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 | ||
d2235e56 | 103 | void sig_show_status(int sig) |
4c6d91e8 JA |
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 | */ | |
82407585 RP |
449 | min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth); |
450 | if ((full && !min_evts) || !td->o.iodepth_batch_complete_min) | |
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 | ||
82407585 RP |
554 | static inline int io_in_polling(struct thread_data *td) |
555 | { | |
556 | return !td->o.iodepth_batch_complete_min && | |
557 | !td->o.iodepth_batch_complete_max; | |
558 | } | |
559 | ||
2e1df07d JA |
560 | /* |
561 | * The main verify engine. Runs over the writes we previously submitted, | |
562 | * reads the blocks back in, and checks the crc/md5 of the data. | |
563 | */ | |
100f49f1 | 564 | static void do_verify(struct thread_data *td, uint64_t verify_bytes) |
2e1df07d JA |
565 | { |
566 | struct fio_file *f; | |
567 | struct io_u *io_u; | |
568 | int ret, min_events; | |
569 | unsigned int i; | |
570 | ||
571 | dprint(FD_VERIFY, "starting loop\n"); | |
572 | ||
573 | /* | |
574 | * sync io first and invalidate cache, to make sure we really | |
575 | * read from disk. | |
576 | */ | |
577 | for_each_file(td, f, i) { | |
578 | if (!fio_file_open(f)) | |
579 | continue; | |
580 | if (fio_io_sync(td, f)) | |
581 | break; | |
582 | if (file_invalidate_cache(td, f)) | |
583 | break; | |
584 | } | |
585 | ||
c97f1ad6 JA |
586 | check_update_rusage(td); |
587 | ||
2e1df07d JA |
588 | if (td->error) |
589 | return; | |
590 | ||
591 | td_set_runstate(td, TD_VERIFYING); | |
592 | ||
593 | io_u = NULL; | |
594 | while (!td->terminate) { | |
fbccf46c | 595 | enum fio_ddir ddir; |
e9d512d8 | 596 | int full; |
2e1df07d JA |
597 | |
598 | update_tv_cache(td); | |
c97f1ad6 | 599 | check_update_rusage(td); |
2e1df07d JA |
600 | |
601 | if (runtime_exceeded(td, &td->tv_cache)) { | |
602 | __update_tv_cache(td); | |
603 | if (runtime_exceeded(td, &td->tv_cache)) { | |
ebea2133 | 604 | fio_mark_td_terminate(td); |
2e1df07d JA |
605 | break; |
606 | } | |
607 | } | |
608 | ||
9e684a49 DE |
609 | if (flow_threshold_exceeded(td)) |
610 | continue; | |
611 | ||
44cbc6da JA |
612 | if (!td->o.experimental_verify) { |
613 | io_u = __get_io_u(td); | |
614 | if (!io_u) | |
615 | break; | |
2e1df07d | 616 | |
44cbc6da JA |
617 | if (get_next_verify(td, io_u)) { |
618 | put_io_u(td, io_u); | |
619 | break; | |
620 | } | |
2e1df07d | 621 | |
44cbc6da JA |
622 | if (td_io_prep(td, io_u)) { |
623 | put_io_u(td, io_u); | |
624 | break; | |
625 | } | |
626 | } else { | |
55312f9f | 627 | if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes) |
100f49f1 JA |
628 | break; |
629 | ||
bcd5abfa | 630 | while ((io_u = get_io_u(td)) != NULL) { |
002fe734 JA |
631 | if (IS_ERR(io_u)) { |
632 | io_u = NULL; | |
633 | ret = FIO_Q_BUSY; | |
634 | goto reap; | |
635 | } | |
636 | ||
bcd5abfa JA |
637 | /* |
638 | * We are only interested in the places where | |
639 | * we wrote or trimmed IOs. Turn those into | |
640 | * reads for verification purposes. | |
641 | */ | |
642 | if (io_u->ddir == DDIR_READ) { | |
643 | /* | |
644 | * Pretend we issued it for rwmix | |
645 | * accounting | |
646 | */ | |
647 | td->io_issues[DDIR_READ]++; | |
648 | put_io_u(td, io_u); | |
649 | continue; | |
650 | } else if (io_u->ddir == DDIR_TRIM) { | |
651 | io_u->ddir = DDIR_READ; | |
a9da8ab2 | 652 | io_u_set(io_u, IO_U_F_TRIMMED); |
bcd5abfa JA |
653 | break; |
654 | } else if (io_u->ddir == DDIR_WRITE) { | |
655 | io_u->ddir = DDIR_READ; | |
656 | break; | |
657 | } else { | |
658 | put_io_u(td, io_u); | |
659 | continue; | |
660 | } | |
661 | } | |
44cbc6da | 662 | |
bcd5abfa | 663 | if (!io_u) |
44cbc6da | 664 | break; |
2e1df07d JA |
665 | } |
666 | ||
ca09be4b JA |
667 | if (verify_state_should_stop(td, io_u)) { |
668 | put_io_u(td, io_u); | |
669 | break; | |
670 | } | |
671 | ||
2e1df07d JA |
672 | if (td->o.verify_async) |
673 | io_u->end_io = verify_io_u_async; | |
674 | else | |
675 | io_u->end_io = verify_io_u; | |
676 | ||
fbccf46c | 677 | ddir = io_u->ddir; |
10adc4a7 GG |
678 | if (!td->o.disable_slat) |
679 | fio_gettime(&io_u->start_time, NULL); | |
fbccf46c | 680 | |
2e1df07d | 681 | ret = td_io_queue(td, io_u); |
2e1df07d | 682 | |
e9d512d8 | 683 | if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL)) |
2e1df07d JA |
684 | break; |
685 | ||
686 | /* | |
687 | * if we can queue more, do so. but check if there are | |
688 | * completed io_u's first. Note that we can get BUSY even | |
689 | * without IO queued, if the system is resource starved. | |
690 | */ | |
002fe734 | 691 | reap: |
2e1df07d | 692 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
82407585 | 693 | if (full || io_in_polling(td)) |
55312f9f | 694 | ret = wait_for_completions(td, NULL); |
2e1df07d | 695 | |
2e1df07d JA |
696 | if (ret < 0) |
697 | break; | |
698 | } | |
699 | ||
c97f1ad6 JA |
700 | check_update_rusage(td); |
701 | ||
2e1df07d JA |
702 | if (!td->error) { |
703 | min_events = td->cur_depth; | |
704 | ||
705 | if (min_events) | |
55312f9f | 706 | ret = io_u_queued_complete(td, min_events); |
2e1df07d JA |
707 | } else |
708 | cleanup_pending_aio(td); | |
709 | ||
710 | td_set_runstate(td, TD_RUNNING); | |
711 | ||
712 | dprint(FD_VERIFY, "exiting loop\n"); | |
713 | } | |
714 | ||
3939fe85 JA |
715 | static unsigned int exceeds_number_ios(struct thread_data *td) |
716 | { | |
717 | unsigned long long number_ios; | |
718 | ||
719 | if (!td->o.number_ios) | |
720 | return 0; | |
721 | ||
cf8a46a7 | 722 | number_ios = ddir_rw_sum(td->io_blocks); |
3939fe85 JA |
723 | number_ios += td->io_u_queued + td->io_u_in_flight; |
724 | ||
cf8a46a7 | 725 | return number_ios >= (td->o.number_ios * td->loops); |
3939fe85 JA |
726 | } |
727 | ||
e28dd2cf | 728 | static int io_issue_bytes_exceeded(struct thread_data *td) |
f7078f7b | 729 | { |
77731b29 | 730 | unsigned long long bytes, limit; |
f7078f7b JA |
731 | |
732 | if (td_rw(td)) | |
74d6277f | 733 | bytes = td->io_issue_bytes[DDIR_READ] + td->io_issue_bytes[DDIR_WRITE]; |
f7078f7b | 734 | else if (td_write(td)) |
74d6277f | 735 | bytes = td->io_issue_bytes[DDIR_WRITE]; |
6eaf09d6 | 736 | else if (td_read(td)) |
74d6277f | 737 | bytes = td->io_issue_bytes[DDIR_READ]; |
f7078f7b | 738 | else |
74d6277f | 739 | bytes = td->io_issue_bytes[DDIR_TRIM]; |
f7078f7b | 740 | |
77731b29 JA |
741 | if (td->o.io_limit) |
742 | limit = td->o.io_limit; | |
743 | else | |
744 | limit = td->o.size; | |
745 | ||
cf8a46a7 | 746 | limit *= td->loops; |
77731b29 | 747 | return bytes >= limit || exceeds_number_ios(td); |
f7078f7b JA |
748 | } |
749 | ||
e28dd2cf JE |
750 | static int io_complete_bytes_exceeded(struct thread_data *td) |
751 | { | |
752 | unsigned long long bytes, limit; | |
753 | ||
754 | if (td_rw(td)) | |
755 | bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE]; | |
756 | else if (td_write(td)) | |
757 | bytes = td->this_io_bytes[DDIR_WRITE]; | |
758 | else if (td_read(td)) | |
759 | bytes = td->this_io_bytes[DDIR_READ]; | |
760 | else | |
761 | bytes = td->this_io_bytes[DDIR_TRIM]; | |
762 | ||
763 | if (td->o.io_limit) | |
764 | limit = td->o.io_limit; | |
765 | else | |
766 | limit = td->o.size; | |
767 | ||
cf8a46a7 | 768 | limit *= td->loops; |
e28dd2cf JE |
769 | return bytes >= limit || exceeds_number_ios(td); |
770 | } | |
771 | ||
50a8ce86 D |
772 | /* |
773 | * used to calculate the next io time for rate control | |
774 | * | |
775 | */ | |
776 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) | |
777 | { | |
778 | uint64_t secs, remainder, bps, bytes; | |
779 | ||
780 | assert(!(td->flags & TD_F_CHILD)); | |
781 | bytes = td->rate_io_issue_bytes[ddir]; | |
782 | bps = td->rate_bps[ddir]; | |
783 | if (bps) { | |
784 | secs = bytes / bps; | |
785 | remainder = bytes % bps; | |
786 | return remainder * 1000000 / bps + secs * 1000000; | |
787 | } else | |
788 | return 0; | |
789 | } | |
790 | ||
2e1df07d JA |
791 | /* |
792 | * Main IO worker function. It retrieves io_u's to process and queues | |
793 | * and reaps them, checking for rate and errors along the way. | |
100f49f1 JA |
794 | * |
795 | * Returns number of bytes written and trimmed. | |
2e1df07d | 796 | */ |
100f49f1 | 797 | static uint64_t do_io(struct thread_data *td) |
2e1df07d JA |
798 | { |
799 | unsigned int i; | |
800 | int ret = 0; | |
c2703bf3 | 801 | uint64_t total_bytes, bytes_issued = 0; |
2e1df07d JA |
802 | |
803 | if (in_ramp_time(td)) | |
804 | td_set_runstate(td, TD_RAMP); | |
805 | else | |
806 | td_set_runstate(td, TD_RUNNING); | |
807 | ||
3e260a46 JA |
808 | lat_target_init(td); |
809 | ||
1e564979 JE |
810 | total_bytes = td->o.size; |
811 | /* | |
812 | * Allow random overwrite workloads to write up to io_limit | |
813 | * before starting verification phase as 'size' doesn't apply. | |
814 | */ | |
815 | if (td_write(td) && td_random(td) && td->o.norandommap) | |
816 | total_bytes = max(total_bytes, (uint64_t) td->o.io_limit); | |
78a6469c JA |
817 | /* |
818 | * If verify_backlog is enabled, we'll run the verify in this | |
819 | * handler as well. For that case, we may need up to twice the | |
820 | * amount of bytes. | |
821 | */ | |
78a6469c JA |
822 | if (td->o.verify != VERIFY_NONE && |
823 | (td_write(td) && td->o.verify_backlog)) | |
c2703bf3 JA |
824 | total_bytes += td->o.size; |
825 | ||
82a90686 | 826 | /* In trimwrite mode, each byte is trimmed and then written, so |
0e4dd95c | 827 | * allow total_bytes to be twice as big */ |
82a90686 | 828 | if (td_trimwrite(td)) |
0e4dd95c DE |
829 | total_bytes += td->total_io_size; |
830 | ||
f7078f7b | 831 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
e28dd2cf | 832 | (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) || |
c04e4661 | 833 | td->o.time_based) { |
2e1df07d | 834 | struct timeval comp_time; |
2e1df07d | 835 | struct io_u *io_u; |
e9d512d8 | 836 | int full; |
2e1df07d JA |
837 | enum fio_ddir ddir; |
838 | ||
c97f1ad6 JA |
839 | check_update_rusage(td); |
840 | ||
7d7803fa | 841 | if (td->terminate || td->done) |
2e1df07d JA |
842 | break; |
843 | ||
844 | update_tv_cache(td); | |
845 | ||
846 | if (runtime_exceeded(td, &td->tv_cache)) { | |
847 | __update_tv_cache(td); | |
848 | if (runtime_exceeded(td, &td->tv_cache)) { | |
ebea2133 | 849 | fio_mark_td_terminate(td); |
2e1df07d JA |
850 | break; |
851 | } | |
852 | } | |
853 | ||
9e684a49 DE |
854 | if (flow_threshold_exceeded(td)) |
855 | continue; | |
856 | ||
c2703bf3 | 857 | if (bytes_issued >= total_bytes) |
20876c53 JC |
858 | break; |
859 | ||
2e1df07d | 860 | io_u = get_io_u(td); |
002fe734 JA |
861 | if (IS_ERR_OR_NULL(io_u)) { |
862 | int err = PTR_ERR(io_u); | |
863 | ||
864 | io_u = NULL; | |
865 | if (err == -EBUSY) { | |
866 | ret = FIO_Q_BUSY; | |
867 | goto reap; | |
868 | } | |
3e260a46 JA |
869 | if (td->o.latency_target) |
870 | goto reap; | |
2e1df07d | 871 | break; |
3e260a46 | 872 | } |
2e1df07d JA |
873 | |
874 | ddir = io_u->ddir; | |
875 | ||
876 | /* | |
82af2a7c JA |
877 | * Add verification end_io handler if: |
878 | * - Asked to verify (!td_rw(td)) | |
879 | * - Or the io_u is from our verify list (mixed write/ver) | |
2e1df07d JA |
880 | */ |
881 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && | |
82af2a7c | 882 | ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { |
c4b6117b PV |
883 | |
884 | if (!td->o.verify_pattern_bytes) { | |
d6b72507 | 885 | io_u->rand_seed = __rand(&td->verify_state); |
c4b6117b | 886 | if (sizeof(int) != sizeof(long *)) |
d6b72507 | 887 | io_u->rand_seed *= __rand(&td->verify_state); |
c4b6117b PV |
888 | } |
889 | ||
ca09be4b JA |
890 | if (verify_state_should_stop(td, io_u)) { |
891 | put_io_u(td, io_u); | |
892 | break; | |
893 | } | |
894 | ||
2e1df07d JA |
895 | if (td->o.verify_async) |
896 | io_u->end_io = verify_io_u_async; | |
897 | else | |
898 | io_u->end_io = verify_io_u; | |
899 | td_set_runstate(td, TD_VERIFYING); | |
900 | } else if (in_ramp_time(td)) | |
901 | td_set_runstate(td, TD_RAMP); | |
902 | else | |
903 | td_set_runstate(td, TD_RUNNING); | |
904 | ||
9a50c5c5 | 905 | /* |
f9401285 JA |
906 | * Always log IO before it's issued, so we know the specific |
907 | * order of it. The logged unit will track when the IO has | |
908 | * completed. | |
9a50c5c5 | 909 | */ |
c4b6117b PV |
910 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
911 | td->o.do_verify && | |
912 | td->o.verify != VERIFY_NONE && | |
f9401285 | 913 | !td->o.experimental_verify) |
c4b6117b PV |
914 | log_io_piece(td, io_u); |
915 | ||
a9da8ab2 JA |
916 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
917 | if (td->error) | |
918 | break; | |
919 | ret = workqueue_enqueue(&td->io_wq, io_u); | |
50a8ce86 D |
920 | |
921 | if (should_check_rate(td)) | |
922 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
923 | ||
a9da8ab2 JA |
924 | } else { |
925 | ret = td_io_queue(td, io_u); | |
2e1df07d | 926 | |
50a8ce86 D |
927 | if (should_check_rate(td)) |
928 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
929 | ||
fd727d9d | 930 | if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time)) |
a9da8ab2 | 931 | break; |
2e1df07d | 932 | |
a9da8ab2 JA |
933 | /* |
934 | * See if we need to complete some commands. Note that | |
935 | * we can get BUSY even without IO queued, if the | |
936 | * system is resource starved. | |
937 | */ | |
3e260a46 | 938 | reap: |
a9da8ab2 JA |
939 | full = queue_full(td) || |
940 | (ret == FIO_Q_BUSY && td->cur_depth); | |
82407585 | 941 | if (full || io_in_polling(td)) |
a9da8ab2 JA |
942 | ret = wait_for_completions(td, &comp_time); |
943 | } | |
2e1df07d JA |
944 | if (ret < 0) |
945 | break; | |
55312f9f JA |
946 | if (!ddir_rw_sum(td->bytes_done) && |
947 | !(td->io_ops->flags & FIO_NOIO)) | |
2e1df07d JA |
948 | continue; |
949 | ||
55312f9f JA |
950 | if (!in_ramp_time(td) && should_check_rate(td)) { |
951 | if (check_min_rate(td, &comp_time)) { | |
2e1df07d JA |
952 | if (exitall_on_terminate) |
953 | fio_terminate_threads(td->groupid); | |
954 | td_verror(td, EIO, "check_min_rate"); | |
955 | break; | |
956 | } | |
957 | } | |
3e260a46 JA |
958 | if (!in_ramp_time(td) && td->o.latency_target) |
959 | lat_target_check(td); | |
e155cb64 | 960 | |
2e1df07d JA |
961 | if (td->o.thinktime) { |
962 | unsigned long long b; | |
963 | ||
342f4be4 | 964 | b = ddir_rw_sum(td->io_blocks); |
2e1df07d JA |
965 | if (!(b % td->o.thinktime_blocks)) { |
966 | int left; | |
967 | ||
002e7183 JA |
968 | io_u_quiesce(td); |
969 | ||
2e1df07d JA |
970 | if (td->o.thinktime_spin) |
971 | usec_spin(td->o.thinktime_spin); | |
972 | ||
973 | left = td->o.thinktime - td->o.thinktime_spin; | |
974 | if (left) | |
975 | usec_sleep(td, left); | |
976 | } | |
977 | } | |
978 | } | |
979 | ||
c97f1ad6 JA |
980 | check_update_rusage(td); |
981 | ||
2e1df07d | 982 | if (td->trim_entries) |
4e0a8fa2 | 983 | log_err("fio: %lu trim entries leaked?\n", td->trim_entries); |
2e1df07d JA |
984 | |
985 | if (td->o.fill_device && td->error == ENOSPC) { | |
986 | td->error = 0; | |
ebea2133 | 987 | fio_mark_td_terminate(td); |
2e1df07d JA |
988 | } |
989 | if (!td->error) { | |
990 | struct fio_file *f; | |
991 | ||
a9da8ab2 JA |
992 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
993 | workqueue_flush(&td->io_wq); | |
994 | i = 0; | |
995 | } else | |
996 | i = td->cur_depth; | |
997 | ||
2e1df07d | 998 | if (i) { |
55312f9f | 999 | ret = io_u_queued_complete(td, i); |
2e1df07d JA |
1000 | if (td->o.fill_device && td->error == ENOSPC) |
1001 | td->error = 0; | |
1002 | } | |
1003 | ||
1004 | if (should_fsync(td) && td->o.end_fsync) { | |
1005 | td_set_runstate(td, TD_FSYNCING); | |
1006 | ||
1007 | for_each_file(td, f, i) { | |
61ee0f86 | 1008 | if (!fio_file_fsync(td, f)) |
2e1df07d | 1009 | continue; |
61ee0f86 JA |
1010 | |
1011 | log_err("fio: end_fsync failed for file %s\n", | |
1012 | f->file_name); | |
2e1df07d JA |
1013 | } |
1014 | } | |
1015 | } else | |
1016 | cleanup_pending_aio(td); | |
1017 | ||
1018 | /* | |
1019 | * stop job if we failed doing any IO | |
1020 | */ | |
342f4be4 | 1021 | if (!ddir_rw_sum(td->this_io_bytes)) |
2e1df07d | 1022 | td->done = 1; |
100f49f1 | 1023 | |
55312f9f | 1024 | return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; |
2e1df07d JA |
1025 | } |
1026 | ||
1027 | static void cleanup_io_u(struct thread_data *td) | |
1028 | { | |
2e1df07d JA |
1029 | struct io_u *io_u; |
1030 | ||
2ae0b204 | 1031 | while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) { |
c73ed246 JA |
1032 | |
1033 | if (td->io_ops->io_u_free) | |
1034 | td->io_ops->io_u_free(td, io_u); | |
1035 | ||
2e1df07d JA |
1036 | fio_memfree(io_u, sizeof(*io_u)); |
1037 | } | |
1038 | ||
1039 | free_io_mem(td); | |
2ae0b204 JA |
1040 | |
1041 | io_u_rexit(&td->io_u_requeues); | |
1042 | io_u_qexit(&td->io_u_freelist); | |
1043 | io_u_qexit(&td->io_u_all); | |
ca09be4b JA |
1044 | |
1045 | if (td->last_write_comp) | |
1046 | sfree(td->last_write_comp); | |
2e1df07d JA |
1047 | } |
1048 | ||
1049 | static int init_io_u(struct thread_data *td) | |
1050 | { | |
1051 | struct io_u *io_u; | |
9c42684e | 1052 | unsigned int max_bs, min_write; |
2e1df07d | 1053 | int cl_align, i, max_units; |
2ae0b204 | 1054 | int data_xfer = 1, err; |
2e1df07d JA |
1055 | char *p; |
1056 | ||
1057 | max_units = td->o.iodepth; | |
74f4b020 | 1058 | max_bs = td_max_bs(td); |
9c42684e | 1059 | min_write = td->o.min_bs[DDIR_WRITE]; |
2e1df07d JA |
1060 | td->orig_buffer_size = (unsigned long long) max_bs |
1061 | * (unsigned long long) max_units; | |
1062 | ||
88045e04 | 1063 | if ((td->io_ops->flags & FIO_NOIO) || !(td_read(td) || td_write(td))) |
59d8d0f5 JA |
1064 | data_xfer = 0; |
1065 | ||
2ae0b204 JA |
1066 | err = 0; |
1067 | err += io_u_rinit(&td->io_u_requeues, td->o.iodepth); | |
1068 | err += io_u_qinit(&td->io_u_freelist, td->o.iodepth); | |
1069 | err += io_u_qinit(&td->io_u_all, td->o.iodepth); | |
1070 | ||
1071 | if (err) { | |
1072 | log_err("fio: failed setting up IO queues\n"); | |
1073 | return 1; | |
1074 | } | |
1075 | ||
fd8a09b8 | 1076 | /* |
1077 | * if we may later need to do address alignment, then add any | |
1078 | * possible adjustment here so that we don't cause a buffer | |
1079 | * overflow later. this adjustment may be too much if we get | |
1080 | * lucky and the allocator gives us an aligned address. | |
1081 | */ | |
d01612f3 CM |
1082 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
1083 | (td->io_ops->flags & FIO_RAWIO)) | |
fd8a09b8 | 1084 | td->orig_buffer_size += page_mask + td->o.mem_align; |
1085 | ||
2e1df07d JA |
1086 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { |
1087 | unsigned long bs; | |
1088 | ||
1089 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; | |
1090 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
1091 | } | |
1092 | ||
1093 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { | |
1094 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
1095 | return 1; | |
1096 | } | |
1097 | ||
59d8d0f5 | 1098 | if (data_xfer && allocate_io_mem(td)) |
2e1df07d JA |
1099 | return 1; |
1100 | ||
d01612f3 | 1101 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
2e1df07d JA |
1102 | (td->io_ops->flags & FIO_RAWIO)) |
1103 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; | |
1104 | else | |
1105 | p = td->orig_buffer; | |
1106 | ||
1107 | cl_align = os_cache_line_size(); | |
1108 | ||
1109 | for (i = 0; i < max_units; i++) { | |
1110 | void *ptr; | |
1111 | ||
1112 | if (td->terminate) | |
1113 | return 1; | |
1114 | ||
1115 | ptr = fio_memalign(cl_align, sizeof(*io_u)); | |
1116 | if (!ptr) { | |
1117 | log_err("fio: unable to allocate aligned memory\n"); | |
1118 | break; | |
1119 | } | |
1120 | ||
1121 | io_u = ptr; | |
1122 | memset(io_u, 0, sizeof(*io_u)); | |
2ae0b204 | 1123 | INIT_FLIST_HEAD(&io_u->verify_list); |
2e1df07d JA |
1124 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
1125 | ||
59d8d0f5 | 1126 | if (data_xfer) { |
2e1df07d JA |
1127 | io_u->buf = p; |
1128 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); | |
1129 | ||
1130 | if (td_write(td)) | |
9c42684e | 1131 | io_u_fill_buffer(td, io_u, min_write, max_bs); |
2e1df07d JA |
1132 | if (td_write(td) && td->o.verify_pattern_bytes) { |
1133 | /* | |
1134 | * Fill the buffer with the pattern if we are | |
1135 | * going to be doing writes. | |
1136 | */ | |
ce35b1ec | 1137 | fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
2e1df07d JA |
1138 | } |
1139 | } | |
1140 | ||
1141 | io_u->index = i; | |
1142 | io_u->flags = IO_U_F_FREE; | |
2ae0b204 JA |
1143 | io_u_qpush(&td->io_u_freelist, io_u); |
1144 | ||
1145 | /* | |
1146 | * io_u never leaves this stack, used for iteration of all | |
1147 | * io_u buffers. | |
1148 | */ | |
1149 | io_u_qpush(&td->io_u_all, io_u); | |
c73ed246 JA |
1150 | |
1151 | if (td->io_ops->io_u_init) { | |
1152 | int ret = td->io_ops->io_u_init(td, io_u); | |
1153 | ||
1154 | if (ret) { | |
1155 | log_err("fio: failed to init engine data: %d\n", ret); | |
1156 | return 1; | |
1157 | } | |
1158 | } | |
1159 | ||
2e1df07d JA |
1160 | p += max_bs; |
1161 | } | |
1162 | ||
ca09be4b JA |
1163 | if (td->o.verify != VERIFY_NONE) { |
1164 | td->last_write_comp = scalloc(max_units, sizeof(uint64_t)); | |
1165 | if (!td->last_write_comp) { | |
1166 | log_err("fio: failed to alloc write comp data\n"); | |
1167 | return 1; | |
1168 | } | |
1169 | } | |
1170 | ||
2e1df07d JA |
1171 | return 0; |
1172 | } | |
1173 | ||
1174 | static int switch_ioscheduler(struct thread_data *td) | |
1175 | { | |
1176 | char tmp[256], tmp2[128]; | |
1177 | FILE *f; | |
1178 | int ret; | |
1179 | ||
1180 | if (td->io_ops->flags & FIO_DISKLESSIO) | |
1181 | return 0; | |
1182 | ||
1183 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
1184 | ||
1185 | f = fopen(tmp, "r+"); | |
1186 | if (!f) { | |
1187 | if (errno == ENOENT) { | |
1188 | log_err("fio: os or kernel doesn't support IO scheduler" | |
1189 | " switching\n"); | |
1190 | return 0; | |
1191 | } | |
1192 | td_verror(td, errno, "fopen iosched"); | |
1193 | return 1; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * Set io scheduler. | |
1198 | */ | |
1199 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); | |
1200 | if (ferror(f) || ret != 1) { | |
1201 | td_verror(td, errno, "fwrite"); | |
1202 | fclose(f); | |
1203 | return 1; | |
1204 | } | |
1205 | ||
1206 | rewind(f); | |
1207 | ||
1208 | /* | |
1209 | * Read back and check that the selected scheduler is now the default. | |
1210 | */ | |
b44b9e45 | 1211 | memset(tmp, 0, sizeof(tmp)); |
49c6f33d | 1212 | ret = fread(tmp, sizeof(tmp), 1, f); |
2e1df07d JA |
1213 | if (ferror(f) || ret < 0) { |
1214 | td_verror(td, errno, "fread"); | |
1215 | fclose(f); | |
1216 | return 1; | |
1217 | } | |
b44b9e45 TK |
1218 | /* |
1219 | * either a list of io schedulers or "none\n" is expected. | |
1220 | */ | |
1221 | tmp[strlen(tmp) - 1] = '\0'; | |
49c6f33d | 1222 | |
2e1df07d JA |
1223 | |
1224 | sprintf(tmp2, "[%s]", td->o.ioscheduler); | |
1225 | if (!strstr(tmp, tmp2)) { | |
1226 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); | |
1227 | td_verror(td, EINVAL, "iosched_switch"); | |
1228 | fclose(f); | |
1229 | return 1; | |
1230 | } | |
1231 | ||
1232 | fclose(f); | |
1233 | return 0; | |
1234 | } | |
1235 | ||
1236 | static int keep_running(struct thread_data *td) | |
1237 | { | |
77731b29 JA |
1238 | unsigned long long limit; |
1239 | ||
2e1df07d JA |
1240 | if (td->done) |
1241 | return 0; | |
1242 | if (td->o.time_based) | |
1243 | return 1; | |
1244 | if (td->o.loops) { | |
1245 | td->o.loops--; | |
1246 | return 1; | |
1247 | } | |
3939fe85 JA |
1248 | if (exceeds_number_ios(td)) |
1249 | return 0; | |
26251d8d | 1250 | |
77731b29 JA |
1251 | if (td->o.io_limit) |
1252 | limit = td->o.io_limit; | |
1253 | else | |
1254 | limit = td->o.size; | |
1255 | ||
1256 | if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) { | |
5bd5f71a JA |
1257 | uint64_t diff; |
1258 | ||
1259 | /* | |
1260 | * If the difference is less than the minimum IO size, we | |
1261 | * are done. | |
1262 | */ | |
77731b29 | 1263 | diff = limit - ddir_rw_sum(td->io_bytes); |
74f4b020 | 1264 | if (diff < td_max_bs(td)) |
5bd5f71a JA |
1265 | return 0; |
1266 | ||
002fe734 JA |
1267 | if (fio_files_done(td)) |
1268 | return 0; | |
1269 | ||
2e1df07d | 1270 | return 1; |
5bd5f71a | 1271 | } |
2e1df07d JA |
1272 | |
1273 | return 0; | |
1274 | } | |
1275 | ||
ce486495 | 1276 | static int exec_string(struct thread_options *o, const char *string, const char *mode) |
2e1df07d | 1277 | { |
6c9ce469 | 1278 | size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; |
f491a907 | 1279 | int ret; |
2e1df07d JA |
1280 | char *str; |
1281 | ||
1282 | str = malloc(newlen); | |
ce486495 | 1283 | sprintf(str, "%s &> %s.%s.txt", string, o->name, mode); |
2e1df07d | 1284 | |
ce486495 | 1285 | log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode); |
2e1df07d JA |
1286 | ret = system(str); |
1287 | if (ret == -1) | |
1288 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1289 | ||
1290 | free(str); | |
1291 | return ret; | |
1292 | } | |
1293 | ||
62167762 JC |
1294 | /* |
1295 | * Dry run to compute correct state of numberio for verification. | |
1296 | */ | |
1297 | static uint64_t do_dry_run(struct thread_data *td) | |
1298 | { | |
62167762 JC |
1299 | td_set_runstate(td, TD_RUNNING); |
1300 | ||
1301 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || | |
e28dd2cf | 1302 | (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) { |
62167762 JC |
1303 | struct io_u *io_u; |
1304 | int ret; | |
1305 | ||
1306 | if (td->terminate || td->done) | |
1307 | break; | |
1308 | ||
1309 | io_u = get_io_u(td); | |
1310 | if (!io_u) | |
1311 | break; | |
1312 | ||
a9da8ab2 | 1313 | io_u_set(io_u, IO_U_F_FLIGHT); |
62167762 JC |
1314 | io_u->error = 0; |
1315 | io_u->resid = 0; | |
1316 | if (ddir_rw(acct_ddir(io_u))) | |
1317 | td->io_issues[acct_ddir(io_u)]++; | |
1318 | if (ddir_rw(io_u->ddir)) { | |
1319 | io_u_mark_depth(td, 1); | |
1320 | td->ts.total_io_u[io_u->ddir]++; | |
1321 | } | |
1322 | ||
2e63e96b PV |
1323 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1324 | td->o.do_verify && | |
1325 | td->o.verify != VERIFY_NONE && | |
1326 | !td->o.experimental_verify) | |
1327 | log_io_piece(td, io_u); | |
1328 | ||
55312f9f | 1329 | ret = io_u_sync_complete(td, io_u); |
62167762 JC |
1330 | (void) ret; |
1331 | } | |
1332 | ||
55312f9f | 1333 | return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; |
62167762 JC |
1334 | } |
1335 | ||
a9da8ab2 JA |
1336 | static void io_workqueue_fn(struct thread_data *td, struct io_u *io_u) |
1337 | { | |
1338 | const enum fio_ddir ddir = io_u->ddir; | |
1339 | int ret; | |
1340 | ||
1341 | dprint(FD_RATE, "io_u %p queued by %u\n", io_u, gettid()); | |
1342 | ||
1343 | io_u_set(io_u, IO_U_F_NO_FILE_PUT); | |
1344 | ||
1345 | td->cur_depth++; | |
1346 | ||
1347 | ret = td_io_queue(td, io_u); | |
1348 | ||
1349 | dprint(FD_RATE, "io_u %p ret %d by %u\n", io_u, ret, gettid()); | |
1350 | ||
1351 | io_queue_event(td, io_u, &ret, ddir, NULL, 0, NULL); | |
1352 | ||
1353 | if (ret == FIO_Q_QUEUED) | |
1354 | ret = io_u_queued_complete(td, 1); | |
1355 | ||
1356 | td->cur_depth--; | |
1357 | } | |
1358 | ||
2e1df07d JA |
1359 | /* |
1360 | * Entry point for the thread based jobs. The process based jobs end up | |
1361 | * here as well, after a little setup. | |
1362 | */ | |
1363 | static void *thread_main(void *data) | |
1364 | { | |
95603b74 | 1365 | unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, }; |
2e1df07d | 1366 | struct thread_data *td = data; |
4896473e | 1367 | struct thread_options *o = &td->o; |
2e1df07d JA |
1368 | pthread_condattr_t attr; |
1369 | int clear_state; | |
28727df7 | 1370 | int ret; |
2e1df07d | 1371 | |
4896473e | 1372 | if (!o->use_thread) { |
2e1df07d JA |
1373 | setsid(); |
1374 | td->pid = getpid(); | |
1375 | } else | |
1376 | td->pid = gettid(); | |
1377 | ||
4896473e | 1378 | fio_local_clock_init(o->use_thread); |
5d879392 | 1379 | |
2e1df07d JA |
1380 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
1381 | ||
122c7725 JA |
1382 | if (is_backend) |
1383 | fio_server_send_start(td); | |
1384 | ||
2e1df07d JA |
1385 | INIT_FLIST_HEAD(&td->io_log_list); |
1386 | INIT_FLIST_HEAD(&td->io_hist_list); | |
1387 | INIT_FLIST_HEAD(&td->verify_list); | |
1388 | INIT_FLIST_HEAD(&td->trim_list); | |
1ae83d45 | 1389 | INIT_FLIST_HEAD(&td->next_rand_list); |
2e1df07d JA |
1390 | pthread_mutex_init(&td->io_u_lock, NULL); |
1391 | td->io_hist_tree = RB_ROOT; | |
1392 | ||
1393 | pthread_condattr_init(&attr); | |
1394 | pthread_cond_init(&td->verify_cond, &attr); | |
1395 | pthread_cond_init(&td->free_cond, &attr); | |
1396 | ||
1397 | td_set_runstate(td, TD_INITIALIZED); | |
1398 | dprint(FD_MUTEX, "up startup_mutex\n"); | |
1399 | fio_mutex_up(startup_mutex); | |
1400 | dprint(FD_MUTEX, "wait on td->mutex\n"); | |
1401 | fio_mutex_down(td->mutex); | |
1402 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); | |
1403 | ||
2e1df07d JA |
1404 | /* |
1405 | * A new gid requires privilege, so we need to do this before setting | |
1406 | * the uid. | |
1407 | */ | |
4896473e | 1408 | if (o->gid != -1U && setgid(o->gid)) { |
2e1df07d JA |
1409 | td_verror(td, errno, "setgid"); |
1410 | goto err; | |
1411 | } | |
4896473e | 1412 | if (o->uid != -1U && setuid(o->uid)) { |
2e1df07d JA |
1413 | td_verror(td, errno, "setuid"); |
1414 | goto err; | |
1415 | } | |
1416 | ||
1417 | /* | |
1418 | * If we have a gettimeofday() thread, make sure we exclude that | |
1419 | * thread from this job | |
1420 | */ | |
4896473e JA |
1421 | if (o->gtod_cpu) |
1422 | fio_cpu_clear(&o->cpumask, o->gtod_cpu); | |
2e1df07d JA |
1423 | |
1424 | /* | |
1425 | * Set affinity first, in case it has an impact on the memory | |
1426 | * allocations. | |
1427 | */ | |
b2a9e649 | 1428 | if (fio_option_is_set(o, cpumask)) { |
c2acfbac | 1429 | if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { |
30cb4c65 | 1430 | ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); |
c2acfbac JA |
1431 | if (!ret) { |
1432 | log_err("fio: no CPUs set\n"); | |
1433 | log_err("fio: Try increasing number of available CPUs\n"); | |
1434 | td_verror(td, EINVAL, "cpus_split"); | |
1435 | goto err; | |
1436 | } | |
1437 | } | |
28727df7 JA |
1438 | ret = fio_setaffinity(td->pid, o->cpumask); |
1439 | if (ret == -1) { | |
4896473e JA |
1440 | td_verror(td, errno, "cpu_set_affinity"); |
1441 | goto err; | |
1442 | } | |
2e1df07d JA |
1443 | } |
1444 | ||
67bf9823 | 1445 | #ifdef CONFIG_LIBNUMA |
d0b937ed | 1446 | /* numa node setup */ |
b2a9e649 JA |
1447 | if (fio_option_is_set(o, numa_cpunodes) || |
1448 | fio_option_is_set(o, numa_memnodes)) { | |
43522848 | 1449 | struct bitmask *mask; |
d0b937ed YR |
1450 | |
1451 | if (numa_available() < 0) { | |
1452 | td_verror(td, errno, "Does not support NUMA API\n"); | |
1453 | goto err; | |
1454 | } | |
1455 | ||
b2a9e649 | 1456 | if (fio_option_is_set(o, numa_cpunodes)) { |
43522848 DG |
1457 | mask = numa_parse_nodestring(o->numa_cpunodes); |
1458 | ret = numa_run_on_node_mask(mask); | |
1459 | numa_free_nodemask(mask); | |
d0b937ed YR |
1460 | if (ret == -1) { |
1461 | td_verror(td, errno, \ | |
1462 | "numa_run_on_node_mask failed\n"); | |
1463 | goto err; | |
1464 | } | |
1465 | } | |
1466 | ||
b2a9e649 | 1467 | if (fio_option_is_set(o, numa_memnodes)) { |
43522848 DG |
1468 | mask = NULL; |
1469 | if (o->numa_memnodes) | |
1470 | mask = numa_parse_nodestring(o->numa_memnodes); | |
1471 | ||
4896473e | 1472 | switch (o->numa_mem_mode) { |
d0b937ed | 1473 | case MPOL_INTERLEAVE: |
43522848 | 1474 | numa_set_interleave_mask(mask); |
d0b937ed YR |
1475 | break; |
1476 | case MPOL_BIND: | |
43522848 | 1477 | numa_set_membind(mask); |
d0b937ed YR |
1478 | break; |
1479 | case MPOL_LOCAL: | |
1480 | numa_set_localalloc(); | |
1481 | break; | |
1482 | case MPOL_PREFERRED: | |
4896473e | 1483 | numa_set_preferred(o->numa_mem_prefer_node); |
d0b937ed YR |
1484 | break; |
1485 | case MPOL_DEFAULT: | |
1486 | default: | |
1487 | break; | |
1488 | } | |
1489 | ||
43522848 DG |
1490 | if (mask) |
1491 | numa_free_nodemask(mask); | |
1492 | ||
d0b937ed YR |
1493 | } |
1494 | } | |
1495 | #endif | |
1496 | ||
9a3f1100 JA |
1497 | if (fio_pin_memory(td)) |
1498 | goto err; | |
1499 | ||
2e1df07d JA |
1500 | /* |
1501 | * May alter parameters that init_io_u() will use, so we need to | |
1502 | * do this first. | |
1503 | */ | |
1504 | if (init_iolog(td)) | |
1505 | goto err; | |
1506 | ||
1507 | if (init_io_u(td)) | |
1508 | goto err; | |
1509 | ||
4896473e | 1510 | if (o->verify_async && verify_async_init(td)) |
2e1df07d JA |
1511 | goto err; |
1512 | ||
27d74836 JA |
1513 | if (fio_option_is_set(o, ioprio) || |
1514 | fio_option_is_set(o, ioprio_class)) { | |
28727df7 JA |
1515 | ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio); |
1516 | if (ret == -1) { | |
2e1df07d JA |
1517 | td_verror(td, errno, "ioprio_set"); |
1518 | goto err; | |
1519 | } | |
1520 | } | |
1521 | ||
4896473e | 1522 | if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
2e1df07d JA |
1523 | goto err; |
1524 | ||
649c10c5 | 1525 | errno = 0; |
4896473e | 1526 | if (nice(o->nice) == -1 && errno != 0) { |
2e1df07d JA |
1527 | td_verror(td, errno, "nice"); |
1528 | goto err; | |
1529 | } | |
1530 | ||
4896473e | 1531 | if (o->ioscheduler && switch_ioscheduler(td)) |
2e1df07d JA |
1532 | goto err; |
1533 | ||
4896473e | 1534 | if (!o->create_serialize && setup_files(td)) |
2e1df07d JA |
1535 | goto err; |
1536 | ||
1537 | if (td_io_init(td)) | |
1538 | goto err; | |
1539 | ||
1540 | if (init_random_map(td)) | |
1541 | goto err; | |
1542 | ||
ce486495 | 1543 | if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun")) |
4896473e | 1544 | goto err; |
2e1df07d | 1545 | |
4896473e | 1546 | if (o->pre_read) { |
2e1df07d JA |
1547 | if (pre_read_files(td) < 0) |
1548 | goto err; | |
1549 | } | |
1550 | ||
aee2ab67 JA |
1551 | if (td->flags & TD_F_COMPRESS_LOG) |
1552 | tp_init(&td->tp_data); | |
1553 | ||
dc5bfbb2 JA |
1554 | fio_verify_init(td); |
1555 | ||
a9da8ab2 JA |
1556 | if ((o->io_submit_mode == IO_MODE_OFFLOAD) && |
1557 | workqueue_init(td, &td->io_wq, io_workqueue_fn, td->o.iodepth)) | |
1558 | goto err; | |
1559 | ||
2e1df07d | 1560 | fio_gettime(&td->epoch, NULL); |
44404c5a | 1561 | fio_getrusage(&td->ru_start); |
ac28d905 JA |
1562 | memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch)); |
1563 | memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch)); | |
1564 | ||
1565 | if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || | |
1566 | o->ratemin[DDIR_TRIM]) { | |
1567 | memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, | |
1568 | sizeof(td->bw_sample_time)); | |
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, | |
1572 | sizeof(td->bw_sample_time)); | |
1573 | } | |
1574 | ||
2e1df07d JA |
1575 | clear_state = 0; |
1576 | while (keep_running(td)) { | |
100f49f1 JA |
1577 | uint64_t verify_bytes; |
1578 | ||
2e1df07d | 1579 | fio_gettime(&td->start, NULL); |
2e1df07d JA |
1580 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
1581 | ||
2e1df07d | 1582 | if (clear_state) |
ac28d905 | 1583 | clear_io_state(td, 0); |
2e1df07d JA |
1584 | |
1585 | prune_io_piece_log(td); | |
1586 | ||
62167762 JC |
1587 | if (td->o.verify_only && (td_write(td) || td_rw(td))) |
1588 | verify_bytes = do_dry_run(td); | |
1589 | else | |
1590 | verify_bytes = do_io(td); | |
2e1df07d JA |
1591 | |
1592 | clear_state = 1; | |
1593 | ||
40c666c8 JA |
1594 | /* |
1595 | * Make sure we've successfully updated the rusage stats | |
1596 | * before waiting on the stat mutex. Otherwise we could have | |
1597 | * the stat thread holding stat mutex and waiting for | |
1598 | * the rusage_sem, which would never get upped because | |
1599 | * this thread is waiting for the stat mutex. | |
1600 | */ | |
1601 | check_update_rusage(td); | |
1602 | ||
e5437a07 | 1603 | fio_mutex_down(stat_mutex); |
95603b74 BF |
1604 | if (td_read(td) && td->io_bytes[DDIR_READ]) |
1605 | update_runtime(td, elapsed_us, DDIR_READ); | |
1606 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) | |
1607 | update_runtime(td, elapsed_us, DDIR_WRITE); | |
1608 | if (td_trim(td) && td->io_bytes[DDIR_TRIM]) | |
1609 | update_runtime(td, elapsed_us, DDIR_TRIM); | |
e5437a07 VT |
1610 | fio_gettime(&td->start, NULL); |
1611 | fio_mutex_up(stat_mutex); | |
2e1df07d JA |
1612 | |
1613 | if (td->error || td->terminate) | |
1614 | break; | |
1615 | ||
4896473e JA |
1616 | if (!o->do_verify || |
1617 | o->verify == VERIFY_NONE || | |
2e1df07d JA |
1618 | (td->io_ops->flags & FIO_UNIDIR)) |
1619 | continue; | |
1620 | ||
ac28d905 | 1621 | clear_io_state(td, 0); |
2e1df07d JA |
1622 | |
1623 | fio_gettime(&td->start, NULL); | |
1624 | ||
100f49f1 | 1625 | do_verify(td, verify_bytes); |
2e1df07d | 1626 | |
40c666c8 JA |
1627 | /* |
1628 | * See comment further up for why this is done here. | |
1629 | */ | |
1630 | check_update_rusage(td); | |
1631 | ||
e5437a07 | 1632 | fio_mutex_down(stat_mutex); |
95603b74 | 1633 | update_runtime(td, elapsed_us, DDIR_READ); |
e5437a07 VT |
1634 | fio_gettime(&td->start, NULL); |
1635 | fio_mutex_up(stat_mutex); | |
2e1df07d JA |
1636 | |
1637 | if (td->error || td->terminate) | |
1638 | break; | |
1639 | } | |
1640 | ||
1641 | update_rusage_stat(td); | |
2e1df07d | 1642 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
6eaf09d6 SL |
1643 | td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ]; |
1644 | td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; | |
1645 | td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; | |
2e1df07d | 1646 | |
ca09be4b | 1647 | if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && |
d264264a JA |
1648 | (td->o.verify != VERIFY_NONE && td_write(td))) |
1649 | verify_save_state(td->thread_number); | |
ca09be4b | 1650 | |
9a3f1100 JA |
1651 | fio_unpin_memory(td); |
1652 | ||
905e3d4f | 1653 | fio_writeout_logs(td); |
2e1df07d | 1654 | |
a9da8ab2 JA |
1655 | if (o->io_submit_mode == IO_MODE_OFFLOAD) |
1656 | workqueue_exit(&td->io_wq); | |
1657 | ||
aee2ab67 JA |
1658 | if (td->flags & TD_F_COMPRESS_LOG) |
1659 | tp_exit(&td->tp_data); | |
1660 | ||
4896473e | 1661 | if (o->exec_postrun) |
ce486495 | 1662 | exec_string(o, o->exec_postrun, (const char *)"postrun"); |
2e1df07d JA |
1663 | |
1664 | if (exitall_on_terminate) | |
1665 | fio_terminate_threads(td->groupid); | |
1666 | ||
1667 | err: | |
1668 | if (td->error) | |
1669 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, | |
1670 | td->verror); | |
1671 | ||
4896473e | 1672 | if (o->verify_async) |
2e1df07d JA |
1673 | verify_async_exit(td); |
1674 | ||
1675 | close_and_free_files(td); | |
2e1df07d | 1676 | cleanup_io_u(td); |
32dbca2c | 1677 | close_ioengine(td); |
2e1df07d | 1678 | cgroup_shutdown(td, &cgroup_mnt); |
ca09be4b | 1679 | verify_free_state(td); |
2e1df07d | 1680 | |
b2a9e649 | 1681 | if (fio_option_is_set(o, cpumask)) { |
8a1db9a1 JA |
1682 | ret = fio_cpuset_exit(&o->cpumask); |
1683 | if (ret) | |
1684 | td_verror(td, ret, "fio_cpuset_exit"); | |
2e1df07d JA |
1685 | } |
1686 | ||
1687 | /* | |
1688 | * do this very late, it will log file closing as well | |
1689 | */ | |
4896473e | 1690 | if (o->write_iolog_file) |
2e1df07d JA |
1691 | write_iolog_close(td); |
1692 | ||
ea66e04f JA |
1693 | fio_mutex_remove(td->mutex); |
1694 | td->mutex = NULL; | |
1695 | ||
2e1df07d | 1696 | td_set_runstate(td, TD_EXITED); |
fda2cfac JA |
1697 | |
1698 | /* | |
1699 | * Do this last after setting our runstate to exited, so we | |
1700 | * know that the stat thread is signaled. | |
1701 | */ | |
1702 | check_update_rusage(td); | |
1703 | ||
e43606c2 | 1704 | return (void *) (uintptr_t) td->error; |
2e1df07d JA |
1705 | } |
1706 | ||
1707 | ||
1708 | /* | |
1709 | * We cannot pass the td data into a forked process, so attach the td and | |
1710 | * pass it to the thread worker. | |
1711 | */ | |
1712 | static int fork_main(int shmid, int offset) | |
1713 | { | |
1714 | struct thread_data *td; | |
1715 | void *data, *ret; | |
1716 | ||
c8931876 | 1717 | #if !defined(__hpux) && !defined(CONFIG_NO_SHM) |
2e1df07d JA |
1718 | data = shmat(shmid, NULL, 0); |
1719 | if (data == (void *) -1) { | |
1720 | int __err = errno; | |
1721 | ||
1722 | perror("shmat"); | |
1723 | return __err; | |
1724 | } | |
1725 | #else | |
1726 | /* | |
1727 | * HP-UX inherits shm mappings? | |
1728 | */ | |
1729 | data = threads; | |
1730 | #endif | |
1731 | ||
1732 | td = data + offset * sizeof(struct thread_data); | |
1733 | ret = thread_main(td); | |
1734 | shmdt(data); | |
e43606c2 | 1735 | return (int) (uintptr_t) ret; |
2e1df07d JA |
1736 | } |
1737 | ||
cba5460c JA |
1738 | static void dump_td_info(struct thread_data *td) |
1739 | { | |
1740 | log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to " | |
1741 | "be stuck. Doing forceful exit of this job.\n", td->o.name, | |
1742 | (unsigned long) time_since_now(&td->terminate_time)); | |
1743 | } | |
1744 | ||
2e1df07d JA |
1745 | /* |
1746 | * Run over the job map and reap the threads that have exited, if any. | |
1747 | */ | |
1748 | static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, | |
1749 | unsigned int *m_rate) | |
1750 | { | |
1751 | struct thread_data *td; | |
1752 | unsigned int cputhreads, realthreads, pending; | |
1753 | int i, status, ret; | |
1754 | ||
1755 | /* | |
1756 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1757 | */ | |
1758 | realthreads = pending = cputhreads = 0; | |
1759 | for_each_td(td, i) { | |
1760 | int flags = 0; | |
1761 | ||
1762 | /* | |
1763 | * ->io_ops is NULL for a thread that has closed its | |
1764 | * io engine | |
1765 | */ | |
1766 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) | |
1767 | cputhreads++; | |
1768 | else | |
1769 | realthreads++; | |
1770 | ||
1771 | if (!td->pid) { | |
1772 | pending++; | |
1773 | continue; | |
1774 | } | |
1775 | if (td->runstate == TD_REAPED) | |
1776 | continue; | |
1777 | if (td->o.use_thread) { | |
1778 | if (td->runstate == TD_EXITED) { | |
1779 | td_set_runstate(td, TD_REAPED); | |
1780 | goto reaped; | |
1781 | } | |
1782 | continue; | |
1783 | } | |
1784 | ||
1785 | flags = WNOHANG; | |
1786 | if (td->runstate == TD_EXITED) | |
1787 | flags = 0; | |
1788 | ||
1789 | /* | |
1790 | * check if someone quit or got killed in an unusual way | |
1791 | */ | |
1792 | ret = waitpid(td->pid, &status, flags); | |
1793 | if (ret < 0) { | |
1794 | if (errno == ECHILD) { | |
1795 | log_err("fio: pid=%d disappeared %d\n", | |
1796 | (int) td->pid, td->runstate); | |
a5e371a6 | 1797 | td->sig = ECHILD; |
2e1df07d JA |
1798 | td_set_runstate(td, TD_REAPED); |
1799 | goto reaped; | |
1800 | } | |
1801 | perror("waitpid"); | |
1802 | } else if (ret == td->pid) { | |
1803 | if (WIFSIGNALED(status)) { | |
1804 | int sig = WTERMSIG(status); | |
1805 | ||
36d80bc7 | 1806 | if (sig != SIGTERM && sig != SIGUSR2) |
2e1df07d JA |
1807 | log_err("fio: pid=%d, got signal=%d\n", |
1808 | (int) td->pid, sig); | |
a5e371a6 | 1809 | td->sig = sig; |
2e1df07d JA |
1810 | td_set_runstate(td, TD_REAPED); |
1811 | goto reaped; | |
1812 | } | |
1813 | if (WIFEXITED(status)) { | |
1814 | if (WEXITSTATUS(status) && !td->error) | |
1815 | td->error = WEXITSTATUS(status); | |
1816 | ||
1817 | td_set_runstate(td, TD_REAPED); | |
1818 | goto reaped; | |
1819 | } | |
1820 | } | |
1821 | ||
cba5460c JA |
1822 | /* |
1823 | * If the job is stuck, do a forceful timeout of it and | |
1824 | * move on. | |
1825 | */ | |
1826 | if (td->terminate && | |
1827 | time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { | |
1828 | dump_td_info(td); | |
1829 | td_set_runstate(td, TD_REAPED); | |
1830 | goto reaped; | |
1831 | } | |
1832 | ||
2e1df07d JA |
1833 | /* |
1834 | * thread is not dead, continue | |
1835 | */ | |
1836 | pending++; | |
1837 | continue; | |
1838 | reaped: | |
1839 | (*nr_running)--; | |
342f4be4 JA |
1840 | (*m_rate) -= ddir_rw_sum(td->o.ratemin); |
1841 | (*t_rate) -= ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
1842 | if (!td->pid) |
1843 | pending--; | |
1844 | ||
1845 | if (td->error) | |
1846 | exit_value++; | |
1847 | ||
1848 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
4a88752a | 1849 | profile_td_exit(td); |
2e1df07d JA |
1850 | } |
1851 | ||
1852 | if (*nr_running == cputhreads && !pending && realthreads) | |
1853 | fio_terminate_threads(TERMINATE_ALL); | |
1854 | } | |
1855 | ||
ca09be4b JA |
1856 | static int __check_trigger_file(void) |
1857 | { | |
1858 | struct stat sb; | |
1859 | ||
1860 | if (!trigger_file) | |
1861 | return 0; | |
1862 | ||
1863 | if (stat(trigger_file, &sb)) | |
1864 | return 0; | |
1865 | ||
1866 | if (unlink(trigger_file) < 0) | |
1867 | log_err("fio: failed to unlink %s: %s\n", trigger_file, | |
1868 | strerror(errno)); | |
1869 | ||
1870 | return 1; | |
1871 | } | |
1872 | ||
1873 | static int trigger_timedout(void) | |
1874 | { | |
1875 | if (trigger_timeout) | |
1876 | return time_since_genesis() >= trigger_timeout; | |
1877 | ||
1878 | return 0; | |
1879 | } | |
1880 | ||
1881 | void exec_trigger(const char *cmd) | |
1882 | { | |
1883 | int ret; | |
1884 | ||
1885 | if (!cmd) | |
1886 | return; | |
1887 | ||
1888 | ret = system(cmd); | |
1889 | if (ret == -1) | |
1890 | log_err("fio: failed executing %s trigger\n", cmd); | |
1891 | } | |
1892 | ||
1893 | void check_trigger_file(void) | |
1894 | { | |
1895 | if (__check_trigger_file() || trigger_timedout()) { | |
796fb3ce JA |
1896 | if (nr_clients) |
1897 | fio_clients_send_trigger(trigger_remote_cmd); | |
1898 | else { | |
d264264a | 1899 | verify_save_state(IO_LIST_ALL); |
ca09be4b JA |
1900 | fio_terminate_threads(TERMINATE_ALL); |
1901 | exec_trigger(trigger_cmd); | |
1902 | } | |
1903 | } | |
1904 | } | |
1905 | ||
1906 | static int fio_verify_load_state(struct thread_data *td) | |
1907 | { | |
1908 | int ret; | |
1909 | ||
1910 | if (!td->o.verify_state) | |
1911 | return 0; | |
1912 | ||
1913 | if (is_backend) { | |
1914 | void *data; | |
c3546b53 | 1915 | int ver; |
ca09be4b JA |
1916 | |
1917 | ret = fio_server_get_verify_state(td->o.name, | |
c3546b53 | 1918 | td->thread_number - 1, &data, &ver); |
ca09be4b | 1919 | if (!ret) |
c3546b53 | 1920 | verify_convert_assign_state(td, data, ver); |
ca09be4b JA |
1921 | } else |
1922 | ret = verify_load_state(td, "local"); | |
1923 | ||
1924 | return ret; | |
1925 | } | |
1926 | ||
06464907 JA |
1927 | static void do_usleep(unsigned int usecs) |
1928 | { | |
1929 | check_for_running_stats(); | |
ca09be4b | 1930 | check_trigger_file(); |
06464907 JA |
1931 | usleep(usecs); |
1932 | } | |
1933 | ||
e81ecca3 JA |
1934 | static int check_mount_writes(struct thread_data *td) |
1935 | { | |
1936 | struct fio_file *f; | |
1937 | unsigned int i; | |
1938 | ||
1939 | if (!td_write(td) || td->o.allow_mounted_write) | |
1940 | return 0; | |
1941 | ||
1942 | for_each_file(td, f, i) { | |
1943 | if (f->filetype != FIO_TYPE_BD) | |
1944 | continue; | |
1945 | if (device_is_mounted(f->file_name)) | |
1946 | goto mounted; | |
1947 | } | |
1948 | ||
1949 | return 0; | |
1950 | mounted: | |
1951 | log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.", f->file_name); | |
1952 | return 1; | |
1953 | } | |
1954 | ||
2e1df07d JA |
1955 | /* |
1956 | * Main function for kicking off and reaping jobs, as needed. | |
1957 | */ | |
1958 | static void run_threads(void) | |
1959 | { | |
1960 | struct thread_data *td; | |
2e1df07d | 1961 | unsigned int i, todo, nr_running, m_rate, t_rate, nr_started; |
0de5b26f | 1962 | uint64_t spent; |
2e1df07d | 1963 | |
2e1df07d JA |
1964 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1965 | return; | |
334185e9 | 1966 | |
f2a2ce0e | 1967 | fio_idle_prof_init(); |
2e1df07d JA |
1968 | |
1969 | set_sig_handlers(); | |
1970 | ||
3a5f6bde JA |
1971 | nr_thread = nr_process = 0; |
1972 | for_each_td(td, i) { | |
e81ecca3 JA |
1973 | if (check_mount_writes(td)) |
1974 | return; | |
3a5f6bde JA |
1975 | if (td->o.use_thread) |
1976 | nr_thread++; | |
1977 | else | |
1978 | nr_process++; | |
1979 | } | |
1980 | ||
01cfefcc | 1981 | if (output_format & FIO_OUTPUT_NORMAL) { |
2e1df07d JA |
1982 | log_info("Starting "); |
1983 | if (nr_thread) | |
1984 | log_info("%d thread%s", nr_thread, | |
1985 | nr_thread > 1 ? "s" : ""); | |
1986 | if (nr_process) { | |
1987 | if (nr_thread) | |
1988 | log_info(" and "); | |
1989 | log_info("%d process%s", nr_process, | |
1990 | nr_process > 1 ? "es" : ""); | |
1991 | } | |
1992 | log_info("\n"); | |
e411c301 | 1993 | log_info_flush(); |
2e1df07d JA |
1994 | } |
1995 | ||
1996 | todo = thread_number; | |
1997 | nr_running = 0; | |
1998 | nr_started = 0; | |
1999 | m_rate = t_rate = 0; | |
2000 | ||
2001 | for_each_td(td, i) { | |
2002 | print_status_init(td->thread_number - 1); | |
2003 | ||
2004 | if (!td->o.create_serialize) | |
2005 | continue; | |
2006 | ||
ca09be4b JA |
2007 | if (fio_verify_load_state(td)) |
2008 | goto reap; | |
2009 | ||
2e1df07d JA |
2010 | /* |
2011 | * do file setup here so it happens sequentially, | |
2012 | * we don't want X number of threads getting their | |
2013 | * client data interspersed on disk | |
2014 | */ | |
2015 | if (setup_files(td)) { | |
ca09be4b | 2016 | reap: |
2e1df07d JA |
2017 | exit_value++; |
2018 | if (td->error) | |
2019 | log_err("fio: pid=%d, err=%d/%s\n", | |
2020 | (int) td->pid, td->error, td->verror); | |
2021 | td_set_runstate(td, TD_REAPED); | |
2022 | todo--; | |
2023 | } else { | |
2024 | struct fio_file *f; | |
2025 | unsigned int j; | |
2026 | ||
2027 | /* | |
2028 | * for sharing to work, each job must always open | |
2029 | * its own files. so close them, if we opened them | |
2030 | * for creation | |
2031 | */ | |
2032 | for_each_file(td, f, j) { | |
2033 | if (fio_file_open(f)) | |
2034 | td_io_close_file(td, f); | |
2035 | } | |
2036 | } | |
2037 | } | |
2038 | ||
f2a2ce0e HL |
2039 | /* start idle threads before io threads start to run */ |
2040 | fio_idle_prof_start(); | |
2041 | ||
2e1df07d JA |
2042 | set_genesis_time(); |
2043 | ||
2044 | while (todo) { | |
2045 | struct thread_data *map[REAL_MAX_JOBS]; | |
2046 | struct timeval this_start; | |
2047 | int this_jobs = 0, left; | |
2048 | ||
2049 | /* | |
2050 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
2051 | */ | |
2052 | for_each_td(td, i) { | |
2053 | if (td->runstate != TD_NOT_CREATED) | |
2054 | continue; | |
2055 | ||
2056 | /* | |
2057 | * never got a chance to start, killed by other | |
2058 | * thread for some reason | |
2059 | */ | |
2060 | if (td->terminate) { | |
2061 | todo--; | |
2062 | continue; | |
2063 | } | |
2064 | ||
2065 | if (td->o.start_delay) { | |
0de5b26f | 2066 | spent = utime_since_genesis(); |
2e1df07d | 2067 | |
74454ce4 | 2068 | if (td->o.start_delay > spent) |
2e1df07d JA |
2069 | continue; |
2070 | } | |
2071 | ||
2072 | if (td->o.stonewall && (nr_started || nr_running)) { | |
2073 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
2074 | td->o.name); | |
2075 | break; | |
2076 | } | |
2077 | ||
2078 | init_disk_util(td); | |
2079 | ||
c97f1ad6 JA |
2080 | td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED); |
2081 | td->update_rusage = 0; | |
2082 | ||
2e1df07d JA |
2083 | /* |
2084 | * Set state to created. Thread will transition | |
2085 | * to TD_INITIALIZED when it's done setting up. | |
2086 | */ | |
2087 | td_set_runstate(td, TD_CREATED); | |
2088 | map[this_jobs++] = td; | |
2089 | nr_started++; | |
2090 | ||
2091 | if (td->o.use_thread) { | |
2092 | int ret; | |
2093 | ||
2094 | dprint(FD_PROCESS, "will pthread_create\n"); | |
2095 | ret = pthread_create(&td->thread, NULL, | |
2096 | thread_main, td); | |
2097 | if (ret) { | |
2098 | log_err("pthread_create: %s\n", | |
2099 | strerror(ret)); | |
2100 | nr_started--; | |
2101 | break; | |
2102 | } | |
2103 | ret = pthread_detach(td->thread); | |
2104 | if (ret) | |
2105 | log_err("pthread_detach: %s", | |
2106 | strerror(ret)); | |
2107 | } else { | |
2108 | pid_t pid; | |
2109 | dprint(FD_PROCESS, "will fork\n"); | |
2110 | pid = fork(); | |
2111 | if (!pid) { | |
2112 | int ret = fork_main(shm_id, i); | |
2113 | ||
2114 | _exit(ret); | |
2115 | } else if (i == fio_debug_jobno) | |
2116 | *fio_debug_jobp = pid; | |
2117 | } | |
2118 | dprint(FD_MUTEX, "wait on startup_mutex\n"); | |
2119 | if (fio_mutex_down_timeout(startup_mutex, 10)) { | |
2120 | log_err("fio: job startup hung? exiting.\n"); | |
2121 | fio_terminate_threads(TERMINATE_ALL); | |
2122 | fio_abort = 1; | |
2123 | nr_started--; | |
2124 | break; | |
2125 | } | |
2126 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
2127 | } | |
2128 | ||
2129 | /* | |
2130 | * Wait for the started threads to transition to | |
2131 | * TD_INITIALIZED. | |
2132 | */ | |
2133 | fio_gettime(&this_start, NULL); | |
2134 | left = this_jobs; | |
2135 | while (left && !fio_abort) { | |
2136 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
2137 | break; | |
2138 | ||
06464907 | 2139 | do_usleep(100000); |
2e1df07d JA |
2140 | |
2141 | for (i = 0; i < this_jobs; i++) { | |
2142 | td = map[i]; | |
2143 | if (!td) | |
2144 | continue; | |
2145 | if (td->runstate == TD_INITIALIZED) { | |
2146 | map[i] = NULL; | |
2147 | left--; | |
2148 | } else if (td->runstate >= TD_EXITED) { | |
2149 | map[i] = NULL; | |
2150 | left--; | |
2151 | todo--; | |
2152 | nr_running++; /* work-around... */ | |
2153 | } | |
2154 | } | |
2155 | } | |
2156 | ||
2157 | if (left) { | |
4e87c37a JA |
2158 | log_err("fio: %d job%s failed to start\n", left, |
2159 | left > 1 ? "s" : ""); | |
2e1df07d JA |
2160 | for (i = 0; i < this_jobs; i++) { |
2161 | td = map[i]; | |
2162 | if (!td) | |
2163 | continue; | |
2164 | kill(td->pid, SIGTERM); | |
2165 | } | |
2166 | break; | |
2167 | } | |
2168 | ||
2169 | /* | |
2170 | * start created threads (TD_INITIALIZED -> TD_RUNNING). | |
2171 | */ | |
2172 | for_each_td(td, i) { | |
2173 | if (td->runstate != TD_INITIALIZED) | |
2174 | continue; | |
2175 | ||
2176 | if (in_ramp_time(td)) | |
2177 | td_set_runstate(td, TD_RAMP); | |
2178 | else | |
2179 | td_set_runstate(td, TD_RUNNING); | |
2180 | nr_running++; | |
2181 | nr_started--; | |
342f4be4 JA |
2182 | m_rate += ddir_rw_sum(td->o.ratemin); |
2183 | t_rate += ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
2184 | todo--; |
2185 | fio_mutex_up(td->mutex); | |
2186 | } | |
2187 | ||
2188 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2189 | ||
122c7725 | 2190 | if (todo) |
06464907 | 2191 | do_usleep(100000); |
2e1df07d JA |
2192 | } |
2193 | ||
2194 | while (nr_running) { | |
2195 | reap_threads(&nr_running, &t_rate, &m_rate); | |
06464907 | 2196 | do_usleep(10000); |
2e1df07d JA |
2197 | } |
2198 | ||
f2a2ce0e HL |
2199 | fio_idle_prof_stop(); |
2200 | ||
2e1df07d | 2201 | update_io_ticks(); |
2e1df07d JA |
2202 | } |
2203 | ||
5ddc6707 | 2204 | static void wait_for_helper_thread_exit(void) |
9ec7779f | 2205 | { |
8aab824f JA |
2206 | void *ret; |
2207 | ||
98b4b0a2 | 2208 | helper_exit = 1; |
5ddc6707 JA |
2209 | pthread_cond_signal(&helper_cond); |
2210 | pthread_join(helper_thread, &ret); | |
9ec7779f JA |
2211 | } |
2212 | ||
27357187 JA |
2213 | static void free_disk_util(void) |
2214 | { | |
27357187 | 2215 | disk_util_prune_entries(); |
8aab824f | 2216 | |
5ddc6707 | 2217 | pthread_cond_destroy(&helper_cond); |
27357187 JA |
2218 | } |
2219 | ||
5ddc6707 | 2220 | static void *helper_thread_main(void *data) |
2e1df07d | 2221 | { |
9ec7779f JA |
2222 | int ret = 0; |
2223 | ||
2e1df07d JA |
2224 | fio_mutex_up(startup_mutex); |
2225 | ||
8aab824f JA |
2226 | while (!ret) { |
2227 | uint64_t sec = DISK_UTIL_MSEC / 1000; | |
2228 | uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000; | |
2229 | struct timespec ts; | |
2230 | struct timeval tv; | |
2231 | ||
2232 | gettimeofday(&tv, NULL); | |
2233 | ts.tv_sec = tv.tv_sec + sec; | |
2234 | ts.tv_nsec = (tv.tv_usec * 1000) + nsec; | |
f893b76d | 2235 | |
823952af | 2236 | if (ts.tv_nsec >= 1000000000ULL) { |
8aab824f JA |
2237 | ts.tv_nsec -= 1000000000ULL; |
2238 | ts.tv_sec++; | |
2239 | } | |
2240 | ||
5ddc6707 | 2241 | pthread_cond_timedwait(&helper_cond, &helper_lock, &ts); |
8aab824f | 2242 | |
9ec7779f | 2243 | ret = update_io_ticks(); |
2e1df07d | 2244 | |
5ddc6707 JA |
2245 | if (helper_do_stat) { |
2246 | helper_do_stat = 0; | |
2247 | __show_running_run_stats(); | |
2248 | } | |
2249 | ||
2e1df07d JA |
2250 | if (!is_backend) |
2251 | print_thread_status(); | |
2252 | } | |
2253 | ||
2254 | return NULL; | |
2255 | } | |
2256 | ||
5ddc6707 | 2257 | static int create_helper_thread(void) |
2e1df07d JA |
2258 | { |
2259 | int ret; | |
2260 | ||
9ec7779f JA |
2261 | setup_disk_util(); |
2262 | ||
5ddc6707 JA |
2263 | pthread_cond_init(&helper_cond, NULL); |
2264 | pthread_mutex_init(&helper_lock, NULL); | |
8aab824f | 2265 | |
5ddc6707 | 2266 | ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL); |
2e1df07d | 2267 | if (ret) { |
5ddc6707 | 2268 | log_err("Can't create helper thread: %s\n", strerror(ret)); |
2e1df07d JA |
2269 | return 1; |
2270 | } | |
2271 | ||
2e1df07d JA |
2272 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
2273 | fio_mutex_down(startup_mutex); | |
2274 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
2275 | return 0; | |
2276 | } | |
2277 | ||
2e1df07d JA |
2278 | int fio_backend(void) |
2279 | { | |
2280 | struct thread_data *td; | |
2281 | int i; | |
2282 | ||
2283 | if (exec_profile) { | |
2284 | if (load_profile(exec_profile)) | |
2285 | return 1; | |
2286 | free(exec_profile); | |
2287 | exec_profile = NULL; | |
2288 | } | |
2289 | if (!thread_number) | |
2290 | return 0; | |
2291 | ||
2292 | if (write_bw_log) { | |
aee2ab67 JA |
2293 | struct log_params p = { |
2294 | .log_type = IO_LOG_TYPE_BW, | |
2295 | }; | |
2296 | ||
2297 | setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log"); | |
2298 | setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log"); | |
2299 | setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); | |
2e1df07d JA |
2300 | } |
2301 | ||
521da527 | 2302 | startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); |
2e1df07d JA |
2303 | if (startup_mutex == NULL) |
2304 | return 1; | |
2e1df07d JA |
2305 | |
2306 | set_genesis_time(); | |
cef9175e | 2307 | stat_init(); |
5ddc6707 | 2308 | create_helper_thread(); |
2e1df07d JA |
2309 | |
2310 | cgroup_list = smalloc(sizeof(*cgroup_list)); | |
2311 | INIT_FLIST_HEAD(cgroup_list); | |
2312 | ||
2313 | run_threads(); | |
2314 | ||
5ddc6707 | 2315 | wait_for_helper_thread_exit(); |
8aab824f | 2316 | |
2e1df07d | 2317 | if (!fio_abort) { |
83f7b64e | 2318 | __show_run_stats(); |
2e1df07d | 2319 | if (write_bw_log) { |
cb7e0ace JA |
2320 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { |
2321 | struct io_log *log = agg_io_log[i]; | |
2322 | ||
3a5db920 | 2323 | flush_log(log, 0); |
518dac09 | 2324 | free_log(log); |
cb7e0ace | 2325 | } |
2e1df07d JA |
2326 | } |
2327 | } | |
2328 | ||
fda2cfac | 2329 | for_each_td(td, i) { |
2e1df07d | 2330 | fio_options_free(td); |
8049adc1 JA |
2331 | if (td->rusage_sem) { |
2332 | fio_mutex_remove(td->rusage_sem); | |
2333 | td->rusage_sem = NULL; | |
2334 | } | |
fda2cfac | 2335 | } |
2e1df07d | 2336 | |
a462baef | 2337 | free_disk_util(); |
2e1df07d JA |
2338 | cgroup_kill(cgroup_list); |
2339 | sfree(cgroup_list); | |
2340 | sfree(cgroup_mnt); | |
2341 | ||
2342 | fio_mutex_remove(startup_mutex); | |
cef9175e | 2343 | stat_exit(); |
2e1df07d JA |
2344 | return exit_value; |
2345 | } |