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