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> |
c06379a6 | 32 | #include <pthread.h> |
2e1df07d JA |
33 | |
34 | #include "fio.h" | |
2e1df07d JA |
35 | #include "smalloc.h" |
36 | #include "verify.h" | |
2e1df07d JA |
37 | #include "diskutil.h" |
38 | #include "cgroup.h" | |
39 | #include "profile.h" | |
40 | #include "lib/rand.h" | |
f7690c4a | 41 | #include "lib/memalign.h" |
2e1df07d | 42 | #include "server.h" |
44404c5a | 43 | #include "lib/getrusage.h" |
f2a2ce0e | 44 | #include "idletime.h" |
002fe734 | 45 | #include "err.h" |
a9da8ab2 | 46 | #include "workqueue.h" |
e81ecca3 | 47 | #include "lib/mountcheck.h" |
40511301 | 48 | #include "rate-submit.h" |
a39fb9ea | 49 | #include "helper_thread.h" |
ae626d4e | 50 | #include "pshared.h" |
07776127 | 51 | #include "zone-dist.h" |
2e1df07d | 52 | |
971caeb1 | 53 | static struct fio_sem *startup_sem; |
2e1df07d | 54 | static struct flist_head *cgroup_list; |
c3dc516a | 55 | static struct cgroup_mnt *cgroup_mnt; |
2e1df07d | 56 | static int exit_value; |
f1867a7f | 57 | static volatile bool fio_abort; |
3a5f6bde JA |
58 | static unsigned int nr_process = 0; |
59 | static unsigned int nr_thread = 0; | |
2e1df07d | 60 | |
6eaf09d6 | 61 | struct io_log *agg_io_log[DDIR_RWDIR_CNT]; |
2e1df07d | 62 | |
a3efc919 JA |
63 | int groupid = 0; |
64 | unsigned int thread_number = 0; | |
108fea77 | 65 | unsigned int stat_number = 0; |
a3efc919 JA |
66 | int shm_id = 0; |
67 | int temp_stall_ts; | |
68 | unsigned long done_secs = 0; | |
c06379a6 | 69 | pthread_mutex_t overlap_check = PTHREAD_MUTEX_INITIALIZER; |
a47591e4 | 70 | |
2e1df07d JA |
71 | #define JOB_START_TIMEOUT (5 * 1000) |
72 | ||
73 | static void sig_int(int sig) | |
74 | { | |
75 | if (threads) { | |
76 | if (is_backend) | |
77 | fio_server_got_signal(sig); | |
78 | else { | |
79 | log_info("\nfio: terminating on signal %d\n", sig); | |
e411c301 | 80 | log_info_flush(); |
2e1df07d JA |
81 | exit_value = 128; |
82 | } | |
83 | ||
64402a8a | 84 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
2e1df07d JA |
85 | } |
86 | } | |
87 | ||
d2235e56 | 88 | void sig_show_status(int sig) |
4c6d91e8 JA |
89 | { |
90 | show_running_run_stats(); | |
91 | } | |
92 | ||
2e1df07d JA |
93 | static void set_sig_handlers(void) |
94 | { | |
95 | struct sigaction act; | |
96 | ||
97 | memset(&act, 0, sizeof(act)); | |
98 | act.sa_handler = sig_int; | |
99 | act.sa_flags = SA_RESTART; | |
100 | sigaction(SIGINT, &act, NULL); | |
101 | ||
102 | memset(&act, 0, sizeof(act)); | |
103 | act.sa_handler = sig_int; | |
104 | act.sa_flags = SA_RESTART; | |
105 | sigaction(SIGTERM, &act, NULL); | |
106 | ||
2f694507 BC |
107 | /* Windows uses SIGBREAK as a quit signal from other applications */ |
108 | #ifdef WIN32 | |
109 | memset(&act, 0, sizeof(act)); | |
110 | act.sa_handler = sig_int; | |
111 | act.sa_flags = SA_RESTART; | |
112 | sigaction(SIGBREAK, &act, NULL); | |
113 | #endif | |
114 | ||
4c6d91e8 JA |
115 | memset(&act, 0, sizeof(act)); |
116 | act.sa_handler = sig_show_status; | |
117 | act.sa_flags = SA_RESTART; | |
118 | sigaction(SIGUSR1, &act, NULL); | |
119 | ||
2e1df07d JA |
120 | if (is_backend) { |
121 | memset(&act, 0, sizeof(act)); | |
122 | act.sa_handler = sig_int; | |
123 | act.sa_flags = SA_RESTART; | |
124 | sigaction(SIGPIPE, &act, NULL); | |
125 | } | |
126 | } | |
127 | ||
128 | /* | |
129 | * Check if we are above the minimum rate given. | |
130 | */ | |
8b6a404c | 131 | static bool __check_min_rate(struct thread_data *td, struct timespec *now, |
e39c0676 | 132 | enum fio_ddir ddir) |
2e1df07d JA |
133 | { |
134 | unsigned long long bytes = 0; | |
135 | unsigned long iops = 0; | |
136 | unsigned long spent; | |
137 | unsigned long rate; | |
138 | unsigned int ratemin = 0; | |
139 | unsigned int rate_iops = 0; | |
140 | unsigned int rate_iops_min = 0; | |
141 | ||
142 | assert(ddir_rw(ddir)); | |
143 | ||
144 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) | |
e39c0676 | 145 | return false; |
2e1df07d JA |
146 | |
147 | /* | |
148 | * allow a 2 second settle period in the beginning | |
149 | */ | |
150 | if (mtime_since(&td->start, now) < 2000) | |
e39c0676 | 151 | return false; |
2e1df07d JA |
152 | |
153 | iops += td->this_io_blocks[ddir]; | |
154 | bytes += td->this_io_bytes[ddir]; | |
155 | ratemin += td->o.ratemin[ddir]; | |
156 | rate_iops += td->o.rate_iops[ddir]; | |
157 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
158 | ||
159 | /* | |
160 | * if rate blocks is set, sample is running | |
161 | */ | |
162 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { | |
163 | spent = mtime_since(&td->lastrate[ddir], now); | |
164 | if (spent < td->o.ratecycle) | |
e39c0676 | 165 | return false; |
2e1df07d | 166 | |
20e37e4a | 167 | if (td->o.rate[ddir] || td->o.ratemin[ddir]) { |
2e1df07d JA |
168 | /* |
169 | * check bandwidth specified rate | |
170 | */ | |
171 | if (bytes < td->rate_bytes[ddir]) { | |
420b104a RE |
172 | log_err("%s: rate_min=%uB/s not met, only transferred %lluB\n", |
173 | td->o.name, ratemin, bytes); | |
e39c0676 | 174 | return true; |
2e1df07d | 175 | } else { |
49cba9b3 JA |
176 | if (spent) |
177 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; | |
178 | else | |
179 | rate = 0; | |
180 | ||
2e1df07d JA |
181 | if (rate < ratemin || |
182 | bytes < td->rate_bytes[ddir]) { | |
420b104a RE |
183 | log_err("%s: rate_min=%uB/s not met, got %luB/s\n", |
184 | td->o.name, ratemin, rate); | |
e39c0676 | 185 | return true; |
2e1df07d JA |
186 | } |
187 | } | |
188 | } else { | |
189 | /* | |
190 | * checks iops specified rate | |
191 | */ | |
192 | if (iops < rate_iops) { | |
420b104a RE |
193 | log_err("%s: rate_iops_min=%u not met, only performed %lu IOs\n", |
194 | td->o.name, rate_iops, iops); | |
e39c0676 | 195 | return true; |
2e1df07d | 196 | } else { |
4c707a3b JA |
197 | if (spent) |
198 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; | |
199 | else | |
200 | rate = 0; | |
201 | ||
2e1df07d JA |
202 | if (rate < rate_iops_min || |
203 | iops < td->rate_blocks[ddir]) { | |
420b104a RE |
204 | log_err("%s: rate_iops_min=%u not met, got %lu IOPS\n", |
205 | td->o.name, rate_iops_min, rate); | |
e39c0676 | 206 | return true; |
2e1df07d JA |
207 | } |
208 | } | |
209 | } | |
210 | } | |
211 | ||
212 | td->rate_bytes[ddir] = bytes; | |
213 | td->rate_blocks[ddir] = iops; | |
214 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
e39c0676 | 215 | return false; |
2e1df07d JA |
216 | } |
217 | ||
8b6a404c | 218 | static bool check_min_rate(struct thread_data *td, struct timespec *now) |
2e1df07d | 219 | { |
e39c0676 | 220 | bool ret = false; |
2e1df07d | 221 | |
55312f9f | 222 | if (td->bytes_done[DDIR_READ]) |
6eaf09d6 | 223 | ret |= __check_min_rate(td, now, DDIR_READ); |
55312f9f | 224 | if (td->bytes_done[DDIR_WRITE]) |
6eaf09d6 | 225 | ret |= __check_min_rate(td, now, DDIR_WRITE); |
55312f9f | 226 | if (td->bytes_done[DDIR_TRIM]) |
6eaf09d6 | 227 | ret |= __check_min_rate(td, now, DDIR_TRIM); |
2e1df07d JA |
228 | |
229 | return ret; | |
230 | } | |
231 | ||
232 | /* | |
233 | * When job exits, we can cancel the in-flight IO if we are using async | |
234 | * io. Attempt to do so. | |
235 | */ | |
236 | static void cleanup_pending_aio(struct thread_data *td) | |
237 | { | |
2e1df07d JA |
238 | int r; |
239 | ||
d28174f0 JA |
240 | if (td->error) |
241 | return; | |
242 | ||
2e1df07d JA |
243 | /* |
244 | * get immediately available events, if any | |
245 | */ | |
55312f9f | 246 | r = io_u_queued_complete(td, 0); |
2e1df07d JA |
247 | if (r < 0) |
248 | return; | |
249 | ||
250 | /* | |
251 | * now cancel remaining active events | |
252 | */ | |
253 | if (td->io_ops->cancel) { | |
2ae0b204 JA |
254 | struct io_u *io_u; |
255 | int i; | |
2e1df07d | 256 | |
2ae0b204 JA |
257 | io_u_qiter(&td->io_u_all, io_u, i) { |
258 | if (io_u->flags & IO_U_F_FLIGHT) { | |
2e1df07d JA |
259 | r = td->io_ops->cancel(td, io_u); |
260 | if (!r) | |
261 | put_io_u(td, io_u); | |
262 | } | |
263 | } | |
264 | } | |
265 | ||
266 | if (td->cur_depth) | |
55312f9f | 267 | r = io_u_queued_complete(td, td->cur_depth); |
2e1df07d JA |
268 | } |
269 | ||
270 | /* | |
271 | * Helper to handle the final sync of a file. Works just like the normal | |
272 | * io path, just does everything sync. | |
273 | */ | |
e39c0676 | 274 | static bool fio_io_sync(struct thread_data *td, struct fio_file *f) |
2e1df07d JA |
275 | { |
276 | struct io_u *io_u = __get_io_u(td); | |
d3b07186 | 277 | enum fio_q_status ret; |
2e1df07d JA |
278 | |
279 | if (!io_u) | |
e39c0676 | 280 | return true; |
2e1df07d JA |
281 | |
282 | io_u->ddir = DDIR_SYNC; | |
283 | io_u->file = f; | |
00615bfb | 284 | io_u_set(td, io_u, IO_U_F_NO_FILE_PUT); |
2e1df07d JA |
285 | |
286 | if (td_io_prep(td, io_u)) { | |
287 | put_io_u(td, io_u); | |
e39c0676 | 288 | return true; |
2e1df07d JA |
289 | } |
290 | ||
291 | requeue: | |
292 | ret = td_io_queue(td, io_u); | |
18f6165e BVA |
293 | switch (ret) { |
294 | case FIO_Q_QUEUED: | |
a80cb54b | 295 | td_io_commit(td); |
55312f9f | 296 | if (io_u_queued_complete(td, 1) < 0) |
e39c0676 | 297 | return true; |
18f6165e BVA |
298 | break; |
299 | case FIO_Q_COMPLETED: | |
2e1df07d JA |
300 | if (io_u->error) { |
301 | td_verror(td, io_u->error, "td_io_queue"); | |
e39c0676 | 302 | return true; |
2e1df07d JA |
303 | } |
304 | ||
55312f9f | 305 | if (io_u_sync_complete(td, io_u) < 0) |
e39c0676 | 306 | return true; |
18f6165e BVA |
307 | break; |
308 | case FIO_Q_BUSY: | |
a80cb54b | 309 | td_io_commit(td); |
2e1df07d JA |
310 | goto requeue; |
311 | } | |
312 | ||
e39c0676 | 313 | return false; |
2e1df07d | 314 | } |
a3efc919 | 315 | |
61ee0f86 JA |
316 | static int fio_file_fsync(struct thread_data *td, struct fio_file *f) |
317 | { | |
00615bfb | 318 | int ret, ret2; |
61ee0f86 JA |
319 | |
320 | if (fio_file_open(f)) | |
321 | return fio_io_sync(td, f); | |
322 | ||
323 | if (td_io_open_file(td, f)) | |
324 | return 1; | |
325 | ||
326 | ret = fio_io_sync(td, f); | |
00615bfb SW |
327 | ret2 = 0; |
328 | if (fio_file_open(f)) | |
329 | ret2 = td_io_close_file(td, f); | |
330 | return (ret || ret2); | |
61ee0f86 JA |
331 | } |
332 | ||
8b6a404c | 333 | static inline void __update_ts_cache(struct thread_data *td) |
2e1df07d | 334 | { |
8b6a404c | 335 | fio_gettime(&td->ts_cache, NULL); |
2e1df07d JA |
336 | } |
337 | ||
8b6a404c | 338 | static inline void update_ts_cache(struct thread_data *td) |
2e1df07d | 339 | { |
8b6a404c VF |
340 | if ((++td->ts_cache_nr & td->ts_cache_mask) == td->ts_cache_mask) |
341 | __update_ts_cache(td); | |
2e1df07d JA |
342 | } |
343 | ||
8b6a404c | 344 | static inline bool runtime_exceeded(struct thread_data *td, struct timespec *t) |
2e1df07d JA |
345 | { |
346 | if (in_ramp_time(td)) | |
e39c0676 | 347 | return false; |
2e1df07d | 348 | if (!td->o.timeout) |
e39c0676 | 349 | return false; |
0de5b26f | 350 | if (utime_since(&td->epoch, t) >= td->o.timeout) |
e39c0676 | 351 | return true; |
2e1df07d | 352 | |
e39c0676 | 353 | return false; |
2e1df07d JA |
354 | } |
355 | ||
95603b74 BF |
356 | /* |
357 | * We need to update the runtime consistently in ms, but keep a running | |
358 | * tally of the current elapsed time in microseconds for sub millisecond | |
359 | * updates. | |
360 | */ | |
361 | static inline void update_runtime(struct thread_data *td, | |
362 | unsigned long long *elapsed_us, | |
363 | const enum fio_ddir ddir) | |
364 | { | |
d5c8ea29 JA |
365 | if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only) |
366 | return; | |
367 | ||
95603b74 BF |
368 | td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000; |
369 | elapsed_us[ddir] += utime_since_now(&td->start); | |
370 | td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000; | |
371 | } | |
372 | ||
e39c0676 JA |
373 | static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir, |
374 | int *retptr) | |
2e1df07d JA |
375 | { |
376 | int ret = *retptr; | |
377 | ||
378 | if (ret < 0 || td->error) { | |
8b28bd41 DM |
379 | int err = td->error; |
380 | enum error_type_bit eb; | |
2e1df07d JA |
381 | |
382 | if (ret < 0) | |
383 | err = -ret; | |
2e1df07d | 384 | |
8b28bd41 DM |
385 | eb = td_error_type(ddir, err); |
386 | if (!(td->o.continue_on_error & (1 << eb))) | |
e39c0676 | 387 | return true; |
2e1df07d | 388 | |
8b28bd41 | 389 | if (td_non_fatal_error(td, eb, err)) { |
2e1df07d JA |
390 | /* |
391 | * Continue with the I/Os in case of | |
392 | * a non fatal error. | |
393 | */ | |
394 | update_error_count(td, err); | |
395 | td_clear_error(td); | |
396 | *retptr = 0; | |
e39c0676 | 397 | return false; |
2e1df07d JA |
398 | } else if (td->o.fill_device && err == ENOSPC) { |
399 | /* | |
400 | * We expect to hit this error if | |
401 | * fill_device option is set. | |
402 | */ | |
403 | td_clear_error(td); | |
ebea2133 | 404 | fio_mark_td_terminate(td); |
e39c0676 | 405 | return true; |
2e1df07d JA |
406 | } else { |
407 | /* | |
408 | * Stop the I/O in case of a fatal | |
409 | * error. | |
410 | */ | |
411 | update_error_count(td, err); | |
e39c0676 | 412 | return true; |
2e1df07d JA |
413 | } |
414 | } | |
415 | ||
e39c0676 | 416 | return false; |
2e1df07d JA |
417 | } |
418 | ||
c97f1ad6 JA |
419 | static void check_update_rusage(struct thread_data *td) |
420 | { | |
421 | if (td->update_rusage) { | |
422 | td->update_rusage = 0; | |
423 | update_rusage_stat(td); | |
971caeb1 | 424 | fio_sem_up(td->rusage_sem); |
c97f1ad6 JA |
425 | } |
426 | } | |
427 | ||
8b6a404c | 428 | static int wait_for_completions(struct thread_data *td, struct timespec *time) |
69fea81e JA |
429 | { |
430 | const int full = queue_full(td); | |
431 | int min_evts = 0; | |
432 | int ret; | |
433 | ||
6be06c46 JA |
434 | if (td->flags & TD_F_REGROW_LOGS) |
435 | return io_u_quiesce(td); | |
1fed2080 | 436 | |
69fea81e JA |
437 | /* |
438 | * if the queue is full, we MUST reap at least 1 event | |
439 | */ | |
82407585 | 440 | min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth); |
4b157ac6 | 441 | if ((full && !min_evts) || !td->o.iodepth_batch_complete_min) |
69fea81e JA |
442 | min_evts = 1; |
443 | ||
46960d0d | 444 | if (time && __should_check_rate(td)) |
69fea81e JA |
445 | fio_gettime(time, NULL); |
446 | ||
447 | do { | |
55312f9f | 448 | ret = io_u_queued_complete(td, min_evts); |
69fea81e JA |
449 | if (ret < 0) |
450 | break; | |
451 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
452 | ||
453 | return ret; | |
454 | } | |
455 | ||
e9d512d8 JA |
456 | int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, |
457 | enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify, | |
8b6a404c | 458 | struct timespec *comp_time) |
e9d512d8 | 459 | { |
e9d512d8 JA |
460 | switch (*ret) { |
461 | case FIO_Q_COMPLETED: | |
462 | if (io_u->error) { | |
463 | *ret = -io_u->error; | |
464 | clear_io_u(td, io_u); | |
465 | } else if (io_u->resid) { | |
5fff9543 | 466 | long long bytes = io_u->xfer_buflen - io_u->resid; |
e9d512d8 JA |
467 | struct fio_file *f = io_u->file; |
468 | ||
469 | if (bytes_issued) | |
470 | *bytes_issued += bytes; | |
471 | ||
472 | if (!from_verify) | |
94a78d52 | 473 | trim_io_piece(io_u); |
e9d512d8 JA |
474 | |
475 | /* | |
476 | * zero read, fail | |
477 | */ | |
478 | if (!bytes) { | |
479 | if (!from_verify) | |
480 | unlog_io_piece(td, io_u); | |
481 | td_verror(td, EIO, "full resid"); | |
482 | put_io_u(td, io_u); | |
483 | break; | |
484 | } | |
485 | ||
486 | io_u->xfer_buflen = io_u->resid; | |
487 | io_u->xfer_buf += bytes; | |
488 | io_u->offset += bytes; | |
489 | ||
490 | if (ddir_rw(io_u->ddir)) | |
491 | td->ts.short_io_u[io_u->ddir]++; | |
492 | ||
e9d512d8 JA |
493 | if (io_u->offset == f->real_file_size) |
494 | goto sync_done; | |
495 | ||
496 | requeue_io_u(td, &io_u); | |
497 | } else { | |
498 | sync_done: | |
46960d0d | 499 | if (comp_time && __should_check_rate(td)) |
e9d512d8 JA |
500 | fio_gettime(comp_time, NULL); |
501 | ||
502 | *ret = io_u_sync_complete(td, io_u); | |
503 | if (*ret < 0) | |
504 | break; | |
505 | } | |
a0e3e5a6 | 506 | |
cf2c8d52 JA |
507 | if (td->flags & TD_F_REGROW_LOGS) |
508 | regrow_logs(td); | |
509 | ||
a0e3e5a6 EW |
510 | /* |
511 | * when doing I/O (not when verifying), | |
512 | * check for any errors that are to be ignored | |
513 | */ | |
514 | if (!from_verify) | |
515 | break; | |
516 | ||
e9d512d8 JA |
517 | return 0; |
518 | case FIO_Q_QUEUED: | |
519 | /* | |
520 | * if the engine doesn't have a commit hook, | |
521 | * the io_u is really queued. if it does have such | |
522 | * a hook, it has to call io_u_queued() itself. | |
523 | */ | |
524 | if (td->io_ops->commit == NULL) | |
525 | io_u_queued(td, io_u); | |
526 | if (bytes_issued) | |
527 | *bytes_issued += io_u->xfer_buflen; | |
528 | break; | |
529 | case FIO_Q_BUSY: | |
530 | if (!from_verify) | |
531 | unlog_io_piece(td, io_u); | |
532 | requeue_io_u(td, &io_u); | |
a80cb54b | 533 | td_io_commit(td); |
e9d512d8 JA |
534 | break; |
535 | default: | |
b0775325 | 536 | assert(*ret < 0); |
e9d512d8 JA |
537 | td_verror(td, -(*ret), "td_io_queue"); |
538 | break; | |
539 | } | |
540 | ||
541 | if (break_on_this_error(td, ddir, ret)) | |
542 | return 1; | |
543 | ||
544 | return 0; | |
545 | } | |
546 | ||
e39c0676 | 547 | static inline bool io_in_polling(struct thread_data *td) |
82407585 RP |
548 | { |
549 | return !td->o.iodepth_batch_complete_min && | |
550 | !td->o.iodepth_batch_complete_max; | |
551 | } | |
39c1c323 | 552 | /* |
553 | * Unlinks files from thread data fio_file structure | |
554 | */ | |
555 | static int unlink_all_files(struct thread_data *td) | |
556 | { | |
557 | struct fio_file *f; | |
558 | unsigned int i; | |
559 | int ret = 0; | |
560 | ||
561 | for_each_file(td, f, i) { | |
562 | if (f->filetype != FIO_TYPE_FILE) | |
563 | continue; | |
564 | ret = td_io_unlink_file(td, f); | |
565 | if (ret) | |
566 | break; | |
567 | } | |
568 | ||
569 | if (ret) | |
570 | td_verror(td, ret, "unlink_all_files"); | |
571 | ||
572 | return ret; | |
573 | } | |
82407585 | 574 | |
997b5680 SW |
575 | /* |
576 | * Check if io_u will overlap an in-flight IO in the queue | |
577 | */ | |
c06379a6 | 578 | bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) |
997b5680 SW |
579 | { |
580 | bool overlap; | |
581 | struct io_u *check_io_u; | |
582 | unsigned long long x1, x2, y1, y2; | |
583 | int i; | |
584 | ||
585 | x1 = io_u->offset; | |
586 | x2 = io_u->offset + io_u->buflen; | |
587 | overlap = false; | |
588 | io_u_qiter(q, check_io_u, i) { | |
589 | if (check_io_u->flags & IO_U_F_FLIGHT) { | |
590 | y1 = check_io_u->offset; | |
591 | y2 = check_io_u->offset + check_io_u->buflen; | |
592 | ||
593 | if (x1 < y2 && y1 < x2) { | |
594 | overlap = true; | |
5fff9543 | 595 | dprint(FD_IO, "in-flight overlap: %llu/%llu, %llu/%llu\n", |
997b5680 SW |
596 | x1, io_u->buflen, |
597 | y1, check_io_u->buflen); | |
598 | break; | |
599 | } | |
600 | } | |
601 | } | |
602 | ||
603 | return overlap; | |
604 | } | |
605 | ||
d3b07186 | 606 | static enum fio_q_status io_u_submit(struct thread_data *td, struct io_u *io_u) |
08b2b386 JA |
607 | { |
608 | /* | |
609 | * Check for overlap if the user asked us to, and we have | |
610 | * at least one IO in flight besides this one. | |
611 | */ | |
612 | if (td->o.serialize_overlap && td->cur_depth > 1 && | |
613 | in_flight_overlap(&td->io_u_all, io_u)) | |
614 | return FIO_Q_BUSY; | |
615 | ||
616 | return td_io_queue(td, io_u); | |
617 | } | |
618 | ||
2e1df07d JA |
619 | /* |
620 | * The main verify engine. Runs over the writes we previously submitted, | |
621 | * reads the blocks back in, and checks the crc/md5 of the data. | |
622 | */ | |
100f49f1 | 623 | static void do_verify(struct thread_data *td, uint64_t verify_bytes) |
2e1df07d JA |
624 | { |
625 | struct fio_file *f; | |
626 | struct io_u *io_u; | |
627 | int ret, min_events; | |
628 | unsigned int i; | |
629 | ||
630 | dprint(FD_VERIFY, "starting loop\n"); | |
631 | ||
632 | /* | |
633 | * sync io first and invalidate cache, to make sure we really | |
634 | * read from disk. | |
635 | */ | |
636 | for_each_file(td, f, i) { | |
637 | if (!fio_file_open(f)) | |
638 | continue; | |
639 | if (fio_io_sync(td, f)) | |
640 | break; | |
641 | if (file_invalidate_cache(td, f)) | |
642 | break; | |
643 | } | |
644 | ||
c97f1ad6 JA |
645 | check_update_rusage(td); |
646 | ||
2e1df07d JA |
647 | if (td->error) |
648 | return; | |
649 | ||
8f380de2 CJ |
650 | /* |
651 | * verify_state needs to be reset before verification | |
652 | * proceeds so that expected random seeds match actual | |
653 | * random seeds in headers. The main loop will reset | |
654 | * all random number generators if randrepeat is set. | |
655 | */ | |
656 | if (!td->o.rand_repeatable) | |
657 | td_fill_verify_state_seed(td); | |
658 | ||
2e1df07d JA |
659 | td_set_runstate(td, TD_VERIFYING); |
660 | ||
661 | io_u = NULL; | |
662 | while (!td->terminate) { | |
fbccf46c | 663 | enum fio_ddir ddir; |
e9d512d8 | 664 | int full; |
2e1df07d | 665 | |
8b6a404c | 666 | update_ts_cache(td); |
c97f1ad6 | 667 | check_update_rusage(td); |
2e1df07d | 668 | |
8b6a404c VF |
669 | if (runtime_exceeded(td, &td->ts_cache)) { |
670 | __update_ts_cache(td); | |
671 | if (runtime_exceeded(td, &td->ts_cache)) { | |
ebea2133 | 672 | fio_mark_td_terminate(td); |
2e1df07d JA |
673 | break; |
674 | } | |
675 | } | |
676 | ||
9e684a49 DE |
677 | if (flow_threshold_exceeded(td)) |
678 | continue; | |
679 | ||
44cbc6da JA |
680 | if (!td->o.experimental_verify) { |
681 | io_u = __get_io_u(td); | |
682 | if (!io_u) | |
683 | break; | |
2e1df07d | 684 | |
44cbc6da JA |
685 | if (get_next_verify(td, io_u)) { |
686 | put_io_u(td, io_u); | |
687 | break; | |
688 | } | |
2e1df07d | 689 | |
44cbc6da JA |
690 | if (td_io_prep(td, io_u)) { |
691 | put_io_u(td, io_u); | |
692 | break; | |
693 | } | |
694 | } else { | |
55312f9f | 695 | if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes) |
100f49f1 JA |
696 | break; |
697 | ||
bcd5abfa | 698 | while ((io_u = get_io_u(td)) != NULL) { |
059b61f2 | 699 | if (IS_ERR_OR_NULL(io_u)) { |
002fe734 JA |
700 | io_u = NULL; |
701 | ret = FIO_Q_BUSY; | |
702 | goto reap; | |
703 | } | |
704 | ||
bcd5abfa JA |
705 | /* |
706 | * We are only interested in the places where | |
707 | * we wrote or trimmed IOs. Turn those into | |
708 | * reads for verification purposes. | |
709 | */ | |
710 | if (io_u->ddir == DDIR_READ) { | |
711 | /* | |
712 | * Pretend we issued it for rwmix | |
713 | * accounting | |
714 | */ | |
715 | td->io_issues[DDIR_READ]++; | |
716 | put_io_u(td, io_u); | |
717 | continue; | |
718 | } else if (io_u->ddir == DDIR_TRIM) { | |
719 | io_u->ddir = DDIR_READ; | |
1651e431 | 720 | io_u_set(td, io_u, IO_U_F_TRIMMED); |
bcd5abfa JA |
721 | break; |
722 | } else if (io_u->ddir == DDIR_WRITE) { | |
723 | io_u->ddir = DDIR_READ; | |
5c5c33c1 | 724 | populate_verify_io_u(td, io_u); |
bcd5abfa JA |
725 | break; |
726 | } else { | |
727 | put_io_u(td, io_u); | |
728 | continue; | |
729 | } | |
730 | } | |
44cbc6da | 731 | |
bcd5abfa | 732 | if (!io_u) |
44cbc6da | 733 | break; |
2e1df07d JA |
734 | } |
735 | ||
ca09be4b JA |
736 | if (verify_state_should_stop(td, io_u)) { |
737 | put_io_u(td, io_u); | |
738 | break; | |
739 | } | |
740 | ||
2e1df07d JA |
741 | if (td->o.verify_async) |
742 | io_u->end_io = verify_io_u_async; | |
743 | else | |
744 | io_u->end_io = verify_io_u; | |
745 | ||
fbccf46c | 746 | ddir = io_u->ddir; |
10adc4a7 GG |
747 | if (!td->o.disable_slat) |
748 | fio_gettime(&io_u->start_time, NULL); | |
fbccf46c | 749 | |
08b2b386 | 750 | ret = io_u_submit(td, io_u); |
2e1df07d | 751 | |
e9d512d8 | 752 | if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL)) |
2e1df07d JA |
753 | break; |
754 | ||
755 | /* | |
756 | * if we can queue more, do so. but check if there are | |
757 | * completed io_u's first. Note that we can get BUSY even | |
758 | * without IO queued, if the system is resource starved. | |
759 | */ | |
002fe734 | 760 | reap: |
2e1df07d | 761 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
82407585 | 762 | if (full || io_in_polling(td)) |
55312f9f | 763 | ret = wait_for_completions(td, NULL); |
2e1df07d | 764 | |
2e1df07d JA |
765 | if (ret < 0) |
766 | break; | |
767 | } | |
768 | ||
c97f1ad6 JA |
769 | check_update_rusage(td); |
770 | ||
2e1df07d JA |
771 | if (!td->error) { |
772 | min_events = td->cur_depth; | |
773 | ||
774 | if (min_events) | |
55312f9f | 775 | ret = io_u_queued_complete(td, min_events); |
2e1df07d JA |
776 | } else |
777 | cleanup_pending_aio(td); | |
778 | ||
779 | td_set_runstate(td, TD_RUNNING); | |
780 | ||
781 | dprint(FD_VERIFY, "exiting loop\n"); | |
782 | } | |
783 | ||
e39c0676 | 784 | static bool exceeds_number_ios(struct thread_data *td) |
3939fe85 JA |
785 | { |
786 | unsigned long long number_ios; | |
787 | ||
788 | if (!td->o.number_ios) | |
e39c0676 | 789 | return false; |
3939fe85 | 790 | |
cf8a46a7 | 791 | number_ios = ddir_rw_sum(td->io_blocks); |
3939fe85 JA |
792 | number_ios += td->io_u_queued + td->io_u_in_flight; |
793 | ||
cf8a46a7 | 794 | return number_ios >= (td->o.number_ios * td->loops); |
3939fe85 JA |
795 | } |
796 | ||
926098e0 | 797 | static bool io_bytes_exceeded(struct thread_data *td, uint64_t *this_bytes) |
f7078f7b | 798 | { |
77731b29 | 799 | unsigned long long bytes, limit; |
f7078f7b JA |
800 | |
801 | if (td_rw(td)) | |
926098e0 | 802 | bytes = this_bytes[DDIR_READ] + this_bytes[DDIR_WRITE]; |
f7078f7b | 803 | else if (td_write(td)) |
926098e0 | 804 | bytes = this_bytes[DDIR_WRITE]; |
6eaf09d6 | 805 | else if (td_read(td)) |
926098e0 | 806 | bytes = this_bytes[DDIR_READ]; |
f7078f7b | 807 | else |
926098e0 | 808 | bytes = this_bytes[DDIR_TRIM]; |
f7078f7b | 809 | |
5be9bf09 TK |
810 | if (td->o.io_size) |
811 | limit = td->o.io_size; | |
77731b29 JA |
812 | else |
813 | limit = td->o.size; | |
814 | ||
cf8a46a7 | 815 | limit *= td->loops; |
77731b29 | 816 | return bytes >= limit || exceeds_number_ios(td); |
f7078f7b JA |
817 | } |
818 | ||
926098e0 | 819 | static bool io_issue_bytes_exceeded(struct thread_data *td) |
e28dd2cf | 820 | { |
926098e0 JA |
821 | return io_bytes_exceeded(td, td->io_issue_bytes); |
822 | } | |
e28dd2cf | 823 | |
926098e0 JA |
824 | static bool io_complete_bytes_exceeded(struct thread_data *td) |
825 | { | |
826 | return io_bytes_exceeded(td, td->this_io_bytes); | |
e28dd2cf JE |
827 | } |
828 | ||
50a8ce86 D |
829 | /* |
830 | * used to calculate the next io time for rate control | |
831 | * | |
832 | */ | |
833 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) | |
834 | { | |
1a9bf814 | 835 | uint64_t bps = td->rate_bps[ddir]; |
50a8ce86 D |
836 | |
837 | assert(!(td->flags & TD_F_CHILD)); | |
ff6bb260 | 838 | |
6de65959 | 839 | if (td->o.rate_process == RATE_PROCESS_POISSON) { |
1a9bf814 JA |
840 | uint64_t val, iops; |
841 | ||
ff6bb260 | 842 | iops = bps / td->o.bs[ddir]; |
a77d139b | 843 | val = (int64_t) (1000000 / iops) * |
6c8611c7 | 844 | -logf(__rand_0_1(&td->poisson_state[ddir])); |
a77d139b | 845 | if (val) { |
98e51d7c JA |
846 | dprint(FD_RATE, "poisson rate iops=%llu, ddir=%d\n", |
847 | (unsigned long long) 1000000 / val, | |
848 | ddir); | |
a77d139b | 849 | } |
6c8611c7 JA |
850 | td->last_usec[ddir] += val; |
851 | return td->last_usec[ddir]; | |
ff6bb260 | 852 | } else if (bps) { |
1a9bf814 JA |
853 | uint64_t bytes = td->rate_io_issue_bytes[ddir]; |
854 | uint64_t secs = bytes / bps; | |
855 | uint64_t remainder = bytes % bps; | |
856 | ||
50a8ce86 | 857 | return remainder * 1000000 / bps + secs * 1000000; |
e7b24047 JA |
858 | } |
859 | ||
860 | return 0; | |
50a8ce86 D |
861 | } |
862 | ||
1a9bf814 JA |
863 | static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir) |
864 | { | |
865 | unsigned long long b; | |
866 | uint64_t total; | |
867 | int left; | |
868 | ||
869 | b = ddir_rw_sum(td->io_blocks); | |
870 | if (b % td->o.thinktime_blocks) | |
871 | return; | |
872 | ||
873 | io_u_quiesce(td); | |
874 | ||
875 | total = 0; | |
876 | if (td->o.thinktime_spin) | |
877 | total = usec_spin(td->o.thinktime_spin); | |
878 | ||
879 | left = td->o.thinktime - total; | |
880 | if (left) | |
881 | total += usec_sleep(td, left); | |
882 | ||
883 | /* | |
884 | * If we're ignoring thinktime for the rate, add the number of bytes | |
1aa39b0c JA |
885 | * we would have done while sleeping, minus one block to ensure we |
886 | * start issuing immediately after the sleep. | |
1a9bf814 | 887 | */ |
1aa39b0c | 888 | if (total && td->rate_bps[ddir] && td->o.rate_ign_think) { |
f2aedb17 | 889 | uint64_t missed = (td->rate_bps[ddir] * total) / 1000000ULL; |
62b858f6 JA |
890 | uint64_t bs = td->o.min_bs[ddir]; |
891 | uint64_t usperop = bs * 1000000ULL / td->rate_bps[ddir]; | |
f2aedb17 JA |
892 | uint64_t over; |
893 | ||
62b858f6 JA |
894 | if (usperop <= total) |
895 | over = bs; | |
f2aedb17 | 896 | else |
62b858f6 | 897 | over = (usperop - total) / usperop * -bs; |
f2aedb17 JA |
898 | |
899 | td->rate_io_issue_bytes[ddir] += (missed - over); | |
182cab24 JA |
900 | /* adjust for rate_process=poisson */ |
901 | td->last_usec[ddir] += total; | |
1aa39b0c | 902 | } |
1a9bf814 JA |
903 | } |
904 | ||
2e1df07d JA |
905 | /* |
906 | * Main IO worker function. It retrieves io_u's to process and queues | |
907 | * and reaps them, checking for rate and errors along the way. | |
100f49f1 JA |
908 | * |
909 | * Returns number of bytes written and trimmed. | |
2e1df07d | 910 | */ |
95b03be0 | 911 | static void do_io(struct thread_data *td, uint64_t *bytes_done) |
2e1df07d JA |
912 | { |
913 | unsigned int i; | |
914 | int ret = 0; | |
c2703bf3 | 915 | uint64_t total_bytes, bytes_issued = 0; |
70c68076 | 916 | |
95b03be0 JA |
917 | for (i = 0; i < DDIR_RWDIR_CNT; i++) |
918 | bytes_done[i] = td->bytes_done[i]; | |
2e1df07d JA |
919 | |
920 | if (in_ramp_time(td)) | |
921 | td_set_runstate(td, TD_RAMP); | |
922 | else | |
923 | td_set_runstate(td, TD_RUNNING); | |
924 | ||
3e260a46 JA |
925 | lat_target_init(td); |
926 | ||
1e564979 JE |
927 | total_bytes = td->o.size; |
928 | /* | |
5be9bf09 | 929 | * Allow random overwrite workloads to write up to io_size |
1e564979 JE |
930 | * before starting verification phase as 'size' doesn't apply. |
931 | */ | |
932 | if (td_write(td) && td_random(td) && td->o.norandommap) | |
5be9bf09 | 933 | total_bytes = max(total_bytes, (uint64_t) td->o.io_size); |
78a6469c JA |
934 | /* |
935 | * If verify_backlog is enabled, we'll run the verify in this | |
936 | * handler as well. For that case, we may need up to twice the | |
937 | * amount of bytes. | |
938 | */ | |
78a6469c JA |
939 | if (td->o.verify != VERIFY_NONE && |
940 | (td_write(td) && td->o.verify_backlog)) | |
c2703bf3 JA |
941 | total_bytes += td->o.size; |
942 | ||
82a90686 | 943 | /* In trimwrite mode, each byte is trimmed and then written, so |
0e4dd95c | 944 | * allow total_bytes to be twice as big */ |
82a90686 | 945 | if (td_trimwrite(td)) |
0e4dd95c DE |
946 | total_bytes += td->total_io_size; |
947 | ||
f7078f7b | 948 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
e28dd2cf | 949 | (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) || |
c04e4661 | 950 | td->o.time_based) { |
8b6a404c | 951 | struct timespec comp_time; |
2e1df07d | 952 | struct io_u *io_u; |
e9d512d8 | 953 | int full; |
2e1df07d JA |
954 | enum fio_ddir ddir; |
955 | ||
c97f1ad6 JA |
956 | check_update_rusage(td); |
957 | ||
7d7803fa | 958 | if (td->terminate || td->done) |
2e1df07d JA |
959 | break; |
960 | ||
8b6a404c | 961 | update_ts_cache(td); |
2e1df07d | 962 | |
8b6a404c VF |
963 | if (runtime_exceeded(td, &td->ts_cache)) { |
964 | __update_ts_cache(td); | |
965 | if (runtime_exceeded(td, &td->ts_cache)) { | |
ebea2133 | 966 | fio_mark_td_terminate(td); |
2e1df07d JA |
967 | break; |
968 | } | |
969 | } | |
970 | ||
9e684a49 DE |
971 | if (flow_threshold_exceeded(td)) |
972 | continue; | |
973 | ||
f1a32461 JA |
974 | /* |
975 | * Break if we exceeded the bytes. The exception is time | |
976 | * based runs, but we still need to break out of the loop | |
977 | * for those to run verification, if enabled. | |
98e7161c | 978 | * Jobs read from iolog do not use this stop condition. |
f1a32461 JA |
979 | */ |
980 | if (bytes_issued >= total_bytes && | |
98e7161c | 981 | !td->o.read_iolog_file && |
f1a32461 JA |
982 | (!td->o.time_based || |
983 | (td->o.time_based && td->o.verify != VERIFY_NONE))) | |
20876c53 JC |
984 | break; |
985 | ||
2e1df07d | 986 | io_u = get_io_u(td); |
002fe734 JA |
987 | if (IS_ERR_OR_NULL(io_u)) { |
988 | int err = PTR_ERR(io_u); | |
989 | ||
990 | io_u = NULL; | |
1a9bf814 | 991 | ddir = DDIR_INVAL; |
002fe734 JA |
992 | if (err == -EBUSY) { |
993 | ret = FIO_Q_BUSY; | |
994 | goto reap; | |
995 | } | |
3e260a46 JA |
996 | if (td->o.latency_target) |
997 | goto reap; | |
2e1df07d | 998 | break; |
3e260a46 | 999 | } |
2e1df07d | 1000 | |
5c5c33c1 BVA |
1001 | if (io_u->ddir == DDIR_WRITE && td->flags & TD_F_DO_VERIFY) |
1002 | populate_verify_io_u(td, io_u); | |
1003 | ||
2e1df07d JA |
1004 | ddir = io_u->ddir; |
1005 | ||
1006 | /* | |
82af2a7c JA |
1007 | * Add verification end_io handler if: |
1008 | * - Asked to verify (!td_rw(td)) | |
1009 | * - Or the io_u is from our verify list (mixed write/ver) | |
2e1df07d JA |
1010 | */ |
1011 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && | |
82af2a7c | 1012 | ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { |
c4b6117b PV |
1013 | |
1014 | if (!td->o.verify_pattern_bytes) { | |
d6b72507 | 1015 | io_u->rand_seed = __rand(&td->verify_state); |
c4b6117b | 1016 | if (sizeof(int) != sizeof(long *)) |
d6b72507 | 1017 | io_u->rand_seed *= __rand(&td->verify_state); |
c4b6117b PV |
1018 | } |
1019 | ||
ca09be4b JA |
1020 | if (verify_state_should_stop(td, io_u)) { |
1021 | put_io_u(td, io_u); | |
1022 | break; | |
1023 | } | |
1024 | ||
2e1df07d JA |
1025 | if (td->o.verify_async) |
1026 | io_u->end_io = verify_io_u_async; | |
1027 | else | |
1028 | io_u->end_io = verify_io_u; | |
1029 | td_set_runstate(td, TD_VERIFYING); | |
1030 | } else if (in_ramp_time(td)) | |
1031 | td_set_runstate(td, TD_RAMP); | |
1032 | else | |
1033 | td_set_runstate(td, TD_RUNNING); | |
1034 | ||
9a50c5c5 | 1035 | /* |
f9401285 JA |
1036 | * Always log IO before it's issued, so we know the specific |
1037 | * order of it. The logged unit will track when the IO has | |
1038 | * completed. | |
9a50c5c5 | 1039 | */ |
c4b6117b PV |
1040 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1041 | td->o.do_verify && | |
1042 | td->o.verify != VERIFY_NONE && | |
f9401285 | 1043 | !td->o.experimental_verify) |
c4b6117b PV |
1044 | log_io_piece(td, io_u); |
1045 | ||
a9da8ab2 | 1046 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
5fff9543 | 1047 | const unsigned long long blen = io_u->xfer_buflen; |
a43f4461 | 1048 | const enum fio_ddir __ddir = acct_ddir(io_u); |
0c5df5f9 | 1049 | |
a9da8ab2 JA |
1050 | if (td->error) |
1051 | break; | |
0c5df5f9 | 1052 | |
26de50cf JA |
1053 | workqueue_enqueue(&td->io_wq, &io_u->work); |
1054 | ret = FIO_Q_QUEUED; | |
50a8ce86 | 1055 | |
a43f4461 JA |
1056 | if (ddir_rw(__ddir)) { |
1057 | td->io_issues[__ddir]++; | |
1058 | td->io_issue_bytes[__ddir] += blen; | |
1059 | td->rate_io_issue_bytes[__ddir] += blen; | |
0c5df5f9 JA |
1060 | } |
1061 | ||
50a8ce86 | 1062 | if (should_check_rate(td)) |
a43f4461 | 1063 | td->rate_next_io_time[__ddir] = usec_for_io(td, __ddir); |
50a8ce86 | 1064 | |
a9da8ab2 | 1065 | } else { |
08b2b386 | 1066 | ret = io_u_submit(td, io_u); |
2e1df07d | 1067 | |
50a8ce86 D |
1068 | if (should_check_rate(td)) |
1069 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
1070 | ||
fd727d9d | 1071 | if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time)) |
a9da8ab2 | 1072 | break; |
2e1df07d | 1073 | |
a9da8ab2 JA |
1074 | /* |
1075 | * See if we need to complete some commands. Note that | |
1076 | * we can get BUSY even without IO queued, if the | |
1077 | * system is resource starved. | |
1078 | */ | |
3e260a46 | 1079 | reap: |
a9da8ab2 JA |
1080 | full = queue_full(td) || |
1081 | (ret == FIO_Q_BUSY && td->cur_depth); | |
82407585 | 1082 | if (full || io_in_polling(td)) |
a9da8ab2 JA |
1083 | ret = wait_for_completions(td, &comp_time); |
1084 | } | |
2e1df07d JA |
1085 | if (ret < 0) |
1086 | break; | |
55312f9f | 1087 | if (!ddir_rw_sum(td->bytes_done) && |
9b87f09b | 1088 | !td_ioengine_flagged(td, FIO_NOIO)) |
2e1df07d JA |
1089 | continue; |
1090 | ||
55312f9f JA |
1091 | if (!in_ramp_time(td) && should_check_rate(td)) { |
1092 | if (check_min_rate(td, &comp_time)) { | |
f9cafb12 | 1093 | if (exitall_on_terminate || td->o.exitall_error) |
64402a8a | 1094 | fio_terminate_threads(td->groupid, td->o.exit_what); |
2e1df07d JA |
1095 | td_verror(td, EIO, "check_min_rate"); |
1096 | break; | |
1097 | } | |
1098 | } | |
3e260a46 JA |
1099 | if (!in_ramp_time(td) && td->o.latency_target) |
1100 | lat_target_check(td); | |
e155cb64 | 1101 | |
1a9bf814 JA |
1102 | if (ddir_rw(ddir) && td->o.thinktime) |
1103 | handle_thinktime(td, ddir); | |
2e1df07d JA |
1104 | } |
1105 | ||
c97f1ad6 JA |
1106 | check_update_rusage(td); |
1107 | ||
2e1df07d | 1108 | if (td->trim_entries) |
4e0a8fa2 | 1109 | log_err("fio: %lu trim entries leaked?\n", td->trim_entries); |
2e1df07d JA |
1110 | |
1111 | if (td->o.fill_device && td->error == ENOSPC) { | |
1112 | td->error = 0; | |
ebea2133 | 1113 | fio_mark_td_terminate(td); |
2e1df07d JA |
1114 | } |
1115 | if (!td->error) { | |
1116 | struct fio_file *f; | |
1117 | ||
a9da8ab2 JA |
1118 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
1119 | workqueue_flush(&td->io_wq); | |
1120 | i = 0; | |
1121 | } else | |
1122 | i = td->cur_depth; | |
1123 | ||
2e1df07d | 1124 | if (i) { |
55312f9f | 1125 | ret = io_u_queued_complete(td, i); |
2e1df07d JA |
1126 | if (td->o.fill_device && td->error == ENOSPC) |
1127 | td->error = 0; | |
1128 | } | |
1129 | ||
00615bfb | 1130 | if (should_fsync(td) && (td->o.end_fsync || td->o.fsync_on_close)) { |
2e1df07d JA |
1131 | td_set_runstate(td, TD_FSYNCING); |
1132 | ||
1133 | for_each_file(td, f, i) { | |
61ee0f86 | 1134 | if (!fio_file_fsync(td, f)) |
2e1df07d | 1135 | continue; |
61ee0f86 JA |
1136 | |
1137 | log_err("fio: end_fsync failed for file %s\n", | |
1138 | f->file_name); | |
2e1df07d JA |
1139 | } |
1140 | } | |
1141 | } else | |
1142 | cleanup_pending_aio(td); | |
1143 | ||
1144 | /* | |
1145 | * stop job if we failed doing any IO | |
1146 | */ | |
342f4be4 | 1147 | if (!ddir_rw_sum(td->this_io_bytes)) |
2e1df07d | 1148 | td->done = 1; |
100f49f1 | 1149 | |
95b03be0 JA |
1150 | for (i = 0; i < DDIR_RWDIR_CNT; i++) |
1151 | bytes_done[i] = td->bytes_done[i] - bytes_done[i]; | |
2e1df07d JA |
1152 | } |
1153 | ||
94a6e1bb JA |
1154 | static void free_file_completion_logging(struct thread_data *td) |
1155 | { | |
1156 | struct fio_file *f; | |
1157 | unsigned int i; | |
1158 | ||
1159 | for_each_file(td, f, i) { | |
1160 | if (!f->last_write_comp) | |
1161 | break; | |
1162 | sfree(f->last_write_comp); | |
1163 | } | |
1164 | } | |
1165 | ||
1166 | static int init_file_completion_logging(struct thread_data *td, | |
1167 | unsigned int depth) | |
1168 | { | |
1169 | struct fio_file *f; | |
1170 | unsigned int i; | |
1171 | ||
1172 | if (td->o.verify == VERIFY_NONE || !td->o.verify_state_save) | |
1173 | return 0; | |
1174 | ||
1175 | for_each_file(td, f, i) { | |
1176 | f->last_write_comp = scalloc(depth, sizeof(uint64_t)); | |
1177 | if (!f->last_write_comp) | |
1178 | goto cleanup; | |
1179 | } | |
1180 | ||
1181 | return 0; | |
1182 | ||
1183 | cleanup: | |
1184 | free_file_completion_logging(td); | |
1185 | log_err("fio: failed to alloc write comp data\n"); | |
1186 | return 1; | |
1187 | } | |
1188 | ||
2e1df07d JA |
1189 | static void cleanup_io_u(struct thread_data *td) |
1190 | { | |
2e1df07d JA |
1191 | struct io_u *io_u; |
1192 | ||
2ae0b204 | 1193 | while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) { |
c73ed246 JA |
1194 | |
1195 | if (td->io_ops->io_u_free) | |
1196 | td->io_ops->io_u_free(td, io_u); | |
1197 | ||
3114b675 | 1198 | fio_memfree(io_u, sizeof(*io_u), td_offload_overlap(td)); |
2e1df07d JA |
1199 | } |
1200 | ||
1201 | free_io_mem(td); | |
2ae0b204 JA |
1202 | |
1203 | io_u_rexit(&td->io_u_requeues); | |
3114b675 VF |
1204 | io_u_qexit(&td->io_u_freelist, false); |
1205 | io_u_qexit(&td->io_u_all, td_offload_overlap(td)); | |
ca09be4b | 1206 | |
94a6e1bb | 1207 | free_file_completion_logging(td); |
2e1df07d JA |
1208 | } |
1209 | ||
1210 | static int init_io_u(struct thread_data *td) | |
1211 | { | |
1212 | struct io_u *io_u; | |
2e1df07d | 1213 | int cl_align, i, max_units; |
71e6e5a2 | 1214 | int err; |
2e1df07d JA |
1215 | |
1216 | max_units = td->o.iodepth; | |
59d8d0f5 | 1217 | |
2ae0b204 | 1218 | err = 0; |
34851ad5 | 1219 | err += !io_u_rinit(&td->io_u_requeues, td->o.iodepth); |
3114b675 VF |
1220 | err += !io_u_qinit(&td->io_u_freelist, td->o.iodepth, false); |
1221 | err += !io_u_qinit(&td->io_u_all, td->o.iodepth, td_offload_overlap(td)); | |
2ae0b204 JA |
1222 | |
1223 | if (err) { | |
1224 | log_err("fio: failed setting up IO queues\n"); | |
1225 | return 1; | |
1226 | } | |
1227 | ||
71e6e5a2 AK |
1228 | cl_align = os_cache_line_size(); |
1229 | ||
1230 | for (i = 0; i < max_units; i++) { | |
1231 | void *ptr; | |
1232 | ||
1233 | if (td->terminate) | |
1234 | return 1; | |
1235 | ||
3114b675 | 1236 | ptr = fio_memalign(cl_align, sizeof(*io_u), td_offload_overlap(td)); |
71e6e5a2 AK |
1237 | if (!ptr) { |
1238 | log_err("fio: unable to allocate aligned memory\n"); | |
ab43b8cb | 1239 | return 1; |
71e6e5a2 AK |
1240 | } |
1241 | ||
1242 | io_u = ptr; | |
1243 | memset(io_u, 0, sizeof(*io_u)); | |
1244 | INIT_FLIST_HEAD(&io_u->verify_list); | |
1245 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); | |
1246 | ||
1247 | io_u->index = i; | |
1248 | io_u->flags = IO_U_F_FREE; | |
1249 | io_u_qpush(&td->io_u_freelist, io_u); | |
1250 | ||
1251 | /* | |
1252 | * io_u never leaves this stack, used for iteration of all | |
1253 | * io_u buffers. | |
1254 | */ | |
1255 | io_u_qpush(&td->io_u_all, io_u); | |
1256 | ||
1257 | if (td->io_ops->io_u_init) { | |
1258 | int ret = td->io_ops->io_u_init(td, io_u); | |
1259 | ||
1260 | if (ret) { | |
1261 | log_err("fio: failed to init engine data: %d\n", ret); | |
1262 | return 1; | |
1263 | } | |
1264 | } | |
1265 | } | |
1266 | ||
1267 | init_io_u_buffers(td); | |
1268 | ||
1269 | if (init_file_completion_logging(td, max_units)) | |
1270 | return 1; | |
1271 | ||
1272 | return 0; | |
1273 | } | |
1274 | ||
1275 | int init_io_u_buffers(struct thread_data *td) | |
1276 | { | |
1277 | struct io_u *io_u; | |
1278 | unsigned long long max_bs, min_write; | |
1279 | int i, max_units; | |
1280 | int data_xfer = 1; | |
1281 | char *p; | |
1282 | ||
1283 | max_units = td->o.iodepth; | |
1284 | max_bs = td_max_bs(td); | |
1285 | min_write = td->o.min_bs[DDIR_WRITE]; | |
1286 | td->orig_buffer_size = (unsigned long long) max_bs | |
1287 | * (unsigned long long) max_units; | |
1288 | ||
1289 | if (td_ioengine_flagged(td, FIO_NOIO) || !(td_read(td) || td_write(td))) | |
1290 | data_xfer = 0; | |
1291 | ||
fd8a09b8 | 1292 | /* |
1293 | * if we may later need to do address alignment, then add any | |
1294 | * possible adjustment here so that we don't cause a buffer | |
1295 | * overflow later. this adjustment may be too much if we get | |
1296 | * lucky and the allocator gives us an aligned address. | |
1297 | */ | |
d01612f3 | 1298 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
9b87f09b | 1299 | td_ioengine_flagged(td, FIO_RAWIO)) |
fd8a09b8 | 1300 | td->orig_buffer_size += page_mask + td->o.mem_align; |
1301 | ||
2e1df07d | 1302 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { |
5fff9543 | 1303 | unsigned long long bs; |
2e1df07d JA |
1304 | |
1305 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; | |
1306 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
1307 | } | |
1308 | ||
1309 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { | |
1310 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
1311 | return 1; | |
1312 | } | |
1313 | ||
59d8d0f5 | 1314 | if (data_xfer && allocate_io_mem(td)) |
2e1df07d JA |
1315 | return 1; |
1316 | ||
d01612f3 | 1317 | if (td->o.odirect || td->o.mem_align || td->o.oatomic || |
9b87f09b | 1318 | td_ioengine_flagged(td, FIO_RAWIO)) |
248c9436 | 1319 | p = PTR_ALIGN(td->orig_buffer, page_mask) + td->o.mem_align; |
2e1df07d JA |
1320 | else |
1321 | p = td->orig_buffer; | |
1322 | ||
2e1df07d | 1323 | for (i = 0; i < max_units; i++) { |
71e6e5a2 | 1324 | io_u = td->io_u_all.io_us[i]; |
2e1df07d JA |
1325 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
1326 | ||
59d8d0f5 | 1327 | if (data_xfer) { |
2e1df07d JA |
1328 | io_u->buf = p; |
1329 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); | |
1330 | ||
1331 | if (td_write(td)) | |
9c42684e | 1332 | io_u_fill_buffer(td, io_u, min_write, max_bs); |
2e1df07d JA |
1333 | if (td_write(td) && td->o.verify_pattern_bytes) { |
1334 | /* | |
1335 | * Fill the buffer with the pattern if we are | |
1336 | * going to be doing writes. | |
1337 | */ | |
ce35b1ec | 1338 | fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
2e1df07d JA |
1339 | } |
1340 | } | |
2e1df07d JA |
1341 | p += max_bs; |
1342 | } | |
1343 | ||
1344 | return 0; | |
1345 | } | |
1346 | ||
bd4d9bdc TK |
1347 | /* |
1348 | * This function is Linux specific. | |
1349 | * FIO_HAVE_IOSCHED_SWITCH enabled currently means it's Linux. | |
1350 | */ | |
2e1df07d JA |
1351 | static int switch_ioscheduler(struct thread_data *td) |
1352 | { | |
db815fb3 | 1353 | #ifdef FIO_HAVE_IOSCHED_SWITCH |
178ef27a | 1354 | char tmp[256], tmp2[128], *p; |
2e1df07d JA |
1355 | FILE *f; |
1356 | int ret; | |
1357 | ||
9b87f09b | 1358 | if (td_ioengine_flagged(td, FIO_DISKLESSIO)) |
2e1df07d JA |
1359 | return 0; |
1360 | ||
bd4d9bdc TK |
1361 | assert(td->files && td->files[0]); |
1362 | sprintf(tmp, "%s/queue/scheduler", td->files[0]->du->sysfs_root); | |
2e1df07d JA |
1363 | |
1364 | f = fopen(tmp, "r+"); | |
1365 | if (!f) { | |
1366 | if (errno == ENOENT) { | |
1367 | log_err("fio: os or kernel doesn't support IO scheduler" | |
1368 | " switching\n"); | |
1369 | return 0; | |
1370 | } | |
1371 | td_verror(td, errno, "fopen iosched"); | |
1372 | return 1; | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * Set io scheduler. | |
1377 | */ | |
1378 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); | |
1379 | if (ferror(f) || ret != 1) { | |
1380 | td_verror(td, errno, "fwrite"); | |
1381 | fclose(f); | |
1382 | return 1; | |
1383 | } | |
1384 | ||
1385 | rewind(f); | |
1386 | ||
1387 | /* | |
1388 | * Read back and check that the selected scheduler is now the default. | |
1389 | */ | |
0e23d456 | 1390 | ret = fread(tmp, 1, sizeof(tmp) - 1, f); |
2e1df07d JA |
1391 | if (ferror(f) || ret < 0) { |
1392 | td_verror(td, errno, "fread"); | |
1393 | fclose(f); | |
1394 | return 1; | |
1395 | } | |
0e23d456 | 1396 | tmp[ret] = '\0'; |
b44b9e45 | 1397 | /* |
178ef27a BVA |
1398 | * either a list of io schedulers or "none\n" is expected. Strip the |
1399 | * trailing newline. | |
b44b9e45 | 1400 | */ |
178ef27a BVA |
1401 | p = tmp; |
1402 | strsep(&p, "\n"); | |
49c6f33d | 1403 | |
67f28a8c TK |
1404 | /* |
1405 | * Write to "none" entry doesn't fail, so check the result here. | |
1406 | */ | |
1407 | if (!strcmp(tmp, "none")) { | |
1408 | log_err("fio: io scheduler is not tunable\n"); | |
1409 | fclose(f); | |
1410 | return 0; | |
1411 | } | |
2e1df07d JA |
1412 | |
1413 | sprintf(tmp2, "[%s]", td->o.ioscheduler); | |
1414 | if (!strstr(tmp, tmp2)) { | |
1415 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); | |
1416 | td_verror(td, EINVAL, "iosched_switch"); | |
1417 | fclose(f); | |
1418 | return 1; | |
1419 | } | |
1420 | ||
1421 | fclose(f); | |
1422 | return 0; | |
db815fb3 TK |
1423 | #else |
1424 | return 0; | |
1425 | #endif | |
2e1df07d JA |
1426 | } |
1427 | ||
e39c0676 | 1428 | static bool keep_running(struct thread_data *td) |
2e1df07d | 1429 | { |
77731b29 JA |
1430 | unsigned long long limit; |
1431 | ||
2e1df07d | 1432 | if (td->done) |
e39c0676 | 1433 | return false; |
98413be7 SW |
1434 | if (td->terminate) |
1435 | return false; | |
2e1df07d | 1436 | if (td->o.time_based) |
e39c0676 | 1437 | return true; |
2e1df07d JA |
1438 | if (td->o.loops) { |
1439 | td->o.loops--; | |
e39c0676 | 1440 | return true; |
2e1df07d | 1441 | } |
3939fe85 | 1442 | if (exceeds_number_ios(td)) |
e39c0676 | 1443 | return false; |
26251d8d | 1444 | |
5be9bf09 TK |
1445 | if (td->o.io_size) |
1446 | limit = td->o.io_size; | |
77731b29 JA |
1447 | else |
1448 | limit = td->o.size; | |
1449 | ||
1450 | if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) { | |
5bd5f71a JA |
1451 | uint64_t diff; |
1452 | ||
1453 | /* | |
589d5d4f | 1454 | * If the difference is less than the maximum IO size, we |
5bd5f71a JA |
1455 | * are done. |
1456 | */ | |
77731b29 | 1457 | diff = limit - ddir_rw_sum(td->io_bytes); |
74f4b020 | 1458 | if (diff < td_max_bs(td)) |
e39c0676 | 1459 | return false; |
5bd5f71a | 1460 | |
5be9bf09 | 1461 | if (fio_files_done(td) && !td->o.io_size) |
e39c0676 | 1462 | return false; |
002fe734 | 1463 | |
e39c0676 | 1464 | return true; |
5bd5f71a | 1465 | } |
2e1df07d | 1466 | |
e39c0676 | 1467 | return false; |
2e1df07d JA |
1468 | } |
1469 | ||
ce486495 | 1470 | static int exec_string(struct thread_options *o, const char *string, const char *mode) |
2e1df07d | 1471 | { |
321d0c07 | 1472 | size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 13 + 1; |
f491a907 | 1473 | int ret; |
2e1df07d JA |
1474 | char *str; |
1475 | ||
1476 | str = malloc(newlen); | |
321d0c07 | 1477 | sprintf(str, "%s > %s.%s.txt 2>&1", string, o->name, mode); |
2e1df07d | 1478 | |
ce486495 | 1479 | log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode); |
2e1df07d JA |
1480 | ret = system(str); |
1481 | if (ret == -1) | |
1482 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1483 | ||
1484 | free(str); | |
1485 | return ret; | |
1486 | } | |
1487 | ||
62167762 JC |
1488 | /* |
1489 | * Dry run to compute correct state of numberio for verification. | |
1490 | */ | |
1491 | static uint64_t do_dry_run(struct thread_data *td) | |
1492 | { | |
62167762 JC |
1493 | td_set_runstate(td, TD_RUNNING); |
1494 | ||
1495 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || | |
e28dd2cf | 1496 | (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) { |
62167762 JC |
1497 | struct io_u *io_u; |
1498 | int ret; | |
1499 | ||
1500 | if (td->terminate || td->done) | |
1501 | break; | |
1502 | ||
1503 | io_u = get_io_u(td); | |
99955d3d | 1504 | if (IS_ERR_OR_NULL(io_u)) |
62167762 JC |
1505 | break; |
1506 | ||
1651e431 | 1507 | io_u_set(td, io_u, IO_U_F_FLIGHT); |
62167762 JC |
1508 | io_u->error = 0; |
1509 | io_u->resid = 0; | |
1510 | if (ddir_rw(acct_ddir(io_u))) | |
1511 | td->io_issues[acct_ddir(io_u)]++; | |
1512 | if (ddir_rw(io_u->ddir)) { | |
1513 | io_u_mark_depth(td, 1); | |
1514 | td->ts.total_io_u[io_u->ddir]++; | |
1515 | } | |
1516 | ||
2e63e96b PV |
1517 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1518 | td->o.do_verify && | |
1519 | td->o.verify != VERIFY_NONE && | |
1520 | !td->o.experimental_verify) | |
1521 | log_io_piece(td, io_u); | |
1522 | ||
55312f9f | 1523 | ret = io_u_sync_complete(td, io_u); |
62167762 JC |
1524 | (void) ret; |
1525 | } | |
1526 | ||
55312f9f | 1527 | return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; |
62167762 JC |
1528 | } |
1529 | ||
24660963 JA |
1530 | struct fork_data { |
1531 | struct thread_data *td; | |
1532 | struct sk_out *sk_out; | |
1533 | }; | |
1534 | ||
2e1df07d JA |
1535 | /* |
1536 | * Entry point for the thread based jobs. The process based jobs end up | |
1537 | * here as well, after a little setup. | |
1538 | */ | |
1539 | static void *thread_main(void *data) | |
1540 | { | |
24660963 | 1541 | struct fork_data *fd = data; |
95603b74 | 1542 | unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, }; |
24660963 | 1543 | struct thread_data *td = fd->td; |
4896473e | 1544 | struct thread_options *o = &td->o; |
24660963 | 1545 | struct sk_out *sk_out = fd->sk_out; |
48366f3a | 1546 | uint64_t bytes_done[DDIR_RWDIR_CNT]; |
ca0122d8 | 1547 | int deadlock_loop_cnt; |
ff3aa922 | 1548 | bool clear_state; |
28727df7 | 1549 | int ret; |
2e1df07d | 1550 | |
24660963 JA |
1551 | sk_out_assign(sk_out); |
1552 | free(fd); | |
1553 | ||
4896473e | 1554 | if (!o->use_thread) { |
2e1df07d JA |
1555 | setsid(); |
1556 | td->pid = getpid(); | |
1557 | } else | |
1558 | td->pid = gettid(); | |
1559 | ||
a0eba820 | 1560 | fio_local_clock_init(); |
5d879392 | 1561 | |
2e1df07d JA |
1562 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
1563 | ||
122c7725 JA |
1564 | if (is_backend) |
1565 | fio_server_send_start(td); | |
1566 | ||
2e1df07d JA |
1567 | INIT_FLIST_HEAD(&td->io_log_list); |
1568 | INIT_FLIST_HEAD(&td->io_hist_list); | |
1569 | INIT_FLIST_HEAD(&td->verify_list); | |
1570 | INIT_FLIST_HEAD(&td->trim_list); | |
2e1df07d JA |
1571 | td->io_hist_tree = RB_ROOT; |
1572 | ||
34febb23 | 1573 | ret = mutex_cond_init_pshared(&td->io_u_lock, &td->free_cond); |
f9e5b5ee | 1574 | if (ret) { |
34febb23 | 1575 | td_verror(td, ret, "mutex_cond_init_pshared"); |
f9e5b5ee JK |
1576 | goto err; |
1577 | } | |
34febb23 | 1578 | ret = cond_init_pshared(&td->verify_cond); |
f9e5b5ee | 1579 | if (ret) { |
34febb23 | 1580 | td_verror(td, ret, "mutex_cond_pshared"); |
f9e5b5ee JK |
1581 | goto err; |
1582 | } | |
2e1df07d JA |
1583 | |
1584 | td_set_runstate(td, TD_INITIALIZED); | |
971caeb1 BVA |
1585 | dprint(FD_MUTEX, "up startup_sem\n"); |
1586 | fio_sem_up(startup_sem); | |
1587 | dprint(FD_MUTEX, "wait on td->sem\n"); | |
1588 | fio_sem_down(td->sem); | |
1589 | dprint(FD_MUTEX, "done waiting on td->sem\n"); | |
2e1df07d | 1590 | |
2e1df07d JA |
1591 | /* |
1592 | * A new gid requires privilege, so we need to do this before setting | |
1593 | * the uid. | |
1594 | */ | |
4896473e | 1595 | if (o->gid != -1U && setgid(o->gid)) { |
2e1df07d JA |
1596 | td_verror(td, errno, "setgid"); |
1597 | goto err; | |
1598 | } | |
4896473e | 1599 | if (o->uid != -1U && setuid(o->uid)) { |
2e1df07d JA |
1600 | td_verror(td, errno, "setuid"); |
1601 | goto err; | |
1602 | } | |
1603 | ||
07776127 JA |
1604 | td_zone_gen_index(td); |
1605 | ||
2354d810 JA |
1606 | /* |
1607 | * Do this early, we don't want the compress threads to be limited | |
1608 | * to the same CPUs as the IO workers. So do this before we set | |
1609 | * any potential CPU affinity | |
1610 | */ | |
1611 | if (iolog_compress_init(td, sk_out)) | |
1612 | goto err; | |
1613 | ||
2e1df07d JA |
1614 | /* |
1615 | * If we have a gettimeofday() thread, make sure we exclude that | |
1616 | * thread from this job | |
1617 | */ | |
4896473e JA |
1618 | if (o->gtod_cpu) |
1619 | fio_cpu_clear(&o->cpumask, o->gtod_cpu); | |
2e1df07d JA |
1620 | |
1621 | /* | |
1622 | * Set affinity first, in case it has an impact on the memory | |
1623 | * allocations. | |
1624 | */ | |
b2a9e649 | 1625 | if (fio_option_is_set(o, cpumask)) { |
c2acfbac | 1626 | if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { |
30cb4c65 | 1627 | ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); |
c2acfbac JA |
1628 | if (!ret) { |
1629 | log_err("fio: no CPUs set\n"); | |
1630 | log_err("fio: Try increasing number of available CPUs\n"); | |
1631 | td_verror(td, EINVAL, "cpus_split"); | |
1632 | goto err; | |
1633 | } | |
1634 | } | |
28727df7 JA |
1635 | ret = fio_setaffinity(td->pid, o->cpumask); |
1636 | if (ret == -1) { | |
4896473e JA |
1637 | td_verror(td, errno, "cpu_set_affinity"); |
1638 | goto err; | |
1639 | } | |
2e1df07d JA |
1640 | } |
1641 | ||
67bf9823 | 1642 | #ifdef CONFIG_LIBNUMA |
d0b937ed | 1643 | /* numa node setup */ |
b2a9e649 JA |
1644 | if (fio_option_is_set(o, numa_cpunodes) || |
1645 | fio_option_is_set(o, numa_memnodes)) { | |
43522848 | 1646 | struct bitmask *mask; |
d0b937ed YR |
1647 | |
1648 | if (numa_available() < 0) { | |
1649 | td_verror(td, errno, "Does not support NUMA API\n"); | |
1650 | goto err; | |
1651 | } | |
1652 | ||
b2a9e649 | 1653 | if (fio_option_is_set(o, numa_cpunodes)) { |
43522848 DG |
1654 | mask = numa_parse_nodestring(o->numa_cpunodes); |
1655 | ret = numa_run_on_node_mask(mask); | |
1656 | numa_free_nodemask(mask); | |
d0b937ed YR |
1657 | if (ret == -1) { |
1658 | td_verror(td, errno, \ | |
1659 | "numa_run_on_node_mask failed\n"); | |
1660 | goto err; | |
1661 | } | |
1662 | } | |
1663 | ||
b2a9e649 | 1664 | if (fio_option_is_set(o, numa_memnodes)) { |
43522848 DG |
1665 | mask = NULL; |
1666 | if (o->numa_memnodes) | |
1667 | mask = numa_parse_nodestring(o->numa_memnodes); | |
1668 | ||
4896473e | 1669 | switch (o->numa_mem_mode) { |
d0b937ed | 1670 | case MPOL_INTERLEAVE: |
43522848 | 1671 | numa_set_interleave_mask(mask); |
d0b937ed YR |
1672 | break; |
1673 | case MPOL_BIND: | |
43522848 | 1674 | numa_set_membind(mask); |
d0b937ed YR |
1675 | break; |
1676 | case MPOL_LOCAL: | |
1677 | numa_set_localalloc(); | |
1678 | break; | |
1679 | case MPOL_PREFERRED: | |
4896473e | 1680 | numa_set_preferred(o->numa_mem_prefer_node); |
d0b937ed YR |
1681 | break; |
1682 | case MPOL_DEFAULT: | |
1683 | default: | |
1684 | break; | |
1685 | } | |
1686 | ||
43522848 DG |
1687 | if (mask) |
1688 | numa_free_nodemask(mask); | |
1689 | ||
d0b937ed YR |
1690 | } |
1691 | } | |
1692 | #endif | |
1693 | ||
9a3f1100 JA |
1694 | if (fio_pin_memory(td)) |
1695 | goto err; | |
1696 | ||
2e1df07d JA |
1697 | /* |
1698 | * May alter parameters that init_io_u() will use, so we need to | |
1699 | * do this first. | |
1700 | */ | |
b153f94a | 1701 | if (!init_iolog(td)) |
2e1df07d JA |
1702 | goto err; |
1703 | ||
08dc3bd5 JA |
1704 | if (td_io_init(td)) |
1705 | goto err; | |
1706 | ||
2e1df07d JA |
1707 | if (init_io_u(td)) |
1708 | goto err; | |
1709 | ||
2041bd34 JA |
1710 | if (td->io_ops->post_init && td->io_ops->post_init(td)) |
1711 | goto err; | |
1712 | ||
4896473e | 1713 | if (o->verify_async && verify_async_init(td)) |
2e1df07d JA |
1714 | goto err; |
1715 | ||
27d74836 JA |
1716 | if (fio_option_is_set(o, ioprio) || |
1717 | fio_option_is_set(o, ioprio_class)) { | |
28727df7 JA |
1718 | ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio); |
1719 | if (ret == -1) { | |
2e1df07d JA |
1720 | td_verror(td, errno, "ioprio_set"); |
1721 | goto err; | |
1722 | } | |
1723 | } | |
1724 | ||
4896473e | 1725 | if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
2e1df07d JA |
1726 | goto err; |
1727 | ||
649c10c5 | 1728 | errno = 0; |
4896473e | 1729 | if (nice(o->nice) == -1 && errno != 0) { |
2e1df07d JA |
1730 | td_verror(td, errno, "nice"); |
1731 | goto err; | |
1732 | } | |
1733 | ||
4896473e | 1734 | if (o->ioscheduler && switch_ioscheduler(td)) |
2e1df07d JA |
1735 | goto err; |
1736 | ||
4896473e | 1737 | if (!o->create_serialize && setup_files(td)) |
2e1df07d JA |
1738 | goto err; |
1739 | ||
4d832322 | 1740 | if (!init_random_map(td)) |
2e1df07d JA |
1741 | goto err; |
1742 | ||
ce486495 | 1743 | if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun")) |
4896473e | 1744 | goto err; |
2e1df07d | 1745 | |
c139f3b4 JA |
1746 | if (o->pre_read && !pre_read_files(td)) |
1747 | goto err; | |
2e1df07d | 1748 | |
dc5bfbb2 JA |
1749 | fio_verify_init(td); |
1750 | ||
24660963 | 1751 | if (rate_submit_init(td, sk_out)) |
a9da8ab2 JA |
1752 | goto err; |
1753 | ||
3aea75b1 | 1754 | set_epoch_time(td, o->log_unix_epoch); |
44404c5a | 1755 | fio_getrusage(&td->ru_start); |
ac28d905 JA |
1756 | memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch)); |
1757 | memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch)); | |
16e56d25 | 1758 | memcpy(&td->ss.prev_time, &td->epoch, sizeof(td->epoch)); |
ac28d905 JA |
1759 | |
1760 | if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || | |
1761 | o->ratemin[DDIR_TRIM]) { | |
1762 | memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time, | |
1763 | sizeof(td->bw_sample_time)); | |
1764 | memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time, | |
1765 | sizeof(td->bw_sample_time)); | |
1766 | memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time, | |
1767 | sizeof(td->bw_sample_time)); | |
1768 | } | |
1769 | ||
48366f3a | 1770 | memset(bytes_done, 0, sizeof(bytes_done)); |
c16035aa | 1771 | clear_state = false; |
48366f3a | 1772 | |
2e1df07d | 1773 | while (keep_running(td)) { |
100f49f1 JA |
1774 | uint64_t verify_bytes; |
1775 | ||
2e1df07d | 1776 | fio_gettime(&td->start, NULL); |
8b6a404c | 1777 | memcpy(&td->ts_cache, &td->start, sizeof(td->start)); |
2e1df07d | 1778 | |
39c1c323 | 1779 | if (clear_state) { |
ac28d905 | 1780 | clear_io_state(td, 0); |
2e1df07d | 1781 | |
39c1c323 | 1782 | if (o->unlink_each_loop && unlink_all_files(td)) |
1783 | break; | |
1784 | } | |
1785 | ||
2e1df07d JA |
1786 | prune_io_piece_log(td); |
1787 | ||
dcf9844e | 1788 | if (td->o.verify_only && td_write(td)) |
62167762 | 1789 | verify_bytes = do_dry_run(td); |
095196b1 | 1790 | else { |
95b03be0 JA |
1791 | do_io(td, bytes_done); |
1792 | ||
1793 | if (!ddir_rw_sum(bytes_done)) { | |
095196b1 | 1794 | fio_mark_td_terminate(td); |
95b03be0 JA |
1795 | verify_bytes = 0; |
1796 | } else { | |
1797 | verify_bytes = bytes_done[DDIR_WRITE] + | |
1798 | bytes_done[DDIR_TRIM]; | |
1799 | } | |
095196b1 | 1800 | } |
2e1df07d | 1801 | |
49f0b8e0 JA |
1802 | /* |
1803 | * If we took too long to shut down, the main thread could | |
1804 | * already consider us reaped/exited. If that happens, break | |
1805 | * out and clean up. | |
1806 | */ | |
1807 | if (td->runstate >= TD_EXITED) | |
1808 | break; | |
1809 | ||
c16035aa | 1810 | clear_state = true; |
2e1df07d | 1811 | |
40c666c8 JA |
1812 | /* |
1813 | * Make sure we've successfully updated the rusage stats | |
1814 | * before waiting on the stat mutex. Otherwise we could have | |
1815 | * the stat thread holding stat mutex and waiting for | |
1816 | * the rusage_sem, which would never get upped because | |
1817 | * this thread is waiting for the stat mutex. | |
1818 | */ | |
ca0122d8 | 1819 | deadlock_loop_cnt = 0; |
39d13e67 JA |
1820 | do { |
1821 | check_update_rusage(td); | |
971caeb1 | 1822 | if (!fio_sem_down_trylock(stat_sem)) |
39d13e67 JA |
1823 | break; |
1824 | usleep(1000); | |
ca0122d8 | 1825 | if (deadlock_loop_cnt++ > 5000) { |
971caeb1 | 1826 | log_err("fio seems to be stuck grabbing stat_sem, forcibly exiting\n"); |
43f248c4 | 1827 | td->error = EDEADLK; |
ca0122d8 TT |
1828 | goto err; |
1829 | } | |
39d13e67 | 1830 | } while (1); |
40c666c8 | 1831 | |
95603b74 BF |
1832 | if (td_read(td) && td->io_bytes[DDIR_READ]) |
1833 | update_runtime(td, elapsed_us, DDIR_READ); | |
1834 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) | |
1835 | update_runtime(td, elapsed_us, DDIR_WRITE); | |
1836 | if (td_trim(td) && td->io_bytes[DDIR_TRIM]) | |
1837 | update_runtime(td, elapsed_us, DDIR_TRIM); | |
e5437a07 | 1838 | fio_gettime(&td->start, NULL); |
971caeb1 | 1839 | fio_sem_up(stat_sem); |
2e1df07d JA |
1840 | |
1841 | if (td->error || td->terminate) | |
1842 | break; | |
1843 | ||
4896473e JA |
1844 | if (!o->do_verify || |
1845 | o->verify == VERIFY_NONE || | |
9b87f09b | 1846 | td_ioengine_flagged(td, FIO_UNIDIR)) |
2e1df07d JA |
1847 | continue; |
1848 | ||
ac28d905 | 1849 | clear_io_state(td, 0); |
2e1df07d JA |
1850 | |
1851 | fio_gettime(&td->start, NULL); | |
1852 | ||
100f49f1 | 1853 | do_verify(td, verify_bytes); |
2e1df07d | 1854 | |
40c666c8 JA |
1855 | /* |
1856 | * See comment further up for why this is done here. | |
1857 | */ | |
1858 | check_update_rusage(td); | |
1859 | ||
971caeb1 | 1860 | fio_sem_down(stat_sem); |
95603b74 | 1861 | update_runtime(td, elapsed_us, DDIR_READ); |
e5437a07 | 1862 | fio_gettime(&td->start, NULL); |
971caeb1 | 1863 | fio_sem_up(stat_sem); |
2e1df07d JA |
1864 | |
1865 | if (td->error || td->terminate) | |
1866 | break; | |
1867 | } | |
1868 | ||
817ae977 VF |
1869 | /* |
1870 | * Acquire this lock if we were doing overlap checking in | |
1871 | * offload mode so that we don't clean up this job while | |
1872 | * another thread is checking its io_u's for overlap | |
1873 | */ | |
1368d95b | 1874 | if (td_offload_overlap(td)) |
c06379a6 | 1875 | pthread_mutex_lock(&overlap_check); |
2cea0b4c | 1876 | td_set_runstate(td, TD_FINISHING); |
1368d95b | 1877 | if (td_offload_overlap(td)) |
c06379a6 | 1878 | pthread_mutex_unlock(&overlap_check); |
2cea0b4c | 1879 | |
2e1df07d | 1880 | update_rusage_stat(td); |
2e1df07d | 1881 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
6eaf09d6 SL |
1882 | td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ]; |
1883 | td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE]; | |
1884 | td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM]; | |
2e1df07d | 1885 | |
ca09be4b | 1886 | if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && |
d264264a JA |
1887 | (td->o.verify != VERIFY_NONE && td_write(td))) |
1888 | verify_save_state(td->thread_number); | |
ca09be4b | 1889 | |
9a3f1100 JA |
1890 | fio_unpin_memory(td); |
1891 | ||
a47591e4 | 1892 | td_writeout_logs(td, true); |
2e1df07d | 1893 | |
155f2f02 | 1894 | iolog_compress_exit(td); |
103b174e | 1895 | rate_submit_exit(td); |
aee2ab67 | 1896 | |
4896473e | 1897 | if (o->exec_postrun) |
ce486495 | 1898 | exec_string(o, o->exec_postrun, (const char *)"postrun"); |
2e1df07d | 1899 | |
f9cafb12 | 1900 | if (exitall_on_terminate || (o->exitall_error && td->error)) |
64402a8a | 1901 | fio_terminate_threads(td->groupid, td->o.exit_what); |
2e1df07d JA |
1902 | |
1903 | err: | |
1904 | if (td->error) | |
1905 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, | |
1906 | td->verror); | |
1907 | ||
4896473e | 1908 | if (o->verify_async) |
2e1df07d JA |
1909 | verify_async_exit(td); |
1910 | ||
1911 | close_and_free_files(td); | |
2e1df07d | 1912 | cleanup_io_u(td); |
32dbca2c | 1913 | close_ioengine(td); |
c3dc516a | 1914 | cgroup_shutdown(td, cgroup_mnt); |
ca09be4b | 1915 | verify_free_state(td); |
07776127 | 1916 | td_zone_free_index(td); |
7c4d0fb7 | 1917 | |
b2a9e649 | 1918 | if (fio_option_is_set(o, cpumask)) { |
8a1db9a1 JA |
1919 | ret = fio_cpuset_exit(&o->cpumask); |
1920 | if (ret) | |
1921 | td_verror(td, ret, "fio_cpuset_exit"); | |
2e1df07d JA |
1922 | } |
1923 | ||
1924 | /* | |
1925 | * do this very late, it will log file closing as well | |
1926 | */ | |
4896473e | 1927 | if (o->write_iolog_file) |
2e1df07d | 1928 | write_iolog_close(td); |
98e7161c AK |
1929 | if (td->io_log_rfile) |
1930 | fclose(td->io_log_rfile); | |
2e1df07d JA |
1931 | |
1932 | td_set_runstate(td, TD_EXITED); | |
fda2cfac JA |
1933 | |
1934 | /* | |
1935 | * Do this last after setting our runstate to exited, so we | |
1936 | * know that the stat thread is signaled. | |
1937 | */ | |
1938 | check_update_rusage(td); | |
1939 | ||
24660963 | 1940 | sk_out_drop(); |
e43606c2 | 1941 | return (void *) (uintptr_t) td->error; |
2e1df07d JA |
1942 | } |
1943 | ||
2e1df07d JA |
1944 | /* |
1945 | * Run over the job map and reap the threads that have exited, if any. | |
1946 | */ | |
90eff1c9 SW |
1947 | static void reap_threads(unsigned int *nr_running, uint64_t *t_rate, |
1948 | uint64_t *m_rate) | |
2e1df07d JA |
1949 | { |
1950 | struct thread_data *td; | |
1951 | unsigned int cputhreads, realthreads, pending; | |
1952 | int i, status, ret; | |
1953 | ||
1954 | /* | |
1955 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1956 | */ | |
1957 | realthreads = pending = cputhreads = 0; | |
1958 | for_each_td(td, i) { | |
1959 | int flags = 0; | |
1960 | ||
8847ae4c | 1961 | if (!strcmp(td->o.ioengine, "cpuio")) |
2e1df07d JA |
1962 | cputhreads++; |
1963 | else | |
1964 | realthreads++; | |
1965 | ||
1966 | if (!td->pid) { | |
1967 | pending++; | |
1968 | continue; | |
1969 | } | |
1970 | if (td->runstate == TD_REAPED) | |
1971 | continue; | |
1972 | if (td->o.use_thread) { | |
1973 | if (td->runstate == TD_EXITED) { | |
1974 | td_set_runstate(td, TD_REAPED); | |
1975 | goto reaped; | |
1976 | } | |
1977 | continue; | |
1978 | } | |
1979 | ||
1980 | flags = WNOHANG; | |
1981 | if (td->runstate == TD_EXITED) | |
1982 | flags = 0; | |
1983 | ||
1984 | /* | |
1985 | * check if someone quit or got killed in an unusual way | |
1986 | */ | |
1987 | ret = waitpid(td->pid, &status, flags); | |
1988 | if (ret < 0) { | |
1989 | if (errno == ECHILD) { | |
1990 | log_err("fio: pid=%d disappeared %d\n", | |
1991 | (int) td->pid, td->runstate); | |
a5e371a6 | 1992 | td->sig = ECHILD; |
2e1df07d JA |
1993 | td_set_runstate(td, TD_REAPED); |
1994 | goto reaped; | |
1995 | } | |
1996 | perror("waitpid"); | |
1997 | } else if (ret == td->pid) { | |
1998 | if (WIFSIGNALED(status)) { | |
1999 | int sig = WTERMSIG(status); | |
2000 | ||
36d80bc7 | 2001 | if (sig != SIGTERM && sig != SIGUSR2) |
2e1df07d JA |
2002 | log_err("fio: pid=%d, got signal=%d\n", |
2003 | (int) td->pid, sig); | |
a5e371a6 | 2004 | td->sig = sig; |
2e1df07d JA |
2005 | td_set_runstate(td, TD_REAPED); |
2006 | goto reaped; | |
2007 | } | |
2008 | if (WIFEXITED(status)) { | |
2009 | if (WEXITSTATUS(status) && !td->error) | |
2010 | td->error = WEXITSTATUS(status); | |
2011 | ||
2012 | td_set_runstate(td, TD_REAPED); | |
2013 | goto reaped; | |
2014 | } | |
2015 | } | |
2016 | ||
cba5460c JA |
2017 | /* |
2018 | * If the job is stuck, do a forceful timeout of it and | |
2019 | * move on. | |
2020 | */ | |
2021 | if (td->terminate && | |
2c45a4ac | 2022 | td->runstate < TD_FSYNCING && |
cba5460c | 2023 | time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { |
86310a1f TK |
2024 | log_err("fio: job '%s' (state=%d) hasn't exited in " |
2025 | "%lu seconds, it appears to be stuck. Doing " | |
2026 | "forceful exit of this job.\n", | |
2027 | td->o.name, td->runstate, | |
2028 | (unsigned long) time_since_now(&td->terminate_time)); | |
cba5460c JA |
2029 | td_set_runstate(td, TD_REAPED); |
2030 | goto reaped; | |
2031 | } | |
2032 | ||
2e1df07d JA |
2033 | /* |
2034 | * thread is not dead, continue | |
2035 | */ | |
2036 | pending++; | |
2037 | continue; | |
2038 | reaped: | |
2039 | (*nr_running)--; | |
342f4be4 JA |
2040 | (*m_rate) -= ddir_rw_sum(td->o.ratemin); |
2041 | (*t_rate) -= ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
2042 | if (!td->pid) |
2043 | pending--; | |
2044 | ||
2045 | if (td->error) | |
2046 | exit_value++; | |
2047 | ||
2048 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
4a88752a | 2049 | profile_td_exit(td); |
2e1df07d JA |
2050 | } |
2051 | ||
2052 | if (*nr_running == cputhreads && !pending && realthreads) | |
64402a8a | 2053 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
2e1df07d JA |
2054 | } |
2055 | ||
e39c0676 | 2056 | static bool __check_trigger_file(void) |
ca09be4b JA |
2057 | { |
2058 | struct stat sb; | |
2059 | ||
2060 | if (!trigger_file) | |
e39c0676 | 2061 | return false; |
ca09be4b JA |
2062 | |
2063 | if (stat(trigger_file, &sb)) | |
e39c0676 | 2064 | return false; |
ca09be4b JA |
2065 | |
2066 | if (unlink(trigger_file) < 0) | |
2067 | log_err("fio: failed to unlink %s: %s\n", trigger_file, | |
2068 | strerror(errno)); | |
2069 | ||
e39c0676 | 2070 | return true; |
ca09be4b JA |
2071 | } |
2072 | ||
e39c0676 | 2073 | static bool trigger_timedout(void) |
ca09be4b JA |
2074 | { |
2075 | if (trigger_timeout) | |
98413be7 SW |
2076 | if (time_since_genesis() >= trigger_timeout) { |
2077 | trigger_timeout = 0; | |
2078 | return true; | |
2079 | } | |
ca09be4b | 2080 | |
e39c0676 | 2081 | return false; |
ca09be4b JA |
2082 | } |
2083 | ||
2084 | void exec_trigger(const char *cmd) | |
2085 | { | |
2086 | int ret; | |
2087 | ||
98413be7 | 2088 | if (!cmd || cmd[0] == '\0') |
ca09be4b JA |
2089 | return; |
2090 | ||
2091 | ret = system(cmd); | |
2092 | if (ret == -1) | |
2093 | log_err("fio: failed executing %s trigger\n", cmd); | |
2094 | } | |
2095 | ||
2096 | void check_trigger_file(void) | |
2097 | { | |
2098 | if (__check_trigger_file() || trigger_timedout()) { | |
796fb3ce JA |
2099 | if (nr_clients) |
2100 | fio_clients_send_trigger(trigger_remote_cmd); | |
2101 | else { | |
d264264a | 2102 | verify_save_state(IO_LIST_ALL); |
64402a8a | 2103 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
ca09be4b JA |
2104 | exec_trigger(trigger_cmd); |
2105 | } | |
2106 | } | |
2107 | } | |
2108 | ||
2109 | static int fio_verify_load_state(struct thread_data *td) | |
2110 | { | |
2111 | int ret; | |
2112 | ||
2113 | if (!td->o.verify_state) | |
2114 | return 0; | |
2115 | ||
2116 | if (is_backend) { | |
2117 | void *data; | |
2118 | ||
2119 | ret = fio_server_get_verify_state(td->o.name, | |
94a6e1bb | 2120 | td->thread_number - 1, &data); |
ca09be4b | 2121 | if (!ret) |
94a6e1bb | 2122 | verify_assign_state(td, data); |
ca09be4b JA |
2123 | } else |
2124 | ret = verify_load_state(td, "local"); | |
2125 | ||
2126 | return ret; | |
2127 | } | |
2128 | ||
06464907 JA |
2129 | static void do_usleep(unsigned int usecs) |
2130 | { | |
2131 | check_for_running_stats(); | |
ca09be4b | 2132 | check_trigger_file(); |
06464907 JA |
2133 | usleep(usecs); |
2134 | } | |
2135 | ||
e39c0676 | 2136 | static bool check_mount_writes(struct thread_data *td) |
e81ecca3 JA |
2137 | { |
2138 | struct fio_file *f; | |
2139 | unsigned int i; | |
2140 | ||
2141 | if (!td_write(td) || td->o.allow_mounted_write) | |
e39c0676 | 2142 | return false; |
e81ecca3 | 2143 | |
5e81f9c1 TK |
2144 | /* |
2145 | * If FIO_HAVE_CHARDEV_SIZE is defined, it's likely that chrdevs | |
2146 | * are mkfs'd and mounted. | |
2147 | */ | |
e81ecca3 | 2148 | for_each_file(td, f, i) { |
5e81f9c1 TK |
2149 | #ifdef FIO_HAVE_CHARDEV_SIZE |
2150 | if (f->filetype != FIO_TYPE_BLOCK && f->filetype != FIO_TYPE_CHAR) | |
2151 | #else | |
686fbd31 | 2152 | if (f->filetype != FIO_TYPE_BLOCK) |
5e81f9c1 | 2153 | #endif |
e81ecca3 JA |
2154 | continue; |
2155 | if (device_is_mounted(f->file_name)) | |
2156 | goto mounted; | |
2157 | } | |
2158 | ||
e39c0676 | 2159 | return false; |
e81ecca3 | 2160 | mounted: |
c5839011 | 2161 | log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.\n", f->file_name); |
e39c0676 | 2162 | return true; |
e81ecca3 JA |
2163 | } |
2164 | ||
9cc8cb91 AK |
2165 | static bool waitee_running(struct thread_data *me) |
2166 | { | |
2167 | const char *waitee = me->o.wait_for; | |
2168 | const char *self = me->o.name; | |
2169 | struct thread_data *td; | |
2170 | int i; | |
2171 | ||
2172 | if (!waitee) | |
2173 | return false; | |
2174 | ||
2175 | for_each_td(td, i) { | |
2176 | if (!strcmp(td->o.name, self) || strcmp(td->o.name, waitee)) | |
2177 | continue; | |
2178 | ||
2179 | if (td->runstate < TD_EXITED) { | |
2180 | dprint(FD_PROCESS, "%s fenced by %s(%s)\n", | |
2181 | self, td->o.name, | |
2182 | runstate_to_name(td->runstate)); | |
2183 | return true; | |
2184 | } | |
2185 | } | |
2186 | ||
2187 | dprint(FD_PROCESS, "%s: %s completed, can run\n", self, waitee); | |
2188 | return false; | |
2189 | } | |
2190 | ||
2e1df07d JA |
2191 | /* |
2192 | * Main function for kicking off and reaping jobs, as needed. | |
2193 | */ | |
24660963 | 2194 | static void run_threads(struct sk_out *sk_out) |
2e1df07d JA |
2195 | { |
2196 | struct thread_data *td; | |
90eff1c9 SW |
2197 | unsigned int i, todo, nr_running, nr_started; |
2198 | uint64_t m_rate, t_rate; | |
0de5b26f | 2199 | uint64_t spent; |
2e1df07d | 2200 | |
2e1df07d JA |
2201 | if (fio_gtod_offload && fio_start_gtod_thread()) |
2202 | return; | |
334185e9 | 2203 | |
f2a2ce0e | 2204 | fio_idle_prof_init(); |
2e1df07d JA |
2205 | |
2206 | set_sig_handlers(); | |
2207 | ||
3a5f6bde JA |
2208 | nr_thread = nr_process = 0; |
2209 | for_each_td(td, i) { | |
e81ecca3 JA |
2210 | if (check_mount_writes(td)) |
2211 | return; | |
3a5f6bde JA |
2212 | if (td->o.use_thread) |
2213 | nr_thread++; | |
2214 | else | |
2215 | nr_process++; | |
2216 | } | |
2217 | ||
01cfefcc | 2218 | if (output_format & FIO_OUTPUT_NORMAL) { |
a41fc529 JA |
2219 | struct buf_output out; |
2220 | ||
2221 | buf_output_init(&out); | |
c2fcb6bc | 2222 | __log_buf(&out, "Starting "); |
2e1df07d | 2223 | if (nr_thread) |
c2fcb6bc | 2224 | __log_buf(&out, "%d thread%s", nr_thread, |
2e1df07d JA |
2225 | nr_thread > 1 ? "s" : ""); |
2226 | if (nr_process) { | |
2227 | if (nr_thread) | |
c2fcb6bc JA |
2228 | __log_buf(&out, " and "); |
2229 | __log_buf(&out, "%d process%s", nr_process, | |
2e1df07d JA |
2230 | nr_process > 1 ? "es" : ""); |
2231 | } | |
c2fcb6bc | 2232 | __log_buf(&out, "\n"); |
a41fc529 | 2233 | log_info_buf(out.buf, out.buflen); |
d5bb0a89 | 2234 | buf_output_free(&out); |
2e1df07d JA |
2235 | } |
2236 | ||
2237 | todo = thread_number; | |
2238 | nr_running = 0; | |
2239 | nr_started = 0; | |
2240 | m_rate = t_rate = 0; | |
2241 | ||
2242 | for_each_td(td, i) { | |
2243 | print_status_init(td->thread_number - 1); | |
2244 | ||
2245 | if (!td->o.create_serialize) | |
2246 | continue; | |
2247 | ||
ca09be4b JA |
2248 | if (fio_verify_load_state(td)) |
2249 | goto reap; | |
2250 | ||
2e1df07d JA |
2251 | /* |
2252 | * do file setup here so it happens sequentially, | |
2253 | * we don't want X number of threads getting their | |
2254 | * client data interspersed on disk | |
2255 | */ | |
2256 | if (setup_files(td)) { | |
ca09be4b | 2257 | reap: |
2e1df07d JA |
2258 | exit_value++; |
2259 | if (td->error) | |
2260 | log_err("fio: pid=%d, err=%d/%s\n", | |
2261 | (int) td->pid, td->error, td->verror); | |
2262 | td_set_runstate(td, TD_REAPED); | |
2263 | todo--; | |
2264 | } else { | |
2265 | struct fio_file *f; | |
2266 | unsigned int j; | |
2267 | ||
2268 | /* | |
2269 | * for sharing to work, each job must always open | |
2270 | * its own files. so close them, if we opened them | |
2271 | * for creation | |
2272 | */ | |
2273 | for_each_file(td, f, j) { | |
2274 | if (fio_file_open(f)) | |
2275 | td_io_close_file(td, f); | |
2276 | } | |
2277 | } | |
2278 | } | |
2279 | ||
f2a2ce0e HL |
2280 | /* start idle threads before io threads start to run */ |
2281 | fio_idle_prof_start(); | |
2282 | ||
2e1df07d JA |
2283 | set_genesis_time(); |
2284 | ||
2285 | while (todo) { | |
2286 | struct thread_data *map[REAL_MAX_JOBS]; | |
8b6a404c | 2287 | struct timespec this_start; |
2e1df07d | 2288 | int this_jobs = 0, left; |
ce8b6139 | 2289 | struct fork_data *fd; |
2e1df07d JA |
2290 | |
2291 | /* | |
2292 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
2293 | */ | |
2294 | for_each_td(td, i) { | |
2295 | if (td->runstate != TD_NOT_CREATED) | |
2296 | continue; | |
2297 | ||
2298 | /* | |
2299 | * never got a chance to start, killed by other | |
2300 | * thread for some reason | |
2301 | */ | |
2302 | if (td->terminate) { | |
2303 | todo--; | |
2304 | continue; | |
2305 | } | |
2306 | ||
2307 | if (td->o.start_delay) { | |
0de5b26f | 2308 | spent = utime_since_genesis(); |
2e1df07d | 2309 | |
74454ce4 | 2310 | if (td->o.start_delay > spent) |
2e1df07d JA |
2311 | continue; |
2312 | } | |
2313 | ||
2314 | if (td->o.stonewall && (nr_started || nr_running)) { | |
2315 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
2316 | td->o.name); | |
2317 | break; | |
2318 | } | |
2319 | ||
9cc8cb91 AK |
2320 | if (waitee_running(td)) { |
2321 | dprint(FD_PROCESS, "%s: waiting for %s\n", | |
2322 | td->o.name, td->o.wait_for); | |
5c74fc76 | 2323 | continue; |
9cc8cb91 AK |
2324 | } |
2325 | ||
2e1df07d JA |
2326 | init_disk_util(td); |
2327 | ||
971caeb1 | 2328 | td->rusage_sem = fio_sem_init(FIO_SEM_LOCKED); |
c97f1ad6 JA |
2329 | td->update_rusage = 0; |
2330 | ||
2e1df07d JA |
2331 | /* |
2332 | * Set state to created. Thread will transition | |
2333 | * to TD_INITIALIZED when it's done setting up. | |
2334 | */ | |
2335 | td_set_runstate(td, TD_CREATED); | |
2336 | map[this_jobs++] = td; | |
2337 | nr_started++; | |
2338 | ||
ce8b6139 JA |
2339 | fd = calloc(1, sizeof(*fd)); |
2340 | fd->td = td; | |
2341 | fd->sk_out = sk_out; | |
2342 | ||
2e1df07d JA |
2343 | if (td->o.use_thread) { |
2344 | int ret; | |
2345 | ||
2346 | dprint(FD_PROCESS, "will pthread_create\n"); | |
2347 | ret = pthread_create(&td->thread, NULL, | |
24660963 | 2348 | thread_main, fd); |
2e1df07d JA |
2349 | if (ret) { |
2350 | log_err("pthread_create: %s\n", | |
2351 | strerror(ret)); | |
24660963 | 2352 | free(fd); |
2e1df07d JA |
2353 | nr_started--; |
2354 | break; | |
2355 | } | |
493dd6c1 | 2356 | fd = NULL; |
2e1df07d JA |
2357 | ret = pthread_detach(td->thread); |
2358 | if (ret) | |
2359 | log_err("pthread_detach: %s", | |
2360 | strerror(ret)); | |
2361 | } else { | |
2362 | pid_t pid; | |
2363 | dprint(FD_PROCESS, "will fork\n"); | |
2364 | pid = fork(); | |
2365 | if (!pid) { | |
ce8b6139 | 2366 | int ret; |
2e1df07d | 2367 | |
ce8b6139 | 2368 | ret = (int)(uintptr_t)thread_main(fd); |
2e1df07d JA |
2369 | _exit(ret); |
2370 | } else if (i == fio_debug_jobno) | |
2371 | *fio_debug_jobp = pid; | |
2372 | } | |
971caeb1 BVA |
2373 | dprint(FD_MUTEX, "wait on startup_sem\n"); |
2374 | if (fio_sem_down_timeout(startup_sem, 10000)) { | |
2e1df07d | 2375 | log_err("fio: job startup hung? exiting.\n"); |
64402a8a | 2376 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
f1867a7f | 2377 | fio_abort = true; |
2e1df07d | 2378 | nr_started--; |
6541f4e0 | 2379 | free(fd); |
2e1df07d JA |
2380 | break; |
2381 | } | |
971caeb1 | 2382 | dprint(FD_MUTEX, "done waiting on startup_sem\n"); |
2e1df07d JA |
2383 | } |
2384 | ||
2385 | /* | |
2386 | * Wait for the started threads to transition to | |
2387 | * TD_INITIALIZED. | |
2388 | */ | |
2389 | fio_gettime(&this_start, NULL); | |
2390 | left = this_jobs; | |
2391 | while (left && !fio_abort) { | |
2392 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
2393 | break; | |
2394 | ||
06464907 | 2395 | do_usleep(100000); |
2e1df07d JA |
2396 | |
2397 | for (i = 0; i < this_jobs; i++) { | |
2398 | td = map[i]; | |
2399 | if (!td) | |
2400 | continue; | |
2401 | if (td->runstate == TD_INITIALIZED) { | |
2402 | map[i] = NULL; | |
2403 | left--; | |
2404 | } else if (td->runstate >= TD_EXITED) { | |
2405 | map[i] = NULL; | |
2406 | left--; | |
2407 | todo--; | |
2408 | nr_running++; /* work-around... */ | |
2409 | } | |
2410 | } | |
2411 | } | |
2412 | ||
2413 | if (left) { | |
4e87c37a JA |
2414 | log_err("fio: %d job%s failed to start\n", left, |
2415 | left > 1 ? "s" : ""); | |
2e1df07d JA |
2416 | for (i = 0; i < this_jobs; i++) { |
2417 | td = map[i]; | |
2418 | if (!td) | |
2419 | continue; | |
2420 | kill(td->pid, SIGTERM); | |
2421 | } | |
2422 | break; | |
2423 | } | |
2424 | ||
2425 | /* | |
2426 | * start created threads (TD_INITIALIZED -> TD_RUNNING). | |
2427 | */ | |
2428 | for_each_td(td, i) { | |
2429 | if (td->runstate != TD_INITIALIZED) | |
2430 | continue; | |
2431 | ||
2432 | if (in_ramp_time(td)) | |
2433 | td_set_runstate(td, TD_RAMP); | |
2434 | else | |
2435 | td_set_runstate(td, TD_RUNNING); | |
2436 | nr_running++; | |
2437 | nr_started--; | |
342f4be4 JA |
2438 | m_rate += ddir_rw_sum(td->o.ratemin); |
2439 | t_rate += ddir_rw_sum(td->o.rate); | |
2e1df07d | 2440 | todo--; |
971caeb1 | 2441 | fio_sem_up(td->sem); |
2e1df07d JA |
2442 | } |
2443 | ||
2444 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2445 | ||
122c7725 | 2446 | if (todo) |
06464907 | 2447 | do_usleep(100000); |
2e1df07d JA |
2448 | } |
2449 | ||
2450 | while (nr_running) { | |
2451 | reap_threads(&nr_running, &t_rate, &m_rate); | |
06464907 | 2452 | do_usleep(10000); |
2e1df07d JA |
2453 | } |
2454 | ||
f2a2ce0e HL |
2455 | fio_idle_prof_stop(); |
2456 | ||
2e1df07d | 2457 | update_io_ticks(); |
2e1df07d JA |
2458 | } |
2459 | ||
27357187 JA |
2460 | static void free_disk_util(void) |
2461 | { | |
27357187 | 2462 | disk_util_prune_entries(); |
a39fb9ea | 2463 | helper_thread_destroy(); |
2e1df07d JA |
2464 | } |
2465 | ||
24660963 | 2466 | int fio_backend(struct sk_out *sk_out) |
2e1df07d JA |
2467 | { |
2468 | struct thread_data *td; | |
2469 | int i; | |
2470 | ||
2471 | if (exec_profile) { | |
2472 | if (load_profile(exec_profile)) | |
2473 | return 1; | |
2474 | free(exec_profile); | |
2475 | exec_profile = NULL; | |
2476 | } | |
2477 | if (!thread_number) | |
2478 | return 0; | |
2479 | ||
2480 | if (write_bw_log) { | |
aee2ab67 JA |
2481 | struct log_params p = { |
2482 | .log_type = IO_LOG_TYPE_BW, | |
2483 | }; | |
2484 | ||
2485 | setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log"); | |
2486 | setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log"); | |
2487 | setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); | |
2e1df07d JA |
2488 | } |
2489 | ||
971caeb1 | 2490 | startup_sem = fio_sem_init(FIO_SEM_LOCKED); |
24a97130 LC |
2491 | if (!sk_out) |
2492 | is_local_backend = true; | |
971caeb1 | 2493 | if (startup_sem == NULL) |
2e1df07d | 2494 | return 1; |
2e1df07d JA |
2495 | |
2496 | set_genesis_time(); | |
cef9175e | 2497 | stat_init(); |
998e9ebb BVA |
2498 | if (helper_thread_create(startup_sem, sk_out)) |
2499 | log_err("fio: failed to create helper thread\n"); | |
2e1df07d JA |
2500 | |
2501 | cgroup_list = smalloc(sizeof(*cgroup_list)); | |
b2e6494c JA |
2502 | if (cgroup_list) |
2503 | INIT_FLIST_HEAD(cgroup_list); | |
2e1df07d | 2504 | |
24660963 | 2505 | run_threads(sk_out); |
2e1df07d | 2506 | |
a39fb9ea | 2507 | helper_thread_exit(); |
8aab824f | 2508 | |
2e1df07d | 2509 | if (!fio_abort) { |
83f7b64e | 2510 | __show_run_stats(); |
2e1df07d | 2511 | if (write_bw_log) { |
cb7e0ace JA |
2512 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { |
2513 | struct io_log *log = agg_io_log[i]; | |
2514 | ||
60a25727 | 2515 | flush_log(log, false); |
518dac09 | 2516 | free_log(log); |
cb7e0ace | 2517 | } |
2e1df07d JA |
2518 | } |
2519 | } | |
2520 | ||
fda2cfac | 2521 | for_each_td(td, i) { |
c27cc65f | 2522 | steadystate_free(td); |
2e1df07d | 2523 | fio_options_free(td); |
8049adc1 | 2524 | if (td->rusage_sem) { |
971caeb1 | 2525 | fio_sem_remove(td->rusage_sem); |
8049adc1 JA |
2526 | td->rusage_sem = NULL; |
2527 | } | |
971caeb1 BVA |
2528 | fio_sem_remove(td->sem); |
2529 | td->sem = NULL; | |
fda2cfac | 2530 | } |
2e1df07d | 2531 | |
a462baef | 2532 | free_disk_util(); |
b2e6494c JA |
2533 | if (cgroup_list) { |
2534 | cgroup_kill(cgroup_list); | |
2535 | sfree(cgroup_list); | |
2536 | } | |
2e1df07d | 2537 | |
971caeb1 | 2538 | fio_sem_remove(startup_sem); |
cef9175e | 2539 | stat_exit(); |
2e1df07d JA |
2540 | return exit_value; |
2541 | } |