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 | |
d3cc4ebf JA |
73 | unsigned long arch_flags = 0; |
74 | ||
bb3884d8 JA |
75 | struct io_log *agg_io_log[2]; |
76 | ||
ebac4655 | 77 | #define TERMINATE_ALL (-1) |
75154845 | 78 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 79 | |
b29ee5b3 | 80 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 81 | { |
d8fab1b6 JA |
82 | if (td->runstate == runstate) |
83 | return; | |
84 | ||
5921e80c JA |
85 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
86 | td->runstate, runstate); | |
6ce15a32 JA |
87 | td->runstate = runstate; |
88 | } | |
89 | ||
390c40e2 | 90 | static void terminate_threads(int group_id) |
ebac4655 | 91 | { |
34572e28 | 92 | struct thread_data *td; |
ebac4655 JA |
93 | int i; |
94 | ||
6f88bcb0 JA |
95 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
96 | ||
34572e28 | 97 | for_each_td(td, i) { |
ebac4655 | 98 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 99 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 100 | td->o.name, (int) td->pid); |
ad830ed7 JA |
101 | td->terminate = 1; |
102 | td->o.start_delay = 0; | |
103 | ||
7a93ab19 JA |
104 | /* |
105 | * if the thread is running, just let it exit | |
106 | */ | |
6aa27816 MC |
107 | if (!td->pid) |
108 | continue; | |
109 | else if (td->runstate < TD_RAMP) | |
03e20d68 | 110 | kill(td->pid, SIGTERM); |
f63f7f3b JA |
111 | else { |
112 | struct ioengine_ops *ops = td->io_ops; | |
113 | ||
03e20d68 BC |
114 | if (ops && (ops->flags & FIO_SIGTERM)) |
115 | kill(td->pid, SIGTERM); | |
f63f7f3b | 116 | } |
ebac4655 JA |
117 | } |
118 | } | |
119 | } | |
120 | ||
697a606c JA |
121 | static void sig_int(int sig) |
122 | { | |
123 | if (threads) { | |
4ceb30d4 | 124 | log_info("\nfio: terminating on signal %d\n", sig); |
5ec10eaa | 125 | fflush(stdout); |
deaa1af2 | 126 | exit_value = 128; |
5ec10eaa | 127 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
128 | } |
129 | } | |
130 | ||
7985701b | 131 | static void *disk_thread_main(void *data) |
a5b01f1b | 132 | { |
7985701b JA |
133 | fio_mutex_up(startup_mutex); |
134 | ||
135 | while (threads) { | |
136 | usleep(DISK_UTIL_MSEC * 1000); | |
9a793c2f JA |
137 | if (!threads) |
138 | break; | |
7985701b JA |
139 | update_io_ticks(); |
140 | print_thread_status(); | |
141 | } | |
142 | ||
143 | return NULL; | |
a5b01f1b JA |
144 | } |
145 | ||
7985701b | 146 | static int create_disk_util_thread(void) |
a5b01f1b | 147 | { |
7985701b | 148 | int ret; |
a5b01f1b | 149 | |
69204d6e | 150 | ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); |
7985701b JA |
151 | if (ret) { |
152 | log_err("Can't create disk util thread: %s\n", strerror(ret)); | |
153 | return 1; | |
154 | } | |
a5b01f1b | 155 | |
7985701b JA |
156 | ret = pthread_detach(disk_util_thread); |
157 | if (ret) { | |
158 | log_err("Can't detatch disk util thread: %s\n", strerror(ret)); | |
159 | return 1; | |
160 | } | |
03e20d68 | 161 | |
7985701b JA |
162 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
163 | fio_mutex_down(startup_mutex); | |
164 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
165 | return 0; | |
a5b01f1b JA |
166 | } |
167 | ||
697a606c JA |
168 | static void set_sig_handlers(void) |
169 | { | |
170 | struct sigaction act; | |
171 | ||
697a606c JA |
172 | memset(&act, 0, sizeof(act)); |
173 | act.sa_handler = sig_int; | |
174 | act.sa_flags = SA_RESTART; | |
175 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 176 | |
5f8f9726 | 177 | memset(&act, 0, sizeof(act)); |
b20b6be4 | 178 | act.sa_handler = sig_int; |
5f8f9726 | 179 | act.sa_flags = SA_RESTART; |
03e20d68 | 180 | sigaction(SIGTERM, &act, NULL); |
697a606c JA |
181 | } |
182 | ||
906c8d75 JA |
183 | /* |
184 | * Check if we are above the minimum rate given. | |
185 | */ | |
581e7141 | 186 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
bd93e34b | 187 | enum fio_ddir ddir) |
ebac4655 | 188 | { |
0904200b | 189 | unsigned long long bytes = 0; |
4e991c23 | 190 | unsigned long iops = 0; |
ebac4655 JA |
191 | unsigned long spent; |
192 | unsigned long rate; | |
581e7141 JA |
193 | unsigned int ratemin = 0; |
194 | unsigned int rate_iops = 0; | |
195 | unsigned int rate_iops_min = 0; | |
ebac4655 | 196 | |
ff58fced JA |
197 | assert(ddir_rw(ddir)); |
198 | ||
d982d873 JA |
199 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
200 | return 0; | |
201 | ||
ebac4655 JA |
202 | /* |
203 | * allow a 2 second settle period in the beginning | |
204 | */ | |
205 | if (mtime_since(&td->start, now) < 2000) | |
206 | return 0; | |
207 | ||
581e7141 JA |
208 | iops += td->io_blocks[ddir]; |
209 | bytes += td->this_io_bytes[ddir]; | |
210 | ratemin += td->o.ratemin[ddir]; | |
211 | rate_iops += td->o.rate_iops[ddir]; | |
212 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 213 | |
ebac4655 JA |
214 | /* |
215 | * if rate blocks is set, sample is running | |
216 | */ | |
581e7141 JA |
217 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
218 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 219 | if (spent < td->o.ratecycle) |
ebac4655 JA |
220 | return 0; |
221 | ||
581e7141 | 222 | if (td->o.rate[ddir]) { |
4e991c23 JA |
223 | /* |
224 | * check bandwidth specified rate | |
225 | */ | |
581e7141 | 226 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 227 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 228 | ratemin); |
4e991c23 JA |
229 | return 1; |
230 | } else { | |
581e7141 JA |
231 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
232 | if (rate < ratemin || | |
233 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 234 | log_err("%s: min rate %u not met, got" |
b22989b9 | 235 | " %luKB/sec\n", td->o.name, |
581e7141 | 236 | ratemin, rate); |
4e991c23 JA |
237 | return 1; |
238 | } | |
239 | } | |
413dd459 | 240 | } else { |
4e991c23 JA |
241 | /* |
242 | * checks iops specified rate | |
243 | */ | |
581e7141 | 244 | if (iops < rate_iops) { |
5ec10eaa | 245 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 246 | td->o.name, rate_iops); |
413dd459 | 247 | return 1; |
4e991c23 | 248 | } else { |
581e7141 JA |
249 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
250 | if (rate < rate_iops_min || | |
251 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
252 | log_err("%s: min iops rate %u not met," |
253 | " got %lu\n", td->o.name, | |
581e7141 | 254 | rate_iops_min, rate); |
4e991c23 | 255 | } |
413dd459 | 256 | } |
ebac4655 JA |
257 | } |
258 | } | |
259 | ||
581e7141 JA |
260 | td->rate_bytes[ddir] = bytes; |
261 | td->rate_blocks[ddir] = iops; | |
262 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
263 | return 0; |
264 | } | |
265 | ||
581e7141 JA |
266 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
267 | unsigned long *bytes_done) | |
268 | { | |
269 | int ret = 0; | |
270 | ||
271 | if (bytes_done[0]) | |
272 | ret |= __check_min_rate(td, now, 0); | |
273 | if (bytes_done[1]) | |
274 | ret |= __check_min_rate(td, now, 1); | |
275 | ||
276 | return ret; | |
277 | } | |
278 | ||
ebac4655 JA |
279 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
280 | { | |
2dc1bbeb | 281 | if (!td->o.timeout) |
ebac4655 | 282 | return 0; |
2dc1bbeb | 283 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
284 | return 1; |
285 | ||
286 | return 0; | |
287 | } | |
288 | ||
906c8d75 JA |
289 | /* |
290 | * When job exits, we can cancel the in-flight IO if we are using async | |
291 | * io. Attempt to do so. | |
292 | */ | |
ebac4655 JA |
293 | static void cleanup_pending_aio(struct thread_data *td) |
294 | { | |
01743ee1 | 295 | struct flist_head *entry, *n; |
ebac4655 JA |
296 | struct io_u *io_u; |
297 | int r; | |
298 | ||
299 | /* | |
300 | * get immediately available events, if any | |
301 | */ | |
581e7141 | 302 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
303 | if (r < 0) |
304 | return; | |
ebac4655 JA |
305 | |
306 | /* | |
307 | * now cancel remaining active events | |
308 | */ | |
2866c82d | 309 | if (td->io_ops->cancel) { |
01743ee1 JA |
310 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
311 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 312 | |
0c6e7517 JA |
313 | /* |
314 | * if the io_u isn't in flight, then that generally | |
315 | * means someone leaked an io_u. complain but fix | |
316 | * it up, so we don't stall here. | |
317 | */ | |
318 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
319 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 320 | put_io_u(td, io_u); |
0c6e7517 JA |
321 | } else { |
322 | r = td->io_ops->cancel(td, io_u); | |
323 | if (!r) | |
324 | put_io_u(td, io_u); | |
325 | } | |
ebac4655 JA |
326 | } |
327 | } | |
328 | ||
97601024 | 329 | if (td->cur_depth) |
581e7141 | 330 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
331 | } |
332 | ||
858a3d47 JA |
333 | /* |
334 | * Helper to handle the final sync of a file. Works just like the normal | |
335 | * io path, just does everything sync. | |
336 | */ | |
337 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
338 | { | |
339 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
340 | int ret; |
341 | ||
342 | if (!io_u) | |
343 | return 1; | |
344 | ||
345 | io_u->ddir = DDIR_SYNC; | |
346 | io_u->file = f; | |
347 | ||
348 | if (td_io_prep(td, io_u)) { | |
349 | put_io_u(td, io_u); | |
350 | return 1; | |
351 | } | |
352 | ||
755200a3 | 353 | requeue: |
858a3d47 | 354 | ret = td_io_queue(td, io_u); |
36167d82 | 355 | if (ret < 0) { |
e1161c32 | 356 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 357 | put_io_u(td, io_u); |
858a3d47 | 358 | return 1; |
36167d82 | 359 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 360 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 361 | return 1; |
36167d82 JA |
362 | } else if (ret == FIO_Q_COMPLETED) { |
363 | if (io_u->error) { | |
e1161c32 | 364 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
365 | return 1; |
366 | } | |
858a3d47 | 367 | |
581e7141 | 368 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 369 | return 1; |
755200a3 JA |
370 | } else if (ret == FIO_Q_BUSY) { |
371 | if (td_io_commit(td)) | |
372 | return 1; | |
373 | goto requeue; | |
858a3d47 JA |
374 | } |
375 | ||
376 | return 0; | |
377 | } | |
378 | ||
687c6a23 JA |
379 | static inline void __update_tv_cache(struct thread_data *td) |
380 | { | |
381 | fio_gettime(&td->tv_cache, NULL); | |
382 | } | |
383 | ||
993bf48b JA |
384 | static inline void update_tv_cache(struct thread_data *td) |
385 | { | |
386 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
687c6a23 | 387 | __update_tv_cache(td); |
993bf48b JA |
388 | } |
389 | ||
f2bba182 RR |
390 | static int break_on_this_error(struct thread_data *td, int *retptr) |
391 | { | |
392 | int ret = *retptr; | |
393 | ||
394 | if (ret < 0 || td->error) { | |
395 | int err; | |
396 | ||
1ec99eea JA |
397 | if (!td->o.continue_on_error) |
398 | return 1; | |
f2bba182 RR |
399 | |
400 | if (ret < 0) | |
401 | err = -ret; | |
402 | else | |
403 | err = td->error; | |
404 | ||
f2bba182 RR |
405 | if (td_non_fatal_error(err)) { |
406 | /* | |
407 | * Continue with the I/Os in case of | |
408 | * a non fatal error. | |
409 | */ | |
8a4f8a24 | 410 | update_error_count(td, err); |
f2bba182 RR |
411 | td_clear_error(td); |
412 | *retptr = 0; | |
413 | return 0; | |
8a4f8a24 RR |
414 | } else if (td->o.fill_device && err == ENOSPC) { |
415 | /* | |
416 | * We expect to hit this error if | |
417 | * fill_device option is set. | |
418 | */ | |
419 | td_clear_error(td); | |
420 | td->terminate = 1; | |
421 | return 1; | |
f2bba182 RR |
422 | } else { |
423 | /* | |
424 | * Stop the I/O in case of a fatal | |
425 | * error. | |
426 | */ | |
8a4f8a24 | 427 | update_error_count(td, err); |
f2bba182 RR |
428 | return 1; |
429 | } | |
430 | } | |
431 | ||
432 | return 0; | |
433 | } | |
434 | ||
906c8d75 | 435 | /* |
34403fb1 | 436 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
437 | * reads the blocks back in, and checks the crc/md5 of the data. |
438 | */ | |
1e97cce9 | 439 | static void do_verify(struct thread_data *td) |
ebac4655 | 440 | { |
53cdc686 | 441 | struct fio_file *f; |
36167d82 | 442 | struct io_u *io_u; |
af52b345 JA |
443 | int ret, min_events; |
444 | unsigned int i; | |
e5b401d4 | 445 | |
008618a8 JA |
446 | dprint(FD_VERIFY, "starting loop\n"); |
447 | ||
e5b401d4 JA |
448 | /* |
449 | * sync io first and invalidate cache, to make sure we really | |
450 | * read from disk. | |
451 | */ | |
452 | for_each_file(td, f, i) { | |
d6aed795 | 453 | if (!fio_file_open(f)) |
3d7b485f | 454 | continue; |
b2fdda43 JA |
455 | if (fio_io_sync(td, f)) |
456 | break; | |
457 | if (file_invalidate_cache(td, f)) | |
458 | break; | |
e5b401d4 | 459 | } |
ebac4655 | 460 | |
b2fdda43 JA |
461 | if (td->error) |
462 | return; | |
463 | ||
ebac4655 JA |
464 | td_set_runstate(td, TD_VERIFYING); |
465 | ||
36167d82 JA |
466 | io_u = NULL; |
467 | while (!td->terminate) { | |
4950421a | 468 | int ret2, full; |
5451792e | 469 | |
993bf48b JA |
470 | update_tv_cache(td); |
471 | ||
472 | if (runtime_exceeded(td, &td->tv_cache)) { | |
687c6a23 JA |
473 | __update_tv_cache(td); |
474 | if (runtime_exceeded(td, &td->tv_cache)) { | |
475 | td->terminate = 1; | |
476 | break; | |
477 | } | |
069c2918 | 478 | } |
02bcaa8c | 479 | |
dcb4830d RR |
480 | io_u = __get_io_u(td); |
481 | if (!io_u) | |
482 | break; | |
483 | ||
069c2918 JA |
484 | if (get_next_verify(td, io_u)) { |
485 | put_io_u(td, io_u); | |
ebac4655 | 486 | break; |
069c2918 | 487 | } |
ebac4655 | 488 | |
069c2918 JA |
489 | if (td_io_prep(td, io_u)) { |
490 | put_io_u(td, io_u); | |
53cdc686 | 491 | break; |
069c2918 | 492 | } |
d7762cf8 | 493 | |
e8462bd8 JA |
494 | if (td->o.verify_async) |
495 | io_u->end_io = verify_io_u_async; | |
496 | else | |
497 | io_u->end_io = verify_io_u; | |
53cdc686 | 498 | |
11786802 | 499 | ret = td_io_queue(td, io_u); |
36167d82 JA |
500 | switch (ret) { |
501 | case FIO_Q_COMPLETED: | |
f2bba182 | 502 | if (io_u->error) { |
22819ec2 | 503 | ret = -io_u->error; |
f2bba182 RR |
504 | clear_io_u(td, io_u); |
505 | } else if (io_u->resid) { | |
36167d82 | 506 | int bytes = io_u->xfer_buflen - io_u->resid; |
ebac4655 | 507 | |
9e9d164e JA |
508 | /* |
509 | * zero read, fail | |
510 | */ | |
511 | if (!bytes) { | |
41d9f0fd | 512 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
513 | put_io_u(td, io_u); |
514 | break; | |
515 | } | |
8400d9b2 | 516 | |
36167d82 JA |
517 | io_u->xfer_buflen = io_u->resid; |
518 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
519 | io_u->offset += bytes; |
520 | ||
ff58fced JA |
521 | if (ddir_rw(io_u->ddir)) |
522 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 523 | |
2b13e716 | 524 | f = io_u->file; |
d460eb31 | 525 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
526 | goto sync_done; |
527 | ||
11786802 JA |
528 | requeue_io_u(td, &io_u); |
529 | } else { | |
8400d9b2 | 530 | sync_done: |
581e7141 | 531 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
532 | if (ret < 0) |
533 | break; | |
36167d82 | 534 | } |
36167d82 JA |
535 | continue; |
536 | case FIO_Q_QUEUED: | |
537 | break; | |
755200a3 JA |
538 | case FIO_Q_BUSY: |
539 | requeue_io_u(td, &io_u); | |
5451792e JA |
540 | ret2 = td_io_commit(td); |
541 | if (ret2 < 0) | |
542 | ret = ret2; | |
755200a3 | 543 | break; |
36167d82 JA |
544 | default: |
545 | assert(ret < 0); | |
e1161c32 | 546 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
547 | break; |
548 | } | |
549 | ||
f2bba182 | 550 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
551 | break; |
552 | ||
ebac4655 | 553 | /* |
3af6ef39 | 554 | * if we can queue more, do so. but check if there are |
ccc02050 JA |
555 | * completed io_u's first. Note that we can get BUSY even |
556 | * without IO queued, if the system is resource starved. | |
ebac4655 | 557 | */ |
ccc02050 | 558 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 559 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
560 | min_events = min(td->o.iodepth_batch_complete, |
561 | td->cur_depth); | |
1f84f1d9 | 562 | if (full && !min_events && td->o.iodepth_batch_complete != 0) |
838bc709 | 563 | min_events = 1; |
e916b390 | 564 | |
4950421a JA |
565 | do { |
566 | /* | |
567 | * Reap required number of io units, if any, | |
568 | * and do the verification on them through | |
569 | * the callback handler | |
570 | */ | |
581e7141 | 571 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
572 | ret = -1; |
573 | break; | |
574 | } | |
575 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
576 | } | |
577 | if (ret < 0) | |
ebac4655 | 578 | break; |
36167d82 | 579 | } |
ebac4655 | 580 | |
c01c0395 JA |
581 | if (!td->error) { |
582 | min_events = td->cur_depth; | |
583 | ||
584 | if (min_events) | |
581e7141 | 585 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 586 | } else |
ebac4655 JA |
587 | cleanup_pending_aio(td); |
588 | ||
589 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
590 | |
591 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
592 | } |
593 | ||
32cd46a0 | 594 | /* |
906c8d75 | 595 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
596 | * and reaps them, checking for rate and errors along the way. |
597 | */ | |
ebac4655 JA |
598 | static void do_io(struct thread_data *td) |
599 | { | |
af52b345 JA |
600 | unsigned int i; |
601 | int ret = 0; | |
ebac4655 | 602 | |
b29ee5b3 JA |
603 | if (in_ramp_time(td)) |
604 | td_set_runstate(td, TD_RAMP); | |
605 | else | |
606 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 607 | |
457bf399 | 608 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
a917a8b3 | 609 | (!flist_empty(&td->trim_list)) || |
457bf399 | 610 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { |
97601024 | 611 | struct timeval comp_time; |
581e7141 | 612 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 613 | int min_evts = 0; |
ebac4655 | 614 | struct io_u *io_u; |
4950421a | 615 | int ret2, full; |
ebac4655 JA |
616 | |
617 | if (td->terminate) | |
618 | break; | |
619 | ||
993bf48b | 620 | update_tv_cache(td); |
97601024 | 621 | |
993bf48b | 622 | if (runtime_exceeded(td, &td->tv_cache)) { |
687c6a23 JA |
623 | __update_tv_cache(td); |
624 | if (runtime_exceeded(td, &td->tv_cache)) { | |
625 | td->terminate = 1; | |
626 | break; | |
627 | } | |
97601024 | 628 | } |
36167d82 | 629 | |
dcb4830d RR |
630 | io_u = get_io_u(td); |
631 | if (!io_u) | |
632 | break; | |
633 | ||
b67740d3 JA |
634 | /* |
635 | * Add verification end_io handler, if asked to verify | |
636 | * a previously written file. | |
637 | */ | |
715d2b3e JA |
638 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
639 | !td_rw(td)) { | |
e8462bd8 JA |
640 | if (td->o.verify_async) |
641 | io_u->end_io = verify_io_u_async; | |
642 | else | |
643 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 644 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
645 | } else if (in_ramp_time(td)) |
646 | td_set_runstate(td, TD_RAMP); | |
647 | else | |
d8fab1b6 | 648 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 649 | |
11786802 | 650 | ret = td_io_queue(td, io_u); |
36167d82 JA |
651 | switch (ret) { |
652 | case FIO_Q_COMPLETED: | |
f2bba182 | 653 | if (io_u->error) { |
5451792e | 654 | ret = -io_u->error; |
f2bba182 RR |
655 | clear_io_u(td, io_u); |
656 | } else if (io_u->resid) { | |
36167d82 | 657 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 658 | struct fio_file *f = io_u->file; |
36167d82 | 659 | |
9e9d164e JA |
660 | /* |
661 | * zero read, fail | |
662 | */ | |
663 | if (!bytes) { | |
41d9f0fd | 664 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
665 | put_io_u(td, io_u); |
666 | break; | |
667 | } | |
668 | ||
cec6b55d | 669 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 670 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
671 | io_u->offset += bytes; |
672 | ||
ff58fced JA |
673 | if (ddir_rw(io_u->ddir)) |
674 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 675 | |
d460eb31 | 676 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
677 | goto sync_done; |
678 | ||
11786802 JA |
679 | requeue_io_u(td, &io_u); |
680 | } else { | |
5a7c5680 | 681 | sync_done: |
581e7141 JA |
682 | if (__should_check_rate(td, 0) || |
683 | __should_check_rate(td, 1)) | |
993bf48b JA |
684 | fio_gettime(&comp_time, NULL); |
685 | ||
581e7141 JA |
686 | ret = io_u_sync_complete(td, io_u, bytes_done); |
687 | if (ret < 0) | |
688 | break; | |
cec6b55d | 689 | } |
36167d82 JA |
690 | break; |
691 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
692 | /* |
693 | * if the engine doesn't have a commit hook, | |
694 | * the io_u is really queued. if it does have such | |
695 | * a hook, it has to call io_u_queued() itself. | |
696 | */ | |
697 | if (td->io_ops->commit == NULL) | |
698 | io_u_queued(td, io_u); | |
36167d82 | 699 | break; |
755200a3 JA |
700 | case FIO_Q_BUSY: |
701 | requeue_io_u(td, &io_u); | |
5451792e JA |
702 | ret2 = td_io_commit(td); |
703 | if (ret2 < 0) | |
704 | ret = ret2; | |
755200a3 | 705 | break; |
36167d82 JA |
706 | default: |
707 | assert(ret < 0); | |
708 | put_io_u(td, io_u); | |
709 | break; | |
ebac4655 JA |
710 | } |
711 | ||
f2bba182 | 712 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
713 | break; |
714 | ||
97601024 | 715 | /* |
ccc02050 JA |
716 | * See if we need to complete some commands. Note that we |
717 | * can get BUSY even without IO queued, if the system is | |
718 | * resource starved. | |
97601024 | 719 | */ |
ccc02050 | 720 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 721 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
722 | min_evts = min(td->o.iodepth_batch_complete, |
723 | td->cur_depth); | |
1f84f1d9 | 724 | if (full && !min_evts && td->o.iodepth_batch_complete != 0) |
36167d82 | 725 | min_evts = 1; |
4950421a | 726 | |
581e7141 JA |
727 | if (__should_check_rate(td, 0) || |
728 | __should_check_rate(td, 1)) | |
993bf48b | 729 | fio_gettime(&comp_time, NULL); |
4950421a JA |
730 | |
731 | do { | |
581e7141 JA |
732 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
733 | if (ret < 0) | |
4950421a JA |
734 | break; |
735 | ||
4950421a | 736 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
737 | } |
738 | ||
4950421a JA |
739 | if (ret < 0) |
740 | break; | |
581e7141 | 741 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
742 | continue; |
743 | ||
581e7141 JA |
744 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
745 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
746 | if (exitall_on_terminate) |
747 | terminate_threads(td->groupid); | |
748 | td_verror(td, EIO, "check_min_rate"); | |
749 | break; | |
750 | } | |
ebac4655 JA |
751 | } |
752 | ||
2dc1bbeb | 753 | if (td->o.thinktime) { |
9c1f7434 JA |
754 | unsigned long long b; |
755 | ||
756 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 757 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
758 | int left; |
759 | ||
2dc1bbeb | 760 | if (td->o.thinktime_spin) |
6dd6f2cd | 761 | usec_spin(td->o.thinktime_spin); |
48097d5c | 762 | |
2dc1bbeb | 763 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
764 | if (left) |
765 | usec_sleep(td, left); | |
766 | } | |
9c1f7434 | 767 | } |
ebac4655 JA |
768 | } |
769 | ||
d708ac5e JA |
770 | if (td->trim_entries) |
771 | printf("trim entries %ld\n", td->trim_entries); | |
a917a8b3 | 772 | |
aa31f1f1 SL |
773 | if (td->o.fill_device && td->error == ENOSPC) { |
774 | td->error = 0; | |
775 | td->terminate = 1; | |
776 | } | |
4d2413c6 | 777 | if (!td->error) { |
3d7c391d JA |
778 | struct fio_file *f; |
779 | ||
c01c0395 | 780 | i = td->cur_depth; |
d06093e7 | 781 | if (i) { |
581e7141 | 782 | ret = io_u_queued_complete(td, i, NULL); |
d06093e7 SL |
783 | if (td->o.fill_device && td->error == ENOSPC) |
784 | td->error = 0; | |
785 | } | |
ebac4655 | 786 | |
2dc1bbeb | 787 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 788 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
789 | |
790 | for_each_file(td, f, i) { | |
d6aed795 | 791 | if (!fio_file_open(f)) |
3d7b485f | 792 | continue; |
858a3d47 | 793 | fio_io_sync(td, f); |
3d7b485f | 794 | } |
84585003 | 795 | } |
c01c0395 JA |
796 | } else |
797 | cleanup_pending_aio(td); | |
2ba1c290 JA |
798 | |
799 | /* | |
800 | * stop job if we failed doing any IO | |
801 | */ | |
802 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
803 | td->done = 1; | |
ebac4655 JA |
804 | } |
805 | ||
ebac4655 JA |
806 | static void cleanup_io_u(struct thread_data *td) |
807 | { | |
01743ee1 | 808 | struct flist_head *entry, *n; |
ebac4655 JA |
809 | struct io_u *io_u; |
810 | ||
01743ee1 JA |
811 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
812 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 813 | |
01743ee1 | 814 | flist_del(&io_u->list); |
91aea6ec | 815 | fio_memfree(io_u, sizeof(*io_u)); |
ebac4655 JA |
816 | } |
817 | ||
2f9ade3c | 818 | free_io_mem(td); |
ebac4655 JA |
819 | } |
820 | ||
821 | static int init_io_u(struct thread_data *td) | |
822 | { | |
823 | struct io_u *io_u; | |
a00735e6 | 824 | unsigned int max_bs; |
eb7ccf38 | 825 | int cl_align, i, max_units; |
ebac4655 JA |
826 | char *p; |
827 | ||
1e3c09af | 828 | max_units = td->o.iodepth; |
2dc1bbeb | 829 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
830 | td->orig_buffer_size = (unsigned long long) max_bs |
831 | * (unsigned long long) max_units; | |
832 | ||
833 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
834 | unsigned long bs; | |
74b025b0 | 835 | |
5ec10eaa JA |
836 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
837 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
838 | } | |
ebac4655 | 839 | |
c3990645 JA |
840 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
841 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
842 | return 1; | |
843 | } | |
844 | ||
2f9ade3c JA |
845 | if (allocate_io_mem(td)) |
846 | return 1; | |
ebac4655 | 847 | |
1c04bf78 JA |
848 | if (td->o.odirect || td->o.mem_align || |
849 | (td->io_ops->flags & FIO_RAWIO)) | |
d529ee19 | 850 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
851 | else |
852 | p = td->orig_buffer; | |
853 | ||
eb7ccf38 JA |
854 | cl_align = os_cache_line_size(); |
855 | ||
ebac4655 | 856 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
857 | void *ptr; |
858 | ||
db1cd90b JA |
859 | if (td->terminate) |
860 | return 1; | |
eb7ccf38 | 861 | |
91aea6ec JA |
862 | ptr = fio_memalign(cl_align, sizeof(*io_u)); |
863 | if (!ptr) { | |
864 | log_err("fio: unable to allocate aligned memory\n"); | |
eb7ccf38 JA |
865 | break; |
866 | } | |
867 | ||
868 | io_u = ptr; | |
ebac4655 | 869 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 870 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 871 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 872 | |
b4c5e1ac JA |
873 | if (!(td->io_ops->flags & FIO_NOIO)) { |
874 | io_u->buf = p + max_bs * i; | |
d529ee19 | 875 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 876 | |
3545a109 | 877 | if (td_write(td)) |
5973cafb | 878 | io_u_fill_buffer(td, io_u, max_bs); |
3545a109 | 879 | if (td_write(td) && td->o.verify_pattern_bytes) { |
cbe8d756 RR |
880 | /* |
881 | * Fill the buffer with the pattern if we are | |
882 | * going to be doing writes. | |
883 | */ | |
7d9fb455 | 884 | fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
cbe8d756 | 885 | } |
b4c5e1ac | 886 | } |
6b9cea23 | 887 | |
b1ff3403 | 888 | io_u->index = i; |
0c6e7517 | 889 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 890 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
891 | } |
892 | ||
893 | return 0; | |
894 | } | |
895 | ||
da86774e JA |
896 | static int switch_ioscheduler(struct thread_data *td) |
897 | { | |
898 | char tmp[256], tmp2[128]; | |
899 | FILE *f; | |
900 | int ret; | |
901 | ||
ba0fbe10 | 902 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
903 | return 0; |
904 | ||
da86774e JA |
905 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
906 | ||
907 | f = fopen(tmp, "r+"); | |
908 | if (!f) { | |
cbf5c1d7 | 909 | if (errno == ENOENT) { |
5ec10eaa JA |
910 | log_err("fio: os or kernel doesn't support IO scheduler" |
911 | " switching\n"); | |
cbf5c1d7 JA |
912 | return 0; |
913 | } | |
914 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
915 | return 1; |
916 | } | |
917 | ||
918 | /* | |
919 | * Set io scheduler. | |
920 | */ | |
2dc1bbeb | 921 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 922 | if (ferror(f) || ret != 1) { |
e1161c32 | 923 | td_verror(td, errno, "fwrite"); |
da86774e JA |
924 | fclose(f); |
925 | return 1; | |
926 | } | |
927 | ||
928 | rewind(f); | |
929 | ||
930 | /* | |
931 | * Read back and check that the selected scheduler is now the default. | |
932 | */ | |
933 | ret = fread(tmp, 1, sizeof(tmp), f); | |
934 | if (ferror(f) || ret < 0) { | |
e1161c32 | 935 | td_verror(td, errno, "fread"); |
da86774e JA |
936 | fclose(f); |
937 | return 1; | |
938 | } | |
939 | ||
2dc1bbeb | 940 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 941 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 942 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 943 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
944 | fclose(f); |
945 | return 1; | |
946 | } | |
947 | ||
948 | fclose(f); | |
949 | return 0; | |
950 | } | |
951 | ||
492158cf JA |
952 | static int keep_running(struct thread_data *td) |
953 | { | |
954 | unsigned long long io_done; | |
955 | ||
20e354ef JA |
956 | if (td->done) |
957 | return 0; | |
492158cf JA |
958 | if (td->o.time_based) |
959 | return 1; | |
960 | if (td->o.loops) { | |
961 | td->o.loops--; | |
962 | return 1; | |
963 | } | |
964 | ||
5ec10eaa JA |
965 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
966 | + td->io_skip_bytes; | |
492158cf JA |
967 | if (io_done < td->o.size) |
968 | return 1; | |
969 | ||
970 | return 0; | |
971 | } | |
972 | ||
b29ee5b3 | 973 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 974 | { |
756867bd | 975 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 976 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 977 | td->zone_bytes = 0; |
581e7141 JA |
978 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
979 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 980 | |
c1324df1 JA |
981 | td->last_was_sync = 0; |
982 | ||
2ba1c290 JA |
983 | /* |
984 | * reset file done count if we are to start over | |
985 | */ | |
986 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 987 | td->nr_done_files = 0; |
b29ee5b3 JA |
988 | } |
989 | ||
990 | void reset_all_stats(struct thread_data *td) | |
991 | { | |
992 | struct timeval tv; | |
993 | int i; | |
994 | ||
995 | reset_io_counters(td); | |
996 | ||
997 | for (i = 0; i < 2; i++) { | |
998 | td->io_bytes[i] = 0; | |
999 | td->io_blocks[i] = 0; | |
1000 | td->io_issues[i] = 0; | |
1001 | td->ts.total_io_u[i] = 0; | |
1002 | } | |
9b836561 | 1003 | |
b29ee5b3 | 1004 | fio_gettime(&tv, NULL); |
cda99fa0 SSP |
1005 | td->ts.runtime[0] = 0; |
1006 | td->ts.runtime[1] = 0; | |
b29ee5b3 JA |
1007 | memcpy(&td->epoch, &tv, sizeof(tv)); |
1008 | memcpy(&td->start, &tv, sizeof(tv)); | |
1009 | } | |
1010 | ||
df9c26b1 | 1011 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
1012 | { |
1013 | struct fio_file *f; | |
1014 | unsigned int i; | |
b29ee5b3 JA |
1015 | |
1016 | reset_io_counters(td); | |
281f9b63 | 1017 | |
24ffd2c2 | 1018 | close_files(td); |
df9c26b1 | 1019 | for_each_file(td, f, i) |
d6aed795 | 1020 | fio_file_clear_done(f); |
f0153948 JA |
1021 | |
1022 | /* | |
1023 | * Set the same seed to get repeatable runs | |
1024 | */ | |
1025 | td_fill_rand_seeds(td); | |
ebac4655 JA |
1026 | } |
1027 | ||
398e8c4d JA |
1028 | static int exec_string(const char *string) |
1029 | { | |
1030 | int ret, newlen = strlen(string) + 1 + 8; | |
1031 | char *str; | |
1032 | ||
1033 | str = malloc(newlen); | |
1034 | sprintf(str, "sh -c %s", string); | |
1035 | ||
1036 | ret = system(str); | |
1037 | if (ret == -1) | |
1038 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1039 | ||
1040 | free(str); | |
1041 | return ret; | |
1042 | } | |
1043 | ||
906c8d75 JA |
1044 | /* |
1045 | * Entry point for the thread based jobs. The process based jobs end up | |
1046 | * here as well, after a little setup. | |
1047 | */ | |
ebac4655 JA |
1048 | static void *thread_main(void *data) |
1049 | { | |
cda99fa0 | 1050 | unsigned long long elapsed; |
ebac4655 | 1051 | struct thread_data *td = data; |
e8462bd8 | 1052 | pthread_condattr_t attr; |
a978ba68 | 1053 | int clear_state; |
ebac4655 | 1054 | |
47f767c1 | 1055 | if (!td->o.use_thread) { |
ebac4655 | 1056 | setsid(); |
47f767c1 JA |
1057 | td->pid = getpid(); |
1058 | } else | |
1059 | td->pid = gettid(); | |
ebac4655 | 1060 | |
5921e80c | 1061 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1062 | |
01743ee1 JA |
1063 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1064 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1065 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1066 | INIT_FLIST_HEAD(&td->io_log_list); | |
1067 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 | 1068 | INIT_FLIST_HEAD(&td->verify_list); |
0d29de83 | 1069 | INIT_FLIST_HEAD(&td->trim_list); |
e8462bd8 | 1070 | pthread_mutex_init(&td->io_u_lock, NULL); |
bb5d7d0b | 1071 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1072 | |
e8462bd8 JA |
1073 | pthread_condattr_init(&attr); |
1074 | pthread_cond_init(&td->verify_cond, &attr); | |
1075 | pthread_cond_init(&td->free_cond, &attr); | |
1076 | ||
db1cd90b | 1077 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1078 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1079 | fio_mutex_up(startup_mutex); |
29adda3c | 1080 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1081 | fio_mutex_down(td->mutex); |
29adda3c | 1082 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1083 | |
1084 | /* | |
cdd18ad8 | 1085 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1086 | * eating a file descriptor |
1087 | */ | |
cdd18ad8 | 1088 | fio_mutex_remove(td->mutex); |
db1cd90b | 1089 | |
dfc6e751 SH |
1090 | /* |
1091 | * A new gid requires privilege, so we need to do this before setting | |
1092 | * the uid. | |
1093 | */ | |
e0b0d892 JA |
1094 | if (td->o.gid != -1U && setgid(td->o.gid)) { |
1095 | td_verror(td, errno, "setgid"); | |
1096 | goto err; | |
1097 | } | |
dfc6e751 SH |
1098 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1099 | td_verror(td, errno, "setuid"); | |
1100 | goto err; | |
1101 | } | |
e0b0d892 | 1102 | |
be18a94f JA |
1103 | /* |
1104 | * If we have a gettimeofday() thread, make sure we exclude that | |
1105 | * thread from this job | |
1106 | */ | |
1107 | if (td->o.gtod_cpu) | |
1108 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); | |
1109 | ||
1110 | /* | |
1111 | * Set affinity first, in case it has an impact on the memory | |
1112 | * allocations. | |
1113 | */ | |
1114 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
1115 | td_verror(td, errno, "cpu_set_affinity"); | |
1116 | goto err; | |
1117 | } | |
1118 | ||
d84f8d49 JA |
1119 | /* |
1120 | * May alter parameters that init_io_u() will use, so we need to | |
1121 | * do this first. | |
1122 | */ | |
1123 | if (init_iolog(td)) | |
1124 | goto err; | |
1125 | ||
ebac4655 | 1126 | if (init_io_u(td)) |
db1cd90b | 1127 | goto err; |
ebac4655 | 1128 | |
e8462bd8 JA |
1129 | if (td->o.verify_async && verify_async_init(td)) |
1130 | goto err; | |
1131 | ||
ac684785 | 1132 | if (td->ioprio_set) { |
ebac4655 | 1133 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1134 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1135 | goto err; |
ebac4655 JA |
1136 | } |
1137 | } | |
1138 | ||
6adb38a1 | 1139 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1140 | goto err; |
1141 | ||
2dc1bbeb | 1142 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1143 | td_verror(td, errno, "nice"); |
db1cd90b | 1144 | goto err; |
b6f4d880 JA |
1145 | } |
1146 | ||
2dc1bbeb | 1147 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1148 | goto err; |
37f56873 | 1149 | |
2dc1bbeb | 1150 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1151 | goto err; |
1152 | ||
7d6c5283 JA |
1153 | if (td_io_init(td)) |
1154 | goto err; | |
1155 | ||
68727076 JA |
1156 | if (init_random_map(td)) |
1157 | goto err; | |
1158 | ||
2dc1bbeb | 1159 | if (td->o.exec_prerun) { |
398e8c4d | 1160 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1161 | goto err; |
1162 | } | |
4e0ba8af | 1163 | |
afad68f7 ZY |
1164 | if (td->o.pre_read) { |
1165 | if (pre_read_files(td) < 0) | |
1166 | goto err; | |
1167 | } | |
1168 | ||
69008999 | 1169 | fio_gettime(&td->epoch, NULL); |
756867bd | 1170 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 | 1171 | |
a978ba68 | 1172 | clear_state = 0; |
492158cf | 1173 | while (keep_running(td)) { |
02bcaa8c | 1174 | fio_gettime(&td->start, NULL); |
c5d3d790 JA |
1175 | memcpy(&td->ts.stat_sample_time[0], &td->start, |
1176 | sizeof(td->start)); | |
1177 | memcpy(&td->ts.stat_sample_time[1], &td->start, | |
1178 | sizeof(td->start)); | |
993bf48b | 1179 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1180 | |
cf2dd8d6 | 1181 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
5ec10eaa JA |
1182 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
1183 | sizeof(td->lastrate)); | |
ebac4655 | 1184 | |
df9c26b1 JA |
1185 | if (clear_state) |
1186 | clear_io_state(td); | |
a978ba68 | 1187 | |
ebac4655 JA |
1188 | prune_io_piece_log(td); |
1189 | ||
ba0fbe10 | 1190 | do_io(td); |
ebac4655 | 1191 | |
a978ba68 JA |
1192 | clear_state = 1; |
1193 | ||
38d77cae | 1194 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1195 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1196 | td->ts.runtime[DDIR_READ] += elapsed; |
38d77cae JA |
1197 | } |
1198 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1199 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1200 | td->ts.runtime[DDIR_WRITE] += elapsed; |
38d77cae | 1201 | } |
5ec10eaa | 1202 | |
ebac4655 JA |
1203 | if (td->error || td->terminate) |
1204 | break; | |
1205 | ||
e84c73a8 SL |
1206 | if (!td->o.do_verify || |
1207 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1208 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1209 | continue; |
1210 | ||
df9c26b1 | 1211 | clear_io_state(td); |
a978ba68 | 1212 | |
02bcaa8c | 1213 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1214 | |
1215 | do_verify(td); | |
1216 | ||
cda99fa0 | 1217 | td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1218 | |
1219 | if (td->error || td->terminate) | |
1220 | break; | |
1221 | } | |
1222 | ||
36dff966 | 1223 | update_rusage_stat(td); |
cda99fa0 SSP |
1224 | td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000; |
1225 | td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000; | |
756867bd JA |
1226 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1227 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1228 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1229 | ||
d2bbb8d9 | 1230 | fio_mutex_down(writeout_mutex); |
e3cedca7 | 1231 | if (td->ts.bw_log) { |
f484470b JA |
1232 | if (td->o.bw_log_file) { |
1233 | finish_log_named(td, td->ts.bw_log, | |
1234 | td->o.bw_log_file, "bw"); | |
1235 | } else | |
e3cedca7 JA |
1236 | finish_log(td, td->ts.bw_log, "bw"); |
1237 | } | |
02af0988 JA |
1238 | if (td->ts.lat_log) { |
1239 | if (td->o.lat_log_file) { | |
1240 | finish_log_named(td, td->ts.lat_log, | |
1241 | td->o.lat_log_file, "lat"); | |
1242 | } else | |
1243 | finish_log(td, td->ts.lat_log, "lat"); | |
1244 | } | |
e3cedca7 | 1245 | if (td->ts.slat_log) { |
f484470b JA |
1246 | if (td->o.lat_log_file) { |
1247 | finish_log_named(td, td->ts.slat_log, | |
1ca95738 | 1248 | td->o.lat_log_file, "slat"); |
f484470b | 1249 | } else |
e3cedca7 JA |
1250 | finish_log(td, td->ts.slat_log, "slat"); |
1251 | } | |
1252 | if (td->ts.clat_log) { | |
f484470b JA |
1253 | if (td->o.lat_log_file) { |
1254 | finish_log_named(td, td->ts.clat_log, | |
1255 | td->o.lat_log_file, "clat"); | |
1256 | } else | |
e3cedca7 JA |
1257 | finish_log(td, td->ts.clat_log, "clat"); |
1258 | } | |
d2bbb8d9 | 1259 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1260 | if (td->o.exec_postrun) |
1261 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1262 | |
1263 | if (exitall_on_terminate) | |
390c40e2 | 1264 | terminate_threads(td->groupid); |
ebac4655 JA |
1265 | |
1266 | err: | |
5bf13a5a | 1267 | if (td->error) |
4ceb30d4 | 1268 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1269 | td->verror); |
b4707ed7 JA |
1270 | |
1271 | if (td->o.verify_async) | |
1272 | verify_async_exit(td); | |
1273 | ||
24ffd2c2 | 1274 | close_and_free_files(td); |
2866c82d | 1275 | close_ioengine(td); |
ebac4655 | 1276 | cleanup_io_u(td); |
6adb38a1 | 1277 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1278 | |
d2ce18b5 JA |
1279 | if (td->o.cpumask_set) { |
1280 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1281 | ||
1282 | td_verror(td, ret, "fio_cpuset_exit"); | |
1283 | } | |
4e78e405 | 1284 | |
f29b25a3 JA |
1285 | /* |
1286 | * do this very late, it will log file closing as well | |
1287 | */ | |
1288 | if (td->o.write_iolog_file) | |
1289 | write_iolog_close(td); | |
1290 | ||
d23bb327 | 1291 | options_mem_free(td); |
ebac4655 | 1292 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1293 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1294 | } |
1295 | ||
906c8d75 JA |
1296 | /* |
1297 | * We cannot pass the td data into a forked process, so attach the td and | |
1298 | * pass it to the thread worker. | |
1299 | */ | |
a6418147 | 1300 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1301 | { |
1302 | struct thread_data *td; | |
a6418147 | 1303 | void *data, *ret; |
ebac4655 | 1304 | |
65258961 | 1305 | #ifndef __hpux |
ebac4655 JA |
1306 | data = shmat(shmid, NULL, 0); |
1307 | if (data == (void *) -1) { | |
a6418147 JA |
1308 | int __err = errno; |
1309 | ||
ebac4655 | 1310 | perror("shmat"); |
a6418147 | 1311 | return __err; |
ebac4655 | 1312 | } |
65258961 JA |
1313 | #else |
1314 | /* | |
1315 | * HP-UX inherits shm mappings? | |
1316 | */ | |
1317 | data = threads; | |
1318 | #endif | |
ebac4655 JA |
1319 | |
1320 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1321 | ret = thread_main(td); |
ebac4655 | 1322 | shmdt(data); |
43d76807 | 1323 | return (int) (unsigned long) ret; |
ebac4655 JA |
1324 | } |
1325 | ||
906c8d75 JA |
1326 | /* |
1327 | * Run over the job map and reap the threads that have exited, if any. | |
1328 | */ | |
ebac4655 JA |
1329 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1330 | { | |
34572e28 | 1331 | struct thread_data *td; |
1f809d15 | 1332 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1333 | |
1334 | /* | |
1335 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1336 | */ | |
1f809d15 | 1337 | realthreads = pending = cputhreads = 0; |
34572e28 | 1338 | for_each_td(td, i) { |
3707f45b | 1339 | int flags = 0; |
a2f77c9f | 1340 | |
84585003 JA |
1341 | /* |
1342 | * ->io_ops is NULL for a thread that has closed its | |
1343 | * io engine | |
1344 | */ | |
ba0fbe10 | 1345 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1346 | cputhreads++; |
1f809d15 JA |
1347 | else |
1348 | realthreads++; | |
b990b5c0 | 1349 | |
a1dedc1f AC |
1350 | if (!td->pid) { |
1351 | pending++; | |
1352 | continue; | |
1353 | } | |
1354 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1355 | continue; |
2dc1bbeb | 1356 | if (td->o.use_thread) { |
3707f45b JA |
1357 | if (td->runstate == TD_EXITED) { |
1358 | td_set_runstate(td, TD_REAPED); | |
1359 | goto reaped; | |
1360 | } | |
1361 | continue; | |
1362 | } | |
a2f77c9f JA |
1363 | |
1364 | flags = WNOHANG; | |
1365 | if (td->runstate == TD_EXITED) | |
1366 | flags = 0; | |
1367 | ||
1368 | /* | |
1369 | * check if someone quit or got killed in an unusual way | |
1370 | */ | |
1371 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1372 | if (ret < 0) { |
a2f77c9f | 1373 | if (errno == ECHILD) { |
5921e80c JA |
1374 | log_err("fio: pid=%d disappeared %d\n", |
1375 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1376 | td_set_runstate(td, TD_REAPED); |
1377 | goto reaped; | |
1378 | } | |
1379 | perror("waitpid"); | |
1380 | } else if (ret == td->pid) { | |
1381 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1382 | int sig = WTERMSIG(status); |
1383 | ||
03e20d68 | 1384 | if (sig != SIGTERM) |
5ec10eaa | 1385 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1386 | (int) td->pid, sig); |
fab6aa71 JA |
1387 | td_set_runstate(td, TD_REAPED); |
1388 | goto reaped; | |
1389 | } | |
a2f77c9f JA |
1390 | if (WIFEXITED(status)) { |
1391 | if (WEXITSTATUS(status) && !td->error) | |
1392 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1393 | |
a2f77c9f JA |
1394 | td_set_runstate(td, TD_REAPED); |
1395 | goto reaped; | |
a6418147 JA |
1396 | } |
1397 | } | |
ebac4655 | 1398 | |
a2f77c9f JA |
1399 | /* |
1400 | * thread is not dead, continue | |
1401 | */ | |
e48676ba | 1402 | pending++; |
a2f77c9f | 1403 | continue; |
fab6aa71 | 1404 | reaped: |
ebac4655 | 1405 | (*nr_running)--; |
581e7141 JA |
1406 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1407 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1408 | if (!td->pid) |
1409 | pending--; | |
a2f77c9f JA |
1410 | |
1411 | if (td->error) | |
1412 | exit_value++; | |
d3eeeabc JA |
1413 | |
1414 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1415 | } |
b990b5c0 | 1416 | |
1f809d15 | 1417 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1418 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1419 | } |
1420 | ||
be4ecfdf JA |
1421 | static void *gtod_thread_main(void *data) |
1422 | { | |
1423 | fio_mutex_up(startup_mutex); | |
1424 | ||
1425 | /* | |
1426 | * As long as we have jobs around, update the clock. It would be nice | |
1427 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1428 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1429 | * it - we don't want to lose precision. | |
1430 | */ | |
1431 | while (threads) { | |
1432 | fio_gtod_update(); | |
1433 | nop; | |
1434 | } | |
1435 | ||
1436 | return NULL; | |
1437 | } | |
1438 | ||
1439 | static int fio_start_gtod_thread(void) | |
1440 | { | |
304a47c7 | 1441 | pthread_attr_t attr; |
f42c153d JA |
1442 | int ret; |
1443 | ||
304a47c7 VA |
1444 | pthread_attr_init(&attr); |
1445 | pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); | |
1446 | ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); | |
1447 | pthread_attr_destroy(&attr); | |
f42c153d JA |
1448 | if (ret) { |
1449 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1450 | return 1; |
1451 | } | |
f42c153d JA |
1452 | |
1453 | ret = pthread_detach(gtod_thread); | |
1454 | if (ret) { | |
1455 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1456 | return 1; |
1457 | } | |
1458 | ||
29adda3c | 1459 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1460 | fio_mutex_down(startup_mutex); |
29adda3c | 1461 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1462 | return 0; |
1463 | } | |
1464 | ||
906c8d75 JA |
1465 | /* |
1466 | * Main function for kicking off and reaping jobs, as needed. | |
1467 | */ | |
ebac4655 JA |
1468 | static void run_threads(void) |
1469 | { | |
ebac4655 JA |
1470 | struct thread_data *td; |
1471 | unsigned long spent; | |
1472 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1473 | |
2f9ade3c JA |
1474 | if (fio_pin_memory()) |
1475 | return; | |
ebac4655 | 1476 | |
be4ecfdf JA |
1477 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1478 | return; | |
1479 | ||
c6ae0a5b | 1480 | if (!terse_output) { |
4ceb30d4 | 1481 | log_info("Starting "); |
9cedf167 | 1482 | if (nr_thread) |
4ceb30d4 | 1483 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1484 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1485 | if (nr_process) { |
1486 | if (nr_thread) | |
1487 | printf(" and "); | |
4ceb30d4 | 1488 | log_info("%d process%s", nr_process, |
5ec10eaa | 1489 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1490 | } |
4ceb30d4 | 1491 | log_info("\n"); |
c6ae0a5b JA |
1492 | fflush(stdout); |
1493 | } | |
c04f7ec3 | 1494 | |
697a606c | 1495 | set_sig_handlers(); |
4efa970e | 1496 | |
ebac4655 JA |
1497 | todo = thread_number; |
1498 | nr_running = 0; | |
1499 | nr_started = 0; | |
1500 | m_rate = t_rate = 0; | |
1501 | ||
34572e28 | 1502 | for_each_td(td, i) { |
263e529f | 1503 | print_status_init(td->thread_number - 1); |
ebac4655 | 1504 | |
067430bb | 1505 | if (!td->o.create_serialize) |
ebac4655 JA |
1506 | continue; |
1507 | ||
1508 | /* | |
1509 | * do file setup here so it happens sequentially, | |
1510 | * we don't want X number of threads getting their | |
1511 | * client data interspersed on disk | |
1512 | */ | |
53cdc686 | 1513 | if (setup_files(td)) { |
5bf13a5a JA |
1514 | exit_value++; |
1515 | if (td->error) | |
5921e80c JA |
1516 | log_err("fio: pid=%d, err=%d/%s\n", |
1517 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1518 | td_set_runstate(td, TD_REAPED); |
1519 | todo--; | |
c69e8875 JA |
1520 | } else { |
1521 | struct fio_file *f; | |
2b13e716 | 1522 | unsigned int j; |
c69e8875 JA |
1523 | |
1524 | /* | |
1525 | * for sharing to work, each job must always open | |
1526 | * its own files. so close them, if we opened them | |
1527 | * for creation | |
1528 | */ | |
2b13e716 | 1529 | for_each_file(td, f, j) { |
22a57ba8 JA |
1530 | if (fio_file_open(f)) |
1531 | td_io_close_file(td, f); | |
22a57ba8 | 1532 | } |
5ec10eaa | 1533 | } |
ebac4655 JA |
1534 | } |
1535 | ||
a2f77c9f JA |
1536 | set_genesis_time(); |
1537 | ||
ebac4655 | 1538 | while (todo) { |
fca70358 | 1539 | struct thread_data *map[REAL_MAX_JOBS]; |
75154845 JA |
1540 | struct timeval this_start; |
1541 | int this_jobs = 0, left; | |
1542 | ||
ebac4655 JA |
1543 | /* |
1544 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1545 | */ | |
34572e28 | 1546 | for_each_td(td, i) { |
ebac4655 JA |
1547 | if (td->runstate != TD_NOT_CREATED) |
1548 | continue; | |
1549 | ||
1550 | /* | |
1551 | * never got a chance to start, killed by other | |
1552 | * thread for some reason | |
1553 | */ | |
1554 | if (td->terminate) { | |
1555 | todo--; | |
1556 | continue; | |
1557 | } | |
1558 | ||
2dc1bbeb | 1559 | if (td->o.start_delay) { |
263e529f | 1560 | spent = mtime_since_genesis(); |
ebac4655 | 1561 | |
2dc1bbeb | 1562 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1563 | continue; |
1564 | } | |
1565 | ||
6f88bcb0 JA |
1566 | if (td->o.stonewall && (nr_started || nr_running)) { |
1567 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1568 | td->o.name); | |
ebac4655 | 1569 | break; |
6f88bcb0 | 1570 | } |
ebac4655 | 1571 | |
067430bb JA |
1572 | init_disk_util(td); |
1573 | ||
75154845 JA |
1574 | /* |
1575 | * Set state to created. Thread will transition | |
1576 | * to TD_INITIALIZED when it's done setting up. | |
1577 | */ | |
ebac4655 | 1578 | td_set_runstate(td, TD_CREATED); |
75154845 | 1579 | map[this_jobs++] = td; |
ebac4655 JA |
1580 | nr_started++; |
1581 | ||
2dc1bbeb | 1582 | if (td->o.use_thread) { |
f42c153d JA |
1583 | int ret; |
1584 | ||
ee56ad50 | 1585 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1586 | ret = pthread_create(&td->thread, NULL, |
1587 | thread_main, td); | |
1588 | if (ret) { | |
1589 | log_err("pthread_create: %s\n", | |
1590 | strerror(ret)); | |
ebac4655 | 1591 | nr_started--; |
c8bb6faf | 1592 | break; |
ebac4655 | 1593 | } |
f42c153d JA |
1594 | ret = pthread_detach(td->thread); |
1595 | if (ret) | |
1596 | log_err("pthread_detach: %s", | |
1597 | strerror(ret)); | |
ebac4655 | 1598 | } else { |
5e1d306e | 1599 | pid_t pid; |
ee56ad50 | 1600 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1601 | pid = fork(); |
1602 | if (!pid) { | |
a6418147 JA |
1603 | int ret = fork_main(shm_id, i); |
1604 | ||
5e1d306e JA |
1605 | _exit(ret); |
1606 | } else if (i == fio_debug_jobno) | |
1607 | *fio_debug_jobp = pid; | |
ebac4655 | 1608 | } |
29adda3c | 1609 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1610 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1611 | log_err("fio: job startup hung? exiting.\n"); | |
1612 | terminate_threads(TERMINATE_ALL); | |
1613 | fio_abort = 1; | |
1614 | nr_started--; | |
1615 | break; | |
1616 | } | |
29adda3c | 1617 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1618 | } |
1619 | ||
1620 | /* | |
75154845 JA |
1621 | * Wait for the started threads to transition to |
1622 | * TD_INITIALIZED. | |
ebac4655 | 1623 | */ |
02bcaa8c | 1624 | fio_gettime(&this_start, NULL); |
75154845 | 1625 | left = this_jobs; |
6ce15a32 | 1626 | while (left && !fio_abort) { |
75154845 JA |
1627 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1628 | break; | |
1629 | ||
1630 | usleep(100000); | |
1631 | ||
1632 | for (i = 0; i < this_jobs; i++) { | |
1633 | td = map[i]; | |
1634 | if (!td) | |
1635 | continue; | |
b6f4d880 | 1636 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1637 | map[i] = NULL; |
1638 | left--; | |
b6f4d880 JA |
1639 | } else if (td->runstate >= TD_EXITED) { |
1640 | map[i] = NULL; | |
1641 | left--; | |
1642 | todo--; | |
1643 | nr_running++; /* work-around... */ | |
75154845 JA |
1644 | } |
1645 | } | |
1646 | } | |
1647 | ||
1648 | if (left) { | |
3b70d7e5 | 1649 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1650 | for (i = 0; i < this_jobs; i++) { |
1651 | td = map[i]; | |
1652 | if (!td) | |
1653 | continue; | |
1654 | kill(td->pid, SIGTERM); | |
1655 | } | |
1656 | break; | |
1657 | } | |
1658 | ||
1659 | /* | |
b6f4d880 | 1660 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1661 | */ |
34572e28 | 1662 | for_each_td(td, i) { |
75154845 | 1663 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1664 | continue; |
1665 | ||
b29ee5b3 JA |
1666 | if (in_ramp_time(td)) |
1667 | td_set_runstate(td, TD_RAMP); | |
1668 | else | |
1669 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1670 | nr_running++; |
1671 | nr_started--; | |
581e7141 JA |
1672 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1673 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1674 | todo--; |
cdd18ad8 | 1675 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1676 | } |
1677 | ||
1678 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1679 | ||
1680 | if (todo) | |
1681 | usleep(100000); | |
1682 | } | |
1683 | ||
1684 | while (nr_running) { | |
1685 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1686 | usleep(10000); | |
1687 | } | |
1688 | ||
1689 | update_io_ticks(); | |
2f9ade3c | 1690 | fio_unpin_memory(); |
ebac4655 JA |
1691 | } |
1692 | ||
d3cc4ebf | 1693 | int main(int argc, char *argv[], char *envp[]) |
ebac4655 | 1694 | { |
29d610e1 JA |
1695 | long ps; |
1696 | ||
d3cc4ebf JA |
1697 | arch_init(envp); |
1698 | ||
2e5cdb11 JA |
1699 | sinit(); |
1700 | ||
dbe1125e JA |
1701 | /* |
1702 | * We need locale for number printing, if it isn't set then just | |
1703 | * go with the US format. | |
1704 | */ | |
1705 | if (!getenv("LC_NUMERIC")) | |
1706 | setlocale(LC_NUMERIC, "en_US"); | |
1707 | ||
29d610e1 JA |
1708 | ps = sysconf(_SC_PAGESIZE); |
1709 | if (ps < 0) { | |
1710 | log_err("Failed to get page size\n"); | |
1711 | return 1; | |
1712 | } | |
1713 | ||
cfc99db7 | 1714 | page_size = ps; |
29d610e1 JA |
1715 | page_mask = ps - 1; |
1716 | ||
74929ac2 JA |
1717 | fio_keywords_init(); |
1718 | ||
1719 | if (parse_options(argc, argv)) | |
1720 | return 1; | |
1721 | ||
07b3232d JA |
1722 | if (exec_profile && load_profile(exec_profile)) |
1723 | return 1; | |
1724 | ||
74929ac2 JA |
1725 | if (!thread_number) |
1726 | return 0; | |
1727 | ||
bb3884d8 JA |
1728 | if (write_bw_log) { |
1729 | setup_log(&agg_io_log[DDIR_READ]); | |
1730 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1731 | } | |
1732 | ||
cdd18ad8 | 1733 | startup_mutex = fio_mutex_init(0); |
03e20d68 BC |
1734 | if (startup_mutex == NULL) |
1735 | return 1; | |
d2bbb8d9 | 1736 | writeout_mutex = fio_mutex_init(1); |
03e20d68 BC |
1737 | if (writeout_mutex == NULL) |
1738 | return 1; | |
07739b57 | 1739 | |
a2f77c9f | 1740 | set_genesis_time(); |
7985701b | 1741 | create_disk_util_thread(); |
ebac4655 | 1742 | |
dae5341c JA |
1743 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1744 | INIT_FLIST_HEAD(cgroup_list); | |
1745 | ||
ebac4655 | 1746 | run_threads(); |
6ce15a32 | 1747 | |
bb3884d8 | 1748 | if (!fio_abort) { |
6ce15a32 | 1749 | show_run_stats(); |
bb3884d8 | 1750 | if (write_bw_log) { |
5ec10eaa JA |
1751 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1752 | __finish_log(agg_io_log[DDIR_WRITE], | |
1753 | "agg-write_bw.log"); | |
bb3884d8 JA |
1754 | } |
1755 | } | |
ebac4655 | 1756 | |
dae5341c JA |
1757 | cgroup_kill(cgroup_list); |
1758 | sfree(cgroup_list); | |
61eb313e | 1759 | sfree(cgroup_mnt); |
39f22027 | 1760 | |
cdd18ad8 | 1761 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1762 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1763 | return exit_value; |
ebac4655 | 1764 | } |