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" |
0d8cc753 | 52 | #include "fio_time.h" |
2e1df07d | 53 | |
971caeb1 | 54 | static struct fio_sem *startup_sem; |
2e1df07d | 55 | static struct flist_head *cgroup_list; |
c3dc516a | 56 | static struct cgroup_mnt *cgroup_mnt; |
2e1df07d | 57 | static int exit_value; |
f1867a7f | 58 | static volatile bool fio_abort; |
3a5f6bde JA |
59 | static unsigned int nr_process = 0; |
60 | static unsigned int nr_thread = 0; | |
2e1df07d | 61 | |
6eaf09d6 | 62 | struct io_log *agg_io_log[DDIR_RWDIR_CNT]; |
2e1df07d | 63 | |
a3efc919 JA |
64 | int groupid = 0; |
65 | unsigned int thread_number = 0; | |
2fbe1e19 JA |
66 | unsigned int nr_segments = 0; |
67 | unsigned int cur_segment = 0; | |
108fea77 | 68 | unsigned int stat_number = 0; |
a3efc919 JA |
69 | int temp_stall_ts; |
70 | unsigned long done_secs = 0; | |
a38c70a8 BVA |
71 | #ifdef PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP |
72 | pthread_mutex_t overlap_check = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP; | |
73 | #else | |
c06379a6 | 74 | pthread_mutex_t overlap_check = PTHREAD_MUTEX_INITIALIZER; |
a38c70a8 | 75 | #endif |
a47591e4 | 76 | |
2e1df07d JA |
77 | #define JOB_START_TIMEOUT (5 * 1000) |
78 | ||
79 | static void sig_int(int sig) | |
80 | { | |
2fbe1e19 | 81 | if (nr_segments) { |
2e1df07d JA |
82 | if (is_backend) |
83 | fio_server_got_signal(sig); | |
84 | else { | |
85 | log_info("\nfio: terminating on signal %d\n", sig); | |
e411c301 | 86 | log_info_flush(); |
2e1df07d JA |
87 | exit_value = 128; |
88 | } | |
89 | ||
64402a8a | 90 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
2e1df07d JA |
91 | } |
92 | } | |
93 | ||
0dea1f55 BP |
94 | #ifdef WIN32 |
95 | static void sig_break(int sig) | |
96 | { | |
0dea1f55 BP |
97 | sig_int(sig); |
98 | ||
99 | /** | |
100 | * Windows terminates all job processes on SIGBREAK after the handler | |
101 | * returns, so give them time to wrap-up and give stats | |
102 | */ | |
da8f124f | 103 | for_each_td(td) { |
0dea1f55 BP |
104 | while (td->runstate < TD_EXITED) |
105 | sleep(1); | |
da8f124f | 106 | } end_for_each(); |
0dea1f55 BP |
107 | } |
108 | #endif | |
109 | ||
d2235e56 | 110 | void sig_show_status(int sig) |
4c6d91e8 JA |
111 | { |
112 | show_running_run_stats(); | |
113 | } | |
114 | ||
2e1df07d JA |
115 | static void set_sig_handlers(void) |
116 | { | |
117 | struct sigaction act; | |
118 | ||
119 | memset(&act, 0, sizeof(act)); | |
120 | act.sa_handler = sig_int; | |
121 | act.sa_flags = SA_RESTART; | |
122 | sigaction(SIGINT, &act, NULL); | |
123 | ||
124 | memset(&act, 0, sizeof(act)); | |
125 | act.sa_handler = sig_int; | |
126 | act.sa_flags = SA_RESTART; | |
127 | sigaction(SIGTERM, &act, NULL); | |
128 | ||
2f694507 BC |
129 | /* Windows uses SIGBREAK as a quit signal from other applications */ |
130 | #ifdef WIN32 | |
131 | memset(&act, 0, sizeof(act)); | |
0dea1f55 | 132 | act.sa_handler = sig_break; |
2f694507 BC |
133 | act.sa_flags = SA_RESTART; |
134 | sigaction(SIGBREAK, &act, NULL); | |
135 | #endif | |
136 | ||
4c6d91e8 JA |
137 | memset(&act, 0, sizeof(act)); |
138 | act.sa_handler = sig_show_status; | |
139 | act.sa_flags = SA_RESTART; | |
140 | sigaction(SIGUSR1, &act, NULL); | |
141 | ||
2e1df07d JA |
142 | if (is_backend) { |
143 | memset(&act, 0, sizeof(act)); | |
144 | act.sa_handler = sig_int; | |
145 | act.sa_flags = SA_RESTART; | |
146 | sigaction(SIGPIPE, &act, NULL); | |
147 | } | |
148 | } | |
149 | ||
150 | /* | |
151 | * Check if we are above the minimum rate given. | |
152 | */ | |
8b6a404c | 153 | static bool __check_min_rate(struct thread_data *td, struct timespec *now, |
e39c0676 | 154 | enum fio_ddir ddir) |
2e1df07d | 155 | { |
f5204b8a | 156 | unsigned long long current_rate_check_bytes = td->this_io_bytes[ddir]; |
157 | unsigned long current_rate_check_blocks = td->this_io_blocks[ddir]; | |
158 | unsigned long long option_rate_bytes_min = td->o.ratemin[ddir]; | |
159 | unsigned int option_rate_iops_min = td->o.rate_iops_min[ddir]; | |
2e1df07d JA |
160 | |
161 | assert(ddir_rw(ddir)); | |
162 | ||
163 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) | |
e39c0676 | 164 | return false; |
2e1df07d JA |
165 | |
166 | /* | |
167 | * allow a 2 second settle period in the beginning | |
168 | */ | |
169 | if (mtime_since(&td->start, now) < 2000) | |
e39c0676 | 170 | return false; |
2e1df07d | 171 | |
2e1df07d | 172 | /* |
f5204b8a | 173 | * if last_rate_check_blocks or last_rate_check_bytes is set, |
174 | * we can compute a rate per ratecycle | |
2e1df07d | 175 | */ |
f5204b8a | 176 | if (td->last_rate_check_bytes[ddir] || td->last_rate_check_blocks[ddir]) { |
177 | unsigned long spent = mtime_since(&td->last_rate_check_time[ddir], now); | |
178 | if (spent < td->o.ratecycle || spent==0) | |
e39c0676 | 179 | return false; |
2e1df07d | 180 | |
f5204b8a | 181 | if (td->o.ratemin[ddir]) { |
2e1df07d JA |
182 | /* |
183 | * check bandwidth specified rate | |
184 | */ | |
f5204b8a | 185 | unsigned long long current_rate_bytes = |
186 | ((current_rate_check_bytes - td->last_rate_check_bytes[ddir]) * 1000) / spent; | |
187 | if (current_rate_bytes < option_rate_bytes_min) { | |
188 | log_err("%s: rate_min=%lluB/s not met, got %lluB/s\n", | |
189 | td->o.name, option_rate_bytes_min, current_rate_bytes); | |
e39c0676 | 190 | return true; |
2e1df07d JA |
191 | } |
192 | } else { | |
193 | /* | |
194 | * checks iops specified rate | |
195 | */ | |
f5204b8a | 196 | unsigned long long current_rate_iops = |
197 | ((current_rate_check_blocks - td->last_rate_check_blocks[ddir]) * 1000) / spent; | |
198 | ||
199 | if (current_rate_iops < option_rate_iops_min) { | |
200 | log_err("%s: rate_iops_min=%u not met, got %llu IOPS\n", | |
201 | td->o.name, option_rate_iops_min, current_rate_iops); | |
e39c0676 | 202 | return true; |
2e1df07d JA |
203 | } |
204 | } | |
205 | } | |
206 | ||
f5204b8a | 207 | td->last_rate_check_bytes[ddir] = current_rate_check_bytes; |
208 | td->last_rate_check_blocks[ddir] = current_rate_check_blocks; | |
209 | memcpy(&td->last_rate_check_time[ddir], now, sizeof(*now)); | |
e39c0676 | 210 | return false; |
2e1df07d JA |
211 | } |
212 | ||
8b6a404c | 213 | static bool check_min_rate(struct thread_data *td, struct timespec *now) |
2e1df07d | 214 | { |
e39c0676 | 215 | bool ret = false; |
2e1df07d | 216 | |
dde39f8c AD |
217 | for_each_rw_ddir(ddir) { |
218 | if (td->bytes_done[ddir]) | |
219 | ret |= __check_min_rate(td, now, ddir); | |
220 | } | |
2e1df07d JA |
221 | |
222 | return ret; | |
223 | } | |
224 | ||
225 | /* | |
226 | * When job exits, we can cancel the in-flight IO if we are using async | |
227 | * io. Attempt to do so. | |
228 | */ | |
229 | static void cleanup_pending_aio(struct thread_data *td) | |
230 | { | |
2e1df07d JA |
231 | int r; |
232 | ||
233 | /* | |
234 | * get immediately available events, if any | |
235 | */ | |
55312f9f | 236 | r = io_u_queued_complete(td, 0); |
2e1df07d JA |
237 | |
238 | /* | |
239 | * now cancel remaining active events | |
240 | */ | |
241 | if (td->io_ops->cancel) { | |
2ae0b204 JA |
242 | struct io_u *io_u; |
243 | int i; | |
2e1df07d | 244 | |
2ae0b204 JA |
245 | io_u_qiter(&td->io_u_all, io_u, i) { |
246 | if (io_u->flags & IO_U_F_FLIGHT) { | |
2e1df07d JA |
247 | r = td->io_ops->cancel(td, io_u); |
248 | if (!r) | |
249 | put_io_u(td, io_u); | |
250 | } | |
251 | } | |
252 | } | |
253 | ||
254 | if (td->cur_depth) | |
55312f9f | 255 | r = io_u_queued_complete(td, td->cur_depth); |
2e1df07d JA |
256 | } |
257 | ||
258 | /* | |
259 | * Helper to handle the final sync of a file. Works just like the normal | |
260 | * io path, just does everything sync. | |
261 | */ | |
e39c0676 | 262 | static bool fio_io_sync(struct thread_data *td, struct fio_file *f) |
2e1df07d JA |
263 | { |
264 | struct io_u *io_u = __get_io_u(td); | |
d3b07186 | 265 | enum fio_q_status ret; |
2e1df07d JA |
266 | |
267 | if (!io_u) | |
e39c0676 | 268 | return true; |
2e1df07d JA |
269 | |
270 | io_u->ddir = DDIR_SYNC; | |
271 | io_u->file = f; | |
00615bfb | 272 | io_u_set(td, io_u, IO_U_F_NO_FILE_PUT); |
2e1df07d JA |
273 | |
274 | if (td_io_prep(td, io_u)) { | |
275 | put_io_u(td, io_u); | |
e39c0676 | 276 | return true; |
2e1df07d JA |
277 | } |
278 | ||
279 | requeue: | |
280 | ret = td_io_queue(td, io_u); | |
18f6165e BVA |
281 | switch (ret) { |
282 | case FIO_Q_QUEUED: | |
a80cb54b | 283 | td_io_commit(td); |
55312f9f | 284 | if (io_u_queued_complete(td, 1) < 0) |
e39c0676 | 285 | return true; |
18f6165e BVA |
286 | break; |
287 | case FIO_Q_COMPLETED: | |
2e1df07d JA |
288 | if (io_u->error) { |
289 | td_verror(td, io_u->error, "td_io_queue"); | |
e39c0676 | 290 | return true; |
2e1df07d JA |
291 | } |
292 | ||
55312f9f | 293 | if (io_u_sync_complete(td, io_u) < 0) |
e39c0676 | 294 | return true; |
18f6165e BVA |
295 | break; |
296 | case FIO_Q_BUSY: | |
a80cb54b | 297 | td_io_commit(td); |
2e1df07d JA |
298 | goto requeue; |
299 | } | |
300 | ||
e39c0676 | 301 | return false; |
2e1df07d | 302 | } |
a3efc919 | 303 | |
61ee0f86 JA |
304 | static int fio_file_fsync(struct thread_data *td, struct fio_file *f) |
305 | { | |
00615bfb | 306 | int ret, ret2; |
61ee0f86 JA |
307 | |
308 | if (fio_file_open(f)) | |
309 | return fio_io_sync(td, f); | |
310 | ||
311 | if (td_io_open_file(td, f)) | |
312 | return 1; | |
313 | ||
314 | ret = fio_io_sync(td, f); | |
00615bfb SW |
315 | ret2 = 0; |
316 | if (fio_file_open(f)) | |
317 | ret2 = td_io_close_file(td, f); | |
318 | return (ret || ret2); | |
61ee0f86 JA |
319 | } |
320 | ||
8b6a404c | 321 | static inline void __update_ts_cache(struct thread_data *td) |
2e1df07d | 322 | { |
8b6a404c | 323 | fio_gettime(&td->ts_cache, NULL); |
2e1df07d JA |
324 | } |
325 | ||
8b6a404c | 326 | static inline void update_ts_cache(struct thread_data *td) |
2e1df07d | 327 | { |
8b6a404c VF |
328 | if ((++td->ts_cache_nr & td->ts_cache_mask) == td->ts_cache_mask) |
329 | __update_ts_cache(td); | |
2e1df07d JA |
330 | } |
331 | ||
8b6a404c | 332 | static inline bool runtime_exceeded(struct thread_data *td, struct timespec *t) |
2e1df07d JA |
333 | { |
334 | if (in_ramp_time(td)) | |
e39c0676 | 335 | return false; |
2e1df07d | 336 | if (!td->o.timeout) |
e39c0676 | 337 | return false; |
0de5b26f | 338 | if (utime_since(&td->epoch, t) >= td->o.timeout) |
e39c0676 | 339 | return true; |
2e1df07d | 340 | |
e39c0676 | 341 | return false; |
2e1df07d JA |
342 | } |
343 | ||
95603b74 BF |
344 | /* |
345 | * We need to update the runtime consistently in ms, but keep a running | |
346 | * tally of the current elapsed time in microseconds for sub millisecond | |
347 | * updates. | |
348 | */ | |
349 | static inline void update_runtime(struct thread_data *td, | |
350 | unsigned long long *elapsed_us, | |
351 | const enum fio_ddir ddir) | |
352 | { | |
d5c8ea29 JA |
353 | if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only) |
354 | return; | |
355 | ||
95603b74 BF |
356 | td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000; |
357 | elapsed_us[ddir] += utime_since_now(&td->start); | |
358 | td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000; | |
359 | } | |
360 | ||
e39c0676 JA |
361 | static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir, |
362 | int *retptr) | |
2e1df07d JA |
363 | { |
364 | int ret = *retptr; | |
365 | ||
366 | if (ret < 0 || td->error) { | |
8b28bd41 DM |
367 | int err = td->error; |
368 | enum error_type_bit eb; | |
2e1df07d JA |
369 | |
370 | if (ret < 0) | |
371 | err = -ret; | |
2e1df07d | 372 | |
8b28bd41 DM |
373 | eb = td_error_type(ddir, err); |
374 | if (!(td->o.continue_on_error & (1 << eb))) | |
e39c0676 | 375 | return true; |
2e1df07d | 376 | |
8b28bd41 | 377 | if (td_non_fatal_error(td, eb, err)) { |
2e1df07d JA |
378 | /* |
379 | * Continue with the I/Os in case of | |
380 | * a non fatal error. | |
381 | */ | |
382 | update_error_count(td, err); | |
383 | td_clear_error(td); | |
384 | *retptr = 0; | |
e39c0676 | 385 | return false; |
418f5399 | 386 | } else if (td->o.fill_device && (err == ENOSPC || err == EDQUOT)) { |
2e1df07d JA |
387 | /* |
388 | * We expect to hit this error if | |
389 | * fill_device option is set. | |
390 | */ | |
391 | td_clear_error(td); | |
ebea2133 | 392 | fio_mark_td_terminate(td); |
e39c0676 | 393 | return true; |
2e1df07d JA |
394 | } else { |
395 | /* | |
396 | * Stop the I/O in case of a fatal | |
397 | * error. | |
398 | */ | |
399 | update_error_count(td, err); | |
e39c0676 | 400 | return true; |
2e1df07d JA |
401 | } |
402 | } | |
403 | ||
e39c0676 | 404 | return false; |
2e1df07d JA |
405 | } |
406 | ||
c97f1ad6 JA |
407 | static void check_update_rusage(struct thread_data *td) |
408 | { | |
409 | if (td->update_rusage) { | |
410 | td->update_rusage = 0; | |
411 | update_rusage_stat(td); | |
971caeb1 | 412 | fio_sem_up(td->rusage_sem); |
c97f1ad6 JA |
413 | } |
414 | } | |
415 | ||
8b6a404c | 416 | static int wait_for_completions(struct thread_data *td, struct timespec *time) |
69fea81e JA |
417 | { |
418 | const int full = queue_full(td); | |
419 | int min_evts = 0; | |
420 | int ret; | |
421 | ||
6be06c46 JA |
422 | if (td->flags & TD_F_REGROW_LOGS) |
423 | return io_u_quiesce(td); | |
1fed2080 | 424 | |
69fea81e JA |
425 | /* |
426 | * if the queue is full, we MUST reap at least 1 event | |
427 | */ | |
82407585 | 428 | min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth); |
4b157ac6 | 429 | if ((full && !min_evts) || !td->o.iodepth_batch_complete_min) |
69fea81e JA |
430 | min_evts = 1; |
431 | ||
90711edc | 432 | if (time && should_check_rate(td)) |
69fea81e JA |
433 | fio_gettime(time, NULL); |
434 | ||
435 | do { | |
55312f9f | 436 | ret = io_u_queued_complete(td, min_evts); |
69fea81e JA |
437 | if (ret < 0) |
438 | break; | |
439 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
440 | ||
441 | return ret; | |
442 | } | |
443 | ||
e9d512d8 JA |
444 | int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret, |
445 | enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify, | |
8b6a404c | 446 | struct timespec *comp_time) |
e9d512d8 | 447 | { |
e9d512d8 JA |
448 | switch (*ret) { |
449 | case FIO_Q_COMPLETED: | |
450 | if (io_u->error) { | |
451 | *ret = -io_u->error; | |
452 | clear_io_u(td, io_u); | |
453 | } else if (io_u->resid) { | |
5fff9543 | 454 | long long bytes = io_u->xfer_buflen - io_u->resid; |
e9d512d8 JA |
455 | struct fio_file *f = io_u->file; |
456 | ||
457 | if (bytes_issued) | |
458 | *bytes_issued += bytes; | |
459 | ||
460 | if (!from_verify) | |
94a78d52 | 461 | trim_io_piece(io_u); |
e9d512d8 JA |
462 | |
463 | /* | |
464 | * zero read, fail | |
465 | */ | |
466 | if (!bytes) { | |
467 | if (!from_verify) | |
468 | unlog_io_piece(td, io_u); | |
469 | td_verror(td, EIO, "full resid"); | |
fc37e3de | 470 | clear_io_u(td, io_u); |
e9d512d8 JA |
471 | break; |
472 | } | |
473 | ||
474 | io_u->xfer_buflen = io_u->resid; | |
475 | io_u->xfer_buf += bytes; | |
476 | io_u->offset += bytes; | |
477 | ||
478 | if (ddir_rw(io_u->ddir)) | |
479 | td->ts.short_io_u[io_u->ddir]++; | |
480 | ||
e9d512d8 JA |
481 | if (io_u->offset == f->real_file_size) |
482 | goto sync_done; | |
483 | ||
484 | requeue_io_u(td, &io_u); | |
485 | } else { | |
486 | sync_done: | |
90711edc | 487 | if (comp_time && should_check_rate(td)) |
e9d512d8 JA |
488 | fio_gettime(comp_time, NULL); |
489 | ||
490 | *ret = io_u_sync_complete(td, io_u); | |
491 | if (*ret < 0) | |
492 | break; | |
493 | } | |
a0e3e5a6 | 494 | |
cf2c8d52 JA |
495 | if (td->flags & TD_F_REGROW_LOGS) |
496 | regrow_logs(td); | |
497 | ||
a0e3e5a6 EW |
498 | /* |
499 | * when doing I/O (not when verifying), | |
500 | * check for any errors that are to be ignored | |
501 | */ | |
502 | if (!from_verify) | |
503 | break; | |
504 | ||
e9d512d8 JA |
505 | return 0; |
506 | case FIO_Q_QUEUED: | |
507 | /* | |
508 | * if the engine doesn't have a commit hook, | |
509 | * the io_u is really queued. if it does have such | |
510 | * a hook, it has to call io_u_queued() itself. | |
511 | */ | |
512 | if (td->io_ops->commit == NULL) | |
513 | io_u_queued(td, io_u); | |
514 | if (bytes_issued) | |
515 | *bytes_issued += io_u->xfer_buflen; | |
516 | break; | |
517 | case FIO_Q_BUSY: | |
518 | if (!from_verify) | |
519 | unlog_io_piece(td, io_u); | |
520 | requeue_io_u(td, &io_u); | |
a80cb54b | 521 | td_io_commit(td); |
e9d512d8 JA |
522 | break; |
523 | default: | |
b0775325 | 524 | assert(*ret < 0); |
e9d512d8 JA |
525 | td_verror(td, -(*ret), "td_io_queue"); |
526 | break; | |
527 | } | |
528 | ||
529 | if (break_on_this_error(td, ddir, ret)) | |
530 | return 1; | |
531 | ||
532 | return 0; | |
533 | } | |
534 | ||
e39c0676 | 535 | static inline bool io_in_polling(struct thread_data *td) |
82407585 RP |
536 | { |
537 | return !td->o.iodepth_batch_complete_min && | |
538 | !td->o.iodepth_batch_complete_max; | |
539 | } | |
39c1c323 | 540 | /* |
541 | * Unlinks files from thread data fio_file structure | |
542 | */ | |
543 | static int unlink_all_files(struct thread_data *td) | |
544 | { | |
545 | struct fio_file *f; | |
546 | unsigned int i; | |
547 | int ret = 0; | |
548 | ||
549 | for_each_file(td, f, i) { | |
550 | if (f->filetype != FIO_TYPE_FILE) | |
551 | continue; | |
552 | ret = td_io_unlink_file(td, f); | |
553 | if (ret) | |
554 | break; | |
555 | } | |
556 | ||
557 | if (ret) | |
558 | td_verror(td, ret, "unlink_all_files"); | |
559 | ||
560 | return ret; | |
561 | } | |
82407585 | 562 | |
997b5680 SW |
563 | /* |
564 | * Check if io_u will overlap an in-flight IO in the queue | |
565 | */ | |
c06379a6 | 566 | bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u) |
997b5680 SW |
567 | { |
568 | bool overlap; | |
569 | struct io_u *check_io_u; | |
570 | unsigned long long x1, x2, y1, y2; | |
571 | int i; | |
572 | ||
573 | x1 = io_u->offset; | |
574 | x2 = io_u->offset + io_u->buflen; | |
575 | overlap = false; | |
576 | io_u_qiter(q, check_io_u, i) { | |
577 | if (check_io_u->flags & IO_U_F_FLIGHT) { | |
578 | y1 = check_io_u->offset; | |
579 | y2 = check_io_u->offset + check_io_u->buflen; | |
580 | ||
581 | if (x1 < y2 && y1 < x2) { | |
582 | overlap = true; | |
5fff9543 | 583 | dprint(FD_IO, "in-flight overlap: %llu/%llu, %llu/%llu\n", |
997b5680 SW |
584 | x1, io_u->buflen, |
585 | y1, check_io_u->buflen); | |
586 | break; | |
587 | } | |
588 | } | |
589 | } | |
590 | ||
591 | return overlap; | |
592 | } | |
593 | ||
d3b07186 | 594 | static enum fio_q_status io_u_submit(struct thread_data *td, struct io_u *io_u) |
08b2b386 JA |
595 | { |
596 | /* | |
597 | * Check for overlap if the user asked us to, and we have | |
598 | * at least one IO in flight besides this one. | |
599 | */ | |
600 | if (td->o.serialize_overlap && td->cur_depth > 1 && | |
601 | in_flight_overlap(&td->io_u_all, io_u)) | |
602 | return FIO_Q_BUSY; | |
603 | ||
604 | return td_io_queue(td, io_u); | |
605 | } | |
606 | ||
2e1df07d JA |
607 | /* |
608 | * The main verify engine. Runs over the writes we previously submitted, | |
609 | * reads the blocks back in, and checks the crc/md5 of the data. | |
610 | */ | |
100f49f1 | 611 | static void do_verify(struct thread_data *td, uint64_t verify_bytes) |
2e1df07d JA |
612 | { |
613 | struct fio_file *f; | |
614 | struct io_u *io_u; | |
615 | int ret, min_events; | |
616 | unsigned int i; | |
617 | ||
618 | dprint(FD_VERIFY, "starting loop\n"); | |
619 | ||
620 | /* | |
621 | * sync io first and invalidate cache, to make sure we really | |
622 | * read from disk. | |
623 | */ | |
624 | for_each_file(td, f, i) { | |
625 | if (!fio_file_open(f)) | |
626 | continue; | |
627 | if (fio_io_sync(td, f)) | |
628 | break; | |
629 | if (file_invalidate_cache(td, f)) | |
630 | break; | |
631 | } | |
632 | ||
c97f1ad6 JA |
633 | check_update_rusage(td); |
634 | ||
2e1df07d JA |
635 | if (td->error) |
636 | return; | |
637 | ||
638 | td_set_runstate(td, TD_VERIFYING); | |
639 | ||
640 | io_u = NULL; | |
641 | while (!td->terminate) { | |
fbccf46c | 642 | enum fio_ddir ddir; |
e9d512d8 | 643 | int full; |
2e1df07d | 644 | |
8b6a404c | 645 | update_ts_cache(td); |
c97f1ad6 | 646 | check_update_rusage(td); |
2e1df07d | 647 | |
8b6a404c VF |
648 | if (runtime_exceeded(td, &td->ts_cache)) { |
649 | __update_ts_cache(td); | |
650 | if (runtime_exceeded(td, &td->ts_cache)) { | |
ebea2133 | 651 | fio_mark_td_terminate(td); |
2e1df07d JA |
652 | break; |
653 | } | |
654 | } | |
655 | ||
9e684a49 DE |
656 | if (flow_threshold_exceeded(td)) |
657 | continue; | |
658 | ||
44cbc6da JA |
659 | if (!td->o.experimental_verify) { |
660 | io_u = __get_io_u(td); | |
661 | if (!io_u) | |
662 | break; | |
2e1df07d | 663 | |
44cbc6da JA |
664 | if (get_next_verify(td, io_u)) { |
665 | put_io_u(td, io_u); | |
666 | break; | |
667 | } | |
2e1df07d | 668 | |
44cbc6da JA |
669 | if (td_io_prep(td, io_u)) { |
670 | put_io_u(td, io_u); | |
671 | break; | |
672 | } | |
673 | } else { | |
191d6634 | 674 | if (td->bytes_verified + td->o.rw_min_bs > verify_bytes) |
100f49f1 JA |
675 | break; |
676 | ||
bcd5abfa | 677 | while ((io_u = get_io_u(td)) != NULL) { |
059b61f2 | 678 | if (IS_ERR_OR_NULL(io_u)) { |
002fe734 JA |
679 | io_u = NULL; |
680 | ret = FIO_Q_BUSY; | |
681 | goto reap; | |
682 | } | |
683 | ||
bcd5abfa JA |
684 | /* |
685 | * We are only interested in the places where | |
686 | * we wrote or trimmed IOs. Turn those into | |
687 | * reads for verification purposes. | |
688 | */ | |
689 | if (io_u->ddir == DDIR_READ) { | |
690 | /* | |
691 | * Pretend we issued it for rwmix | |
692 | * accounting | |
693 | */ | |
694 | td->io_issues[DDIR_READ]++; | |
695 | put_io_u(td, io_u); | |
696 | continue; | |
697 | } else if (io_u->ddir == DDIR_TRIM) { | |
698 | io_u->ddir = DDIR_READ; | |
1651e431 | 699 | io_u_set(td, io_u, IO_U_F_TRIMMED); |
bcd5abfa JA |
700 | break; |
701 | } else if (io_u->ddir == DDIR_WRITE) { | |
702 | io_u->ddir = DDIR_READ; | |
28921bc0 SK |
703 | io_u->numberio = td->verify_read_issues; |
704 | td->verify_read_issues++; | |
5c5c33c1 | 705 | populate_verify_io_u(td, io_u); |
bcd5abfa JA |
706 | break; |
707 | } else { | |
708 | put_io_u(td, io_u); | |
709 | continue; | |
710 | } | |
711 | } | |
44cbc6da | 712 | |
bcd5abfa | 713 | if (!io_u) |
44cbc6da | 714 | break; |
2e1df07d JA |
715 | } |
716 | ||
ca09be4b JA |
717 | if (verify_state_should_stop(td, io_u)) { |
718 | put_io_u(td, io_u); | |
719 | break; | |
720 | } | |
721 | ||
2e1df07d JA |
722 | if (td->o.verify_async) |
723 | io_u->end_io = verify_io_u_async; | |
724 | else | |
725 | io_u->end_io = verify_io_u; | |
726 | ||
fbccf46c | 727 | ddir = io_u->ddir; |
10adc4a7 GG |
728 | if (!td->o.disable_slat) |
729 | fio_gettime(&io_u->start_time, NULL); | |
fbccf46c | 730 | |
08b2b386 | 731 | ret = io_u_submit(td, io_u); |
2e1df07d | 732 | |
e9d512d8 | 733 | if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL)) |
2e1df07d JA |
734 | break; |
735 | ||
736 | /* | |
737 | * if we can queue more, do so. but check if there are | |
738 | * completed io_u's first. Note that we can get BUSY even | |
739 | * without IO queued, if the system is resource starved. | |
740 | */ | |
002fe734 | 741 | reap: |
2e1df07d | 742 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
82407585 | 743 | if (full || io_in_polling(td)) |
55312f9f | 744 | ret = wait_for_completions(td, NULL); |
2e1df07d | 745 | |
2e1df07d JA |
746 | if (ret < 0) |
747 | break; | |
748 | } | |
749 | ||
c97f1ad6 JA |
750 | check_update_rusage(td); |
751 | ||
2e1df07d JA |
752 | if (!td->error) { |
753 | min_events = td->cur_depth; | |
754 | ||
755 | if (min_events) | |
55312f9f | 756 | ret = io_u_queued_complete(td, min_events); |
2e1df07d JA |
757 | } else |
758 | cleanup_pending_aio(td); | |
759 | ||
760 | td_set_runstate(td, TD_RUNNING); | |
761 | ||
762 | dprint(FD_VERIFY, "exiting loop\n"); | |
763 | } | |
764 | ||
e39c0676 | 765 | static bool exceeds_number_ios(struct thread_data *td) |
3939fe85 JA |
766 | { |
767 | unsigned long long number_ios; | |
768 | ||
769 | if (!td->o.number_ios) | |
e39c0676 | 770 | return false; |
3939fe85 | 771 | |
cf8a46a7 | 772 | number_ios = ddir_rw_sum(td->io_blocks); |
3939fe85 JA |
773 | number_ios += td->io_u_queued + td->io_u_in_flight; |
774 | ||
cf8a46a7 | 775 | return number_ios >= (td->o.number_ios * td->loops); |
3939fe85 JA |
776 | } |
777 | ||
926098e0 | 778 | static bool io_bytes_exceeded(struct thread_data *td, uint64_t *this_bytes) |
f7078f7b | 779 | { |
77731b29 | 780 | unsigned long long bytes, limit; |
f7078f7b JA |
781 | |
782 | if (td_rw(td)) | |
926098e0 | 783 | bytes = this_bytes[DDIR_READ] + this_bytes[DDIR_WRITE]; |
f7078f7b | 784 | else if (td_write(td)) |
926098e0 | 785 | bytes = this_bytes[DDIR_WRITE]; |
6eaf09d6 | 786 | else if (td_read(td)) |
926098e0 | 787 | bytes = this_bytes[DDIR_READ]; |
f7078f7b | 788 | else |
926098e0 | 789 | bytes = this_bytes[DDIR_TRIM]; |
f7078f7b | 790 | |
5be9bf09 TK |
791 | if (td->o.io_size) |
792 | limit = td->o.io_size; | |
77731b29 JA |
793 | else |
794 | limit = td->o.size; | |
795 | ||
cf8a46a7 | 796 | limit *= td->loops; |
77731b29 | 797 | return bytes >= limit || exceeds_number_ios(td); |
f7078f7b JA |
798 | } |
799 | ||
926098e0 | 800 | static bool io_issue_bytes_exceeded(struct thread_data *td) |
e28dd2cf | 801 | { |
926098e0 JA |
802 | return io_bytes_exceeded(td, td->io_issue_bytes); |
803 | } | |
e28dd2cf | 804 | |
926098e0 JA |
805 | static bool io_complete_bytes_exceeded(struct thread_data *td) |
806 | { | |
807 | return io_bytes_exceeded(td, td->this_io_bytes); | |
e28dd2cf JE |
808 | } |
809 | ||
50a8ce86 D |
810 | /* |
811 | * used to calculate the next io time for rate control | |
812 | * | |
813 | */ | |
814 | static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir) | |
815 | { | |
1a9bf814 | 816 | uint64_t bps = td->rate_bps[ddir]; |
50a8ce86 D |
817 | |
818 | assert(!(td->flags & TD_F_CHILD)); | |
ff6bb260 | 819 | |
6de65959 | 820 | if (td->o.rate_process == RATE_PROCESS_POISSON) { |
1a9bf814 JA |
821 | uint64_t val, iops; |
822 | ||
d6299558 | 823 | iops = bps / td->o.min_bs[ddir]; |
a77d139b | 824 | val = (int64_t) (1000000 / iops) * |
6c8611c7 | 825 | -logf(__rand_0_1(&td->poisson_state[ddir])); |
a77d139b | 826 | if (val) { |
98e51d7c JA |
827 | dprint(FD_RATE, "poisson rate iops=%llu, ddir=%d\n", |
828 | (unsigned long long) 1000000 / val, | |
829 | ddir); | |
a77d139b | 830 | } |
6c8611c7 JA |
831 | td->last_usec[ddir] += val; |
832 | return td->last_usec[ddir]; | |
ff6bb260 | 833 | } else if (bps) { |
1a9bf814 JA |
834 | uint64_t bytes = td->rate_io_issue_bytes[ddir]; |
835 | uint64_t secs = bytes / bps; | |
836 | uint64_t remainder = bytes % bps; | |
837 | ||
50a8ce86 | 838 | return remainder * 1000000 / bps + secs * 1000000; |
e7b24047 JA |
839 | } |
840 | ||
841 | return 0; | |
50a8ce86 D |
842 | } |
843 | ||
f7942acd SK |
844 | static void init_thinktime(struct thread_data *td) |
845 | { | |
846 | if (td->o.thinktime_blocks_type == THINKTIME_BLOCKS_TYPE_COMPLETE) | |
847 | td->thinktime_blocks_counter = td->io_blocks; | |
848 | else | |
849 | td->thinktime_blocks_counter = td->io_issues; | |
850 | td->last_thinktime = td->epoch; | |
851 | td->last_thinktime_blocks = 0; | |
852 | } | |
853 | ||
7216b664 HQ |
854 | static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir, |
855 | struct timespec *time) | |
1a9bf814 JA |
856 | { |
857 | unsigned long long b; | |
f9fc7a27 | 858 | unsigned long long runtime_left; |
1a9bf814 JA |
859 | uint64_t total; |
860 | int left; | |
f7942acd SK |
861 | struct timespec now; |
862 | bool stall = false; | |
863 | ||
864 | if (td->o.thinktime_iotime) { | |
865 | fio_gettime(&now, NULL); | |
866 | if (utime_since(&td->last_thinktime, &now) | |
f9fc7a27 | 867 | >= td->o.thinktime_iotime) { |
f7942acd SK |
868 | stall = true; |
869 | } else if (!fio_option_is_set(&td->o, thinktime_blocks)) { | |
870 | /* | |
871 | * When thinktime_iotime is set and thinktime_blocks is | |
872 | * not set, skip the thinktime_blocks check, since | |
873 | * thinktime_blocks default value 1 does not work | |
874 | * together with thinktime_iotime. | |
875 | */ | |
876 | return; | |
877 | } | |
878 | ||
879 | } | |
1a9bf814 | 880 | |
33f42c20 | 881 | b = ddir_rw_sum(td->thinktime_blocks_counter); |
f7942acd SK |
882 | if (b >= td->last_thinktime_blocks + td->o.thinktime_blocks) |
883 | stall = true; | |
884 | ||
885 | if (!stall) | |
1a9bf814 JA |
886 | return; |
887 | ||
888 | io_u_quiesce(td); | |
889 | ||
f9fc7a27 AK |
890 | left = td->o.thinktime_spin; |
891 | if (td->o.timeout) { | |
892 | runtime_left = td->o.timeout - utime_since_now(&td->epoch); | |
893 | if (runtime_left < (unsigned long long)left) | |
894 | left = runtime_left; | |
895 | } | |
896 | ||
1a9bf814 | 897 | total = 0; |
f9fc7a27 AK |
898 | if (left) |
899 | total = usec_spin(left); | |
1a9bf814 | 900 | |
14adf6e3 MK |
901 | /* |
902 | * usec_spin() might run for slightly longer than intended in a VM | |
903 | * where the vCPU could get descheduled or the hypervisor could steal | |
904 | * CPU time. Ensure "left" doesn't become negative. | |
905 | */ | |
906 | if (total < td->o.thinktime) | |
907 | left = td->o.thinktime - total; | |
908 | else | |
909 | left = 0; | |
910 | ||
f9fc7a27 AK |
911 | if (td->o.timeout) { |
912 | runtime_left = td->o.timeout - utime_since_now(&td->epoch); | |
913 | if (runtime_left < (unsigned long long)left) | |
914 | left = runtime_left; | |
915 | } | |
916 | ||
1a9bf814 JA |
917 | if (left) |
918 | total += usec_sleep(td, left); | |
919 | ||
920 | /* | |
921 | * If we're ignoring thinktime for the rate, add the number of bytes | |
1aa39b0c JA |
922 | * we would have done while sleeping, minus one block to ensure we |
923 | * start issuing immediately after the sleep. | |
1a9bf814 | 924 | */ |
1aa39b0c | 925 | if (total && td->rate_bps[ddir] && td->o.rate_ign_think) { |
f2aedb17 | 926 | uint64_t missed = (td->rate_bps[ddir] * total) / 1000000ULL; |
62b858f6 JA |
927 | uint64_t bs = td->o.min_bs[ddir]; |
928 | uint64_t usperop = bs * 1000000ULL / td->rate_bps[ddir]; | |
f2aedb17 JA |
929 | uint64_t over; |
930 | ||
62b858f6 JA |
931 | if (usperop <= total) |
932 | over = bs; | |
f2aedb17 | 933 | else |
62b858f6 | 934 | over = (usperop - total) / usperop * -bs; |
f2aedb17 JA |
935 | |
936 | td->rate_io_issue_bytes[ddir] += (missed - over); | |
182cab24 JA |
937 | /* adjust for rate_process=poisson */ |
938 | td->last_usec[ddir] += total; | |
1aa39b0c | 939 | } |
7216b664 HQ |
940 | |
941 | if (time && should_check_rate(td)) | |
942 | fio_gettime(time, NULL); | |
f7942acd SK |
943 | |
944 | td->last_thinktime_blocks = b; | |
f9fc7a27 AK |
945 | if (td->o.thinktime_iotime) { |
946 | fio_gettime(&now, NULL); | |
f7942acd | 947 | td->last_thinktime = now; |
f9fc7a27 | 948 | } |
1a9bf814 JA |
949 | } |
950 | ||
2e1df07d JA |
951 | /* |
952 | * Main IO worker function. It retrieves io_u's to process and queues | |
953 | * and reaps them, checking for rate and errors along the way. | |
100f49f1 JA |
954 | * |
955 | * Returns number of bytes written and trimmed. | |
2e1df07d | 956 | */ |
95b03be0 | 957 | static void do_io(struct thread_data *td, uint64_t *bytes_done) |
2e1df07d JA |
958 | { |
959 | unsigned int i; | |
960 | int ret = 0; | |
c2703bf3 | 961 | uint64_t total_bytes, bytes_issued = 0; |
70c68076 | 962 | |
95b03be0 JA |
963 | for (i = 0; i < DDIR_RWDIR_CNT; i++) |
964 | bytes_done[i] = td->bytes_done[i]; | |
2e1df07d JA |
965 | |
966 | if (in_ramp_time(td)) | |
967 | td_set_runstate(td, TD_RAMP); | |
968 | else | |
969 | td_set_runstate(td, TD_RUNNING); | |
970 | ||
3e260a46 JA |
971 | lat_target_init(td); |
972 | ||
1e564979 JE |
973 | total_bytes = td->o.size; |
974 | /* | |
5be9bf09 | 975 | * Allow random overwrite workloads to write up to io_size |
1e564979 JE |
976 | * before starting verification phase as 'size' doesn't apply. |
977 | */ | |
978 | if (td_write(td) && td_random(td) && td->o.norandommap) | |
5be9bf09 | 979 | total_bytes = max(total_bytes, (uint64_t) td->o.io_size); |
78a6469c JA |
980 | /* |
981 | * If verify_backlog is enabled, we'll run the verify in this | |
982 | * handler as well. For that case, we may need up to twice the | |
983 | * amount of bytes. | |
984 | */ | |
78a6469c JA |
985 | if (td->o.verify != VERIFY_NONE && |
986 | (td_write(td) && td->o.verify_backlog)) | |
c2703bf3 JA |
987 | total_bytes += td->o.size; |
988 | ||
82a90686 | 989 | /* In trimwrite mode, each byte is trimmed and then written, so |
08996af4 AK |
990 | * allow total_bytes or number of ios to be twice as big */ |
991 | if (td_trimwrite(td)) { | |
0e4dd95c | 992 | total_bytes += td->total_io_size; |
08996af4 AK |
993 | td->o.number_ios *= 2; |
994 | } | |
0e4dd95c | 995 | |
f7078f7b | 996 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
e28dd2cf | 997 | (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) || |
c04e4661 | 998 | td->o.time_based) { |
8b6a404c | 999 | struct timespec comp_time; |
2e1df07d | 1000 | struct io_u *io_u; |
e9d512d8 | 1001 | int full; |
2e1df07d JA |
1002 | enum fio_ddir ddir; |
1003 | ||
c97f1ad6 JA |
1004 | check_update_rusage(td); |
1005 | ||
7d7803fa | 1006 | if (td->terminate || td->done) |
2e1df07d JA |
1007 | break; |
1008 | ||
8b6a404c | 1009 | update_ts_cache(td); |
2e1df07d | 1010 | |
8b6a404c VF |
1011 | if (runtime_exceeded(td, &td->ts_cache)) { |
1012 | __update_ts_cache(td); | |
1013 | if (runtime_exceeded(td, &td->ts_cache)) { | |
ebea2133 | 1014 | fio_mark_td_terminate(td); |
2e1df07d JA |
1015 | break; |
1016 | } | |
1017 | } | |
1018 | ||
9e684a49 DE |
1019 | if (flow_threshold_exceeded(td)) |
1020 | continue; | |
1021 | ||
f1a32461 JA |
1022 | /* |
1023 | * Break if we exceeded the bytes. The exception is time | |
1024 | * based runs, but we still need to break out of the loop | |
1025 | * for those to run verification, if enabled. | |
98e7161c | 1026 | * Jobs read from iolog do not use this stop condition. |
f1a32461 JA |
1027 | */ |
1028 | if (bytes_issued >= total_bytes && | |
98e7161c | 1029 | !td->o.read_iolog_file && |
f1a32461 JA |
1030 | (!td->o.time_based || |
1031 | (td->o.time_based && td->o.verify != VERIFY_NONE))) | |
20876c53 JC |
1032 | break; |
1033 | ||
2e1df07d | 1034 | io_u = get_io_u(td); |
002fe734 JA |
1035 | if (IS_ERR_OR_NULL(io_u)) { |
1036 | int err = PTR_ERR(io_u); | |
1037 | ||
1038 | io_u = NULL; | |
1a9bf814 | 1039 | ddir = DDIR_INVAL; |
002fe734 JA |
1040 | if (err == -EBUSY) { |
1041 | ret = FIO_Q_BUSY; | |
1042 | goto reap; | |
1043 | } | |
3e260a46 JA |
1044 | if (td->o.latency_target) |
1045 | goto reap; | |
2e1df07d | 1046 | break; |
3e260a46 | 1047 | } |
2e1df07d | 1048 | |
28921bc0 | 1049 | if (io_u->ddir == DDIR_WRITE && td->flags & TD_F_DO_VERIFY) { |
ad21b61d H |
1050 | if (!(io_u->flags & IO_U_F_PATTERN_DONE)) { |
1051 | io_u_set(td, io_u, IO_U_F_PATTERN_DONE); | |
1052 | io_u->numberio = td->io_issues[io_u->ddir]; | |
1053 | populate_verify_io_u(td, io_u); | |
1054 | } | |
28921bc0 | 1055 | } |
5c5c33c1 | 1056 | |
2e1df07d JA |
1057 | ddir = io_u->ddir; |
1058 | ||
1059 | /* | |
82af2a7c JA |
1060 | * Add verification end_io handler if: |
1061 | * - Asked to verify (!td_rw(td)) | |
1062 | * - Or the io_u is from our verify list (mixed write/ver) | |
2e1df07d JA |
1063 | */ |
1064 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && | |
82af2a7c | 1065 | ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) { |
c4b6117b | 1066 | |
ca09be4b JA |
1067 | if (verify_state_should_stop(td, io_u)) { |
1068 | put_io_u(td, io_u); | |
1069 | break; | |
1070 | } | |
1071 | ||
2e1df07d JA |
1072 | if (td->o.verify_async) |
1073 | io_u->end_io = verify_io_u_async; | |
1074 | else | |
1075 | io_u->end_io = verify_io_u; | |
1076 | td_set_runstate(td, TD_VERIFYING); | |
1077 | } else if (in_ramp_time(td)) | |
1078 | td_set_runstate(td, TD_RAMP); | |
1079 | else | |
1080 | td_set_runstate(td, TD_RUNNING); | |
1081 | ||
9a50c5c5 | 1082 | /* |
f9401285 JA |
1083 | * Always log IO before it's issued, so we know the specific |
1084 | * order of it. The logged unit will track when the IO has | |
1085 | * completed. | |
9a50c5c5 | 1086 | */ |
c4b6117b PV |
1087 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1088 | td->o.do_verify && | |
1089 | td->o.verify != VERIFY_NONE && | |
f9401285 | 1090 | !td->o.experimental_verify) |
c4b6117b PV |
1091 | log_io_piece(td, io_u); |
1092 | ||
a9da8ab2 | 1093 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
5fff9543 | 1094 | const unsigned long long blen = io_u->xfer_buflen; |
a43f4461 | 1095 | const enum fio_ddir __ddir = acct_ddir(io_u); |
0c5df5f9 | 1096 | |
a9da8ab2 JA |
1097 | if (td->error) |
1098 | break; | |
0c5df5f9 | 1099 | |
26de50cf JA |
1100 | workqueue_enqueue(&td->io_wq, &io_u->work); |
1101 | ret = FIO_Q_QUEUED; | |
50a8ce86 | 1102 | |
a43f4461 JA |
1103 | if (ddir_rw(__ddir)) { |
1104 | td->io_issues[__ddir]++; | |
1105 | td->io_issue_bytes[__ddir] += blen; | |
1106 | td->rate_io_issue_bytes[__ddir] += blen; | |
0c5df5f9 JA |
1107 | } |
1108 | ||
5b28b75b | 1109 | if (should_check_rate(td)) { |
a43f4461 | 1110 | td->rate_next_io_time[__ddir] = usec_for_io(td, __ddir); |
5b28b75b | 1111 | fio_gettime(&comp_time, NULL); |
1112 | } | |
50a8ce86 | 1113 | |
a9da8ab2 | 1114 | } else { |
08b2b386 | 1115 | ret = io_u_submit(td, io_u); |
2e1df07d | 1116 | |
50a8ce86 D |
1117 | if (should_check_rate(td)) |
1118 | td->rate_next_io_time[ddir] = usec_for_io(td, ddir); | |
1119 | ||
fd727d9d | 1120 | if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time)) |
a9da8ab2 | 1121 | break; |
2e1df07d | 1122 | |
a9da8ab2 JA |
1123 | /* |
1124 | * See if we need to complete some commands. Note that | |
1125 | * we can get BUSY even without IO queued, if the | |
1126 | * system is resource starved. | |
1127 | */ | |
3e260a46 | 1128 | reap: |
a9da8ab2 JA |
1129 | full = queue_full(td) || |
1130 | (ret == FIO_Q_BUSY && td->cur_depth); | |
82407585 | 1131 | if (full || io_in_polling(td)) |
a9da8ab2 JA |
1132 | ret = wait_for_completions(td, &comp_time); |
1133 | } | |
2e1df07d JA |
1134 | if (ret < 0) |
1135 | break; | |
33f42c20 | 1136 | |
0d8cc753 CL |
1137 | if (ddir_rw(ddir) && td->o.thinkcycles) |
1138 | cycles_spin(td->o.thinkcycles); | |
1139 | ||
33f42c20 | 1140 | if (ddir_rw(ddir) && td->o.thinktime) |
7216b664 | 1141 | handle_thinktime(td, ddir, &comp_time); |
33f42c20 | 1142 | |
55312f9f | 1143 | if (!ddir_rw_sum(td->bytes_done) && |
9b87f09b | 1144 | !td_ioengine_flagged(td, FIO_NOIO)) |
2e1df07d JA |
1145 | continue; |
1146 | ||
55312f9f JA |
1147 | if (!in_ramp_time(td) && should_check_rate(td)) { |
1148 | if (check_min_rate(td, &comp_time)) { | |
f9cafb12 | 1149 | if (exitall_on_terminate || td->o.exitall_error) |
64402a8a | 1150 | fio_terminate_threads(td->groupid, td->o.exit_what); |
2e1df07d JA |
1151 | td_verror(td, EIO, "check_min_rate"); |
1152 | break; | |
1153 | } | |
1154 | } | |
3e260a46 JA |
1155 | if (!in_ramp_time(td) && td->o.latency_target) |
1156 | lat_target_check(td); | |
2e1df07d JA |
1157 | } |
1158 | ||
c97f1ad6 JA |
1159 | check_update_rusage(td); |
1160 | ||
2e1df07d | 1161 | if (td->trim_entries) |
4e0a8fa2 | 1162 | log_err("fio: %lu trim entries leaked?\n", td->trim_entries); |
2e1df07d | 1163 | |
418f5399 | 1164 | if (td->o.fill_device && (td->error == ENOSPC || td->error == EDQUOT)) { |
2e1df07d | 1165 | td->error = 0; |
ebea2133 | 1166 | fio_mark_td_terminate(td); |
2e1df07d JA |
1167 | } |
1168 | if (!td->error) { | |
1169 | struct fio_file *f; | |
1170 | ||
a9da8ab2 JA |
1171 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) { |
1172 | workqueue_flush(&td->io_wq); | |
1173 | i = 0; | |
1174 | } else | |
1175 | i = td->cur_depth; | |
1176 | ||
2e1df07d | 1177 | if (i) { |
55312f9f | 1178 | ret = io_u_queued_complete(td, i); |
418f5399 MB |
1179 | if (td->o.fill_device && |
1180 | (td->error == ENOSPC || td->error == EDQUOT)) | |
2e1df07d JA |
1181 | td->error = 0; |
1182 | } | |
1183 | ||
00615bfb | 1184 | if (should_fsync(td) && (td->o.end_fsync || td->o.fsync_on_close)) { |
2e1df07d JA |
1185 | td_set_runstate(td, TD_FSYNCING); |
1186 | ||
1187 | for_each_file(td, f, i) { | |
61ee0f86 | 1188 | if (!fio_file_fsync(td, f)) |
2e1df07d | 1189 | continue; |
61ee0f86 JA |
1190 | |
1191 | log_err("fio: end_fsync failed for file %s\n", | |
1192 | f->file_name); | |
2e1df07d JA |
1193 | } |
1194 | } | |
5b28b75b | 1195 | } else { |
1196 | if (td->o.io_submit_mode == IO_MODE_OFFLOAD) | |
1197 | workqueue_flush(&td->io_wq); | |
2e1df07d | 1198 | cleanup_pending_aio(td); |
5b28b75b | 1199 | } |
2e1df07d JA |
1200 | |
1201 | /* | |
1202 | * stop job if we failed doing any IO | |
1203 | */ | |
342f4be4 | 1204 | if (!ddir_rw_sum(td->this_io_bytes)) |
2e1df07d | 1205 | td->done = 1; |
100f49f1 | 1206 | |
95b03be0 JA |
1207 | for (i = 0; i < DDIR_RWDIR_CNT; i++) |
1208 | bytes_done[i] = td->bytes_done[i] - bytes_done[i]; | |
2e1df07d JA |
1209 | } |
1210 | ||
94a6e1bb JA |
1211 | static void free_file_completion_logging(struct thread_data *td) |
1212 | { | |
1213 | struct fio_file *f; | |
1214 | unsigned int i; | |
1215 | ||
1216 | for_each_file(td, f, i) { | |
1217 | if (!f->last_write_comp) | |
1218 | break; | |
1219 | sfree(f->last_write_comp); | |
1220 | } | |
1221 | } | |
1222 | ||
1223 | static int init_file_completion_logging(struct thread_data *td, | |
1224 | unsigned int depth) | |
1225 | { | |
1226 | struct fio_file *f; | |
1227 | unsigned int i; | |
1228 | ||
1229 | if (td->o.verify == VERIFY_NONE || !td->o.verify_state_save) | |
1230 | return 0; | |
1231 | ||
1232 | for_each_file(td, f, i) { | |
1233 | f->last_write_comp = scalloc(depth, sizeof(uint64_t)); | |
1234 | if (!f->last_write_comp) | |
1235 | goto cleanup; | |
1236 | } | |
1237 | ||
1238 | return 0; | |
1239 | ||
1240 | cleanup: | |
1241 | free_file_completion_logging(td); | |
1242 | log_err("fio: failed to alloc write comp data\n"); | |
1243 | return 1; | |
1244 | } | |
1245 | ||
2e1df07d JA |
1246 | static void cleanup_io_u(struct thread_data *td) |
1247 | { | |
2e1df07d JA |
1248 | struct io_u *io_u; |
1249 | ||
2ae0b204 | 1250 | while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) { |
c73ed246 JA |
1251 | |
1252 | if (td->io_ops->io_u_free) | |
1253 | td->io_ops->io_u_free(td, io_u); | |
1254 | ||
3114b675 | 1255 | fio_memfree(io_u, sizeof(*io_u), td_offload_overlap(td)); |
2e1df07d JA |
1256 | } |
1257 | ||
1258 | free_io_mem(td); | |
2ae0b204 JA |
1259 | |
1260 | io_u_rexit(&td->io_u_requeues); | |
3114b675 VF |
1261 | io_u_qexit(&td->io_u_freelist, false); |
1262 | io_u_qexit(&td->io_u_all, td_offload_overlap(td)); | |
ca09be4b | 1263 | |
94a6e1bb | 1264 | free_file_completion_logging(td); |
2e1df07d JA |
1265 | } |
1266 | ||
1267 | static int init_io_u(struct thread_data *td) | |
1268 | { | |
1269 | struct io_u *io_u; | |
2e1df07d | 1270 | int cl_align, i, max_units; |
71e6e5a2 | 1271 | int err; |
2e1df07d JA |
1272 | |
1273 | max_units = td->o.iodepth; | |
59d8d0f5 | 1274 | |
2ae0b204 | 1275 | err = 0; |
34851ad5 | 1276 | err += !io_u_rinit(&td->io_u_requeues, td->o.iodepth); |
3114b675 VF |
1277 | err += !io_u_qinit(&td->io_u_freelist, td->o.iodepth, false); |
1278 | err += !io_u_qinit(&td->io_u_all, td->o.iodepth, td_offload_overlap(td)); | |
2ae0b204 JA |
1279 | |
1280 | if (err) { | |
1281 | log_err("fio: failed setting up IO queues\n"); | |
1282 | return 1; | |
1283 | } | |
1284 | ||
71e6e5a2 AK |
1285 | cl_align = os_cache_line_size(); |
1286 | ||
1287 | for (i = 0; i < max_units; i++) { | |
1288 | void *ptr; | |
1289 | ||
1290 | if (td->terminate) | |
1291 | return 1; | |
1292 | ||
3114b675 | 1293 | ptr = fio_memalign(cl_align, sizeof(*io_u), td_offload_overlap(td)); |
71e6e5a2 AK |
1294 | if (!ptr) { |
1295 | log_err("fio: unable to allocate aligned memory\n"); | |
ab43b8cb | 1296 | return 1; |
71e6e5a2 AK |
1297 | } |
1298 | ||
1299 | io_u = ptr; | |
1300 | memset(io_u, 0, sizeof(*io_u)); | |
1301 | INIT_FLIST_HEAD(&io_u->verify_list); | |
1302 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); | |
1303 | ||
1304 | io_u->index = i; | |
1305 | io_u->flags = IO_U_F_FREE; | |
1306 | io_u_qpush(&td->io_u_freelist, io_u); | |
1307 | ||
1308 | /* | |
1309 | * io_u never leaves this stack, used for iteration of all | |
1310 | * io_u buffers. | |
1311 | */ | |
1312 | io_u_qpush(&td->io_u_all, io_u); | |
1313 | ||
1314 | if (td->io_ops->io_u_init) { | |
1315 | int ret = td->io_ops->io_u_init(td, io_u); | |
1316 | ||
1317 | if (ret) { | |
1318 | log_err("fio: failed to init engine data: %d\n", ret); | |
1319 | return 1; | |
1320 | } | |
1321 | } | |
1322 | } | |
1323 | ||
6d8fe6e8 SK |
1324 | if (init_io_u_buffers(td)) |
1325 | return 1; | |
71e6e5a2 AK |
1326 | |
1327 | if (init_file_completion_logging(td, max_units)) | |
1328 | return 1; | |
1329 | ||
1330 | return 0; | |
1331 | } | |
1332 | ||
1333 | int init_io_u_buffers(struct thread_data *td) | |
1334 | { | |
1335 | struct io_u *io_u; | |
b3251e31 | 1336 | unsigned long long max_bs, min_write, trim_bs = 0; |
71e6e5a2 AK |
1337 | int i, max_units; |
1338 | int data_xfer = 1; | |
1339 | char *p; | |
1340 | ||
1341 | max_units = td->o.iodepth; | |
1342 | max_bs = td_max_bs(td); | |
1343 | min_write = td->o.min_bs[DDIR_WRITE]; | |
1344 | td->orig_buffer_size = (unsigned long long) max_bs | |
1345 | * (unsigned long long) max_units; | |
1346 | ||
b3251e31 AK |
1347 | if (td_trim(td) && td->o.num_range > 1) { |
1348 | trim_bs = td->o.num_range * sizeof(struct trim_range); | |
1349 | td->orig_buffer_size = trim_bs | |
1350 | * (unsigned long long) max_units; | |
1351 | } | |
1352 | ||
1353 | /* | |
1354 | * For reads, writes, and multi-range trim operations we need a | |
1355 | * data buffer | |
1356 | */ | |
1357 | if (td_ioengine_flagged(td, FIO_NOIO) || | |
1358 | !(td_read(td) || td_write(td) || (td_trim(td) && td->o.num_range > 1))) | |
71e6e5a2 AK |
1359 | data_xfer = 0; |
1360 | ||
fd8a09b8 | 1361 | /* |
1362 | * if we may later need to do address alignment, then add any | |
1363 | * possible adjustment here so that we don't cause a buffer | |
1364 | * overflow later. this adjustment may be too much if we get | |
1365 | * lucky and the allocator gives us an aligned address. | |
1366 | */ | |
a25ba6c6 | 1367 | if (td->o.odirect || td->o.mem_align || |
9b87f09b | 1368 | td_ioengine_flagged(td, FIO_RAWIO)) |
fd8a09b8 | 1369 | td->orig_buffer_size += page_mask + td->o.mem_align; |
1370 | ||
2e1df07d | 1371 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { |
5fff9543 | 1372 | unsigned long long bs; |
2e1df07d JA |
1373 | |
1374 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; | |
1375 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
1376 | } | |
1377 | ||
1378 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { | |
1379 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
1380 | return 1; | |
1381 | } | |
1382 | ||
59d8d0f5 | 1383 | if (data_xfer && allocate_io_mem(td)) |
2e1df07d JA |
1384 | return 1; |
1385 | ||
a25ba6c6 | 1386 | if (td->o.odirect || td->o.mem_align || |
9b87f09b | 1387 | td_ioengine_flagged(td, FIO_RAWIO)) |
248c9436 | 1388 | p = PTR_ALIGN(td->orig_buffer, page_mask) + td->o.mem_align; |
2e1df07d JA |
1389 | else |
1390 | p = td->orig_buffer; | |
1391 | ||
2e1df07d | 1392 | for (i = 0; i < max_units; i++) { |
71e6e5a2 | 1393 | io_u = td->io_u_all.io_us[i]; |
2e1df07d JA |
1394 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
1395 | ||
59d8d0f5 | 1396 | if (data_xfer) { |
2e1df07d JA |
1397 | io_u->buf = p; |
1398 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); | |
1399 | ||
1400 | if (td_write(td)) | |
9c42684e | 1401 | io_u_fill_buffer(td, io_u, min_write, max_bs); |
2e1df07d JA |
1402 | if (td_write(td) && td->o.verify_pattern_bytes) { |
1403 | /* | |
1404 | * Fill the buffer with the pattern if we are | |
1405 | * going to be doing writes. | |
1406 | */ | |
ce35b1ec | 1407 | fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
2e1df07d JA |
1408 | } |
1409 | } | |
b3251e31 AK |
1410 | if (td_trim(td) && td->o.num_range > 1) |
1411 | p += trim_bs; | |
1412 | else | |
1413 | p += max_bs; | |
2e1df07d JA |
1414 | } |
1415 | ||
1416 | return 0; | |
1417 | } | |
1418 | ||
5592e992 | 1419 | #ifdef FIO_HAVE_IOSCHED_SWITCH |
bd4d9bdc | 1420 | /* |
5592e992 | 1421 | * These functions are Linux specific. |
bd4d9bdc TK |
1422 | * FIO_HAVE_IOSCHED_SWITCH enabled currently means it's Linux. |
1423 | */ | |
5592e992 | 1424 | static int set_ioscheduler(struct thread_data *td, struct fio_file *file) |
2e1df07d | 1425 | { |
178ef27a | 1426 | char tmp[256], tmp2[128], *p; |
2e1df07d JA |
1427 | FILE *f; |
1428 | int ret; | |
1429 | ||
5592e992 DLM |
1430 | assert(file->du && file->du->sysfs_root); |
1431 | sprintf(tmp, "%s/queue/scheduler", file->du->sysfs_root); | |
2e1df07d JA |
1432 | |
1433 | f = fopen(tmp, "r+"); | |
1434 | if (!f) { | |
1435 | if (errno == ENOENT) { | |
1436 | log_err("fio: os or kernel doesn't support IO scheduler" | |
1437 | " switching\n"); | |
1438 | return 0; | |
1439 | } | |
1440 | td_verror(td, errno, "fopen iosched"); | |
1441 | return 1; | |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * Set io scheduler. | |
1446 | */ | |
1447 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); | |
1448 | if (ferror(f) || ret != 1) { | |
1449 | td_verror(td, errno, "fwrite"); | |
1450 | fclose(f); | |
1451 | return 1; | |
1452 | } | |
1453 | ||
1454 | rewind(f); | |
1455 | ||
1456 | /* | |
1457 | * Read back and check that the selected scheduler is now the default. | |
1458 | */ | |
0e23d456 | 1459 | ret = fread(tmp, 1, sizeof(tmp) - 1, f); |
2e1df07d JA |
1460 | if (ferror(f) || ret < 0) { |
1461 | td_verror(td, errno, "fread"); | |
1462 | fclose(f); | |
1463 | return 1; | |
1464 | } | |
0e23d456 | 1465 | tmp[ret] = '\0'; |
b44b9e45 | 1466 | /* |
178ef27a BVA |
1467 | * either a list of io schedulers or "none\n" is expected. Strip the |
1468 | * trailing newline. | |
b44b9e45 | 1469 | */ |
178ef27a BVA |
1470 | p = tmp; |
1471 | strsep(&p, "\n"); | |
49c6f33d | 1472 | |
67f28a8c TK |
1473 | /* |
1474 | * Write to "none" entry doesn't fail, so check the result here. | |
1475 | */ | |
1476 | if (!strcmp(tmp, "none")) { | |
1477 | log_err("fio: io scheduler is not tunable\n"); | |
1478 | fclose(f); | |
1479 | return 0; | |
1480 | } | |
2e1df07d JA |
1481 | |
1482 | sprintf(tmp2, "[%s]", td->o.ioscheduler); | |
1483 | if (!strstr(tmp, tmp2)) { | |
da44720c | 1484 | log_err("fio: unable to set io scheduler to %s\n", td->o.ioscheduler); |
2e1df07d JA |
1485 | td_verror(td, EINVAL, "iosched_switch"); |
1486 | fclose(f); | |
1487 | return 1; | |
1488 | } | |
1489 | ||
1490 | fclose(f); | |
1491 | return 0; | |
5592e992 DLM |
1492 | } |
1493 | ||
1494 | static int switch_ioscheduler(struct thread_data *td) | |
1495 | { | |
1496 | struct fio_file *f; | |
1497 | unsigned int i; | |
1498 | int ret = 0; | |
1499 | ||
1500 | if (td_ioengine_flagged(td, FIO_DISKLESSIO)) | |
1501 | return 0; | |
1502 | ||
1503 | assert(td->files && td->files[0]); | |
1504 | ||
1505 | for_each_file(td, f, i) { | |
1506 | ||
1507 | /* Only consider regular files and block device files */ | |
1508 | switch (f->filetype) { | |
1509 | case FIO_TYPE_FILE: | |
1510 | case FIO_TYPE_BLOCK: | |
1511 | /* | |
1512 | * Make sure that the device hosting the file could | |
1513 | * be determined. | |
1514 | */ | |
1515 | if (!f->du) | |
1516 | continue; | |
1517 | break; | |
1518 | case FIO_TYPE_CHAR: | |
1519 | case FIO_TYPE_PIPE: | |
1520 | default: | |
1521 | continue; | |
1522 | } | |
1523 | ||
1524 | ret = set_ioscheduler(td, f); | |
1525 | if (ret) | |
1526 | return ret; | |
1527 | } | |
1528 | ||
1529 | return 0; | |
1530 | } | |
1531 | ||
db815fb3 | 1532 | #else |
5592e992 DLM |
1533 | |
1534 | static int switch_ioscheduler(struct thread_data *td) | |
1535 | { | |
db815fb3 | 1536 | return 0; |
2e1df07d JA |
1537 | } |
1538 | ||
5592e992 DLM |
1539 | #endif /* FIO_HAVE_IOSCHED_SWITCH */ |
1540 | ||
e39c0676 | 1541 | static bool keep_running(struct thread_data *td) |
2e1df07d | 1542 | { |
77731b29 JA |
1543 | unsigned long long limit; |
1544 | ||
2e1df07d | 1545 | if (td->done) |
e39c0676 | 1546 | return false; |
98413be7 SW |
1547 | if (td->terminate) |
1548 | return false; | |
2e1df07d | 1549 | if (td->o.time_based) |
e39c0676 | 1550 | return true; |
2e1df07d JA |
1551 | if (td->o.loops) { |
1552 | td->o.loops--; | |
e39c0676 | 1553 | return true; |
2e1df07d | 1554 | } |
3939fe85 | 1555 | if (exceeds_number_ios(td)) |
e39c0676 | 1556 | return false; |
26251d8d | 1557 | |
5be9bf09 TK |
1558 | if (td->o.io_size) |
1559 | limit = td->o.io_size; | |
77731b29 JA |
1560 | else |
1561 | limit = td->o.size; | |
1562 | ||
1563 | if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) { | |
5bd5f71a JA |
1564 | uint64_t diff; |
1565 | ||
1566 | /* | |
589d5d4f | 1567 | * If the difference is less than the maximum IO size, we |
5bd5f71a JA |
1568 | * are done. |
1569 | */ | |
77731b29 | 1570 | diff = limit - ddir_rw_sum(td->io_bytes); |
74f4b020 | 1571 | if (diff < td_max_bs(td)) |
e39c0676 | 1572 | return false; |
5bd5f71a | 1573 | |
5be9bf09 | 1574 | if (fio_files_done(td) && !td->o.io_size) |
e39c0676 | 1575 | return false; |
002fe734 | 1576 | |
e39c0676 | 1577 | return true; |
5bd5f71a | 1578 | } |
2e1df07d | 1579 | |
e39c0676 | 1580 | return false; |
2e1df07d JA |
1581 | } |
1582 | ||
b7e4a53c BVA |
1583 | static int exec_string(struct thread_options *o, const char *string, |
1584 | const char *mode) | |
2e1df07d | 1585 | { |
f491a907 | 1586 | int ret; |
2e1df07d JA |
1587 | char *str; |
1588 | ||
b7e4a53c BVA |
1589 | if (asprintf(&str, "%s > %s.%s.txt 2>&1", string, o->name, mode) < 0) |
1590 | return -1; | |
2e1df07d | 1591 | |
b7e4a53c BVA |
1592 | log_info("%s : Saving output of %s in %s.%s.txt\n", o->name, mode, |
1593 | o->name, mode); | |
2e1df07d JA |
1594 | ret = system(str); |
1595 | if (ret == -1) | |
1596 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1597 | ||
1598 | free(str); | |
1599 | return ret; | |
1600 | } | |
1601 | ||
62167762 JC |
1602 | /* |
1603 | * Dry run to compute correct state of numberio for verification. | |
1604 | */ | |
1605 | static uint64_t do_dry_run(struct thread_data *td) | |
1606 | { | |
62167762 JC |
1607 | td_set_runstate(td, TD_RUNNING); |
1608 | ||
1609 | while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || | |
e28dd2cf | 1610 | (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) { |
62167762 JC |
1611 | struct io_u *io_u; |
1612 | int ret; | |
1613 | ||
1614 | if (td->terminate || td->done) | |
1615 | break; | |
1616 | ||
1617 | io_u = get_io_u(td); | |
99955d3d | 1618 | if (IS_ERR_OR_NULL(io_u)) |
62167762 JC |
1619 | break; |
1620 | ||
1651e431 | 1621 | io_u_set(td, io_u, IO_U_F_FLIGHT); |
62167762 JC |
1622 | io_u->error = 0; |
1623 | io_u->resid = 0; | |
1624 | if (ddir_rw(acct_ddir(io_u))) | |
1625 | td->io_issues[acct_ddir(io_u)]++; | |
1626 | if (ddir_rw(io_u->ddir)) { | |
1627 | io_u_mark_depth(td, 1); | |
1628 | td->ts.total_io_u[io_u->ddir]++; | |
1629 | } | |
1630 | ||
2e63e96b PV |
1631 | if (td_write(td) && io_u->ddir == DDIR_WRITE && |
1632 | td->o.do_verify && | |
1633 | td->o.verify != VERIFY_NONE && | |
1634 | !td->o.experimental_verify) | |
1635 | log_io_piece(td, io_u); | |
1636 | ||
55312f9f | 1637 | ret = io_u_sync_complete(td, io_u); |
62167762 JC |
1638 | (void) ret; |
1639 | } | |
1640 | ||
55312f9f | 1641 | return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM]; |
62167762 JC |
1642 | } |
1643 | ||
24660963 JA |
1644 | struct fork_data { |
1645 | struct thread_data *td; | |
1646 | struct sk_out *sk_out; | |
1647 | }; | |
1648 | ||
2e1df07d JA |
1649 | /* |
1650 | * Entry point for the thread based jobs. The process based jobs end up | |
1651 | * here as well, after a little setup. | |
1652 | */ | |
1653 | static void *thread_main(void *data) | |
1654 | { | |
24660963 | 1655 | struct fork_data *fd = data; |
95603b74 | 1656 | unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, }; |
24660963 | 1657 | struct thread_data *td = fd->td; |
4896473e | 1658 | struct thread_options *o = &td->o; |
24660963 | 1659 | struct sk_out *sk_out = fd->sk_out; |
48366f3a | 1660 | uint64_t bytes_done[DDIR_RWDIR_CNT]; |
ca0122d8 | 1661 | int deadlock_loop_cnt; |
ff3aa922 | 1662 | bool clear_state; |
83276370 | 1663 | int ret; |
2e1df07d | 1664 | |
24660963 JA |
1665 | sk_out_assign(sk_out); |
1666 | free(fd); | |
1667 | ||
4896473e | 1668 | if (!o->use_thread) { |
2e1df07d JA |
1669 | setsid(); |
1670 | td->pid = getpid(); | |
1671 | } else | |
1672 | td->pid = gettid(); | |
1673 | ||
a0eba820 | 1674 | fio_local_clock_init(); |
5d879392 | 1675 | |
2e1df07d JA |
1676 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
1677 | ||
122c7725 JA |
1678 | if (is_backend) |
1679 | fio_server_send_start(td); | |
1680 | ||
2e1df07d JA |
1681 | INIT_FLIST_HEAD(&td->io_log_list); |
1682 | INIT_FLIST_HEAD(&td->io_hist_list); | |
1683 | INIT_FLIST_HEAD(&td->verify_list); | |
1684 | INIT_FLIST_HEAD(&td->trim_list); | |
2e1df07d JA |
1685 | td->io_hist_tree = RB_ROOT; |
1686 | ||
34febb23 | 1687 | ret = mutex_cond_init_pshared(&td->io_u_lock, &td->free_cond); |
f9e5b5ee | 1688 | if (ret) { |
34febb23 | 1689 | td_verror(td, ret, "mutex_cond_init_pshared"); |
f9e5b5ee JK |
1690 | goto err; |
1691 | } | |
34febb23 | 1692 | ret = cond_init_pshared(&td->verify_cond); |
f9e5b5ee | 1693 | if (ret) { |
34febb23 | 1694 | td_verror(td, ret, "mutex_cond_pshared"); |
f9e5b5ee JK |
1695 | goto err; |
1696 | } | |
2e1df07d JA |
1697 | |
1698 | td_set_runstate(td, TD_INITIALIZED); | |
971caeb1 BVA |
1699 | dprint(FD_MUTEX, "up startup_sem\n"); |
1700 | fio_sem_up(startup_sem); | |
1701 | dprint(FD_MUTEX, "wait on td->sem\n"); | |
1702 | fio_sem_down(td->sem); | |
1703 | dprint(FD_MUTEX, "done waiting on td->sem\n"); | |
2e1df07d | 1704 | |
2e1df07d JA |
1705 | /* |
1706 | * A new gid requires privilege, so we need to do this before setting | |
1707 | * the uid. | |
1708 | */ | |
4896473e | 1709 | if (o->gid != -1U && setgid(o->gid)) { |
2e1df07d JA |
1710 | td_verror(td, errno, "setgid"); |
1711 | goto err; | |
1712 | } | |
4896473e | 1713 | if (o->uid != -1U && setuid(o->uid)) { |
2e1df07d JA |
1714 | td_verror(td, errno, "setuid"); |
1715 | goto err; | |
1716 | } | |
1717 | ||
07776127 JA |
1718 | td_zone_gen_index(td); |
1719 | ||
2354d810 JA |
1720 | /* |
1721 | * Do this early, we don't want the compress threads to be limited | |
1722 | * to the same CPUs as the IO workers. So do this before we set | |
1723 | * any potential CPU affinity | |
1724 | */ | |
1725 | if (iolog_compress_init(td, sk_out)) | |
1726 | goto err; | |
1727 | ||
2e1df07d JA |
1728 | /* |
1729 | * If we have a gettimeofday() thread, make sure we exclude that | |
1730 | * thread from this job | |
1731 | */ | |
4896473e JA |
1732 | if (o->gtod_cpu) |
1733 | fio_cpu_clear(&o->cpumask, o->gtod_cpu); | |
2e1df07d JA |
1734 | |
1735 | /* | |
1736 | * Set affinity first, in case it has an impact on the memory | |
1737 | * allocations. | |
1738 | */ | |
b2a9e649 | 1739 | if (fio_option_is_set(o, cpumask)) { |
c2acfbac | 1740 | if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) { |
30cb4c65 | 1741 | ret = fio_cpus_split(&o->cpumask, td->thread_number - 1); |
c2acfbac JA |
1742 | if (!ret) { |
1743 | log_err("fio: no CPUs set\n"); | |
1744 | log_err("fio: Try increasing number of available CPUs\n"); | |
1745 | td_verror(td, EINVAL, "cpus_split"); | |
1746 | goto err; | |
1747 | } | |
1748 | } | |
28727df7 JA |
1749 | ret = fio_setaffinity(td->pid, o->cpumask); |
1750 | if (ret == -1) { | |
4896473e JA |
1751 | td_verror(td, errno, "cpu_set_affinity"); |
1752 | goto err; | |
1753 | } | |
2e1df07d JA |
1754 | } |
1755 | ||
67bf9823 | 1756 | #ifdef CONFIG_LIBNUMA |
d0b937ed | 1757 | /* numa node setup */ |
b2a9e649 JA |
1758 | if (fio_option_is_set(o, numa_cpunodes) || |
1759 | fio_option_is_set(o, numa_memnodes)) { | |
43522848 | 1760 | struct bitmask *mask; |
d0b937ed YR |
1761 | |
1762 | if (numa_available() < 0) { | |
1763 | td_verror(td, errno, "Does not support NUMA API\n"); | |
1764 | goto err; | |
1765 | } | |
1766 | ||
b2a9e649 | 1767 | if (fio_option_is_set(o, numa_cpunodes)) { |
43522848 DG |
1768 | mask = numa_parse_nodestring(o->numa_cpunodes); |
1769 | ret = numa_run_on_node_mask(mask); | |
1770 | numa_free_nodemask(mask); | |
d0b937ed YR |
1771 | if (ret == -1) { |
1772 | td_verror(td, errno, \ | |
1773 | "numa_run_on_node_mask failed\n"); | |
1774 | goto err; | |
1775 | } | |
1776 | } | |
1777 | ||
b2a9e649 | 1778 | if (fio_option_is_set(o, numa_memnodes)) { |
43522848 DG |
1779 | mask = NULL; |
1780 | if (o->numa_memnodes) | |
1781 | mask = numa_parse_nodestring(o->numa_memnodes); | |
1782 | ||
4896473e | 1783 | switch (o->numa_mem_mode) { |
d0b937ed | 1784 | case MPOL_INTERLEAVE: |
43522848 | 1785 | numa_set_interleave_mask(mask); |
d0b937ed YR |
1786 | break; |
1787 | case MPOL_BIND: | |
43522848 | 1788 | numa_set_membind(mask); |
d0b937ed YR |
1789 | break; |
1790 | case MPOL_LOCAL: | |
1791 | numa_set_localalloc(); | |
1792 | break; | |
1793 | case MPOL_PREFERRED: | |
4896473e | 1794 | numa_set_preferred(o->numa_mem_prefer_node); |
d0b937ed YR |
1795 | break; |
1796 | case MPOL_DEFAULT: | |
1797 | default: | |
1798 | break; | |
1799 | } | |
1800 | ||
43522848 DG |
1801 | if (mask) |
1802 | numa_free_nodemask(mask); | |
1803 | ||
d0b937ed YR |
1804 | } |
1805 | } | |
1806 | #endif | |
1807 | ||
9a3f1100 JA |
1808 | if (fio_pin_memory(td)) |
1809 | goto err; | |
1810 | ||
2e1df07d JA |
1811 | /* |
1812 | * May alter parameters that init_io_u() will use, so we need to | |
1813 | * do this first. | |
1814 | */ | |
b153f94a | 1815 | if (!init_iolog(td)) |
2e1df07d JA |
1816 | goto err; |
1817 | ||
61b20c5c NC |
1818 | /* ioprio_set() has to be done before td_io_init() */ |
1819 | if (fio_option_is_set(o, ioprio) || | |
860462da DLM |
1820 | fio_option_is_set(o, ioprio_class) || |
1821 | fio_option_is_set(o, ioprio_hint)) { | |
b3048c1b | 1822 | ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, |
860462da | 1823 | o->ioprio, o->ioprio_hint); |
61b20c5c NC |
1824 | if (ret == -1) { |
1825 | td_verror(td, errno, "ioprio_set"); | |
1826 | goto err; | |
1827 | } | |
860462da DLM |
1828 | td->ioprio = ioprio_value(o->ioprio_class, o->ioprio, |
1829 | o->ioprio_hint); | |
2df7f081 | 1830 | td->ts.ioprio = td->ioprio; |
61b20c5c NC |
1831 | } |
1832 | ||
08dc3bd5 JA |
1833 | if (td_io_init(td)) |
1834 | goto err; | |
1835 | ||
b8d07a36 | 1836 | if (td_ioengine_flagged(td, FIO_SYNCIO) && td->o.iodepth > 1 && td->o.io_submit_mode != IO_MODE_OFFLOAD) { |
349d5f6c AD |
1837 | log_info("note: both iodepth >= 1 and synchronous I/O engine " |
1838 | "are selected, queue depth will be capped at 1\n"); | |
1839 | } | |
1840 | ||
2e1df07d JA |
1841 | if (init_io_u(td)) |
1842 | goto err; | |
1843 | ||
2041bd34 JA |
1844 | if (td->io_ops->post_init && td->io_ops->post_init(td)) |
1845 | goto err; | |
1846 | ||
4896473e | 1847 | if (o->verify_async && verify_async_init(td)) |
2e1df07d JA |
1848 | goto err; |
1849 | ||
4896473e | 1850 | if (o->cgroup && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
2e1df07d JA |
1851 | goto err; |
1852 | ||
649c10c5 | 1853 | errno = 0; |
4896473e | 1854 | if (nice(o->nice) == -1 && errno != 0) { |
2e1df07d JA |
1855 | td_verror(td, errno, "nice"); |
1856 | goto err; | |
1857 | } | |
1858 | ||
4896473e | 1859 | if (o->ioscheduler && switch_ioscheduler(td)) |
2e1df07d JA |
1860 | goto err; |
1861 | ||
4896473e | 1862 | if (!o->create_serialize && setup_files(td)) |
2e1df07d JA |
1863 | goto err; |
1864 | ||
4d832322 | 1865 | if (!init_random_map(td)) |
2e1df07d JA |
1866 | goto err; |
1867 | ||
0518f09f | 1868 | if (o->exec_prerun && exec_string(o, o->exec_prerun, "prerun")) |
4896473e | 1869 | goto err; |
2e1df07d | 1870 | |
c139f3b4 JA |
1871 | if (o->pre_read && !pre_read_files(td)) |
1872 | goto err; | |
2e1df07d | 1873 | |
dc5bfbb2 JA |
1874 | fio_verify_init(td); |
1875 | ||
24660963 | 1876 | if (rate_submit_init(td, sk_out)) |
a9da8ab2 JA |
1877 | goto err; |
1878 | ||
12d325ca | 1879 | set_epoch_time(td, o->log_alternate_epoch_clock_id, o->job_start_clock_id); |
44404c5a | 1880 | fio_getrusage(&td->ru_start); |
ac28d905 JA |
1881 | memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch)); |
1882 | memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch)); | |
16e56d25 | 1883 | memcpy(&td->ss.prev_time, &td->epoch, sizeof(td->epoch)); |
ac28d905 | 1884 | |
f7942acd SK |
1885 | init_thinktime(td); |
1886 | ||
ac28d905 JA |
1887 | if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] || |
1888 | o->ratemin[DDIR_TRIM]) { | |
f5204b8a | 1889 | memcpy(&td->last_rate_check_time[DDIR_READ], &td->bw_sample_time, |
ac28d905 | 1890 | sizeof(td->bw_sample_time)); |
f5204b8a | 1891 | memcpy(&td->last_rate_check_time[DDIR_WRITE], &td->bw_sample_time, |
ac28d905 | 1892 | sizeof(td->bw_sample_time)); |
f5204b8a | 1893 | memcpy(&td->last_rate_check_time[DDIR_TRIM], &td->bw_sample_time, |
ac28d905 JA |
1894 | sizeof(td->bw_sample_time)); |
1895 | } | |
1896 | ||
48366f3a | 1897 | memset(bytes_done, 0, sizeof(bytes_done)); |
c16035aa | 1898 | clear_state = false; |
48366f3a | 1899 | |
2e1df07d | 1900 | while (keep_running(td)) { |
100f49f1 JA |
1901 | uint64_t verify_bytes; |
1902 | ||
2e1df07d | 1903 | fio_gettime(&td->start, NULL); |
8b6a404c | 1904 | memcpy(&td->ts_cache, &td->start, sizeof(td->start)); |
2e1df07d | 1905 | |
39c1c323 | 1906 | if (clear_state) { |
ac28d905 | 1907 | clear_io_state(td, 0); |
2e1df07d | 1908 | |
39c1c323 | 1909 | if (o->unlink_each_loop && unlink_all_files(td)) |
1910 | break; | |
1911 | } | |
1912 | ||
2e1df07d JA |
1913 | prune_io_piece_log(td); |
1914 | ||
dcf9844e | 1915 | if (td->o.verify_only && td_write(td)) |
62167762 | 1916 | verify_bytes = do_dry_run(td); |
095196b1 | 1917 | else { |
50eaefa5 H |
1918 | if (!td->o.rand_repeatable) |
1919 | /* save verify rand state to replay hdr seeds later at verify */ | |
1920 | frand_copy(&td->verify_state_last_do_io, &td->verify_state); | |
95b03be0 | 1921 | do_io(td, bytes_done); |
50eaefa5 H |
1922 | if (!td->o.rand_repeatable) |
1923 | frand_copy(&td->verify_state, &td->verify_state_last_do_io); | |
95b03be0 | 1924 | if (!ddir_rw_sum(bytes_done)) { |
095196b1 | 1925 | fio_mark_td_terminate(td); |
95b03be0 JA |
1926 | verify_bytes = 0; |
1927 | } else { | |
1928 | verify_bytes = bytes_done[DDIR_WRITE] + | |
1929 | bytes_done[DDIR_TRIM]; | |
1930 | } | |
095196b1 | 1931 | } |
2e1df07d | 1932 | |
49f0b8e0 JA |
1933 | /* |
1934 | * If we took too long to shut down, the main thread could | |
1935 | * already consider us reaped/exited. If that happens, break | |
1936 | * out and clean up. | |
1937 | */ | |
1938 | if (td->runstate >= TD_EXITED) | |
1939 | break; | |
1940 | ||
c16035aa | 1941 | clear_state = true; |
2e1df07d | 1942 | |
40c666c8 JA |
1943 | /* |
1944 | * Make sure we've successfully updated the rusage stats | |
1945 | * before waiting on the stat mutex. Otherwise we could have | |
1946 | * the stat thread holding stat mutex and waiting for | |
1947 | * the rusage_sem, which would never get upped because | |
1948 | * this thread is waiting for the stat mutex. | |
1949 | */ | |
ca0122d8 | 1950 | deadlock_loop_cnt = 0; |
39d13e67 JA |
1951 | do { |
1952 | check_update_rusage(td); | |
971caeb1 | 1953 | if (!fio_sem_down_trylock(stat_sem)) |
39d13e67 JA |
1954 | break; |
1955 | usleep(1000); | |
ca0122d8 | 1956 | if (deadlock_loop_cnt++ > 5000) { |
971caeb1 | 1957 | log_err("fio seems to be stuck grabbing stat_sem, forcibly exiting\n"); |
43f248c4 | 1958 | td->error = EDEADLK; |
ca0122d8 TT |
1959 | goto err; |
1960 | } | |
39d13e67 | 1961 | } while (1); |
40c666c8 | 1962 | |
615c794c H |
1963 | if (td->io_bytes[DDIR_READ] && (td_read(td) || |
1964 | ((td->flags & TD_F_VER_BACKLOG) && td_write(td)))) | |
95603b74 BF |
1965 | update_runtime(td, elapsed_us, DDIR_READ); |
1966 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) | |
1967 | update_runtime(td, elapsed_us, DDIR_WRITE); | |
1968 | if (td_trim(td) && td->io_bytes[DDIR_TRIM]) | |
1969 | update_runtime(td, elapsed_us, DDIR_TRIM); | |
e5437a07 | 1970 | fio_gettime(&td->start, NULL); |
971caeb1 | 1971 | fio_sem_up(stat_sem); |
2e1df07d JA |
1972 | |
1973 | if (td->error || td->terminate) | |
1974 | break; | |
1975 | ||
4896473e JA |
1976 | if (!o->do_verify || |
1977 | o->verify == VERIFY_NONE || | |
9b87f09b | 1978 | td_ioengine_flagged(td, FIO_UNIDIR)) |
2e1df07d JA |
1979 | continue; |
1980 | ||
2d44f2c6 SK |
1981 | clear_io_state(td, 0); |
1982 | ||
2e1df07d JA |
1983 | fio_gettime(&td->start, NULL); |
1984 | ||
100f49f1 | 1985 | do_verify(td, verify_bytes); |
2e1df07d | 1986 | |
40c666c8 JA |
1987 | /* |
1988 | * See comment further up for why this is done here. | |
1989 | */ | |
1990 | check_update_rusage(td); | |
1991 | ||
971caeb1 | 1992 | fio_sem_down(stat_sem); |
95603b74 | 1993 | update_runtime(td, elapsed_us, DDIR_READ); |
e5437a07 | 1994 | fio_gettime(&td->start, NULL); |
971caeb1 | 1995 | fio_sem_up(stat_sem); |
2e1df07d JA |
1996 | |
1997 | if (td->error || td->terminate) | |
1998 | break; | |
1999 | } | |
2000 | ||
817ae977 VF |
2001 | /* |
2002 | * Acquire this lock if we were doing overlap checking in | |
2003 | * offload mode so that we don't clean up this job while | |
2004 | * another thread is checking its io_u's for overlap | |
2005 | */ | |
a38c70a8 | 2006 | if (td_offload_overlap(td)) { |
83276370 DP |
2007 | int res; |
2008 | ||
2009 | res = pthread_mutex_lock(&overlap_check); | |
2010 | if (res) { | |
2011 | td->error = errno; | |
2012 | goto err; | |
2013 | } | |
a38c70a8 | 2014 | } |
2cea0b4c | 2015 | td_set_runstate(td, TD_FINISHING); |
a38c70a8 | 2016 | if (td_offload_overlap(td)) { |
83276370 DP |
2017 | int res; |
2018 | ||
a38c70a8 | 2019 | res = pthread_mutex_unlock(&overlap_check); |
83276370 DP |
2020 | if (res) { |
2021 | td->error = errno; | |
2022 | goto err; | |
2023 | } | |
a38c70a8 | 2024 | } |
2cea0b4c | 2025 | |
2e1df07d | 2026 | update_rusage_stat(td); |
2e1df07d | 2027 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
dde39f8c AD |
2028 | for_each_rw_ddir(ddir) { |
2029 | td->ts.io_bytes[ddir] = td->io_bytes[ddir]; | |
2030 | } | |
2e1df07d | 2031 | |
ca09be4b | 2032 | if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) && |
d264264a JA |
2033 | (td->o.verify != VERIFY_NONE && td_write(td))) |
2034 | verify_save_state(td->thread_number); | |
ca09be4b | 2035 | |
9a3f1100 JA |
2036 | fio_unpin_memory(td); |
2037 | ||
a47591e4 | 2038 | td_writeout_logs(td, true); |
2e1df07d | 2039 | |
155f2f02 | 2040 | iolog_compress_exit(td); |
103b174e | 2041 | rate_submit_exit(td); |
aee2ab67 | 2042 | |
4896473e | 2043 | if (o->exec_postrun) |
0518f09f | 2044 | exec_string(o, o->exec_postrun, "postrun"); |
2e1df07d | 2045 | |
f9cafb12 | 2046 | if (exitall_on_terminate || (o->exitall_error && td->error)) |
64402a8a | 2047 | fio_terminate_threads(td->groupid, td->o.exit_what); |
2e1df07d JA |
2048 | |
2049 | err: | |
2050 | if (td->error) | |
2051 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, | |
2052 | td->verror); | |
2053 | ||
4896473e | 2054 | if (o->verify_async) |
2e1df07d JA |
2055 | verify_async_exit(td); |
2056 | ||
2057 | close_and_free_files(td); | |
2e1df07d | 2058 | cleanup_io_u(td); |
32dbca2c | 2059 | close_ioengine(td); |
c3dc516a | 2060 | cgroup_shutdown(td, cgroup_mnt); |
ca09be4b | 2061 | verify_free_state(td); |
07776127 | 2062 | td_zone_free_index(td); |
7c4d0fb7 | 2063 | |
b2a9e649 | 2064 | if (fio_option_is_set(o, cpumask)) { |
8a1db9a1 JA |
2065 | ret = fio_cpuset_exit(&o->cpumask); |
2066 | if (ret) | |
2067 | td_verror(td, ret, "fio_cpuset_exit"); | |
2e1df07d JA |
2068 | } |
2069 | ||
2070 | /* | |
2071 | * do this very late, it will log file closing as well | |
2072 | */ | |
4896473e | 2073 | if (o->write_iolog_file) |
2e1df07d | 2074 | write_iolog_close(td); |
98e7161c AK |
2075 | if (td->io_log_rfile) |
2076 | fclose(td->io_log_rfile); | |
2e1df07d JA |
2077 | |
2078 | td_set_runstate(td, TD_EXITED); | |
fda2cfac JA |
2079 | |
2080 | /* | |
2081 | * Do this last after setting our runstate to exited, so we | |
2082 | * know that the stat thread is signaled. | |
2083 | */ | |
2084 | check_update_rusage(td); | |
2085 | ||
24660963 | 2086 | sk_out_drop(); |
e43606c2 | 2087 | return (void *) (uintptr_t) td->error; |
2e1df07d JA |
2088 | } |
2089 | ||
2e1df07d JA |
2090 | /* |
2091 | * Run over the job map and reap the threads that have exited, if any. | |
2092 | */ | |
90eff1c9 SW |
2093 | static void reap_threads(unsigned int *nr_running, uint64_t *t_rate, |
2094 | uint64_t *m_rate) | |
2e1df07d | 2095 | { |
2e1df07d | 2096 | unsigned int cputhreads, realthreads, pending; |
5b7565b0 | 2097 | int ret; |
2e1df07d JA |
2098 | |
2099 | /* | |
2100 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
2101 | */ | |
2102 | realthreads = pending = cputhreads = 0; | |
da8f124f | 2103 | for_each_td(td) { |
5b7565b0 | 2104 | int flags = 0, status; |
2e1df07d | 2105 | |
74ee1904 | 2106 | if (!strcmp(td->o.ioengine, "cpuio")) |
2e1df07d JA |
2107 | cputhreads++; |
2108 | else | |
2109 | realthreads++; | |
2110 | ||
2111 | if (!td->pid) { | |
2112 | pending++; | |
2113 | continue; | |
2114 | } | |
2115 | if (td->runstate == TD_REAPED) | |
2116 | continue; | |
2117 | if (td->o.use_thread) { | |
2118 | if (td->runstate == TD_EXITED) { | |
2119 | td_set_runstate(td, TD_REAPED); | |
2120 | goto reaped; | |
2121 | } | |
2122 | continue; | |
2123 | } | |
2124 | ||
2125 | flags = WNOHANG; | |
2126 | if (td->runstate == TD_EXITED) | |
2127 | flags = 0; | |
2128 | ||
2129 | /* | |
2130 | * check if someone quit or got killed in an unusual way | |
2131 | */ | |
2132 | ret = waitpid(td->pid, &status, flags); | |
2133 | if (ret < 0) { | |
2134 | if (errno == ECHILD) { | |
2135 | log_err("fio: pid=%d disappeared %d\n", | |
2136 | (int) td->pid, td->runstate); | |
a5e371a6 | 2137 | td->sig = ECHILD; |
2e1df07d JA |
2138 | td_set_runstate(td, TD_REAPED); |
2139 | goto reaped; | |
2140 | } | |
2141 | perror("waitpid"); | |
2142 | } else if (ret == td->pid) { | |
2143 | if (WIFSIGNALED(status)) { | |
2144 | int sig = WTERMSIG(status); | |
2145 | ||
36d80bc7 | 2146 | if (sig != SIGTERM && sig != SIGUSR2) |
2e1df07d JA |
2147 | log_err("fio: pid=%d, got signal=%d\n", |
2148 | (int) td->pid, sig); | |
a5e371a6 | 2149 | td->sig = sig; |
2e1df07d JA |
2150 | td_set_runstate(td, TD_REAPED); |
2151 | goto reaped; | |
2152 | } | |
2153 | if (WIFEXITED(status)) { | |
2154 | if (WEXITSTATUS(status) && !td->error) | |
2155 | td->error = WEXITSTATUS(status); | |
2156 | ||
2157 | td_set_runstate(td, TD_REAPED); | |
2158 | goto reaped; | |
2159 | } | |
2160 | } | |
2161 | ||
cba5460c JA |
2162 | /* |
2163 | * If the job is stuck, do a forceful timeout of it and | |
2164 | * move on. | |
2165 | */ | |
2166 | if (td->terminate && | |
2c45a4ac | 2167 | td->runstate < TD_FSYNCING && |
cba5460c | 2168 | time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) { |
86310a1f TK |
2169 | log_err("fio: job '%s' (state=%d) hasn't exited in " |
2170 | "%lu seconds, it appears to be stuck. Doing " | |
2171 | "forceful exit of this job.\n", | |
2172 | td->o.name, td->runstate, | |
2173 | (unsigned long) time_since_now(&td->terminate_time)); | |
cba5460c JA |
2174 | td_set_runstate(td, TD_REAPED); |
2175 | goto reaped; | |
2176 | } | |
2177 | ||
2e1df07d JA |
2178 | /* |
2179 | * thread is not dead, continue | |
2180 | */ | |
2181 | pending++; | |
2182 | continue; | |
2183 | reaped: | |
2184 | (*nr_running)--; | |
342f4be4 JA |
2185 | (*m_rate) -= ddir_rw_sum(td->o.ratemin); |
2186 | (*t_rate) -= ddir_rw_sum(td->o.rate); | |
2e1df07d JA |
2187 | if (!td->pid) |
2188 | pending--; | |
2189 | ||
2190 | if (td->error) | |
2191 | exit_value++; | |
2192 | ||
2193 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
4a88752a | 2194 | profile_td_exit(td); |
42e26e94 | 2195 | flow_exit_job(td); |
da8f124f | 2196 | } end_for_each(); |
2e1df07d JA |
2197 | |
2198 | if (*nr_running == cputhreads && !pending && realthreads) | |
64402a8a | 2199 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
2e1df07d JA |
2200 | } |
2201 | ||
e39c0676 | 2202 | static bool __check_trigger_file(void) |
ca09be4b JA |
2203 | { |
2204 | struct stat sb; | |
2205 | ||
2206 | if (!trigger_file) | |
e39c0676 | 2207 | return false; |
ca09be4b JA |
2208 | |
2209 | if (stat(trigger_file, &sb)) | |
e39c0676 | 2210 | return false; |
ca09be4b JA |
2211 | |
2212 | if (unlink(trigger_file) < 0) | |
2213 | log_err("fio: failed to unlink %s: %s\n", trigger_file, | |
2214 | strerror(errno)); | |
2215 | ||
e39c0676 | 2216 | return true; |
ca09be4b JA |
2217 | } |
2218 | ||
e39c0676 | 2219 | static bool trigger_timedout(void) |
ca09be4b JA |
2220 | { |
2221 | if (trigger_timeout) | |
98413be7 SW |
2222 | if (time_since_genesis() >= trigger_timeout) { |
2223 | trigger_timeout = 0; | |
2224 | return true; | |
2225 | } | |
ca09be4b | 2226 | |
e39c0676 | 2227 | return false; |
ca09be4b JA |
2228 | } |
2229 | ||
2230 | void exec_trigger(const char *cmd) | |
2231 | { | |
2232 | int ret; | |
2233 | ||
98413be7 | 2234 | if (!cmd || cmd[0] == '\0') |
ca09be4b JA |
2235 | return; |
2236 | ||
2237 | ret = system(cmd); | |
2238 | if (ret == -1) | |
2239 | log_err("fio: failed executing %s trigger\n", cmd); | |
2240 | } | |
2241 | ||
2242 | void check_trigger_file(void) | |
2243 | { | |
2244 | if (__check_trigger_file() || trigger_timedout()) { | |
796fb3ce JA |
2245 | if (nr_clients) |
2246 | fio_clients_send_trigger(trigger_remote_cmd); | |
2247 | else { | |
d264264a | 2248 | verify_save_state(IO_LIST_ALL); |
64402a8a | 2249 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
ca09be4b JA |
2250 | exec_trigger(trigger_cmd); |
2251 | } | |
2252 | } | |
2253 | } | |
2254 | ||
2255 | static int fio_verify_load_state(struct thread_data *td) | |
2256 | { | |
2257 | int ret; | |
2258 | ||
2259 | if (!td->o.verify_state) | |
2260 | return 0; | |
2261 | ||
2262 | if (is_backend) { | |
2263 | void *data; | |
2264 | ||
2265 | ret = fio_server_get_verify_state(td->o.name, | |
94a6e1bb | 2266 | td->thread_number - 1, &data); |
ca09be4b | 2267 | if (!ret) |
94a6e1bb | 2268 | verify_assign_state(td, data); |
67710881 AK |
2269 | } else { |
2270 | char prefix[PATH_MAX]; | |
2271 | ||
2272 | if (aux_path) | |
2273 | sprintf(prefix, "%s%clocal", aux_path, | |
2274 | FIO_OS_PATH_SEPARATOR); | |
2275 | else | |
2276 | strcpy(prefix, "local"); | |
2277 | ret = verify_load_state(td, prefix); | |
2278 | } | |
ca09be4b JA |
2279 | |
2280 | return ret; | |
2281 | } | |
2282 | ||
06464907 JA |
2283 | static void do_usleep(unsigned int usecs) |
2284 | { | |
2285 | check_for_running_stats(); | |
ca09be4b | 2286 | check_trigger_file(); |
06464907 JA |
2287 | usleep(usecs); |
2288 | } | |
2289 | ||
e39c0676 | 2290 | static bool check_mount_writes(struct thread_data *td) |
e81ecca3 JA |
2291 | { |
2292 | struct fio_file *f; | |
2293 | unsigned int i; | |
2294 | ||
2295 | if (!td_write(td) || td->o.allow_mounted_write) | |
e39c0676 | 2296 | return false; |
e81ecca3 | 2297 | |
5e81f9c1 TK |
2298 | /* |
2299 | * If FIO_HAVE_CHARDEV_SIZE is defined, it's likely that chrdevs | |
2300 | * are mkfs'd and mounted. | |
2301 | */ | |
e81ecca3 | 2302 | for_each_file(td, f, i) { |
5e81f9c1 TK |
2303 | #ifdef FIO_HAVE_CHARDEV_SIZE |
2304 | if (f->filetype != FIO_TYPE_BLOCK && f->filetype != FIO_TYPE_CHAR) | |
2305 | #else | |
686fbd31 | 2306 | if (f->filetype != FIO_TYPE_BLOCK) |
5e81f9c1 | 2307 | #endif |
e81ecca3 JA |
2308 | continue; |
2309 | if (device_is_mounted(f->file_name)) | |
2310 | goto mounted; | |
2311 | } | |
2312 | ||
e39c0676 | 2313 | return false; |
e81ecca3 | 2314 | mounted: |
c5839011 | 2315 | log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.\n", f->file_name); |
e39c0676 | 2316 | return true; |
e81ecca3 JA |
2317 | } |
2318 | ||
9cc8cb91 AK |
2319 | static bool waitee_running(struct thread_data *me) |
2320 | { | |
2321 | const char *waitee = me->o.wait_for; | |
2322 | const char *self = me->o.name; | |
9cc8cb91 AK |
2323 | |
2324 | if (!waitee) | |
2325 | return false; | |
2326 | ||
da8f124f | 2327 | for_each_td(td) { |
9cc8cb91 AK |
2328 | if (!strcmp(td->o.name, self) || strcmp(td->o.name, waitee)) |
2329 | continue; | |
2330 | ||
2331 | if (td->runstate < TD_EXITED) { | |
2332 | dprint(FD_PROCESS, "%s fenced by %s(%s)\n", | |
2333 | self, td->o.name, | |
2334 | runstate_to_name(td->runstate)); | |
2335 | return true; | |
2336 | } | |
da8f124f | 2337 | } end_for_each(); |
9cc8cb91 AK |
2338 | |
2339 | dprint(FD_PROCESS, "%s: %s completed, can run\n", self, waitee); | |
2340 | return false; | |
2341 | } | |
2342 | ||
2e1df07d JA |
2343 | /* |
2344 | * Main function for kicking off and reaping jobs, as needed. | |
2345 | */ | |
24660963 | 2346 | static void run_threads(struct sk_out *sk_out) |
2e1df07d JA |
2347 | { |
2348 | struct thread_data *td; | |
90eff1c9 SW |
2349 | unsigned int i, todo, nr_running, nr_started; |
2350 | uint64_t m_rate, t_rate; | |
0de5b26f | 2351 | uint64_t spent; |
2e1df07d | 2352 | |
2e1df07d JA |
2353 | if (fio_gtod_offload && fio_start_gtod_thread()) |
2354 | return; | |
334185e9 | 2355 | |
f2a2ce0e | 2356 | fio_idle_prof_init(); |
2e1df07d JA |
2357 | |
2358 | set_sig_handlers(); | |
2359 | ||
3a5f6bde | 2360 | nr_thread = nr_process = 0; |
da8f124f | 2361 | for_each_td(td) { |
e81ecca3 JA |
2362 | if (check_mount_writes(td)) |
2363 | return; | |
3a5f6bde JA |
2364 | if (td->o.use_thread) |
2365 | nr_thread++; | |
2366 | else | |
2367 | nr_process++; | |
da8f124f | 2368 | } end_for_each(); |
3a5f6bde | 2369 | |
01cfefcc | 2370 | if (output_format & FIO_OUTPUT_NORMAL) { |
a41fc529 JA |
2371 | struct buf_output out; |
2372 | ||
2373 | buf_output_init(&out); | |
c2fcb6bc | 2374 | __log_buf(&out, "Starting "); |
2e1df07d | 2375 | if (nr_thread) |
c2fcb6bc | 2376 | __log_buf(&out, "%d thread%s", nr_thread, |
2e1df07d JA |
2377 | nr_thread > 1 ? "s" : ""); |
2378 | if (nr_process) { | |
2379 | if (nr_thread) | |
c2fcb6bc JA |
2380 | __log_buf(&out, " and "); |
2381 | __log_buf(&out, "%d process%s", nr_process, | |
2e1df07d JA |
2382 | nr_process > 1 ? "es" : ""); |
2383 | } | |
c2fcb6bc | 2384 | __log_buf(&out, "\n"); |
a41fc529 | 2385 | log_info_buf(out.buf, out.buflen); |
d5bb0a89 | 2386 | buf_output_free(&out); |
2e1df07d JA |
2387 | } |
2388 | ||
2389 | todo = thread_number; | |
2390 | nr_running = 0; | |
2391 | nr_started = 0; | |
2392 | m_rate = t_rate = 0; | |
2393 | ||
da8f124f | 2394 | for_each_td(td) { |
2e1df07d JA |
2395 | print_status_init(td->thread_number - 1); |
2396 | ||
d5e3f7cb | 2397 | if (!td->o.create_serialize) |
2e1df07d JA |
2398 | continue; |
2399 | ||
ca09be4b JA |
2400 | if (fio_verify_load_state(td)) |
2401 | goto reap; | |
2402 | ||
2e1df07d JA |
2403 | /* |
2404 | * do file setup here so it happens sequentially, | |
2405 | * we don't want X number of threads getting their | |
2406 | * client data interspersed on disk | |
2407 | */ | |
2408 | if (setup_files(td)) { | |
ca09be4b | 2409 | reap: |
2e1df07d JA |
2410 | exit_value++; |
2411 | if (td->error) | |
2412 | log_err("fio: pid=%d, err=%d/%s\n", | |
2413 | (int) td->pid, td->error, td->verror); | |
2414 | td_set_runstate(td, TD_REAPED); | |
2415 | todo--; | |
2416 | } else { | |
2417 | struct fio_file *f; | |
2418 | unsigned int j; | |
2419 | ||
2420 | /* | |
2421 | * for sharing to work, each job must always open | |
2422 | * its own files. so close them, if we opened them | |
2423 | * for creation | |
2424 | */ | |
2425 | for_each_file(td, f, j) { | |
2426 | if (fio_file_open(f)) | |
2427 | td_io_close_file(td, f); | |
2428 | } | |
2429 | } | |
da8f124f | 2430 | } end_for_each(); |
2e1df07d | 2431 | |
f2a2ce0e HL |
2432 | /* start idle threads before io threads start to run */ |
2433 | fio_idle_prof_start(); | |
2434 | ||
2e1df07d JA |
2435 | set_genesis_time(); |
2436 | ||
2437 | while (todo) { | |
2438 | struct thread_data *map[REAL_MAX_JOBS]; | |
8b6a404c | 2439 | struct timespec this_start; |
2e1df07d | 2440 | int this_jobs = 0, left; |
ce8b6139 | 2441 | struct fork_data *fd; |
2e1df07d JA |
2442 | |
2443 | /* | |
2444 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
2445 | */ | |
da8f124f | 2446 | for_each_td(td) { |
2e1df07d JA |
2447 | if (td->runstate != TD_NOT_CREATED) |
2448 | continue; | |
2449 | ||
2450 | /* | |
2451 | * never got a chance to start, killed by other | |
2452 | * thread for some reason | |
2453 | */ | |
2454 | if (td->terminate) { | |
2455 | todo--; | |
2456 | continue; | |
2457 | } | |
2458 | ||
2459 | if (td->o.start_delay) { | |
0de5b26f | 2460 | spent = utime_since_genesis(); |
2e1df07d | 2461 | |
74454ce4 | 2462 | if (td->o.start_delay > spent) |
2e1df07d JA |
2463 | continue; |
2464 | } | |
2465 | ||
2466 | if (td->o.stonewall && (nr_started || nr_running)) { | |
2467 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
2468 | td->o.name); | |
2469 | break; | |
2470 | } | |
2471 | ||
9cc8cb91 AK |
2472 | if (waitee_running(td)) { |
2473 | dprint(FD_PROCESS, "%s: waiting for %s\n", | |
2474 | td->o.name, td->o.wait_for); | |
5c74fc76 | 2475 | continue; |
9cc8cb91 AK |
2476 | } |
2477 | ||
2e1df07d JA |
2478 | init_disk_util(td); |
2479 | ||
971caeb1 | 2480 | td->rusage_sem = fio_sem_init(FIO_SEM_LOCKED); |
c97f1ad6 JA |
2481 | td->update_rusage = 0; |
2482 | ||
2e1df07d JA |
2483 | /* |
2484 | * Set state to created. Thread will transition | |
2485 | * to TD_INITIALIZED when it's done setting up. | |
2486 | */ | |
2487 | td_set_runstate(td, TD_CREATED); | |
2488 | map[this_jobs++] = td; | |
2489 | nr_started++; | |
2490 | ||
ce8b6139 JA |
2491 | fd = calloc(1, sizeof(*fd)); |
2492 | fd->td = td; | |
2493 | fd->sk_out = sk_out; | |
2494 | ||
2e1df07d JA |
2495 | if (td->o.use_thread) { |
2496 | int ret; | |
2497 | ||
2498 | dprint(FD_PROCESS, "will pthread_create\n"); | |
2499 | ret = pthread_create(&td->thread, NULL, | |
24660963 | 2500 | thread_main, fd); |
2e1df07d JA |
2501 | if (ret) { |
2502 | log_err("pthread_create: %s\n", | |
2503 | strerror(ret)); | |
24660963 | 2504 | free(fd); |
2e1df07d JA |
2505 | nr_started--; |
2506 | break; | |
2507 | } | |
493dd6c1 | 2508 | fd = NULL; |
2e1df07d JA |
2509 | ret = pthread_detach(td->thread); |
2510 | if (ret) | |
2511 | log_err("pthread_detach: %s", | |
2512 | strerror(ret)); | |
2513 | } else { | |
2514 | pid_t pid; | |
824912be | 2515 | void *eo; |
2e1df07d | 2516 | dprint(FD_PROCESS, "will fork\n"); |
824912be JA |
2517 | eo = td->eo; |
2518 | read_barrier(); | |
2e1df07d JA |
2519 | pid = fork(); |
2520 | if (!pid) { | |
ce8b6139 | 2521 | int ret; |
2e1df07d | 2522 | |
ce8b6139 | 2523 | ret = (int)(uintptr_t)thread_main(fd); |
2e1df07d | 2524 | _exit(ret); |
da8f124f | 2525 | } else if (__td_index == fio_debug_jobno) |
2e1df07d | 2526 | *fio_debug_jobp = pid; |
824912be | 2527 | free(eo); |
36aac34e DP |
2528 | free(fd); |
2529 | fd = NULL; | |
2e1df07d | 2530 | } |
971caeb1 BVA |
2531 | dprint(FD_MUTEX, "wait on startup_sem\n"); |
2532 | if (fio_sem_down_timeout(startup_sem, 10000)) { | |
2e1df07d | 2533 | log_err("fio: job startup hung? exiting.\n"); |
64402a8a | 2534 | fio_terminate_threads(TERMINATE_ALL, TERMINATE_ALL); |
f1867a7f | 2535 | fio_abort = true; |
2e1df07d | 2536 | nr_started--; |
6541f4e0 | 2537 | free(fd); |
2e1df07d JA |
2538 | break; |
2539 | } | |
971caeb1 | 2540 | dprint(FD_MUTEX, "done waiting on startup_sem\n"); |
da8f124f | 2541 | } end_for_each(); |
2e1df07d JA |
2542 | |
2543 | /* | |
2544 | * Wait for the started threads to transition to | |
2545 | * TD_INITIALIZED. | |
2546 | */ | |
2547 | fio_gettime(&this_start, NULL); | |
2548 | left = this_jobs; | |
2549 | while (left && !fio_abort) { | |
2550 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
2551 | break; | |
2552 | ||
06464907 | 2553 | do_usleep(100000); |
2e1df07d JA |
2554 | |
2555 | for (i = 0; i < this_jobs; i++) { | |
2556 | td = map[i]; | |
2557 | if (!td) | |
2558 | continue; | |
2559 | if (td->runstate == TD_INITIALIZED) { | |
2560 | map[i] = NULL; | |
2561 | left--; | |
2562 | } else if (td->runstate >= TD_EXITED) { | |
2563 | map[i] = NULL; | |
2564 | left--; | |
2565 | todo--; | |
2566 | nr_running++; /* work-around... */ | |
2567 | } | |
2568 | } | |
2569 | } | |
2570 | ||
2571 | if (left) { | |
4e87c37a JA |
2572 | log_err("fio: %d job%s failed to start\n", left, |
2573 | left > 1 ? "s" : ""); | |
2e1df07d JA |
2574 | for (i = 0; i < this_jobs; i++) { |
2575 | td = map[i]; | |
2576 | if (!td) | |
2577 | continue; | |
2578 | kill(td->pid, SIGTERM); | |
2579 | } | |
2580 | break; | |
2581 | } | |
2582 | ||
2583 | /* | |
2584 | * start created threads (TD_INITIALIZED -> TD_RUNNING). | |
2585 | */ | |
da8f124f | 2586 | for_each_td(td) { |
2e1df07d JA |
2587 | if (td->runstate != TD_INITIALIZED) |
2588 | continue; | |
2589 | ||
2590 | if (in_ramp_time(td)) | |
2591 | td_set_runstate(td, TD_RAMP); | |
2592 | else | |
2593 | td_set_runstate(td, TD_RUNNING); | |
2594 | nr_running++; | |
2595 | nr_started--; | |
342f4be4 JA |
2596 | m_rate += ddir_rw_sum(td->o.ratemin); |
2597 | t_rate += ddir_rw_sum(td->o.rate); | |
2e1df07d | 2598 | todo--; |
971caeb1 | 2599 | fio_sem_up(td->sem); |
da8f124f | 2600 | } end_for_each(); |
2e1df07d JA |
2601 | |
2602 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2603 | ||
122c7725 | 2604 | if (todo) |
06464907 | 2605 | do_usleep(100000); |
2e1df07d JA |
2606 | } |
2607 | ||
2608 | while (nr_running) { | |
2609 | reap_threads(&nr_running, &t_rate, &m_rate); | |
06464907 | 2610 | do_usleep(10000); |
2e1df07d JA |
2611 | } |
2612 | ||
f2a2ce0e HL |
2613 | fio_idle_prof_stop(); |
2614 | ||
2e1df07d | 2615 | update_io_ticks(); |
2e1df07d JA |
2616 | } |
2617 | ||
27357187 JA |
2618 | static void free_disk_util(void) |
2619 | { | |
27357187 | 2620 | disk_util_prune_entries(); |
a39fb9ea | 2621 | helper_thread_destroy(); |
2e1df07d JA |
2622 | } |
2623 | ||
24660963 | 2624 | int fio_backend(struct sk_out *sk_out) |
2e1df07d | 2625 | { |
2e1df07d | 2626 | int i; |
2e1df07d JA |
2627 | if (exec_profile) { |
2628 | if (load_profile(exec_profile)) | |
2629 | return 1; | |
2630 | free(exec_profile); | |
2631 | exec_profile = NULL; | |
2632 | } | |
2633 | if (!thread_number) | |
2634 | return 0; | |
2635 | ||
2636 | if (write_bw_log) { | |
aee2ab67 JA |
2637 | struct log_params p = { |
2638 | .log_type = IO_LOG_TYPE_BW, | |
2639 | }; | |
2640 | ||
2641 | setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log"); | |
2642 | setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log"); | |
2643 | setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log"); | |
2e1df07d JA |
2644 | } |
2645 | ||
c49cfc76 BD |
2646 | if (init_global_dedupe_working_set_seeds()) { |
2647 | log_err("fio: failed to initialize global dedupe working set\n"); | |
2648 | return 1; | |
2649 | } | |
2650 | ||
971caeb1 | 2651 | startup_sem = fio_sem_init(FIO_SEM_LOCKED); |
24a97130 LC |
2652 | if (!sk_out) |
2653 | is_local_backend = true; | |
971caeb1 | 2654 | if (startup_sem == NULL) |
2e1df07d | 2655 | return 1; |
2e1df07d JA |
2656 | |
2657 | set_genesis_time(); | |
cef9175e | 2658 | stat_init(); |
998e9ebb BVA |
2659 | if (helper_thread_create(startup_sem, sk_out)) |
2660 | log_err("fio: failed to create helper thread\n"); | |
2e1df07d JA |
2661 | |
2662 | cgroup_list = smalloc(sizeof(*cgroup_list)); | |
b2e6494c JA |
2663 | if (cgroup_list) |
2664 | INIT_FLIST_HEAD(cgroup_list); | |
2e1df07d | 2665 | |
24660963 | 2666 | run_threads(sk_out); |
2e1df07d | 2667 | |
a39fb9ea | 2668 | helper_thread_exit(); |
8aab824f | 2669 | |
2e1df07d | 2670 | if (!fio_abort) { |
83f7b64e | 2671 | __show_run_stats(); |
2e1df07d | 2672 | if (write_bw_log) { |
cb7e0ace JA |
2673 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { |
2674 | struct io_log *log = agg_io_log[i]; | |
2675 | ||
60a25727 | 2676 | flush_log(log, false); |
518dac09 | 2677 | free_log(log); |
cb7e0ace | 2678 | } |
2e1df07d JA |
2679 | } |
2680 | } | |
2681 | ||
da8f124f | 2682 | for_each_td(td) { |
f0547200 NC |
2683 | struct thread_stat *ts = &td->ts; |
2684 | ||
2685 | free_clat_prio_stats(ts); | |
c27cc65f | 2686 | steadystate_free(td); |
2e1df07d | 2687 | fio_options_free(td); |
8009f543 | 2688 | fio_dump_options_free(td); |
8049adc1 | 2689 | if (td->rusage_sem) { |
971caeb1 | 2690 | fio_sem_remove(td->rusage_sem); |
8049adc1 JA |
2691 | td->rusage_sem = NULL; |
2692 | } | |
971caeb1 BVA |
2693 | fio_sem_remove(td->sem); |
2694 | td->sem = NULL; | |
da8f124f | 2695 | } end_for_each(); |
2e1df07d | 2696 | |
a462baef | 2697 | free_disk_util(); |
b2e6494c JA |
2698 | if (cgroup_list) { |
2699 | cgroup_kill(cgroup_list); | |
2700 | sfree(cgroup_list); | |
2701 | } | |
2e1df07d | 2702 | |
971caeb1 | 2703 | fio_sem_remove(startup_sem); |
cef9175e | 2704 | stat_exit(); |
2e1df07d JA |
2705 | return exit_value; |
2706 | } |