SIGUSR1, &act, NULL); | |
134 | ||
2e1df07d JA |
135 | if (is_backend) { |
136 | memset(&act, 0, sizeof(act)); | |
137 | act.sa_handler = sig_int; | |
138 | act.sa_flags = SA_RESTART; | |
139 | sigaction(SIGPIPE, &act, NULL); | |
140 | } | |
141 | } | |
142 | ||
143 | /* | |
144 | * Check if we are above the minimum rate given. | |
145 | */ | |
146 | static int __check_min_rate(struct thread_data *td, struct timeval *now, | |
147 | enum fio_ddir ddir) | |
148 | { | |
149 | unsigned long long bytes = 0; | |
150 | unsigned long iops = 0; | |
151 | unsigned long spent; | |
152 | unsigned long rate; | |
153 | unsigned int ratemin = 0; | |
154 | unsigned int rate_iops = 0; | |
155 | unsigned int rate_iops_min = 0; | |
156 | ||
157 | assert(ddir_rw(ddir)); | |
158 | ||
159 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) | |
160 | return 0; | |
161 | ||
162 | /* | |
163 | * allow a 2 second settle period in the beginning | |
164 | */ | |
165 | if (mtime_since(&td->start, now) < 2000) | |
166 | return 0; | |
167 | ||
168 | iops += td->this_io_blocks[ddir]; | |
169 | bytes += td->this_io_bytes[ddir]; | |
170 | ratemin += td->o.ratemin[ddir]; | |
171 | rate_iops += td->o.rate_iops[ddir]; | |
172 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
173 | ||
174 | /* | |
175 | * if rate blocks is set, sample is running | |
176 | */ | |
177 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { | |
178 | spent = mtime_since(&td->lastrate[ddir], now); | |
179 | if (spent < td->o.ratecycle) | |
180 | return 0; | |
181 | ||
20e37e4a | 182 | if (td->o.rate[ddir] || td->o.ratemin[ddir]) { |
2e1df07d JA |
183 | /* |
184 | * check bandwidth specified rate | |
185 | */ | |
186 | if (bytes < td->rate_bytes[ddir]) { | |
187 | log_err("%s: min rate %u not met\n", td->o.name, | |
188 | ratemin); | |
189 | return 1; | |
190 | } else { | |
49cba9b3 JA |
191 | if (spent) |
192 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; | |
193 | else | |
194 | rate = 0; | |
195 | ||
2e1df07d JA |
196 | if (rate < ratemin || |
197 | bytes < td->rate_bytes[ddir]) { | |
198 | log_err("%s: min rate %u not met, got" | |
199 | " %luKB/sec\n", td->o.name, | |
200 | ratemin, rate); | |
201 | return 1; | |
202 | } | |
203 | } | |
204 | } else { | |
205 | /* | |
206 | * checks iops specified rate | |
207 | */ | |
208 | if (iops < rate_iops) { | |
209 | log_err("%s: min iops rate %u not met\n", | |
210 | td->o.name, rate_iops); | |
211 | return 1; | |
212 | } else { | |
4c707a3b JA |
213 | if (spent) |
214 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; | |
215 | else | |
216 | rate = 0; | |
217 | ||
2e1df07d JA |
218 | if (rate < rate_iops_min || |
219 | iops < td->rate_blocks[ddir]) { | |
220 | log_err("%s: min iops rate %u not met," | |
221 | " got %lu\n", td->o.name, | |
222 | rate_iops_min, rate); | |
20e37e4a | 223 | return 1; |
2e1df07d JA |
224 | } |
225 | } | |
226 | } | |
227 | } | |
228 | ||
229 | td->rate_bytes[ddir] = bytes; | |
230 | td->rate_blocks[ddir] = iops; | |
231 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
232 | return 0; | |
233 | } | |
234 | ||
55312f9f | 235 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
2e1df07d JA |
236 | { |
237 | int ret = 0; | |
238 | ||
55312f9f | 239 | if (td->bytes_done[DDIR_READ]) |
6eaf09d6 | 240 | ret |= __check_min_rate(td, now, DDIR_READ); |
55312f9f | 241 | if (td->bytes_done[DDIR_WRITE]) |
6eaf09d6 | 242 | ret |= __check_min_rate(td, now, DDIR_WRITE); |
55312f9f | 243 | if (td->bytes_done[DDIR_TRIM]) |
6eaf09d6 | 244 | ret |= __check_min_rate(td, now, DDIR_TRIM); |
2e1df07d JA |
245 | |
246 | return ret; | |
247 | } | |
248 | ||
249 | /* | |
250 | * When job exits, we can cancel the in-flight IO if we are using async | |
251 | * io. Attempt to do so. | |
252 | */ | |
253 | static void cleanup_pending_aio(struct thread_data *td) | |
254 | { | |
2e1df07d JA |
255 | int r; |
256 | ||
257 | /* | |
258 | * get immediately available events, if any | |
259 | */ | |
55312f9f | 260 | r = io_u_queued_complete(td, 0); |
2e1df07d JA |
261 | if (r < 0) |
262 | return; | |
263 | ||
264 | /* | |
265 | * now cancel remaining active events | |
266 | */ | |
267 | if (td->io_ops->cancel) { | |
2ae0b204 JA |
268 | struct io_u *io_u; |
269 | int i; | |
2e1df07d | 270 | |
2ae0b204 JA |
271 | io_u_qiter(&td->io_u_all, io_u, i) { |
272 | if (io_u->flags & IO_U_F_FLIGHT) { | |
2e1df07d JA |
273 | r = td->io_ops->cancel(td, io_u); |
274 | if (!r) | |
275 | put_io_u(td, io_u); | |
276 | } | |
277 | } | |
278 | } | |
279 | ||
280 | if (td->cur_depth) | |
55312f9f | 281 | r = io_u_queued_complete(td, td->cur_depth); |
2e1df07d JA |
282 | } |
283 | ||
284 | /* | |
285 | * Helper to handle the final sync of a file. Works just like the normal | |
286 | * io path, just does everything sync. | |
287 | */ | |
288 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
289 | { | |
290 | struct io_u *io_u = __get_io_u(td); | |
291 | int ret; | |
292 | ||
293 | if (!io_u) | |
294 | return 1; | |
295 | ||
296 | io_u->ddir = DDIR_SYNC; | |
297 | io_u->file = f; | |
298 | ||
299 | if (td_io_prep(td, io_u)) { | |
300 | put_io_u(td, io_u); | |
301 | return 1; | |
302 | } | |
303 | ||
304 | requeue: | |
305 | ret = td_io_queue(td, io_u); | |
306 | if (ret < 0) { | |
307 | td_verror(td, io_u->error, "td_io_queue"); | |
308 | put_io_u(td, io_u); | |
309 | return 1; | |
310 | } else if (ret == FIO_Q_QUEUED) { | |
55312f9f | 311 | if (io_u_queued_complete(td, 1) < 0) |
2e1df07d JA |
312 | return 1; |
313 | } else if (ret == FIO_Q_COMPLETED) { | |
314 | if (io_u->error) { | |
315 | td_verror(td, io_u->error, "td_io_queue"); | |
316 | return 1; | |
317 | } | |
318 | ||
55312f9f | 319 | if (io_u_sync_complete(td, io_u) < 0) |
2e1df07d JA |
320 | return 1; |
321 | } else if (ret == FIO_Q_BUSY) { | |
322 | if (td_io_commit(td)) | |
323 | return 1; | |
324 | goto requeue; | |
325 | } | |
326 | ||
327 | return 0; | |
328 | } | |
a3efc919 | 329 | |
61ee0f86 JA |
330 | static int fio_file_fsync(struct thread_data *td, struct fio_file *f) |
331 | { | |
332 | int ret; | |
333 | ||
334 | if (fio_file_open(f)) | |
335 | return fio_io_sync(td, f); | |
336 | ||
337 | if (td_io_open_file(td, f)) | |
338 | return 1; | |
339 | ||
340 | ret = fio_io_sync(td, f); | |
341 | td_io_close_file(td, f); | |
342 | return ret; | |
343 | } | |
344 | ||
2e1df07d JA |
345 | static inline void __update_tv_cache(struct thread_data *td) |
346 | { | |
347 | fio_gettime(&td->tv_cache, NULL); | |
348 | } | |
349 | ||
350 | static inline void update_tv_cache(struct thread_data *td) | |
351 | { | |
352 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
353 | __update_tv_cache(td); | |
354 | } | |
355 | ||
356 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
357 | { | |
358 | if (in_ramp_time(td)) | |
359 | return 0; | |
360 | if (!td->o.timeout) | |
361 | return 0; | |
0de5b26f | 362 | if (utime_since(&td->epoch, t) >= td->o.timeout) |
2e1df07d JA |
363 | return 1; |
364 | ||
365 | return 0; | |
366 | } | |
367 | ||
95603b74 BF |
368 | /* |
369 | * We need to update the runtime consistently in ms, but keep a running | |
370 | * tally of the current elapsed time in microseconds for sub millisecond | |
371 | * updates. | |
372 | */ | |
373 | static inline void update_runtime(struct thread_data *td, | |
374 | unsigned long long *elapsed_us, | |
375 | const enum fio_ddir ddir) | |
376 | { | |
d5c8ea29 JA |
377 | if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only) |
378 | return; | |
379 | ||
95603b74 BF |
380 | td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000; |
381 | elapsed_us[ddir] += utime_since_now(&td->start); | |
382 | td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000; | |
383 | } | |
384 | ||
2e1df07d JA |
385 | static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, |
386 | int *retptr) | |
387 | { | |
388 | int ret = *retptr; | |
389 | ||
390 | if (ret < 0 || td->error) { | |
8b28bd41 DM |
391 | int err = td->error; |
392 | enum error_type_bit eb; | |
2e1df07d JA |
393 | |
394 | if (ret < 0) | |
395 | err = -ret; | |
2e1df07d | 396 | |
8b28bd41 DM |
397 | eb = td_error_type(ddir, err); |
398 | if (!(td->o.continue_on_error & (1 << eb))) | |
2e1df07d JA |
399 | return 1; |
400 | ||
8b28bd41 | 401 | if (td_non_fatal_error(td, eb, err)) { |
2e1df07d JA |
402 | /* |
403 | * Continue with the I/Os in case of | |
404 | * a non fatal error. | |
405 | */ | |
406 | update_error_count(td, err); | |
407 | td_clear_error(td); | |
408 | *retptr = 0; | |
409 | return 0; | |
410 | } else if (td->o.fill_device && err == ENOSPC) { | |
411 | /* | |
412 | * We expect to hit this error if | |
413 | * fill_device option is set. | |
414 | */ | |
415 | td_clear_error(td); | |
ebea2133 | 416 | fio_mark_td_terminate(td); |
2e1df07d JA |
417 | return 1; |
418 | } else { | |
419 | /* | |
420 | * Stop the I/O in case of a fatal | |
421 | * error. | |
422 | */ | |
423 | update_error_count(td, err); | |
424 | return 1; | |
425 | } | |
426 | } | |
427 | ||
428 | return 0; | |
429 | } | |
430 | ||
c97f1ad6 JA |
431 | static void check_update_rusage(struct thread_data *td) |
432 | { | |
433 | if (td->update_rusage) { | |
434 | td->update_rusage = 0; | |
435 | update_rusage_stat(td); | |
436 | fio_mutex_up(td->rusage_sem); | |
437 | } | |
438 | } | |
439 | ||
55312f9f | 440 | static int wait_for_completions(struct thread_data *td, struct timeval *time) |
69fea81e JA |
441 | { |
442 | const int full = queue_full(td); | |
443 | int min_evts = 0; | |
444 | int ret; | |
445 | ||
446 | /* | |
447 | * if the queue is full, we MUST reap at least 1 event | |
448 | */ | |
449 | min_evts = min(td->o.iodepth_batch_complete, td->cur_depth); | |
77e7b330 | 450 | if ((full && !min_evts) || !td->o.iodepth_batch_complete) |
69fea81e JA |
451 | min_evts = 1; |
452 | ||
66b6c5ef | 453 | if (time && (__should_check_rate(td, DDIR_READ) || |
69fea81e JA |
454 | __should_check_rate(td, DDIR_WRITE) || |
455 | __should_check_rate(td, DDIR_TRIM))) | |
456 | fio_gettime(time, NULL); | |
457 | ||
458 | do { | |
55312f9f | 459 | ret = io_u_queued_complete(td, min_evts); |
69fea81e JA |
460 | if (ret < 0) |
461 | break; | |
462 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
463 | ||
464 | return ret; | |
465 | } | |
466 | ||
e9d512d8 JA |
467 | int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, |
468 | enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify, | |
469 | struct timeval *comp_time) | |
470 | { | |
471 | int ret2; | |
472 | ||
473 | switch (*ret) { | |
474 | case FIO_Q_COMPLETED: | |
475 | if (io_u->error) { | |
476 | *ret = -io_u->error; | |
477 | clear_io_u(td, io_u); | |
478 | } else if (io_u->resid) { | |
479 | int bytes = io_u->xfer_buflen - io_u->resid; | |
480 | struct fio_file *f = io_u->file; | |
481 | ||
482 | if (bytes_issued) | |
483 | *bytes_issued += bytes; | |
484 | ||
485 | if (!from_verify) | |
486 | trim_io_piece(td, io_u); | |
487 | ||
488 | /* | |
489 | * zero read, fail | |
490 | */ | |
491 | if (!bytes) { | |
492 | if (!from_verify) | |
493 | unlog_io_piece(td, io_u); | |
494 | td_verror(td, EIO, "full resid"); | |
495 | put_io_u(td, io_u); | |
496 | break; | |
497 | } | |
498 | ||
499 | io_u->xfer_buflen = io_u->resid; | |
500 | io_u->xfer_buf += bytes; | |
501 | io_u->offset += bytes; | |
502 | ||
503 | if (ddir_rw(io_u->ddir)) | |
504 | td->ts.short_io_u[io_u->ddir]++; | |
505 | ||
506 | f = io_u->file; | |
507 | if (io_u->offset == f->real_file_size) | |
508 | goto sync_done; | |
509 | ||
510 | requeue_io_u(td, &io_u); | |
511 | } else { | |
512 | sync_done: | |
513 | if (comp_time && (__should_check_rate(td, DDIR_READ) || | |
514 | __should_check_rate(td, DDIR_WRITE) || | |
515 | __should_check_rate(td, DDIR_TRIM))) | |
516 | fio_gettime(comp_time, NULL); | |
517 | ||
518 | *ret = io_u_sync_complete(td, io_u); | |
519 | if (*ret < 0) | |
520 | break; | |
521 | } | |
522 | return 0; | |
523 | case FIO_Q_QUEUED: | |
524 | /* | |
525 | * if the engine doesn't have a commit hook, | |
526 | * the io_u is really queued. if it does have such | |
527 | * a hook, it has to call io_u_queued() itself. | |
528 | */ | |
529 | if (td->io_ops->commit == NULL) | |
530 | io_u_queued(td, io_u); | |
531 | if (bytes_issued) | |
532 | *bytes_issued += io_u->xfer_buflen; | |
533 | break; | |
534 | case FIO_Q_BUSY: | |
535 | if (!from_verify) | |
536 | unlog_io_piece(td, io_u); | |
537 | requeue_io_u(td, &io_u); | |
538 | ret2 = td_io_commit(td); | |
539 | if (ret2 < 0) | |
540 | *ret = ret2; | |
541 | break; | |
542 | default: | |
b0775325 | 543 | assert(*ret < 0); |
e9d512d8 JA |
544 | td_verror(td, -(*ret), "td_io_queue"); |
545 | break; | |
546 | } | |
547 | ||
548 | if (break_on_this_error(td, ddir, ret)) | |
549 | return 1; | |
550 | ||
551 | return 0; | |
552 | } | |
553 | ||
2e1df07d JA |
554 | /* |
555 | * The main verify engine. Runs over the writes we previously submitted, | |
556 | * reads the blocks back in, and checks the crc/md5 of the data. | |
557 | */ | |
100f49f1 | 558 | static void do_verify(struct thread_data *td, uint64_t verify_bytes) |
2e1df07d JA |
559 | { |
560 | struct fio_file *f; | |
561 | struct io_u *io_u; | |
562 | int ret, min_events; | |
563 | unsigned int i; | |
564 | ||
565 | dprint(FD_VERIFY, "starting loop\n"); | |
566 | ||
567 | /* | |
568 | * sync io first and invalidate cache, to make sure we really | |
569 | * read from disk. | |
570 | */ | |
571 | for_each_file(td, f, i) { | |
572 | if (!fio_file_open(f)) | |
573 | continue; | |
574 | if (fio_io_sync(td, f)) | |
575 | break; | |
576 | if (file_invalidate_cache(td, f)) | |
577 | break; | |
578 | } | |
579 | ||
c97f1ad6 JA |
580 | check_update_rusage(td); |
581 | ||
2e1df07d JA |
582 | if (td->error) |
583 | return; | |
584 | ||
585 | td_set_runstate(td, TD_VERIFYING); | |
586 | ||
587 | io_u = NULL; | |
588 | while (!td->terminate) { | |
fbccf46c | 589 | enum fio_ddir ddir; |
e9d512d8 | 590 | int full; |
2e1df07d JA |
591 | |
592 | update_tv_cache(td); | |
c97f1ad6 | 593 | check_update_rusage(td); |
2e1df07d JA |
594 | |
595 | if (runtime_exceeded(td, &td->tv_cache)) { | |
596 | __update_tv_cache(td); | |
597 | if (runtime_exceeded(td, &td->tv_cache)) { | |
ebea2133 | 598 | fio_mark_td_terminate(td); |
2e1df07d JA |
599 | break; |
600 | } | |
601 | } | |
602 | ||
9e684a49 DE |
603 | if (flow_threshold_exceeded(td)) |
604 | continue; | |
605 | ||
44cbc6da JA |
606 | if (!td->o.experimental_verify) { |
607 | io_u = __get_io_u(td); | |
608 | if (!io_u) | |
609 | break; | |
2e1df07d | 610 | |
44cbc6da JA |
611 | if (get_next_verify(td, io_u)) { |
612 | put_io_u(td, io_u); | |
613 | break; | |
614 | } | |
2e1df07d | 615 | |
44cbc6da JA |
616 | if (td_io_prep(td, io_u)) { |
617 | put_io_u(td, io_u); | |
618 | break; | |
619 | } | |
620 | } else { | |
55312f9f | 621 | if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes) |
100f49f1 JA |
622 | break; |
623 | ||
bcd5abfa | 624 | while ((io_u = get_io_u(td)) != NULL) { |
002fe734 JA |
625 | if (IS_ERR(io_u)) { |
626 | io_u = NULL; | |
627 | ret = FIO_Q_BUSY; | |
628 | goto reap; | |
629 | } | |
630 | ||
bcd5abfa JA |
631 | /* |
632 | * We are only interested in the places where | |
633 | * we wrote or trimmed IOs. Turn those into | |
634 | * reads for verification purposes. | |
635 | */ | |
636 | if (io_u->ddir == DDIR_READ) { | |
637 | /* | |
638 | * Pretend we issued it for rwmix | |
639 | * accounting | |
640 | */ | |
641 | td->io_issues[DDIR_READ]++; | |
642 | put_io_u(td, io_u); | |
643 | continue; | |
644 | } else if (io_u->ddir == DDIR_TRIM) { | |
645 | io_u->ddir = DDIR_READ; | |
a9da8ab2 | 646 | io_u_set(io_u, IO_U_F_TRIMMED); |
bcd5abfa JA |
647 | break; |
648 | } else if (io_u->ddir == DDIR_WRITE) { | |
649 | io_u->ddir = DDIR_READ; | |
650 | break; | |
651 | } else { | |
652 | put_io_u(td, io_u); | |
653 | continue; | |
654 | } | |
655 | } | |
44cbc6da | 656 | |
bcd5abfa | 657 | if (!io_u) |
44cbc6da | 658 | break; |
2e1df07d JA |
659 | } |
660 | ||
ca09be4b JA |
661 | if (verify_state_should_stop(td, io_u)) { |
662 | put_io_u(td, io_u); | |
663 | break; | |
664 | } | |
665 | ||
2e1df07d JA |
666 | if (td->o.verify_async) |
667 | io_u->end_io = verify_io_u_async; | |
668 | else | |
669 | io_u->end_io = verify_io_u; | |
670 | ||
fbccf46c | 671 | ddir = io_u->ddir; |
10adc4a7 GG |
672 | if (!td->o.disable_slat) |
673 | fio_gettime(&io_u->start_time, NULL); | |
fbccf46c | 674 | |
2e1df07d | 675 | ret = td_io_queue(td, io_u); |
2e1df07d | 676 | |
e9d512d8 | 677 | if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL)) |
2e1df07d JA |
678 | break; |
679 | ||
680 | /* | |
681 | * if we can queue more, do so. but check if there are | |
682 | * completed io_u's first. Note that we can get BUSY even | |
683 | * without IO queued, if the system is resource starved. | |
684 | */ | |
002fe734 | 685 | reap: |
2e1df07d | 686 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
69fea81e | 687 | if (full || !td->o.iodepth_batch_complete) |
55312f9f | 688 | ret = wait_for_completions(td, NULL); |
2e1df07d | 689 | |
2e1df07d JA |
690 | if (ret < 0) |
691 | break; | |
692 | } | |
693 | ||
c97f1ad6 JA |
694 | check_update_rusage(td); |
695 | ||
2e1df07d JA |
696 | if (!td->error) { |
697 | min_events = td->cur_depth; | |
698 | ||
699 | if (min_events) | |
55312f9f | 700 | ret = io_u_queued_complete(td, min_events); |
2e1df07d JA |
701 | } else |
702 | cleanup_pending_aio(td); | |
703 | ||
704 | td_set_runstate(td, TD_RUNNING); | |
705 | ||
706 | dprint(FD_VERIFY, "exiting loop\n"); | |
707 | } | |
708 | ||
3939fe85 JA |
709 | static unsigned int exceeds_number_ios(struct thread_data *td) |
710 | { | |
711 | unsigned long long number_ios; | |
712 | ||
713 | if (!td->o.number_ios) | |
714 | return 0; | |
715 | ||
cf8a46a7 | 716 | number_ios = ddir_rw_sum(td->io_blocks); |
3939fe85 JA |
717 | number_ios += td->io_u_queued + td->io_u_in_flight; |
718 | ||
cf8a46a7 | 719 | return number_ios >= (td->o.number_ios * td->loops); |
3939fe85 JA |
720 | } |
721 | ||
e28dd2cf | 722 | static int io_issue_bytes_exceeded(struct thread_data *td) |
f7078f7b | 723 | { |
77731b29 | 724 | unsigned long long bytes, limit; |
f7078f7b JA |
725 | |
726 | if (td_rw(td)) | |
74d6277f | 727 | bytes = td->io_issue_bytes[DDIR_READ] + td->io_issue_bytes[DDIR_WRITE]; |
f7078f7b | 728 | else if (td_write(td)) |
74d6277f | 729 | bytes = td->io_issue_bytes[DDIR_WRITE]; |
6eaf09d6 | 730 | else if (td_read(td)) |
74d6277f | 731 | bytes = td->io_issue_bytes[DDIR_READ]; |
f7078f7b | 732 | else |
74d6277f | 733 | bytes = td->io_issue_bytes[DDIR_TRIM]; |
f7078f7b | 734 | |
77731b29 JA |
735 | if (td->o.io_limit) |
736 | limit = td->o.io_limit; | |
737 | else | |
738 | limit = td->o.size; | |
739 | ||
cf8a46a7 | 740 | limit *= td->loops; |
77731b29 | 741 | return bytes >= limit || exceeds_number_ios(td); |
f7078f7b JA |
742 | } |
743 | ||
e28dd2cf JE |
744 | static int io_complete_bytes_exceeded(struct thread_data *td) |
745 | { | |
746 | unsigned long long bytes, limit; | |
747 | ||
748 | if (td_rw(td)) | |
749 | bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE]; | |
750 | else if (td_write(td)) | |
751 | bytes = td->this_io_bytes[DDIR_WRITE]; | |
752 | else if (td_read(td)) | |
753 | bytes = td->this_io_bytes[DDIR_READ]; | |
754 | else | |
755 | bytes = td->this_io_bytes[DDIR_TRIM]; | |
756 | ||
757 | if (td->o.io_limit) | |
758 | limit = td->o.io_limit; | |
759 | else | |
760 | limit = td->o.size; | |
761 | ||
cf8a46a7 | 762 | limit *= td->loops; |
e28dd2cf JE |
763 | return bytes >= limit || exceeds_number_ios(td); |
764 | } | |
765 | ||
50a8ce86 D |
766 | /* |
767 | * used to calculate the next io time for rate control | |
768 | * | |
769 | */ | |
770 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) | |
771 | { | |
772 | uint64_t secs, remainder, bps, bytes; | |
773 | ||
774 | assert(!(td->flags & TD_F_CHILD)); | |
775 | bytes = td->rate_io_issue_bytes[ddir]; | |
776 | bps = td->rate_bps[ddir]; | |
777 | if (bps) { | |
778 | secs = bytes / bps; | |
779 | remainder = bytes % bps; | |
780 | return remainder * 1000000 / bps + secs * 1000000; | |
781 | } else | |
782 | return 0; | |
783 | } | |
784 | ||
2e1df07d JA |
785 | /* |
786 | * Main IO worker function. It retrieves io_u's to process and queues | |
787 | * and reaps them, checking for rate and errors along the way. | |
100f49f1 JA |
788 | * |
789 | * Returns number of bytes written and trimmed. | |
2e1df07d | 790 | */ |
100f49f1 | 791 | static uint64_t do_io(struct thread_data *td) |
2e1df07d JA |
792 | { |
793 | unsigned int i; | |
794 | int ret = 0; | |
c2703bf3 | 795 | uint64_t total_bytes, bytes_issued = 0; |
2e1df07d JA |
796 | |
797 | if (in_ramp_time(td)) | |
798 | td_set_runstate(td, TD_RAMP); | |
799 | else | |
800 | td_set_runstate(td, TD_RUNNING); | |
801 | ||
3e260a46 JA |
802 | lat_target_init(td); |
803 | ||
1e564979 JE |
804 | total_bytes = td->o.size; |
805 | /* | |
806 | * Allow random overwrite workloads to write up to io_limit | |
807 | * before starting verification phase as 'size' doesn't apply. | |
808 | */ | |
809 | if (td_write(td) && td_random(td) && td->o.norandommap) | |
810 | total_bytes = max(total_bytes, (uint64_t) td->o.io_limit); | |
78a6469c JA |
811 | /* |
812 | * If verify_backlog is enabled, we'll run the verify in this | |
813 | * handler as well. For that case, we may need up to twice the | |
814 | * amount of bytes. | |
815 | */ | |
78a6469c JA |
816 | if (td->o.verify != VERIFY_NONE && |
817 | (td_write(td) && td->o.verify_backlog)) | |
c2703bf3 JA |
818 | total_bytes += td->o.size; |
819 | ||
82a90686 | 820 | /* In trimwrite mode, each byte is trimmed and then written, so |
0e4dd95c | 821 | * allow total_bytes to be twice as big */ |
82a90686 | 822 | if (td_trimwrite(td)) |
0e4dd95c DE |
823 | total_bytes += td->total_io_size; |
824 | ||
f7078f7b | 825 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
e28dd2cf | 826 | (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) || |
c04e4661 | 827 | td->o.time_based) { |
2e1df07d | 828 | struct timeval comp_time; |
2e1df07d | 829 | struct io_u *io_u; |
e9d512d8 | 830 | int full; |
2e1df07d JA |
831 | enum fio_ddir ddir; |
832 | ||
c97f1ad6 JA |
833 | check_update_rusage(td); |
834 | ||
7d7803fa | 835 | if (td->terminate || td->done) |
2e1df07d JA |
836 | break; |
837 | ||
838 | update_tv_cache(td); | |
839 | ||
840 | if (runtime_exceeded(td, &td->tv_cache)) { | |
841 | __update_tv_cache(td); | |
842 | if (runtime_exceeded(td, &td->tv_cache)) { | |
ebea2133 | 843 | fio_mark_td_terminate(td); |
2e1df07d JA |
844 | break; |
845 | } | |
846 | } | |
847 | ||
9e684a49 DE |
848 | if (flow_threshold_exceeded(td)) |
849 | continue; | |
850 | ||
c2703bf3 | 851 | if (bytes_issued >= total_bytes) |
20876c53 JC |
852 | break; |
853 | ||
2e1df07d | 854 | io_u = get_io_u(td); |
002fe734 JA |
855 | if (IS_ERR_OR_NULL(io_u)) { |
856 | int err = PTR_ERR(io_u); | |
857 | ||
858 | io_u = NULL; | |
859 | if (err == -EBUSY) { | |
860 | ret = FIO_Q_BUSY; | |
861 | goto reap; | |
862 | } | |
3e260a46 JA |
863 | if (td->o.latency_target) |
864 | goto reap; | |
2e1df07d | 865 | break; |
3e260a46 | 866 | } |
2e1df07d JA |
867 | |
868 | ddir = io_u->ddir; | |
869 | ||
870 | /* | |
82af2a7c JA |
871 | * Add verification end_io handler if: |
872 | * - Asked to verify (!td_rw(td)) | |
873 | * - Or the io_u is from our verify list (mixed write/ver) | |
2e1df07d JA |
874 | */ |
875 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && | |
82af2a7c | 876 | ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { |
c4b6117b PV |
877 | |
878 | if (!td->o.verify_pattern_bytes) { | |
d6b72507 | 879 | io_u->rand_seed = __rand(&td->verify_state); |
c4b6117b | 880 | if (sizeof(int) != sizeof(long *)) |
d6b72507 | 881 | io_u->rand_seed *= __rand(&td->verify_state); |
c4b6117b PV |
882 | } |
883 | ||
ca09be4b JA |
884 | if (verify_state_should_stop(td, io_u)) { |
885 | put_io_u(td, io_u); | |
886 | break; | |
887 | } | |
888 | ||
2e1df07d JA |
889 | if (td->o.verify_async) |
890 | io_u->end_io = verify_io_u_async; | |
891 | else | |
892 | io_u->end_io = verify_io_u; | |
893 | td_set_runstate(td, TD_VERIFYING); | |
894 | } else if (in_ramp_time(td)) | |
895 | td_set_runstate(td, TD_RAMP); | |
896 | else | |
897 | td_set_runstate(td, TD_RUNNING); | |
898 | ||
9a50c5c5 | 899 | /* |
f9401285 JA |
900 | * Always log IO before it's issued, so we know the specific |
901 | * order of it. The logged unit will track when the IO has | |
902 | * completed. | |
9a50c5c5 | 903 | */ |
c4b6117b PV |
904 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
905 | td->o.do_verify && | |
906 | td->o.verify != VERIFY_NONE && | |
f9401285 | 907 | !td->o.experimental_verify) |
c4b6117b PV |
908 | log_io_piece(td, io_u); |
909 | ||
a9da8ab2 JA |
910 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
911 | if (td->error) | |
912 | break; | |
913 | ret = workqueue_enqueue(&td->io_wq, io_u); | |
50a8ce86 D |
914 | |
915 | if (should_check_rate(td)) | |
916 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
917 | ||
a9da8ab2 JA |
918 | } else { |
919 | ret = td_io_queue(td, io_u); | |
2e1df07d | 920 | |
50a8ce86 D |
921 | if (should_check_rate(td)) |
922 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
923 | ||
fd727d9d | 924 | if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time)) |
a9da8ab2 | 925 | break; |
2e1df07d | 926 | |
a9da8ab2 JA |
927 | /* |
928 | * See if we need to complete some commands. Note that | |
929 | * we can get BUSY even without IO queued, if the | |
930 | * system is resource starved. | |
931 | */ | |
3e260a46 | 932 | reap: |
a9da8ab2 JA |
933 | full = queue_full(td) || |
934 | (ret == FIO_Q_BUSY && td->cur_depth); | |
935 | if (full || !td->o.iodepth_batch_complete) | |
936 | ret = wait_for_completions(td, &comp_time); | |
937 | } | |
2e1df07d JA |
938 | if (ret < 0) |
939 | break; | |
55312f9f JA |
940 | if (!ddir_rw_sum(td->bytes_done) && |
941 | !(td->io_ops->flags & FIO_NOIO)) | |
2e1df07d JA |
942 | continue; |
943 | ||
55312f9f JA |
944 | if (!in_ramp_time(td) && should_check_rate(td)) { |
945 | if (check_min_rate(td, &comp_time)) { | |
2e1df07d JA |
946 | if (exitall_on_terminate) |
947 | fio_terminate_threads(td->groupid); | |
948 | td_verror(td, EIO, "check_min_rate"); | |
949 | break; | |
950 | } | |
951 | } | |
3e260a46 JA |
952 | if (!in_ramp_time(td) && td->o.latency_target) |
953 | lat_target_check(td); | |
e155cb64 | 954 | |
2e1df07d JA |
955 | if (td->o.thinktime) { |
956 | unsigned long long b; | |
957 | ||
342f4be4 | 958 | b = ddir_rw_sum(td->io_blocks); |
2e1df07d JA |
959 | if (!(b % td->o.thinktime_blocks)) { |
960 | int left; | |
961 | ||
002e7183 JA |
962 | io_u_quiesce(td); |
963 | ||
2e1df07d JA |
964 | if (td->o.thinktime_spin) |
965 | usec_spin(td->o.thinktime_spin); | |
966 | ||
967 | left = td->o.thinktime - td->o.thinktime_spin; | |
968 | if (left) | |
969 | usec_sleep(td, left); | |
970 | } | |
971 | } | |
972 | } | |
973 | ||
c97f1ad6 JA |
974 | check_update_rusage(td); |
975 | ||
2e1df07d | 976 | if (td->trim_entries) |
4e0a8fa2 | 977 | log_err("fio: %lu trim entries leaked?\n", td->trim_entries); |
2e1df07d JA |
978 | |
979 | if (td->o.fill_device && td->error == ENOSPC) { | |
980 | td->error = 0; | |
ebea2133 | 981 | fio_mark_td_terminate(td); |
2e1df07d JA |
982 | } |
983 | if (!td->error) { | |
984 | struct fio_file *f; | |
985 | ||
a9da8ab2 JA |
986 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
987 | workqueue_flush(&td->io_wq); | |
988 | i = 0; | |
989 | } else | |
990 | i = td->cur_depth; | |
991 | ||
2e1df07d | 992 | if (i) { |
55312f9f | 993 | ret = io_u_queued_complete(td, i); |
2e1df07d JA |
994 | if (td->o.fill_device && td->error == ENOSPC) |
995 | td->error = 0; | |
996 | } | |
997 | ||
998 | if (should_fsync(td) && td->o.end_fsync) { | |
999 | td_set_runstate(td, TD_FSYNCING); | |
1000 | ||
1001 | for_each_file(td, f, i) { | |
61ee0f86 | 1002 | if (!fio_file_fsync(td, f)) |
2e1df07d | 1003 | continue; |
61ee0f86 JA |
1004 | |
1005 | log_err("fio: end_fsync failed for file %s\n", | |
1006 | f->file_name); | |
2e1df07d JA |
1007 | } |
1008 | } | |
1009 | } else | |
1010 | cleanup_pending_aio(td); | |
1011 | ||
1012 | /* | |
1013 | * stop job if we failed doing any IO | |
1014 | */ | |
342f4be4 | 1015 | if (!ddir_rw_sum(td->this_io_bytes)) |
2e1df07d | 1016 | td->done = 1; |
100f49f1 | 1017 | |
55312f9f | 1018 | return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; |
2e1df07d JA |
1019 | } |
1020 | ||
1021 | static void cleanup_io_u(struct thread_data *td) | |
1022 | { | |
2e1df07d JA |
1023 | struct io_u *io_u; |
1024 | ||
2ae0b204 | 1025 | while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) { |
c73ed246 JA |
1026 | |
1027 | if (td->io_ops->io_u_free) | |
1028 | td->io_ops->io_u_free(td, io_u); | |
1029 | ||
2e1df07d JA |
1030 | fio_memfree(io_u, sizeof(*io_u)); |
1031 | } | |
1032 | ||
1033 | free_io_mem(td); | |
2ae0b204 JA |
1034 | |
1035 | io_u_rexit(&td->io_u_requeues); | |
1036 | io_u_qexit(&td->io_u_freelist); | |
1037 | io_u_qexit(&td->io_u_all); | |
ca09be4b JA |
1038 | |
1039 | if (td->last_write_comp) | |
1040 | sfree(td->last_write_comp); | |
2e1df07d JA |
1041 | } |
1042 | ||
1043 | static int init_io_u(struct thread_data *td) | |
1044 | { | |
1045 | struct io_u *io_u; | |
9c42684e | 1046 | unsigned int max_bs, min_write; |
2e1df07d | 1047 | int cl_align, i, max_units; |
2ae0b204 | 1048 | int data_xfer = 1, err; |
2e1df07d JA |
1049 | char *p; |
1050 | ||
1051 | max_units = td->o.iodepth; | |
74f4b020 | 1052 | max_bs = td_max_bs(td); |
9c42684e | 1053 | min_write = td->o.min_bs[DDIR_WRITE]; |
2e1df07d JA |
1054 | td->orig_buffer_size = (unsigned long long) max_bs |
1055 | * (unsigned long long) max_units; | |
1056 | ||
88045e04 | 1057 | if ((td->io_ops->flags & FIO_NOIO) || !(td_read(td) || td_write(td))) |
59d8d0f5 JA |
1058 | data_xfer = 0; |
1059 | ||
2ae0b204 JA |
1060 | err = 0; |
1061 | err += io_u_rinit(&td->io_u_requeues, td->o.iodepth); | |
1062 | err += io_u_qinit(&td->io_u_freelist, td->o.iodepth); | |
1063 | err += io_u_qinit(&td->io_u_all, td->o.iodepth); | |
1064 | ||
1065 | if (err) { | |
1066 | log_err("fio: failed setting up IO queues\n"); | |
1067 | return 1; | |
1068 | } | |
1069 | ||
fd8a09b8 | 1070 | /* |
1071 | * if we may later need to do address alignment, then add any | |
1072 | * possible adjustment here so that we don't cause a buffer | |
1073 | * overflow later. this adjustment may be too much if we get | |
1074 | * lucky and the allocator gives us an aligned address. | |
1075 | */ | |
d01612f3 CM |
1076 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
1077 | (td->io_ops->flags & FIO_RAWIO)) | |
fd8a09b8 | 1078 | td->orig_buffer_size += page_mask + td->o.mem_align; |
1079 | ||
2e1df07d JA |
1080 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { |
1081 | unsigned long bs; | |
1082 | ||
1083 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; | |
1084 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
1085 | } | |
1086 | ||
1087 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { | |
1088 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
1089 | return 1; | |
1090 | } | |
1091 | ||
59d8d0f5 | 1092 | if (data_xfer && allocate_io_mem(td)) |
2e1df07d JA |
1093 | return 1; |
1094 | ||
d01612f3 | 1095 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
2e1df07d JA |
1096 | (td->io_ops->flags & FIO_RAWIO)) |
1097 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; | |
1098 | else | |
1099 | p = td->orig_buffer; | |
1100 | ||
1101 | cl_align = os_cache_line_size(); | |
1102 | ||
1103 | for (i = 0; i < max_units; i++) { | |
1104 | void *ptr; | |
1105 | ||
1106 | if (td->terminate) | |
1107 | return 1; | |
1108 | ||
1109 | ptr = fio_memalign(cl_align, sizeof(*io_u)); | |
1110 | if (!ptr) { | |
1111 | log_err("fio: unable to allocate aligned memory\n"); | |
1112 | break; | |
1113 | } | |
1114 | ||
1115 | io_u = ptr; | |
1116 | memset(io_u, 0, sizeof(*io_u)); | |
2ae0b204 | 1117 | INIT_FLIST_HEAD(&io_u->verify_list); |
2e1df07d JA |
1118 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
1119 | ||
59d8d0f5 | 1120 | if (data_xfer) { |
2e1df07d JA |
1121 | io_u->buf = p; |
1122 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); | |
1123 | ||
1124 | if (td_write(td)) | |
9c42684e | 1125 | io_u_fill_buffer(td, io_u, min_write, max_bs); |
2e1df07d JA |
1126 | if (td_write(td) && td->o.verify_pattern_bytes) { |
1127 | /* | |
1128 | * Fill the buffer with the pattern if we are | |
1129 | * going to be doing writes. | |
1130 | */ | |
ce35b1ec | 1131 | fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
2e1df07d JA |
1132 | } |
1133 | } | |
1134 | ||
1135 | io_u->index = i; | |
1136 | io_u->flags = IO_U_F_FREE; | |
2ae0b204 JA |
1137 | io_u_qpush(&td->io_u_freelist, io_u); |
1138 | ||
1139 | /* | |
1140 | * io_u never leaves this stack, used for iteration of all | |
1141 | * io_u buffers. | |
1142 | */ | |
1143 | io_u_qpush(&td->io_u_all, io_u); | |
c73ed246 JA |
1144 | |
1145 | if (td->io_ops->io_u_init) { | |
1146 | int ret = td->io_ops->io_u_init(td, io_u); | |
1147 | ||
1148 | if (ret) { | |
1149 | log_err("fio: failed to init engine data: %d\n", ret); | |
1150 | return 1; | |
1151 | } | |
1152 | } | |
1153 | ||
2e1df07d JA |
1154 | p += max_bs; |
1155 | } | |
1156 | ||
ca09be4b JA |
1157 | if (td->o.verify != VERIFY_NONE) { |
1158 | td->last_write_comp = scalloc(max_units, sizeof(uint64_t)); | |
1159 | if (!td->last_write_comp) { | |
1160 | log_err("fio: failed to alloc write comp data\n"); | |
1161 | return 1; | |
1162 | } | |
1163 | } | |
1164 | ||
2e1df07d JA |
1165 | return 0; |
1166 | } | |
1167 | ||
1168 | static int switch_ioscheduler(struct thread_data *td) | |
1169 | { | |
1170 | char tmp[256], tmp2[128]; | |
1171 | FILE *f; | |
1172 | int ret; | |
1173 | ||
1174 | if (td->io_ops->flags & FIO_DISKLESSIO) | |
1175 | return 0; | |
1176 | ||
1177 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
1178 | ||
1179 | f = fopen(tmp, "r+"); | |
1180 | if (!f) { | |
1181 | if (errno == ENOENT) { | |
1182 | log_err("fio: os or kernel doesn't support IO scheduler" | |
1183 | " switching\n"); | |
1184 | return 0; | |
1185 | } | |
1186 | td_verror(td, errno, "fopen iosched"); | |
1187 | return 1; | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * Set io scheduler. | |
1192 | */ | |
1193 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); | |
1194 | if (ferror(f) || ret != 1) { | |
1195 | td_verror(td, errno, "fwrite"); | |
1196 | fclose(f); | |
1197 | return 1; | |
1198 | } | |
1199 | ||
1200 | rewind(f); | |
1201 | ||
1202 | /* | |
1203 | * Read back and check that the selected scheduler is now the default. | |
1204 | */ | |
b44b9e45 | 1205 | memset(tmp, 0, sizeof(tmp)); |
49c6f33d | 1206 | ret = fread(tmp, sizeof(tmp), 1, f); |
2e1df07d JA |
1207 | if (ferror(f) || ret < 0) { |
1208 | td_verror(td, errno, "fread"); | |
1209 | fclose(f); | |
1210 | return 1; | |
1211 | } | |
b44b9e45 TK |
1212 | /* |
1213 | * either a list of io schedulers or "none\n" is expected. | |
1214 | */ | |
1215 | tmp[strlen(tmp) - 1] = '\0'; | |
49c6f33d | 1216 | |
2e1df07d JA |
1217 | |
1218 | sprintf(tmp2, "[%s]", td->o.ioscheduler); | |
1219 | if (!strstr(tmp, tmp2)) { | |
1220 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); | |
1221 | td_verror(td, EINVAL, "iosched_switch"); | |
1222 | fclose(f); | |
1223 | return 1; | |
1224 | } | |
1225 | ||
1226 | fclose(f); | |
1227 | return 0; | |
1228 | } | |
1229 | ||
1230 | static int keep_running(struct thread_data *td) | |
1231 | { | |
77731b29 JA |
1232 | unsigned long long limit; |
1233 | ||
2e1df07d JA |
1234 | if (td->done) |
1235 | return 0; | |
1236 | if (td->o.time_based) | |
1237 | return 1; | |
1238 | if (td->o.loops) { | |
1239 | td->o.loops--; | |
1240 | return 1; | |
1241 | } | |
3939fe85 JA |
1242 | if (exceeds_number_ios(td)) |
1243 | return 0; | |
26251d8d | 1244 | |
77731b29 JA |
1245 | if (td->o.io_limit) |
1246 | limit = td->o.io_limit; | |
1247 | else | |
1248 | limit = td->o.size; | |
1249 | ||
1250 | if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) { | |
5bd5f71a JA |
1251 | uint64_t diff; |
1252 | ||
1253 | /* | |
1254 | * If the difference is less than the minimum IO size, we | |
1255 | * are done. | |
1256 | */ | |
77731b29 | 1257 | diff = limit - ddir_rw_sum(td->io_bytes); |
74f4b020 | 1258 | if (diff < td_max_bs(td)) |
5bd5f71a JA |
1259 | return 0; |
1260 | ||
002fe734 JA |
1261 | if (fio_files_done(td)) |
1262 | return 0; | |
1263 | ||
2e1df07d | 1264 | return 1; |
5bd5f71a | 1265 | } |
2e1df07d JA |
1266 | |
1267 | return 0; | |
1268 | } | |
1269 | ||
ce486495 | 1270 | static int exec_string(struct thread_options *o, const char *string, const char *mode) |
2e1df07d | 1271 | { |
6c9ce469 | 1272 | size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; |
f491a907 | 1273 | int ret; |
2e1df07d JA |
1274 | char *str; |
1275 | ||
1276 | str = malloc(newlen); | |
ce486495 | 1277 | sprintf(str, "%s &> %s.%s.txt", string, o->name, mode); |
2e1df07d | 1278 | |
ce486495 | 1279 | log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode); |
2e1df07d JA |
1280 | ret = system(str); |
1281 | if (ret == -1) | |
1282 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1283 | ||
1284 | free(str); | |
1285 | return ret; | |
1286 | } | |
1287 | ||
62167762 JC |
1288 | /* |
1289 | * Dry run to compute correct state of numberio for verification. | |
1290 | */ | |
1291 | static uint64_t do_dry_run(struct thread_data *td) | |
1292 | { | |
62167762 JC |
1293 | td_set_runstate(td, TD_RUNNING); |
1294 | ||
1295 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || | |
e28dd2cf | 1296 | (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) { |
62167762 JC |
1297 | struct io_u *io_u; |
1298 | int ret; | |
1299 | ||
1300 | if (td->terminate || td->done) | |
1301 | break; | |
1302 | ||
1303 | io_u = get_io_u(td); | |
1304 | if (!io_u) | |
1305 | break; | |
1306 | ||
a9da8ab2 | 1307 | io_u_set(io_u, IO_U_F_FLIGHT); |
62167762 JC |
1308 | io_u->error = 0; |
1309 | io_u->resid = 0; | |
1310 | if (ddir_rw(acct_ddir(io_u))) | |
1311 | td->io_issues[acct_ddir(io_u)]++; | |
1312 | if (ddir_rw(io_u->ddir)) { | |
1313 | io_u_mark_depth(td, 1); | |
1314 | td->ts.total_io_u[io_u->ddir]++; | |
1315 | } | |
1316 | ||
2e63e96b PV |
1317 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1318 | td->o.do_verify && | |
1319 | td->o.verify != VERIFY_NONE && | |
1320 | !td->o.experimental_verify) | |
1321 | log_io_piece(td, io_u); | |
1322 | ||
55312f9f | 1323 | ret = io_u_sync_complete(td, io_u); |
62167762 JC |
1324 | (void) ret; |
1325 | } | |
1326 | ||
55312f9f | 1327 | return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; |
62167762 JC |
1328 | } |
1329 | ||
a9da8ab2 JA |
1330 | static void io_workqueue_fn(struct thread_data *td, struct io_u *io_u) |
1331 | { | |
1332 | const enum fio_ddir ddir = io_u->ddir; | |
1333 | int ret; | |
1334 | ||
1335 | dprint(FD_RATE, "io_u %p queued by %u\n", io_u, gettid()); | |
1336 | ||
1337 | io_u_set(io_u, IO_U_F_NO_FILE_PUT); | |
1338 | ||
1339 | td->cur_depth++; | |
1340 | ||
1341 | ret = td_io_queue(td, io_u); | |
1342 | ||
1343 | dprint(FD_RATE, "io_u %p ret %d by %u\n", io_u, ret, gettid()); | |
1344 | ||
1345 | io_queue_event(td, io_u, &ret, ddir, NULL, 0, NULL); | |
1346 | ||
1347 | if (ret == FIO_Q_QUEUED) | |
1348 | ret = io_u_queued_complete(td, 1); | |
1349 | ||
1350 | td->cur_depth--; | |
1351 | } | |
1352 | ||
2e1df07d JA |
1353 | /* |
1354 | * Entry point for the thread based jobs. The process based jobs end up | |
1355 | * here as well, after a little setup. | |
1356 | */ | |
1357 | static void *thread_main(void *data) | |
1358 | { | |
95603b74 | 1359 | unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, }; |
2e1df07d | 1360 | struct thread_data *td = data; |
4896473e | 1361 | struct thread_options *o = &td->o; |
2e1df07d JA |
1362 | pthread_condattr_t attr; |
1363 | int clear_state; | |
28727df7 | 1364 | int ret; |
2e1df07d | 1365 | |
4896473e | 1366 | if (!o->use_thread) { |
2e1df07d JA |
1367 | setsid(); |
1368 | td->pid = getpid(); | |
1369 | } else | |
1370 | td->pid = gettid(); | |
1371 | ||
4896473e | 1372 | fio_local_clock_init(o->use_thread); |
5d879392 | 1373 | |
2e1df07d JA |
1374 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
1375 | ||
122c7725 JA |
1376 | if (is_backend) |
1377 | fio_server_send_start(td); | |
1378 | ||
2e1df07d JA |
1379 | INIT_FLIST_HEAD(&td->io_log_list); |
1380 | INIT_FLIST_HEAD(&td->io_hist_list); | |
1381 | INIT_FLIST_HEAD(&td->verify_list); | |
1382 | INIT_FLIST_HEAD(&td->trim_list); | |
1ae83d45 | 1383 | INIT_FLIST_HEAD(&td->next_rand_list); |
2e1df07d JA |
1384 | pthread_mutex_init(&td->io_u_lock, NULL); |
1385 | td->io_hist_tree = RB_ROOT; | |
1386 | ||
1387 | pthread_condattr_init(&attr); | |
1388 | pthread_cond_init(&td->verify_cond, &attr); | |
1389 | pthread_cond_init(&td->free_cond, &attr); | |
1390 | ||
1391 | td_set_runstate(td, TD_INITIALIZED); | |
1392 | dprint(FD_MUTEX, "up startup_mutex\n"); | |
1393 | fio_mutex_up(startup_mutex); | |
1394 | dprint(FD_MUTEX, "wait on td->mutex\n"); | |
1395 | fio_mutex_down(td->mutex); | |
1396 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); | |
1397 | ||
2e1df07d JA |
1398 | /* |
1399 | * A new gid requires privilege, so we need to do this before setting | |
1400 | * the uid. | |
1401 | */ | |
4896473e | 1402 | if (o->gid != -1U && setgid(o->gid)) { |
2e1df07d JA |
1403 | td_verror(td, errno, "setgid"); |
1404 | goto err; | |
1405 | } | |
4896473e | 1406 | if (o->uid != -1U && setuid(o->uid)) { |
2e1df07d JA |
1407 | td_verror(td, errno, "setuid"); |
1408 | goto err; | |
1409 | } | |
1410 | ||
1411 | /* | |
1412 | * If we have a gettimeofday() thread, make sure we exclude that | |
1413 | * thread from this job | |
1414 | */ | |
4896473e JA |
1415 | if (o->gtod_cpu) |
1416 | fio_cpu_clear(&o->cpumask, o->gtod_cpu); | |
2e1df07d JA |
1417 | |
1418 | /* | |
1419 | * Set affinity first, in case it has an impact on the memory | |
1420 | * allocations. | |
1421 | */ | |
b2a9e649 | 1422 | if (fio_option_is_set(o, cpumask)) { |
c2acfbac | 1423 | if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { |
30cb4c65 | 1424 | ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); |
c2acfbac JA |
1425 | if (!ret) { |
1426 | log_err("fio: no CPUs set\n"); | |
1427 | log_err("fio: Try increasing number of available CPUs\n"); | |
1428 | td_verror(td, EINVAL, "cpus_split"); | |
1429 | goto err; | |
1430 | } | |
1431 | } | |
28727df7 JA |
1432 | ret = fio_setaffinity(td->pid, o->cpumask); |
1433 | if (ret == -1) { | |
4896473e JA |
1434 | td_verror(td, errno, "cpu_set_affinity"); |
1435 | goto err; | |
1436 | } | |
2e1df07d JA |
1437 | } |
1438 | ||
67bf9823 | 1439 | #ifdef CONFIG_LIBNUMA |
d0b937ed | 1440 | /* numa node setup */ |
b2a9e649 JA |
1441 | if (fio_option_is_set(o, numa_cpunodes) || |
1442 | fio_option_is_set(o, numa_memnodes)) { | |
43522848 | 1443 | struct bitmask *mask; |
d0b937ed YR |
1444 | |
1445 | if (numa_available() < 0) { | |
1446 | td_verror(td, errno, "Does not support NUMA API\n"); | |
1447 | goto err; | |
1448 | } | |
1449 | ||
b2a9e649 | 1450 | if (fio_option_is_set(o, numa_cpunodes)) { |
43522848 DG |
1451 | mask = numa_parse_nodestring(o->numa_cpunodes); |
1452 | ret = numa_run_on_node_mask(mask); | |
1453 | numa_free_nodemask(mask); | |
d0b937ed YR |
1454 | if (ret == -1) { |
1455 | td_verror(td, errno, \ | |
1456 | "numa_run_on_node_mask failed\n"); | |
1457 | goto err; | |
1458 | } | |
1459 | } | |
1460 | ||
b2a9e649 | 1461 | if (fio_option_is_set(o, numa_memnodes)) { |
43522848 DG |
1462 | mask = NULL; |
1463 | if (o->numa_memnodes) | |
1464 | mask = numa_parse_nodestring(o->numa_memnodes); | |
1465 | ||
4896473e | 1466 | switch (o->numa_mem_mode) { |
d0b937ed | 1467 | case MPOL_INTERLEAVE: |
43522848 | 1468 | numa_set_interleave_mask(mask); |
d0b937ed YR |
1469 | break; |
1470 | case MPOL_BIND: | |
43522848 | 1471 | numa_set_membind(mask); |
d0b937ed YR |
1472 | break; |
1473 | case MPOL_LOCAL: | |
1474 | numa_set_localalloc(); | |
1475 | break; | |
1476 | case MPOL_PREFERRED: | |
4896473e | 1477 | numa_set_preferred(o->numa_mem_prefer_node); |
d0b937ed YR |
1478 | break; |
1479 | case MPOL_DEFAULT: | |
1480 | default: | |
1481 | break; | |
1482 | } | |
1483 | ||
43522848 DG |
1484 | if (mask) |
1485 | numa_free_nodemask(mask); | |
1486 | ||
d0b937ed YR |
1487 | } |
1488 | } | |
1489 | #endif | |
1490 | ||
9a3f1100 JA |
1491 | if (fio_pin_memory(td)) |
1492 | goto err; | |
1493 | ||
2e1df07d JA |
1494 | /* |
1495 | * May alter parameters that init_io_u() will use, so we need to | |
1496 | * do this first. | |
1497 | */ | |
1498 | if (init_iolog(td)) | |
1499 | goto err; | |
1500 | ||
1501 | if (init_io_u(td)) | |
1502 | goto err; | |
1503 | ||
4896473e | 1504 | if (o->verify_async && verify_async_init(td)) |
2e1df07d JA |
1505 | goto err; |
1506 | ||
27d74836 JA |
1507 | if (fio_option_is_set(o, ioprio) || |
1508 | fio_option_is_set(o, ioprio_class)) { | |
28727df7 JA |
1509 | ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio); |
1510 | if (ret == -1) { | |
2e1df07d JA |
1511 | td_verror(td, errno, "ioprio_set"); |
1512 | goto err; | |
1513 | } | |
1514 | } | |
1515 | ||
4896473e | 1516 | if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
2e1df07d JA |
1517 | goto err; |
1518 | ||
649c10c5 | 1519 | errno = 0; |
4896473e | 1520 | if (nice(o->nice) == -1 && errno != 0) { |
2e1df07d JA |
1521 | td_verror(td, errno, "nice"); |
1522 | goto err; | |
1523 | } | |
1524 | ||
4896473e | 1525 | if (o->ioscheduler && switch_ioscheduler(td)) |
2e1df07d JA |
1526 | goto err; |
1527 | ||
4896473e | 1528 | if (!o->create_serialize && setup_files(td)) |
2e1df07d JA |
1529 | goto err; |
1530 | ||
1531 | if (td_io_init(td)) | |
1532 | goto err; | |
1533 | ||
1534 | if (init_random_map(td)) | |
1535 | goto err; | |
1536 | ||
ce486495 | 1537 | if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun")) |
4896473e | 1538 | goto err; |
2e1df07d | 1539 | |
4896473e | 1540 | if (o->pre_read) { |
2e1df07d JA |
1541 | if (pre_read_files(td) < 0) |
1542 | goto err; | |
1543 | } | |
1544 | ||
aee2ab67 JA |
1545 | if (td->flags & TD_F_COMPRESS_LOG) |
1546 | tp_init(&td->tp_data); | |
1547 | ||
dc5bfbb2 JA |
1548 | fio_verify_init(td); |
1549 | ||
a9da8ab2 JA |
1550 | if ((o->io_submit_mode == IO_MODE_OFFLOAD) && |
1551 | workqueue_init(td, &td->io_wq, io_workqueue_fn, td->o.iodepth)) | |
1552 | goto err; | |
1553 | ||
2e1df07d | 1554 | fio_gettime(&td->epoch, NULL); |
44404c5a | 1555 | fio_getrusage(&td->ru_start); |
2e1df07d JA |
1556 | clear_state = 0; |
1557 | while (keep_running(td)) { | |
100f49f1 JA |
1558 | uint64_t verify_bytes; |
1559 | ||
2e1df07d JA |
1560 | fio_gettime(&td->start, NULL); |
1561 | memcpy(&td->bw_sample_time, &td->start, sizeof(td->start)); | |
1562 | memcpy(&td->iops_sample_time, &td->start, sizeof(td->start)); | |
1563 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); | |
1564 | ||
4896473e JA |
1565 | if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || |
1566 | o->ratemin[DDIR_TRIM]) { | |
6eaf09d6 | 1567 | memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, |
2e1df07d | 1568 | sizeof(td->bw_sample_time)); |
6eaf09d6 SL |
1569 | memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, |
1570 | sizeof(td->bw_sample_time)); | |
1571 | memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time, | |
2e1df07d JA |
1572 | sizeof(td->bw_sample_time)); |
1573 | } | |
1574 | ||
1575 | if (clear_state) | |
1576 | clear_io_state(td); | |
1577 | ||
1578 | prune_io_piece_log(td); | |
1579 | ||
62167762 JC |
1580 | if (td->o.verify_only && (td_write(td) || td_rw(td))) |
1581 | verify_bytes = do_dry_run(td); | |
1582 | else | |
1583 | verify_bytes = do_io(td); | |
2e1df07d JA |
1584 | |
1585 | clear_state = 1; | |
1586 | ||
40c666c8 JA |
1587 | /* |
1588 | * Make sure we've successfully updated the rusage stats | |
1589 | * before waiting on the stat mutex. Otherwise we could have | |
1590 | * the stat thread holding stat mutex and waiting for | |
1591 | * the rusage_sem, which would never get upped because | |
1592 | * this thread is waiting for the stat mutex. | |
1593 | */ | |
1594 | check_update_rusage(td); | |
1595 | ||
e5437a07 | 1596 | fio_mutex_down(stat_mutex); |
95603b74 BF |
1597 | if (td_read(td) && td->io_bytes[DDIR_READ]) |
1598 | update_runtime(td, elapsed_us, DDIR_READ); | |
1599 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) | |
1600 | update_runtime(td, elapsed_us, DDIR_WRITE); | |
1601 | if (td_trim(td) && td->io_bytes[DDIR_TRIM]) | |
1602 | update_runtime(td, elapsed_us, DDIR_TRIM); | |
e5437a07 VT |
1603 | fio_gettime(&td->start, NULL); |
1604 | fio_mutex_up(stat_mutex); | |
2e1df07d JA |
1605 | |
1606 | if (td->error || td->terminate) | |
1607 | break; | |
1608 | ||
4896473e JA |
1609 | if (!o->do_verify || |
1610 | o->verify == VERIFY_NONE || | |
2e1df07d JA |
1611 | (td->io_ops->flags & FIO_UNIDIR)) |
1612 | continue; | |
1613 | ||
1614 | clear_io_state(td); | |
1615 | ||
1616 | fio_gettime(&td->start, NULL); | |
1617 | ||
100f49f1 | 1618 | do_verify(td, verify_bytes); |
2e1df07d | 1619 | |
40c666c8 JA |
1620 | /* |
1621 | * See comment further up for why this is done here. | |
1622 | */ | |
1623 | check_update_rusage(td); | |
1624 | ||
e5437a07 | 1625 | fio_mutex_down(stat_mutex); |
95603b74 | 1626 | update_runtime(td, elapsed_us, DDIR_READ); |
e5437a07 VT |
1627 | fio_gettime(&td->start, NULL); |
1628 | fio_mutex_up(stat_mutex); | |
2e1df07d JA |
1629 | |
1630 | if (td->error || td->terminate) | |
1631 | break; | |
1632 | } | |
1633 | ||
1634 | update_rusage_stat(td); | |
2e1df07d | 1635 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
6eaf09d6 SL |
1636 | td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ]; |
1637 | td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; | |
1638 | td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; | |
2e1df07d | 1639 | |
ca09be4b | 1640 | if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && |
d264264a JA |
1641 | (td->o.verify != VERIFY_NONE && td_write(td))) |
1642 | verify_save_state(td->thread_number); | |
ca09be4b | 1643 | |
9a3f1100 JA |
1644 | fio_unpin_memory(td); |
1645 | ||
905e3d4f | 1646 | fio_writeout_logs(td); |
2e1df07d | 1647 | |
a9da8ab2 JA |
1648 | if (o->io_submit_mode == IO_MODE_OFFLOAD) |
1649 | workqueue_exit(&td->io_wq); | |
1650 | ||
aee2ab67 JA |
1651 | if (td->flags & TD_F_COMPRESS_LOG) |
1652 | tp_exit(&td->tp_data); | |
1653 | ||
4896473e | 1654 | if (o->exec_postrun) |
ce486495 | 1655 | exec_string(o, o->exec_postrun, (const char *)"postrun"); |
2e1df07d JA |
1656 | |
1657 | if (exitall_on_terminate) | |
1658 | fio_terminate_threads(td->groupid); | |
1659 | ||
1660 | err: | |
1661 | if (td->error) | |
1662 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, | |
1663 | td->verror); | |
1664 | ||
4896473e | 1665 | if (o->verify_async) |
2e1df07d JA |
1666 | verify_async_exit(td); |
1667 | ||
1668 | close_and_free_files(td); | |
2e1df07d | 1669 | cleanup_io_u(td); |
32dbca2c | 1670 | close_ioengine(td); |
2e1df07d | 1671 | cgroup_shutdown(td, &cgroup_mnt); |
ca09be4b | 1672 | verify_free_state(td); |
2e1df07d | 1673 | |
b2a9e649 | 1674 | if (fio_option_is_set(o, cpumask)) { |
8a1db9a1 JA |
1675 | ret = fio_cpuset_exit(&o->cpumask); |
1676 | if (ret) | |
1677 | td_verror(td, ret, "fio_cpuset_exit"); | |
2e1df07d JA |
1678 | } |
1679 | ||
1680 | /* | |
1681 | * do this very late, it will log file closing as well | |
1682 | */ | |
4896473e | 1683 | if (o->write_iolog_file) |
2e1df07d JA |
1684 | write_iolog_close(td); |
1685 | ||
ea66e04f JA |
1686 | fio_mutex_remove(td->mutex); |
1687 | td->mutex = NULL; | |
1688 | ||
2e1df07d | 1689 | td_set_runstate(td, TD_EXITED); |
fda2cfac JA |
1690 | |
1691 | /* | |
1692 | * Do this last after setting our runstate to exited, so we | |
1693 | * know that the stat thread is signaled. | |
1694 | */ | |
1695 | check_update_rusage(td); | |
1696 | ||
e43606c2 | 1697 | return (void *) (uintptr_t) td->error; |
2e1df07d JA |
1698 | } |
1699 | ||
1700 | ||
1701 | /* | |
1702 | * We cannot pass the td data into a forked process, so attach the td and | |
1703 | * pass it to the thread worker. | |
1704 | */ | |
1705 | static int fork_main(int shmid, int offset) | |
1706 | { | |
1707 | struct thread_data *td; | |
1708 | void *data, *ret; | |
1709 | ||
c8931876 | 1710 | #if !defined(__hpux) && !defined(CONFIG_NO_SHM) |
2e1df07d JA |
1711 | data = shmat(shmid, NULL, 0); |
1712 | if (data == (void *) -1) { | |
1713 | int __err = errno; | |
1714 | ||
1715 | perror("shmat"); | |
1716 | return __err; | |
1717 | } | |
1718 | #else | |
1719 | /* | |
1720 | * HP-UX inherits shm mappings? | |
1721 | */ | |
1722 | data = threads; | |
1723 | #endif | |
1724 | ||
1725 | td = data + offset * sizeof(struct thread_data); | |
1726 | ret = thread_main(td); | |
1727 | shmdt(data); | |
e43606c2 | 1728 | return (int) (uintptr_t) ret; |
2e1df07d JA |
1729 | } |
1730 | ||
cba5460c JA |
1731 | static void dump_td_info(struct thread_data *td) |
1732 | { | |
1733 | log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to " | |
1734 | "be stuck. Doing forceful exit of this job.\n", td->o.name, | |
1735 | (unsigned long) time_since_now(&td->terminate_time)); | |
1736 | } | |
1737 | ||
2e1df07d JA |
1738 | /* |
1739 | * Run over the job map and reap the threads that have exited, if any. | |
1740 | */ | |
1741 | static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, | |
1742 | unsigned int *m_rate) | |
1743 | { | |
1744 | struct thread_data *td; | |
1745 | unsigned int cputhreads, realthreads, pending; | |
1746 | int i, status, ret; | |
1747 | ||
1748 | /* | |
1749 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1750 | */ | |
1751 | realthreads = pending = cputhreads = 0; | |
1752 | for_each_td(td, i) { | |
1753 | int flags = 0; | |
1754 | ||
1755 | /* | |
1756 | * ->io_ops is NULL for a thread that has closed its | |
1757 | * io engine | |
1758 | */ | |
1759 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) | |
1760 | cputhreads++; | |
1761 | else | |
1762 | realthreads++; | |
1763 | ||
1764 | if (!td->pid) { | |
1765 | pending++; | |
1766 | continue; | |
1767 | } | |
1768 | if (td->runstate == TD_REAPED) | |
1769 | continue; | |
1770 | if (td->o.use_thread) { | |
1771 | if (td->runstate == TD_EXITED) { | |
1772 | td_set_runstate(td, TD_REAPED); | |
1773 | goto reaped; | |
1774 | } | |
1775 | continue; | |
1776 | } | |
1777 | ||
1778 | flags = WNOHANG; | |
1779 | if (td->runstate == TD_EXITED) | |
1780 | flags = 0; | |
1781 | ||
1782 | /* | |
1783 | * check if someone quit or got killed in an unusual way | |
1784 | */ | |
1785 | ret = waitpid(td->pid, &status, flags); | |
1786 | if (ret < 0) { | |
1787 | if (errno == ECHILD) { | |
1788 | log_err("fio: pid=%d disappeared %d\n", | |
1789 | (int) td->pid, td->runstate); | |
a5e371a6 | 1790 | td->sig = ECHILD; |
2e1df07d JA |
1791 | td_set_runstate(td, TD_REAPED); |
1792 | goto reaped; | |
1793 | } | |
1794 | perror("waitpid"); | |
1795 | } else if (ret == td->pid) { | |
1796 | if (WIFSIGNALED(status)) { | |
1797 | int sig = WTERMSIG(status); | |
1798 | ||
36d80bc7 | 1799 | if (sig != SIGTERM && sig != SIGUSR2) |
2e1df07d JA |
1800 | log_err("fio: pid=%d, got signal=%d\n", |
1801 | (int) td->pid, sig); | |
a5e371a6 | 1802 | td->sig = sig; |
2e1df07d JA |
1803 | td_set_runstate(td, TD_REAPED); |
1804 | goto reaped; | |
1805 | } | |
1806 | if (WIFEXITED(status)) { | |
1807 | if (WEXITSTATUS(status) && !td->error) | |
1808 | td->error = WEXITSTATUS(status); | |
1809 | ||
1810 | td_set_runstate(td, TD_REAPED); | |
1811 | goto reaped; | |
1812 | } | |
1813 | } | |
1814 | ||
cba5460c JA |
1815 | /* |
1816 | * If the job is stuck, do a forceful timeout of it and | |
1817 | * move on. | |
1818 | */ | |
1819 | if (td->terminate && | |
1820 | time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { | |
1821 | dump_td_info(td); | |
1822 | td_set_runstate(td, TD_REAPED); | |
1823 | goto reaped; | |
1824 | } | |
1825 | ||
2e1df07d JA |
1826 | /* |
1827 | * thread is not dead, continue | |
1828 | */ | |
1829 | pending++; | |
1830 | continue; | |
1831 | reaped: | |
1832 | (*nr_running)--; | |
342f4be4 JA |
1833 | (*m_rate) -= ddir_rw_sum(td->o.ratemin); |
1834 | (*t_rate) -= ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
1835 | if (!td->pid) |
1836 | pending--; | |
1837 | ||
1838 | if (td->error) | |
1839 | exit_value++; | |
1840 | ||
1841 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
4a88752a | 1842 | profile_td_exit(td); |
2e1df07d JA |
1843 | } |
1844 | ||
1845 | if (*nr_running == cputhreads && !pending && realthreads) | |
1846 | fio_terminate_threads(TERMINATE_ALL); | |
1847 | } | |
1848 | ||
ca09be4b JA |
1849 | static int __check_trigger_file(void) |
1850 | { | |
1851 | struct stat sb; | |
1852 | ||
1853 | if (!trigger_file) | |
1854 | return 0; | |
1855 | ||
1856 | if (stat(trigger_file, &sb)) | |
1857 | return 0; | |
1858 | ||
1859 | if (unlink(trigger_file) < 0) | |
1860 | log_err("fio: failed to unlink %s: %s\n", trigger_file, | |
1861 | strerror(errno)); | |
1862 | ||
1863 | return 1; | |
1864 | } | |
1865 | ||
1866 | static int trigger_timedout(void) | |
1867 | { | |
1868 | if (trigger_timeout) | |
1869 | return time_since_genesis() >= trigger_timeout; | |
1870 | ||
1871 | return 0; | |
1872 | } | |
1873 | ||
1874 | void exec_trigger(const char *cmd) | |
1875 | { | |
1876 | int ret; | |
1877 | ||
1878 | if (!cmd) | |
1879 | return; | |
1880 | ||
1881 | ret = system(cmd); | |
1882 | if (ret == -1) | |
1883 | log_err("fio: failed executing %s trigger\n", cmd); | |
1884 | } | |
1885 | ||
1886 | void check_trigger_file(void) | |
1887 | { | |
1888 | if (__check_trigger_file() || trigger_timedout()) { | |
796fb3ce JA |
1889 | if (nr_clients) |
1890 | fio_clients_send_trigger(trigger_remote_cmd); | |
1891 | else { | |
d264264a | 1892 | verify_save_state(IO_LIST_ALL); |
ca09be4b JA |
1893 | fio_terminate_threads(TERMINATE_ALL); |
1894 | exec_trigger(trigger_cmd); | |
1895 | } | |
1896 | } | |
1897 | } | |
1898 | ||
1899 | static int fio_verify_load_state(struct thread_data *td) | |
1900 | { | |
1901 | int ret; | |
1902 | ||
1903 | if (!td->o.verify_state) | |
1904 | return 0; | |
1905 | ||
1906 | if (is_backend) { | |
1907 | void *data; | |
c3546b53 | 1908 | int ver; |
ca09be4b JA |
1909 | |
1910 | ret = fio_server_get_verify_state(td->o.name, | |
c3546b53 | 1911 | td->thread_number - 1, &data, &ver); |
ca09be4b | 1912 | if (!ret) |
c3546b53 | 1913 | verify_convert_assign_state(td, data, ver); |
ca09be4b JA |
1914 | } else |
1915 | ret = verify_load_state(td, "local"); | |
1916 | ||
1917 | return ret; | |
1918 | } | |
1919 | ||
06464907 JA |
1920 | static void do_usleep(unsigned int usecs) |
1921 | { | |
1922 | check_for_running_stats(); | |
ca09be4b | 1923 | check_trigger_file(); |
06464907 JA |
1924 | usleep(usecs); |
1925 | } | |
1926 | ||
e81ecca3 JA |
1927 | static int check_mount_writes(struct thread_data *td) |
1928 | { | |
1929 | struct fio_file *f; | |
1930 | unsigned int i; | |
1931 | ||
1932 | if (!td_write(td) || td->o.allow_mounted_write) | |
1933 | return 0; | |
1934 | ||
1935 | for_each_file(td, f, i) { | |
1936 | if (f->filetype != FIO_TYPE_BD) | |
1937 | continue; | |
1938 | if (device_is_mounted(f->file_name)) | |
1939 | goto mounted; | |
1940 | } | |
1941 | ||
1942 | return 0; | |
1943 | mounted: | |
1944 | log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.", f->file_name); | |
1945 | return 1; | |
1946 | } | |
1947 | ||
2e1df07d JA |
1948 | /* |
1949 | * Main function for kicking off and reaping jobs, as needed. | |
1950 | */ | |
1951 | static void run_threads(void) | |
1952 | { | |
1953 | struct thread_data *td; | |
2e1df07d | 1954 | unsigned int i, todo, nr_running, m_rate, t_rate, nr_started; |
0de5b26f | 1955 | uint64_t spent; |
2e1df07d | 1956 | |
2e1df07d JA |
1957 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1958 | return; | |
334185e9 | 1959 | |
f2a2ce0e | 1960 | fio_idle_prof_init(); |
2e1df07d JA |
1961 | |
1962 | set_sig_handlers(); | |
1963 | ||
3a5f6bde JA |
1964 | nr_thread = nr_process = 0; |
1965 | for_each_td(td, i) { | |
e81ecca3 JA |
1966 | if (check_mount_writes(td)) |
1967 | return; | |
3a5f6bde JA |
1968 | if (td->o.use_thread) |
1969 | nr_thread++; | |
1970 | else | |
1971 | nr_process++; | |
1972 | } | |
1973 | ||
f3afa57e | 1974 | if (output_format == FIO_OUTPUT_NORMAL) { |
2e1df07d JA |
1975 | log_info("Starting "); |
1976 | if (nr_thread) | |
1977 | log_info("%d thread%s", nr_thread, | |
1978 | nr_thread > 1 ? "s" : ""); | |
1979 | if (nr_process) { | |
1980 | if (nr_thread) | |
1981 | log_info(" and "); | |
1982 | log_info("%d process%s", nr_process, | |
1983 | nr_process > 1 ? "es" : ""); | |
1984 | } | |
1985 | log_info("\n"); | |
e411c301 | 1986 | log_info_flush(); |
2e1df07d JA |
1987 | } |
1988 | ||
1989 | todo = thread_number; | |
1990 | nr_running = 0; | |
1991 | nr_started = 0; | |
1992 | m_rate = t_rate = 0; | |
1993 | ||
1994 | for_each_td(td, i) { | |
1995 | print_status_init(td->thread_number - 1); | |
1996 | ||
1997 | if (!td->o.create_serialize) | |
1998 | continue; | |
1999 | ||
ca09be4b JA |
2000 | if (fio_verify_load_state(td)) |
2001 | goto reap; | |
2002 | ||
2e1df07d JA |
2003 | /* |
2004 | * do file setup here so it happens sequentially, | |
2005 | * we don't want X number of threads getting their | |
2006 | * client data interspersed on disk | |
2007 | */ | |
2008 | if (setup_files(td)) { | |
ca09be4b | 2009 | reap: |
2e1df07d JA |
2010 | exit_value++; |
2011 | if (td->error) | |
2012 | log_err("fio: pid=%d, err=%d/%s\n", | |
2013 | (int) td->pid, td->error, td->verror); | |
2014 | td_set_runstate(td, TD_REAPED); | |
2015 | todo--; | |
2016 | } else { | |
2017 | struct fio_file *f; | |
2018 | unsigned int j; | |
2019 | ||
2020 | /* | |
2021 | * for sharing to work, each job must always open | |
2022 | * its own files. so close them, if we opened them | |
2023 | * for creation | |
2024 | */ | |
2025 | for_each_file(td, f, j) { | |
2026 | if (fio_file_open(f)) | |
2027 | td_io_close_file(td, f); | |
2028 | } | |
2029 | } | |
2030 | } | |
2031 | ||
f2a2ce0e HL |
2032 | /* start idle threads before io threads start to run */ |
2033 | fio_idle_prof_start(); | |
2034 | ||
2e1df07d JA |
2035 | set_genesis_time(); |
2036 | ||
2037 | while (todo) { | |
2038 | struct thread_data *map[REAL_MAX_JOBS]; | |
2039 | struct timeval this_start; | |
2040 | int this_jobs = 0, left; | |
2041 | ||
2042 | /* | |
2043 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
2044 | */ | |
2045 | for_each_td(td, i) { | |
2046 | if (td->runstate != TD_NOT_CREATED) | |
2047 | continue; | |
2048 | ||
2049 | /* | |
2050 | * never got a chance to start, killed by other | |
2051 | * thread for some reason | |
2052 | */ | |
2053 | if (td->terminate) { | |
2054 | todo--; | |
2055 | continue; | |
2056 | } | |
2057 | ||
2058 | if (td->o.start_delay) { | |
0de5b26f | 2059 | spent = utime_since_genesis(); |
2e1df07d | 2060 | |
74454ce4 | 2061 | if (td->o.start_delay > spent) |
2e1df07d JA |
2062 | continue; |
2063 | } | |
2064 | ||
2065 | if (td->o.stonewall && (nr_started || nr_running)) { | |
2066 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
2067 | td->o.name); | |
2068 | break; | |
2069 | } | |
2070 | ||
2071 | init_disk_util(td); | |
2072 | ||
c97f1ad6 JA |
2073 | td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED); |
2074 | td->update_rusage = 0; | |
2075 | ||
2e1df07d JA |
2076 | /* |
2077 | * Set state to created. Thread will transition | |
2078 | * to TD_INITIALIZED when it's done setting up. | |
2079 | */ | |
2080 | td_set_runstate(td, TD_CREATED); | |
2081 | map[this_jobs++] = td; | |
2082 | nr_started++; | |
2083 | ||
2084 | if (td->o.use_thread) { | |
2085 | int ret; | |
2086 | ||
2087 | dprint(FD_PROCESS, "will pthread_create\n"); | |
2088 | ret = pthread_create(&td->thread, NULL, | |
2089 | thread_main, td); | |
2090 | if (ret) { | |
2091 | log_err("pthread_create: %s\n", | |
2092 | strerror(ret)); | |
2093 | nr_started--; | |
2094 | break; | |
2095 | } | |
2096 | ret = pthread_detach(td->thread); | |
2097 | if (ret) | |
2098 | log_err("pthread_detach: %s", | |
2099 | strerror(ret)); | |
2100 | } else { | |
2101 | pid_t pid; | |
2102 | dprint(FD_PROCESS, "will fork\n"); | |
2103 | pid = fork(); | |
2104 | if (!pid) { | |
2105 | int ret = fork_main(shm_id, i); | |
2106 | ||
2107 | _exit(ret); | |
2108 | } else if (i == fio_debug_jobno) | |
2109 | *fio_debug_jobp = pid; | |
2110 | } | |
2111 | dprint(FD_MUTEX, "wait on startup_mutex\n"); | |
2112 | if (fio_mutex_down_timeout(startup_mutex, 10)) { | |
2113 | log_err("fio: job startup hung? exiting.\n"); | |
2114 | fio_terminate_threads(TERMINATE_ALL); | |
2115 | fio_abort = 1; | |
2116 | nr_started--; | |
2117 | break; | |
2118 | } | |
2119 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
2120 | } | |
2121 | ||
2122 | /* | |
2123 | * Wait for the started threads to transition to | |
2124 | * TD_INITIALIZED. | |
2125 | */ | |
2126 | fio_gettime(&this_start, NULL); | |
2127 | left = this_jobs; | |
2128 | while (left && !fio_abort) { | |
2129 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
2130 | break; | |
2131 | ||
06464907 | 2132 | do_usleep(100000); |
2e1df07d JA |
2133 | |
2134 | for (i = 0; i < this_jobs; i++) { | |
2135 | td = map[i]; | |
2136 | if (!td) | |
2137 | continue; | |
2138 | if (td->runstate == TD_INITIALIZED) { | |
2139 | map[i] = NULL; | |
2140 | left--; | |
2141 | } else if (td->runstate >= TD_EXITED) { | |
2142 | map[i] = NULL; | |
2143 | left--; | |
2144 | todo--; | |
2145 | nr_running++; /* work-around... */ | |
2146 | } | |
2147 | } | |
2148 | } | |
2149 | ||
2150 | if (left) { | |
4e87c37a JA |
2151 | log_err("fio: %d job%s failed to start\n", left, |
2152 | left > 1 ? "s" : ""); | |
2e1df07d JA |
2153 | for (i = 0; i < this_jobs; i++) { |
2154 | td = map[i]; | |
2155 | if (!td) | |
2156 | continue; | |
2157 | kill(td->pid, SIGTERM); | |
2158 | } | |
2159 | break; | |
2160 | } | |
2161 | ||
2162 | /* | |
2163 | * start created threads (TD_INITIALIZED -> TD_RUNNING). | |
2164 | */ | |
2165 | for_each_td(td, i) { | |
2166 | if (td->runstate != TD_INITIALIZED) | |
2167 | continue; | |
2168 | ||
2169 | if (in_ramp_time(td)) | |
2170 | td_set_runstate(td, TD_RAMP); | |
2171 | else | |
2172 | td_set_runstate(td, TD_RUNNING); | |
2173 | nr_running++; | |
2174 | nr_started--; | |
342f4be4 JA |
2175 | m_rate += ddir_rw_sum(td->o.ratemin); |
2176 | t_rate += ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
2177 | todo--; |
2178 | fio_mutex_up(td->mutex); | |
2179 | } | |
2180 | ||
2181 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2182 | ||
122c7725 | 2183 | if (todo) |
06464907 | 2184 | do_usleep(100000); |
2e1df07d JA |
2185 | } |
2186 | ||
2187 | while (nr_running) { | |
2188 | reap_threads(&nr_running, &t_rate, &m_rate); | |
06464907 | 2189 | do_usleep(10000); |
2e1df07d JA |
2190 | } |
2191 | ||
f2a2ce0e HL |
2192 | fio_idle_prof_stop(); |
2193 | ||
2e1df07d | 2194 | update_io_ticks(); |
2e1df07d JA |
2195 | } |
2196 | ||
5ddc6707 | 2197 | static void wait_for_helper_thread_exit(void) |
9ec7779f | 2198 | { |
8aab824f JA |
2199 | void *ret; |
2200 | ||
98b4b0a2 | 2201 | helper_exit = 1; |
5ddc6707 JA |
2202 | pthread_cond_signal(&helper_cond); |
2203 | pthread_join(helper_thread, &ret); | |
9ec7779f JA |
2204 | } |
2205 | ||
27357187 JA |
2206 | static void free_disk_util(void) |
2207 | { | |
27357187 | 2208 | disk_util_prune_entries(); |
8aab824f | 2209 | |
5ddc6707 | 2210 | pthread_cond_destroy(&helper_cond); |
27357187 JA |
2211 | } |
2212 | ||
5ddc6707 | 2213 | static void *helper_thread_main(void *data) |
2e1df07d | 2214 | { |
9ec7779f JA |
2215 | int ret = 0; |
2216 | ||
2e1df07d JA |
2217 | fio_mutex_up(startup_mutex); |
2218 | ||
8aab824f JA |
2219 | while (!ret) { |
2220 | uint64_t sec = DISK_UTIL_MSEC / 1000; | |
2221 | uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000; | |
2222 | struct timespec ts; | |
2223 | struct timeval tv; | |
2224 | ||
2225 | gettimeofday(&tv, NULL); | |
2226 | ts.tv_sec = tv.tv_sec + sec; | |
2227 | ts.tv_nsec = (tv.tv_usec * 1000) + nsec; | |
f893b76d | 2228 | |
823952af | 2229 | if (ts.tv_nsec >= 1000000000ULL) { |
8aab824f JA |
2230 | ts.tv_nsec -= 1000000000ULL; |
2231 | ts.tv_sec++; | |
2232 | } | |
2233 | ||
5ddc6707 | 2234 | pthread_cond_timedwait(&helper_cond, &helper_lock, &ts); |
8aab824f | 2235 | |
9ec7779f | 2236 | ret = update_io_ticks(); |
2e1df07d | 2237 | |
5ddc6707 JA |
2238 | if (helper_do_stat) { |
2239 | helper_do_stat = 0; | |
2240 | __show_running_run_stats(); | |
2241 | } | |
2242 | ||
2e1df07d JA |
2243 | if (!is_backend) |
2244 | print_thread_status(); | |
2245 | } | |
2246 | ||
2247 | return NULL; | |
2248 | } | |
2249 | ||
5ddc6707 | 2250 | static int create_helper_thread(void) |
2e1df07d JA |
2251 | { |
2252 | int ret; | |
2253 | ||
9ec7779f JA |
2254 | setup_disk_util(); |
2255 | ||
5ddc6707 JA |
2256 | pthread_cond_init(&helper_cond, NULL); |
2257 | pthread_mutex_init(&helper_lock, NULL); | |
8aab824f | 2258 | |
5ddc6707 | 2259 | ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL); |
2e1df07d | 2260 | if (ret) { |
5ddc6707 | 2261 | log_err("Can't create helper thread: %s\n", strerror(ret)); |
2e1df07d JA |
2262 | return 1; |
2263 | } | |
2264 | ||
2e1df07d JA |
2265 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
2266 | fio_mutex_down(startup_mutex); | |
2267 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
2268 | return 0; | |
2269 | } | |
2270 | ||
2e1df07d JA |
2271 | int fio_backend(void) |
2272 | { | |
2273 | struct thread_data *td; | |
2274 | int i; | |
2275 | ||
2276 | if (exec_profile) { | |
2277 | if (load_profile(exec_profile)) | |
2278 | return 1; | |
2279 | free(exec_profile); | |
2280 | exec_profile = NULL; | |
2281 | } | |
2282 | if (!thread_number) | |
2283 | return 0; | |
2284 | ||
2285 | if (write_bw_log) { | |
aee2ab67 JA |
2286 | struct log_params p = { |
2287 | .log_type = IO_LOG_TYPE_BW, | |
2288 | }; | |
2289 | ||
2290 | setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log"); | |
2291 | setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log"); | |
2292 | setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); | |
2e1df07d JA |
2293 | } |
2294 | ||
521da527 | 2295 | startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); |
2e1df07d JA |
2296 | if (startup_mutex == NULL) |
2297 | return 1; | |
2e1df07d JA |
2298 | |
2299 | set_genesis_time(); | |
cef9175e | 2300 | stat_init(); |
5ddc6707 | 2301 | create_helper_thread(); |
2e1df07d JA |
2302 | |
2303 | cgroup_list = smalloc(sizeof(*cgroup_list)); | |
2304 | INIT_FLIST_HEAD(cgroup_list); | |
2305 | ||
2306 | run_threads(); | |
2307 | ||
5ddc6707 | 2308 | wait_for_helper_thread_exit(); |
8aab824f | 2309 | |
2e1df07d | 2310 | if (!fio_abort) { |
83f7b64e | 2311 | __show_run_stats(); |
2e1df07d | 2312 | if (write_bw_log) { |
cb7e0ace JA |
2313 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { |
2314 | struct io_log *log = agg_io_log[i]; | |
2315 | ||
3a5db920 | 2316 | flush_log(log, 0); |
518dac09 | 2317 | free_log(log); |
cb7e0ace | 2318 | } |
2e1df07d JA |
2319 | } |
2320 | } | |
2321 | ||
fda2cfac | 2322 | for_each_td(td, i) { |
2e1df07d | 2323 | fio_options_free(td); |
8049adc1 JA |
2324 | if (td->rusage_sem) { |
2325 | fio_mutex_remove(td->rusage_sem); | |
2326 | td->rusage_sem = NULL; | |
2327 | } | |
fda2cfac | 2328 | } |
2e1df07d | 2329 | |
a462baef | 2330 | free_disk_util(); |
2e1df07d JA |
2331 | cgroup_kill(cgroup_list); |
2332 | sfree(cgroup_list); | |
2333 | sfree(cgroup_mnt); | |
2334 | ||
2335 | fio_mutex_remove(startup_mutex); | |
cef9175e | 2336 | stat_exit(); |
2e1df07d JA |
2337 | return exit_value; |
2338 | } |