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 | */ | |
6aa27816 MC |
105 | if (!td->pid) |
106 | continue; | |
107 | else if (td->runstate < TD_RAMP) | |
03e20d68 | 108 | kill(td->pid, SIGTERM); |
f63f7f3b JA |
109 | else { |
110 | struct ioengine_ops *ops = td->io_ops; | |
111 | ||
03e20d68 BC |
112 | if (ops && (ops->flags & FIO_SIGTERM)) |
113 | kill(td->pid, SIGTERM); | |
f63f7f3b | 114 | } |
ebac4655 JA |
115 | } |
116 | } | |
117 | } | |
118 | ||
697a606c JA |
119 | static void sig_int(int sig) |
120 | { | |
121 | if (threads) { | |
4ceb30d4 | 122 | log_info("\nfio: terminating on signal %d\n", sig); |
5ec10eaa | 123 | fflush(stdout); |
deaa1af2 | 124 | exit_value = 128; |
5ec10eaa | 125 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
126 | } |
127 | } | |
128 | ||
7985701b | 129 | static void *disk_thread_main(void *data) |
a5b01f1b | 130 | { |
7985701b JA |
131 | fio_mutex_up(startup_mutex); |
132 | ||
133 | while (threads) { | |
134 | usleep(DISK_UTIL_MSEC * 1000); | |
9a793c2f JA |
135 | if (!threads) |
136 | break; | |
7985701b JA |
137 | update_io_ticks(); |
138 | print_thread_status(); | |
139 | } | |
140 | ||
141 | return NULL; | |
a5b01f1b JA |
142 | } |
143 | ||
7985701b | 144 | static int create_disk_util_thread(void) |
a5b01f1b | 145 | { |
7985701b | 146 | int ret; |
a5b01f1b | 147 | |
69204d6e | 148 | ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); |
7985701b JA |
149 | if (ret) { |
150 | log_err("Can't create disk util thread: %s\n", strerror(ret)); | |
151 | return 1; | |
152 | } | |
a5b01f1b | 153 | |
7985701b JA |
154 | ret = pthread_detach(disk_util_thread); |
155 | if (ret) { | |
156 | log_err("Can't detatch disk util thread: %s\n", strerror(ret)); | |
157 | return 1; | |
158 | } | |
03e20d68 | 159 | |
7985701b JA |
160 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
161 | fio_mutex_down(startup_mutex); | |
162 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
163 | return 0; | |
a5b01f1b JA |
164 | } |
165 | ||
697a606c JA |
166 | static void set_sig_handlers(void) |
167 | { | |
168 | struct sigaction act; | |
169 | ||
697a606c JA |
170 | memset(&act, 0, sizeof(act)); |
171 | act.sa_handler = sig_int; | |
172 | act.sa_flags = SA_RESTART; | |
173 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 174 | |
5f8f9726 | 175 | memset(&act, 0, sizeof(act)); |
b20b6be4 | 176 | act.sa_handler = sig_int; |
5f8f9726 | 177 | act.sa_flags = SA_RESTART; |
03e20d68 | 178 | sigaction(SIGTERM, &act, NULL); |
697a606c JA |
179 | } |
180 | ||
906c8d75 JA |
181 | /* |
182 | * Check if we are above the minimum rate given. | |
183 | */ | |
581e7141 | 184 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
bd93e34b | 185 | enum fio_ddir ddir) |
ebac4655 | 186 | { |
0904200b | 187 | unsigned long long bytes = 0; |
4e991c23 | 188 | unsigned long iops = 0; |
ebac4655 JA |
189 | unsigned long spent; |
190 | unsigned long rate; | |
581e7141 JA |
191 | unsigned int ratemin = 0; |
192 | unsigned int rate_iops = 0; | |
193 | unsigned int rate_iops_min = 0; | |
ebac4655 | 194 | |
ff58fced JA |
195 | assert(ddir_rw(ddir)); |
196 | ||
d982d873 JA |
197 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
198 | return 0; | |
199 | ||
ebac4655 JA |
200 | /* |
201 | * allow a 2 second settle period in the beginning | |
202 | */ | |
203 | if (mtime_since(&td->start, now) < 2000) | |
204 | return 0; | |
205 | ||
581e7141 JA |
206 | iops += td->io_blocks[ddir]; |
207 | bytes += td->this_io_bytes[ddir]; | |
208 | ratemin += td->o.ratemin[ddir]; | |
209 | rate_iops += td->o.rate_iops[ddir]; | |
210 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 211 | |
ebac4655 JA |
212 | /* |
213 | * if rate blocks is set, sample is running | |
214 | */ | |
581e7141 JA |
215 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
216 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 217 | if (spent < td->o.ratecycle) |
ebac4655 JA |
218 | return 0; |
219 | ||
581e7141 | 220 | if (td->o.rate[ddir]) { |
4e991c23 JA |
221 | /* |
222 | * check bandwidth specified rate | |
223 | */ | |
581e7141 | 224 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 225 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 226 | ratemin); |
4e991c23 JA |
227 | return 1; |
228 | } else { | |
581e7141 JA |
229 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
230 | if (rate < ratemin || | |
231 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 232 | log_err("%s: min rate %u not met, got" |
b22989b9 | 233 | " %luKB/sec\n", td->o.name, |
581e7141 | 234 | ratemin, rate); |
4e991c23 JA |
235 | return 1; |
236 | } | |
237 | } | |
413dd459 | 238 | } else { |
4e991c23 JA |
239 | /* |
240 | * checks iops specified rate | |
241 | */ | |
581e7141 | 242 | if (iops < rate_iops) { |
5ec10eaa | 243 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 244 | td->o.name, rate_iops); |
413dd459 | 245 | return 1; |
4e991c23 | 246 | } else { |
581e7141 JA |
247 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
248 | if (rate < rate_iops_min || | |
249 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
250 | log_err("%s: min iops rate %u not met," |
251 | " got %lu\n", td->o.name, | |
581e7141 | 252 | rate_iops_min, rate); |
4e991c23 | 253 | } |
413dd459 | 254 | } |
ebac4655 JA |
255 | } |
256 | } | |
257 | ||
581e7141 JA |
258 | td->rate_bytes[ddir] = bytes; |
259 | td->rate_blocks[ddir] = iops; | |
260 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
261 | return 0; |
262 | } | |
263 | ||
581e7141 JA |
264 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
265 | unsigned long *bytes_done) | |
266 | { | |
267 | int ret = 0; | |
268 | ||
269 | if (bytes_done[0]) | |
270 | ret |= __check_min_rate(td, now, 0); | |
271 | if (bytes_done[1]) | |
272 | ret |= __check_min_rate(td, now, 1); | |
273 | ||
274 | return ret; | |
275 | } | |
276 | ||
ebac4655 JA |
277 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
278 | { | |
2dc1bbeb | 279 | if (!td->o.timeout) |
ebac4655 | 280 | return 0; |
2dc1bbeb | 281 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
282 | return 1; |
283 | ||
284 | return 0; | |
285 | } | |
286 | ||
906c8d75 JA |
287 | /* |
288 | * When job exits, we can cancel the in-flight IO if we are using async | |
289 | * io. Attempt to do so. | |
290 | */ | |
ebac4655 JA |
291 | static void cleanup_pending_aio(struct thread_data *td) |
292 | { | |
01743ee1 | 293 | struct flist_head *entry, *n; |
ebac4655 JA |
294 | struct io_u *io_u; |
295 | int r; | |
296 | ||
297 | /* | |
298 | * get immediately available events, if any | |
299 | */ | |
581e7141 | 300 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
301 | if (r < 0) |
302 | return; | |
ebac4655 JA |
303 | |
304 | /* | |
305 | * now cancel remaining active events | |
306 | */ | |
2866c82d | 307 | if (td->io_ops->cancel) { |
01743ee1 JA |
308 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
309 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 310 | |
0c6e7517 JA |
311 | /* |
312 | * if the io_u isn't in flight, then that generally | |
313 | * means someone leaked an io_u. complain but fix | |
314 | * it up, so we don't stall here. | |
315 | */ | |
316 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
317 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 318 | put_io_u(td, io_u); |
0c6e7517 JA |
319 | } else { |
320 | r = td->io_ops->cancel(td, io_u); | |
321 | if (!r) | |
322 | put_io_u(td, io_u); | |
323 | } | |
ebac4655 JA |
324 | } |
325 | } | |
326 | ||
97601024 | 327 | if (td->cur_depth) |
581e7141 | 328 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
329 | } |
330 | ||
858a3d47 JA |
331 | /* |
332 | * Helper to handle the final sync of a file. Works just like the normal | |
333 | * io path, just does everything sync. | |
334 | */ | |
335 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
336 | { | |
337 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
338 | int ret; |
339 | ||
340 | if (!io_u) | |
341 | return 1; | |
342 | ||
343 | io_u->ddir = DDIR_SYNC; | |
344 | io_u->file = f; | |
345 | ||
346 | if (td_io_prep(td, io_u)) { | |
347 | put_io_u(td, io_u); | |
348 | return 1; | |
349 | } | |
350 | ||
755200a3 | 351 | requeue: |
858a3d47 | 352 | ret = td_io_queue(td, io_u); |
36167d82 | 353 | if (ret < 0) { |
e1161c32 | 354 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 355 | put_io_u(td, io_u); |
858a3d47 | 356 | return 1; |
36167d82 | 357 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 358 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 359 | return 1; |
36167d82 JA |
360 | } else if (ret == FIO_Q_COMPLETED) { |
361 | if (io_u->error) { | |
e1161c32 | 362 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
363 | return 1; |
364 | } | |
858a3d47 | 365 | |
581e7141 | 366 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 367 | return 1; |
755200a3 JA |
368 | } else if (ret == FIO_Q_BUSY) { |
369 | if (td_io_commit(td)) | |
370 | return 1; | |
371 | goto requeue; | |
858a3d47 JA |
372 | } |
373 | ||
374 | return 0; | |
375 | } | |
376 | ||
687c6a23 JA |
377 | static inline void __update_tv_cache(struct thread_data *td) |
378 | { | |
379 | fio_gettime(&td->tv_cache, NULL); | |
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) | |
687c6a23 | 385 | __update_tv_cache(td); |
993bf48b JA |
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)) { | |
687c6a23 JA |
471 | __update_tv_cache(td); |
472 | if (runtime_exceeded(td, &td->tv_cache)) { | |
473 | td->terminate = 1; | |
474 | break; | |
475 | } | |
069c2918 | 476 | } |
02bcaa8c | 477 | |
dcb4830d RR |
478 | io_u = __get_io_u(td); |
479 | if (!io_u) | |
480 | break; | |
481 | ||
069c2918 JA |
482 | if (get_next_verify(td, io_u)) { |
483 | put_io_u(td, io_u); | |
ebac4655 | 484 | break; |
069c2918 | 485 | } |
ebac4655 | 486 | |
069c2918 JA |
487 | if (td_io_prep(td, io_u)) { |
488 | put_io_u(td, io_u); | |
53cdc686 | 489 | break; |
069c2918 | 490 | } |
d7762cf8 | 491 | |
e8462bd8 JA |
492 | if (td->o.verify_async) |
493 | io_u->end_io = verify_io_u_async; | |
494 | else | |
495 | io_u->end_io = verify_io_u; | |
53cdc686 | 496 | |
11786802 | 497 | ret = td_io_queue(td, io_u); |
36167d82 JA |
498 | switch (ret) { |
499 | case FIO_Q_COMPLETED: | |
f2bba182 | 500 | if (io_u->error) { |
22819ec2 | 501 | ret = -io_u->error; |
f2bba182 RR |
502 | clear_io_u(td, io_u); |
503 | } else if (io_u->resid) { | |
36167d82 | 504 | int bytes = io_u->xfer_buflen - io_u->resid; |
ebac4655 | 505 | |
9e9d164e JA |
506 | /* |
507 | * zero read, fail | |
508 | */ | |
509 | if (!bytes) { | |
41d9f0fd | 510 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
511 | put_io_u(td, io_u); |
512 | break; | |
513 | } | |
8400d9b2 | 514 | |
36167d82 JA |
515 | io_u->xfer_buflen = io_u->resid; |
516 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
517 | io_u->offset += bytes; |
518 | ||
ff58fced JA |
519 | if (ddir_rw(io_u->ddir)) |
520 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 521 | |
2b13e716 | 522 | f = io_u->file; |
d460eb31 | 523 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
524 | goto sync_done; |
525 | ||
11786802 JA |
526 | requeue_io_u(td, &io_u); |
527 | } else { | |
8400d9b2 | 528 | sync_done: |
581e7141 | 529 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
530 | if (ret < 0) |
531 | break; | |
36167d82 | 532 | } |
36167d82 JA |
533 | continue; |
534 | case FIO_Q_QUEUED: | |
535 | break; | |
755200a3 JA |
536 | case FIO_Q_BUSY: |
537 | requeue_io_u(td, &io_u); | |
5451792e JA |
538 | ret2 = td_io_commit(td); |
539 | if (ret2 < 0) | |
540 | ret = ret2; | |
755200a3 | 541 | break; |
36167d82 JA |
542 | default: |
543 | assert(ret < 0); | |
e1161c32 | 544 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
545 | break; |
546 | } | |
547 | ||
f2bba182 | 548 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
549 | break; |
550 | ||
ebac4655 | 551 | /* |
3af6ef39 | 552 | * if we can queue more, do so. but check if there are |
ccc02050 JA |
553 | * completed io_u's first. Note that we can get BUSY even |
554 | * without IO queued, if the system is resource starved. | |
ebac4655 | 555 | */ |
ccc02050 | 556 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 557 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
558 | min_events = min(td->o.iodepth_batch_complete, |
559 | td->cur_depth); | |
1f84f1d9 | 560 | if (full && !min_events && td->o.iodepth_batch_complete != 0) |
838bc709 | 561 | min_events = 1; |
e916b390 | 562 | |
4950421a JA |
563 | do { |
564 | /* | |
565 | * Reap required number of io units, if any, | |
566 | * and do the verification on them through | |
567 | * the callback handler | |
568 | */ | |
581e7141 | 569 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
570 | ret = -1; |
571 | break; | |
572 | } | |
573 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
574 | } | |
575 | if (ret < 0) | |
ebac4655 | 576 | break; |
36167d82 | 577 | } |
ebac4655 | 578 | |
c01c0395 JA |
579 | if (!td->error) { |
580 | min_events = td->cur_depth; | |
581 | ||
582 | if (min_events) | |
581e7141 | 583 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 584 | } else |
ebac4655 JA |
585 | cleanup_pending_aio(td); |
586 | ||
587 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
588 | |
589 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
590 | } |
591 | ||
32cd46a0 | 592 | /* |
906c8d75 | 593 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
594 | * and reaps them, checking for rate and errors along the way. |
595 | */ | |
ebac4655 JA |
596 | static void do_io(struct thread_data *td) |
597 | { | |
af52b345 JA |
598 | unsigned int i; |
599 | int ret = 0; | |
ebac4655 | 600 | |
b29ee5b3 JA |
601 | if (in_ramp_time(td)) |
602 | td_set_runstate(td, TD_RAMP); | |
603 | else | |
604 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 605 | |
457bf399 | 606 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
a917a8b3 | 607 | (!flist_empty(&td->trim_list)) || |
457bf399 | 608 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { |
97601024 | 609 | struct timeval comp_time; |
581e7141 | 610 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 611 | int min_evts = 0; |
ebac4655 | 612 | struct io_u *io_u; |
4950421a | 613 | int ret2, full; |
ebac4655 JA |
614 | |
615 | if (td->terminate) | |
616 | break; | |
617 | ||
993bf48b | 618 | update_tv_cache(td); |
97601024 | 619 | |
993bf48b | 620 | if (runtime_exceeded(td, &td->tv_cache)) { |
687c6a23 JA |
621 | __update_tv_cache(td); |
622 | if (runtime_exceeded(td, &td->tv_cache)) { | |
623 | td->terminate = 1; | |
624 | break; | |
625 | } | |
97601024 | 626 | } |
36167d82 | 627 | |
dcb4830d RR |
628 | io_u = get_io_u(td); |
629 | if (!io_u) | |
630 | break; | |
631 | ||
b67740d3 JA |
632 | /* |
633 | * Add verification end_io handler, if asked to verify | |
634 | * a previously written file. | |
635 | */ | |
715d2b3e JA |
636 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
637 | !td_rw(td)) { | |
e8462bd8 JA |
638 | if (td->o.verify_async) |
639 | io_u->end_io = verify_io_u_async; | |
640 | else | |
641 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 642 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
643 | } else if (in_ramp_time(td)) |
644 | td_set_runstate(td, TD_RAMP); | |
645 | else | |
d8fab1b6 | 646 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 647 | |
11786802 | 648 | ret = td_io_queue(td, io_u); |
36167d82 JA |
649 | switch (ret) { |
650 | case FIO_Q_COMPLETED: | |
f2bba182 | 651 | if (io_u->error) { |
5451792e | 652 | ret = -io_u->error; |
f2bba182 RR |
653 | clear_io_u(td, io_u); |
654 | } else if (io_u->resid) { | |
36167d82 | 655 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 656 | struct fio_file *f = io_u->file; |
36167d82 | 657 | |
9e9d164e JA |
658 | /* |
659 | * zero read, fail | |
660 | */ | |
661 | if (!bytes) { | |
41d9f0fd | 662 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
663 | put_io_u(td, io_u); |
664 | break; | |
665 | } | |
666 | ||
cec6b55d | 667 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 668 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
669 | io_u->offset += bytes; |
670 | ||
ff58fced JA |
671 | if (ddir_rw(io_u->ddir)) |
672 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 673 | |
d460eb31 | 674 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
675 | goto sync_done; |
676 | ||
11786802 JA |
677 | requeue_io_u(td, &io_u); |
678 | } else { | |
5a7c5680 | 679 | sync_done: |
581e7141 JA |
680 | if (__should_check_rate(td, 0) || |
681 | __should_check_rate(td, 1)) | |
993bf48b JA |
682 | fio_gettime(&comp_time, NULL); |
683 | ||
581e7141 JA |
684 | ret = io_u_sync_complete(td, io_u, bytes_done); |
685 | if (ret < 0) | |
686 | break; | |
cec6b55d | 687 | } |
36167d82 JA |
688 | break; |
689 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
690 | /* |
691 | * if the engine doesn't have a commit hook, | |
692 | * the io_u is really queued. if it does have such | |
693 | * a hook, it has to call io_u_queued() itself. | |
694 | */ | |
695 | if (td->io_ops->commit == NULL) | |
696 | io_u_queued(td, io_u); | |
36167d82 | 697 | break; |
755200a3 JA |
698 | case FIO_Q_BUSY: |
699 | requeue_io_u(td, &io_u); | |
5451792e JA |
700 | ret2 = td_io_commit(td); |
701 | if (ret2 < 0) | |
702 | ret = ret2; | |
755200a3 | 703 | break; |
36167d82 JA |
704 | default: |
705 | assert(ret < 0); | |
706 | put_io_u(td, io_u); | |
707 | break; | |
ebac4655 JA |
708 | } |
709 | ||
f2bba182 | 710 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
711 | break; |
712 | ||
97601024 | 713 | /* |
ccc02050 JA |
714 | * See if we need to complete some commands. Note that we |
715 | * can get BUSY even without IO queued, if the system is | |
716 | * resource starved. | |
97601024 | 717 | */ |
ccc02050 | 718 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 719 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
720 | min_evts = min(td->o.iodepth_batch_complete, |
721 | td->cur_depth); | |
1f84f1d9 | 722 | if (full && !min_evts && td->o.iodepth_batch_complete != 0) |
36167d82 | 723 | min_evts = 1; |
4950421a | 724 | |
581e7141 JA |
725 | if (__should_check_rate(td, 0) || |
726 | __should_check_rate(td, 1)) | |
993bf48b | 727 | fio_gettime(&comp_time, NULL); |
4950421a JA |
728 | |
729 | do { | |
581e7141 JA |
730 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
731 | if (ret < 0) | |
4950421a JA |
732 | break; |
733 | ||
4950421a | 734 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
735 | } |
736 | ||
4950421a JA |
737 | if (ret < 0) |
738 | break; | |
581e7141 | 739 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
740 | continue; |
741 | ||
581e7141 JA |
742 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
743 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
744 | if (exitall_on_terminate) |
745 | terminate_threads(td->groupid); | |
746 | td_verror(td, EIO, "check_min_rate"); | |
747 | break; | |
748 | } | |
ebac4655 JA |
749 | } |
750 | ||
2dc1bbeb | 751 | if (td->o.thinktime) { |
9c1f7434 JA |
752 | unsigned long long b; |
753 | ||
754 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 755 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
756 | int left; |
757 | ||
2dc1bbeb | 758 | if (td->o.thinktime_spin) |
6dd6f2cd | 759 | usec_spin(td->o.thinktime_spin); |
48097d5c | 760 | |
2dc1bbeb | 761 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
762 | if (left) |
763 | usec_sleep(td, left); | |
764 | } | |
9c1f7434 | 765 | } |
ebac4655 JA |
766 | } |
767 | ||
d708ac5e JA |
768 | if (td->trim_entries) |
769 | printf("trim entries %ld\n", td->trim_entries); | |
a917a8b3 | 770 | |
aa31f1f1 SL |
771 | if (td->o.fill_device && td->error == ENOSPC) { |
772 | td->error = 0; | |
773 | td->terminate = 1; | |
774 | } | |
4d2413c6 | 775 | if (!td->error) { |
3d7c391d JA |
776 | struct fio_file *f; |
777 | ||
c01c0395 | 778 | i = td->cur_depth; |
d06093e7 | 779 | if (i) { |
581e7141 | 780 | ret = io_u_queued_complete(td, i, NULL); |
d06093e7 SL |
781 | if (td->o.fill_device && td->error == ENOSPC) |
782 | td->error = 0; | |
783 | } | |
ebac4655 | 784 | |
2dc1bbeb | 785 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 786 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
787 | |
788 | for_each_file(td, f, i) { | |
d6aed795 | 789 | if (!fio_file_open(f)) |
3d7b485f | 790 | continue; |
858a3d47 | 791 | fio_io_sync(td, f); |
3d7b485f | 792 | } |
84585003 | 793 | } |
c01c0395 JA |
794 | } else |
795 | cleanup_pending_aio(td); | |
2ba1c290 JA |
796 | |
797 | /* | |
798 | * stop job if we failed doing any IO | |
799 | */ | |
800 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
801 | td->done = 1; | |
ebac4655 JA |
802 | } |
803 | ||
ebac4655 JA |
804 | static void cleanup_io_u(struct thread_data *td) |
805 | { | |
01743ee1 | 806 | struct flist_head *entry, *n; |
ebac4655 JA |
807 | struct io_u *io_u; |
808 | ||
01743ee1 JA |
809 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
810 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 811 | |
01743ee1 | 812 | flist_del(&io_u->list); |
91aea6ec | 813 | fio_memfree(io_u, sizeof(*io_u)); |
ebac4655 JA |
814 | } |
815 | ||
2f9ade3c | 816 | free_io_mem(td); |
ebac4655 JA |
817 | } |
818 | ||
819 | static int init_io_u(struct thread_data *td) | |
820 | { | |
821 | struct io_u *io_u; | |
a00735e6 | 822 | unsigned int max_bs; |
eb7ccf38 | 823 | int cl_align, i, max_units; |
ebac4655 JA |
824 | char *p; |
825 | ||
1e3c09af | 826 | max_units = td->o.iodepth; |
2dc1bbeb | 827 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
828 | td->orig_buffer_size = (unsigned long long) max_bs |
829 | * (unsigned long long) max_units; | |
830 | ||
831 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
832 | unsigned long bs; | |
74b025b0 | 833 | |
5ec10eaa JA |
834 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
835 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
836 | } | |
ebac4655 | 837 | |
c3990645 JA |
838 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
839 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
840 | return 1; | |
841 | } | |
842 | ||
2f9ade3c JA |
843 | if (allocate_io_mem(td)) |
844 | return 1; | |
ebac4655 | 845 | |
1c04bf78 JA |
846 | if (td->o.odirect || td->o.mem_align || |
847 | (td->io_ops->flags & FIO_RAWIO)) | |
d529ee19 | 848 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
849 | else |
850 | p = td->orig_buffer; | |
851 | ||
eb7ccf38 JA |
852 | cl_align = os_cache_line_size(); |
853 | ||
ebac4655 | 854 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
855 | void *ptr; |
856 | ||
db1cd90b JA |
857 | if (td->terminate) |
858 | return 1; | |
eb7ccf38 | 859 | |
91aea6ec JA |
860 | ptr = fio_memalign(cl_align, sizeof(*io_u)); |
861 | if (!ptr) { | |
862 | log_err("fio: unable to allocate aligned memory\n"); | |
eb7ccf38 JA |
863 | break; |
864 | } | |
865 | ||
866 | io_u = ptr; | |
ebac4655 | 867 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 868 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 869 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 870 | |
b4c5e1ac JA |
871 | if (!(td->io_ops->flags & FIO_NOIO)) { |
872 | io_u->buf = p + max_bs * i; | |
d529ee19 | 873 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 874 | |
5973cafb JA |
875 | if (td_write(td) && !td->o.refill_buffers) |
876 | io_u_fill_buffer(td, io_u, max_bs); | |
cbe8d756 RR |
877 | else if (td_write(td) && td->o.verify_pattern_bytes) { |
878 | /* | |
879 | * Fill the buffer with the pattern if we are | |
880 | * going to be doing writes. | |
881 | */ | |
7d9fb455 | 882 | fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
cbe8d756 | 883 | } |
b4c5e1ac | 884 | } |
6b9cea23 | 885 | |
b1ff3403 | 886 | io_u->index = i; |
0c6e7517 | 887 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 888 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
889 | } |
890 | ||
891 | return 0; | |
892 | } | |
893 | ||
da86774e JA |
894 | static int switch_ioscheduler(struct thread_data *td) |
895 | { | |
896 | char tmp[256], tmp2[128]; | |
897 | FILE *f; | |
898 | int ret; | |
899 | ||
ba0fbe10 | 900 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
901 | return 0; |
902 | ||
da86774e JA |
903 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
904 | ||
905 | f = fopen(tmp, "r+"); | |
906 | if (!f) { | |
cbf5c1d7 | 907 | if (errno == ENOENT) { |
5ec10eaa JA |
908 | log_err("fio: os or kernel doesn't support IO scheduler" |
909 | " switching\n"); | |
cbf5c1d7 JA |
910 | return 0; |
911 | } | |
912 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
913 | return 1; |
914 | } | |
915 | ||
916 | /* | |
917 | * Set io scheduler. | |
918 | */ | |
2dc1bbeb | 919 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 920 | if (ferror(f) || ret != 1) { |
e1161c32 | 921 | td_verror(td, errno, "fwrite"); |
da86774e JA |
922 | fclose(f); |
923 | return 1; | |
924 | } | |
925 | ||
926 | rewind(f); | |
927 | ||
928 | /* | |
929 | * Read back and check that the selected scheduler is now the default. | |
930 | */ | |
931 | ret = fread(tmp, 1, sizeof(tmp), f); | |
932 | if (ferror(f) || ret < 0) { | |
e1161c32 | 933 | td_verror(td, errno, "fread"); |
da86774e JA |
934 | fclose(f); |
935 | return 1; | |
936 | } | |
937 | ||
2dc1bbeb | 938 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 939 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 940 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 941 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
942 | fclose(f); |
943 | return 1; | |
944 | } | |
945 | ||
946 | fclose(f); | |
947 | return 0; | |
948 | } | |
949 | ||
492158cf JA |
950 | static int keep_running(struct thread_data *td) |
951 | { | |
952 | unsigned long long io_done; | |
953 | ||
20e354ef JA |
954 | if (td->done) |
955 | return 0; | |
492158cf JA |
956 | if (td->o.time_based) |
957 | return 1; | |
958 | if (td->o.loops) { | |
959 | td->o.loops--; | |
960 | return 1; | |
961 | } | |
962 | ||
5ec10eaa JA |
963 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
964 | + td->io_skip_bytes; | |
492158cf JA |
965 | if (io_done < td->o.size) |
966 | return 1; | |
967 | ||
968 | return 0; | |
969 | } | |
970 | ||
b29ee5b3 | 971 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 972 | { |
756867bd | 973 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 974 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 975 | td->zone_bytes = 0; |
581e7141 JA |
976 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
977 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 978 | |
c1324df1 JA |
979 | td->last_was_sync = 0; |
980 | ||
2ba1c290 JA |
981 | /* |
982 | * reset file done count if we are to start over | |
983 | */ | |
984 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 985 | td->nr_done_files = 0; |
b29ee5b3 JA |
986 | } |
987 | ||
988 | void reset_all_stats(struct thread_data *td) | |
989 | { | |
990 | struct timeval tv; | |
991 | int i; | |
992 | ||
993 | reset_io_counters(td); | |
994 | ||
995 | for (i = 0; i < 2; i++) { | |
996 | td->io_bytes[i] = 0; | |
997 | td->io_blocks[i] = 0; | |
998 | td->io_issues[i] = 0; | |
999 | td->ts.total_io_u[i] = 0; | |
1000 | } | |
9b836561 | 1001 | |
b29ee5b3 | 1002 | fio_gettime(&tv, NULL); |
cda99fa0 SSP |
1003 | td->ts.runtime[0] = 0; |
1004 | td->ts.runtime[1] = 0; | |
b29ee5b3 JA |
1005 | memcpy(&td->epoch, &tv, sizeof(tv)); |
1006 | memcpy(&td->start, &tv, sizeof(tv)); | |
1007 | } | |
1008 | ||
df9c26b1 | 1009 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
1010 | { |
1011 | struct fio_file *f; | |
1012 | unsigned int i; | |
b29ee5b3 JA |
1013 | |
1014 | reset_io_counters(td); | |
281f9b63 | 1015 | |
24ffd2c2 | 1016 | close_files(td); |
df9c26b1 | 1017 | for_each_file(td, f, i) |
d6aed795 | 1018 | fio_file_clear_done(f); |
f0153948 JA |
1019 | |
1020 | /* | |
1021 | * Set the same seed to get repeatable runs | |
1022 | */ | |
1023 | td_fill_rand_seeds(td); | |
ebac4655 JA |
1024 | } |
1025 | ||
398e8c4d JA |
1026 | static int exec_string(const char *string) |
1027 | { | |
1028 | int ret, newlen = strlen(string) + 1 + 8; | |
1029 | char *str; | |
1030 | ||
1031 | str = malloc(newlen); | |
1032 | sprintf(str, "sh -c %s", string); | |
1033 | ||
1034 | ret = system(str); | |
1035 | if (ret == -1) | |
1036 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1037 | ||
1038 | free(str); | |
1039 | return ret; | |
1040 | } | |
1041 | ||
906c8d75 JA |
1042 | /* |
1043 | * Entry point for the thread based jobs. The process based jobs end up | |
1044 | * here as well, after a little setup. | |
1045 | */ | |
ebac4655 JA |
1046 | static void *thread_main(void *data) |
1047 | { | |
cda99fa0 | 1048 | unsigned long long elapsed; |
ebac4655 | 1049 | struct thread_data *td = data; |
e8462bd8 | 1050 | pthread_condattr_t attr; |
a978ba68 | 1051 | int clear_state; |
ebac4655 | 1052 | |
47f767c1 | 1053 | if (!td->o.use_thread) { |
ebac4655 | 1054 | setsid(); |
47f767c1 JA |
1055 | td->pid = getpid(); |
1056 | } else | |
1057 | td->pid = gettid(); | |
ebac4655 | 1058 | |
5921e80c | 1059 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1060 | |
01743ee1 JA |
1061 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1062 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1063 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1064 | INIT_FLIST_HEAD(&td->io_log_list); | |
1065 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 | 1066 | INIT_FLIST_HEAD(&td->verify_list); |
0d29de83 | 1067 | INIT_FLIST_HEAD(&td->trim_list); |
e8462bd8 | 1068 | pthread_mutex_init(&td->io_u_lock, NULL); |
bb5d7d0b | 1069 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1070 | |
e8462bd8 JA |
1071 | pthread_condattr_init(&attr); |
1072 | pthread_cond_init(&td->verify_cond, &attr); | |
1073 | pthread_cond_init(&td->free_cond, &attr); | |
1074 | ||
db1cd90b | 1075 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1076 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1077 | fio_mutex_up(startup_mutex); |
29adda3c | 1078 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1079 | fio_mutex_down(td->mutex); |
29adda3c | 1080 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1081 | |
1082 | /* | |
cdd18ad8 | 1083 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1084 | * eating a file descriptor |
1085 | */ | |
cdd18ad8 | 1086 | fio_mutex_remove(td->mutex); |
db1cd90b | 1087 | |
dfc6e751 SH |
1088 | /* |
1089 | * A new gid requires privilege, so we need to do this before setting | |
1090 | * the uid. | |
1091 | */ | |
e0b0d892 JA |
1092 | if (td->o.gid != -1U && setgid(td->o.gid)) { |
1093 | td_verror(td, errno, "setgid"); | |
1094 | goto err; | |
1095 | } | |
dfc6e751 SH |
1096 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1097 | td_verror(td, errno, "setuid"); | |
1098 | goto err; | |
1099 | } | |
e0b0d892 | 1100 | |
be18a94f JA |
1101 | /* |
1102 | * If we have a gettimeofday() thread, make sure we exclude that | |
1103 | * thread from this job | |
1104 | */ | |
1105 | if (td->o.gtod_cpu) | |
1106 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); | |
1107 | ||
1108 | /* | |
1109 | * Set affinity first, in case it has an impact on the memory | |
1110 | * allocations. | |
1111 | */ | |
1112 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
1113 | td_verror(td, errno, "cpu_set_affinity"); | |
1114 | goto err; | |
1115 | } | |
1116 | ||
d84f8d49 JA |
1117 | /* |
1118 | * May alter parameters that init_io_u() will use, so we need to | |
1119 | * do this first. | |
1120 | */ | |
1121 | if (init_iolog(td)) | |
1122 | goto err; | |
1123 | ||
ebac4655 | 1124 | if (init_io_u(td)) |
db1cd90b | 1125 | goto err; |
ebac4655 | 1126 | |
e8462bd8 JA |
1127 | if (td->o.verify_async && verify_async_init(td)) |
1128 | goto err; | |
1129 | ||
ac684785 | 1130 | if (td->ioprio_set) { |
ebac4655 | 1131 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1132 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1133 | goto err; |
ebac4655 JA |
1134 | } |
1135 | } | |
1136 | ||
6adb38a1 | 1137 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1138 | goto err; |
1139 | ||
2dc1bbeb | 1140 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1141 | td_verror(td, errno, "nice"); |
db1cd90b | 1142 | goto err; |
b6f4d880 JA |
1143 | } |
1144 | ||
2dc1bbeb | 1145 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1146 | goto err; |
37f56873 | 1147 | |
2dc1bbeb | 1148 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1149 | goto err; |
1150 | ||
7d6c5283 JA |
1151 | if (td_io_init(td)) |
1152 | goto err; | |
1153 | ||
68727076 JA |
1154 | if (init_random_map(td)) |
1155 | goto err; | |
1156 | ||
2dc1bbeb | 1157 | if (td->o.exec_prerun) { |
398e8c4d | 1158 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1159 | goto err; |
1160 | } | |
4e0ba8af | 1161 | |
afad68f7 ZY |
1162 | if (td->o.pre_read) { |
1163 | if (pre_read_files(td) < 0) | |
1164 | goto err; | |
1165 | } | |
1166 | ||
69008999 | 1167 | fio_gettime(&td->epoch, NULL); |
756867bd | 1168 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 | 1169 | |
a978ba68 | 1170 | clear_state = 0; |
492158cf | 1171 | while (keep_running(td)) { |
02bcaa8c | 1172 | fio_gettime(&td->start, NULL); |
c5d3d790 JA |
1173 | memcpy(&td->ts.stat_sample_time[0], &td->start, |
1174 | sizeof(td->start)); | |
1175 | memcpy(&td->ts.stat_sample_time[1], &td->start, | |
1176 | sizeof(td->start)); | |
993bf48b | 1177 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1178 | |
cf2dd8d6 | 1179 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
5ec10eaa JA |
1180 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
1181 | sizeof(td->lastrate)); | |
ebac4655 | 1182 | |
df9c26b1 JA |
1183 | if (clear_state) |
1184 | clear_io_state(td); | |
a978ba68 | 1185 | |
ebac4655 JA |
1186 | prune_io_piece_log(td); |
1187 | ||
ba0fbe10 | 1188 | do_io(td); |
ebac4655 | 1189 | |
a978ba68 JA |
1190 | clear_state = 1; |
1191 | ||
38d77cae | 1192 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1193 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1194 | td->ts.runtime[DDIR_READ] += elapsed; |
38d77cae JA |
1195 | } |
1196 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1197 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1198 | td->ts.runtime[DDIR_WRITE] += elapsed; |
38d77cae | 1199 | } |
5ec10eaa | 1200 | |
ebac4655 JA |
1201 | if (td->error || td->terminate) |
1202 | break; | |
1203 | ||
e84c73a8 SL |
1204 | if (!td->o.do_verify || |
1205 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1206 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1207 | continue; |
1208 | ||
df9c26b1 | 1209 | clear_io_state(td); |
a978ba68 | 1210 | |
02bcaa8c | 1211 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1212 | |
1213 | do_verify(td); | |
1214 | ||
cda99fa0 | 1215 | td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1216 | |
1217 | if (td->error || td->terminate) | |
1218 | break; | |
1219 | } | |
1220 | ||
36dff966 | 1221 | update_rusage_stat(td); |
cda99fa0 SSP |
1222 | td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000; |
1223 | td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000; | |
756867bd JA |
1224 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1225 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1226 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1227 | ||
d2bbb8d9 | 1228 | fio_mutex_down(writeout_mutex); |
e3cedca7 | 1229 | if (td->ts.bw_log) { |
f484470b JA |
1230 | if (td->o.bw_log_file) { |
1231 | finish_log_named(td, td->ts.bw_log, | |
1232 | td->o.bw_log_file, "bw"); | |
1233 | } else | |
e3cedca7 JA |
1234 | finish_log(td, td->ts.bw_log, "bw"); |
1235 | } | |
02af0988 JA |
1236 | if (td->ts.lat_log) { |
1237 | if (td->o.lat_log_file) { | |
1238 | finish_log_named(td, td->ts.lat_log, | |
1239 | td->o.lat_log_file, "lat"); | |
1240 | } else | |
1241 | finish_log(td, td->ts.lat_log, "lat"); | |
1242 | } | |
e3cedca7 | 1243 | if (td->ts.slat_log) { |
f484470b JA |
1244 | if (td->o.lat_log_file) { |
1245 | finish_log_named(td, td->ts.slat_log, | |
1ca95738 | 1246 | td->o.lat_log_file, "slat"); |
f484470b | 1247 | } else |
e3cedca7 JA |
1248 | finish_log(td, td->ts.slat_log, "slat"); |
1249 | } | |
1250 | if (td->ts.clat_log) { | |
f484470b JA |
1251 | if (td->o.lat_log_file) { |
1252 | finish_log_named(td, td->ts.clat_log, | |
1253 | td->o.lat_log_file, "clat"); | |
1254 | } else | |
e3cedca7 JA |
1255 | finish_log(td, td->ts.clat_log, "clat"); |
1256 | } | |
d2bbb8d9 | 1257 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1258 | if (td->o.exec_postrun) |
1259 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1260 | |
1261 | if (exitall_on_terminate) | |
390c40e2 | 1262 | terminate_threads(td->groupid); |
ebac4655 JA |
1263 | |
1264 | err: | |
5bf13a5a | 1265 | if (td->error) |
4ceb30d4 | 1266 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1267 | td->verror); |
b4707ed7 JA |
1268 | |
1269 | if (td->o.verify_async) | |
1270 | verify_async_exit(td); | |
1271 | ||
24ffd2c2 | 1272 | close_and_free_files(td); |
2866c82d | 1273 | close_ioengine(td); |
ebac4655 | 1274 | cleanup_io_u(td); |
6adb38a1 | 1275 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1276 | |
d2ce18b5 JA |
1277 | if (td->o.cpumask_set) { |
1278 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1279 | ||
1280 | td_verror(td, ret, "fio_cpuset_exit"); | |
1281 | } | |
4e78e405 | 1282 | |
f29b25a3 JA |
1283 | /* |
1284 | * do this very late, it will log file closing as well | |
1285 | */ | |
1286 | if (td->o.write_iolog_file) | |
1287 | write_iolog_close(td); | |
1288 | ||
d23bb327 | 1289 | options_mem_free(td); |
ebac4655 | 1290 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1291 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1292 | } |
1293 | ||
906c8d75 JA |
1294 | /* |
1295 | * We cannot pass the td data into a forked process, so attach the td and | |
1296 | * pass it to the thread worker. | |
1297 | */ | |
a6418147 | 1298 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1299 | { |
1300 | struct thread_data *td; | |
a6418147 | 1301 | void *data, *ret; |
ebac4655 | 1302 | |
65258961 | 1303 | #ifndef __hpux |
ebac4655 JA |
1304 | data = shmat(shmid, NULL, 0); |
1305 | if (data == (void *) -1) { | |
a6418147 JA |
1306 | int __err = errno; |
1307 | ||
ebac4655 | 1308 | perror("shmat"); |
a6418147 | 1309 | return __err; |
ebac4655 | 1310 | } |
65258961 JA |
1311 | #else |
1312 | /* | |
1313 | * HP-UX inherits shm mappings? | |
1314 | */ | |
1315 | data = threads; | |
1316 | #endif | |
ebac4655 JA |
1317 | |
1318 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1319 | ret = thread_main(td); |
ebac4655 | 1320 | shmdt(data); |
43d76807 | 1321 | return (int) (unsigned long) ret; |
ebac4655 JA |
1322 | } |
1323 | ||
906c8d75 JA |
1324 | /* |
1325 | * Run over the job map and reap the threads that have exited, if any. | |
1326 | */ | |
ebac4655 JA |
1327 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1328 | { | |
34572e28 | 1329 | struct thread_data *td; |
1f809d15 | 1330 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1331 | |
1332 | /* | |
1333 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1334 | */ | |
1f809d15 | 1335 | realthreads = pending = cputhreads = 0; |
34572e28 | 1336 | for_each_td(td, i) { |
3707f45b | 1337 | int flags = 0; |
a2f77c9f | 1338 | |
84585003 JA |
1339 | /* |
1340 | * ->io_ops is NULL for a thread that has closed its | |
1341 | * io engine | |
1342 | */ | |
ba0fbe10 | 1343 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1344 | cputhreads++; |
1f809d15 JA |
1345 | else |
1346 | realthreads++; | |
b990b5c0 | 1347 | |
a1dedc1f AC |
1348 | if (!td->pid) { |
1349 | pending++; | |
1350 | continue; | |
1351 | } | |
1352 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1353 | continue; |
2dc1bbeb | 1354 | if (td->o.use_thread) { |
3707f45b JA |
1355 | if (td->runstate == TD_EXITED) { |
1356 | td_set_runstate(td, TD_REAPED); | |
1357 | goto reaped; | |
1358 | } | |
1359 | continue; | |
1360 | } | |
a2f77c9f JA |
1361 | |
1362 | flags = WNOHANG; | |
1363 | if (td->runstate == TD_EXITED) | |
1364 | flags = 0; | |
1365 | ||
1366 | /* | |
1367 | * check if someone quit or got killed in an unusual way | |
1368 | */ | |
1369 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1370 | if (ret < 0) { |
a2f77c9f | 1371 | if (errno == ECHILD) { |
5921e80c JA |
1372 | log_err("fio: pid=%d disappeared %d\n", |
1373 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1374 | td_set_runstate(td, TD_REAPED); |
1375 | goto reaped; | |
1376 | } | |
1377 | perror("waitpid"); | |
1378 | } else if (ret == td->pid) { | |
1379 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1380 | int sig = WTERMSIG(status); |
1381 | ||
03e20d68 | 1382 | if (sig != SIGTERM) |
5ec10eaa | 1383 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1384 | (int) td->pid, sig); |
fab6aa71 JA |
1385 | td_set_runstate(td, TD_REAPED); |
1386 | goto reaped; | |
1387 | } | |
a2f77c9f JA |
1388 | if (WIFEXITED(status)) { |
1389 | if (WEXITSTATUS(status) && !td->error) | |
1390 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1391 | |
a2f77c9f JA |
1392 | td_set_runstate(td, TD_REAPED); |
1393 | goto reaped; | |
a6418147 JA |
1394 | } |
1395 | } | |
ebac4655 | 1396 | |
a2f77c9f JA |
1397 | /* |
1398 | * thread is not dead, continue | |
1399 | */ | |
e48676ba | 1400 | pending++; |
a2f77c9f | 1401 | continue; |
fab6aa71 | 1402 | reaped: |
ebac4655 | 1403 | (*nr_running)--; |
581e7141 JA |
1404 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1405 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1406 | if (!td->pid) |
1407 | pending--; | |
a2f77c9f JA |
1408 | |
1409 | if (td->error) | |
1410 | exit_value++; | |
d3eeeabc JA |
1411 | |
1412 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1413 | } |
b990b5c0 | 1414 | |
1f809d15 | 1415 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1416 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1417 | } |
1418 | ||
be4ecfdf JA |
1419 | static void *gtod_thread_main(void *data) |
1420 | { | |
1421 | fio_mutex_up(startup_mutex); | |
1422 | ||
1423 | /* | |
1424 | * As long as we have jobs around, update the clock. It would be nice | |
1425 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1426 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1427 | * it - we don't want to lose precision. | |
1428 | */ | |
1429 | while (threads) { | |
1430 | fio_gtod_update(); | |
1431 | nop; | |
1432 | } | |
1433 | ||
1434 | return NULL; | |
1435 | } | |
1436 | ||
1437 | static int fio_start_gtod_thread(void) | |
1438 | { | |
304a47c7 | 1439 | pthread_attr_t attr; |
f42c153d JA |
1440 | int ret; |
1441 | ||
304a47c7 VA |
1442 | pthread_attr_init(&attr); |
1443 | pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); | |
1444 | ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); | |
1445 | pthread_attr_destroy(&attr); | |
f42c153d JA |
1446 | if (ret) { |
1447 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1448 | return 1; |
1449 | } | |
f42c153d JA |
1450 | |
1451 | ret = pthread_detach(gtod_thread); | |
1452 | if (ret) { | |
1453 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1454 | return 1; |
1455 | } | |
1456 | ||
29adda3c | 1457 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1458 | fio_mutex_down(startup_mutex); |
29adda3c | 1459 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1460 | return 0; |
1461 | } | |
1462 | ||
906c8d75 JA |
1463 | /* |
1464 | * Main function for kicking off and reaping jobs, as needed. | |
1465 | */ | |
ebac4655 JA |
1466 | static void run_threads(void) |
1467 | { | |
ebac4655 JA |
1468 | struct thread_data *td; |
1469 | unsigned long spent; | |
1470 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1471 | |
2f9ade3c JA |
1472 | if (fio_pin_memory()) |
1473 | return; | |
ebac4655 | 1474 | |
be4ecfdf JA |
1475 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1476 | return; | |
1477 | ||
c6ae0a5b | 1478 | if (!terse_output) { |
4ceb30d4 | 1479 | log_info("Starting "); |
9cedf167 | 1480 | if (nr_thread) |
4ceb30d4 | 1481 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1482 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1483 | if (nr_process) { |
1484 | if (nr_thread) | |
1485 | printf(" and "); | |
4ceb30d4 | 1486 | log_info("%d process%s", nr_process, |
5ec10eaa | 1487 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1488 | } |
4ceb30d4 | 1489 | log_info("\n"); |
c6ae0a5b JA |
1490 | fflush(stdout); |
1491 | } | |
c04f7ec3 | 1492 | |
697a606c | 1493 | set_sig_handlers(); |
4efa970e | 1494 | |
ebac4655 JA |
1495 | todo = thread_number; |
1496 | nr_running = 0; | |
1497 | nr_started = 0; | |
1498 | m_rate = t_rate = 0; | |
1499 | ||
34572e28 | 1500 | for_each_td(td, i) { |
263e529f | 1501 | print_status_init(td->thread_number - 1); |
ebac4655 | 1502 | |
2dc1bbeb | 1503 | if (!td->o.create_serialize) { |
380cf265 | 1504 | init_disk_util(td); |
ebac4655 | 1505 | continue; |
380cf265 | 1506 | } |
ebac4655 JA |
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 | } |
380cf265 JA |
1534 | |
1535 | init_disk_util(td); | |
ebac4655 JA |
1536 | } |
1537 | ||
a2f77c9f JA |
1538 | set_genesis_time(); |
1539 | ||
ebac4655 | 1540 | while (todo) { |
fca70358 | 1541 | struct thread_data *map[REAL_MAX_JOBS]; |
75154845 JA |
1542 | struct timeval this_start; |
1543 | int this_jobs = 0, left; | |
1544 | ||
ebac4655 JA |
1545 | /* |
1546 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1547 | */ | |
34572e28 | 1548 | for_each_td(td, i) { |
ebac4655 JA |
1549 | if (td->runstate != TD_NOT_CREATED) |
1550 | continue; | |
1551 | ||
1552 | /* | |
1553 | * never got a chance to start, killed by other | |
1554 | * thread for some reason | |
1555 | */ | |
1556 | if (td->terminate) { | |
1557 | todo--; | |
1558 | continue; | |
1559 | } | |
1560 | ||
2dc1bbeb | 1561 | if (td->o.start_delay) { |
263e529f | 1562 | spent = mtime_since_genesis(); |
ebac4655 | 1563 | |
2dc1bbeb | 1564 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1565 | continue; |
1566 | } | |
1567 | ||
6f88bcb0 JA |
1568 | if (td->o.stonewall && (nr_started || nr_running)) { |
1569 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1570 | td->o.name); | |
ebac4655 | 1571 | break; |
6f88bcb0 | 1572 | } |
ebac4655 | 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 | ||
ebac4655 JA |
1693 | int main(int argc, char *argv[]) |
1694 | { | |
29d610e1 JA |
1695 | long ps; |
1696 | ||
2e5cdb11 | 1697 | sinit(); |
d54a144d | 1698 | init_rand(&__fio_rand_state); |
2e5cdb11 | 1699 | |
dbe1125e JA |
1700 | /* |
1701 | * We need locale for number printing, if it isn't set then just | |
1702 | * go with the US format. | |
1703 | */ | |
1704 | if (!getenv("LC_NUMERIC")) | |
1705 | setlocale(LC_NUMERIC, "en_US"); | |
1706 | ||
29d610e1 JA |
1707 | ps = sysconf(_SC_PAGESIZE); |
1708 | if (ps < 0) { | |
1709 | log_err("Failed to get page size\n"); | |
1710 | return 1; | |
1711 | } | |
1712 | ||
cfc99db7 | 1713 | page_size = ps; |
29d610e1 JA |
1714 | page_mask = ps - 1; |
1715 | ||
74929ac2 JA |
1716 | fio_keywords_init(); |
1717 | ||
1718 | if (parse_options(argc, argv)) | |
1719 | return 1; | |
1720 | ||
07b3232d JA |
1721 | if (exec_profile && load_profile(exec_profile)) |
1722 | return 1; | |
1723 | ||
74929ac2 JA |
1724 | if (!thread_number) |
1725 | return 0; | |
1726 | ||
bb3884d8 JA |
1727 | if (write_bw_log) { |
1728 | setup_log(&agg_io_log[DDIR_READ]); | |
1729 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1730 | } | |
1731 | ||
cdd18ad8 | 1732 | startup_mutex = fio_mutex_init(0); |
03e20d68 BC |
1733 | if (startup_mutex == NULL) |
1734 | return 1; | |
d2bbb8d9 | 1735 | writeout_mutex = fio_mutex_init(1); |
03e20d68 BC |
1736 | if (writeout_mutex == NULL) |
1737 | return 1; | |
07739b57 | 1738 | |
a2f77c9f | 1739 | set_genesis_time(); |
7985701b | 1740 | create_disk_util_thread(); |
ebac4655 | 1741 | |
dae5341c JA |
1742 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1743 | INIT_FLIST_HEAD(cgroup_list); | |
1744 | ||
ebac4655 | 1745 | run_threads(); |
6ce15a32 | 1746 | |
bb3884d8 | 1747 | if (!fio_abort) { |
6ce15a32 | 1748 | show_run_stats(); |
bb3884d8 | 1749 | if (write_bw_log) { |
5ec10eaa JA |
1750 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1751 | __finish_log(agg_io_log[DDIR_WRITE], | |
1752 | "agg-write_bw.log"); | |
bb3884d8 JA |
1753 | } |
1754 | } | |
ebac4655 | 1755 | |
dae5341c JA |
1756 | cgroup_kill(cgroup_list); |
1757 | sfree(cgroup_list); | |
61eb313e | 1758 | sfree(cgroup_mnt); |
39f22027 | 1759 | |
cdd18ad8 | 1760 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1761 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1762 | return exit_value; |
ebac4655 | 1763 | } |