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 | |
fa07eaa6 | 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. |
2e1df07d JA |
22 | * |
23 | */ | |
24 | #include <unistd.h> | |
2e1df07d | 25 | #include <string.h> |
2e1df07d | 26 | #include <signal.h> |
2e1df07d | 27 | #include <assert.h> |
e43606c2 | 28 | #include <inttypes.h> |
2e1df07d JA |
29 | #include <sys/stat.h> |
30 | #include <sys/wait.h> | |
ff6bb260 | 31 | #include <math.h> |
2e1df07d JA |
32 | |
33 | #include "fio.h" | |
2e1df07d JA |
34 | #include "smalloc.h" |
35 | #include "verify.h" | |
2e1df07d JA |
36 | #include "diskutil.h" |
37 | #include "cgroup.h" | |
38 | #include "profile.h" | |
39 | #include "lib/rand.h" | |
f7690c4a | 40 | #include "lib/memalign.h" |
2e1df07d | 41 | #include "server.h" |
44404c5a | 42 | #include "lib/getrusage.h" |
f2a2ce0e | 43 | #include "idletime.h" |
002fe734 | 44 | #include "err.h" |
a9da8ab2 | 45 | #include "workqueue.h" |
e81ecca3 | 46 | #include "lib/mountcheck.h" |
40511301 | 47 | #include "rate-submit.h" |
a39fb9ea | 48 | #include "helper_thread.h" |
ae626d4e | 49 | #include "pshared.h" |
2e1df07d | 50 | |
971caeb1 | 51 | static struct fio_sem *startup_sem; |
2e1df07d JA |
52 | static struct flist_head *cgroup_list; |
53 | static char *cgroup_mnt; | |
54 | static int exit_value; | |
55 | static volatile int fio_abort; | |
3a5f6bde JA |
56 | static unsigned int nr_process = 0; |
57 | static unsigned int nr_thread = 0; | |
2e1df07d | 58 | |
6eaf09d6 | 59 | struct io_log *agg_io_log[DDIR_RWDIR_CNT]; |
2e1df07d | 60 | |
a3efc919 JA |
61 | int groupid = 0; |
62 | unsigned int thread_number = 0; | |
108fea77 | 63 | unsigned int stat_number = 0; |
a3efc919 JA |
64 | int shm_id = 0; |
65 | int temp_stall_ts; | |
66 | unsigned long done_secs = 0; | |
a47591e4 | 67 | |
2e1df07d JA |
68 | #define JOB_START_TIMEOUT (5 * 1000) |
69 | ||
70 | static void sig_int(int sig) | |
71 | { | |
72 | if (threads) { | |
73 | if (is_backend) | |
74 | fio_server_got_signal(sig); | |
75 | else { | |
76 | log_info("\nfio: terminating on signal %d\n", sig); | |
e411c301 | 77 | log_info_flush(); |
2e1df07d JA |
78 | exit_value = 128; |
79 | } | |
80 | ||
81 | fio_terminate_threads(TERMINATE_ALL); | |
82 | } | |
83 | } | |
84 | ||
d2235e56 | 85 | void sig_show_status(int sig) |
4c6d91e8 JA |
86 | { |
87 | show_running_run_stats(); | |
88 | } | |
89 | ||
2e1df07d JA |
90 | static void set_sig_handlers(void) |
91 | { | |
92 | struct sigaction act; | |
93 | ||
94 | memset(&act, 0, sizeof(act)); | |
95 | act.sa_handler = sig_int; | |
96 | act.sa_flags = SA_RESTART; | |
97 | sigaction(SIGINT, &act, NULL); | |
98 | ||
99 | memset(&act, 0, sizeof(act)); | |
100 | act.sa_handler = sig_int; | |
101 | act.sa_flags = SA_RESTART; | |
102 | sigaction(SIGTERM, &act, NULL); | |
103 | ||
2f694507 BC |
104 | /* Windows uses SIGBREAK as a quit signal from other applications */ |
105 | #ifdef WIN32 | |
106 | memset(&act, 0, sizeof(act)); | |
107 | act.sa_handler = sig_int; | |
108 | act.sa_flags = SA_RESTART; | |
109 | sigaction(SIGBREAK, &act, NULL); | |
110 | #endif | |
111 | ||
4c6d91e8 JA |
112 | memset(&act, 0, sizeof(act)); |
113 | act.sa_handler = sig_show_status; | |
114 | act.sa_flags = SA_RESTART; | |
115 | sigaction(SIGUSR1, &act, NULL); | |
116 | ||
2e1df07d JA |
117 | if (is_backend) { |
118 | memset(&act, 0, sizeof(act)); | |
119 | act.sa_handler = sig_int; | |
120 | act.sa_flags = SA_RESTART; | |
121 | sigaction(SIGPIPE, &act, NULL); | |
122 | } | |
123 | } | |
124 | ||
125 | /* | |
126 | * Check if we are above the minimum rate given. | |
127 | */ | |
8b6a404c | 128 | static bool __check_min_rate(struct thread_data *td, struct timespec *now, |
e39c0676 | 129 | enum fio_ddir ddir) |
2e1df07d JA |
130 | { |
131 | unsigned long long bytes = 0; | |
132 | unsigned long iops = 0; | |
133 | unsigned long spent; | |
134 | unsigned long rate; | |
135 | unsigned int ratemin = 0; | |
136 | unsigned int rate_iops = 0; | |
137 | unsigned int rate_iops_min = 0; | |
138 | ||
139 | assert(ddir_rw(ddir)); | |
140 | ||
141 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) | |
e39c0676 | 142 | return false; |
2e1df07d JA |
143 | |
144 | /* | |
145 | * allow a 2 second settle period in the beginning | |
146 | */ | |
147 | if (mtime_since(&td->start, now) < 2000) | |
e39c0676 | 148 | return false; |
2e1df07d JA |
149 | |
150 | iops += td->this_io_blocks[ddir]; | |
151 | bytes += td->this_io_bytes[ddir]; | |
152 | ratemin += td->o.ratemin[ddir]; | |
153 | rate_iops += td->o.rate_iops[ddir]; | |
154 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
155 | ||
156 | /* | |
157 | * if rate blocks is set, sample is running | |
158 | */ | |
159 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { | |
160 | spent = mtime_since(&td->lastrate[ddir], now); | |
161 | if (spent < td->o.ratecycle) | |
e39c0676 | 162 | return false; |
2e1df07d | 163 | |
20e37e4a | 164 | if (td->o.rate[ddir] || td->o.ratemin[ddir]) { |
2e1df07d JA |
165 | /* |
166 | * check bandwidth specified rate | |
167 | */ | |
168 | if (bytes < td->rate_bytes[ddir]) { | |
420b104a RE |
169 | log_err("%s: rate_min=%uB/s not met, only transferred %lluB\n", |
170 | td->o.name, ratemin, bytes); | |
e39c0676 | 171 | return true; |
2e1df07d | 172 | } else { |
49cba9b3 JA |
173 | if (spent) |
174 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; | |
175 | else | |
176 | rate = 0; | |
177 | ||
2e1df07d JA |
178 | if (rate < ratemin || |
179 | bytes < td->rate_bytes[ddir]) { | |
420b104a RE |
180 | log_err("%s: rate_min=%uB/s not met, got %luB/s\n", |
181 | td->o.name, ratemin, rate); | |
e39c0676 | 182 | return true; |
2e1df07d JA |
183 | } |
184 | } | |
185 | } else { | |
186 | /* | |
187 | * checks iops specified rate | |
188 | */ | |
189 | if (iops < rate_iops) { | |
420b104a RE |
190 | log_err("%s: rate_iops_min=%u not met, only performed %lu IOs\n", |
191 | td->o.name, rate_iops, iops); | |
e39c0676 | 192 | return true; |
2e1df07d | 193 | } else { |
4c707a3b JA |
194 | if (spent) |
195 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; | |
196 | else | |
197 | rate = 0; | |
198 | ||
2e1df07d JA |
199 | if (rate < rate_iops_min || |
200 | iops < td->rate_blocks[ddir]) { | |
420b104a RE |
201 | log_err("%s: rate_iops_min=%u not met, got %lu IOPS\n", |
202 | td->o.name, rate_iops_min, rate); | |
e39c0676 | 203 | return true; |
2e1df07d JA |
204 | } |
205 | } | |
206 | } | |
207 | } | |
208 | ||
209 | td->rate_bytes[ddir] = bytes; | |
210 | td->rate_blocks[ddir] = iops; | |
211 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
e39c0676 | 212 | return false; |
2e1df07d JA |
213 | } |
214 | ||
8b6a404c | 215 | static bool check_min_rate(struct thread_data *td, struct timespec *now) |
2e1df07d | 216 | { |
e39c0676 | 217 | bool ret = false; |
2e1df07d | 218 | |
55312f9f | 219 | if (td->bytes_done[DDIR_READ]) |
6eaf09d6 | 220 | ret |= __check_min_rate(td, now, DDIR_READ); |
55312f9f | 221 | if (td->bytes_done[DDIR_WRITE]) |
6eaf09d6 | 222 | ret |= __check_min_rate(td, now, DDIR_WRITE); |
55312f9f | 223 | if (td->bytes_done[DDIR_TRIM]) |
6eaf09d6 | 224 | ret |= __check_min_rate(td, now, DDIR_TRIM); |
2e1df07d JA |
225 | |
226 | return ret; | |
227 | } | |
228 | ||
229 | /* | |
230 | * When job exits, we can cancel the in-flight IO if we are using async | |
231 | * io. Attempt to do so. | |
232 | */ | |
233 | static void cleanup_pending_aio(struct thread_data *td) | |
234 | { | |
2e1df07d JA |
235 | int r; |
236 | ||
237 | /* | |
238 | * get immediately available events, if any | |
239 | */ | |
55312f9f | 240 | r = io_u_queued_complete(td, 0); |
2e1df07d JA |
241 | if (r < 0) |
242 | return; | |
243 | ||
244 | /* | |
245 | * now cancel remaining active events | |
246 | */ | |
247 | if (td->io_ops->cancel) { | |
2ae0b204 JA |
248 | struct io_u *io_u; |
249 | int i; | |
2e1df07d | 250 | |
2ae0b204 JA |
251 | io_u_qiter(&td->io_u_all, io_u, i) { |
252 | if (io_u->flags & IO_U_F_FLIGHT) { | |
2e1df07d JA |
253 | r = td->io_ops->cancel(td, io_u); |
254 | if (!r) | |
255 | put_io_u(td, io_u); | |
256 | } | |
257 | } | |
258 | } | |
259 | ||
260 | if (td->cur_depth) | |
55312f9f | 261 | r = io_u_queued_complete(td, td->cur_depth); |
2e1df07d JA |
262 | } |
263 | ||
264 | /* | |
265 | * Helper to handle the final sync of a file. Works just like the normal | |
266 | * io path, just does everything sync. | |
267 | */ | |
e39c0676 | 268 | static bool fio_io_sync(struct thread_data *td, struct fio_file *f) |
2e1df07d JA |
269 | { |
270 | struct io_u *io_u = __get_io_u(td); | |
271 | int ret; | |
272 | ||
273 | if (!io_u) | |
e39c0676 | 274 | return true; |
2e1df07d JA |
275 | |
276 | io_u->ddir = DDIR_SYNC; | |
277 | io_u->file = f; | |
278 | ||
279 | if (td_io_prep(td, io_u)) { | |
280 | put_io_u(td, io_u); | |
e39c0676 | 281 | return true; |
2e1df07d JA |
282 | } |
283 | ||
284 | requeue: | |
285 | ret = td_io_queue(td, io_u); | |
286 | if (ret < 0) { | |
287 | td_verror(td, io_u->error, "td_io_queue"); | |
288 | put_io_u(td, io_u); | |
e39c0676 | 289 | return true; |
2e1df07d | 290 | } else if (ret == FIO_Q_QUEUED) { |
a80cb54b | 291 | td_io_commit(td); |
55312f9f | 292 | if (io_u_queued_complete(td, 1) < 0) |
e39c0676 | 293 | return true; |
2e1df07d JA |
294 | } else if (ret == FIO_Q_COMPLETED) { |
295 | if (io_u->error) { | |
296 | td_verror(td, io_u->error, "td_io_queue"); | |
e39c0676 | 297 | return true; |
2e1df07d JA |
298 | } |
299 | ||
55312f9f | 300 | if (io_u_sync_complete(td, io_u) < 0) |
e39c0676 | 301 | return true; |
2e1df07d | 302 | } else if (ret == FIO_Q_BUSY) { |
a80cb54b | 303 | td_io_commit(td); |
2e1df07d JA |
304 | goto requeue; |
305 | } | |
306 | ||
e39c0676 | 307 | return false; |
2e1df07d | 308 | } |
a3efc919 | 309 | |
61ee0f86 JA |
310 | static int fio_file_fsync(struct thread_data *td, struct fio_file *f) |
311 | { | |
312 | int ret; | |
313 | ||
314 | if (fio_file_open(f)) | |
315 | return fio_io_sync(td, f); | |
316 | ||
317 | if (td_io_open_file(td, f)) | |
318 | return 1; | |
319 | ||
320 | ret = fio_io_sync(td, f); | |
321 | td_io_close_file(td, f); | |
322 | return ret; | |
323 | } | |
324 | ||
8b6a404c | 325 | static inline void __update_ts_cache(struct thread_data *td) |
2e1df07d | 326 | { |
8b6a404c | 327 | fio_gettime(&td->ts_cache, NULL); |
2e1df07d JA |
328 | } |
329 | ||
8b6a404c | 330 | static inline void update_ts_cache(struct thread_data *td) |
2e1df07d | 331 | { |
8b6a404c VF |
332 | if ((++td->ts_cache_nr & td->ts_cache_mask) == td->ts_cache_mask) |
333 | __update_ts_cache(td); | |
2e1df07d JA |
334 | } |
335 | ||
8b6a404c | 336 | static inline bool runtime_exceeded(struct thread_data *td, struct timespec *t) |
2e1df07d JA |
337 | { |
338 | if (in_ramp_time(td)) | |
e39c0676 | 339 | return false; |
2e1df07d | 340 | if (!td->o.timeout) |
e39c0676 | 341 | return false; |
0de5b26f | 342 | if (utime_since(&td->epoch, t) >= td->o.timeout) |
e39c0676 | 343 | return true; |
2e1df07d | 344 | |
e39c0676 | 345 | return false; |
2e1df07d JA |
346 | } |
347 | ||
95603b74 BF |
348 | /* |
349 | * We need to update the runtime consistently in ms, but keep a running | |
350 | * tally of the current elapsed time in microseconds for sub millisecond | |
351 | * updates. | |
352 | */ | |
353 | static inline void update_runtime(struct thread_data *td, | |
354 | unsigned long long *elapsed_us, | |
355 | const enum fio_ddir ddir) | |
356 | { | |
d5c8ea29 JA |
357 | if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only) |
358 | return; | |
359 | ||
95603b74 BF |
360 | td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000; |
361 | elapsed_us[ddir] += utime_since_now(&td->start); | |
362 | td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000; | |
363 | } | |
364 | ||
e39c0676 JA |
365 | static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir, |
366 | int *retptr) | |
2e1df07d JA |
367 | { |
368 | int ret = *retptr; | |
369 | ||
370 | if (ret < 0 || td->error) { | |
8b28bd41 DM |
371 | int err = td->error; |
372 | enum error_type_bit eb; | |
2e1df07d JA |
373 | |
374 | if (ret < 0) | |
375 | err = -ret; | |
2e1df07d | 376 | |
8b28bd41 DM |
377 | eb = td_error_type(ddir, err); |
378 | if (!(td->o.continue_on_error & (1 << eb))) | |
e39c0676 | 379 | return true; |
2e1df07d | 380 | |
8b28bd41 | 381 | if (td_non_fatal_error(td, eb, err)) { |
2e1df07d JA |
382 | /* |
383 | * Continue with the I/Os in case of | |
384 | * a non fatal error. | |
385 | */ | |
386 | update_error_count(td, err); | |
387 | td_clear_error(td); | |
388 | *retptr = 0; | |
e39c0676 | 389 | return false; |
2e1df07d JA |
390 | } else if (td->o.fill_device && err == ENOSPC) { |
391 | /* | |
392 | * We expect to hit this error if | |
393 | * fill_device option is set. | |
394 | */ | |
395 | td_clear_error(td); | |
ebea2133 | 396 | fio_mark_td_terminate(td); |
e39c0676 | 397 | return true; |
2e1df07d JA |
398 | } else { |
399 | /* | |
400 | * Stop the I/O in case of a fatal | |
401 | * error. | |
402 | */ | |
403 | update_error_count(td, err); | |
e39c0676 | 404 | return true; |
2e1df07d JA |
405 | } |
406 | } | |
407 | ||
e39c0676 | 408 | return false; |
2e1df07d JA |
409 | } |
410 | ||
c97f1ad6 JA |
411 | static void check_update_rusage(struct thread_data *td) |
412 | { | |
413 | if (td->update_rusage) { | |
414 | td->update_rusage = 0; | |
415 | update_rusage_stat(td); | |
971caeb1 | 416 | fio_sem_up(td->rusage_sem); |
c97f1ad6 JA |
417 | } |
418 | } | |
419 | ||
8b6a404c | 420 | static int wait_for_completions(struct thread_data *td, struct timespec *time) |
69fea81e JA |
421 | { |
422 | const int full = queue_full(td); | |
423 | int min_evts = 0; | |
424 | int ret; | |
425 | ||
6be06c46 JA |
426 | if (td->flags & TD_F_REGROW_LOGS) |
427 | return io_u_quiesce(td); | |
1fed2080 | 428 | |
69fea81e JA |
429 | /* |
430 | * if the queue is full, we MUST reap at least 1 event | |
431 | */ | |
82407585 | 432 | min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth); |
4b157ac6 | 433 | if ((full && !min_evts) || !td->o.iodepth_batch_complete_min) |
69fea81e JA |
434 | min_evts = 1; |
435 | ||
66b6c5ef | 436 | if (time && (__should_check_rate(td, DDIR_READ) || |
69fea81e JA |
437 | __should_check_rate(td, DDIR_WRITE) || |
438 | __should_check_rate(td, DDIR_TRIM))) | |
439 | fio_gettime(time, NULL); | |
440 | ||
441 | do { | |
55312f9f | 442 | ret = io_u_queued_complete(td, min_evts); |
69fea81e JA |
443 | if (ret < 0) |
444 | break; | |
445 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
446 | ||
447 | return ret; | |
448 | } | |
449 | ||
e9d512d8 JA |
450 | int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, |
451 | enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify, | |
8b6a404c | 452 | struct timespec *comp_time) |
e9d512d8 | 453 | { |
e9d512d8 JA |
454 | switch (*ret) { |
455 | case FIO_Q_COMPLETED: | |
456 | if (io_u->error) { | |
457 | *ret = -io_u->error; | |
458 | clear_io_u(td, io_u); | |
459 | } else if (io_u->resid) { | |
460 | int bytes = io_u->xfer_buflen - io_u->resid; | |
461 | struct fio_file *f = io_u->file; | |
462 | ||
463 | if (bytes_issued) | |
464 | *bytes_issued += bytes; | |
465 | ||
466 | if (!from_verify) | |
467 | trim_io_piece(td, io_u); | |
468 | ||
469 | /* | |
470 | * zero read, fail | |
471 | */ | |
472 | if (!bytes) { | |
473 | if (!from_verify) | |
474 | unlog_io_piece(td, io_u); | |
475 | td_verror(td, EIO, "full resid"); | |
476 | put_io_u(td, io_u); | |
477 | break; | |
478 | } | |
479 | ||
480 | io_u->xfer_buflen = io_u->resid; | |
481 | io_u->xfer_buf += bytes; | |
482 | io_u->offset += bytes; | |
483 | ||
484 | if (ddir_rw(io_u->ddir)) | |
485 | td->ts.short_io_u[io_u->ddir]++; | |
486 | ||
e9d512d8 JA |
487 | if (io_u->offset == f->real_file_size) |
488 | goto sync_done; | |
489 | ||
490 | requeue_io_u(td, &io_u); | |
491 | } else { | |
492 | sync_done: | |
493 | if (comp_time && (__should_check_rate(td, DDIR_READ) || | |
494 | __should_check_rate(td, DDIR_WRITE) || | |
495 | __should_check_rate(td, DDIR_TRIM))) | |
496 | fio_gettime(comp_time, NULL); | |
497 | ||
498 | *ret = io_u_sync_complete(td, io_u); | |
499 | if (*ret < 0) | |
500 | break; | |
501 | } | |
a0e3e5a6 | 502 | |
cf2c8d52 JA |
503 | if (td->flags & TD_F_REGROW_LOGS) |
504 | regrow_logs(td); | |
505 | ||
a0e3e5a6 EW |
506 | /* |
507 | * when doing I/O (not when verifying), | |
508 | * check for any errors that are to be ignored | |
509 | */ | |
510 | if (!from_verify) | |
511 | break; | |
512 | ||
e9d512d8 JA |
513 | return 0; |
514 | case FIO_Q_QUEUED: | |
515 | /* | |
516 | * if the engine doesn't have a commit hook, | |
517 | * the io_u is really queued. if it does have such | |
518 | * a hook, it has to call io_u_queued() itself. | |
519 | */ | |
520 | if (td->io_ops->commit == NULL) | |
521 | io_u_queued(td, io_u); | |
522 | if (bytes_issued) | |
523 | *bytes_issued += io_u->xfer_buflen; | |
524 | break; | |
525 | case FIO_Q_BUSY: | |
526 | if (!from_verify) | |
527 | unlog_io_piece(td, io_u); | |
528 | requeue_io_u(td, &io_u); | |
a80cb54b | 529 | td_io_commit(td); |
e9d512d8 JA |
530 | break; |
531 | default: | |
b0775325 | 532 | assert(*ret < 0); |
e9d512d8 JA |
533 | td_verror(td, -(*ret), "td_io_queue"); |
534 | break; | |
535 | } | |
536 | ||
537 | if (break_on_this_error(td, ddir, ret)) | |
538 | return 1; | |
539 | ||
540 | return 0; | |
541 | } | |
542 | ||
e39c0676 | 543 | static inline bool io_in_polling(struct thread_data *td) |
82407585 RP |
544 | { |
545 | return !td->o.iodepth_batch_complete_min && | |
546 | !td->o.iodepth_batch_complete_max; | |
547 | } | |
39c1c323 | 548 | /* |
549 | * Unlinks files from thread data fio_file structure | |
550 | */ | |
551 | static int unlink_all_files(struct thread_data *td) | |
552 | { | |
553 | struct fio_file *f; | |
554 | unsigned int i; | |
555 | int ret = 0; | |
556 | ||
557 | for_each_file(td, f, i) { | |
558 | if (f->filetype != FIO_TYPE_FILE) | |
559 | continue; | |
560 | ret = td_io_unlink_file(td, f); | |
561 | if (ret) | |
562 | break; | |
563 | } | |
564 | ||
565 | if (ret) | |
566 | td_verror(td, ret, "unlink_all_files"); | |
567 | ||
568 | return ret; | |
569 | } | |
82407585 | 570 | |
997b5680 SW |
571 | /* |
572 | * Check if io_u will overlap an in-flight IO in the queue | |
573 | */ | |
574 | static bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) | |
575 | { | |
576 | bool overlap; | |
577 | struct io_u *check_io_u; | |
578 | unsigned long long x1, x2, y1, y2; | |
579 | int i; | |
580 | ||
581 | x1 = io_u->offset; | |
582 | x2 = io_u->offset + io_u->buflen; | |
583 | overlap = false; | |
584 | io_u_qiter(q, check_io_u, i) { | |
585 | if (check_io_u->flags & IO_U_F_FLIGHT) { | |
586 | y1 = check_io_u->offset; | |
587 | y2 = check_io_u->offset + check_io_u->buflen; | |
588 | ||
589 | if (x1 < y2 && y1 < x2) { | |
590 | overlap = true; | |
591 | dprint(FD_IO, "in-flight overlap: %llu/%lu, %llu/%lu\n", | |
592 | x1, io_u->buflen, | |
593 | y1, check_io_u->buflen); | |
594 | break; | |
595 | } | |
596 | } | |
597 | } | |
598 | ||
599 | return overlap; | |
600 | } | |
601 | ||
08b2b386 JA |
602 | static int io_u_submit(struct thread_data *td, struct io_u *io_u) |
603 | { | |
604 | /* | |
605 | * Check for overlap if the user asked us to, and we have | |
606 | * at least one IO in flight besides this one. | |
607 | */ | |
608 | if (td->o.serialize_overlap && td->cur_depth > 1 && | |
609 | in_flight_overlap(&td->io_u_all, io_u)) | |
610 | return FIO_Q_BUSY; | |
611 | ||
612 | return td_io_queue(td, io_u); | |
613 | } | |
614 | ||
2e1df07d JA |
615 | /* |
616 | * The main verify engine. Runs over the writes we previously submitted, | |
617 | * reads the blocks back in, and checks the crc/md5 of the data. | |
618 | */ | |
100f49f1 | 619 | static void do_verify(struct thread_data *td, uint64_t verify_bytes) |
2e1df07d JA |
620 | { |
621 | struct fio_file *f; | |
622 | struct io_u *io_u; | |
623 | int ret, min_events; | |
624 | unsigned int i; | |
625 | ||
626 | dprint(FD_VERIFY, "starting loop\n"); | |
627 | ||
628 | /* | |
629 | * sync io first and invalidate cache, to make sure we really | |
630 | * read from disk. | |
631 | */ | |
632 | for_each_file(td, f, i) { | |
633 | if (!fio_file_open(f)) | |
634 | continue; | |
635 | if (fio_io_sync(td, f)) | |
636 | break; | |
637 | if (file_invalidate_cache(td, f)) | |
638 | break; | |
639 | } | |
640 | ||
c97f1ad6 JA |
641 | check_update_rusage(td); |
642 | ||
2e1df07d JA |
643 | if (td->error) |
644 | return; | |
645 | ||
8f380de2 CJ |
646 | /* |
647 | * verify_state needs to be reset before verification | |
648 | * proceeds so that expected random seeds match actual | |
649 | * random seeds in headers. The main loop will reset | |
650 | * all random number generators if randrepeat is set. | |
651 | */ | |
652 | if (!td->o.rand_repeatable) | |
653 | td_fill_verify_state_seed(td); | |
654 | ||
2e1df07d JA |
655 | td_set_runstate(td, TD_VERIFYING); |
656 | ||
657 | io_u = NULL; | |
658 | while (!td->terminate) { | |
fbccf46c | 659 | enum fio_ddir ddir; |
e9d512d8 | 660 | int full; |
2e1df07d | 661 | |
8b6a404c | 662 | update_ts_cache(td); |
c97f1ad6 | 663 | check_update_rusage(td); |
2e1df07d | 664 | |
8b6a404c VF |
665 | if (runtime_exceeded(td, &td->ts_cache)) { |
666 | __update_ts_cache(td); | |
667 | if (runtime_exceeded(td, &td->ts_cache)) { | |
ebea2133 | 668 | fio_mark_td_terminate(td); |
2e1df07d JA |
669 | break; |
670 | } | |
671 | } | |
672 | ||
9e684a49 DE |
673 | if (flow_threshold_exceeded(td)) |
674 | continue; | |
675 | ||
44cbc6da JA |
676 | if (!td->o.experimental_verify) { |
677 | io_u = __get_io_u(td); | |
678 | if (!io_u) | |
679 | break; | |
2e1df07d | 680 | |
44cbc6da JA |
681 | if (get_next_verify(td, io_u)) { |
682 | put_io_u(td, io_u); | |
683 | break; | |
684 | } | |
2e1df07d | 685 | |
44cbc6da JA |
686 | if (td_io_prep(td, io_u)) { |
687 | put_io_u(td, io_u); | |
688 | break; | |
689 | } | |
690 | } else { | |
55312f9f | 691 | if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes) |
100f49f1 JA |
692 | break; |
693 | ||
bcd5abfa | 694 | while ((io_u = get_io_u(td)) != NULL) { |
059b61f2 | 695 | if (IS_ERR_OR_NULL(io_u)) { |
002fe734 JA |
696 | io_u = NULL; |
697 | ret = FIO_Q_BUSY; | |
698 | goto reap; | |
699 | } | |
700 | ||
bcd5abfa JA |
701 | /* |
702 | * We are only interested in the places where | |
703 | * we wrote or trimmed IOs. Turn those into | |
704 | * reads for verification purposes. | |
705 | */ | |
706 | if (io_u->ddir == DDIR_READ) { | |
707 | /* | |
708 | * Pretend we issued it for rwmix | |
709 | * accounting | |
710 | */ | |
711 | td->io_issues[DDIR_READ]++; | |
712 | put_io_u(td, io_u); | |
713 | continue; | |
714 | } else if (io_u->ddir == DDIR_TRIM) { | |
715 | io_u->ddir = DDIR_READ; | |
1651e431 | 716 | io_u_set(td, io_u, IO_U_F_TRIMMED); |
bcd5abfa JA |
717 | break; |
718 | } else if (io_u->ddir == DDIR_WRITE) { | |
719 | io_u->ddir = DDIR_READ; | |
5c5c33c1 | 720 | populate_verify_io_u(td, io_u); |
bcd5abfa JA |
721 | break; |
722 | } else { | |
723 | put_io_u(td, io_u); | |
724 | continue; | |
725 | } | |
726 | } | |
44cbc6da | 727 | |
bcd5abfa | 728 | if (!io_u) |
44cbc6da | 729 | break; |
2e1df07d JA |
730 | } |
731 | ||
ca09be4b JA |
732 | if (verify_state_should_stop(td, io_u)) { |
733 | put_io_u(td, io_u); | |
734 | break; | |
735 | } | |
736 | ||
2e1df07d JA |
737 | if (td->o.verify_async) |
738 | io_u->end_io = verify_io_u_async; | |
739 | else | |
740 | io_u->end_io = verify_io_u; | |
741 | ||
fbccf46c | 742 | ddir = io_u->ddir; |
10adc4a7 GG |
743 | if (!td->o.disable_slat) |
744 | fio_gettime(&io_u->start_time, NULL); | |
fbccf46c | 745 | |
08b2b386 | 746 | ret = io_u_submit(td, io_u); |
2e1df07d | 747 | |
e9d512d8 | 748 | if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL)) |
2e1df07d JA |
749 | break; |
750 | ||
751 | /* | |
752 | * if we can queue more, do so. but check if there are | |
753 | * completed io_u's first. Note that we can get BUSY even | |
754 | * without IO queued, if the system is resource starved. | |
755 | */ | |
002fe734 | 756 | reap: |
2e1df07d | 757 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
82407585 | 758 | if (full || io_in_polling(td)) |
55312f9f | 759 | ret = wait_for_completions(td, NULL); |
2e1df07d | 760 | |
2e1df07d JA |
761 | if (ret < 0) |
762 | break; | |
763 | } | |
764 | ||
c97f1ad6 JA |
765 | check_update_rusage(td); |
766 | ||
2e1df07d JA |
767 | if (!td->error) { |
768 | min_events = td->cur_depth; | |
769 | ||
770 | if (min_events) | |
55312f9f | 771 | ret = io_u_queued_complete(td, min_events); |
2e1df07d JA |
772 | } else |
773 | cleanup_pending_aio(td); | |
774 | ||
775 | td_set_runstate(td, TD_RUNNING); | |
776 | ||
777 | dprint(FD_VERIFY, "exiting loop\n"); | |
778 | } | |
779 | ||
e39c0676 | 780 | static bool exceeds_number_ios(struct thread_data *td) |
3939fe85 JA |
781 | { |
782 | unsigned long long number_ios; | |
783 | ||
784 | if (!td->o.number_ios) | |
e39c0676 | 785 | return false; |
3939fe85 | 786 | |
cf8a46a7 | 787 | number_ios = ddir_rw_sum(td->io_blocks); |
3939fe85 JA |
788 | number_ios += td->io_u_queued + td->io_u_in_flight; |
789 | ||
cf8a46a7 | 790 | return number_ios >= (td->o.number_ios * td->loops); |
3939fe85 JA |
791 | } |
792 | ||
926098e0 | 793 | static bool io_bytes_exceeded(struct thread_data *td, uint64_t *this_bytes) |
f7078f7b | 794 | { |
77731b29 | 795 | unsigned long long bytes, limit; |
f7078f7b JA |
796 | |
797 | if (td_rw(td)) | |
926098e0 | 798 | bytes = this_bytes[DDIR_READ] + this_bytes[DDIR_WRITE]; |
f7078f7b | 799 | else if (td_write(td)) |
926098e0 | 800 | bytes = this_bytes[DDIR_WRITE]; |
6eaf09d6 | 801 | else if (td_read(td)) |
926098e0 | 802 | bytes = this_bytes[DDIR_READ]; |
f7078f7b | 803 | else |
926098e0 | 804 | bytes = this_bytes[DDIR_TRIM]; |
f7078f7b | 805 | |
5be9bf09 TK |
806 | if (td->o.io_size) |
807 | limit = td->o.io_size; | |
77731b29 JA |
808 | else |
809 | limit = td->o.size; | |
810 | ||
cf8a46a7 | 811 | limit *= td->loops; |
77731b29 | 812 | return bytes >= limit || exceeds_number_ios(td); |
f7078f7b JA |
813 | } |
814 | ||
926098e0 | 815 | static bool io_issue_bytes_exceeded(struct thread_data *td) |
e28dd2cf | 816 | { |
926098e0 JA |
817 | return io_bytes_exceeded(td, td->io_issue_bytes); |
818 | } | |
e28dd2cf | 819 | |
926098e0 JA |
820 | static bool io_complete_bytes_exceeded(struct thread_data *td) |
821 | { | |
822 | return io_bytes_exceeded(td, td->this_io_bytes); | |
e28dd2cf JE |
823 | } |
824 | ||
50a8ce86 D |
825 | /* |
826 | * used to calculate the next io time for rate control | |
827 | * | |
828 | */ | |
829 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) | |
830 | { | |
1a9bf814 | 831 | uint64_t bps = td->rate_bps[ddir]; |
50a8ce86 D |
832 | |
833 | assert(!(td->flags & TD_F_CHILD)); | |
ff6bb260 | 834 | |
6de65959 | 835 | if (td->o.rate_process == RATE_PROCESS_POISSON) { |
1a9bf814 JA |
836 | uint64_t val, iops; |
837 | ||
ff6bb260 | 838 | iops = bps / td->o.bs[ddir]; |
a77d139b | 839 | val = (int64_t) (1000000 / iops) * |
6c8611c7 | 840 | -logf(__rand_0_1(&td->poisson_state[ddir])); |
a77d139b | 841 | if (val) { |
98e51d7c JA |
842 | dprint(FD_RATE, "poisson rate iops=%llu, ddir=%d\n", |
843 | (unsigned long long) 1000000 / val, | |
844 | ddir); | |
a77d139b | 845 | } |
6c8611c7 JA |
846 | td->last_usec[ddir] += val; |
847 | return td->last_usec[ddir]; | |
ff6bb260 | 848 | } else if (bps) { |
1a9bf814 JA |
849 | uint64_t bytes = td->rate_io_issue_bytes[ddir]; |
850 | uint64_t secs = bytes / bps; | |
851 | uint64_t remainder = bytes % bps; | |
852 | ||
50a8ce86 | 853 | return remainder * 1000000 / bps + secs * 1000000; |
e7b24047 JA |
854 | } |
855 | ||
856 | return 0; | |
50a8ce86 D |
857 | } |
858 | ||
1a9bf814 JA |
859 | static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir) |
860 | { | |
861 | unsigned long long b; | |
862 | uint64_t total; | |
863 | int left; | |
864 | ||
865 | b = ddir_rw_sum(td->io_blocks); | |
866 | if (b % td->o.thinktime_blocks) | |
867 | return; | |
868 | ||
869 | io_u_quiesce(td); | |
870 | ||
871 | total = 0; | |
872 | if (td->o.thinktime_spin) | |
873 | total = usec_spin(td->o.thinktime_spin); | |
874 | ||
875 | left = td->o.thinktime - total; | |
876 | if (left) | |
877 | total += usec_sleep(td, left); | |
878 | ||
879 | /* | |
880 | * If we're ignoring thinktime for the rate, add the number of bytes | |
1aa39b0c JA |
881 | * we would have done while sleeping, minus one block to ensure we |
882 | * start issuing immediately after the sleep. | |
1a9bf814 | 883 | */ |
1aa39b0c | 884 | if (total && td->rate_bps[ddir] && td->o.rate_ign_think) { |
f2aedb17 | 885 | uint64_t missed = (td->rate_bps[ddir] * total) / 1000000ULL; |
62b858f6 JA |
886 | uint64_t bs = td->o.min_bs[ddir]; |
887 | uint64_t usperop = bs * 1000000ULL / td->rate_bps[ddir]; | |
f2aedb17 JA |
888 | uint64_t over; |
889 | ||
62b858f6 JA |
890 | if (usperop <= total) |
891 | over = bs; | |
f2aedb17 | 892 | else |
62b858f6 | 893 | over = (usperop - total) / usperop * -bs; |
f2aedb17 JA |
894 | |
895 | td->rate_io_issue_bytes[ddir] += (missed - over); | |
1aa39b0c | 896 | } |
1a9bf814 JA |
897 | } |
898 | ||
2e1df07d JA |
899 | /* |
900 | * Main IO worker function. It retrieves io_u's to process and queues | |
901 | * and reaps them, checking for rate and errors along the way. | |
100f49f1 JA |
902 | * |
903 | * Returns number of bytes written and trimmed. | |
2e1df07d | 904 | */ |
95b03be0 | 905 | static void do_io(struct thread_data *td, uint64_t *bytes_done) |
2e1df07d JA |
906 | { |
907 | unsigned int i; | |
908 | int ret = 0; | |
c2703bf3 | 909 | uint64_t total_bytes, bytes_issued = 0; |
70c68076 | 910 | |
95b03be0 JA |
911 | for (i = 0; i < DDIR_RWDIR_CNT; i++) |
912 | bytes_done[i] = td->bytes_done[i]; | |
2e1df07d JA |
913 | |
914 | if (in_ramp_time(td)) | |
915 | td_set_runstate(td, TD_RAMP); | |
916 | else | |
917 | td_set_runstate(td, TD_RUNNING); | |
918 | ||
3e260a46 JA |
919 | lat_target_init(td); |
920 | ||
1e564979 JE |
921 | total_bytes = td->o.size; |
922 | /* | |
5be9bf09 | 923 | * Allow random overwrite workloads to write up to io_size |
1e564979 JE |
924 | * before starting verification phase as 'size' doesn't apply. |
925 | */ | |
926 | if (td_write(td) && td_random(td) && td->o.norandommap) | |
5be9bf09 | 927 | total_bytes = max(total_bytes, (uint64_t) td->o.io_size); |
78a6469c JA |
928 | /* |
929 | * If verify_backlog is enabled, we'll run the verify in this | |
930 | * handler as well. For that case, we may need up to twice the | |
931 | * amount of bytes. | |
932 | */ | |
78a6469c JA |
933 | if (td->o.verify != VERIFY_NONE && |
934 | (td_write(td) && td->o.verify_backlog)) | |
c2703bf3 JA |
935 | total_bytes += td->o.size; |
936 | ||
82a90686 | 937 | /* In trimwrite mode, each byte is trimmed and then written, so |
0e4dd95c | 938 | * allow total_bytes to be twice as big */ |
82a90686 | 939 | if (td_trimwrite(td)) |
0e4dd95c DE |
940 | total_bytes += td->total_io_size; |
941 | ||
f7078f7b | 942 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
e28dd2cf | 943 | (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) || |
c04e4661 | 944 | td->o.time_based) { |
8b6a404c | 945 | struct timespec comp_time; |
2e1df07d | 946 | struct io_u *io_u; |
e9d512d8 | 947 | int full; |
2e1df07d JA |
948 | enum fio_ddir ddir; |
949 | ||
c97f1ad6 JA |
950 | check_update_rusage(td); |
951 | ||
7d7803fa | 952 | if (td->terminate || td->done) |
2e1df07d JA |
953 | break; |
954 | ||
8b6a404c | 955 | update_ts_cache(td); |
2e1df07d | 956 | |
8b6a404c VF |
957 | if (runtime_exceeded(td, &td->ts_cache)) { |
958 | __update_ts_cache(td); | |
959 | if (runtime_exceeded(td, &td->ts_cache)) { | |
ebea2133 | 960 | fio_mark_td_terminate(td); |
2e1df07d JA |
961 | break; |
962 | } | |
963 | } | |
964 | ||
9e684a49 DE |
965 | if (flow_threshold_exceeded(td)) |
966 | continue; | |
967 | ||
f1a32461 JA |
968 | /* |
969 | * Break if we exceeded the bytes. The exception is time | |
970 | * based runs, but we still need to break out of the loop | |
971 | * for those to run verification, if enabled. | |
972 | */ | |
973 | if (bytes_issued >= total_bytes && | |
974 | (!td->o.time_based || | |
975 | (td->o.time_based && td->o.verify != VERIFY_NONE))) | |
20876c53 JC |
976 | break; |
977 | ||
2e1df07d | 978 | io_u = get_io_u(td); |
002fe734 JA |
979 | if (IS_ERR_OR_NULL(io_u)) { |
980 | int err = PTR_ERR(io_u); | |
981 | ||
982 | io_u = NULL; | |
1a9bf814 | 983 | ddir = DDIR_INVAL; |
002fe734 JA |
984 | if (err == -EBUSY) { |
985 | ret = FIO_Q_BUSY; | |
986 | goto reap; | |
987 | } | |
3e260a46 JA |
988 | if (td->o.latency_target) |
989 | goto reap; | |
2e1df07d | 990 | break; |
3e260a46 | 991 | } |
2e1df07d | 992 | |
5c5c33c1 BVA |
993 | if (io_u->ddir == DDIR_WRITE && td->flags & TD_F_DO_VERIFY) |
994 | populate_verify_io_u(td, io_u); | |
995 | ||
2e1df07d JA |
996 | ddir = io_u->ddir; |
997 | ||
998 | /* | |
82af2a7c JA |
999 | * Add verification end_io handler if: |
1000 | * - Asked to verify (!td_rw(td)) | |
1001 | * - Or the io_u is from our verify list (mixed write/ver) | |
2e1df07d JA |
1002 | */ |
1003 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && | |
82af2a7c | 1004 | ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { |
c4b6117b PV |
1005 | |
1006 | if (!td->o.verify_pattern_bytes) { | |
d6b72507 | 1007 | io_u->rand_seed = __rand(&td->verify_state); |
c4b6117b | 1008 | if (sizeof(int) != sizeof(long *)) |
d6b72507 | 1009 | io_u->rand_seed *= __rand(&td->verify_state); |
c4b6117b PV |
1010 | } |
1011 | ||
ca09be4b JA |
1012 | if (verify_state_should_stop(td, io_u)) { |
1013 | put_io_u(td, io_u); | |
1014 | break; | |
1015 | } | |
1016 | ||
2e1df07d JA |
1017 | if (td->o.verify_async) |
1018 | io_u->end_io = verify_io_u_async; | |
1019 | else | |
1020 | io_u->end_io = verify_io_u; | |
1021 | td_set_runstate(td, TD_VERIFYING); | |
1022 | } else if (in_ramp_time(td)) | |
1023 | td_set_runstate(td, TD_RAMP); | |
1024 | else | |
1025 | td_set_runstate(td, TD_RUNNING); | |
1026 | ||
9a50c5c5 | 1027 | /* |
f9401285 JA |
1028 | * Always log IO before it's issued, so we know the specific |
1029 | * order of it. The logged unit will track when the IO has | |
1030 | * completed. | |
9a50c5c5 | 1031 | */ |
c4b6117b PV |
1032 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1033 | td->o.do_verify && | |
1034 | td->o.verify != VERIFY_NONE && | |
f9401285 | 1035 | !td->o.experimental_verify) |
c4b6117b PV |
1036 | log_io_piece(td, io_u); |
1037 | ||
a9da8ab2 | 1038 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
0c5df5f9 JA |
1039 | const unsigned long blen = io_u->xfer_buflen; |
1040 | const enum fio_ddir ddir = acct_ddir(io_u); | |
1041 | ||
a9da8ab2 JA |
1042 | if (td->error) |
1043 | break; | |
0c5df5f9 | 1044 | |
26de50cf JA |
1045 | workqueue_enqueue(&td->io_wq, &io_u->work); |
1046 | ret = FIO_Q_QUEUED; | |
50a8ce86 | 1047 | |
26de50cf | 1048 | if (ddir_rw(ddir)) { |
0c5df5f9 JA |
1049 | td->io_issues[ddir]++; |
1050 | td->io_issue_bytes[ddir] += blen; | |
1051 | td->rate_io_issue_bytes[ddir] += blen; | |
1052 | } | |
1053 | ||
50a8ce86 D |
1054 | if (should_check_rate(td)) |
1055 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
1056 | ||
a9da8ab2 | 1057 | } else { |
08b2b386 | 1058 | ret = io_u_submit(td, io_u); |
2e1df07d | 1059 | |
50a8ce86 D |
1060 | if (should_check_rate(td)) |
1061 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
1062 | ||
fd727d9d | 1063 | if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time)) |
a9da8ab2 | 1064 | break; |
2e1df07d | 1065 | |
a9da8ab2 JA |
1066 | /* |
1067 | * See if we need to complete some commands. Note that | |
1068 | * we can get BUSY even without IO queued, if the | |
1069 | * system is resource starved. | |
1070 | */ | |
3e260a46 | 1071 | reap: |
a9da8ab2 JA |
1072 | full = queue_full(td) || |
1073 | (ret == FIO_Q_BUSY && td->cur_depth); | |
82407585 | 1074 | if (full || io_in_polling(td)) |
a9da8ab2 JA |
1075 | ret = wait_for_completions(td, &comp_time); |
1076 | } | |
2e1df07d JA |
1077 | if (ret < 0) |
1078 | break; | |
55312f9f | 1079 | if (!ddir_rw_sum(td->bytes_done) && |
9b87f09b | 1080 | !td_ioengine_flagged(td, FIO_NOIO)) |
2e1df07d JA |
1081 | continue; |
1082 | ||
55312f9f JA |
1083 | if (!in_ramp_time(td) && should_check_rate(td)) { |
1084 | if (check_min_rate(td, &comp_time)) { | |
f9cafb12 | 1085 | if (exitall_on_terminate || td->o.exitall_error) |
2e1df07d JA |
1086 | fio_terminate_threads(td->groupid); |
1087 | td_verror(td, EIO, "check_min_rate"); | |
1088 | break; | |
1089 | } | |
1090 | } | |
3e260a46 JA |
1091 | if (!in_ramp_time(td) && td->o.latency_target) |
1092 | lat_target_check(td); | |
e155cb64 | 1093 | |
1a9bf814 JA |
1094 | if (ddir_rw(ddir) && td->o.thinktime) |
1095 | handle_thinktime(td, ddir); | |
2e1df07d JA |
1096 | } |
1097 | ||
c97f1ad6 JA |
1098 | check_update_rusage(td); |
1099 | ||
2e1df07d | 1100 | if (td->trim_entries) |
4e0a8fa2 | 1101 | log_err("fio: %lu trim entries leaked?\n", td->trim_entries); |
2e1df07d JA |
1102 | |
1103 | if (td->o.fill_device && td->error == ENOSPC) { | |
1104 | td->error = 0; | |
ebea2133 | 1105 | fio_mark_td_terminate(td); |
2e1df07d JA |
1106 | } |
1107 | if (!td->error) { | |
1108 | struct fio_file *f; | |
1109 | ||
a9da8ab2 JA |
1110 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
1111 | workqueue_flush(&td->io_wq); | |
1112 | i = 0; | |
1113 | } else | |
1114 | i = td->cur_depth; | |
1115 | ||
2e1df07d | 1116 | if (i) { |
55312f9f | 1117 | ret = io_u_queued_complete(td, i); |
2e1df07d JA |
1118 | if (td->o.fill_device && td->error == ENOSPC) |
1119 | td->error = 0; | |
1120 | } | |
1121 | ||
1122 | if (should_fsync(td) && td->o.end_fsync) { | |
1123 | td_set_runstate(td, TD_FSYNCING); | |
1124 | ||
1125 | for_each_file(td, f, i) { | |
61ee0f86 | 1126 | if (!fio_file_fsync(td, f)) |
2e1df07d | 1127 | continue; |
61ee0f86 JA |
1128 | |
1129 | log_err("fio: end_fsync failed for file %s\n", | |
1130 | f->file_name); | |
2e1df07d JA |
1131 | } |
1132 | } | |
1133 | } else | |
1134 | cleanup_pending_aio(td); | |
1135 | ||
1136 | /* | |
1137 | * stop job if we failed doing any IO | |
1138 | */ | |
342f4be4 | 1139 | if (!ddir_rw_sum(td->this_io_bytes)) |
2e1df07d | 1140 | td->done = 1; |
100f49f1 | 1141 | |
95b03be0 JA |
1142 | for (i = 0; i < DDIR_RWDIR_CNT; i++) |
1143 | bytes_done[i] = td->bytes_done[i] - bytes_done[i]; | |
2e1df07d JA |
1144 | } |
1145 | ||
94a6e1bb JA |
1146 | static void free_file_completion_logging(struct thread_data *td) |
1147 | { | |
1148 | struct fio_file *f; | |
1149 | unsigned int i; | |
1150 | ||
1151 | for_each_file(td, f, i) { | |
1152 | if (!f->last_write_comp) | |
1153 | break; | |
1154 | sfree(f->last_write_comp); | |
1155 | } | |
1156 | } | |
1157 | ||
1158 | static int init_file_completion_logging(struct thread_data *td, | |
1159 | unsigned int depth) | |
1160 | { | |
1161 | struct fio_file *f; | |
1162 | unsigned int i; | |
1163 | ||
1164 | if (td->o.verify == VERIFY_NONE || !td->o.verify_state_save) | |
1165 | return 0; | |
1166 | ||
1167 | for_each_file(td, f, i) { | |
1168 | f->last_write_comp = scalloc(depth, sizeof(uint64_t)); | |
1169 | if (!f->last_write_comp) | |
1170 | goto cleanup; | |
1171 | } | |
1172 | ||
1173 | return 0; | |
1174 | ||
1175 | cleanup: | |
1176 | free_file_completion_logging(td); | |
1177 | log_err("fio: failed to alloc write comp data\n"); | |
1178 | return 1; | |
1179 | } | |
1180 | ||
2e1df07d JA |
1181 | static void cleanup_io_u(struct thread_data *td) |
1182 | { | |
2e1df07d JA |
1183 | struct io_u *io_u; |
1184 | ||
2ae0b204 | 1185 | while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) { |
c73ed246 JA |
1186 | |
1187 | if (td->io_ops->io_u_free) | |
1188 | td->io_ops->io_u_free(td, io_u); | |
1189 | ||
2e1df07d JA |
1190 | fio_memfree(io_u, sizeof(*io_u)); |
1191 | } | |
1192 | ||
1193 | free_io_mem(td); | |
2ae0b204 JA |
1194 | |
1195 | io_u_rexit(&td->io_u_requeues); | |
1196 | io_u_qexit(&td->io_u_freelist); | |
1197 | io_u_qexit(&td->io_u_all); | |
ca09be4b | 1198 | |
94a6e1bb | 1199 | free_file_completion_logging(td); |
2e1df07d JA |
1200 | } |
1201 | ||
1202 | static int init_io_u(struct thread_data *td) | |
1203 | { | |
1204 | struct io_u *io_u; | |
9c42684e | 1205 | unsigned int max_bs, min_write; |
2e1df07d | 1206 | int cl_align, i, max_units; |
2ae0b204 | 1207 | int data_xfer = 1, err; |
2e1df07d JA |
1208 | char *p; |
1209 | ||
1210 | max_units = td->o.iodepth; | |
74f4b020 | 1211 | max_bs = td_max_bs(td); |
9c42684e | 1212 | min_write = td->o.min_bs[DDIR_WRITE]; |
2e1df07d JA |
1213 | td->orig_buffer_size = (unsigned long long) max_bs |
1214 | * (unsigned long long) max_units; | |
1215 | ||
9b87f09b | 1216 | if (td_ioengine_flagged(td, FIO_NOIO) || !(td_read(td) || td_write(td))) |
59d8d0f5 JA |
1217 | data_xfer = 0; |
1218 | ||
2ae0b204 | 1219 | err = 0; |
34851ad5 JA |
1220 | err += !io_u_rinit(&td->io_u_requeues, td->o.iodepth); |
1221 | err += !io_u_qinit(&td->io_u_freelist, td->o.iodepth); | |
1222 | err += !io_u_qinit(&td->io_u_all, td->o.iodepth); | |
2ae0b204 JA |
1223 | |
1224 | if (err) { | |
1225 | log_err("fio: failed setting up IO queues\n"); | |
1226 | return 1; | |
1227 | } | |
1228 | ||
fd8a09b8 | 1229 | /* |
1230 | * if we may later need to do address alignment, then add any | |
1231 | * possible adjustment here so that we don't cause a buffer | |
1232 | * overflow later. this adjustment may be too much if we get | |
1233 | * lucky and the allocator gives us an aligned address. | |
1234 | */ | |
d01612f3 | 1235 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
9b87f09b | 1236 | td_ioengine_flagged(td, FIO_RAWIO)) |
fd8a09b8 | 1237 | td->orig_buffer_size += page_mask + td->o.mem_align; |
1238 | ||
2e1df07d JA |
1239 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { |
1240 | unsigned long bs; | |
1241 | ||
1242 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; | |
1243 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
1244 | } | |
1245 | ||
1246 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { | |
1247 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
1248 | return 1; | |
1249 | } | |
1250 | ||
59d8d0f5 | 1251 | if (data_xfer && allocate_io_mem(td)) |
2e1df07d JA |
1252 | return 1; |
1253 | ||
d01612f3 | 1254 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
9b87f09b | 1255 | td_ioengine_flagged(td, FIO_RAWIO)) |
248c9436 | 1256 | p = PTR_ALIGN(td->orig_buffer, page_mask) + td->o.mem_align; |
2e1df07d JA |
1257 | else |
1258 | p = td->orig_buffer; | |
1259 | ||
1260 | cl_align = os_cache_line_size(); | |
1261 | ||
1262 | for (i = 0; i < max_units; i++) { | |
1263 | void *ptr; | |
1264 | ||
1265 | if (td->terminate) | |
1266 | return 1; | |
1267 | ||
1268 | ptr = fio_memalign(cl_align, sizeof(*io_u)); | |
1269 | if (!ptr) { | |
1270 | log_err("fio: unable to allocate aligned memory\n"); | |
1271 | break; | |
1272 | } | |
1273 | ||
1274 | io_u = ptr; | |
1275 | memset(io_u, 0, sizeof(*io_u)); | |
2ae0b204 | 1276 | INIT_FLIST_HEAD(&io_u->verify_list); |
2e1df07d JA |
1277 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
1278 | ||
59d8d0f5 | 1279 | if (data_xfer) { |
2e1df07d JA |
1280 | io_u->buf = p; |
1281 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); | |
1282 | ||
1283 | if (td_write(td)) | |
9c42684e | 1284 | io_u_fill_buffer(td, io_u, min_write, max_bs); |
2e1df07d JA |
1285 | if (td_write(td) && td->o.verify_pattern_bytes) { |
1286 | /* | |
1287 | * Fill the buffer with the pattern if we are | |
1288 | * going to be doing writes. | |
1289 | */ | |
ce35b1ec | 1290 | fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
2e1df07d JA |
1291 | } |
1292 | } | |
1293 | ||
1294 | io_u->index = i; | |
1295 | io_u->flags = IO_U_F_FREE; | |
2ae0b204 JA |
1296 | io_u_qpush(&td->io_u_freelist, io_u); |
1297 | ||
1298 | /* | |
1299 | * io_u never leaves this stack, used for iteration of all | |
1300 | * io_u buffers. | |
1301 | */ | |
1302 | io_u_qpush(&td->io_u_all, io_u); | |
c73ed246 JA |
1303 | |
1304 | if (td->io_ops->io_u_init) { | |
1305 | int ret = td->io_ops->io_u_init(td, io_u); | |
1306 | ||
1307 | if (ret) { | |
1308 | log_err("fio: failed to init engine data: %d\n", ret); | |
1309 | return 1; | |
1310 | } | |
1311 | } | |
1312 | ||
2e1df07d JA |
1313 | p += max_bs; |
1314 | } | |
1315 | ||
94a6e1bb JA |
1316 | if (init_file_completion_logging(td, max_units)) |
1317 | return 1; | |
ca09be4b | 1318 | |
2e1df07d JA |
1319 | return 0; |
1320 | } | |
1321 | ||
bd4d9bdc TK |
1322 | /* |
1323 | * This function is Linux specific. | |
1324 | * FIO_HAVE_IOSCHED_SWITCH enabled currently means it's Linux. | |
1325 | */ | |
2e1df07d JA |
1326 | static int switch_ioscheduler(struct thread_data *td) |
1327 | { | |
db815fb3 | 1328 | #ifdef FIO_HAVE_IOSCHED_SWITCH |
178ef27a | 1329 | char tmp[256], tmp2[128], *p; |
2e1df07d JA |
1330 | FILE *f; |
1331 | int ret; | |
1332 | ||
9b87f09b | 1333 | if (td_ioengine_flagged(td, FIO_DISKLESSIO)) |
2e1df07d JA |
1334 | return 0; |
1335 | ||
bd4d9bdc TK |
1336 | assert(td->files && td->files[0]); |
1337 | sprintf(tmp, "%s/queue/scheduler", td->files[0]->du->sysfs_root); | |
2e1df07d JA |
1338 | |
1339 | f = fopen(tmp, "r+"); | |
1340 | if (!f) { | |
1341 | if (errno == ENOENT) { | |
1342 | log_err("fio: os or kernel doesn't support IO scheduler" | |
1343 | " switching\n"); | |
1344 | return 0; | |
1345 | } | |
1346 | td_verror(td, errno, "fopen iosched"); | |
1347 | return 1; | |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * Set io scheduler. | |
1352 | */ | |
1353 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); | |
1354 | if (ferror(f) || ret != 1) { | |
1355 | td_verror(td, errno, "fwrite"); | |
1356 | fclose(f); | |
1357 | return 1; | |
1358 | } | |
1359 | ||
1360 | rewind(f); | |
1361 | ||
1362 | /* | |
1363 | * Read back and check that the selected scheduler is now the default. | |
1364 | */ | |
0e23d456 | 1365 | ret = fread(tmp, 1, sizeof(tmp) - 1, f); |
2e1df07d JA |
1366 | if (ferror(f) || ret < 0) { |
1367 | td_verror(td, errno, "fread"); | |
1368 | fclose(f); | |
1369 | return 1; | |
1370 | } | |
0e23d456 | 1371 | tmp[ret] = '\0'; |
b44b9e45 | 1372 | /* |
178ef27a BVA |
1373 | * either a list of io schedulers or "none\n" is expected. Strip the |
1374 | * trailing newline. | |
b44b9e45 | 1375 | */ |
178ef27a BVA |
1376 | p = tmp; |
1377 | strsep(&p, "\n"); | |
49c6f33d | 1378 | |
67f28a8c TK |
1379 | /* |
1380 | * Write to "none" entry doesn't fail, so check the result here. | |
1381 | */ | |
1382 | if (!strcmp(tmp, "none")) { | |
1383 | log_err("fio: io scheduler is not tunable\n"); | |
1384 | fclose(f); | |
1385 | return 0; | |
1386 | } | |
2e1df07d JA |
1387 | |
1388 | sprintf(tmp2, "[%s]", td->o.ioscheduler); | |
1389 | if (!strstr(tmp, tmp2)) { | |
1390 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); | |
1391 | td_verror(td, EINVAL, "iosched_switch"); | |
1392 | fclose(f); | |
1393 | return 1; | |
1394 | } | |
1395 | ||
1396 | fclose(f); | |
1397 | return 0; | |
db815fb3 TK |
1398 | #else |
1399 | return 0; | |
1400 | #endif | |
2e1df07d JA |
1401 | } |
1402 | ||
e39c0676 | 1403 | static bool keep_running(struct thread_data *td) |
2e1df07d | 1404 | { |
77731b29 JA |
1405 | unsigned long long limit; |
1406 | ||
2e1df07d | 1407 | if (td->done) |
e39c0676 | 1408 | return false; |
98413be7 SW |
1409 | if (td->terminate) |
1410 | return false; | |
2e1df07d | 1411 | if (td->o.time_based) |
e39c0676 | 1412 | return true; |
2e1df07d JA |
1413 | if (td->o.loops) { |
1414 | td->o.loops--; | |
e39c0676 | 1415 | return true; |
2e1df07d | 1416 | } |
3939fe85 | 1417 | if (exceeds_number_ios(td)) |
e39c0676 | 1418 | return false; |
26251d8d | 1419 | |
5be9bf09 TK |
1420 | if (td->o.io_size) |
1421 | limit = td->o.io_size; | |
77731b29 JA |
1422 | else |
1423 | limit = td->o.size; | |
1424 | ||
1425 | if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) { | |
5bd5f71a JA |
1426 | uint64_t diff; |
1427 | ||
1428 | /* | |
589d5d4f | 1429 | * If the difference is less than the maximum IO size, we |
5bd5f71a JA |
1430 | * are done. |
1431 | */ | |
77731b29 | 1432 | diff = limit - ddir_rw_sum(td->io_bytes); |
74f4b020 | 1433 | if (diff < td_max_bs(td)) |
e39c0676 | 1434 | return false; |
5bd5f71a | 1435 | |
5be9bf09 | 1436 | if (fio_files_done(td) && !td->o.io_size) |
e39c0676 | 1437 | return false; |
002fe734 | 1438 | |
e39c0676 | 1439 | return true; |
5bd5f71a | 1440 | } |
2e1df07d | 1441 | |
e39c0676 | 1442 | return false; |
2e1df07d JA |
1443 | } |
1444 | ||
ce486495 | 1445 | static int exec_string(struct thread_options *o, const char *string, const char *mode) |
2e1df07d | 1446 | { |
6c9ce469 | 1447 | size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; |
f491a907 | 1448 | int ret; |
2e1df07d JA |
1449 | char *str; |
1450 | ||
1451 | str = malloc(newlen); | |
ce486495 | 1452 | sprintf(str, "%s &> %s.%s.txt", string, o->name, mode); |
2e1df07d | 1453 | |
ce486495 | 1454 | log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode); |
2e1df07d JA |
1455 | ret = system(str); |
1456 | if (ret == -1) | |
1457 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1458 | ||
1459 | free(str); | |
1460 | return ret; | |
1461 | } | |
1462 | ||
62167762 JC |
1463 | /* |
1464 | * Dry run to compute correct state of numberio for verification. | |
1465 | */ | |
1466 | static uint64_t do_dry_run(struct thread_data *td) | |
1467 | { | |
62167762 JC |
1468 | td_set_runstate(td, TD_RUNNING); |
1469 | ||
1470 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || | |
e28dd2cf | 1471 | (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) { |
62167762 JC |
1472 | struct io_u *io_u; |
1473 | int ret; | |
1474 | ||
1475 | if (td->terminate || td->done) | |
1476 | break; | |
1477 | ||
1478 | io_u = get_io_u(td); | |
99955d3d | 1479 | if (IS_ERR_OR_NULL(io_u)) |
62167762 JC |
1480 | break; |
1481 | ||
1651e431 | 1482 | io_u_set(td, io_u, IO_U_F_FLIGHT); |
62167762 JC |
1483 | io_u->error = 0; |
1484 | io_u->resid = 0; | |
1485 | if (ddir_rw(acct_ddir(io_u))) | |
1486 | td->io_issues[acct_ddir(io_u)]++; | |
1487 | if (ddir_rw(io_u->ddir)) { | |
1488 | io_u_mark_depth(td, 1); | |
1489 | td->ts.total_io_u[io_u->ddir]++; | |
1490 | } | |
1491 | ||
2e63e96b PV |
1492 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1493 | td->o.do_verify && | |
1494 | td->o.verify != VERIFY_NONE && | |
1495 | !td->o.experimental_verify) | |
1496 | log_io_piece(td, io_u); | |
1497 | ||
55312f9f | 1498 | ret = io_u_sync_complete(td, io_u); |
62167762 JC |
1499 | (void) ret; |
1500 | } | |
1501 | ||
55312f9f | 1502 | return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; |
62167762 JC |
1503 | } |
1504 | ||
24660963 JA |
1505 | struct fork_data { |
1506 | struct thread_data *td; | |
1507 | struct sk_out *sk_out; | |
1508 | }; | |
1509 | ||
2e1df07d JA |
1510 | /* |
1511 | * Entry point for the thread based jobs. The process based jobs end up | |
1512 | * here as well, after a little setup. | |
1513 | */ | |
1514 | static void *thread_main(void *data) | |
1515 | { | |
24660963 | 1516 | struct fork_data *fd = data; |
95603b74 | 1517 | unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, }; |
24660963 | 1518 | struct thread_data *td = fd->td; |
4896473e | 1519 | struct thread_options *o = &td->o; |
24660963 | 1520 | struct sk_out *sk_out = fd->sk_out; |
48366f3a | 1521 | uint64_t bytes_done[DDIR_RWDIR_CNT]; |
ca0122d8 | 1522 | int deadlock_loop_cnt; |
c16035aa | 1523 | bool clear_state, did_some_io; |
28727df7 | 1524 | int ret; |
2e1df07d | 1525 | |
24660963 JA |
1526 | sk_out_assign(sk_out); |
1527 | free(fd); | |
1528 | ||
4896473e | 1529 | if (!o->use_thread) { |
2e1df07d JA |
1530 | setsid(); |
1531 | td->pid = getpid(); | |
1532 | } else | |
1533 | td->pid = gettid(); | |
1534 | ||
4896473e | 1535 | fio_local_clock_init(o->use_thread); |
5d879392 | 1536 | |
2e1df07d JA |
1537 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
1538 | ||
122c7725 JA |
1539 | if (is_backend) |
1540 | fio_server_send_start(td); | |
1541 | ||
2e1df07d JA |
1542 | INIT_FLIST_HEAD(&td->io_log_list); |
1543 | INIT_FLIST_HEAD(&td->io_hist_list); | |
1544 | INIT_FLIST_HEAD(&td->verify_list); | |
1545 | INIT_FLIST_HEAD(&td->trim_list); | |
2e1df07d JA |
1546 | td->io_hist_tree = RB_ROOT; |
1547 | ||
34febb23 | 1548 | ret = mutex_cond_init_pshared(&td->io_u_lock, &td->free_cond); |
f9e5b5ee | 1549 | if (ret) { |
34febb23 | 1550 | td_verror(td, ret, "mutex_cond_init_pshared"); |
f9e5b5ee JK |
1551 | goto err; |
1552 | } | |
34febb23 | 1553 | ret = cond_init_pshared(&td->verify_cond); |
f9e5b5ee | 1554 | if (ret) { |
34febb23 | 1555 | td_verror(td, ret, "mutex_cond_pshared"); |
f9e5b5ee JK |
1556 | goto err; |
1557 | } | |
2e1df07d JA |
1558 | |
1559 | td_set_runstate(td, TD_INITIALIZED); | |
971caeb1 BVA |
1560 | dprint(FD_MUTEX, "up startup_sem\n"); |
1561 | fio_sem_up(startup_sem); | |
1562 | dprint(FD_MUTEX, "wait on td->sem\n"); | |
1563 | fio_sem_down(td->sem); | |
1564 | dprint(FD_MUTEX, "done waiting on td->sem\n"); | |
2e1df07d | 1565 | |
2e1df07d JA |
1566 | /* |
1567 | * A new gid requires privilege, so we need to do this before setting | |
1568 | * the uid. | |
1569 | */ | |
4896473e | 1570 | if (o->gid != -1U && setgid(o->gid)) { |
2e1df07d JA |
1571 | td_verror(td, errno, "setgid"); |
1572 | goto err; | |
1573 | } | |
4896473e | 1574 | if (o->uid != -1U && setuid(o->uid)) { |
2e1df07d JA |
1575 | td_verror(td, errno, "setuid"); |
1576 | goto err; | |
1577 | } | |
1578 | ||
2354d810 JA |
1579 | /* |
1580 | * Do this early, we don't want the compress threads to be limited | |
1581 | * to the same CPUs as the IO workers. So do this before we set | |
1582 | * any potential CPU affinity | |
1583 | */ | |
1584 | if (iolog_compress_init(td, sk_out)) | |
1585 | goto err; | |
1586 | ||
2e1df07d JA |
1587 | /* |
1588 | * If we have a gettimeofday() thread, make sure we exclude that | |
1589 | * thread from this job | |
1590 | */ | |
4896473e JA |
1591 | if (o->gtod_cpu) |
1592 | fio_cpu_clear(&o->cpumask, o->gtod_cpu); | |
2e1df07d JA |
1593 | |
1594 | /* | |
1595 | * Set affinity first, in case it has an impact on the memory | |
1596 | * allocations. | |
1597 | */ | |
b2a9e649 | 1598 | if (fio_option_is_set(o, cpumask)) { |
c2acfbac | 1599 | if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { |
30cb4c65 | 1600 | ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); |
c2acfbac JA |
1601 | if (!ret) { |
1602 | log_err("fio: no CPUs set\n"); | |
1603 | log_err("fio: Try increasing number of available CPUs\n"); | |
1604 | td_verror(td, EINVAL, "cpus_split"); | |
1605 | goto err; | |
1606 | } | |
1607 | } | |
28727df7 JA |
1608 | ret = fio_setaffinity(td->pid, o->cpumask); |
1609 | if (ret == -1) { | |
4896473e JA |
1610 | td_verror(td, errno, "cpu_set_affinity"); |
1611 | goto err; | |
1612 | } | |
2e1df07d JA |
1613 | } |
1614 | ||
67bf9823 | 1615 | #ifdef CONFIG_LIBNUMA |
d0b937ed | 1616 | /* numa node setup */ |
b2a9e649 JA |
1617 | if (fio_option_is_set(o, numa_cpunodes) || |
1618 | fio_option_is_set(o, numa_memnodes)) { | |
43522848 | 1619 | struct bitmask *mask; |
d0b937ed YR |
1620 | |
1621 | if (numa_available() < 0) { | |
1622 | td_verror(td, errno, "Does not support NUMA API\n"); | |
1623 | goto err; | |
1624 | } | |
1625 | ||
b2a9e649 | 1626 | if (fio_option_is_set(o, numa_cpunodes)) { |
43522848 DG |
1627 | mask = numa_parse_nodestring(o->numa_cpunodes); |
1628 | ret = numa_run_on_node_mask(mask); | |
1629 | numa_free_nodemask(mask); | |
d0b937ed YR |
1630 | if (ret == -1) { |
1631 | td_verror(td, errno, \ | |
1632 | "numa_run_on_node_mask failed\n"); | |
1633 | goto err; | |
1634 | } | |
1635 | } | |
1636 | ||
b2a9e649 | 1637 | if (fio_option_is_set(o, numa_memnodes)) { |
43522848 DG |
1638 | mask = NULL; |
1639 | if (o->numa_memnodes) | |
1640 | mask = numa_parse_nodestring(o->numa_memnodes); | |
1641 | ||
4896473e | 1642 | switch (o->numa_mem_mode) { |
d0b937ed | 1643 | case MPOL_INTERLEAVE: |
43522848 | 1644 | numa_set_interleave_mask(mask); |
d0b937ed YR |
1645 | break; |
1646 | case MPOL_BIND: | |
43522848 | 1647 | numa_set_membind(mask); |
d0b937ed YR |
1648 | break; |
1649 | case MPOL_LOCAL: | |
1650 | numa_set_localalloc(); | |
1651 | break; | |
1652 | case MPOL_PREFERRED: | |
4896473e | 1653 | numa_set_preferred(o->numa_mem_prefer_node); |
d0b937ed YR |
1654 | break; |
1655 | case MPOL_DEFAULT: | |
1656 | default: | |
1657 | break; | |
1658 | } | |
1659 | ||
43522848 DG |
1660 | if (mask) |
1661 | numa_free_nodemask(mask); | |
1662 | ||
d0b937ed YR |
1663 | } |
1664 | } | |
1665 | #endif | |
1666 | ||
9a3f1100 JA |
1667 | if (fio_pin_memory(td)) |
1668 | goto err; | |
1669 | ||
2e1df07d JA |
1670 | /* |
1671 | * May alter parameters that init_io_u() will use, so we need to | |
1672 | * do this first. | |
1673 | */ | |
1674 | if (init_iolog(td)) | |
1675 | goto err; | |
1676 | ||
1677 | if (init_io_u(td)) | |
1678 | goto err; | |
1679 | ||
4896473e | 1680 | if (o->verify_async && verify_async_init(td)) |
2e1df07d JA |
1681 | goto err; |
1682 | ||
27d74836 JA |
1683 | if (fio_option_is_set(o, ioprio) || |
1684 | fio_option_is_set(o, ioprio_class)) { | |
28727df7 JA |
1685 | ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio); |
1686 | if (ret == -1) { | |
2e1df07d JA |
1687 | td_verror(td, errno, "ioprio_set"); |
1688 | goto err; | |
1689 | } | |
1690 | } | |
1691 | ||
4896473e | 1692 | if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
2e1df07d JA |
1693 | goto err; |
1694 | ||
649c10c5 | 1695 | errno = 0; |
4896473e | 1696 | if (nice(o->nice) == -1 && errno != 0) { |
2e1df07d JA |
1697 | td_verror(td, errno, "nice"); |
1698 | goto err; | |
1699 | } | |
1700 | ||
4896473e | 1701 | if (o->ioscheduler && switch_ioscheduler(td)) |
2e1df07d JA |
1702 | goto err; |
1703 | ||
4896473e | 1704 | if (!o->create_serialize && setup_files(td)) |
2e1df07d JA |
1705 | goto err; |
1706 | ||
1707 | if (td_io_init(td)) | |
1708 | goto err; | |
1709 | ||
4d832322 | 1710 | if (!init_random_map(td)) |
2e1df07d JA |
1711 | goto err; |
1712 | ||
ce486495 | 1713 | if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun")) |
4896473e | 1714 | goto err; |
2e1df07d | 1715 | |
c139f3b4 JA |
1716 | if (o->pre_read && !pre_read_files(td)) |
1717 | goto err; | |
2e1df07d | 1718 | |
dc5bfbb2 JA |
1719 | fio_verify_init(td); |
1720 | ||
24660963 | 1721 | if (rate_submit_init(td, sk_out)) |
a9da8ab2 JA |
1722 | goto err; |
1723 | ||
3aea75b1 | 1724 | set_epoch_time(td, o->log_unix_epoch); |
44404c5a | 1725 | fio_getrusage(&td->ru_start); |
ac28d905 JA |
1726 | memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch)); |
1727 | memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch)); | |
16e56d25 | 1728 | memcpy(&td->ss.prev_time, &td->epoch, sizeof(td->epoch)); |
ac28d905 JA |
1729 | |
1730 | if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || | |
1731 | o->ratemin[DDIR_TRIM]) { | |
1732 | memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, | |
1733 | sizeof(td->bw_sample_time)); | |
1734 | memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, | |
1735 | sizeof(td->bw_sample_time)); | |
1736 | memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time, | |
1737 | sizeof(td->bw_sample_time)); | |
1738 | } | |
1739 | ||
48366f3a | 1740 | memset(bytes_done, 0, sizeof(bytes_done)); |
c16035aa JA |
1741 | clear_state = false; |
1742 | did_some_io = false; | |
48366f3a | 1743 | |
2e1df07d | 1744 | while (keep_running(td)) { |
100f49f1 JA |
1745 | uint64_t verify_bytes; |
1746 | ||
2e1df07d | 1747 | fio_gettime(&td->start, NULL); |
8b6a404c | 1748 | memcpy(&td->ts_cache, &td->start, sizeof(td->start)); |
2e1df07d | 1749 | |
39c1c323 | 1750 | if (clear_state) { |
ac28d905 | 1751 | clear_io_state(td, 0); |
2e1df07d | 1752 | |
39c1c323 | 1753 | if (o->unlink_each_loop && unlink_all_files(td)) |
1754 | break; | |
1755 | } | |
1756 | ||
2e1df07d JA |
1757 | prune_io_piece_log(td); |
1758 | ||
dcf9844e | 1759 | if (td->o.verify_only && td_write(td)) |
62167762 | 1760 | verify_bytes = do_dry_run(td); |
095196b1 | 1761 | else { |
95b03be0 JA |
1762 | do_io(td, bytes_done); |
1763 | ||
1764 | if (!ddir_rw_sum(bytes_done)) { | |
095196b1 | 1765 | fio_mark_td_terminate(td); |
95b03be0 JA |
1766 | verify_bytes = 0; |
1767 | } else { | |
1768 | verify_bytes = bytes_done[DDIR_WRITE] + | |
1769 | bytes_done[DDIR_TRIM]; | |
1770 | } | |
095196b1 | 1771 | } |
2e1df07d | 1772 | |
49f0b8e0 JA |
1773 | /* |
1774 | * If we took too long to shut down, the main thread could | |
1775 | * already consider us reaped/exited. If that happens, break | |
1776 | * out and clean up. | |
1777 | */ | |
1778 | if (td->runstate >= TD_EXITED) | |
1779 | break; | |
1780 | ||
c16035aa | 1781 | clear_state = true; |
2e1df07d | 1782 | |
40c666c8 JA |
1783 | /* |
1784 | * Make sure we've successfully updated the rusage stats | |
1785 | * before waiting on the stat mutex. Otherwise we could have | |
1786 | * the stat thread holding stat mutex and waiting for | |
1787 | * the rusage_sem, which would never get upped because | |
1788 | * this thread is waiting for the stat mutex. | |
1789 | */ | |
ca0122d8 | 1790 | deadlock_loop_cnt = 0; |
39d13e67 JA |
1791 | do { |
1792 | check_update_rusage(td); | |
971caeb1 | 1793 | if (!fio_sem_down_trylock(stat_sem)) |
39d13e67 JA |
1794 | break; |
1795 | usleep(1000); | |
ca0122d8 | 1796 | if (deadlock_loop_cnt++ > 5000) { |
971caeb1 | 1797 | log_err("fio seems to be stuck grabbing stat_sem, forcibly exiting\n"); |
43f248c4 | 1798 | td->error = EDEADLK; |
ca0122d8 TT |
1799 | goto err; |
1800 | } | |
39d13e67 | 1801 | } while (1); |
40c666c8 | 1802 | |
95603b74 BF |
1803 | if (td_read(td) && td->io_bytes[DDIR_READ]) |
1804 | update_runtime(td, elapsed_us, DDIR_READ); | |
1805 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) | |
1806 | update_runtime(td, elapsed_us, DDIR_WRITE); | |
1807 | if (td_trim(td) && td->io_bytes[DDIR_TRIM]) | |
1808 | update_runtime(td, elapsed_us, DDIR_TRIM); | |
e5437a07 | 1809 | fio_gettime(&td->start, NULL); |
971caeb1 | 1810 | fio_sem_up(stat_sem); |
2e1df07d JA |
1811 | |
1812 | if (td->error || td->terminate) | |
1813 | break; | |
1814 | ||
4896473e JA |
1815 | if (!o->do_verify || |
1816 | o->verify == VERIFY_NONE || | |
9b87f09b | 1817 | td_ioengine_flagged(td, FIO_UNIDIR)) |
2e1df07d JA |
1818 | continue; |
1819 | ||
c16035aa JA |
1820 | if (ddir_rw_sum(bytes_done)) |
1821 | did_some_io = true; | |
1822 | ||
ac28d905 | 1823 | clear_io_state(td, 0); |
2e1df07d JA |
1824 | |
1825 | fio_gettime(&td->start, NULL); | |
1826 | ||
100f49f1 | 1827 | do_verify(td, verify_bytes); |
2e1df07d | 1828 | |
40c666c8 JA |
1829 | /* |
1830 | * See comment further up for why this is done here. | |
1831 | */ | |
1832 | check_update_rusage(td); | |
1833 | ||
971caeb1 | 1834 | fio_sem_down(stat_sem); |
95603b74 | 1835 | update_runtime(td, elapsed_us, DDIR_READ); |
e5437a07 | 1836 | fio_gettime(&td->start, NULL); |
971caeb1 | 1837 | fio_sem_up(stat_sem); |
2e1df07d JA |
1838 | |
1839 | if (td->error || td->terminate) | |
1840 | break; | |
1841 | } | |
1842 | ||
48366f3a TK |
1843 | /* |
1844 | * If td ended up with no I/O when it should have had, | |
1845 | * then something went wrong unless FIO_NOIO or FIO_DISKLESSIO. | |
1846 | * (Are we not missing other flags that can be ignored ?) | |
1847 | */ | |
1848 | if ((td->o.size || td->o.io_size) && !ddir_rw_sum(bytes_done) && | |
f7c9bfd5 | 1849 | !did_some_io && !td->o.create_only && |
48366f3a TK |
1850 | !(td_ioengine_flagged(td, FIO_NOIO) || |
1851 | td_ioengine_flagged(td, FIO_DISKLESSIO))) | |
1852 | log_err("%s: No I/O performed by %s, " | |
1853 | "perhaps try --debug=io option for details?\n", | |
1854 | td->o.name, td->io_ops->name); | |
1855 | ||
2cea0b4c JA |
1856 | td_set_runstate(td, TD_FINISHING); |
1857 | ||
2e1df07d | 1858 | update_rusage_stat(td); |
2e1df07d | 1859 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
6eaf09d6 SL |
1860 | td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ]; |
1861 | td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; | |
1862 | td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; | |
2e1df07d | 1863 | |
ca09be4b | 1864 | if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && |
d264264a JA |
1865 | (td->o.verify != VERIFY_NONE && td_write(td))) |
1866 | verify_save_state(td->thread_number); | |
ca09be4b | 1867 | |
9a3f1100 JA |
1868 | fio_unpin_memory(td); |
1869 | ||
a47591e4 | 1870 | td_writeout_logs(td, true); |
2e1df07d | 1871 | |
155f2f02 | 1872 | iolog_compress_exit(td); |
103b174e | 1873 | rate_submit_exit(td); |
aee2ab67 | 1874 | |
4896473e | 1875 | if (o->exec_postrun) |
ce486495 | 1876 | exec_string(o, o->exec_postrun, (const char *)"postrun"); |
2e1df07d | 1877 | |
f9cafb12 | 1878 | if (exitall_on_terminate || (o->exitall_error && td->error)) |
2e1df07d JA |
1879 | fio_terminate_threads(td->groupid); |
1880 | ||
1881 | err: | |
1882 | if (td->error) | |
1883 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, | |
1884 | td->verror); | |
1885 | ||
4896473e | 1886 | if (o->verify_async) |
2e1df07d JA |
1887 | verify_async_exit(td); |
1888 | ||
1889 | close_and_free_files(td); | |
2e1df07d | 1890 | cleanup_io_u(td); |
32dbca2c | 1891 | close_ioengine(td); |
2e1df07d | 1892 | cgroup_shutdown(td, &cgroup_mnt); |
ca09be4b | 1893 | verify_free_state(td); |
2e1df07d | 1894 | |
7c4d0fb7 JA |
1895 | if (td->zone_state_index) { |
1896 | int i; | |
1897 | ||
1898 | for (i = 0; i < DDIR_RWDIR_CNT; i++) | |
1899 | free(td->zone_state_index[i]); | |
1900 | free(td->zone_state_index); | |
1901 | td->zone_state_index = NULL; | |
1902 | } | |
1903 | ||
b2a9e649 | 1904 | if (fio_option_is_set(o, cpumask)) { |
8a1db9a1 JA |
1905 | ret = fio_cpuset_exit(&o->cpumask); |
1906 | if (ret) | |
1907 | td_verror(td, ret, "fio_cpuset_exit"); | |
2e1df07d JA |
1908 | } |
1909 | ||
1910 | /* | |
1911 | * do this very late, it will log file closing as well | |
1912 | */ | |
4896473e | 1913 | if (o->write_iolog_file) |
2e1df07d JA |
1914 | write_iolog_close(td); |
1915 | ||
1916 | td_set_runstate(td, TD_EXITED); | |
fda2cfac JA |
1917 | |
1918 | /* | |
1919 | * Do this last after setting our runstate to exited, so we | |
1920 | * know that the stat thread is signaled. | |
1921 | */ | |
1922 | check_update_rusage(td); | |
1923 | ||
24660963 | 1924 | sk_out_drop(); |
e43606c2 | 1925 | return (void *) (uintptr_t) td->error; |
2e1df07d JA |
1926 | } |
1927 | ||
2e1df07d JA |
1928 | /* |
1929 | * Run over the job map and reap the threads that have exited, if any. | |
1930 | */ | |
90eff1c9 SW |
1931 | static void reap_threads(unsigned int *nr_running, uint64_t *t_rate, |
1932 | uint64_t *m_rate) | |
2e1df07d JA |
1933 | { |
1934 | struct thread_data *td; | |
1935 | unsigned int cputhreads, realthreads, pending; | |
1936 | int i, status, ret; | |
1937 | ||
1938 | /* | |
1939 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1940 | */ | |
1941 | realthreads = pending = cputhreads = 0; | |
1942 | for_each_td(td, i) { | |
1943 | int flags = 0; | |
1944 | ||
8847ae4c | 1945 | if (!strcmp(td->o.ioengine, "cpuio")) |
2e1df07d JA |
1946 | cputhreads++; |
1947 | else | |
1948 | realthreads++; | |
1949 | ||
1950 | if (!td->pid) { | |
1951 | pending++; | |
1952 | continue; | |
1953 | } | |
1954 | if (td->runstate == TD_REAPED) | |
1955 | continue; | |
1956 | if (td->o.use_thread) { | |
1957 | if (td->runstate == TD_EXITED) { | |
1958 | td_set_runstate(td, TD_REAPED); | |
1959 | goto reaped; | |
1960 | } | |
1961 | continue; | |
1962 | } | |
1963 | ||
1964 | flags = WNOHANG; | |
1965 | if (td->runstate == TD_EXITED) | |
1966 | flags = 0; | |
1967 | ||
1968 | /* | |
1969 | * check if someone quit or got killed in an unusual way | |
1970 | */ | |
1971 | ret = waitpid(td->pid, &status, flags); | |
1972 | if (ret < 0) { | |
1973 | if (errno == ECHILD) { | |
1974 | log_err("fio: pid=%d disappeared %d\n", | |
1975 | (int) td->pid, td->runstate); | |
a5e371a6 | 1976 | td->sig = ECHILD; |
2e1df07d JA |
1977 | td_set_runstate(td, TD_REAPED); |
1978 | goto reaped; | |
1979 | } | |
1980 | perror("waitpid"); | |
1981 | } else if (ret == td->pid) { | |
1982 | if (WIFSIGNALED(status)) { | |
1983 | int sig = WTERMSIG(status); | |
1984 | ||
36d80bc7 | 1985 | if (sig != SIGTERM && sig != SIGUSR2) |
2e1df07d JA |
1986 | log_err("fio: pid=%d, got signal=%d\n", |
1987 | (int) td->pid, sig); | |
a5e371a6 | 1988 | td->sig = sig; |
2e1df07d JA |
1989 | td_set_runstate(td, TD_REAPED); |
1990 | goto reaped; | |
1991 | } | |
1992 | if (WIFEXITED(status)) { | |
1993 | if (WEXITSTATUS(status) && !td->error) | |
1994 | td->error = WEXITSTATUS(status); | |
1995 | ||
1996 | td_set_runstate(td, TD_REAPED); | |
1997 | goto reaped; | |
1998 | } | |
1999 | } | |
2000 | ||
cba5460c JA |
2001 | /* |
2002 | * If the job is stuck, do a forceful timeout of it and | |
2003 | * move on. | |
2004 | */ | |
2005 | if (td->terminate && | |
2c45a4ac | 2006 | td->runstate < TD_FSYNCING && |
cba5460c | 2007 | time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { |
86310a1f TK |
2008 | log_err("fio: job '%s' (state=%d) hasn't exited in " |
2009 | "%lu seconds, it appears to be stuck. Doing " | |
2010 | "forceful exit of this job.\n", | |
2011 | td->o.name, td->runstate, | |
2012 | (unsigned long) time_since_now(&td->terminate_time)); | |
cba5460c JA |
2013 | td_set_runstate(td, TD_REAPED); |
2014 | goto reaped; | |
2015 | } | |
2016 | ||
2e1df07d JA |
2017 | /* |
2018 | * thread is not dead, continue | |
2019 | */ | |
2020 | pending++; | |
2021 | continue; | |
2022 | reaped: | |
2023 | (*nr_running)--; | |
342f4be4 JA |
2024 | (*m_rate) -= ddir_rw_sum(td->o.ratemin); |
2025 | (*t_rate) -= ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
2026 | if (!td->pid) |
2027 | pending--; | |
2028 | ||
2029 | if (td->error) | |
2030 | exit_value++; | |
2031 | ||
2032 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
4a88752a | 2033 | profile_td_exit(td); |
2e1df07d JA |
2034 | } |
2035 | ||
2036 | if (*nr_running == cputhreads && !pending && realthreads) | |
2037 | fio_terminate_threads(TERMINATE_ALL); | |
2038 | } | |
2039 | ||
e39c0676 | 2040 | static bool __check_trigger_file(void) |
ca09be4b JA |
2041 | { |
2042 | struct stat sb; | |
2043 | ||
2044 | if (!trigger_file) | |
e39c0676 | 2045 | return false; |
ca09be4b JA |
2046 | |
2047 | if (stat(trigger_file, &sb)) | |
e39c0676 | 2048 | return false; |
ca09be4b JA |
2049 | |
2050 | if (unlink(trigger_file) < 0) | |
2051 | log_err("fio: failed to unlink %s: %s\n", trigger_file, | |
2052 | strerror(errno)); | |
2053 | ||
e39c0676 | 2054 | return true; |
ca09be4b JA |
2055 | } |
2056 | ||
e39c0676 | 2057 | static bool trigger_timedout(void) |
ca09be4b JA |
2058 | { |
2059 | if (trigger_timeout) | |
98413be7 SW |
2060 | if (time_since_genesis() >= trigger_timeout) { |
2061 | trigger_timeout = 0; | |
2062 | return true; | |
2063 | } | |
ca09be4b | 2064 | |
e39c0676 | 2065 | return false; |
ca09be4b JA |
2066 | } |
2067 | ||
2068 | void exec_trigger(const char *cmd) | |
2069 | { | |
2070 | int ret; | |
2071 | ||
98413be7 | 2072 | if (!cmd || cmd[0] == '\0') |
ca09be4b JA |
2073 | return; |
2074 | ||
2075 | ret = system(cmd); | |
2076 | if (ret == -1) | |
2077 | log_err("fio: failed executing %s trigger\n", cmd); | |
2078 | } | |
2079 | ||
2080 | void check_trigger_file(void) | |
2081 | { | |
2082 | if (__check_trigger_file() || trigger_timedout()) { | |
796fb3ce JA |
2083 | if (nr_clients) |
2084 | fio_clients_send_trigger(trigger_remote_cmd); | |
2085 | else { | |
d264264a | 2086 | verify_save_state(IO_LIST_ALL); |
ca09be4b JA |
2087 | fio_terminate_threads(TERMINATE_ALL); |
2088 | exec_trigger(trigger_cmd); | |
2089 | } | |
2090 | } | |
2091 | } | |
2092 | ||
2093 | static int fio_verify_load_state(struct thread_data *td) | |
2094 | { | |
2095 | int ret; | |
2096 | ||
2097 | if (!td->o.verify_state) | |
2098 | return 0; | |
2099 | ||
2100 | if (is_backend) { | |
2101 | void *data; | |
2102 | ||
2103 | ret = fio_server_get_verify_state(td->o.name, | |
94a6e1bb | 2104 | td->thread_number - 1, &data); |
ca09be4b | 2105 | if (!ret) |
94a6e1bb | 2106 | verify_assign_state(td, data); |
ca09be4b JA |
2107 | } else |
2108 | ret = verify_load_state(td, "local"); | |
2109 | ||
2110 | return ret; | |
2111 | } | |
2112 | ||
06464907 JA |
2113 | static void do_usleep(unsigned int usecs) |
2114 | { | |
2115 | check_for_running_stats(); | |
ca09be4b | 2116 | check_trigger_file(); |
06464907 JA |
2117 | usleep(usecs); |
2118 | } | |
2119 | ||
e39c0676 | 2120 | static bool check_mount_writes(struct thread_data *td) |
e81ecca3 JA |
2121 | { |
2122 | struct fio_file *f; | |
2123 | unsigned int i; | |
2124 | ||
2125 | if (!td_write(td) || td->o.allow_mounted_write) | |
e39c0676 | 2126 | return false; |
e81ecca3 | 2127 | |
5e81f9c1 TK |
2128 | /* |
2129 | * If FIO_HAVE_CHARDEV_SIZE is defined, it's likely that chrdevs | |
2130 | * are mkfs'd and mounted. | |
2131 | */ | |
e81ecca3 | 2132 | for_each_file(td, f, i) { |
5e81f9c1 TK |
2133 | #ifdef FIO_HAVE_CHARDEV_SIZE |
2134 | if (f->filetype != FIO_TYPE_BLOCK && f->filetype != FIO_TYPE_CHAR) | |
2135 | #else | |
686fbd31 | 2136 | if (f->filetype != FIO_TYPE_BLOCK) |
5e81f9c1 | 2137 | #endif |
e81ecca3 JA |
2138 | continue; |
2139 | if (device_is_mounted(f->file_name)) | |
2140 | goto mounted; | |
2141 | } | |
2142 | ||
e39c0676 | 2143 | return false; |
e81ecca3 | 2144 | mounted: |
c5839011 | 2145 | log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.\n", f->file_name); |
e39c0676 | 2146 | return true; |
e81ecca3 JA |
2147 | } |
2148 | ||
9cc8cb91 AK |
2149 | static bool waitee_running(struct thread_data *me) |
2150 | { | |
2151 | const char *waitee = me->o.wait_for; | |
2152 | const char *self = me->o.name; | |
2153 | struct thread_data *td; | |
2154 | int i; | |
2155 | ||
2156 | if (!waitee) | |
2157 | return false; | |
2158 | ||
2159 | for_each_td(td, i) { | |
2160 | if (!strcmp(td->o.name, self) || strcmp(td->o.name, waitee)) | |
2161 | continue; | |
2162 | ||
2163 | if (td->runstate < TD_EXITED) { | |
2164 | dprint(FD_PROCESS, "%s fenced by %s(%s)\n", | |
2165 | self, td->o.name, | |
2166 | runstate_to_name(td->runstate)); | |
2167 | return true; | |
2168 | } | |
2169 | } | |
2170 | ||
2171 | dprint(FD_PROCESS, "%s: %s completed, can run\n", self, waitee); | |
2172 | return false; | |
2173 | } | |
2174 | ||
2e1df07d JA |
2175 | /* |
2176 | * Main function for kicking off and reaping jobs, as needed. | |
2177 | */ | |
24660963 | 2178 | static void run_threads(struct sk_out *sk_out) |
2e1df07d JA |
2179 | { |
2180 | struct thread_data *td; | |
90eff1c9 SW |
2181 | unsigned int i, todo, nr_running, nr_started; |
2182 | uint64_t m_rate, t_rate; | |
0de5b26f | 2183 | uint64_t spent; |
2e1df07d | 2184 | |
2e1df07d JA |
2185 | if (fio_gtod_offload && fio_start_gtod_thread()) |
2186 | return; | |
334185e9 | 2187 | |
f2a2ce0e | 2188 | fio_idle_prof_init(); |
2e1df07d JA |
2189 | |
2190 | set_sig_handlers(); | |
2191 | ||
3a5f6bde JA |
2192 | nr_thread = nr_process = 0; |
2193 | for_each_td(td, i) { | |
e81ecca3 JA |
2194 | if (check_mount_writes(td)) |
2195 | return; | |
3a5f6bde JA |
2196 | if (td->o.use_thread) |
2197 | nr_thread++; | |
2198 | else | |
2199 | nr_process++; | |
2200 | } | |
2201 | ||
01cfefcc | 2202 | if (output_format & FIO_OUTPUT_NORMAL) { |
2e1df07d JA |
2203 | log_info("Starting "); |
2204 | if (nr_thread) | |
2205 | log_info("%d thread%s", nr_thread, | |
2206 | nr_thread > 1 ? "s" : ""); | |
2207 | if (nr_process) { | |
2208 | if (nr_thread) | |
2209 | log_info(" and "); | |
2210 | log_info("%d process%s", nr_process, | |
2211 | nr_process > 1 ? "es" : ""); | |
2212 | } | |
2213 | log_info("\n"); | |
e411c301 | 2214 | log_info_flush(); |
2e1df07d JA |
2215 | } |
2216 | ||
2217 | todo = thread_number; | |
2218 | nr_running = 0; | |
2219 | nr_started = 0; | |
2220 | m_rate = t_rate = 0; | |
2221 | ||
2222 | for_each_td(td, i) { | |
2223 | print_status_init(td->thread_number - 1); | |
2224 | ||
2225 | if (!td->o.create_serialize) | |
2226 | continue; | |
2227 | ||
ca09be4b JA |
2228 | if (fio_verify_load_state(td)) |
2229 | goto reap; | |
2230 | ||
2e1df07d JA |
2231 | /* |
2232 | * do file setup here so it happens sequentially, | |
2233 | * we don't want X number of threads getting their | |
2234 | * client data interspersed on disk | |
2235 | */ | |
2236 | if (setup_files(td)) { | |
ca09be4b | 2237 | reap: |
2e1df07d JA |
2238 | exit_value++; |
2239 | if (td->error) | |
2240 | log_err("fio: pid=%d, err=%d/%s\n", | |
2241 | (int) td->pid, td->error, td->verror); | |
2242 | td_set_runstate(td, TD_REAPED); | |
2243 | todo--; | |
2244 | } else { | |
2245 | struct fio_file *f; | |
2246 | unsigned int j; | |
2247 | ||
2248 | /* | |
2249 | * for sharing to work, each job must always open | |
2250 | * its own files. so close them, if we opened them | |
2251 | * for creation | |
2252 | */ | |
2253 | for_each_file(td, f, j) { | |
2254 | if (fio_file_open(f)) | |
2255 | td_io_close_file(td, f); | |
2256 | } | |
2257 | } | |
2258 | } | |
2259 | ||
f2a2ce0e HL |
2260 | /* start idle threads before io threads start to run */ |
2261 | fio_idle_prof_start(); | |
2262 | ||
2e1df07d JA |
2263 | set_genesis_time(); |
2264 | ||
2265 | while (todo) { | |
2266 | struct thread_data *map[REAL_MAX_JOBS]; | |
8b6a404c | 2267 | struct timespec this_start; |
2e1df07d | 2268 | int this_jobs = 0, left; |
ce8b6139 | 2269 | struct fork_data *fd; |
2e1df07d JA |
2270 | |
2271 | /* | |
2272 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
2273 | */ | |
2274 | for_each_td(td, i) { | |
2275 | if (td->runstate != TD_NOT_CREATED) | |
2276 | continue; | |
2277 | ||
2278 | /* | |
2279 | * never got a chance to start, killed by other | |
2280 | * thread for some reason | |
2281 | */ | |
2282 | if (td->terminate) { | |
2283 | todo--; | |
2284 | continue; | |
2285 | } | |
2286 | ||
2287 | if (td->o.start_delay) { | |
0de5b26f | 2288 | spent = utime_since_genesis(); |
2e1df07d | 2289 | |
74454ce4 | 2290 | if (td->o.start_delay > spent) |
2e1df07d JA |
2291 | continue; |
2292 | } | |
2293 | ||
2294 | if (td->o.stonewall && (nr_started || nr_running)) { | |
2295 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
2296 | td->o.name); | |
2297 | break; | |
2298 | } | |
2299 | ||
9cc8cb91 AK |
2300 | if (waitee_running(td)) { |
2301 | dprint(FD_PROCESS, "%s: waiting for %s\n", | |
2302 | td->o.name, td->o.wait_for); | |
5c74fc76 | 2303 | continue; |
9cc8cb91 AK |
2304 | } |
2305 | ||
2e1df07d JA |
2306 | init_disk_util(td); |
2307 | ||
971caeb1 | 2308 | td->rusage_sem = fio_sem_init(FIO_SEM_LOCKED); |
c97f1ad6 JA |
2309 | td->update_rusage = 0; |
2310 | ||
2e1df07d JA |
2311 | /* |
2312 | * Set state to created. Thread will transition | |
2313 | * to TD_INITIALIZED when it's done setting up. | |
2314 | */ | |
2315 | td_set_runstate(td, TD_CREATED); | |
2316 | map[this_jobs++] = td; | |
2317 | nr_started++; | |
2318 | ||
ce8b6139 JA |
2319 | fd = calloc(1, sizeof(*fd)); |
2320 | fd->td = td; | |
2321 | fd->sk_out = sk_out; | |
2322 | ||
2e1df07d JA |
2323 | if (td->o.use_thread) { |
2324 | int ret; | |
2325 | ||
2326 | dprint(FD_PROCESS, "will pthread_create\n"); | |
2327 | ret = pthread_create(&td->thread, NULL, | |
24660963 | 2328 | thread_main, fd); |
2e1df07d JA |
2329 | if (ret) { |
2330 | log_err("pthread_create: %s\n", | |
2331 | strerror(ret)); | |
24660963 | 2332 | free(fd); |
2e1df07d JA |
2333 | nr_started--; |
2334 | break; | |
2335 | } | |
493dd6c1 | 2336 | fd = NULL; |
2e1df07d JA |
2337 | ret = pthread_detach(td->thread); |
2338 | if (ret) | |
2339 | log_err("pthread_detach: %s", | |
2340 | strerror(ret)); | |
2341 | } else { | |
2342 | pid_t pid; | |
2343 | dprint(FD_PROCESS, "will fork\n"); | |
2344 | pid = fork(); | |
2345 | if (!pid) { | |
ce8b6139 | 2346 | int ret; |
2e1df07d | 2347 | |
ce8b6139 | 2348 | ret = (int)(uintptr_t)thread_main(fd); |
2e1df07d JA |
2349 | _exit(ret); |
2350 | } else if (i == fio_debug_jobno) | |
2351 | *fio_debug_jobp = pid; | |
2352 | } | |
971caeb1 BVA |
2353 | dprint(FD_MUTEX, "wait on startup_sem\n"); |
2354 | if (fio_sem_down_timeout(startup_sem, 10000)) { | |
2e1df07d JA |
2355 | log_err("fio: job startup hung? exiting.\n"); |
2356 | fio_terminate_threads(TERMINATE_ALL); | |
2357 | fio_abort = 1; | |
2358 | nr_started--; | |
6541f4e0 | 2359 | free(fd); |
2e1df07d JA |
2360 | break; |
2361 | } | |
971caeb1 | 2362 | dprint(FD_MUTEX, "done waiting on startup_sem\n"); |
2e1df07d JA |
2363 | } |
2364 | ||
2365 | /* | |
2366 | * Wait for the started threads to transition to | |
2367 | * TD_INITIALIZED. | |
2368 | */ | |
2369 | fio_gettime(&this_start, NULL); | |
2370 | left = this_jobs; | |
2371 | while (left && !fio_abort) { | |
2372 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
2373 | break; | |
2374 | ||
06464907 | 2375 | do_usleep(100000); |
2e1df07d JA |
2376 | |
2377 | for (i = 0; i < this_jobs; i++) { | |
2378 | td = map[i]; | |
2379 | if (!td) | |
2380 | continue; | |
2381 | if (td->runstate == TD_INITIALIZED) { | |
2382 | map[i] = NULL; | |
2383 | left--; | |
2384 | } else if (td->runstate >= TD_EXITED) { | |
2385 | map[i] = NULL; | |
2386 | left--; | |
2387 | todo--; | |
2388 | nr_running++; /* work-around... */ | |
2389 | } | |
2390 | } | |
2391 | } | |
2392 | ||
2393 | if (left) { | |
4e87c37a JA |
2394 | log_err("fio: %d job%s failed to start\n", left, |
2395 | left > 1 ? "s" : ""); | |
2e1df07d JA |
2396 | for (i = 0; i < this_jobs; i++) { |
2397 | td = map[i]; | |
2398 | if (!td) | |
2399 | continue; | |
2400 | kill(td->pid, SIGTERM); | |
2401 | } | |
2402 | break; | |
2403 | } | |
2404 | ||
2405 | /* | |
2406 | * start created threads (TD_INITIALIZED -> TD_RUNNING). | |
2407 | */ | |
2408 | for_each_td(td, i) { | |
2409 | if (td->runstate != TD_INITIALIZED) | |
2410 | continue; | |
2411 | ||
2412 | if (in_ramp_time(td)) | |
2413 | td_set_runstate(td, TD_RAMP); | |
2414 | else | |
2415 | td_set_runstate(td, TD_RUNNING); | |
2416 | nr_running++; | |
2417 | nr_started--; | |
342f4be4 JA |
2418 | m_rate += ddir_rw_sum(td->o.ratemin); |
2419 | t_rate += ddir_rw_sum(td->o.rate); | |
2e1df07d | 2420 | todo--; |
971caeb1 | 2421 | fio_sem_up(td->sem); |
2e1df07d JA |
2422 | } |
2423 | ||
2424 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2425 | ||
122c7725 | 2426 | if (todo) |
06464907 | 2427 | do_usleep(100000); |
2e1df07d JA |
2428 | } |
2429 | ||
2430 | while (nr_running) { | |
2431 | reap_threads(&nr_running, &t_rate, &m_rate); | |
06464907 | 2432 | do_usleep(10000); |
2e1df07d JA |
2433 | } |
2434 | ||
f2a2ce0e HL |
2435 | fio_idle_prof_stop(); |
2436 | ||
2e1df07d | 2437 | update_io_ticks(); |
2e1df07d JA |
2438 | } |
2439 | ||
27357187 JA |
2440 | static void free_disk_util(void) |
2441 | { | |
27357187 | 2442 | disk_util_prune_entries(); |
a39fb9ea | 2443 | helper_thread_destroy(); |
2e1df07d JA |
2444 | } |
2445 | ||
24660963 | 2446 | int fio_backend(struct sk_out *sk_out) |
2e1df07d JA |
2447 | { |
2448 | struct thread_data *td; | |
2449 | int i; | |
2450 | ||
2451 | if (exec_profile) { | |
2452 | if (load_profile(exec_profile)) | |
2453 | return 1; | |
2454 | free(exec_profile); | |
2455 | exec_profile = NULL; | |
2456 | } | |
2457 | if (!thread_number) | |
2458 | return 0; | |
2459 | ||
2460 | if (write_bw_log) { | |
aee2ab67 JA |
2461 | struct log_params p = { |
2462 | .log_type = IO_LOG_TYPE_BW, | |
2463 | }; | |
2464 | ||
2465 | setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log"); | |
2466 | setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log"); | |
2467 | setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); | |
2e1df07d JA |
2468 | } |
2469 | ||
971caeb1 BVA |
2470 | startup_sem = fio_sem_init(FIO_SEM_LOCKED); |
2471 | if (startup_sem == NULL) | |
2e1df07d | 2472 | return 1; |
2e1df07d JA |
2473 | |
2474 | set_genesis_time(); | |
cef9175e | 2475 | stat_init(); |
971caeb1 | 2476 | helper_thread_create(startup_sem, sk_out); |
2e1df07d JA |
2477 | |
2478 | cgroup_list = smalloc(sizeof(*cgroup_list)); | |
b2e6494c JA |
2479 | if (cgroup_list) |
2480 | INIT_FLIST_HEAD(cgroup_list); | |
2e1df07d | 2481 | |
24660963 | 2482 | run_threads(sk_out); |
2e1df07d | 2483 | |
a39fb9ea | 2484 | helper_thread_exit(); |
8aab824f | 2485 | |
2e1df07d | 2486 | if (!fio_abort) { |
83f7b64e | 2487 | __show_run_stats(); |
2e1df07d | 2488 | if (write_bw_log) { |
cb7e0ace JA |
2489 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { |
2490 | struct io_log *log = agg_io_log[i]; | |
2491 | ||
60a25727 | 2492 | flush_log(log, false); |
518dac09 | 2493 | free_log(log); |
cb7e0ace | 2494 | } |
2e1df07d JA |
2495 | } |
2496 | } | |
2497 | ||
fda2cfac | 2498 | for_each_td(td, i) { |
c27cc65f | 2499 | steadystate_free(td); |
2e1df07d | 2500 | fio_options_free(td); |
8049adc1 | 2501 | if (td->rusage_sem) { |
971caeb1 | 2502 | fio_sem_remove(td->rusage_sem); |
8049adc1 JA |
2503 | td->rusage_sem = NULL; |
2504 | } | |
971caeb1 BVA |
2505 | fio_sem_remove(td->sem); |
2506 | td->sem = NULL; | |
fda2cfac | 2507 | } |
2e1df07d | 2508 | |
a462baef | 2509 | free_disk_util(); |
b2e6494c JA |
2510 | if (cgroup_list) { |
2511 | cgroup_kill(cgroup_list); | |
2512 | sfree(cgroup_list); | |
2513 | } | |
2e1df07d JA |
2514 | sfree(cgroup_mnt); |
2515 | ||
971caeb1 | 2516 | fio_sem_remove(startup_sem); |
cef9175e | 2517 | stat_exit(); |
2e1df07d JA |
2518 | return exit_value; |
2519 | } |