Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
aae22ca7 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
ebac4655 | 6 | * |
8e9fe637 JA |
7 | * The license below covers all files distributed with fio unless otherwise |
8 | * noted in the file itself. | |
9 | * | |
ebac4655 | 10 | * This program is free software; you can redistribute it and/or modify |
8e9fe637 JA |
11 | * it under the terms of the GNU General Public License version 2 as |
12 | * published by the Free Software Foundation. | |
ebac4655 JA |
13 | * |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
21 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
22 | * | |
23 | */ | |
ebac4655 JA |
24 | #include <unistd.h> |
25 | #include <fcntl.h> | |
26 | #include <string.h> | |
9b836561 | 27 | #include <limits.h> |
ebac4655 JA |
28 | #include <signal.h> |
29 | #include <time.h> | |
dbe1125e | 30 | #include <locale.h> |
36167d82 | 31 | #include <assert.h> |
a5b01f1b | 32 | #include <time.h> |
ebac4655 JA |
33 | #include <sys/stat.h> |
34 | #include <sys/wait.h> | |
35 | #include <sys/ipc.h> | |
36 | #include <sys/shm.h> | |
ebac4655 JA |
37 | #include <sys/mman.h> |
38 | ||
39 | #include "fio.h" | |
210f43b3 | 40 | #include "hash.h" |
2e5cdb11 | 41 | #include "smalloc.h" |
4f5af7b2 | 42 | #include "verify.h" |
a917a8b3 | 43 | #include "trim.h" |
7c9b1bce | 44 | #include "diskutil.h" |
a696fa2a | 45 | #include "cgroup.h" |
07b3232d | 46 | #include "profile.h" |
d54a144d | 47 | #include "lib/rand.h" |
91aea6ec | 48 | #include "memalign.h" |
ebac4655 | 49 | |
cfc99db7 JA |
50 | unsigned long page_mask; |
51 | unsigned long page_size; | |
d529ee19 JA |
52 | |
53 | #define PAGE_ALIGN(buf) \ | |
29d610e1 | 54 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) |
ebac4655 JA |
55 | |
56 | int groupid = 0; | |
57 | int thread_number = 0; | |
9cedf167 JA |
58 | int nr_process = 0; |
59 | int nr_thread = 0; | |
ebac4655 | 60 | int shm_id = 0; |
53cdc686 | 61 | int temp_stall_ts; |
d3eeeabc | 62 | unsigned long done_secs = 0; |
ebac4655 | 63 | |
cdd18ad8 | 64 | static struct fio_mutex *startup_mutex; |
d2bbb8d9 | 65 | static struct fio_mutex *writeout_mutex; |
6ce15a32 | 66 | static volatile int fio_abort; |
437c9b71 | 67 | static int exit_value; |
be4ecfdf | 68 | static pthread_t gtod_thread; |
7985701b | 69 | static pthread_t disk_util_thread; |
dae5341c | 70 | static struct flist_head *cgroup_list; |
6adb38a1 | 71 | static char *cgroup_mnt; |
ebac4655 | 72 | |
bb3884d8 JA |
73 | struct io_log *agg_io_log[2]; |
74 | ||
ebac4655 | 75 | #define TERMINATE_ALL (-1) |
75154845 | 76 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 77 | |
b29ee5b3 | 78 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 79 | { |
d8fab1b6 JA |
80 | if (td->runstate == runstate) |
81 | return; | |
82 | ||
5921e80c JA |
83 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
84 | td->runstate, runstate); | |
6ce15a32 JA |
85 | td->runstate = runstate; |
86 | } | |
87 | ||
390c40e2 | 88 | static void terminate_threads(int group_id) |
ebac4655 | 89 | { |
34572e28 | 90 | struct thread_data *td; |
ebac4655 JA |
91 | int i; |
92 | ||
6f88bcb0 JA |
93 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
94 | ||
34572e28 | 95 | for_each_td(td, i) { |
ebac4655 | 96 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 97 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 98 | td->o.name, (int) td->pid); |
ad830ed7 JA |
99 | td->terminate = 1; |
100 | td->o.start_delay = 0; | |
101 | ||
7a93ab19 JA |
102 | /* |
103 | * if the thread is running, just let it exit | |
104 | */ | |
105 | if (td->runstate < TD_RUNNING) | |
03e20d68 | 106 | kill(td->pid, SIGTERM); |
f63f7f3b JA |
107 | else { |
108 | struct ioengine_ops *ops = td->io_ops; | |
109 | ||
03e20d68 BC |
110 | if (ops && (ops->flags & FIO_SIGTERM)) |
111 | kill(td->pid, SIGTERM); | |
f63f7f3b | 112 | } |
ebac4655 JA |
113 | } |
114 | } | |
115 | } | |
116 | ||
5f8f9726 JA |
117 | /* |
118 | * Happens on thread runs with ctrl-c, ignore our own SIGQUIT | |
119 | */ | |
120 | static void sig_quit(int sig) | |
121 | { | |
122 | } | |
123 | ||
697a606c JA |
124 | static void sig_int(int sig) |
125 | { | |
126 | if (threads) { | |
4ceb30d4 | 127 | log_info("\nfio: terminating on signal %d\n", sig); |
5ec10eaa | 128 | fflush(stdout); |
deaa1af2 | 129 | exit_value = 128; |
5ec10eaa | 130 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
131 | } |
132 | } | |
133 | ||
7985701b | 134 | static void *disk_thread_main(void *data) |
a5b01f1b | 135 | { |
7985701b JA |
136 | fio_mutex_up(startup_mutex); |
137 | ||
138 | while (threads) { | |
139 | usleep(DISK_UTIL_MSEC * 1000); | |
9a793c2f JA |
140 | if (!threads) |
141 | break; | |
7985701b JA |
142 | update_io_ticks(); |
143 | print_thread_status(); | |
144 | } | |
145 | ||
146 | return NULL; | |
a5b01f1b JA |
147 | } |
148 | ||
7985701b | 149 | static int create_disk_util_thread(void) |
a5b01f1b | 150 | { |
7985701b | 151 | int ret; |
a5b01f1b | 152 | |
69204d6e | 153 | ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); |
7985701b JA |
154 | if (ret) { |
155 | log_err("Can't create disk util thread: %s\n", strerror(ret)); | |
156 | return 1; | |
157 | } | |
a5b01f1b | 158 | |
7985701b JA |
159 | ret = pthread_detach(disk_util_thread); |
160 | if (ret) { | |
161 | log_err("Can't detatch disk util thread: %s\n", strerror(ret)); | |
162 | return 1; | |
163 | } | |
03e20d68 | 164 | |
7985701b JA |
165 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
166 | fio_mutex_down(startup_mutex); | |
167 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
168 | return 0; | |
a5b01f1b JA |
169 | } |
170 | ||
697a606c JA |
171 | static void set_sig_handlers(void) |
172 | { | |
173 | struct sigaction act; | |
174 | ||
697a606c JA |
175 | memset(&act, 0, sizeof(act)); |
176 | act.sa_handler = sig_int; | |
177 | act.sa_flags = SA_RESTART; | |
178 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 179 | |
5f8f9726 JA |
180 | memset(&act, 0, sizeof(act)); |
181 | act.sa_handler = sig_quit; | |
182 | act.sa_flags = SA_RESTART; | |
03e20d68 | 183 | sigaction(SIGTERM, &act, NULL); |
697a606c JA |
184 | } |
185 | ||
906c8d75 JA |
186 | /* |
187 | * Check if we are above the minimum rate given. | |
188 | */ | |
581e7141 JA |
189 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
190 | enum td_ddir ddir) | |
ebac4655 | 191 | { |
0904200b | 192 | unsigned long long bytes = 0; |
4e991c23 | 193 | unsigned long iops = 0; |
ebac4655 JA |
194 | unsigned long spent; |
195 | unsigned long rate; | |
581e7141 JA |
196 | unsigned int ratemin = 0; |
197 | unsigned int rate_iops = 0; | |
198 | unsigned int rate_iops_min = 0; | |
ebac4655 | 199 | |
ff58fced JA |
200 | assert(ddir_rw(ddir)); |
201 | ||
d982d873 JA |
202 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
203 | return 0; | |
204 | ||
ebac4655 JA |
205 | /* |
206 | * allow a 2 second settle period in the beginning | |
207 | */ | |
208 | if (mtime_since(&td->start, now) < 2000) | |
209 | return 0; | |
210 | ||
581e7141 JA |
211 | iops += td->io_blocks[ddir]; |
212 | bytes += td->this_io_bytes[ddir]; | |
213 | ratemin += td->o.ratemin[ddir]; | |
214 | rate_iops += td->o.rate_iops[ddir]; | |
215 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 216 | |
ebac4655 JA |
217 | /* |
218 | * if rate blocks is set, sample is running | |
219 | */ | |
581e7141 JA |
220 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
221 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 222 | if (spent < td->o.ratecycle) |
ebac4655 JA |
223 | return 0; |
224 | ||
581e7141 | 225 | if (td->o.rate[ddir]) { |
4e991c23 JA |
226 | /* |
227 | * check bandwidth specified rate | |
228 | */ | |
581e7141 | 229 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 230 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 231 | ratemin); |
4e991c23 JA |
232 | return 1; |
233 | } else { | |
581e7141 JA |
234 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
235 | if (rate < ratemin || | |
236 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 237 | log_err("%s: min rate %u not met, got" |
b22989b9 | 238 | " %luKB/sec\n", td->o.name, |
581e7141 | 239 | ratemin, rate); |
4e991c23 JA |
240 | return 1; |
241 | } | |
242 | } | |
413dd459 | 243 | } else { |
4e991c23 JA |
244 | /* |
245 | * checks iops specified rate | |
246 | */ | |
581e7141 | 247 | if (iops < rate_iops) { |
5ec10eaa | 248 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 249 | td->o.name, rate_iops); |
413dd459 | 250 | return 1; |
4e991c23 | 251 | } else { |
581e7141 JA |
252 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
253 | if (rate < rate_iops_min || | |
254 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
255 | log_err("%s: min iops rate %u not met," |
256 | " got %lu\n", td->o.name, | |
581e7141 | 257 | rate_iops_min, rate); |
4e991c23 | 258 | } |
413dd459 | 259 | } |
ebac4655 JA |
260 | } |
261 | } | |
262 | ||
581e7141 JA |
263 | td->rate_bytes[ddir] = bytes; |
264 | td->rate_blocks[ddir] = iops; | |
265 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
266 | return 0; |
267 | } | |
268 | ||
581e7141 JA |
269 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
270 | unsigned long *bytes_done) | |
271 | { | |
272 | int ret = 0; | |
273 | ||
274 | if (bytes_done[0]) | |
275 | ret |= __check_min_rate(td, now, 0); | |
276 | if (bytes_done[1]) | |
277 | ret |= __check_min_rate(td, now, 1); | |
278 | ||
279 | return ret; | |
280 | } | |
281 | ||
ebac4655 JA |
282 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
283 | { | |
2dc1bbeb | 284 | if (!td->o.timeout) |
ebac4655 | 285 | return 0; |
2dc1bbeb | 286 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
287 | return 1; |
288 | ||
289 | return 0; | |
290 | } | |
291 | ||
906c8d75 JA |
292 | /* |
293 | * When job exits, we can cancel the in-flight IO if we are using async | |
294 | * io. Attempt to do so. | |
295 | */ | |
ebac4655 JA |
296 | static void cleanup_pending_aio(struct thread_data *td) |
297 | { | |
01743ee1 | 298 | struct flist_head *entry, *n; |
ebac4655 JA |
299 | struct io_u *io_u; |
300 | int r; | |
301 | ||
302 | /* | |
303 | * get immediately available events, if any | |
304 | */ | |
581e7141 | 305 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
306 | if (r < 0) |
307 | return; | |
ebac4655 JA |
308 | |
309 | /* | |
310 | * now cancel remaining active events | |
311 | */ | |
2866c82d | 312 | if (td->io_ops->cancel) { |
01743ee1 JA |
313 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
314 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 315 | |
0c6e7517 JA |
316 | /* |
317 | * if the io_u isn't in flight, then that generally | |
318 | * means someone leaked an io_u. complain but fix | |
319 | * it up, so we don't stall here. | |
320 | */ | |
321 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
322 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 323 | put_io_u(td, io_u); |
0c6e7517 JA |
324 | } else { |
325 | r = td->io_ops->cancel(td, io_u); | |
326 | if (!r) | |
327 | put_io_u(td, io_u); | |
328 | } | |
ebac4655 JA |
329 | } |
330 | } | |
331 | ||
97601024 | 332 | if (td->cur_depth) |
581e7141 | 333 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
334 | } |
335 | ||
858a3d47 JA |
336 | /* |
337 | * Helper to handle the final sync of a file. Works just like the normal | |
338 | * io path, just does everything sync. | |
339 | */ | |
340 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
341 | { | |
342 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
343 | int ret; |
344 | ||
345 | if (!io_u) | |
346 | return 1; | |
347 | ||
348 | io_u->ddir = DDIR_SYNC; | |
349 | io_u->file = f; | |
350 | ||
351 | if (td_io_prep(td, io_u)) { | |
352 | put_io_u(td, io_u); | |
353 | return 1; | |
354 | } | |
355 | ||
755200a3 | 356 | requeue: |
858a3d47 | 357 | ret = td_io_queue(td, io_u); |
36167d82 | 358 | if (ret < 0) { |
e1161c32 | 359 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 360 | put_io_u(td, io_u); |
858a3d47 | 361 | return 1; |
36167d82 | 362 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 363 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 364 | return 1; |
36167d82 JA |
365 | } else if (ret == FIO_Q_COMPLETED) { |
366 | if (io_u->error) { | |
e1161c32 | 367 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
368 | return 1; |
369 | } | |
858a3d47 | 370 | |
581e7141 | 371 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 372 | return 1; |
755200a3 JA |
373 | } else if (ret == FIO_Q_BUSY) { |
374 | if (td_io_commit(td)) | |
375 | return 1; | |
376 | goto requeue; | |
858a3d47 JA |
377 | } |
378 | ||
379 | return 0; | |
380 | } | |
381 | ||
993bf48b JA |
382 | static inline void update_tv_cache(struct thread_data *td) |
383 | { | |
384 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
385 | fio_gettime(&td->tv_cache, NULL); | |
386 | } | |
387 | ||
f2bba182 RR |
388 | static int break_on_this_error(struct thread_data *td, int *retptr) |
389 | { | |
390 | int ret = *retptr; | |
391 | ||
392 | if (ret < 0 || td->error) { | |
393 | int err; | |
394 | ||
1ec99eea JA |
395 | if (!td->o.continue_on_error) |
396 | return 1; | |
f2bba182 RR |
397 | |
398 | if (ret < 0) | |
399 | err = -ret; | |
400 | else | |
401 | err = td->error; | |
402 | ||
f2bba182 RR |
403 | if (td_non_fatal_error(err)) { |
404 | /* | |
405 | * Continue with the I/Os in case of | |
406 | * a non fatal error. | |
407 | */ | |
8a4f8a24 | 408 | update_error_count(td, err); |
f2bba182 RR |
409 | td_clear_error(td); |
410 | *retptr = 0; | |
411 | return 0; | |
8a4f8a24 RR |
412 | } else if (td->o.fill_device && err == ENOSPC) { |
413 | /* | |
414 | * We expect to hit this error if | |
415 | * fill_device option is set. | |
416 | */ | |
417 | td_clear_error(td); | |
418 | td->terminate = 1; | |
419 | return 1; | |
f2bba182 RR |
420 | } else { |
421 | /* | |
422 | * Stop the I/O in case of a fatal | |
423 | * error. | |
424 | */ | |
8a4f8a24 | 425 | update_error_count(td, err); |
f2bba182 RR |
426 | return 1; |
427 | } | |
428 | } | |
429 | ||
430 | return 0; | |
431 | } | |
432 | ||
906c8d75 | 433 | /* |
34403fb1 | 434 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
435 | * reads the blocks back in, and checks the crc/md5 of the data. |
436 | */ | |
1e97cce9 | 437 | static void do_verify(struct thread_data *td) |
ebac4655 | 438 | { |
53cdc686 | 439 | struct fio_file *f; |
36167d82 | 440 | struct io_u *io_u; |
af52b345 JA |
441 | int ret, min_events; |
442 | unsigned int i; | |
e5b401d4 | 443 | |
008618a8 JA |
444 | dprint(FD_VERIFY, "starting loop\n"); |
445 | ||
e5b401d4 JA |
446 | /* |
447 | * sync io first and invalidate cache, to make sure we really | |
448 | * read from disk. | |
449 | */ | |
450 | for_each_file(td, f, i) { | |
d6aed795 | 451 | if (!fio_file_open(f)) |
3d7b485f | 452 | continue; |
b2fdda43 JA |
453 | if (fio_io_sync(td, f)) |
454 | break; | |
455 | if (file_invalidate_cache(td, f)) | |
456 | break; | |
e5b401d4 | 457 | } |
ebac4655 | 458 | |
b2fdda43 JA |
459 | if (td->error) |
460 | return; | |
461 | ||
ebac4655 JA |
462 | td_set_runstate(td, TD_VERIFYING); |
463 | ||
36167d82 JA |
464 | io_u = NULL; |
465 | while (!td->terminate) { | |
4950421a | 466 | int ret2, full; |
5451792e | 467 | |
993bf48b JA |
468 | update_tv_cache(td); |
469 | ||
470 | if (runtime_exceeded(td, &td->tv_cache)) { | |
90a87d4b | 471 | td->terminate = 1; |
02bcaa8c | 472 | break; |
069c2918 | 473 | } |
02bcaa8c | 474 | |
dcb4830d RR |
475 | io_u = __get_io_u(td); |
476 | if (!io_u) | |
477 | break; | |
478 | ||
069c2918 JA |
479 | if (get_next_verify(td, io_u)) { |
480 | put_io_u(td, io_u); | |
ebac4655 | 481 | break; |
069c2918 | 482 | } |
ebac4655 | 483 | |
069c2918 JA |
484 | if (td_io_prep(td, io_u)) { |
485 | put_io_u(td, io_u); | |
53cdc686 | 486 | break; |
069c2918 | 487 | } |
d7762cf8 | 488 | |
e8462bd8 JA |
489 | if (td->o.verify_async) |
490 | io_u->end_io = verify_io_u_async; | |
491 | else | |
492 | io_u->end_io = verify_io_u; | |
53cdc686 | 493 | |
11786802 | 494 | ret = td_io_queue(td, io_u); |
36167d82 JA |
495 | switch (ret) { |
496 | case FIO_Q_COMPLETED: | |
f2bba182 | 497 | if (io_u->error) { |
22819ec2 | 498 | ret = -io_u->error; |
f2bba182 RR |
499 | clear_io_u(td, io_u); |
500 | } else if (io_u->resid) { | |
36167d82 | 501 | int bytes = io_u->xfer_buflen - io_u->resid; |
ebac4655 | 502 | |
9e9d164e JA |
503 | /* |
504 | * zero read, fail | |
505 | */ | |
506 | if (!bytes) { | |
41d9f0fd | 507 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
508 | put_io_u(td, io_u); |
509 | break; | |
510 | } | |
8400d9b2 | 511 | |
36167d82 JA |
512 | io_u->xfer_buflen = io_u->resid; |
513 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
514 | io_u->offset += bytes; |
515 | ||
ff58fced JA |
516 | if (ddir_rw(io_u->ddir)) |
517 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 518 | |
2b13e716 | 519 | f = io_u->file; |
d460eb31 | 520 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
521 | goto sync_done; |
522 | ||
11786802 JA |
523 | requeue_io_u(td, &io_u); |
524 | } else { | |
8400d9b2 | 525 | sync_done: |
581e7141 | 526 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
527 | if (ret < 0) |
528 | break; | |
36167d82 | 529 | } |
36167d82 JA |
530 | continue; |
531 | case FIO_Q_QUEUED: | |
532 | break; | |
755200a3 JA |
533 | case FIO_Q_BUSY: |
534 | requeue_io_u(td, &io_u); | |
5451792e JA |
535 | ret2 = td_io_commit(td); |
536 | if (ret2 < 0) | |
537 | ret = ret2; | |
755200a3 | 538 | break; |
36167d82 JA |
539 | default: |
540 | assert(ret < 0); | |
e1161c32 | 541 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
542 | break; |
543 | } | |
544 | ||
f2bba182 | 545 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
546 | break; |
547 | ||
ebac4655 | 548 | /* |
3af6ef39 JA |
549 | * if we can queue more, do so. but check if there are |
550 | * completed io_u's first. | |
ebac4655 | 551 | */ |
4950421a JA |
552 | full = queue_full(td) || ret == FIO_Q_BUSY; |
553 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
554 | min_events = min(td->o.iodepth_batch_complete, |
555 | td->cur_depth); | |
4950421a | 556 | if (full && !min_events) |
838bc709 | 557 | min_events = 1; |
e916b390 | 558 | |
4950421a JA |
559 | do { |
560 | /* | |
561 | * Reap required number of io units, if any, | |
562 | * and do the verification on them through | |
563 | * the callback handler | |
564 | */ | |
581e7141 | 565 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
566 | ret = -1; |
567 | break; | |
568 | } | |
569 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
570 | } | |
571 | if (ret < 0) | |
ebac4655 | 572 | break; |
36167d82 | 573 | } |
ebac4655 | 574 | |
c01c0395 JA |
575 | if (!td->error) { |
576 | min_events = td->cur_depth; | |
577 | ||
578 | if (min_events) | |
581e7141 | 579 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 580 | } else |
ebac4655 JA |
581 | cleanup_pending_aio(td); |
582 | ||
583 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
584 | |
585 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
586 | } |
587 | ||
32cd46a0 | 588 | /* |
906c8d75 | 589 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
590 | * and reaps them, checking for rate and errors along the way. |
591 | */ | |
ebac4655 JA |
592 | static void do_io(struct thread_data *td) |
593 | { | |
af52b345 JA |
594 | unsigned int i; |
595 | int ret = 0; | |
ebac4655 | 596 | |
b29ee5b3 JA |
597 | if (in_ramp_time(td)) |
598 | td_set_runstate(td, TD_RAMP); | |
599 | else | |
600 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 601 | |
457bf399 | 602 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
a917a8b3 | 603 | (!flist_empty(&td->trim_list)) || |
457bf399 | 604 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { |
97601024 | 605 | struct timeval comp_time; |
581e7141 | 606 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 607 | int min_evts = 0; |
ebac4655 | 608 | struct io_u *io_u; |
4950421a | 609 | int ret2, full; |
ebac4655 JA |
610 | |
611 | if (td->terminate) | |
612 | break; | |
613 | ||
993bf48b | 614 | update_tv_cache(td); |
97601024 | 615 | |
993bf48b | 616 | if (runtime_exceeded(td, &td->tv_cache)) { |
90a87d4b | 617 | td->terminate = 1; |
97601024 JA |
618 | break; |
619 | } | |
36167d82 | 620 | |
dcb4830d RR |
621 | io_u = get_io_u(td); |
622 | if (!io_u) | |
623 | break; | |
624 | ||
b67740d3 JA |
625 | /* |
626 | * Add verification end_io handler, if asked to verify | |
627 | * a previously written file. | |
628 | */ | |
715d2b3e JA |
629 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
630 | !td_rw(td)) { | |
e8462bd8 JA |
631 | if (td->o.verify_async) |
632 | io_u->end_io = verify_io_u_async; | |
633 | else | |
634 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 635 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
636 | } else if (in_ramp_time(td)) |
637 | td_set_runstate(td, TD_RAMP); | |
638 | else | |
d8fab1b6 | 639 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 640 | |
11786802 | 641 | ret = td_io_queue(td, io_u); |
36167d82 JA |
642 | switch (ret) { |
643 | case FIO_Q_COMPLETED: | |
f2bba182 | 644 | if (io_u->error) { |
5451792e | 645 | ret = -io_u->error; |
f2bba182 RR |
646 | clear_io_u(td, io_u); |
647 | } else if (io_u->resid) { | |
36167d82 | 648 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 649 | struct fio_file *f = io_u->file; |
36167d82 | 650 | |
9e9d164e JA |
651 | /* |
652 | * zero read, fail | |
653 | */ | |
654 | if (!bytes) { | |
41d9f0fd | 655 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
656 | put_io_u(td, io_u); |
657 | break; | |
658 | } | |
659 | ||
cec6b55d | 660 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 661 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
662 | io_u->offset += bytes; |
663 | ||
ff58fced JA |
664 | if (ddir_rw(io_u->ddir)) |
665 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 666 | |
d460eb31 | 667 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
668 | goto sync_done; |
669 | ||
11786802 JA |
670 | requeue_io_u(td, &io_u); |
671 | } else { | |
5a7c5680 | 672 | sync_done: |
581e7141 JA |
673 | if (__should_check_rate(td, 0) || |
674 | __should_check_rate(td, 1)) | |
993bf48b JA |
675 | fio_gettime(&comp_time, NULL); |
676 | ||
581e7141 JA |
677 | ret = io_u_sync_complete(td, io_u, bytes_done); |
678 | if (ret < 0) | |
679 | break; | |
cec6b55d | 680 | } |
36167d82 JA |
681 | break; |
682 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
683 | /* |
684 | * if the engine doesn't have a commit hook, | |
685 | * the io_u is really queued. if it does have such | |
686 | * a hook, it has to call io_u_queued() itself. | |
687 | */ | |
688 | if (td->io_ops->commit == NULL) | |
689 | io_u_queued(td, io_u); | |
36167d82 | 690 | break; |
755200a3 JA |
691 | case FIO_Q_BUSY: |
692 | requeue_io_u(td, &io_u); | |
5451792e JA |
693 | ret2 = td_io_commit(td); |
694 | if (ret2 < 0) | |
695 | ret = ret2; | |
755200a3 | 696 | break; |
36167d82 JA |
697 | default: |
698 | assert(ret < 0); | |
699 | put_io_u(td, io_u); | |
700 | break; | |
ebac4655 JA |
701 | } |
702 | ||
f2bba182 | 703 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
704 | break; |
705 | ||
97601024 JA |
706 | /* |
707 | * See if we need to complete some commands | |
708 | */ | |
4950421a JA |
709 | full = queue_full(td) || ret == FIO_Q_BUSY; |
710 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
711 | min_evts = min(td->o.iodepth_batch_complete, |
712 | td->cur_depth); | |
4950421a | 713 | if (full && !min_evts) |
36167d82 | 714 | min_evts = 1; |
4950421a | 715 | |
581e7141 JA |
716 | if (__should_check_rate(td, 0) || |
717 | __should_check_rate(td, 1)) | |
993bf48b | 718 | fio_gettime(&comp_time, NULL); |
4950421a JA |
719 | |
720 | do { | |
581e7141 JA |
721 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
722 | if (ret < 0) | |
4950421a JA |
723 | break; |
724 | ||
4950421a | 725 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
726 | } |
727 | ||
4950421a JA |
728 | if (ret < 0) |
729 | break; | |
581e7141 | 730 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
731 | continue; |
732 | ||
581e7141 JA |
733 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
734 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
735 | if (exitall_on_terminate) |
736 | terminate_threads(td->groupid); | |
737 | td_verror(td, EIO, "check_min_rate"); | |
738 | break; | |
739 | } | |
ebac4655 JA |
740 | } |
741 | ||
2dc1bbeb | 742 | if (td->o.thinktime) { |
9c1f7434 JA |
743 | unsigned long long b; |
744 | ||
745 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 746 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
747 | int left; |
748 | ||
2dc1bbeb | 749 | if (td->o.thinktime_spin) |
6dd6f2cd | 750 | usec_spin(td->o.thinktime_spin); |
48097d5c | 751 | |
2dc1bbeb | 752 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
753 | if (left) |
754 | usec_sleep(td, left); | |
755 | } | |
9c1f7434 | 756 | } |
ebac4655 JA |
757 | } |
758 | ||
d708ac5e JA |
759 | if (td->trim_entries) |
760 | printf("trim entries %ld\n", td->trim_entries); | |
a917a8b3 | 761 | |
aa31f1f1 SL |
762 | if (td->o.fill_device && td->error == ENOSPC) { |
763 | td->error = 0; | |
764 | td->terminate = 1; | |
765 | } | |
4d2413c6 | 766 | if (!td->error) { |
3d7c391d JA |
767 | struct fio_file *f; |
768 | ||
c01c0395 JA |
769 | i = td->cur_depth; |
770 | if (i) | |
581e7141 | 771 | ret = io_u_queued_complete(td, i, NULL); |
ebac4655 | 772 | |
2dc1bbeb | 773 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 774 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
775 | |
776 | for_each_file(td, f, i) { | |
d6aed795 | 777 | if (!fio_file_open(f)) |
3d7b485f | 778 | continue; |
858a3d47 | 779 | fio_io_sync(td, f); |
3d7b485f | 780 | } |
84585003 | 781 | } |
c01c0395 JA |
782 | } else |
783 | cleanup_pending_aio(td); | |
2ba1c290 JA |
784 | |
785 | /* | |
786 | * stop job if we failed doing any IO | |
787 | */ | |
788 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
789 | td->done = 1; | |
ebac4655 JA |
790 | } |
791 | ||
ebac4655 JA |
792 | static void cleanup_io_u(struct thread_data *td) |
793 | { | |
01743ee1 | 794 | struct flist_head *entry, *n; |
ebac4655 JA |
795 | struct io_u *io_u; |
796 | ||
01743ee1 JA |
797 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
798 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 799 | |
01743ee1 | 800 | flist_del(&io_u->list); |
91aea6ec | 801 | fio_memfree(io_u, sizeof(*io_u)); |
ebac4655 JA |
802 | } |
803 | ||
2f9ade3c | 804 | free_io_mem(td); |
ebac4655 JA |
805 | } |
806 | ||
807 | static int init_io_u(struct thread_data *td) | |
808 | { | |
809 | struct io_u *io_u; | |
a00735e6 | 810 | unsigned int max_bs; |
eb7ccf38 | 811 | int cl_align, i, max_units; |
ebac4655 JA |
812 | char *p; |
813 | ||
1e3c09af | 814 | max_units = td->o.iodepth; |
2dc1bbeb | 815 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
816 | td->orig_buffer_size = (unsigned long long) max_bs |
817 | * (unsigned long long) max_units; | |
818 | ||
819 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
820 | unsigned long bs; | |
74b025b0 | 821 | |
5ec10eaa JA |
822 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
823 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
824 | } | |
ebac4655 | 825 | |
c3990645 JA |
826 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
827 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
828 | return 1; | |
829 | } | |
830 | ||
2f9ade3c JA |
831 | if (allocate_io_mem(td)) |
832 | return 1; | |
ebac4655 | 833 | |
1c04bf78 JA |
834 | if (td->o.odirect || td->o.mem_align || |
835 | (td->io_ops->flags & FIO_RAWIO)) | |
d529ee19 | 836 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
837 | else |
838 | p = td->orig_buffer; | |
839 | ||
eb7ccf38 JA |
840 | cl_align = os_cache_line_size(); |
841 | ||
ebac4655 | 842 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
843 | void *ptr; |
844 | ||
db1cd90b JA |
845 | if (td->terminate) |
846 | return 1; | |
eb7ccf38 | 847 | |
91aea6ec JA |
848 | ptr = fio_memalign(cl_align, sizeof(*io_u)); |
849 | if (!ptr) { | |
850 | log_err("fio: unable to allocate aligned memory\n"); | |
eb7ccf38 JA |
851 | break; |
852 | } | |
853 | ||
854 | io_u = ptr; | |
ebac4655 | 855 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 856 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 857 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 858 | |
b4c5e1ac JA |
859 | if (!(td->io_ops->flags & FIO_NOIO)) { |
860 | io_u->buf = p + max_bs * i; | |
d529ee19 | 861 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 862 | |
5973cafb JA |
863 | if (td_write(td) && !td->o.refill_buffers) |
864 | io_u_fill_buffer(td, io_u, max_bs); | |
cbe8d756 RR |
865 | else if (td_write(td) && td->o.verify_pattern_bytes) { |
866 | /* | |
867 | * Fill the buffer with the pattern if we are | |
868 | * going to be doing writes. | |
869 | */ | |
7d9fb455 | 870 | fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
cbe8d756 | 871 | } |
b4c5e1ac | 872 | } |
6b9cea23 | 873 | |
b1ff3403 | 874 | io_u->index = i; |
0c6e7517 | 875 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 876 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
877 | } |
878 | ||
879 | return 0; | |
880 | } | |
881 | ||
da86774e JA |
882 | static int switch_ioscheduler(struct thread_data *td) |
883 | { | |
884 | char tmp[256], tmp2[128]; | |
885 | FILE *f; | |
886 | int ret; | |
887 | ||
ba0fbe10 | 888 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
889 | return 0; |
890 | ||
da86774e JA |
891 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
892 | ||
893 | f = fopen(tmp, "r+"); | |
894 | if (!f) { | |
cbf5c1d7 | 895 | if (errno == ENOENT) { |
5ec10eaa JA |
896 | log_err("fio: os or kernel doesn't support IO scheduler" |
897 | " switching\n"); | |
cbf5c1d7 JA |
898 | return 0; |
899 | } | |
900 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
901 | return 1; |
902 | } | |
903 | ||
904 | /* | |
905 | * Set io scheduler. | |
906 | */ | |
2dc1bbeb | 907 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 908 | if (ferror(f) || ret != 1) { |
e1161c32 | 909 | td_verror(td, errno, "fwrite"); |
da86774e JA |
910 | fclose(f); |
911 | return 1; | |
912 | } | |
913 | ||
914 | rewind(f); | |
915 | ||
916 | /* | |
917 | * Read back and check that the selected scheduler is now the default. | |
918 | */ | |
919 | ret = fread(tmp, 1, sizeof(tmp), f); | |
920 | if (ferror(f) || ret < 0) { | |
e1161c32 | 921 | td_verror(td, errno, "fread"); |
da86774e JA |
922 | fclose(f); |
923 | return 1; | |
924 | } | |
925 | ||
2dc1bbeb | 926 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 927 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 928 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 929 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
930 | fclose(f); |
931 | return 1; | |
932 | } | |
933 | ||
934 | fclose(f); | |
935 | return 0; | |
936 | } | |
937 | ||
492158cf JA |
938 | static int keep_running(struct thread_data *td) |
939 | { | |
940 | unsigned long long io_done; | |
941 | ||
20e354ef JA |
942 | if (td->done) |
943 | return 0; | |
492158cf JA |
944 | if (td->o.time_based) |
945 | return 1; | |
946 | if (td->o.loops) { | |
947 | td->o.loops--; | |
948 | return 1; | |
949 | } | |
950 | ||
5ec10eaa JA |
951 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
952 | + td->io_skip_bytes; | |
492158cf JA |
953 | if (io_done < td->o.size) |
954 | return 1; | |
955 | ||
956 | return 0; | |
957 | } | |
958 | ||
b29ee5b3 | 959 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 960 | { |
756867bd | 961 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 962 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 963 | td->zone_bytes = 0; |
581e7141 JA |
964 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
965 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 966 | |
c1324df1 JA |
967 | td->last_was_sync = 0; |
968 | ||
2ba1c290 JA |
969 | /* |
970 | * reset file done count if we are to start over | |
971 | */ | |
972 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 973 | td->nr_done_files = 0; |
5bfc35d7 JA |
974 | |
975 | /* | |
976 | * Set the same seed to get repeatable runs | |
977 | */ | |
978 | td_fill_rand_seeds(td); | |
b29ee5b3 JA |
979 | } |
980 | ||
981 | void reset_all_stats(struct thread_data *td) | |
982 | { | |
983 | struct timeval tv; | |
984 | int i; | |
985 | ||
986 | reset_io_counters(td); | |
987 | ||
988 | for (i = 0; i < 2; i++) { | |
989 | td->io_bytes[i] = 0; | |
990 | td->io_blocks[i] = 0; | |
991 | td->io_issues[i] = 0; | |
992 | td->ts.total_io_u[i] = 0; | |
993 | } | |
9b836561 | 994 | |
b29ee5b3 | 995 | fio_gettime(&tv, NULL); |
cda99fa0 SSP |
996 | td->ts.runtime[0] = 0; |
997 | td->ts.runtime[1] = 0; | |
b29ee5b3 JA |
998 | memcpy(&td->epoch, &tv, sizeof(tv)); |
999 | memcpy(&td->start, &tv, sizeof(tv)); | |
1000 | } | |
1001 | ||
df9c26b1 | 1002 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
1003 | { |
1004 | struct fio_file *f; | |
1005 | unsigned int i; | |
b29ee5b3 JA |
1006 | |
1007 | reset_io_counters(td); | |
281f9b63 | 1008 | |
24ffd2c2 | 1009 | close_files(td); |
df9c26b1 | 1010 | for_each_file(td, f, i) |
d6aed795 | 1011 | fio_file_clear_done(f); |
ebac4655 JA |
1012 | } |
1013 | ||
398e8c4d JA |
1014 | static int exec_string(const char *string) |
1015 | { | |
1016 | int ret, newlen = strlen(string) + 1 + 8; | |
1017 | char *str; | |
1018 | ||
1019 | str = malloc(newlen); | |
1020 | sprintf(str, "sh -c %s", string); | |
1021 | ||
1022 | ret = system(str); | |
1023 | if (ret == -1) | |
1024 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1025 | ||
1026 | free(str); | |
1027 | return ret; | |
1028 | } | |
1029 | ||
906c8d75 JA |
1030 | /* |
1031 | * Entry point for the thread based jobs. The process based jobs end up | |
1032 | * here as well, after a little setup. | |
1033 | */ | |
ebac4655 JA |
1034 | static void *thread_main(void *data) |
1035 | { | |
cda99fa0 | 1036 | unsigned long long elapsed; |
ebac4655 | 1037 | struct thread_data *td = data; |
e8462bd8 | 1038 | pthread_condattr_t attr; |
a978ba68 | 1039 | int clear_state; |
ebac4655 | 1040 | |
2dc1bbeb | 1041 | if (!td->o.use_thread) |
ebac4655 JA |
1042 | setsid(); |
1043 | ||
1044 | td->pid = getpid(); | |
1045 | ||
5921e80c | 1046 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1047 | |
01743ee1 JA |
1048 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1049 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1050 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1051 | INIT_FLIST_HEAD(&td->io_log_list); | |
1052 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 | 1053 | INIT_FLIST_HEAD(&td->verify_list); |
0d29de83 | 1054 | INIT_FLIST_HEAD(&td->trim_list); |
e8462bd8 | 1055 | pthread_mutex_init(&td->io_u_lock, NULL); |
bb5d7d0b | 1056 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1057 | |
e8462bd8 JA |
1058 | pthread_condattr_init(&attr); |
1059 | pthread_cond_init(&td->verify_cond, &attr); | |
1060 | pthread_cond_init(&td->free_cond, &attr); | |
1061 | ||
db1cd90b | 1062 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1063 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1064 | fio_mutex_up(startup_mutex); |
29adda3c | 1065 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1066 | fio_mutex_down(td->mutex); |
29adda3c | 1067 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1068 | |
1069 | /* | |
cdd18ad8 | 1070 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1071 | * eating a file descriptor |
1072 | */ | |
cdd18ad8 | 1073 | fio_mutex_remove(td->mutex); |
db1cd90b | 1074 | |
dfc6e751 SH |
1075 | /* |
1076 | * A new gid requires privilege, so we need to do this before setting | |
1077 | * the uid. | |
1078 | */ | |
e0b0d892 JA |
1079 | if (td->o.gid != -1U && setgid(td->o.gid)) { |
1080 | td_verror(td, errno, "setgid"); | |
1081 | goto err; | |
1082 | } | |
dfc6e751 SH |
1083 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1084 | td_verror(td, errno, "setuid"); | |
1085 | goto err; | |
1086 | } | |
e0b0d892 | 1087 | |
d84f8d49 JA |
1088 | /* |
1089 | * May alter parameters that init_io_u() will use, so we need to | |
1090 | * do this first. | |
1091 | */ | |
1092 | if (init_iolog(td)) | |
1093 | goto err; | |
1094 | ||
ebac4655 | 1095 | if (init_io_u(td)) |
db1cd90b | 1096 | goto err; |
ebac4655 | 1097 | |
e8462bd8 JA |
1098 | if (td->o.verify_async && verify_async_init(td)) |
1099 | goto err; | |
1100 | ||
1101 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
e1161c32 | 1102 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 1103 | goto err; |
ebac4655 JA |
1104 | } |
1105 | ||
f484470b JA |
1106 | /* |
1107 | * If we have a gettimeofday() thread, make sure we exclude that | |
1108 | * thread from this job | |
1109 | */ | |
be4ecfdf | 1110 | if (td->o.gtod_cpu) { |
be4ecfdf | 1111 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); |
e8462bd8 | 1112 | if (fio_setaffinity(td->pid, td->o.cpumask) == -1) { |
be4ecfdf JA |
1113 | td_verror(td, errno, "cpu_set_affinity"); |
1114 | goto err; | |
1115 | } | |
1116 | } | |
1117 | ||
ac684785 | 1118 | if (td->ioprio_set) { |
ebac4655 | 1119 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1120 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1121 | goto err; |
ebac4655 JA |
1122 | } |
1123 | } | |
1124 | ||
6adb38a1 | 1125 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1126 | goto err; |
1127 | ||
2dc1bbeb | 1128 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1129 | td_verror(td, errno, "nice"); |
db1cd90b | 1130 | goto err; |
b6f4d880 JA |
1131 | } |
1132 | ||
2dc1bbeb | 1133 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1134 | goto err; |
37f56873 | 1135 | |
2dc1bbeb | 1136 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1137 | goto err; |
1138 | ||
7d6c5283 JA |
1139 | if (td_io_init(td)) |
1140 | goto err; | |
1141 | ||
68727076 JA |
1142 | if (init_random_map(td)) |
1143 | goto err; | |
1144 | ||
2dc1bbeb | 1145 | if (td->o.exec_prerun) { |
398e8c4d | 1146 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1147 | goto err; |
1148 | } | |
4e0ba8af | 1149 | |
afad68f7 ZY |
1150 | if (td->o.pre_read) { |
1151 | if (pre_read_files(td) < 0) | |
1152 | goto err; | |
1153 | } | |
1154 | ||
69008999 | 1155 | fio_gettime(&td->epoch, NULL); |
756867bd | 1156 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 | 1157 | |
a978ba68 | 1158 | clear_state = 0; |
492158cf | 1159 | while (keep_running(td)) { |
02bcaa8c | 1160 | fio_gettime(&td->start, NULL); |
c5d3d790 JA |
1161 | memcpy(&td->ts.stat_sample_time[0], &td->start, |
1162 | sizeof(td->start)); | |
1163 | memcpy(&td->ts.stat_sample_time[1], &td->start, | |
1164 | sizeof(td->start)); | |
993bf48b | 1165 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1166 | |
cf2dd8d6 | 1167 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
5ec10eaa JA |
1168 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
1169 | sizeof(td->lastrate)); | |
ebac4655 | 1170 | |
df9c26b1 JA |
1171 | if (clear_state) |
1172 | clear_io_state(td); | |
a978ba68 | 1173 | |
ebac4655 JA |
1174 | prune_io_piece_log(td); |
1175 | ||
ba0fbe10 | 1176 | do_io(td); |
ebac4655 | 1177 | |
a978ba68 JA |
1178 | clear_state = 1; |
1179 | ||
38d77cae | 1180 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1181 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1182 | td->ts.runtime[DDIR_READ] += elapsed; |
38d77cae JA |
1183 | } |
1184 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1185 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1186 | td->ts.runtime[DDIR_WRITE] += elapsed; |
38d77cae | 1187 | } |
5ec10eaa | 1188 | |
ebac4655 JA |
1189 | if (td->error || td->terminate) |
1190 | break; | |
1191 | ||
e84c73a8 SL |
1192 | if (!td->o.do_verify || |
1193 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1194 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1195 | continue; |
1196 | ||
df9c26b1 | 1197 | clear_io_state(td); |
a978ba68 | 1198 | |
02bcaa8c | 1199 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1200 | |
1201 | do_verify(td); | |
1202 | ||
cda99fa0 | 1203 | td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1204 | |
1205 | if (td->error || td->terminate) | |
1206 | break; | |
1207 | } | |
1208 | ||
36dff966 | 1209 | update_rusage_stat(td); |
cda99fa0 SSP |
1210 | td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000; |
1211 | td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000; | |
756867bd JA |
1212 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1213 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1214 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1215 | ||
d2bbb8d9 | 1216 | fio_mutex_down(writeout_mutex); |
e3cedca7 | 1217 | if (td->ts.bw_log) { |
f484470b JA |
1218 | if (td->o.bw_log_file) { |
1219 | finish_log_named(td, td->ts.bw_log, | |
1220 | td->o.bw_log_file, "bw"); | |
1221 | } else | |
e3cedca7 JA |
1222 | finish_log(td, td->ts.bw_log, "bw"); |
1223 | } | |
02af0988 JA |
1224 | if (td->ts.lat_log) { |
1225 | if (td->o.lat_log_file) { | |
1226 | finish_log_named(td, td->ts.lat_log, | |
1227 | td->o.lat_log_file, "lat"); | |
1228 | } else | |
1229 | finish_log(td, td->ts.lat_log, "lat"); | |
1230 | } | |
e3cedca7 | 1231 | if (td->ts.slat_log) { |
f484470b JA |
1232 | if (td->o.lat_log_file) { |
1233 | finish_log_named(td, td->ts.slat_log, | |
1ca95738 | 1234 | td->o.lat_log_file, "slat"); |
f484470b | 1235 | } else |
e3cedca7 JA |
1236 | finish_log(td, td->ts.slat_log, "slat"); |
1237 | } | |
1238 | if (td->ts.clat_log) { | |
f484470b JA |
1239 | if (td->o.lat_log_file) { |
1240 | finish_log_named(td, td->ts.clat_log, | |
1241 | td->o.lat_log_file, "clat"); | |
1242 | } else | |
e3cedca7 JA |
1243 | finish_log(td, td->ts.clat_log, "clat"); |
1244 | } | |
d2bbb8d9 | 1245 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1246 | if (td->o.exec_postrun) |
1247 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1248 | |
1249 | if (exitall_on_terminate) | |
390c40e2 | 1250 | terminate_threads(td->groupid); |
ebac4655 JA |
1251 | |
1252 | err: | |
5bf13a5a | 1253 | if (td->error) |
4ceb30d4 | 1254 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1255 | td->verror); |
b4707ed7 JA |
1256 | |
1257 | if (td->o.verify_async) | |
1258 | verify_async_exit(td); | |
1259 | ||
24ffd2c2 | 1260 | close_and_free_files(td); |
2866c82d | 1261 | close_ioengine(td); |
ebac4655 | 1262 | cleanup_io_u(td); |
6adb38a1 | 1263 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1264 | |
d2ce18b5 JA |
1265 | if (td->o.cpumask_set) { |
1266 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1267 | ||
1268 | td_verror(td, ret, "fio_cpuset_exit"); | |
1269 | } | |
4e78e405 | 1270 | |
f29b25a3 JA |
1271 | /* |
1272 | * do this very late, it will log file closing as well | |
1273 | */ | |
1274 | if (td->o.write_iolog_file) | |
1275 | write_iolog_close(td); | |
1276 | ||
d23bb327 | 1277 | options_mem_free(td); |
ebac4655 | 1278 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1279 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1280 | } |
1281 | ||
906c8d75 JA |
1282 | /* |
1283 | * We cannot pass the td data into a forked process, so attach the td and | |
1284 | * pass it to the thread worker. | |
1285 | */ | |
a6418147 | 1286 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1287 | { |
1288 | struct thread_data *td; | |
a6418147 | 1289 | void *data, *ret; |
ebac4655 JA |
1290 | |
1291 | data = shmat(shmid, NULL, 0); | |
1292 | if (data == (void *) -1) { | |
a6418147 JA |
1293 | int __err = errno; |
1294 | ||
ebac4655 | 1295 | perror("shmat"); |
a6418147 | 1296 | return __err; |
ebac4655 JA |
1297 | } |
1298 | ||
1299 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1300 | ret = thread_main(td); |
ebac4655 | 1301 | shmdt(data); |
43d76807 | 1302 | return (int) (unsigned long) ret; |
ebac4655 JA |
1303 | } |
1304 | ||
906c8d75 JA |
1305 | /* |
1306 | * Run over the job map and reap the threads that have exited, if any. | |
1307 | */ | |
ebac4655 JA |
1308 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1309 | { | |
34572e28 | 1310 | struct thread_data *td; |
1f809d15 | 1311 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1312 | |
1313 | /* | |
1314 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1315 | */ | |
1f809d15 | 1316 | realthreads = pending = cputhreads = 0; |
34572e28 | 1317 | for_each_td(td, i) { |
3707f45b | 1318 | int flags = 0; |
a2f77c9f | 1319 | |
84585003 JA |
1320 | /* |
1321 | * ->io_ops is NULL for a thread that has closed its | |
1322 | * io engine | |
1323 | */ | |
ba0fbe10 | 1324 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1325 | cputhreads++; |
1f809d15 JA |
1326 | else |
1327 | realthreads++; | |
b990b5c0 | 1328 | |
a1dedc1f AC |
1329 | if (!td->pid) { |
1330 | pending++; | |
1331 | continue; | |
1332 | } | |
1333 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1334 | continue; |
2dc1bbeb | 1335 | if (td->o.use_thread) { |
3707f45b JA |
1336 | if (td->runstate == TD_EXITED) { |
1337 | td_set_runstate(td, TD_REAPED); | |
1338 | goto reaped; | |
1339 | } | |
1340 | continue; | |
1341 | } | |
a2f77c9f JA |
1342 | |
1343 | flags = WNOHANG; | |
1344 | if (td->runstate == TD_EXITED) | |
1345 | flags = 0; | |
1346 | ||
1347 | /* | |
1348 | * check if someone quit or got killed in an unusual way | |
1349 | */ | |
1350 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1351 | if (ret < 0) { |
a2f77c9f | 1352 | if (errno == ECHILD) { |
5921e80c JA |
1353 | log_err("fio: pid=%d disappeared %d\n", |
1354 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1355 | td_set_runstate(td, TD_REAPED); |
1356 | goto reaped; | |
1357 | } | |
1358 | perror("waitpid"); | |
1359 | } else if (ret == td->pid) { | |
1360 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1361 | int sig = WTERMSIG(status); |
1362 | ||
03e20d68 | 1363 | if (sig != SIGTERM) |
5ec10eaa | 1364 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1365 | (int) td->pid, sig); |
fab6aa71 JA |
1366 | td_set_runstate(td, TD_REAPED); |
1367 | goto reaped; | |
1368 | } | |
a2f77c9f JA |
1369 | if (WIFEXITED(status)) { |
1370 | if (WEXITSTATUS(status) && !td->error) | |
1371 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1372 | |
a2f77c9f JA |
1373 | td_set_runstate(td, TD_REAPED); |
1374 | goto reaped; | |
a6418147 JA |
1375 | } |
1376 | } | |
ebac4655 | 1377 | |
a2f77c9f JA |
1378 | /* |
1379 | * thread is not dead, continue | |
1380 | */ | |
e48676ba | 1381 | pending++; |
a2f77c9f | 1382 | continue; |
fab6aa71 | 1383 | reaped: |
ebac4655 | 1384 | (*nr_running)--; |
581e7141 JA |
1385 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1386 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1387 | if (!td->pid) |
1388 | pending--; | |
a2f77c9f JA |
1389 | |
1390 | if (td->error) | |
1391 | exit_value++; | |
d3eeeabc JA |
1392 | |
1393 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1394 | } |
b990b5c0 | 1395 | |
1f809d15 | 1396 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1397 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1398 | } |
1399 | ||
be4ecfdf JA |
1400 | static void *gtod_thread_main(void *data) |
1401 | { | |
1402 | fio_mutex_up(startup_mutex); | |
1403 | ||
1404 | /* | |
1405 | * As long as we have jobs around, update the clock. It would be nice | |
1406 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1407 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1408 | * it - we don't want to lose precision. | |
1409 | */ | |
1410 | while (threads) { | |
1411 | fio_gtod_update(); | |
1412 | nop; | |
1413 | } | |
1414 | ||
1415 | return NULL; | |
1416 | } | |
1417 | ||
1418 | static int fio_start_gtod_thread(void) | |
1419 | { | |
304a47c7 | 1420 | pthread_attr_t attr; |
f42c153d JA |
1421 | int ret; |
1422 | ||
304a47c7 VA |
1423 | pthread_attr_init(&attr); |
1424 | pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); | |
1425 | ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); | |
1426 | pthread_attr_destroy(&attr); | |
f42c153d JA |
1427 | if (ret) { |
1428 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1429 | return 1; |
1430 | } | |
f42c153d JA |
1431 | |
1432 | ret = pthread_detach(gtod_thread); | |
1433 | if (ret) { | |
1434 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1435 | return 1; |
1436 | } | |
1437 | ||
29adda3c | 1438 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1439 | fio_mutex_down(startup_mutex); |
29adda3c | 1440 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1441 | return 0; |
1442 | } | |
1443 | ||
906c8d75 JA |
1444 | /* |
1445 | * Main function for kicking off and reaping jobs, as needed. | |
1446 | */ | |
ebac4655 JA |
1447 | static void run_threads(void) |
1448 | { | |
ebac4655 JA |
1449 | struct thread_data *td; |
1450 | unsigned long spent; | |
1451 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1452 | |
2f9ade3c JA |
1453 | if (fio_pin_memory()) |
1454 | return; | |
ebac4655 | 1455 | |
be4ecfdf JA |
1456 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1457 | return; | |
1458 | ||
c6ae0a5b | 1459 | if (!terse_output) { |
4ceb30d4 | 1460 | log_info("Starting "); |
9cedf167 | 1461 | if (nr_thread) |
4ceb30d4 | 1462 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1463 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1464 | if (nr_process) { |
1465 | if (nr_thread) | |
1466 | printf(" and "); | |
4ceb30d4 | 1467 | log_info("%d process%s", nr_process, |
5ec10eaa | 1468 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1469 | } |
4ceb30d4 | 1470 | log_info("\n"); |
c6ae0a5b JA |
1471 | fflush(stdout); |
1472 | } | |
c04f7ec3 | 1473 | |
697a606c | 1474 | set_sig_handlers(); |
4efa970e | 1475 | |
ebac4655 JA |
1476 | todo = thread_number; |
1477 | nr_running = 0; | |
1478 | nr_started = 0; | |
1479 | m_rate = t_rate = 0; | |
1480 | ||
34572e28 | 1481 | for_each_td(td, i) { |
263e529f | 1482 | print_status_init(td->thread_number - 1); |
ebac4655 | 1483 | |
2dc1bbeb | 1484 | if (!td->o.create_serialize) { |
380cf265 | 1485 | init_disk_util(td); |
ebac4655 | 1486 | continue; |
380cf265 | 1487 | } |
ebac4655 JA |
1488 | |
1489 | /* | |
1490 | * do file setup here so it happens sequentially, | |
1491 | * we don't want X number of threads getting their | |
1492 | * client data interspersed on disk | |
1493 | */ | |
53cdc686 | 1494 | if (setup_files(td)) { |
5bf13a5a JA |
1495 | exit_value++; |
1496 | if (td->error) | |
5921e80c JA |
1497 | log_err("fio: pid=%d, err=%d/%s\n", |
1498 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1499 | td_set_runstate(td, TD_REAPED); |
1500 | todo--; | |
c69e8875 JA |
1501 | } else { |
1502 | struct fio_file *f; | |
2b13e716 | 1503 | unsigned int j; |
c69e8875 JA |
1504 | |
1505 | /* | |
1506 | * for sharing to work, each job must always open | |
1507 | * its own files. so close them, if we opened them | |
1508 | * for creation | |
1509 | */ | |
2b13e716 | 1510 | for_each_file(td, f, j) { |
22a57ba8 JA |
1511 | if (fio_file_open(f)) |
1512 | td_io_close_file(td, f); | |
22a57ba8 | 1513 | } |
5ec10eaa | 1514 | } |
380cf265 JA |
1515 | |
1516 | init_disk_util(td); | |
ebac4655 JA |
1517 | } |
1518 | ||
a2f77c9f JA |
1519 | set_genesis_time(); |
1520 | ||
ebac4655 | 1521 | while (todo) { |
75154845 JA |
1522 | struct thread_data *map[MAX_JOBS]; |
1523 | struct timeval this_start; | |
1524 | int this_jobs = 0, left; | |
1525 | ||
ebac4655 JA |
1526 | /* |
1527 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1528 | */ | |
34572e28 | 1529 | for_each_td(td, i) { |
ebac4655 JA |
1530 | if (td->runstate != TD_NOT_CREATED) |
1531 | continue; | |
1532 | ||
1533 | /* | |
1534 | * never got a chance to start, killed by other | |
1535 | * thread for some reason | |
1536 | */ | |
1537 | if (td->terminate) { | |
1538 | todo--; | |
1539 | continue; | |
1540 | } | |
1541 | ||
2dc1bbeb | 1542 | if (td->o.start_delay) { |
263e529f | 1543 | spent = mtime_since_genesis(); |
ebac4655 | 1544 | |
2dc1bbeb | 1545 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1546 | continue; |
1547 | } | |
1548 | ||
6f88bcb0 JA |
1549 | if (td->o.stonewall && (nr_started || nr_running)) { |
1550 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1551 | td->o.name); | |
ebac4655 | 1552 | break; |
6f88bcb0 | 1553 | } |
ebac4655 | 1554 | |
75154845 JA |
1555 | /* |
1556 | * Set state to created. Thread will transition | |
1557 | * to TD_INITIALIZED when it's done setting up. | |
1558 | */ | |
ebac4655 | 1559 | td_set_runstate(td, TD_CREATED); |
75154845 | 1560 | map[this_jobs++] = td; |
ebac4655 JA |
1561 | nr_started++; |
1562 | ||
2dc1bbeb | 1563 | if (td->o.use_thread) { |
f42c153d JA |
1564 | int ret; |
1565 | ||
ee56ad50 | 1566 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1567 | ret = pthread_create(&td->thread, NULL, |
1568 | thread_main, td); | |
1569 | if (ret) { | |
1570 | log_err("pthread_create: %s\n", | |
1571 | strerror(ret)); | |
ebac4655 | 1572 | nr_started--; |
c8bb6faf | 1573 | break; |
ebac4655 | 1574 | } |
f42c153d JA |
1575 | ret = pthread_detach(td->thread); |
1576 | if (ret) | |
1577 | log_err("pthread_detach: %s", | |
1578 | strerror(ret)); | |
ebac4655 | 1579 | } else { |
5e1d306e | 1580 | pid_t pid; |
ee56ad50 | 1581 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1582 | pid = fork(); |
1583 | if (!pid) { | |
a6418147 JA |
1584 | int ret = fork_main(shm_id, i); |
1585 | ||
5e1d306e JA |
1586 | _exit(ret); |
1587 | } else if (i == fio_debug_jobno) | |
1588 | *fio_debug_jobp = pid; | |
ebac4655 | 1589 | } |
29adda3c | 1590 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1591 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1592 | log_err("fio: job startup hung? exiting.\n"); | |
1593 | terminate_threads(TERMINATE_ALL); | |
1594 | fio_abort = 1; | |
1595 | nr_started--; | |
1596 | break; | |
1597 | } | |
29adda3c | 1598 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1599 | } |
1600 | ||
1601 | /* | |
75154845 JA |
1602 | * Wait for the started threads to transition to |
1603 | * TD_INITIALIZED. | |
ebac4655 | 1604 | */ |
02bcaa8c | 1605 | fio_gettime(&this_start, NULL); |
75154845 | 1606 | left = this_jobs; |
6ce15a32 | 1607 | while (left && !fio_abort) { |
75154845 JA |
1608 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1609 | break; | |
1610 | ||
1611 | usleep(100000); | |
1612 | ||
1613 | for (i = 0; i < this_jobs; i++) { | |
1614 | td = map[i]; | |
1615 | if (!td) | |
1616 | continue; | |
b6f4d880 | 1617 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1618 | map[i] = NULL; |
1619 | left--; | |
b6f4d880 JA |
1620 | } else if (td->runstate >= TD_EXITED) { |
1621 | map[i] = NULL; | |
1622 | left--; | |
1623 | todo--; | |
1624 | nr_running++; /* work-around... */ | |
75154845 JA |
1625 | } |
1626 | } | |
1627 | } | |
1628 | ||
1629 | if (left) { | |
3b70d7e5 | 1630 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1631 | for (i = 0; i < this_jobs; i++) { |
1632 | td = map[i]; | |
1633 | if (!td) | |
1634 | continue; | |
1635 | kill(td->pid, SIGTERM); | |
1636 | } | |
1637 | break; | |
1638 | } | |
1639 | ||
1640 | /* | |
b6f4d880 | 1641 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1642 | */ |
34572e28 | 1643 | for_each_td(td, i) { |
75154845 | 1644 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1645 | continue; |
1646 | ||
b29ee5b3 JA |
1647 | if (in_ramp_time(td)) |
1648 | td_set_runstate(td, TD_RAMP); | |
1649 | else | |
1650 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1651 | nr_running++; |
1652 | nr_started--; | |
581e7141 JA |
1653 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1654 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1655 | todo--; |
cdd18ad8 | 1656 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1657 | } |
1658 | ||
1659 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1660 | ||
1661 | if (todo) | |
1662 | usleep(100000); | |
1663 | } | |
1664 | ||
1665 | while (nr_running) { | |
1666 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1667 | usleep(10000); | |
1668 | } | |
1669 | ||
1670 | update_io_ticks(); | |
2f9ade3c | 1671 | fio_unpin_memory(); |
ebac4655 JA |
1672 | } |
1673 | ||
ebac4655 JA |
1674 | int main(int argc, char *argv[]) |
1675 | { | |
29d610e1 JA |
1676 | long ps; |
1677 | ||
2e5cdb11 | 1678 | sinit(); |
d54a144d | 1679 | init_rand(&__fio_rand_state); |
2e5cdb11 | 1680 | |
dbe1125e JA |
1681 | /* |
1682 | * We need locale for number printing, if it isn't set then just | |
1683 | * go with the US format. | |
1684 | */ | |
1685 | if (!getenv("LC_NUMERIC")) | |
1686 | setlocale(LC_NUMERIC, "en_US"); | |
1687 | ||
29d610e1 JA |
1688 | ps = sysconf(_SC_PAGESIZE); |
1689 | if (ps < 0) { | |
1690 | log_err("Failed to get page size\n"); | |
1691 | return 1; | |
1692 | } | |
1693 | ||
cfc99db7 | 1694 | page_size = ps; |
29d610e1 JA |
1695 | page_mask = ps - 1; |
1696 | ||
74929ac2 JA |
1697 | fio_keywords_init(); |
1698 | ||
1699 | if (parse_options(argc, argv)) | |
1700 | return 1; | |
1701 | ||
07b3232d JA |
1702 | if (exec_profile && load_profile(exec_profile)) |
1703 | return 1; | |
1704 | ||
74929ac2 JA |
1705 | if (!thread_number) |
1706 | return 0; | |
1707 | ||
bb3884d8 JA |
1708 | if (write_bw_log) { |
1709 | setup_log(&agg_io_log[DDIR_READ]); | |
1710 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1711 | } | |
1712 | ||
cdd18ad8 | 1713 | startup_mutex = fio_mutex_init(0); |
03e20d68 BC |
1714 | if (startup_mutex == NULL) |
1715 | return 1; | |
d2bbb8d9 | 1716 | writeout_mutex = fio_mutex_init(1); |
03e20d68 BC |
1717 | if (writeout_mutex == NULL) |
1718 | return 1; | |
07739b57 | 1719 | |
a2f77c9f | 1720 | set_genesis_time(); |
7985701b | 1721 | create_disk_util_thread(); |
ebac4655 | 1722 | |
dae5341c JA |
1723 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1724 | INIT_FLIST_HEAD(cgroup_list); | |
1725 | ||
ebac4655 | 1726 | run_threads(); |
6ce15a32 | 1727 | |
bb3884d8 | 1728 | if (!fio_abort) { |
6ce15a32 | 1729 | show_run_stats(); |
bb3884d8 | 1730 | if (write_bw_log) { |
5ec10eaa JA |
1731 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1732 | __finish_log(agg_io_log[DDIR_WRITE], | |
1733 | "agg-write_bw.log"); | |
bb3884d8 JA |
1734 | } |
1735 | } | |
ebac4655 | 1736 | |
dae5341c JA |
1737 | cgroup_kill(cgroup_list); |
1738 | sfree(cgroup_list); | |
61eb313e | 1739 | sfree(cgroup_mnt); |
39f22027 | 1740 | |
cdd18ad8 | 1741 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1742 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1743 | return exit_value; |
ebac4655 | 1744 | } |