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" |
2e1df07d | 57 | |
5ddc6707 JA |
58 | static pthread_t helper_thread; |
59 | static pthread_mutex_t helper_lock; | |
60 | pthread_cond_t helper_cond; | |
61 | int helper_do_stat = 0; | |
8aab824f | 62 | |
2e1df07d | 63 | static struct fio_mutex *startup_mutex; |
2e1df07d JA |
64 | static struct flist_head *cgroup_list; |
65 | static char *cgroup_mnt; | |
66 | static int exit_value; | |
67 | static volatile int fio_abort; | |
3a5f6bde JA |
68 | static unsigned int nr_process = 0; |
69 | static unsigned int nr_thread = 0; | |
2e1df07d | 70 | |
6eaf09d6 | 71 | struct io_log *agg_io_log[DDIR_RWDIR_CNT]; |
2e1df07d | 72 | |
a3efc919 JA |
73 | int groupid = 0; |
74 | unsigned int thread_number = 0; | |
108fea77 | 75 | unsigned int stat_number = 0; |
a3efc919 JA |
76 | int shm_id = 0; |
77 | int temp_stall_ts; | |
78 | unsigned long done_secs = 0; | |
5ddc6707 | 79 | volatile int helper_exit = 0; |
a3efc919 | 80 | |
2e1df07d | 81 | #define PAGE_ALIGN(buf) \ |
e43606c2 | 82 | (char *) (((uintptr_t) (buf) + page_mask) & ~page_mask) |
2e1df07d JA |
83 | |
84 | #define JOB_START_TIMEOUT (5 * 1000) | |
85 | ||
86 | static void sig_int(int sig) | |
87 | { | |
88 | if (threads) { | |
89 | if (is_backend) | |
90 | fio_server_got_signal(sig); | |
91 | else { | |
92 | log_info("\nfio: terminating on signal %d\n", sig); | |
e411c301 | 93 | log_info_flush(); |
2e1df07d JA |
94 | exit_value = 128; |
95 | } | |
96 | ||
97 | fio_terminate_threads(TERMINATE_ALL); | |
98 | } | |
99 | } | |
100 | ||
4c6d91e8 JA |
101 | static void sig_show_status(int sig) |
102 | { | |
103 | show_running_run_stats(); | |
104 | } | |
105 | ||
2e1df07d JA |
106 | static void set_sig_handlers(void) |
107 | { | |
108 | struct sigaction act; | |
109 | ||
110 | memset(&act, 0, sizeof(act)); | |
111 | act.sa_handler = sig_int; | |
112 | act.sa_flags = SA_RESTART; | |
113 | sigaction(SIGINT, &act, NULL); | |
114 | ||
115 | memset(&act, 0, sizeof(act)); | |
116 | act.sa_handler = sig_int; | |
117 | act.sa_flags = SA_RESTART; | |
118 | sigaction(SIGTERM, &act, NULL); | |
119 | ||
2f694507 BC |
120 | /* Windows uses SIGBREAK as a quit signal from other applications */ |
121 | #ifdef WIN32 | |
122 | memset(&act, 0, sizeof(act)); | |
123 | act.sa_handler = sig_int; | |
124 | act.sa_flags = SA_RESTART; | |
125 | sigaction(SIGBREAK, &act, NULL); | |
126 | #endif | |
127 | ||
4c6d91e8 JA |
128 | memset(&act, 0, sizeof(act)); |
129 | act.sa_handler = sig_show_status; | |
130 | act.sa_flags = SA_RESTART; | |
131 | sigaction(SIGUSR1, &act, NULL); | |
132 | ||
2e1df07d JA |
133 | if (is_backend) { |
134 | memset(&act, 0, sizeof(act)); | |
135 | act.sa_handler = sig_int; | |
136 | act.sa_flags = SA_RESTART; | |
137 | sigaction(SIGPIPE, &act, NULL); | |
138 | } | |
139 | } | |
140 | ||
141 | /* | |
142 | * Check if we are above the minimum rate given. | |
143 | */ | |
144 | static int __check_min_rate(struct thread_data *td, struct timeval *now, | |
145 | enum fio_ddir ddir) | |
146 | { | |
147 | unsigned long long bytes = 0; | |
148 | unsigned long iops = 0; | |
149 | unsigned long spent; | |
150 | unsigned long rate; | |
151 | unsigned int ratemin = 0; | |
152 | unsigned int rate_iops = 0; | |
153 | unsigned int rate_iops_min = 0; | |
154 | ||
155 | assert(ddir_rw(ddir)); | |
156 | ||
157 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) | |
158 | return 0; | |
159 | ||
160 | /* | |
161 | * allow a 2 second settle period in the beginning | |
162 | */ | |
163 | if (mtime_since(&td->start, now) < 2000) | |
164 | return 0; | |
165 | ||
166 | iops += td->this_io_blocks[ddir]; | |
167 | bytes += td->this_io_bytes[ddir]; | |
168 | ratemin += td->o.ratemin[ddir]; | |
169 | rate_iops += td->o.rate_iops[ddir]; | |
170 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
171 | ||
172 | /* | |
173 | * if rate blocks is set, sample is running | |
174 | */ | |
175 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { | |
176 | spent = mtime_since(&td->lastrate[ddir], now); | |
177 | if (spent < td->o.ratecycle) | |
178 | return 0; | |
179 | ||
180 | if (td->o.rate[ddir]) { | |
181 | /* | |
182 | * check bandwidth specified rate | |
183 | */ | |
184 | if (bytes < td->rate_bytes[ddir]) { | |
185 | log_err("%s: min rate %u not met\n", td->o.name, | |
186 | ratemin); | |
187 | return 1; | |
188 | } else { | |
49cba9b3 JA |
189 | if (spent) |
190 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; | |
191 | else | |
192 | rate = 0; | |
193 | ||
2e1df07d JA |
194 | if (rate < ratemin || |
195 | bytes < td->rate_bytes[ddir]) { | |
196 | log_err("%s: min rate %u not met, got" | |
197 | " %luKB/sec\n", td->o.name, | |
198 | ratemin, rate); | |
199 | return 1; | |
200 | } | |
201 | } | |
202 | } else { | |
203 | /* | |
204 | * checks iops specified rate | |
205 | */ | |
206 | if (iops < rate_iops) { | |
207 | log_err("%s: min iops rate %u not met\n", | |
208 | td->o.name, rate_iops); | |
209 | return 1; | |
210 | } else { | |
4c707a3b JA |
211 | if (spent) |
212 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; | |
213 | else | |
214 | rate = 0; | |
215 | ||
2e1df07d JA |
216 | if (rate < rate_iops_min || |
217 | iops < td->rate_blocks[ddir]) { | |
218 | log_err("%s: min iops rate %u not met," | |
219 | " got %lu\n", td->o.name, | |
220 | rate_iops_min, rate); | |
221 | } | |
222 | } | |
223 | } | |
224 | } | |
225 | ||
226 | td->rate_bytes[ddir] = bytes; | |
227 | td->rate_blocks[ddir] = iops; | |
228 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
229 | return 0; | |
230 | } | |
231 | ||
232 | static int check_min_rate(struct thread_data *td, struct timeval *now, | |
100f49f1 | 233 | uint64_t *bytes_done) |
2e1df07d JA |
234 | { |
235 | int ret = 0; | |
236 | ||
6eaf09d6 SL |
237 | if (bytes_done[DDIR_READ]) |
238 | ret |= __check_min_rate(td, now, DDIR_READ); | |
239 | if (bytes_done[DDIR_WRITE]) | |
240 | ret |= __check_min_rate(td, now, DDIR_WRITE); | |
241 | if (bytes_done[DDIR_TRIM]) | |
242 | ret |= __check_min_rate(td, now, DDIR_TRIM); | |
2e1df07d JA |
243 | |
244 | return ret; | |
245 | } | |
246 | ||
247 | /* | |
248 | * When job exits, we can cancel the in-flight IO if we are using async | |
249 | * io. Attempt to do so. | |
250 | */ | |
251 | static void cleanup_pending_aio(struct thread_data *td) | |
252 | { | |
2e1df07d JA |
253 | int r; |
254 | ||
255 | /* | |
256 | * get immediately available events, if any | |
257 | */ | |
258 | r = io_u_queued_complete(td, 0, NULL); | |
259 | if (r < 0) | |
260 | return; | |
261 | ||
262 | /* | |
263 | * now cancel remaining active events | |
264 | */ | |
265 | if (td->io_ops->cancel) { | |
2ae0b204 JA |
266 | struct io_u *io_u; |
267 | int i; | |
2e1df07d | 268 | |
2ae0b204 JA |
269 | io_u_qiter(&td->io_u_all, io_u, i) { |
270 | if (io_u->flags & IO_U_F_FLIGHT) { | |
2e1df07d JA |
271 | r = td->io_ops->cancel(td, io_u); |
272 | if (!r) | |
273 | put_io_u(td, io_u); | |
274 | } | |
275 | } | |
276 | } | |
277 | ||
278 | if (td->cur_depth) | |
279 | r = io_u_queued_complete(td, td->cur_depth, NULL); | |
280 | } | |
281 | ||
282 | /* | |
283 | * Helper to handle the final sync of a file. Works just like the normal | |
284 | * io path, just does everything sync. | |
285 | */ | |
286 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
287 | { | |
288 | struct io_u *io_u = __get_io_u(td); | |
289 | int ret; | |
290 | ||
291 | if (!io_u) | |
292 | return 1; | |
293 | ||
294 | io_u->ddir = DDIR_SYNC; | |
295 | io_u->file = f; | |
296 | ||
297 | if (td_io_prep(td, io_u)) { | |
298 | put_io_u(td, io_u); | |
299 | return 1; | |
300 | } | |
301 | ||
302 | requeue: | |
303 | ret = td_io_queue(td, io_u); | |
304 | if (ret < 0) { | |
305 | td_verror(td, io_u->error, "td_io_queue"); | |
306 | put_io_u(td, io_u); | |
307 | return 1; | |
308 | } else if (ret == FIO_Q_QUEUED) { | |
309 | if (io_u_queued_complete(td, 1, NULL) < 0) | |
310 | return 1; | |
311 | } else if (ret == FIO_Q_COMPLETED) { | |
312 | if (io_u->error) { | |
313 | td_verror(td, io_u->error, "td_io_queue"); | |
314 | return 1; | |
315 | } | |
316 | ||
317 | if (io_u_sync_complete(td, io_u, NULL) < 0) | |
318 | return 1; | |
319 | } else if (ret == FIO_Q_BUSY) { | |
320 | if (td_io_commit(td)) | |
321 | return 1; | |
322 | goto requeue; | |
323 | } | |
324 | ||
325 | return 0; | |
326 | } | |
a3efc919 | 327 | |
61ee0f86 JA |
328 | static int fio_file_fsync(struct thread_data *td, struct fio_file *f) |
329 | { | |
330 | int ret; | |
331 | ||
332 | if (fio_file_open(f)) | |
333 | return fio_io_sync(td, f); | |
334 | ||
335 | if (td_io_open_file(td, f)) | |
336 | return 1; | |
337 | ||
338 | ret = fio_io_sync(td, f); | |
339 | td_io_close_file(td, f); | |
340 | return ret; | |
341 | } | |
342 | ||
2e1df07d JA |
343 | static inline void __update_tv_cache(struct thread_data *td) |
344 | { | |
345 | fio_gettime(&td->tv_cache, NULL); | |
346 | } | |
347 | ||
348 | static inline void update_tv_cache(struct thread_data *td) | |
349 | { | |
350 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
351 | __update_tv_cache(td); | |
352 | } | |
353 | ||
354 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
355 | { | |
356 | if (in_ramp_time(td)) | |
357 | return 0; | |
358 | if (!td->o.timeout) | |
359 | return 0; | |
0de5b26f | 360 | if (utime_since(&td->epoch, t) >= td->o.timeout) |
2e1df07d JA |
361 | return 1; |
362 | ||
363 | return 0; | |
364 | } | |
365 | ||
366 | static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, | |
367 | int *retptr) | |
368 | { | |
369 | int ret = *retptr; | |
370 | ||
371 | if (ret < 0 || td->error) { | |
8b28bd41 DM |
372 | int err = td->error; |
373 | enum error_type_bit eb; | |
2e1df07d JA |
374 | |
375 | if (ret < 0) | |
376 | err = -ret; | |
2e1df07d | 377 | |
8b28bd41 DM |
378 | eb = td_error_type(ddir, err); |
379 | if (!(td->o.continue_on_error & (1 << eb))) | |
2e1df07d JA |
380 | return 1; |
381 | ||
8b28bd41 | 382 | if (td_non_fatal_error(td, eb, err)) { |
2e1df07d JA |
383 | /* |
384 | * Continue with the I/Os in case of | |
385 | * a non fatal error. | |
386 | */ | |
387 | update_error_count(td, err); | |
388 | td_clear_error(td); | |
389 | *retptr = 0; | |
390 | return 0; | |
391 | } else if (td->o.fill_device && err == ENOSPC) { | |
392 | /* | |
393 | * We expect to hit this error if | |
394 | * fill_device option is set. | |
395 | */ | |
396 | td_clear_error(td); | |
ebea2133 | 397 | fio_mark_td_terminate(td); |
2e1df07d JA |
398 | return 1; |
399 | } else { | |
400 | /* | |
401 | * Stop the I/O in case of a fatal | |
402 | * error. | |
403 | */ | |
404 | update_error_count(td, err); | |
405 | return 1; | |
406 | } | |
407 | } | |
408 | ||
409 | return 0; | |
410 | } | |
411 | ||
c97f1ad6 JA |
412 | static void check_update_rusage(struct thread_data *td) |
413 | { | |
414 | if (td->update_rusage) { | |
415 | td->update_rusage = 0; | |
416 | update_rusage_stat(td); | |
417 | fio_mutex_up(td->rusage_sem); | |
418 | } | |
419 | } | |
420 | ||
69fea81e JA |
421 | static int wait_for_completions(struct thread_data *td, struct timeval *time, |
422 | uint64_t *bytes_done) | |
423 | { | |
424 | const int full = queue_full(td); | |
425 | int min_evts = 0; | |
426 | int ret; | |
427 | ||
428 | /* | |
429 | * if the queue is full, we MUST reap at least 1 event | |
430 | */ | |
431 | min_evts = min(td->o.iodepth_batch_complete, td->cur_depth); | |
432 | if (full && !min_evts) | |
433 | min_evts = 1; | |
434 | ||
66b6c5ef | 435 | if (time && (__should_check_rate(td, DDIR_READ) || |
69fea81e JA |
436 | __should_check_rate(td, DDIR_WRITE) || |
437 | __should_check_rate(td, DDIR_TRIM))) | |
438 | fio_gettime(time, NULL); | |
439 | ||
440 | do { | |
441 | ret = io_u_queued_complete(td, min_evts, bytes_done); | |
442 | if (ret < 0) | |
443 | break; | |
444 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
445 | ||
446 | return ret; | |
447 | } | |
448 | ||
2e1df07d JA |
449 | /* |
450 | * The main verify engine. Runs over the writes we previously submitted, | |
451 | * reads the blocks back in, and checks the crc/md5 of the data. | |
452 | */ | |
100f49f1 | 453 | static void do_verify(struct thread_data *td, uint64_t verify_bytes) |
2e1df07d | 454 | { |
100f49f1 | 455 | uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; |
2e1df07d JA |
456 | struct fio_file *f; |
457 | struct io_u *io_u; | |
458 | int ret, min_events; | |
459 | unsigned int i; | |
460 | ||
461 | dprint(FD_VERIFY, "starting loop\n"); | |
462 | ||
463 | /* | |
464 | * sync io first and invalidate cache, to make sure we really | |
465 | * read from disk. | |
466 | */ | |
467 | for_each_file(td, f, i) { | |
468 | if (!fio_file_open(f)) | |
469 | continue; | |
470 | if (fio_io_sync(td, f)) | |
471 | break; | |
472 | if (file_invalidate_cache(td, f)) | |
473 | break; | |
474 | } | |
475 | ||
c97f1ad6 JA |
476 | check_update_rusage(td); |
477 | ||
2e1df07d JA |
478 | if (td->error) |
479 | return; | |
480 | ||
481 | td_set_runstate(td, TD_VERIFYING); | |
482 | ||
483 | io_u = NULL; | |
484 | while (!td->terminate) { | |
fbccf46c | 485 | enum fio_ddir ddir; |
2e1df07d JA |
486 | int ret2, full; |
487 | ||
488 | update_tv_cache(td); | |
c97f1ad6 | 489 | check_update_rusage(td); |
2e1df07d JA |
490 | |
491 | if (runtime_exceeded(td, &td->tv_cache)) { | |
492 | __update_tv_cache(td); | |
493 | if (runtime_exceeded(td, &td->tv_cache)) { | |
ebea2133 | 494 | fio_mark_td_terminate(td); |
2e1df07d JA |
495 | break; |
496 | } | |
497 | } | |
498 | ||
9e684a49 DE |
499 | if (flow_threshold_exceeded(td)) |
500 | continue; | |
501 | ||
44cbc6da JA |
502 | if (!td->o.experimental_verify) { |
503 | io_u = __get_io_u(td); | |
504 | if (!io_u) | |
505 | break; | |
2e1df07d | 506 | |
44cbc6da JA |
507 | if (get_next_verify(td, io_u)) { |
508 | put_io_u(td, io_u); | |
509 | break; | |
510 | } | |
2e1df07d | 511 | |
44cbc6da JA |
512 | if (td_io_prep(td, io_u)) { |
513 | put_io_u(td, io_u); | |
514 | break; | |
515 | } | |
516 | } else { | |
100f49f1 JA |
517 | if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes) |
518 | break; | |
519 | ||
bcd5abfa | 520 | while ((io_u = get_io_u(td)) != NULL) { |
002fe734 JA |
521 | if (IS_ERR(io_u)) { |
522 | io_u = NULL; | |
523 | ret = FIO_Q_BUSY; | |
524 | goto reap; | |
525 | } | |
526 | ||
bcd5abfa JA |
527 | /* |
528 | * We are only interested in the places where | |
529 | * we wrote or trimmed IOs. Turn those into | |
530 | * reads for verification purposes. | |
531 | */ | |
532 | if (io_u->ddir == DDIR_READ) { | |
533 | /* | |
534 | * Pretend we issued it for rwmix | |
535 | * accounting | |
536 | */ | |
537 | td->io_issues[DDIR_READ]++; | |
538 | put_io_u(td, io_u); | |
539 | continue; | |
540 | } else if (io_u->ddir == DDIR_TRIM) { | |
541 | io_u->ddir = DDIR_READ; | |
542 | io_u->flags |= IO_U_F_TRIMMED; | |
543 | break; | |
544 | } else if (io_u->ddir == DDIR_WRITE) { | |
545 | io_u->ddir = DDIR_READ; | |
546 | break; | |
547 | } else { | |
548 | put_io_u(td, io_u); | |
549 | continue; | |
550 | } | |
551 | } | |
44cbc6da | 552 | |
bcd5abfa | 553 | if (!io_u) |
44cbc6da | 554 | break; |
2e1df07d JA |
555 | } |
556 | ||
ca09be4b JA |
557 | if (verify_state_should_stop(td, io_u)) { |
558 | put_io_u(td, io_u); | |
559 | break; | |
560 | } | |
561 | ||
2e1df07d JA |
562 | if (td->o.verify_async) |
563 | io_u->end_io = verify_io_u_async; | |
564 | else | |
565 | io_u->end_io = verify_io_u; | |
566 | ||
fbccf46c JA |
567 | ddir = io_u->ddir; |
568 | ||
2e1df07d JA |
569 | ret = td_io_queue(td, io_u); |
570 | switch (ret) { | |
571 | case FIO_Q_COMPLETED: | |
572 | if (io_u->error) { | |
573 | ret = -io_u->error; | |
574 | clear_io_u(td, io_u); | |
575 | } else if (io_u->resid) { | |
576 | int bytes = io_u->xfer_buflen - io_u->resid; | |
577 | ||
578 | /* | |
579 | * zero read, fail | |
580 | */ | |
581 | if (!bytes) { | |
582 | td_verror(td, EIO, "full resid"); | |
583 | put_io_u(td, io_u); | |
584 | break; | |
585 | } | |
586 | ||
587 | io_u->xfer_buflen = io_u->resid; | |
588 | io_u->xfer_buf += bytes; | |
589 | io_u->offset += bytes; | |
590 | ||
591 | if (ddir_rw(io_u->ddir)) | |
592 | td->ts.short_io_u[io_u->ddir]++; | |
593 | ||
594 | f = io_u->file; | |
595 | if (io_u->offset == f->real_file_size) | |
596 | goto sync_done; | |
597 | ||
598 | requeue_io_u(td, &io_u); | |
599 | } else { | |
600 | sync_done: | |
100f49f1 | 601 | ret = io_u_sync_complete(td, io_u, bytes_done); |
2e1df07d JA |
602 | if (ret < 0) |
603 | break; | |
604 | } | |
605 | continue; | |
606 | case FIO_Q_QUEUED: | |
607 | break; | |
608 | case FIO_Q_BUSY: | |
609 | requeue_io_u(td, &io_u); | |
610 | ret2 = td_io_commit(td); | |
611 | if (ret2 < 0) | |
612 | ret = ret2; | |
613 | break; | |
614 | default: | |
615 | assert(ret < 0); | |
616 | td_verror(td, -ret, "td_io_queue"); | |
617 | break; | |
618 | } | |
619 | ||
fbccf46c | 620 | if (break_on_this_error(td, ddir, &ret)) |
2e1df07d JA |
621 | break; |
622 | ||
623 | /* | |
624 | * if we can queue more, do so. but check if there are | |
625 | * completed io_u's first. Note that we can get BUSY even | |
626 | * without IO queued, if the system is resource starved. | |
627 | */ | |
002fe734 | 628 | reap: |
2e1df07d | 629 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
69fea81e JA |
630 | if (full || !td->o.iodepth_batch_complete) |
631 | ret = wait_for_completions(td, NULL, bytes_done); | |
2e1df07d | 632 | |
2e1df07d JA |
633 | if (ret < 0) |
634 | break; | |
635 | } | |
636 | ||
c97f1ad6 JA |
637 | check_update_rusage(td); |
638 | ||
2e1df07d JA |
639 | if (!td->error) { |
640 | min_events = td->cur_depth; | |
641 | ||
642 | if (min_events) | |
643 | ret = io_u_queued_complete(td, min_events, NULL); | |
644 | } else | |
645 | cleanup_pending_aio(td); | |
646 | ||
647 | td_set_runstate(td, TD_RUNNING); | |
648 | ||
649 | dprint(FD_VERIFY, "exiting loop\n"); | |
650 | } | |
651 | ||
3939fe85 JA |
652 | static unsigned int exceeds_number_ios(struct thread_data *td) |
653 | { | |
654 | unsigned long long number_ios; | |
655 | ||
656 | if (!td->o.number_ios) | |
657 | return 0; | |
658 | ||
659 | number_ios = ddir_rw_sum(td->this_io_blocks); | |
660 | number_ios += td->io_u_queued + td->io_u_in_flight; | |
661 | ||
662 | return number_ios >= td->o.number_ios; | |
663 | } | |
664 | ||
f7078f7b JA |
665 | static int io_bytes_exceeded(struct thread_data *td) |
666 | { | |
77731b29 | 667 | unsigned long long bytes, limit; |
f7078f7b JA |
668 | |
669 | if (td_rw(td)) | |
6eaf09d6 | 670 | bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE]; |
f7078f7b | 671 | else if (td_write(td)) |
6eaf09d6 SL |
672 | bytes = td->this_io_bytes[DDIR_WRITE]; |
673 | else if (td_read(td)) | |
674 | bytes = td->this_io_bytes[DDIR_READ]; | |
f7078f7b | 675 | else |
6eaf09d6 | 676 | bytes = td->this_io_bytes[DDIR_TRIM]; |
f7078f7b | 677 | |
77731b29 JA |
678 | if (td->o.io_limit) |
679 | limit = td->o.io_limit; | |
680 | else | |
681 | limit = td->o.size; | |
682 | ||
683 | return bytes >= limit || exceeds_number_ios(td); | |
f7078f7b JA |
684 | } |
685 | ||
2e1df07d JA |
686 | /* |
687 | * Main IO worker function. It retrieves io_u's to process and queues | |
688 | * and reaps them, checking for rate and errors along the way. | |
100f49f1 JA |
689 | * |
690 | * Returns number of bytes written and trimmed. | |
2e1df07d | 691 | */ |
100f49f1 | 692 | static uint64_t do_io(struct thread_data *td) |
2e1df07d | 693 | { |
100f49f1 | 694 | uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; |
2e1df07d JA |
695 | unsigned int i; |
696 | int ret = 0; | |
c2703bf3 | 697 | uint64_t total_bytes, bytes_issued = 0; |
2e1df07d JA |
698 | |
699 | if (in_ramp_time(td)) | |
700 | td_set_runstate(td, TD_RAMP); | |
701 | else | |
702 | td_set_runstate(td, TD_RUNNING); | |
703 | ||
3e260a46 JA |
704 | lat_target_init(td); |
705 | ||
78a6469c JA |
706 | /* |
707 | * If verify_backlog is enabled, we'll run the verify in this | |
708 | * handler as well. For that case, we may need up to twice the | |
709 | * amount of bytes. | |
710 | */ | |
c2703bf3 | 711 | total_bytes = td->o.size; |
78a6469c JA |
712 | if (td->o.verify != VERIFY_NONE && |
713 | (td_write(td) && td->o.verify_backlog)) | |
c2703bf3 JA |
714 | total_bytes += td->o.size; |
715 | ||
f7078f7b | 716 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
c04e4661 DE |
717 | (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) || |
718 | td->o.time_based) { | |
2e1df07d | 719 | struct timeval comp_time; |
2e1df07d JA |
720 | struct io_u *io_u; |
721 | int ret2, full; | |
722 | enum fio_ddir ddir; | |
723 | ||
c97f1ad6 JA |
724 | check_update_rusage(td); |
725 | ||
7d7803fa | 726 | if (td->terminate || td->done) |
2e1df07d JA |
727 | break; |
728 | ||
729 | update_tv_cache(td); | |
730 | ||
731 | if (runtime_exceeded(td, &td->tv_cache)) { | |
732 | __update_tv_cache(td); | |
733 | if (runtime_exceeded(td, &td->tv_cache)) { | |
ebea2133 | 734 | fio_mark_td_terminate(td); |
2e1df07d JA |
735 | break; |
736 | } | |
737 | } | |
738 | ||
9e684a49 DE |
739 | if (flow_threshold_exceeded(td)) |
740 | continue; | |
741 | ||
c2703bf3 | 742 | if (bytes_issued >= total_bytes) |
20876c53 JC |
743 | break; |
744 | ||
2e1df07d | 745 | io_u = get_io_u(td); |
002fe734 JA |
746 | if (IS_ERR_OR_NULL(io_u)) { |
747 | int err = PTR_ERR(io_u); | |
748 | ||
749 | io_u = NULL; | |
750 | if (err == -EBUSY) { | |
751 | ret = FIO_Q_BUSY; | |
752 | goto reap; | |
753 | } | |
3e260a46 JA |
754 | if (td->o.latency_target) |
755 | goto reap; | |
2e1df07d | 756 | break; |
3e260a46 | 757 | } |
2e1df07d JA |
758 | |
759 | ddir = io_u->ddir; | |
760 | ||
761 | /* | |
82af2a7c JA |
762 | * Add verification end_io handler if: |
763 | * - Asked to verify (!td_rw(td)) | |
764 | * - Or the io_u is from our verify list (mixed write/ver) | |
2e1df07d JA |
765 | */ |
766 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && | |
82af2a7c | 767 | ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { |
c4b6117b PV |
768 | |
769 | if (!td->o.verify_pattern_bytes) { | |
d6b72507 | 770 | io_u->rand_seed = __rand(&td->verify_state); |
c4b6117b | 771 | if (sizeof(int) != sizeof(long *)) |
d6b72507 | 772 | io_u->rand_seed *= __rand(&td->verify_state); |
c4b6117b PV |
773 | } |
774 | ||
ca09be4b JA |
775 | if (verify_state_should_stop(td, io_u)) { |
776 | put_io_u(td, io_u); | |
777 | break; | |
778 | } | |
779 | ||
2e1df07d JA |
780 | if (td->o.verify_async) |
781 | io_u->end_io = verify_io_u_async; | |
782 | else | |
783 | io_u->end_io = verify_io_u; | |
784 | td_set_runstate(td, TD_VERIFYING); | |
785 | } else if (in_ramp_time(td)) | |
786 | td_set_runstate(td, TD_RAMP); | |
787 | else | |
788 | td_set_runstate(td, TD_RUNNING); | |
789 | ||
9a50c5c5 | 790 | /* |
f9401285 JA |
791 | * Always log IO before it's issued, so we know the specific |
792 | * order of it. The logged unit will track when the IO has | |
793 | * completed. | |
9a50c5c5 | 794 | */ |
c4b6117b PV |
795 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
796 | td->o.do_verify && | |
797 | td->o.verify != VERIFY_NONE && | |
f9401285 | 798 | !td->o.experimental_verify) |
c4b6117b PV |
799 | log_io_piece(td, io_u); |
800 | ||
2e1df07d JA |
801 | ret = td_io_queue(td, io_u); |
802 | switch (ret) { | |
803 | case FIO_Q_COMPLETED: | |
804 | if (io_u->error) { | |
805 | ret = -io_u->error; | |
890b6656 | 806 | unlog_io_piece(td, io_u); |
2e1df07d JA |
807 | clear_io_u(td, io_u); |
808 | } else if (io_u->resid) { | |
809 | int bytes = io_u->xfer_buflen - io_u->resid; | |
810 | struct fio_file *f = io_u->file; | |
811 | ||
20876c53 | 812 | bytes_issued += bytes; |
890b6656 JA |
813 | |
814 | trim_io_piece(td, io_u); | |
815 | ||
2e1df07d JA |
816 | /* |
817 | * zero read, fail | |
818 | */ | |
819 | if (!bytes) { | |
890b6656 | 820 | unlog_io_piece(td, io_u); |
2e1df07d JA |
821 | td_verror(td, EIO, "full resid"); |
822 | put_io_u(td, io_u); | |
823 | break; | |
824 | } | |
825 | ||
826 | io_u->xfer_buflen = io_u->resid; | |
827 | io_u->xfer_buf += bytes; | |
828 | io_u->offset += bytes; | |
829 | ||
830 | if (ddir_rw(io_u->ddir)) | |
831 | td->ts.short_io_u[io_u->ddir]++; | |
832 | ||
833 | if (io_u->offset == f->real_file_size) | |
834 | goto sync_done; | |
835 | ||
836 | requeue_io_u(td, &io_u); | |
837 | } else { | |
838 | sync_done: | |
6eaf09d6 SL |
839 | if (__should_check_rate(td, DDIR_READ) || |
840 | __should_check_rate(td, DDIR_WRITE) || | |
841 | __should_check_rate(td, DDIR_TRIM)) | |
2e1df07d JA |
842 | fio_gettime(&comp_time, NULL); |
843 | ||
844 | ret = io_u_sync_complete(td, io_u, bytes_done); | |
845 | if (ret < 0) | |
846 | break; | |
20876c53 | 847 | bytes_issued += io_u->xfer_buflen; |
2e1df07d JA |
848 | } |
849 | break; | |
850 | case FIO_Q_QUEUED: | |
851 | /* | |
852 | * if the engine doesn't have a commit hook, | |
853 | * the io_u is really queued. if it does have such | |
854 | * a hook, it has to call io_u_queued() itself. | |
855 | */ | |
856 | if (td->io_ops->commit == NULL) | |
857 | io_u_queued(td, io_u); | |
20876c53 | 858 | bytes_issued += io_u->xfer_buflen; |
2e1df07d JA |
859 | break; |
860 | case FIO_Q_BUSY: | |
890b6656 | 861 | unlog_io_piece(td, io_u); |
2e1df07d JA |
862 | requeue_io_u(td, &io_u); |
863 | ret2 = td_io_commit(td); | |
864 | if (ret2 < 0) | |
865 | ret = ret2; | |
866 | break; | |
867 | default: | |
868 | assert(ret < 0); | |
869 | put_io_u(td, io_u); | |
870 | break; | |
871 | } | |
872 | ||
873 | if (break_on_this_error(td, ddir, &ret)) | |
874 | break; | |
875 | ||
876 | /* | |
877 | * See if we need to complete some commands. Note that we | |
878 | * can get BUSY even without IO queued, if the system is | |
879 | * resource starved. | |
880 | */ | |
3e260a46 | 881 | reap: |
2e1df07d | 882 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
69fea81e JA |
883 | if (full || !td->o.iodepth_batch_complete) |
884 | ret = wait_for_completions(td, &comp_time, bytes_done); | |
2e1df07d JA |
885 | if (ret < 0) |
886 | break; | |
d5abee06 | 887 | if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO)) |
2e1df07d JA |
888 | continue; |
889 | ||
890 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { | |
891 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
892 | if (exitall_on_terminate) | |
893 | fio_terminate_threads(td->groupid); | |
894 | td_verror(td, EIO, "check_min_rate"); | |
895 | break; | |
896 | } | |
897 | } | |
3e260a46 JA |
898 | if (!in_ramp_time(td) && td->o.latency_target) |
899 | lat_target_check(td); | |
2e1df07d JA |
900 | |
901 | if (td->o.thinktime) { | |
902 | unsigned long long b; | |
903 | ||
342f4be4 | 904 | b = ddir_rw_sum(td->io_blocks); |
2e1df07d JA |
905 | if (!(b % td->o.thinktime_blocks)) { |
906 | int left; | |
907 | ||
002e7183 JA |
908 | io_u_quiesce(td); |
909 | ||
2e1df07d JA |
910 | if (td->o.thinktime_spin) |
911 | usec_spin(td->o.thinktime_spin); | |
912 | ||
913 | left = td->o.thinktime - td->o.thinktime_spin; | |
914 | if (left) | |
915 | usec_sleep(td, left); | |
916 | } | |
917 | } | |
918 | } | |
919 | ||
c97f1ad6 JA |
920 | check_update_rusage(td); |
921 | ||
2e1df07d | 922 | if (td->trim_entries) |
4e0a8fa2 | 923 | log_err("fio: %lu trim entries leaked?\n", td->trim_entries); |
2e1df07d JA |
924 | |
925 | if (td->o.fill_device && td->error == ENOSPC) { | |
926 | td->error = 0; | |
ebea2133 | 927 | fio_mark_td_terminate(td); |
2e1df07d JA |
928 | } |
929 | if (!td->error) { | |
930 | struct fio_file *f; | |
931 | ||
932 | i = td->cur_depth; | |
933 | if (i) { | |
5bd5f71a | 934 | ret = io_u_queued_complete(td, i, bytes_done); |
2e1df07d JA |
935 | if (td->o.fill_device && td->error == ENOSPC) |
936 | td->error = 0; | |
937 | } | |
938 | ||
939 | if (should_fsync(td) && td->o.end_fsync) { | |
940 | td_set_runstate(td, TD_FSYNCING); | |
941 | ||
942 | for_each_file(td, f, i) { | |
61ee0f86 | 943 | if (!fio_file_fsync(td, f)) |
2e1df07d | 944 | continue; |
61ee0f86 JA |
945 | |
946 | log_err("fio: end_fsync failed for file %s\n", | |
947 | f->file_name); | |
2e1df07d JA |
948 | } |
949 | } | |
950 | } else | |
951 | cleanup_pending_aio(td); | |
952 | ||
953 | /* | |
954 | * stop job if we failed doing any IO | |
955 | */ | |
342f4be4 | 956 | if (!ddir_rw_sum(td->this_io_bytes)) |
2e1df07d | 957 | td->done = 1; |
100f49f1 JA |
958 | |
959 | return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; | |
2e1df07d JA |
960 | } |
961 | ||
962 | static void cleanup_io_u(struct thread_data *td) | |
963 | { | |
2e1df07d JA |
964 | struct io_u *io_u; |
965 | ||
2ae0b204 | 966 | while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) { |
c73ed246 JA |
967 | |
968 | if (td->io_ops->io_u_free) | |
969 | td->io_ops->io_u_free(td, io_u); | |
970 | ||
2e1df07d JA |
971 | fio_memfree(io_u, sizeof(*io_u)); |
972 | } | |
973 | ||
974 | free_io_mem(td); | |
2ae0b204 JA |
975 | |
976 | io_u_rexit(&td->io_u_requeues); | |
977 | io_u_qexit(&td->io_u_freelist); | |
978 | io_u_qexit(&td->io_u_all); | |
ca09be4b JA |
979 | |
980 | if (td->last_write_comp) | |
981 | sfree(td->last_write_comp); | |
2e1df07d JA |
982 | } |
983 | ||
984 | static int init_io_u(struct thread_data *td) | |
985 | { | |
986 | struct io_u *io_u; | |
9c42684e | 987 | unsigned int max_bs, min_write; |
2e1df07d | 988 | int cl_align, i, max_units; |
2ae0b204 | 989 | int data_xfer = 1, err; |
2e1df07d JA |
990 | char *p; |
991 | ||
992 | max_units = td->o.iodepth; | |
74f4b020 | 993 | max_bs = td_max_bs(td); |
9c42684e | 994 | min_write = td->o.min_bs[DDIR_WRITE]; |
2e1df07d JA |
995 | td->orig_buffer_size = (unsigned long long) max_bs |
996 | * (unsigned long long) max_units; | |
997 | ||
88045e04 | 998 | if ((td->io_ops->flags & FIO_NOIO) || !(td_read(td) || td_write(td))) |
59d8d0f5 JA |
999 | data_xfer = 0; |
1000 | ||
2ae0b204 JA |
1001 | err = 0; |
1002 | err += io_u_rinit(&td->io_u_requeues, td->o.iodepth); | |
1003 | err += io_u_qinit(&td->io_u_freelist, td->o.iodepth); | |
1004 | err += io_u_qinit(&td->io_u_all, td->o.iodepth); | |
1005 | ||
1006 | if (err) { | |
1007 | log_err("fio: failed setting up IO queues\n"); | |
1008 | return 1; | |
1009 | } | |
1010 | ||
fd8a09b8 | 1011 | /* |
1012 | * if we may later need to do address alignment, then add any | |
1013 | * possible adjustment here so that we don't cause a buffer | |
1014 | * overflow later. this adjustment may be too much if we get | |
1015 | * lucky and the allocator gives us an aligned address. | |
1016 | */ | |
d01612f3 CM |
1017 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
1018 | (td->io_ops->flags & FIO_RAWIO)) | |
fd8a09b8 | 1019 | td->orig_buffer_size += page_mask + td->o.mem_align; |
1020 | ||
2e1df07d JA |
1021 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { |
1022 | unsigned long bs; | |
1023 | ||
1024 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; | |
1025 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
1026 | } | |
1027 | ||
1028 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { | |
1029 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
1030 | return 1; | |
1031 | } | |
1032 | ||
59d8d0f5 | 1033 | if (data_xfer && allocate_io_mem(td)) |
2e1df07d JA |
1034 | return 1; |
1035 | ||
d01612f3 | 1036 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
2e1df07d JA |
1037 | (td->io_ops->flags & FIO_RAWIO)) |
1038 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; | |
1039 | else | |
1040 | p = td->orig_buffer; | |
1041 | ||
1042 | cl_align = os_cache_line_size(); | |
1043 | ||
1044 | for (i = 0; i < max_units; i++) { | |
1045 | void *ptr; | |
1046 | ||
1047 | if (td->terminate) | |
1048 | return 1; | |
1049 | ||
1050 | ptr = fio_memalign(cl_align, sizeof(*io_u)); | |
1051 | if (!ptr) { | |
1052 | log_err("fio: unable to allocate aligned memory\n"); | |
1053 | break; | |
1054 | } | |
1055 | ||
1056 | io_u = ptr; | |
1057 | memset(io_u, 0, sizeof(*io_u)); | |
2ae0b204 | 1058 | INIT_FLIST_HEAD(&io_u->verify_list); |
2e1df07d JA |
1059 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
1060 | ||
59d8d0f5 | 1061 | if (data_xfer) { |
2e1df07d JA |
1062 | io_u->buf = p; |
1063 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); | |
1064 | ||
1065 | if (td_write(td)) | |
9c42684e | 1066 | io_u_fill_buffer(td, io_u, min_write, max_bs); |
2e1df07d JA |
1067 | if (td_write(td) && td->o.verify_pattern_bytes) { |
1068 | /* | |
1069 | * Fill the buffer with the pattern if we are | |
1070 | * going to be doing writes. | |
1071 | */ | |
ce35b1ec | 1072 | fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
2e1df07d JA |
1073 | } |
1074 | } | |
1075 | ||
1076 | io_u->index = i; | |
1077 | io_u->flags = IO_U_F_FREE; | |
2ae0b204 JA |
1078 | io_u_qpush(&td->io_u_freelist, io_u); |
1079 | ||
1080 | /* | |
1081 | * io_u never leaves this stack, used for iteration of all | |
1082 | * io_u buffers. | |
1083 | */ | |
1084 | io_u_qpush(&td->io_u_all, io_u); | |
c73ed246 JA |
1085 | |
1086 | if (td->io_ops->io_u_init) { | |
1087 | int ret = td->io_ops->io_u_init(td, io_u); | |
1088 | ||
1089 | if (ret) { | |
1090 | log_err("fio: failed to init engine data: %d\n", ret); | |
1091 | return 1; | |
1092 | } | |
1093 | } | |
1094 | ||
2e1df07d JA |
1095 | p += max_bs; |
1096 | } | |
1097 | ||
ca09be4b JA |
1098 | if (td->o.verify != VERIFY_NONE) { |
1099 | td->last_write_comp = scalloc(max_units, sizeof(uint64_t)); | |
1100 | if (!td->last_write_comp) { | |
1101 | log_err("fio: failed to alloc write comp data\n"); | |
1102 | return 1; | |
1103 | } | |
1104 | } | |
1105 | ||
2e1df07d JA |
1106 | return 0; |
1107 | } | |
1108 | ||
1109 | static int switch_ioscheduler(struct thread_data *td) | |
1110 | { | |
1111 | char tmp[256], tmp2[128]; | |
1112 | FILE *f; | |
1113 | int ret; | |
1114 | ||
1115 | if (td->io_ops->flags & FIO_DISKLESSIO) | |
1116 | return 0; | |
1117 | ||
1118 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
1119 | ||
1120 | f = fopen(tmp, "r+"); | |
1121 | if (!f) { | |
1122 | if (errno == ENOENT) { | |
1123 | log_err("fio: os or kernel doesn't support IO scheduler" | |
1124 | " switching\n"); | |
1125 | return 0; | |
1126 | } | |
1127 | td_verror(td, errno, "fopen iosched"); | |
1128 | return 1; | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * Set io scheduler. | |
1133 | */ | |
1134 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); | |
1135 | if (ferror(f) || ret != 1) { | |
1136 | td_verror(td, errno, "fwrite"); | |
1137 | fclose(f); | |
1138 | return 1; | |
1139 | } | |
1140 | ||
1141 | rewind(f); | |
1142 | ||
1143 | /* | |
1144 | * Read back and check that the selected scheduler is now the default. | |
1145 | */ | |
49c6f33d | 1146 | ret = fread(tmp, sizeof(tmp), 1, f); |
2e1df07d JA |
1147 | if (ferror(f) || ret < 0) { |
1148 | td_verror(td, errno, "fread"); | |
1149 | fclose(f); | |
1150 | return 1; | |
1151 | } | |
49c6f33d JA |
1152 | tmp[sizeof(tmp) - 1] = '\0'; |
1153 | ||
2e1df07d JA |
1154 | |
1155 | sprintf(tmp2, "[%s]", td->o.ioscheduler); | |
1156 | if (!strstr(tmp, tmp2)) { | |
1157 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); | |
1158 | td_verror(td, EINVAL, "iosched_switch"); | |
1159 | fclose(f); | |
1160 | return 1; | |
1161 | } | |
1162 | ||
1163 | fclose(f); | |
1164 | return 0; | |
1165 | } | |
1166 | ||
1167 | static int keep_running(struct thread_data *td) | |
1168 | { | |
77731b29 JA |
1169 | unsigned long long limit; |
1170 | ||
2e1df07d JA |
1171 | if (td->done) |
1172 | return 0; | |
1173 | if (td->o.time_based) | |
1174 | return 1; | |
1175 | if (td->o.loops) { | |
1176 | td->o.loops--; | |
1177 | return 1; | |
1178 | } | |
3939fe85 JA |
1179 | if (exceeds_number_ios(td)) |
1180 | return 0; | |
26251d8d | 1181 | |
77731b29 JA |
1182 | if (td->o.io_limit) |
1183 | limit = td->o.io_limit; | |
1184 | else | |
1185 | limit = td->o.size; | |
1186 | ||
1187 | if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) { | |
5bd5f71a JA |
1188 | uint64_t diff; |
1189 | ||
1190 | /* | |
1191 | * If the difference is less than the minimum IO size, we | |
1192 | * are done. | |
1193 | */ | |
77731b29 | 1194 | diff = limit - ddir_rw_sum(td->io_bytes); |
74f4b020 | 1195 | if (diff < td_max_bs(td)) |
5bd5f71a JA |
1196 | return 0; |
1197 | ||
002fe734 JA |
1198 | if (fio_files_done(td)) |
1199 | return 0; | |
1200 | ||
2e1df07d | 1201 | return 1; |
5bd5f71a | 1202 | } |
2e1df07d JA |
1203 | |
1204 | return 0; | |
1205 | } | |
1206 | ||
ce486495 | 1207 | static int exec_string(struct thread_options *o, const char *string, const char *mode) |
2e1df07d | 1208 | { |
ce486495 | 1209 | int ret, newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; |
2e1df07d JA |
1210 | char *str; |
1211 | ||
1212 | str = malloc(newlen); | |
ce486495 | 1213 | sprintf(str, "%s &> %s.%s.txt", string, o->name, mode); |
2e1df07d | 1214 | |
ce486495 | 1215 | log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode); |
2e1df07d JA |
1216 | ret = system(str); |
1217 | if (ret == -1) | |
1218 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1219 | ||
1220 | free(str); | |
1221 | return ret; | |
1222 | } | |
1223 | ||
62167762 JC |
1224 | /* |
1225 | * Dry run to compute correct state of numberio for verification. | |
1226 | */ | |
1227 | static uint64_t do_dry_run(struct thread_data *td) | |
1228 | { | |
1229 | uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 }; | |
1230 | ||
1231 | td_set_runstate(td, TD_RUNNING); | |
1232 | ||
1233 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || | |
1234 | (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td)) { | |
1235 | struct io_u *io_u; | |
1236 | int ret; | |
1237 | ||
1238 | if (td->terminate || td->done) | |
1239 | break; | |
1240 | ||
1241 | io_u = get_io_u(td); | |
1242 | if (!io_u) | |
1243 | break; | |
1244 | ||
1245 | io_u->flags |= IO_U_F_FLIGHT; | |
1246 | io_u->error = 0; | |
1247 | io_u->resid = 0; | |
1248 | if (ddir_rw(acct_ddir(io_u))) | |
1249 | td->io_issues[acct_ddir(io_u)]++; | |
1250 | if (ddir_rw(io_u->ddir)) { | |
1251 | io_u_mark_depth(td, 1); | |
1252 | td->ts.total_io_u[io_u->ddir]++; | |
1253 | } | |
1254 | ||
2e63e96b PV |
1255 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1256 | td->o.do_verify && | |
1257 | td->o.verify != VERIFY_NONE && | |
1258 | !td->o.experimental_verify) | |
1259 | log_io_piece(td, io_u); | |
1260 | ||
62167762 JC |
1261 | ret = io_u_sync_complete(td, io_u, bytes_done); |
1262 | (void) ret; | |
1263 | } | |
1264 | ||
1265 | return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM]; | |
1266 | } | |
1267 | ||
2e1df07d JA |
1268 | /* |
1269 | * Entry point for the thread based jobs. The process based jobs end up | |
1270 | * here as well, after a little setup. | |
1271 | */ | |
1272 | static void *thread_main(void *data) | |
1273 | { | |
1274 | unsigned long long elapsed; | |
1275 | struct thread_data *td = data; | |
4896473e | 1276 | struct thread_options *o = &td->o; |
2e1df07d JA |
1277 | pthread_condattr_t attr; |
1278 | int clear_state; | |
28727df7 | 1279 | int ret; |
2e1df07d | 1280 | |
4896473e | 1281 | if (!o->use_thread) { |
2e1df07d JA |
1282 | setsid(); |
1283 | td->pid = getpid(); | |
1284 | } else | |
1285 | td->pid = gettid(); | |
1286 | ||
4896473e | 1287 | fio_local_clock_init(o->use_thread); |
5d879392 | 1288 | |
2e1df07d JA |
1289 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
1290 | ||
122c7725 JA |
1291 | if (is_backend) |
1292 | fio_server_send_start(td); | |
1293 | ||
2e1df07d JA |
1294 | INIT_FLIST_HEAD(&td->io_log_list); |
1295 | INIT_FLIST_HEAD(&td->io_hist_list); | |
1296 | INIT_FLIST_HEAD(&td->verify_list); | |
1297 | INIT_FLIST_HEAD(&td->trim_list); | |
1ae83d45 | 1298 | INIT_FLIST_HEAD(&td->next_rand_list); |
2e1df07d JA |
1299 | pthread_mutex_init(&td->io_u_lock, NULL); |
1300 | td->io_hist_tree = RB_ROOT; | |
1301 | ||
1302 | pthread_condattr_init(&attr); | |
1303 | pthread_cond_init(&td->verify_cond, &attr); | |
1304 | pthread_cond_init(&td->free_cond, &attr); | |
1305 | ||
1306 | td_set_runstate(td, TD_INITIALIZED); | |
1307 | dprint(FD_MUTEX, "up startup_mutex\n"); | |
1308 | fio_mutex_up(startup_mutex); | |
1309 | dprint(FD_MUTEX, "wait on td->mutex\n"); | |
1310 | fio_mutex_down(td->mutex); | |
1311 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); | |
1312 | ||
2e1df07d JA |
1313 | /* |
1314 | * A new gid requires privilege, so we need to do this before setting | |
1315 | * the uid. | |
1316 | */ | |
4896473e | 1317 | if (o->gid != -1U && setgid(o->gid)) { |
2e1df07d JA |
1318 | td_verror(td, errno, "setgid"); |
1319 | goto err; | |
1320 | } | |
4896473e | 1321 | if (o->uid != -1U && setuid(o->uid)) { |
2e1df07d JA |
1322 | td_verror(td, errno, "setuid"); |
1323 | goto err; | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * If we have a gettimeofday() thread, make sure we exclude that | |
1328 | * thread from this job | |
1329 | */ | |
4896473e JA |
1330 | if (o->gtod_cpu) |
1331 | fio_cpu_clear(&o->cpumask, o->gtod_cpu); | |
2e1df07d JA |
1332 | |
1333 | /* | |
1334 | * Set affinity first, in case it has an impact on the memory | |
1335 | * allocations. | |
1336 | */ | |
b2a9e649 | 1337 | if (fio_option_is_set(o, cpumask)) { |
c2acfbac | 1338 | if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { |
30cb4c65 | 1339 | ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); |
c2acfbac JA |
1340 | if (!ret) { |
1341 | log_err("fio: no CPUs set\n"); | |
1342 | log_err("fio: Try increasing number of available CPUs\n"); | |
1343 | td_verror(td, EINVAL, "cpus_split"); | |
1344 | goto err; | |
1345 | } | |
1346 | } | |
28727df7 JA |
1347 | ret = fio_setaffinity(td->pid, o->cpumask); |
1348 | if (ret == -1) { | |
4896473e JA |
1349 | td_verror(td, errno, "cpu_set_affinity"); |
1350 | goto err; | |
1351 | } | |
2e1df07d JA |
1352 | } |
1353 | ||
67bf9823 | 1354 | #ifdef CONFIG_LIBNUMA |
d0b937ed | 1355 | /* numa node setup */ |
b2a9e649 JA |
1356 | if (fio_option_is_set(o, numa_cpunodes) || |
1357 | fio_option_is_set(o, numa_memnodes)) { | |
43522848 | 1358 | struct bitmask *mask; |
d0b937ed YR |
1359 | |
1360 | if (numa_available() < 0) { | |
1361 | td_verror(td, errno, "Does not support NUMA API\n"); | |
1362 | goto err; | |
1363 | } | |
1364 | ||
b2a9e649 | 1365 | if (fio_option_is_set(o, numa_cpunodes)) { |
43522848 DG |
1366 | mask = numa_parse_nodestring(o->numa_cpunodes); |
1367 | ret = numa_run_on_node_mask(mask); | |
1368 | numa_free_nodemask(mask); | |
d0b937ed YR |
1369 | if (ret == -1) { |
1370 | td_verror(td, errno, \ | |
1371 | "numa_run_on_node_mask failed\n"); | |
1372 | goto err; | |
1373 | } | |
1374 | } | |
1375 | ||
b2a9e649 | 1376 | if (fio_option_is_set(o, numa_memnodes)) { |
43522848 DG |
1377 | mask = NULL; |
1378 | if (o->numa_memnodes) | |
1379 | mask = numa_parse_nodestring(o->numa_memnodes); | |
1380 | ||
4896473e | 1381 | switch (o->numa_mem_mode) { |
d0b937ed | 1382 | case MPOL_INTERLEAVE: |
43522848 | 1383 | numa_set_interleave_mask(mask); |
d0b937ed YR |
1384 | break; |
1385 | case MPOL_BIND: | |
43522848 | 1386 | numa_set_membind(mask); |
d0b937ed YR |
1387 | break; |
1388 | case MPOL_LOCAL: | |
1389 | numa_set_localalloc(); | |
1390 | break; | |
1391 | case MPOL_PREFERRED: | |
4896473e | 1392 | numa_set_preferred(o->numa_mem_prefer_node); |
d0b937ed YR |
1393 | break; |
1394 | case MPOL_DEFAULT: | |
1395 | default: | |
1396 | break; | |
1397 | } | |
1398 | ||
43522848 DG |
1399 | if (mask) |
1400 | numa_free_nodemask(mask); | |
1401 | ||
d0b937ed YR |
1402 | } |
1403 | } | |
1404 | #endif | |
1405 | ||
9a3f1100 JA |
1406 | if (fio_pin_memory(td)) |
1407 | goto err; | |
1408 | ||
2e1df07d JA |
1409 | /* |
1410 | * May alter parameters that init_io_u() will use, so we need to | |
1411 | * do this first. | |
1412 | */ | |
1413 | if (init_iolog(td)) | |
1414 | goto err; | |
1415 | ||
1416 | if (init_io_u(td)) | |
1417 | goto err; | |
1418 | ||
4896473e | 1419 | if (o->verify_async && verify_async_init(td)) |
2e1df07d JA |
1420 | goto err; |
1421 | ||
27d74836 JA |
1422 | if (fio_option_is_set(o, ioprio) || |
1423 | fio_option_is_set(o, ioprio_class)) { | |
28727df7 JA |
1424 | ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio); |
1425 | if (ret == -1) { | |
2e1df07d JA |
1426 | td_verror(td, errno, "ioprio_set"); |
1427 | goto err; | |
1428 | } | |
1429 | } | |
1430 | ||
4896473e | 1431 | if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
2e1df07d JA |
1432 | goto err; |
1433 | ||
649c10c5 | 1434 | errno = 0; |
4896473e | 1435 | if (nice(o->nice) == -1 && errno != 0) { |
2e1df07d JA |
1436 | td_verror(td, errno, "nice"); |
1437 | goto err; | |
1438 | } | |
1439 | ||
4896473e | 1440 | if (o->ioscheduler && switch_ioscheduler(td)) |
2e1df07d JA |
1441 | goto err; |
1442 | ||
4896473e | 1443 | if (!o->create_serialize && setup_files(td)) |
2e1df07d JA |
1444 | goto err; |
1445 | ||
1446 | if (td_io_init(td)) | |
1447 | goto err; | |
1448 | ||
1449 | if (init_random_map(td)) | |
1450 | goto err; | |
1451 | ||
ce486495 | 1452 | if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun")) |
4896473e | 1453 | goto err; |
2e1df07d | 1454 | |
4896473e | 1455 | if (o->pre_read) { |
2e1df07d JA |
1456 | if (pre_read_files(td) < 0) |
1457 | goto err; | |
1458 | } | |
1459 | ||
aee2ab67 JA |
1460 | if (td->flags & TD_F_COMPRESS_LOG) |
1461 | tp_init(&td->tp_data); | |
1462 | ||
dc5bfbb2 JA |
1463 | fio_verify_init(td); |
1464 | ||
2e1df07d | 1465 | fio_gettime(&td->epoch, NULL); |
44404c5a | 1466 | fio_getrusage(&td->ru_start); |
2e1df07d JA |
1467 | clear_state = 0; |
1468 | while (keep_running(td)) { | |
100f49f1 JA |
1469 | uint64_t verify_bytes; |
1470 | ||
2e1df07d JA |
1471 | fio_gettime(&td->start, NULL); |
1472 | memcpy(&td->bw_sample_time, &td->start, sizeof(td->start)); | |
1473 | memcpy(&td->iops_sample_time, &td->start, sizeof(td->start)); | |
1474 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); | |
1475 | ||
4896473e JA |
1476 | if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || |
1477 | o->ratemin[DDIR_TRIM]) { | |
6eaf09d6 | 1478 | memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, |
2e1df07d | 1479 | sizeof(td->bw_sample_time)); |
6eaf09d6 SL |
1480 | memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, |
1481 | sizeof(td->bw_sample_time)); | |
1482 | memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time, | |
2e1df07d JA |
1483 | sizeof(td->bw_sample_time)); |
1484 | } | |
1485 | ||
1486 | if (clear_state) | |
1487 | clear_io_state(td); | |
1488 | ||
1489 | prune_io_piece_log(td); | |
1490 | ||
62167762 JC |
1491 | if (td->o.verify_only && (td_write(td) || td_rw(td))) |
1492 | verify_bytes = do_dry_run(td); | |
1493 | else | |
1494 | verify_bytes = do_io(td); | |
2e1df07d JA |
1495 | |
1496 | clear_state = 1; | |
1497 | ||
e5437a07 | 1498 | fio_mutex_down(stat_mutex); |
2e1df07d | 1499 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
e5437a07 | 1500 | elapsed = mtime_since_now(&td->start); |
2e1df07d JA |
1501 | td->ts.runtime[DDIR_READ] += elapsed; |
1502 | } | |
1503 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
e5437a07 | 1504 | elapsed = mtime_since_now(&td->start); |
2e1df07d JA |
1505 | td->ts.runtime[DDIR_WRITE] += elapsed; |
1506 | } | |
6eaf09d6 | 1507 | if (td_trim(td) && td->io_bytes[DDIR_TRIM]) { |
e5437a07 | 1508 | elapsed = mtime_since_now(&td->start); |
6eaf09d6 SL |
1509 | td->ts.runtime[DDIR_TRIM] += elapsed; |
1510 | } | |
e5437a07 VT |
1511 | fio_gettime(&td->start, NULL); |
1512 | fio_mutex_up(stat_mutex); | |
2e1df07d JA |
1513 | |
1514 | if (td->error || td->terminate) | |
1515 | break; | |
1516 | ||
4896473e JA |
1517 | if (!o->do_verify || |
1518 | o->verify == VERIFY_NONE || | |
2e1df07d JA |
1519 | (td->io_ops->flags & FIO_UNIDIR)) |
1520 | continue; | |
1521 | ||
1522 | clear_io_state(td); | |
1523 | ||
1524 | fio_gettime(&td->start, NULL); | |
1525 | ||
100f49f1 | 1526 | do_verify(td, verify_bytes); |
2e1df07d | 1527 | |
e5437a07 VT |
1528 | fio_mutex_down(stat_mutex); |
1529 | td->ts.runtime[DDIR_READ] += mtime_since_now(&td->start); | |
1530 | fio_gettime(&td->start, NULL); | |
1531 | fio_mutex_up(stat_mutex); | |
2e1df07d JA |
1532 | |
1533 | if (td->error || td->terminate) | |
1534 | break; | |
1535 | } | |
1536 | ||
1537 | update_rusage_stat(td); | |
2e1df07d | 1538 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
6eaf09d6 SL |
1539 | td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ]; |
1540 | td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; | |
1541 | td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; | |
2e1df07d | 1542 | |
ca09be4b JA |
1543 | if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && |
1544 | (td->o.verify != VERIFY_NONE && td_write(td))) { | |
1545 | struct all_io_list *state; | |
1546 | size_t sz; | |
1547 | ||
1548 | state = get_all_io_list(td->thread_number, &sz); | |
1549 | if (state) { | |
1550 | __verify_save_state(state, "local"); | |
1551 | free(state); | |
1552 | } | |
1553 | } | |
1554 | ||
9a3f1100 JA |
1555 | fio_unpin_memory(td); |
1556 | ||
905e3d4f | 1557 | fio_writeout_logs(td); |
2e1df07d | 1558 | |
aee2ab67 JA |
1559 | if (td->flags & TD_F_COMPRESS_LOG) |
1560 | tp_exit(&td->tp_data); | |
1561 | ||
4896473e | 1562 | if (o->exec_postrun) |
ce486495 | 1563 | exec_string(o, o->exec_postrun, (const char *)"postrun"); |
2e1df07d JA |
1564 | |
1565 | if (exitall_on_terminate) | |
1566 | fio_terminate_threads(td->groupid); | |
1567 | ||
1568 | err: | |
1569 | if (td->error) | |
1570 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, | |
1571 | td->verror); | |
1572 | ||
4896473e | 1573 | if (o->verify_async) |
2e1df07d JA |
1574 | verify_async_exit(td); |
1575 | ||
1576 | close_and_free_files(td); | |
2e1df07d | 1577 | cleanup_io_u(td); |
32dbca2c | 1578 | close_ioengine(td); |
2e1df07d | 1579 | cgroup_shutdown(td, &cgroup_mnt); |
ca09be4b | 1580 | verify_free_state(td); |
2e1df07d | 1581 | |
b2a9e649 | 1582 | if (fio_option_is_set(o, cpumask)) { |
8a1db9a1 JA |
1583 | ret = fio_cpuset_exit(&o->cpumask); |
1584 | if (ret) | |
1585 | td_verror(td, ret, "fio_cpuset_exit"); | |
2e1df07d JA |
1586 | } |
1587 | ||
1588 | /* | |
1589 | * do this very late, it will log file closing as well | |
1590 | */ | |
4896473e | 1591 | if (o->write_iolog_file) |
2e1df07d JA |
1592 | write_iolog_close(td); |
1593 | ||
ea66e04f JA |
1594 | fio_mutex_remove(td->mutex); |
1595 | td->mutex = NULL; | |
1596 | ||
2e1df07d | 1597 | td_set_runstate(td, TD_EXITED); |
fda2cfac JA |
1598 | |
1599 | /* | |
1600 | * Do this last after setting our runstate to exited, so we | |
1601 | * know that the stat thread is signaled. | |
1602 | */ | |
1603 | check_update_rusage(td); | |
1604 | ||
e43606c2 | 1605 | return (void *) (uintptr_t) td->error; |
2e1df07d JA |
1606 | } |
1607 | ||
1608 | ||
1609 | /* | |
1610 | * We cannot pass the td data into a forked process, so attach the td and | |
1611 | * pass it to the thread worker. | |
1612 | */ | |
1613 | static int fork_main(int shmid, int offset) | |
1614 | { | |
1615 | struct thread_data *td; | |
1616 | void *data, *ret; | |
1617 | ||
c8931876 | 1618 | #if !defined(__hpux) && !defined(CONFIG_NO_SHM) |
2e1df07d JA |
1619 | data = shmat(shmid, NULL, 0); |
1620 | if (data == (void *) -1) { | |
1621 | int __err = errno; | |
1622 | ||
1623 | perror("shmat"); | |
1624 | return __err; | |
1625 | } | |
1626 | #else | |
1627 | /* | |
1628 | * HP-UX inherits shm mappings? | |
1629 | */ | |
1630 | data = threads; | |
1631 | #endif | |
1632 | ||
1633 | td = data + offset * sizeof(struct thread_data); | |
1634 | ret = thread_main(td); | |
1635 | shmdt(data); | |
e43606c2 | 1636 | return (int) (uintptr_t) ret; |
2e1df07d JA |
1637 | } |
1638 | ||
cba5460c JA |
1639 | static void dump_td_info(struct thread_data *td) |
1640 | { | |
1641 | log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to " | |
1642 | "be stuck. Doing forceful exit of this job.\n", td->o.name, | |
1643 | (unsigned long) time_since_now(&td->terminate_time)); | |
1644 | } | |
1645 | ||
2e1df07d JA |
1646 | /* |
1647 | * Run over the job map and reap the threads that have exited, if any. | |
1648 | */ | |
1649 | static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, | |
1650 | unsigned int *m_rate) | |
1651 | { | |
1652 | struct thread_data *td; | |
1653 | unsigned int cputhreads, realthreads, pending; | |
1654 | int i, status, ret; | |
1655 | ||
1656 | /* | |
1657 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1658 | */ | |
1659 | realthreads = pending = cputhreads = 0; | |
1660 | for_each_td(td, i) { | |
1661 | int flags = 0; | |
1662 | ||
1663 | /* | |
1664 | * ->io_ops is NULL for a thread that has closed its | |
1665 | * io engine | |
1666 | */ | |
1667 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) | |
1668 | cputhreads++; | |
1669 | else | |
1670 | realthreads++; | |
1671 | ||
1672 | if (!td->pid) { | |
1673 | pending++; | |
1674 | continue; | |
1675 | } | |
1676 | if (td->runstate == TD_REAPED) | |
1677 | continue; | |
1678 | if (td->o.use_thread) { | |
1679 | if (td->runstate == TD_EXITED) { | |
1680 | td_set_runstate(td, TD_REAPED); | |
1681 | goto reaped; | |
1682 | } | |
1683 | continue; | |
1684 | } | |
1685 | ||
1686 | flags = WNOHANG; | |
1687 | if (td->runstate == TD_EXITED) | |
1688 | flags = 0; | |
1689 | ||
1690 | /* | |
1691 | * check if someone quit or got killed in an unusual way | |
1692 | */ | |
1693 | ret = waitpid(td->pid, &status, flags); | |
1694 | if (ret < 0) { | |
1695 | if (errno == ECHILD) { | |
1696 | log_err("fio: pid=%d disappeared %d\n", | |
1697 | (int) td->pid, td->runstate); | |
a5e371a6 | 1698 | td->sig = ECHILD; |
2e1df07d JA |
1699 | td_set_runstate(td, TD_REAPED); |
1700 | goto reaped; | |
1701 | } | |
1702 | perror("waitpid"); | |
1703 | } else if (ret == td->pid) { | |
1704 | if (WIFSIGNALED(status)) { | |
1705 | int sig = WTERMSIG(status); | |
1706 | ||
36d80bc7 | 1707 | if (sig != SIGTERM && sig != SIGUSR2) |
2e1df07d JA |
1708 | log_err("fio: pid=%d, got signal=%d\n", |
1709 | (int) td->pid, sig); | |
a5e371a6 | 1710 | td->sig = sig; |
2e1df07d JA |
1711 | td_set_runstate(td, TD_REAPED); |
1712 | goto reaped; | |
1713 | } | |
1714 | if (WIFEXITED(status)) { | |
1715 | if (WEXITSTATUS(status) && !td->error) | |
1716 | td->error = WEXITSTATUS(status); | |
1717 | ||
1718 | td_set_runstate(td, TD_REAPED); | |
1719 | goto reaped; | |
1720 | } | |
1721 | } | |
1722 | ||
cba5460c JA |
1723 | /* |
1724 | * If the job is stuck, do a forceful timeout of it and | |
1725 | * move on. | |
1726 | */ | |
1727 | if (td->terminate && | |
1728 | time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { | |
1729 | dump_td_info(td); | |
1730 | td_set_runstate(td, TD_REAPED); | |
1731 | goto reaped; | |
1732 | } | |
1733 | ||
2e1df07d JA |
1734 | /* |
1735 | * thread is not dead, continue | |
1736 | */ | |
1737 | pending++; | |
1738 | continue; | |
1739 | reaped: | |
1740 | (*nr_running)--; | |
342f4be4 JA |
1741 | (*m_rate) -= ddir_rw_sum(td->o.ratemin); |
1742 | (*t_rate) -= ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
1743 | if (!td->pid) |
1744 | pending--; | |
1745 | ||
1746 | if (td->error) | |
1747 | exit_value++; | |
1748 | ||
1749 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
4a88752a | 1750 | profile_td_exit(td); |
2e1df07d JA |
1751 | } |
1752 | ||
1753 | if (*nr_running == cputhreads && !pending && realthreads) | |
1754 | fio_terminate_threads(TERMINATE_ALL); | |
1755 | } | |
1756 | ||
ca09be4b JA |
1757 | static int __check_trigger_file(void) |
1758 | { | |
1759 | struct stat sb; | |
1760 | ||
1761 | if (!trigger_file) | |
1762 | return 0; | |
1763 | ||
1764 | if (stat(trigger_file, &sb)) | |
1765 | return 0; | |
1766 | ||
1767 | if (unlink(trigger_file) < 0) | |
1768 | log_err("fio: failed to unlink %s: %s\n", trigger_file, | |
1769 | strerror(errno)); | |
1770 | ||
1771 | return 1; | |
1772 | } | |
1773 | ||
1774 | static int trigger_timedout(void) | |
1775 | { | |
1776 | if (trigger_timeout) | |
1777 | return time_since_genesis() >= trigger_timeout; | |
1778 | ||
1779 | return 0; | |
1780 | } | |
1781 | ||
1782 | void exec_trigger(const char *cmd) | |
1783 | { | |
1784 | int ret; | |
1785 | ||
1786 | if (!cmd) | |
1787 | return; | |
1788 | ||
1789 | ret = system(cmd); | |
1790 | if (ret == -1) | |
1791 | log_err("fio: failed executing %s trigger\n", cmd); | |
1792 | } | |
1793 | ||
1794 | void check_trigger_file(void) | |
1795 | { | |
1796 | if (__check_trigger_file() || trigger_timedout()) { | |
796fb3ce JA |
1797 | if (nr_clients) |
1798 | fio_clients_send_trigger(trigger_remote_cmd); | |
1799 | else { | |
ca09be4b JA |
1800 | verify_save_state(); |
1801 | fio_terminate_threads(TERMINATE_ALL); | |
1802 | exec_trigger(trigger_cmd); | |
1803 | } | |
1804 | } | |
1805 | } | |
1806 | ||
1807 | static int fio_verify_load_state(struct thread_data *td) | |
1808 | { | |
1809 | int ret; | |
1810 | ||
1811 | if (!td->o.verify_state) | |
1812 | return 0; | |
1813 | ||
1814 | if (is_backend) { | |
1815 | void *data; | |
1816 | ||
1817 | ret = fio_server_get_verify_state(td->o.name, | |
1818 | td->thread_number - 1, &data); | |
1819 | if (!ret) | |
1820 | verify_convert_assign_state(td, data); | |
1821 | } else | |
1822 | ret = verify_load_state(td, "local"); | |
1823 | ||
1824 | return ret; | |
1825 | } | |
1826 | ||
06464907 JA |
1827 | static void do_usleep(unsigned int usecs) |
1828 | { | |
1829 | check_for_running_stats(); | |
ca09be4b | 1830 | check_trigger_file(); |
06464907 JA |
1831 | usleep(usecs); |
1832 | } | |
1833 | ||
2e1df07d JA |
1834 | /* |
1835 | * Main function for kicking off and reaping jobs, as needed. | |
1836 | */ | |
1837 | static void run_threads(void) | |
1838 | { | |
1839 | struct thread_data *td; | |
2e1df07d | 1840 | unsigned int i, todo, nr_running, m_rate, t_rate, nr_started; |
0de5b26f | 1841 | uint64_t spent; |
2e1df07d | 1842 | |
2e1df07d JA |
1843 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1844 | return; | |
334185e9 | 1845 | |
f2a2ce0e | 1846 | fio_idle_prof_init(); |
2e1df07d JA |
1847 | |
1848 | set_sig_handlers(); | |
1849 | ||
3a5f6bde JA |
1850 | nr_thread = nr_process = 0; |
1851 | for_each_td(td, i) { | |
1852 | if (td->o.use_thread) | |
1853 | nr_thread++; | |
1854 | else | |
1855 | nr_process++; | |
1856 | } | |
1857 | ||
f3afa57e | 1858 | if (output_format == FIO_OUTPUT_NORMAL) { |
2e1df07d JA |
1859 | log_info("Starting "); |
1860 | if (nr_thread) | |
1861 | log_info("%d thread%s", nr_thread, | |
1862 | nr_thread > 1 ? "s" : ""); | |
1863 | if (nr_process) { | |
1864 | if (nr_thread) | |
1865 | log_info(" and "); | |
1866 | log_info("%d process%s", nr_process, | |
1867 | nr_process > 1 ? "es" : ""); | |
1868 | } | |
1869 | log_info("\n"); | |
e411c301 | 1870 | log_info_flush(); |
2e1df07d JA |
1871 | } |
1872 | ||
1873 | todo = thread_number; | |
1874 | nr_running = 0; | |
1875 | nr_started = 0; | |
1876 | m_rate = t_rate = 0; | |
1877 | ||
1878 | for_each_td(td, i) { | |
1879 | print_status_init(td->thread_number - 1); | |
1880 | ||
1881 | if (!td->o.create_serialize) | |
1882 | continue; | |
1883 | ||
ca09be4b JA |
1884 | if (fio_verify_load_state(td)) |
1885 | goto reap; | |
1886 | ||
2e1df07d JA |
1887 | /* |
1888 | * do file setup here so it happens sequentially, | |
1889 | * we don't want X number of threads getting their | |
1890 | * client data interspersed on disk | |
1891 | */ | |
1892 | if (setup_files(td)) { | |
ca09be4b | 1893 | reap: |
2e1df07d JA |
1894 | exit_value++; |
1895 | if (td->error) | |
1896 | log_err("fio: pid=%d, err=%d/%s\n", | |
1897 | (int) td->pid, td->error, td->verror); | |
1898 | td_set_runstate(td, TD_REAPED); | |
1899 | todo--; | |
1900 | } else { | |
1901 | struct fio_file *f; | |
1902 | unsigned int j; | |
1903 | ||
1904 | /* | |
1905 | * for sharing to work, each job must always open | |
1906 | * its own files. so close them, if we opened them | |
1907 | * for creation | |
1908 | */ | |
1909 | for_each_file(td, f, j) { | |
1910 | if (fio_file_open(f)) | |
1911 | td_io_close_file(td, f); | |
1912 | } | |
1913 | } | |
1914 | } | |
1915 | ||
f2a2ce0e HL |
1916 | /* start idle threads before io threads start to run */ |
1917 | fio_idle_prof_start(); | |
1918 | ||
2e1df07d JA |
1919 | set_genesis_time(); |
1920 | ||
1921 | while (todo) { | |
1922 | struct thread_data *map[REAL_MAX_JOBS]; | |
1923 | struct timeval this_start; | |
1924 | int this_jobs = 0, left; | |
1925 | ||
1926 | /* | |
1927 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1928 | */ | |
1929 | for_each_td(td, i) { | |
1930 | if (td->runstate != TD_NOT_CREATED) | |
1931 | continue; | |
1932 | ||
1933 | /* | |
1934 | * never got a chance to start, killed by other | |
1935 | * thread for some reason | |
1936 | */ | |
1937 | if (td->terminate) { | |
1938 | todo--; | |
1939 | continue; | |
1940 | } | |
1941 | ||
1942 | if (td->o.start_delay) { | |
0de5b26f | 1943 | spent = utime_since_genesis(); |
2e1df07d | 1944 | |
74454ce4 | 1945 | if (td->o.start_delay > spent) |
2e1df07d JA |
1946 | continue; |
1947 | } | |
1948 | ||
1949 | if (td->o.stonewall && (nr_started || nr_running)) { | |
1950 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1951 | td->o.name); | |
1952 | break; | |
1953 | } | |
1954 | ||
1955 | init_disk_util(td); | |
1956 | ||
c97f1ad6 JA |
1957 | td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED); |
1958 | td->update_rusage = 0; | |
1959 | ||
2e1df07d JA |
1960 | /* |
1961 | * Set state to created. Thread will transition | |
1962 | * to TD_INITIALIZED when it's done setting up. | |
1963 | */ | |
1964 | td_set_runstate(td, TD_CREATED); | |
1965 | map[this_jobs++] = td; | |
1966 | nr_started++; | |
1967 | ||
1968 | if (td->o.use_thread) { | |
1969 | int ret; | |
1970 | ||
1971 | dprint(FD_PROCESS, "will pthread_create\n"); | |
1972 | ret = pthread_create(&td->thread, NULL, | |
1973 | thread_main, td); | |
1974 | if (ret) { | |
1975 | log_err("pthread_create: %s\n", | |
1976 | strerror(ret)); | |
1977 | nr_started--; | |
1978 | break; | |
1979 | } | |
1980 | ret = pthread_detach(td->thread); | |
1981 | if (ret) | |
1982 | log_err("pthread_detach: %s", | |
1983 | strerror(ret)); | |
1984 | } else { | |
1985 | pid_t pid; | |
1986 | dprint(FD_PROCESS, "will fork\n"); | |
1987 | pid = fork(); | |
1988 | if (!pid) { | |
1989 | int ret = fork_main(shm_id, i); | |
1990 | ||
1991 | _exit(ret); | |
1992 | } else if (i == fio_debug_jobno) | |
1993 | *fio_debug_jobp = pid; | |
1994 | } | |
1995 | dprint(FD_MUTEX, "wait on startup_mutex\n"); | |
1996 | if (fio_mutex_down_timeout(startup_mutex, 10)) { | |
1997 | log_err("fio: job startup hung? exiting.\n"); | |
1998 | fio_terminate_threads(TERMINATE_ALL); | |
1999 | fio_abort = 1; | |
2000 | nr_started--; | |
2001 | break; | |
2002 | } | |
2003 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
2004 | } | |
2005 | ||
2006 | /* | |
2007 | * Wait for the started threads to transition to | |
2008 | * TD_INITIALIZED. | |
2009 | */ | |
2010 | fio_gettime(&this_start, NULL); | |
2011 | left = this_jobs; | |
2012 | while (left && !fio_abort) { | |
2013 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
2014 | break; | |
2015 | ||
06464907 | 2016 | do_usleep(100000); |
2e1df07d JA |
2017 | |
2018 | for (i = 0; i < this_jobs; i++) { | |
2019 | td = map[i]; | |
2020 | if (!td) | |
2021 | continue; | |
2022 | if (td->runstate == TD_INITIALIZED) { | |
2023 | map[i] = NULL; | |
2024 | left--; | |
2025 | } else if (td->runstate >= TD_EXITED) { | |
2026 | map[i] = NULL; | |
2027 | left--; | |
2028 | todo--; | |
2029 | nr_running++; /* work-around... */ | |
2030 | } | |
2031 | } | |
2032 | } | |
2033 | ||
2034 | if (left) { | |
4e87c37a JA |
2035 | log_err("fio: %d job%s failed to start\n", left, |
2036 | left > 1 ? "s" : ""); | |
2e1df07d JA |
2037 | for (i = 0; i < this_jobs; i++) { |
2038 | td = map[i]; | |
2039 | if (!td) | |
2040 | continue; | |
2041 | kill(td->pid, SIGTERM); | |
2042 | } | |
2043 | break; | |
2044 | } | |
2045 | ||
2046 | /* | |
2047 | * start created threads (TD_INITIALIZED -> TD_RUNNING). | |
2048 | */ | |
2049 | for_each_td(td, i) { | |
2050 | if (td->runstate != TD_INITIALIZED) | |
2051 | continue; | |
2052 | ||
2053 | if (in_ramp_time(td)) | |
2054 | td_set_runstate(td, TD_RAMP); | |
2055 | else | |
2056 | td_set_runstate(td, TD_RUNNING); | |
2057 | nr_running++; | |
2058 | nr_started--; | |
342f4be4 JA |
2059 | m_rate += ddir_rw_sum(td->o.ratemin); |
2060 | t_rate += ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
2061 | todo--; |
2062 | fio_mutex_up(td->mutex); | |
2063 | } | |
2064 | ||
2065 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2066 | ||
122c7725 | 2067 | if (todo) |
06464907 | 2068 | do_usleep(100000); |
2e1df07d JA |
2069 | } |
2070 | ||
2071 | while (nr_running) { | |
2072 | reap_threads(&nr_running, &t_rate, &m_rate); | |
06464907 | 2073 | do_usleep(10000); |
2e1df07d JA |
2074 | } |
2075 | ||
f2a2ce0e HL |
2076 | fio_idle_prof_stop(); |
2077 | ||
2e1df07d | 2078 | update_io_ticks(); |
2e1df07d JA |
2079 | } |
2080 | ||
5ddc6707 | 2081 | static void wait_for_helper_thread_exit(void) |
9ec7779f | 2082 | { |
8aab824f JA |
2083 | void *ret; |
2084 | ||
98b4b0a2 | 2085 | helper_exit = 1; |
5ddc6707 JA |
2086 | pthread_cond_signal(&helper_cond); |
2087 | pthread_join(helper_thread, &ret); | |
9ec7779f JA |
2088 | } |
2089 | ||
27357187 JA |
2090 | static void free_disk_util(void) |
2091 | { | |
27357187 | 2092 | disk_util_prune_entries(); |
8aab824f | 2093 | |
5ddc6707 | 2094 | pthread_cond_destroy(&helper_cond); |
27357187 JA |
2095 | } |
2096 | ||
5ddc6707 | 2097 | static void *helper_thread_main(void *data) |
2e1df07d | 2098 | { |
9ec7779f JA |
2099 | int ret = 0; |
2100 | ||
2e1df07d JA |
2101 | fio_mutex_up(startup_mutex); |
2102 | ||
8aab824f JA |
2103 | while (!ret) { |
2104 | uint64_t sec = DISK_UTIL_MSEC / 1000; | |
2105 | uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000; | |
2106 | struct timespec ts; | |
2107 | struct timeval tv; | |
2108 | ||
2109 | gettimeofday(&tv, NULL); | |
2110 | ts.tv_sec = tv.tv_sec + sec; | |
2111 | ts.tv_nsec = (tv.tv_usec * 1000) + nsec; | |
f893b76d | 2112 | |
823952af | 2113 | if (ts.tv_nsec >= 1000000000ULL) { |
8aab824f JA |
2114 | ts.tv_nsec -= 1000000000ULL; |
2115 | ts.tv_sec++; | |
2116 | } | |
2117 | ||
5ddc6707 | 2118 | pthread_cond_timedwait(&helper_cond, &helper_lock, &ts); |
8aab824f | 2119 | |
9ec7779f | 2120 | ret = update_io_ticks(); |
2e1df07d | 2121 | |
5ddc6707 JA |
2122 | if (helper_do_stat) { |
2123 | helper_do_stat = 0; | |
2124 | __show_running_run_stats(); | |
2125 | } | |
2126 | ||
2e1df07d JA |
2127 | if (!is_backend) |
2128 | print_thread_status(); | |
2129 | } | |
2130 | ||
2131 | return NULL; | |
2132 | } | |
2133 | ||
5ddc6707 | 2134 | static int create_helper_thread(void) |
2e1df07d JA |
2135 | { |
2136 | int ret; | |
2137 | ||
9ec7779f JA |
2138 | setup_disk_util(); |
2139 | ||
5ddc6707 JA |
2140 | pthread_cond_init(&helper_cond, NULL); |
2141 | pthread_mutex_init(&helper_lock, NULL); | |
8aab824f | 2142 | |
5ddc6707 | 2143 | ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL); |
2e1df07d | 2144 | if (ret) { |
5ddc6707 | 2145 | log_err("Can't create helper thread: %s\n", strerror(ret)); |
2e1df07d JA |
2146 | return 1; |
2147 | } | |
2148 | ||
2e1df07d JA |
2149 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
2150 | fio_mutex_down(startup_mutex); | |
2151 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
2152 | return 0; | |
2153 | } | |
2154 | ||
2e1df07d JA |
2155 | int fio_backend(void) |
2156 | { | |
2157 | struct thread_data *td; | |
2158 | int i; | |
2159 | ||
2160 | if (exec_profile) { | |
2161 | if (load_profile(exec_profile)) | |
2162 | return 1; | |
2163 | free(exec_profile); | |
2164 | exec_profile = NULL; | |
2165 | } | |
2166 | if (!thread_number) | |
2167 | return 0; | |
2168 | ||
2169 | if (write_bw_log) { | |
aee2ab67 JA |
2170 | struct log_params p = { |
2171 | .log_type = IO_LOG_TYPE_BW, | |
2172 | }; | |
2173 | ||
2174 | setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log"); | |
2175 | setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log"); | |
2176 | setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); | |
2e1df07d JA |
2177 | } |
2178 | ||
521da527 | 2179 | startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED); |
2e1df07d JA |
2180 | if (startup_mutex == NULL) |
2181 | return 1; | |
2e1df07d JA |
2182 | |
2183 | set_genesis_time(); | |
cef9175e | 2184 | stat_init(); |
5ddc6707 | 2185 | create_helper_thread(); |
2e1df07d JA |
2186 | |
2187 | cgroup_list = smalloc(sizeof(*cgroup_list)); | |
2188 | INIT_FLIST_HEAD(cgroup_list); | |
2189 | ||
2190 | run_threads(); | |
2191 | ||
5ddc6707 | 2192 | wait_for_helper_thread_exit(); |
8aab824f | 2193 | |
2e1df07d | 2194 | if (!fio_abort) { |
83f7b64e | 2195 | __show_run_stats(); |
2e1df07d | 2196 | if (write_bw_log) { |
cb7e0ace JA |
2197 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { |
2198 | struct io_log *log = agg_io_log[i]; | |
2199 | ||
aee2ab67 | 2200 | flush_log(log); |
518dac09 | 2201 | free_log(log); |
cb7e0ace | 2202 | } |
2e1df07d JA |
2203 | } |
2204 | } | |
2205 | ||
fda2cfac | 2206 | for_each_td(td, i) { |
2e1df07d | 2207 | fio_options_free(td); |
8049adc1 JA |
2208 | if (td->rusage_sem) { |
2209 | fio_mutex_remove(td->rusage_sem); | |
2210 | td->rusage_sem = NULL; | |
2211 | } | |
fda2cfac | 2212 | } |
2e1df07d | 2213 | |
a462baef | 2214 | free_disk_util(); |
2e1df07d JA |
2215 | cgroup_kill(cgroup_list); |
2216 | sfree(cgroup_list); | |
2217 | sfree(cgroup_mnt); | |
2218 | ||
2219 | fio_mutex_remove(startup_mutex); | |
cef9175e | 2220 | stat_exit(); |
2e1df07d JA |
2221 | return exit_value; |
2222 | } |