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> | |
ebac4655 JA |
27 | #include <signal.h> |
28 | #include <time.h> | |
dbe1125e | 29 | #include <locale.h> |
36167d82 | 30 | #include <assert.h> |
ebac4655 JA |
31 | #include <sys/stat.h> |
32 | #include <sys/wait.h> | |
33 | #include <sys/ipc.h> | |
34 | #include <sys/shm.h> | |
ebac4655 JA |
35 | #include <sys/mman.h> |
36 | ||
37 | #include "fio.h" | |
210f43b3 | 38 | #include "hash.h" |
2e5cdb11 | 39 | #include "smalloc.h" |
4f5af7b2 | 40 | #include "verify.h" |
a917a8b3 | 41 | #include "trim.h" |
7c9b1bce | 42 | #include "diskutil.h" |
a696fa2a | 43 | #include "cgroup.h" |
07b3232d | 44 | #include "profile.h" |
d54a144d | 45 | #include "lib/rand.h" |
ebac4655 | 46 | |
cfc99db7 JA |
47 | unsigned long page_mask; |
48 | unsigned long page_size; | |
d529ee19 JA |
49 | |
50 | #define PAGE_ALIGN(buf) \ | |
29d610e1 | 51 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) |
ebac4655 JA |
52 | |
53 | int groupid = 0; | |
54 | int thread_number = 0; | |
9cedf167 JA |
55 | int nr_process = 0; |
56 | int nr_thread = 0; | |
ebac4655 | 57 | int shm_id = 0; |
53cdc686 | 58 | int temp_stall_ts; |
d3eeeabc | 59 | unsigned long done_secs = 0; |
ebac4655 | 60 | |
cdd18ad8 | 61 | static struct fio_mutex *startup_mutex; |
d2bbb8d9 | 62 | static struct fio_mutex *writeout_mutex; |
6ce15a32 | 63 | static volatile int fio_abort; |
437c9b71 | 64 | static int exit_value; |
ccb0fa24 | 65 | static struct itimerval itimer; |
be4ecfdf | 66 | static pthread_t gtod_thread; |
dae5341c | 67 | static struct flist_head *cgroup_list; |
6adb38a1 | 68 | static char *cgroup_mnt; |
ebac4655 | 69 | |
bb3884d8 JA |
70 | struct io_log *agg_io_log[2]; |
71 | ||
ebac4655 | 72 | #define TERMINATE_ALL (-1) |
75154845 | 73 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 74 | |
b29ee5b3 | 75 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 76 | { |
d8fab1b6 JA |
77 | if (td->runstate == runstate) |
78 | return; | |
79 | ||
5921e80c JA |
80 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
81 | td->runstate, runstate); | |
6ce15a32 JA |
82 | td->runstate = runstate; |
83 | } | |
84 | ||
390c40e2 | 85 | static void terminate_threads(int group_id) |
ebac4655 | 86 | { |
34572e28 | 87 | struct thread_data *td; |
ebac4655 JA |
88 | int i; |
89 | ||
6f88bcb0 JA |
90 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
91 | ||
34572e28 | 92 | for_each_td(td, i) { |
ebac4655 | 93 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 94 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 95 | td->o.name, (int) td->pid); |
ad830ed7 JA |
96 | td->terminate = 1; |
97 | td->o.start_delay = 0; | |
98 | ||
7a93ab19 JA |
99 | /* |
100 | * if the thread is running, just let it exit | |
101 | */ | |
102 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 103 | kill(td->pid, SIGQUIT); |
f63f7f3b JA |
104 | else { |
105 | struct ioengine_ops *ops = td->io_ops; | |
106 | ||
107 | if (ops && (ops->flags & FIO_SIGQUIT)) | |
108 | kill(td->pid, SIGQUIT); | |
109 | } | |
ebac4655 JA |
110 | } |
111 | } | |
112 | } | |
113 | ||
ccb0fa24 JA |
114 | static void status_timer_arm(void) |
115 | { | |
116 | itimer.it_value.tv_sec = 0; | |
117 | itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; | |
118 | setitimer(ITIMER_REAL, &itimer, NULL); | |
119 | } | |
120 | ||
02444ad1 | 121 | static void sig_alrm(int fio_unused sig) |
ebac4655 | 122 | { |
697a606c | 123 | if (threads) { |
5ec10eaa | 124 | update_io_ticks(); |
5ec10eaa | 125 | print_thread_status(); |
ccb0fa24 | 126 | status_timer_arm(); |
697a606c JA |
127 | } |
128 | } | |
129 | ||
5f8f9726 JA |
130 | /* |
131 | * Happens on thread runs with ctrl-c, ignore our own SIGQUIT | |
132 | */ | |
133 | static void sig_quit(int sig) | |
134 | { | |
135 | } | |
136 | ||
697a606c JA |
137 | static void sig_int(int sig) |
138 | { | |
139 | if (threads) { | |
4ceb30d4 | 140 | log_info("\nfio: terminating on signal %d\n", sig); |
5ec10eaa JA |
141 | fflush(stdout); |
142 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
143 | } |
144 | } | |
145 | ||
697a606c JA |
146 | static void set_sig_handlers(void) |
147 | { | |
148 | struct sigaction act; | |
149 | ||
150 | memset(&act, 0, sizeof(act)); | |
151 | act.sa_handler = sig_alrm; | |
152 | act.sa_flags = SA_RESTART; | |
153 | sigaction(SIGALRM, &act, NULL); | |
154 | ||
155 | memset(&act, 0, sizeof(act)); | |
156 | act.sa_handler = sig_int; | |
157 | act.sa_flags = SA_RESTART; | |
158 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 159 | |
5f8f9726 JA |
160 | memset(&act, 0, sizeof(act)); |
161 | act.sa_handler = sig_quit; | |
162 | act.sa_flags = SA_RESTART; | |
163 | sigaction(SIGQUIT, &act, NULL); | |
697a606c JA |
164 | } |
165 | ||
906c8d75 JA |
166 | /* |
167 | * Check if we are above the minimum rate given. | |
168 | */ | |
581e7141 JA |
169 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
170 | enum td_ddir ddir) | |
ebac4655 | 171 | { |
0904200b | 172 | unsigned long long bytes = 0; |
4e991c23 | 173 | unsigned long iops = 0; |
ebac4655 JA |
174 | unsigned long spent; |
175 | unsigned long rate; | |
581e7141 JA |
176 | unsigned int ratemin = 0; |
177 | unsigned int rate_iops = 0; | |
178 | unsigned int rate_iops_min = 0; | |
ebac4655 | 179 | |
ff58fced JA |
180 | assert(ddir_rw(ddir)); |
181 | ||
d982d873 JA |
182 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
183 | return 0; | |
184 | ||
ebac4655 JA |
185 | /* |
186 | * allow a 2 second settle period in the beginning | |
187 | */ | |
188 | if (mtime_since(&td->start, now) < 2000) | |
189 | return 0; | |
190 | ||
581e7141 JA |
191 | iops += td->io_blocks[ddir]; |
192 | bytes += td->this_io_bytes[ddir]; | |
193 | ratemin += td->o.ratemin[ddir]; | |
194 | rate_iops += td->o.rate_iops[ddir]; | |
195 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 196 | |
ebac4655 JA |
197 | /* |
198 | * if rate blocks is set, sample is running | |
199 | */ | |
581e7141 JA |
200 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
201 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 202 | if (spent < td->o.ratecycle) |
ebac4655 JA |
203 | return 0; |
204 | ||
581e7141 | 205 | if (td->o.rate[ddir]) { |
4e991c23 JA |
206 | /* |
207 | * check bandwidth specified rate | |
208 | */ | |
581e7141 | 209 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 210 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 211 | ratemin); |
4e991c23 JA |
212 | return 1; |
213 | } else { | |
581e7141 JA |
214 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
215 | if (rate < ratemin || | |
216 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 217 | log_err("%s: min rate %u not met, got" |
b22989b9 | 218 | " %luKB/sec\n", td->o.name, |
581e7141 | 219 | ratemin, rate); |
4e991c23 JA |
220 | return 1; |
221 | } | |
222 | } | |
413dd459 | 223 | } else { |
4e991c23 JA |
224 | /* |
225 | * checks iops specified rate | |
226 | */ | |
581e7141 | 227 | if (iops < rate_iops) { |
5ec10eaa | 228 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 229 | td->o.name, rate_iops); |
413dd459 | 230 | return 1; |
4e991c23 | 231 | } else { |
581e7141 JA |
232 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
233 | if (rate < rate_iops_min || | |
234 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
235 | log_err("%s: min iops rate %u not met," |
236 | " got %lu\n", td->o.name, | |
581e7141 | 237 | rate_iops_min, rate); |
4e991c23 | 238 | } |
413dd459 | 239 | } |
ebac4655 JA |
240 | } |
241 | } | |
242 | ||
581e7141 JA |
243 | td->rate_bytes[ddir] = bytes; |
244 | td->rate_blocks[ddir] = iops; | |
245 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
246 | return 0; |
247 | } | |
248 | ||
581e7141 JA |
249 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
250 | unsigned long *bytes_done) | |
251 | { | |
252 | int ret = 0; | |
253 | ||
254 | if (bytes_done[0]) | |
255 | ret |= __check_min_rate(td, now, 0); | |
256 | if (bytes_done[1]) | |
257 | ret |= __check_min_rate(td, now, 1); | |
258 | ||
259 | return ret; | |
260 | } | |
261 | ||
ebac4655 JA |
262 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
263 | { | |
2dc1bbeb | 264 | if (!td->o.timeout) |
ebac4655 | 265 | return 0; |
2dc1bbeb | 266 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
267 | return 1; |
268 | ||
269 | return 0; | |
270 | } | |
271 | ||
906c8d75 JA |
272 | /* |
273 | * When job exits, we can cancel the in-flight IO if we are using async | |
274 | * io. Attempt to do so. | |
275 | */ | |
ebac4655 JA |
276 | static void cleanup_pending_aio(struct thread_data *td) |
277 | { | |
01743ee1 | 278 | struct flist_head *entry, *n; |
ebac4655 JA |
279 | struct io_u *io_u; |
280 | int r; | |
281 | ||
282 | /* | |
283 | * get immediately available events, if any | |
284 | */ | |
581e7141 | 285 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
286 | if (r < 0) |
287 | return; | |
ebac4655 JA |
288 | |
289 | /* | |
290 | * now cancel remaining active events | |
291 | */ | |
2866c82d | 292 | if (td->io_ops->cancel) { |
01743ee1 JA |
293 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
294 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 295 | |
0c6e7517 JA |
296 | /* |
297 | * if the io_u isn't in flight, then that generally | |
298 | * means someone leaked an io_u. complain but fix | |
299 | * it up, so we don't stall here. | |
300 | */ | |
301 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
302 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 303 | put_io_u(td, io_u); |
0c6e7517 JA |
304 | } else { |
305 | r = td->io_ops->cancel(td, io_u); | |
306 | if (!r) | |
307 | put_io_u(td, io_u); | |
308 | } | |
ebac4655 JA |
309 | } |
310 | } | |
311 | ||
97601024 | 312 | if (td->cur_depth) |
581e7141 | 313 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
314 | } |
315 | ||
858a3d47 JA |
316 | /* |
317 | * Helper to handle the final sync of a file. Works just like the normal | |
318 | * io path, just does everything sync. | |
319 | */ | |
320 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
321 | { | |
322 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
323 | int ret; |
324 | ||
325 | if (!io_u) | |
326 | return 1; | |
327 | ||
328 | io_u->ddir = DDIR_SYNC; | |
329 | io_u->file = f; | |
330 | ||
331 | if (td_io_prep(td, io_u)) { | |
332 | put_io_u(td, io_u); | |
333 | return 1; | |
334 | } | |
335 | ||
755200a3 | 336 | requeue: |
858a3d47 | 337 | ret = td_io_queue(td, io_u); |
36167d82 | 338 | if (ret < 0) { |
e1161c32 | 339 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 340 | put_io_u(td, io_u); |
858a3d47 | 341 | return 1; |
36167d82 | 342 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 343 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 344 | return 1; |
36167d82 JA |
345 | } else if (ret == FIO_Q_COMPLETED) { |
346 | if (io_u->error) { | |
e1161c32 | 347 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
348 | return 1; |
349 | } | |
858a3d47 | 350 | |
581e7141 | 351 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 352 | return 1; |
755200a3 JA |
353 | } else if (ret == FIO_Q_BUSY) { |
354 | if (td_io_commit(td)) | |
355 | return 1; | |
356 | goto requeue; | |
858a3d47 JA |
357 | } |
358 | ||
359 | return 0; | |
360 | } | |
361 | ||
993bf48b JA |
362 | static inline void update_tv_cache(struct thread_data *td) |
363 | { | |
364 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
365 | fio_gettime(&td->tv_cache, NULL); | |
366 | } | |
367 | ||
f2bba182 RR |
368 | static int break_on_this_error(struct thread_data *td, int *retptr) |
369 | { | |
370 | int ret = *retptr; | |
371 | ||
372 | if (ret < 0 || td->error) { | |
373 | int err; | |
374 | ||
1ec99eea JA |
375 | if (!td->o.continue_on_error) |
376 | return 1; | |
f2bba182 RR |
377 | |
378 | if (ret < 0) | |
379 | err = -ret; | |
380 | else | |
381 | err = td->error; | |
382 | ||
f2bba182 RR |
383 | if (td_non_fatal_error(err)) { |
384 | /* | |
385 | * Continue with the I/Os in case of | |
386 | * a non fatal error. | |
387 | */ | |
8a4f8a24 | 388 | update_error_count(td, err); |
f2bba182 RR |
389 | td_clear_error(td); |
390 | *retptr = 0; | |
391 | return 0; | |
8a4f8a24 RR |
392 | } else if (td->o.fill_device && err == ENOSPC) { |
393 | /* | |
394 | * We expect to hit this error if | |
395 | * fill_device option is set. | |
396 | */ | |
397 | td_clear_error(td); | |
398 | td->terminate = 1; | |
399 | return 1; | |
f2bba182 RR |
400 | } else { |
401 | /* | |
402 | * Stop the I/O in case of a fatal | |
403 | * error. | |
404 | */ | |
8a4f8a24 | 405 | update_error_count(td, err); |
f2bba182 RR |
406 | return 1; |
407 | } | |
408 | } | |
409 | ||
410 | return 0; | |
411 | } | |
412 | ||
906c8d75 | 413 | /* |
34403fb1 | 414 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
415 | * reads the blocks back in, and checks the crc/md5 of the data. |
416 | */ | |
1e97cce9 | 417 | static void do_verify(struct thread_data *td) |
ebac4655 | 418 | { |
53cdc686 | 419 | struct fio_file *f; |
36167d82 | 420 | struct io_u *io_u; |
af52b345 JA |
421 | int ret, min_events; |
422 | unsigned int i; | |
e5b401d4 | 423 | |
008618a8 JA |
424 | dprint(FD_VERIFY, "starting loop\n"); |
425 | ||
e5b401d4 JA |
426 | /* |
427 | * sync io first and invalidate cache, to make sure we really | |
428 | * read from disk. | |
429 | */ | |
430 | for_each_file(td, f, i) { | |
d6aed795 | 431 | if (!fio_file_open(f)) |
3d7b485f | 432 | continue; |
b2fdda43 JA |
433 | if (fio_io_sync(td, f)) |
434 | break; | |
435 | if (file_invalidate_cache(td, f)) | |
436 | break; | |
e5b401d4 | 437 | } |
ebac4655 | 438 | |
b2fdda43 JA |
439 | if (td->error) |
440 | return; | |
441 | ||
ebac4655 JA |
442 | td_set_runstate(td, TD_VERIFYING); |
443 | ||
36167d82 JA |
444 | io_u = NULL; |
445 | while (!td->terminate) { | |
4950421a | 446 | int ret2, full; |
5451792e | 447 | |
993bf48b JA |
448 | update_tv_cache(td); |
449 | ||
450 | if (runtime_exceeded(td, &td->tv_cache)) { | |
90a87d4b | 451 | td->terminate = 1; |
02bcaa8c | 452 | break; |
069c2918 | 453 | } |
02bcaa8c | 454 | |
dcb4830d RR |
455 | io_u = __get_io_u(td); |
456 | if (!io_u) | |
457 | break; | |
458 | ||
069c2918 JA |
459 | if (get_next_verify(td, io_u)) { |
460 | put_io_u(td, io_u); | |
ebac4655 | 461 | break; |
069c2918 | 462 | } |
ebac4655 | 463 | |
069c2918 JA |
464 | if (td_io_prep(td, io_u)) { |
465 | put_io_u(td, io_u); | |
53cdc686 | 466 | break; |
069c2918 | 467 | } |
d7762cf8 | 468 | |
e8462bd8 JA |
469 | if (td->o.verify_async) |
470 | io_u->end_io = verify_io_u_async; | |
471 | else | |
472 | io_u->end_io = verify_io_u; | |
53cdc686 | 473 | |
11786802 | 474 | ret = td_io_queue(td, io_u); |
36167d82 JA |
475 | switch (ret) { |
476 | case FIO_Q_COMPLETED: | |
f2bba182 | 477 | if (io_u->error) { |
22819ec2 | 478 | ret = -io_u->error; |
f2bba182 RR |
479 | clear_io_u(td, io_u); |
480 | } else if (io_u->resid) { | |
36167d82 | 481 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 482 | struct fio_file *f = io_u->file; |
ebac4655 | 483 | |
9e9d164e JA |
484 | /* |
485 | * zero read, fail | |
486 | */ | |
487 | if (!bytes) { | |
41d9f0fd | 488 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
489 | put_io_u(td, io_u); |
490 | break; | |
491 | } | |
8400d9b2 | 492 | |
36167d82 JA |
493 | io_u->xfer_buflen = io_u->resid; |
494 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
495 | io_u->offset += bytes; |
496 | ||
ff58fced JA |
497 | if (ddir_rw(io_u->ddir)) |
498 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 499 | |
d460eb31 | 500 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
501 | goto sync_done; |
502 | ||
11786802 JA |
503 | requeue_io_u(td, &io_u); |
504 | } else { | |
8400d9b2 | 505 | sync_done: |
581e7141 | 506 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
507 | if (ret < 0) |
508 | break; | |
36167d82 | 509 | } |
36167d82 JA |
510 | continue; |
511 | case FIO_Q_QUEUED: | |
512 | break; | |
755200a3 JA |
513 | case FIO_Q_BUSY: |
514 | requeue_io_u(td, &io_u); | |
5451792e JA |
515 | ret2 = td_io_commit(td); |
516 | if (ret2 < 0) | |
517 | ret = ret2; | |
755200a3 | 518 | break; |
36167d82 JA |
519 | default: |
520 | assert(ret < 0); | |
e1161c32 | 521 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
522 | break; |
523 | } | |
524 | ||
f2bba182 | 525 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
526 | break; |
527 | ||
ebac4655 | 528 | /* |
3af6ef39 JA |
529 | * if we can queue more, do so. but check if there are |
530 | * completed io_u's first. | |
ebac4655 | 531 | */ |
4950421a JA |
532 | full = queue_full(td) || ret == FIO_Q_BUSY; |
533 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
534 | min_events = min(td->o.iodepth_batch_complete, |
535 | td->cur_depth); | |
4950421a | 536 | if (full && !min_events) |
838bc709 | 537 | min_events = 1; |
e916b390 | 538 | |
4950421a JA |
539 | do { |
540 | /* | |
541 | * Reap required number of io units, if any, | |
542 | * and do the verification on them through | |
543 | * the callback handler | |
544 | */ | |
581e7141 | 545 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
546 | ret = -1; |
547 | break; | |
548 | } | |
549 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
550 | } | |
551 | if (ret < 0) | |
ebac4655 | 552 | break; |
36167d82 | 553 | } |
ebac4655 | 554 | |
c01c0395 JA |
555 | if (!td->error) { |
556 | min_events = td->cur_depth; | |
557 | ||
558 | if (min_events) | |
581e7141 | 559 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 560 | } else |
ebac4655 JA |
561 | cleanup_pending_aio(td); |
562 | ||
563 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
564 | |
565 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
566 | } |
567 | ||
32cd46a0 | 568 | /* |
906c8d75 | 569 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
570 | * and reaps them, checking for rate and errors along the way. |
571 | */ | |
ebac4655 JA |
572 | static void do_io(struct thread_data *td) |
573 | { | |
af52b345 JA |
574 | unsigned int i; |
575 | int ret = 0; | |
ebac4655 | 576 | |
b29ee5b3 JA |
577 | if (in_ramp_time(td)) |
578 | td_set_runstate(td, TD_RAMP); | |
579 | else | |
580 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 581 | |
457bf399 | 582 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
a917a8b3 | 583 | (!flist_empty(&td->trim_list)) || |
457bf399 | 584 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { |
97601024 | 585 | struct timeval comp_time; |
581e7141 | 586 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 587 | int min_evts = 0; |
ebac4655 | 588 | struct io_u *io_u; |
4950421a | 589 | int ret2, full; |
ebac4655 JA |
590 | |
591 | if (td->terminate) | |
592 | break; | |
593 | ||
993bf48b | 594 | update_tv_cache(td); |
97601024 | 595 | |
993bf48b | 596 | if (runtime_exceeded(td, &td->tv_cache)) { |
90a87d4b | 597 | td->terminate = 1; |
97601024 JA |
598 | break; |
599 | } | |
36167d82 | 600 | |
dcb4830d RR |
601 | io_u = get_io_u(td); |
602 | if (!io_u) | |
603 | break; | |
604 | ||
b67740d3 JA |
605 | /* |
606 | * Add verification end_io handler, if asked to verify | |
607 | * a previously written file. | |
608 | */ | |
715d2b3e JA |
609 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
610 | !td_rw(td)) { | |
e8462bd8 JA |
611 | if (td->o.verify_async) |
612 | io_u->end_io = verify_io_u_async; | |
613 | else | |
614 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 615 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
616 | } else if (in_ramp_time(td)) |
617 | td_set_runstate(td, TD_RAMP); | |
618 | else | |
d8fab1b6 | 619 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 620 | |
11786802 | 621 | ret = td_io_queue(td, io_u); |
36167d82 JA |
622 | switch (ret) { |
623 | case FIO_Q_COMPLETED: | |
f2bba182 | 624 | if (io_u->error) { |
5451792e | 625 | ret = -io_u->error; |
f2bba182 RR |
626 | clear_io_u(td, io_u); |
627 | } else if (io_u->resid) { | |
36167d82 | 628 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 629 | struct fio_file *f = io_u->file; |
36167d82 | 630 | |
9e9d164e JA |
631 | /* |
632 | * zero read, fail | |
633 | */ | |
634 | if (!bytes) { | |
41d9f0fd | 635 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
636 | put_io_u(td, io_u); |
637 | break; | |
638 | } | |
639 | ||
cec6b55d | 640 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 641 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
642 | io_u->offset += bytes; |
643 | ||
ff58fced JA |
644 | if (ddir_rw(io_u->ddir)) |
645 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 646 | |
d460eb31 | 647 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
648 | goto sync_done; |
649 | ||
11786802 JA |
650 | requeue_io_u(td, &io_u); |
651 | } else { | |
5a7c5680 | 652 | sync_done: |
581e7141 JA |
653 | if (__should_check_rate(td, 0) || |
654 | __should_check_rate(td, 1)) | |
993bf48b JA |
655 | fio_gettime(&comp_time, NULL); |
656 | ||
581e7141 JA |
657 | ret = io_u_sync_complete(td, io_u, bytes_done); |
658 | if (ret < 0) | |
659 | break; | |
cec6b55d | 660 | } |
36167d82 JA |
661 | break; |
662 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
663 | /* |
664 | * if the engine doesn't have a commit hook, | |
665 | * the io_u is really queued. if it does have such | |
666 | * a hook, it has to call io_u_queued() itself. | |
667 | */ | |
668 | if (td->io_ops->commit == NULL) | |
669 | io_u_queued(td, io_u); | |
36167d82 | 670 | break; |
755200a3 JA |
671 | case FIO_Q_BUSY: |
672 | requeue_io_u(td, &io_u); | |
5451792e JA |
673 | ret2 = td_io_commit(td); |
674 | if (ret2 < 0) | |
675 | ret = ret2; | |
755200a3 | 676 | break; |
36167d82 JA |
677 | default: |
678 | assert(ret < 0); | |
679 | put_io_u(td, io_u); | |
680 | break; | |
ebac4655 JA |
681 | } |
682 | ||
f2bba182 | 683 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
684 | break; |
685 | ||
97601024 JA |
686 | /* |
687 | * See if we need to complete some commands | |
688 | */ | |
4950421a JA |
689 | full = queue_full(td) || ret == FIO_Q_BUSY; |
690 | if (full || !td->o.iodepth_batch_complete) { | |
6a96276c RR |
691 | min_evts = min(td->o.iodepth_batch_complete, |
692 | td->cur_depth); | |
4950421a | 693 | if (full && !min_evts) |
36167d82 | 694 | min_evts = 1; |
4950421a | 695 | |
581e7141 JA |
696 | if (__should_check_rate(td, 0) || |
697 | __should_check_rate(td, 1)) | |
993bf48b | 698 | fio_gettime(&comp_time, NULL); |
4950421a JA |
699 | |
700 | do { | |
581e7141 JA |
701 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
702 | if (ret < 0) | |
4950421a JA |
703 | break; |
704 | ||
4950421a | 705 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
706 | } |
707 | ||
4950421a JA |
708 | if (ret < 0) |
709 | break; | |
581e7141 | 710 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
711 | continue; |
712 | ||
581e7141 JA |
713 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
714 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
715 | if (exitall_on_terminate) |
716 | terminate_threads(td->groupid); | |
717 | td_verror(td, EIO, "check_min_rate"); | |
718 | break; | |
719 | } | |
ebac4655 JA |
720 | } |
721 | ||
2dc1bbeb | 722 | if (td->o.thinktime) { |
9c1f7434 JA |
723 | unsigned long long b; |
724 | ||
725 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 726 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
727 | int left; |
728 | ||
2dc1bbeb | 729 | if (td->o.thinktime_spin) |
6dd6f2cd | 730 | usec_spin(td->o.thinktime_spin); |
48097d5c | 731 | |
2dc1bbeb | 732 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
733 | if (left) |
734 | usec_sleep(td, left); | |
735 | } | |
9c1f7434 | 736 | } |
ebac4655 JA |
737 | } |
738 | ||
d708ac5e JA |
739 | if (td->trim_entries) |
740 | printf("trim entries %ld\n", td->trim_entries); | |
a917a8b3 | 741 | |
aa31f1f1 SL |
742 | if (td->o.fill_device && td->error == ENOSPC) { |
743 | td->error = 0; | |
744 | td->terminate = 1; | |
745 | } | |
4d2413c6 | 746 | if (!td->error) { |
3d7c391d JA |
747 | struct fio_file *f; |
748 | ||
c01c0395 JA |
749 | i = td->cur_depth; |
750 | if (i) | |
581e7141 | 751 | ret = io_u_queued_complete(td, i, NULL); |
ebac4655 | 752 | |
2dc1bbeb | 753 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 754 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
755 | |
756 | for_each_file(td, f, i) { | |
d6aed795 | 757 | if (!fio_file_open(f)) |
3d7b485f | 758 | continue; |
858a3d47 | 759 | fio_io_sync(td, f); |
3d7b485f | 760 | } |
84585003 | 761 | } |
c01c0395 JA |
762 | } else |
763 | cleanup_pending_aio(td); | |
2ba1c290 JA |
764 | |
765 | /* | |
766 | * stop job if we failed doing any IO | |
767 | */ | |
768 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
769 | td->done = 1; | |
ebac4655 JA |
770 | } |
771 | ||
ebac4655 JA |
772 | static void cleanup_io_u(struct thread_data *td) |
773 | { | |
01743ee1 | 774 | struct flist_head *entry, *n; |
ebac4655 JA |
775 | struct io_u *io_u; |
776 | ||
01743ee1 JA |
777 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
778 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 779 | |
01743ee1 | 780 | flist_del(&io_u->list); |
ebac4655 JA |
781 | free(io_u); |
782 | } | |
783 | ||
2f9ade3c | 784 | free_io_mem(td); |
ebac4655 JA |
785 | } |
786 | ||
787 | static int init_io_u(struct thread_data *td) | |
788 | { | |
789 | struct io_u *io_u; | |
a00735e6 | 790 | unsigned int max_bs; |
eb7ccf38 | 791 | int cl_align, i, max_units; |
ebac4655 JA |
792 | char *p; |
793 | ||
1e3c09af | 794 | max_units = td->o.iodepth; |
2dc1bbeb | 795 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
796 | td->orig_buffer_size = (unsigned long long) max_bs |
797 | * (unsigned long long) max_units; | |
798 | ||
799 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
800 | unsigned long bs; | |
74b025b0 | 801 | |
5ec10eaa JA |
802 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
803 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
804 | } | |
ebac4655 | 805 | |
c3990645 JA |
806 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
807 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
808 | return 1; | |
809 | } | |
810 | ||
2f9ade3c JA |
811 | if (allocate_io_mem(td)) |
812 | return 1; | |
ebac4655 | 813 | |
1c04bf78 JA |
814 | if (td->o.odirect || td->o.mem_align || |
815 | (td->io_ops->flags & FIO_RAWIO)) | |
d529ee19 | 816 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
817 | else |
818 | p = td->orig_buffer; | |
819 | ||
eb7ccf38 JA |
820 | cl_align = os_cache_line_size(); |
821 | ||
ebac4655 | 822 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
823 | void *ptr; |
824 | ||
db1cd90b JA |
825 | if (td->terminate) |
826 | return 1; | |
eb7ccf38 JA |
827 | |
828 | if (posix_memalign(&ptr, cl_align, sizeof(*io_u))) { | |
829 | log_err("fio: posix_memalign=%s\n", strerror(errno)); | |
830 | break; | |
831 | } | |
832 | ||
833 | io_u = ptr; | |
ebac4655 | 834 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 835 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 836 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 837 | |
b4c5e1ac JA |
838 | if (!(td->io_ops->flags & FIO_NOIO)) { |
839 | io_u->buf = p + max_bs * i; | |
d529ee19 | 840 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 841 | |
5973cafb JA |
842 | if (td_write(td) && !td->o.refill_buffers) |
843 | io_u_fill_buffer(td, io_u, max_bs); | |
cbe8d756 RR |
844 | else if (td_write(td) && td->o.verify_pattern_bytes) { |
845 | /* | |
846 | * Fill the buffer with the pattern if we are | |
847 | * going to be doing writes. | |
848 | */ | |
849 | fill_pattern(td, io_u->buf, max_bs, io_u); | |
850 | } | |
b4c5e1ac | 851 | } |
6b9cea23 | 852 | |
b1ff3403 | 853 | io_u->index = i; |
0c6e7517 | 854 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 855 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
856 | } |
857 | ||
858 | return 0; | |
859 | } | |
860 | ||
da86774e JA |
861 | static int switch_ioscheduler(struct thread_data *td) |
862 | { | |
863 | char tmp[256], tmp2[128]; | |
864 | FILE *f; | |
865 | int ret; | |
866 | ||
ba0fbe10 | 867 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
868 | return 0; |
869 | ||
da86774e JA |
870 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
871 | ||
872 | f = fopen(tmp, "r+"); | |
873 | if (!f) { | |
cbf5c1d7 | 874 | if (errno == ENOENT) { |
5ec10eaa JA |
875 | log_err("fio: os or kernel doesn't support IO scheduler" |
876 | " switching\n"); | |
cbf5c1d7 JA |
877 | return 0; |
878 | } | |
879 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
880 | return 1; |
881 | } | |
882 | ||
883 | /* | |
884 | * Set io scheduler. | |
885 | */ | |
2dc1bbeb | 886 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 887 | if (ferror(f) || ret != 1) { |
e1161c32 | 888 | td_verror(td, errno, "fwrite"); |
da86774e JA |
889 | fclose(f); |
890 | return 1; | |
891 | } | |
892 | ||
893 | rewind(f); | |
894 | ||
895 | /* | |
896 | * Read back and check that the selected scheduler is now the default. | |
897 | */ | |
898 | ret = fread(tmp, 1, sizeof(tmp), f); | |
899 | if (ferror(f) || ret < 0) { | |
e1161c32 | 900 | td_verror(td, errno, "fread"); |
da86774e JA |
901 | fclose(f); |
902 | return 1; | |
903 | } | |
904 | ||
2dc1bbeb | 905 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 906 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 907 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 908 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
909 | fclose(f); |
910 | return 1; | |
911 | } | |
912 | ||
913 | fclose(f); | |
914 | return 0; | |
915 | } | |
916 | ||
492158cf JA |
917 | static int keep_running(struct thread_data *td) |
918 | { | |
919 | unsigned long long io_done; | |
920 | ||
20e354ef JA |
921 | if (td->done) |
922 | return 0; | |
492158cf JA |
923 | if (td->o.time_based) |
924 | return 1; | |
925 | if (td->o.loops) { | |
926 | td->o.loops--; | |
927 | return 1; | |
928 | } | |
929 | ||
5ec10eaa JA |
930 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
931 | + td->io_skip_bytes; | |
492158cf JA |
932 | if (io_done < td->o.size) |
933 | return 1; | |
934 | ||
935 | return 0; | |
936 | } | |
937 | ||
b29ee5b3 | 938 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 939 | { |
756867bd | 940 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 941 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 942 | td->zone_bytes = 0; |
581e7141 JA |
943 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
944 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 945 | |
c1324df1 JA |
946 | td->last_was_sync = 0; |
947 | ||
2ba1c290 JA |
948 | /* |
949 | * reset file done count if we are to start over | |
950 | */ | |
951 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 952 | td->nr_done_files = 0; |
5bfc35d7 JA |
953 | |
954 | /* | |
955 | * Set the same seed to get repeatable runs | |
956 | */ | |
957 | td_fill_rand_seeds(td); | |
b29ee5b3 JA |
958 | } |
959 | ||
960 | void reset_all_stats(struct thread_data *td) | |
961 | { | |
962 | struct timeval tv; | |
963 | int i; | |
964 | ||
965 | reset_io_counters(td); | |
966 | ||
967 | for (i = 0; i < 2; i++) { | |
968 | td->io_bytes[i] = 0; | |
969 | td->io_blocks[i] = 0; | |
970 | td->io_issues[i] = 0; | |
971 | td->ts.total_io_u[i] = 0; | |
972 | } | |
973 | ||
974 | fio_gettime(&tv, NULL); | |
975 | memcpy(&td->epoch, &tv, sizeof(tv)); | |
976 | memcpy(&td->start, &tv, sizeof(tv)); | |
977 | } | |
978 | ||
df9c26b1 | 979 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
980 | { |
981 | struct fio_file *f; | |
982 | unsigned int i; | |
b29ee5b3 JA |
983 | |
984 | reset_io_counters(td); | |
281f9b63 | 985 | |
24ffd2c2 | 986 | close_files(td); |
df9c26b1 | 987 | for_each_file(td, f, i) |
d6aed795 | 988 | fio_file_clear_done(f); |
ebac4655 JA |
989 | } |
990 | ||
398e8c4d JA |
991 | static int exec_string(const char *string) |
992 | { | |
993 | int ret, newlen = strlen(string) + 1 + 8; | |
994 | char *str; | |
995 | ||
996 | str = malloc(newlen); | |
997 | sprintf(str, "sh -c %s", string); | |
998 | ||
999 | ret = system(str); | |
1000 | if (ret == -1) | |
1001 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1002 | ||
1003 | free(str); | |
1004 | return ret; | |
1005 | } | |
1006 | ||
906c8d75 JA |
1007 | /* |
1008 | * Entry point for the thread based jobs. The process based jobs end up | |
1009 | * here as well, after a little setup. | |
1010 | */ | |
ebac4655 JA |
1011 | static void *thread_main(void *data) |
1012 | { | |
10927316 | 1013 | unsigned long long runtime[2], elapsed; |
ebac4655 | 1014 | struct thread_data *td = data; |
e8462bd8 | 1015 | pthread_condattr_t attr; |
a978ba68 | 1016 | int clear_state; |
ebac4655 | 1017 | |
2dc1bbeb | 1018 | if (!td->o.use_thread) |
ebac4655 JA |
1019 | setsid(); |
1020 | ||
1021 | td->pid = getpid(); | |
1022 | ||
5921e80c | 1023 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1024 | |
01743ee1 JA |
1025 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1026 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1027 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1028 | INIT_FLIST_HEAD(&td->io_log_list); | |
1029 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 | 1030 | INIT_FLIST_HEAD(&td->verify_list); |
0d29de83 | 1031 | INIT_FLIST_HEAD(&td->trim_list); |
e8462bd8 | 1032 | pthread_mutex_init(&td->io_u_lock, NULL); |
bb5d7d0b | 1033 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1034 | |
e8462bd8 JA |
1035 | pthread_condattr_init(&attr); |
1036 | pthread_cond_init(&td->verify_cond, &attr); | |
1037 | pthread_cond_init(&td->free_cond, &attr); | |
1038 | ||
db1cd90b | 1039 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1040 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1041 | fio_mutex_up(startup_mutex); |
29adda3c | 1042 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1043 | fio_mutex_down(td->mutex); |
29adda3c | 1044 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1045 | |
1046 | /* | |
cdd18ad8 | 1047 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1048 | * eating a file descriptor |
1049 | */ | |
cdd18ad8 | 1050 | fio_mutex_remove(td->mutex); |
db1cd90b | 1051 | |
e0b0d892 JA |
1052 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1053 | td_verror(td, errno, "setuid"); | |
1054 | goto err; | |
1055 | } | |
1056 | if (td->o.gid != -1U && setgid(td->o.gid)) { | |
1057 | td_verror(td, errno, "setgid"); | |
1058 | goto err; | |
1059 | } | |
1060 | ||
d84f8d49 JA |
1061 | /* |
1062 | * May alter parameters that init_io_u() will use, so we need to | |
1063 | * do this first. | |
1064 | */ | |
1065 | if (init_iolog(td)) | |
1066 | goto err; | |
1067 | ||
ebac4655 | 1068 | if (init_io_u(td)) |
db1cd90b | 1069 | goto err; |
ebac4655 | 1070 | |
e8462bd8 JA |
1071 | if (td->o.verify_async && verify_async_init(td)) |
1072 | goto err; | |
1073 | ||
1074 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
e1161c32 | 1075 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 1076 | goto err; |
ebac4655 JA |
1077 | } |
1078 | ||
f484470b JA |
1079 | /* |
1080 | * If we have a gettimeofday() thread, make sure we exclude that | |
1081 | * thread from this job | |
1082 | */ | |
be4ecfdf | 1083 | if (td->o.gtod_cpu) { |
be4ecfdf | 1084 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); |
e8462bd8 | 1085 | if (fio_setaffinity(td->pid, td->o.cpumask) == -1) { |
be4ecfdf JA |
1086 | td_verror(td, errno, "cpu_set_affinity"); |
1087 | goto err; | |
1088 | } | |
1089 | } | |
1090 | ||
ac684785 | 1091 | if (td->ioprio_set) { |
ebac4655 | 1092 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1093 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1094 | goto err; |
ebac4655 JA |
1095 | } |
1096 | } | |
1097 | ||
6adb38a1 | 1098 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1099 | goto err; |
1100 | ||
2dc1bbeb | 1101 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1102 | td_verror(td, errno, "nice"); |
db1cd90b | 1103 | goto err; |
b6f4d880 JA |
1104 | } |
1105 | ||
2dc1bbeb | 1106 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1107 | goto err; |
37f56873 | 1108 | |
2dc1bbeb | 1109 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1110 | goto err; |
1111 | ||
7d6c5283 JA |
1112 | if (td_io_init(td)) |
1113 | goto err; | |
1114 | ||
68727076 JA |
1115 | if (init_random_map(td)) |
1116 | goto err; | |
1117 | ||
2dc1bbeb | 1118 | if (td->o.exec_prerun) { |
398e8c4d | 1119 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1120 | goto err; |
1121 | } | |
4e0ba8af | 1122 | |
afad68f7 ZY |
1123 | if (td->o.pre_read) { |
1124 | if (pre_read_files(td) < 0) | |
1125 | goto err; | |
1126 | } | |
1127 | ||
69008999 | 1128 | fio_gettime(&td->epoch, NULL); |
756867bd | 1129 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
1130 | |
1131 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 1132 | clear_state = 0; |
492158cf | 1133 | while (keep_running(td)) { |
02bcaa8c | 1134 | fio_gettime(&td->start, NULL); |
c5d3d790 JA |
1135 | memcpy(&td->ts.stat_sample_time[0], &td->start, |
1136 | sizeof(td->start)); | |
1137 | memcpy(&td->ts.stat_sample_time[1], &td->start, | |
1138 | sizeof(td->start)); | |
993bf48b | 1139 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1140 | |
cf2dd8d6 | 1141 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
5ec10eaa JA |
1142 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
1143 | sizeof(td->lastrate)); | |
ebac4655 | 1144 | |
df9c26b1 JA |
1145 | if (clear_state) |
1146 | clear_io_state(td); | |
a978ba68 | 1147 | |
ebac4655 JA |
1148 | prune_io_piece_log(td); |
1149 | ||
ba0fbe10 | 1150 | do_io(td); |
ebac4655 | 1151 | |
a978ba68 JA |
1152 | clear_state = 1; |
1153 | ||
38d77cae | 1154 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1155 | elapsed = utime_since_now(&td->start); |
38d77cae JA |
1156 | runtime[DDIR_READ] += elapsed; |
1157 | } | |
1158 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1159 | elapsed = utime_since_now(&td->start); |
38d77cae JA |
1160 | runtime[DDIR_WRITE] += elapsed; |
1161 | } | |
5ec10eaa | 1162 | |
ebac4655 JA |
1163 | if (td->error || td->terminate) |
1164 | break; | |
1165 | ||
e84c73a8 SL |
1166 | if (!td->o.do_verify || |
1167 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1168 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1169 | continue; |
1170 | ||
df9c26b1 | 1171 | clear_io_state(td); |
a978ba68 | 1172 | |
02bcaa8c | 1173 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1174 | |
1175 | do_verify(td); | |
1176 | ||
69008999 | 1177 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1178 | |
1179 | if (td->error || td->terminate) | |
1180 | break; | |
1181 | } | |
1182 | ||
36dff966 | 1183 | update_rusage_stat(td); |
47f0cc48 YHJT |
1184 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
1185 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
1186 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1187 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1188 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1189 | ||
d2bbb8d9 | 1190 | fio_mutex_down(writeout_mutex); |
e3cedca7 | 1191 | if (td->ts.bw_log) { |
f484470b JA |
1192 | if (td->o.bw_log_file) { |
1193 | finish_log_named(td, td->ts.bw_log, | |
1194 | td->o.bw_log_file, "bw"); | |
1195 | } else | |
e3cedca7 JA |
1196 | finish_log(td, td->ts.bw_log, "bw"); |
1197 | } | |
02af0988 JA |
1198 | if (td->ts.lat_log) { |
1199 | if (td->o.lat_log_file) { | |
1200 | finish_log_named(td, td->ts.lat_log, | |
1201 | td->o.lat_log_file, "lat"); | |
1202 | } else | |
1203 | finish_log(td, td->ts.lat_log, "lat"); | |
1204 | } | |
e3cedca7 | 1205 | if (td->ts.slat_log) { |
f484470b JA |
1206 | if (td->o.lat_log_file) { |
1207 | finish_log_named(td, td->ts.slat_log, | |
1ca95738 | 1208 | td->o.lat_log_file, "slat"); |
f484470b | 1209 | } else |
e3cedca7 JA |
1210 | finish_log(td, td->ts.slat_log, "slat"); |
1211 | } | |
1212 | if (td->ts.clat_log) { | |
f484470b JA |
1213 | if (td->o.lat_log_file) { |
1214 | finish_log_named(td, td->ts.clat_log, | |
1215 | td->o.lat_log_file, "clat"); | |
1216 | } else | |
e3cedca7 JA |
1217 | finish_log(td, td->ts.clat_log, "clat"); |
1218 | } | |
d2bbb8d9 | 1219 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1220 | if (td->o.exec_postrun) |
1221 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1222 | |
1223 | if (exitall_on_terminate) | |
390c40e2 | 1224 | terminate_threads(td->groupid); |
ebac4655 JA |
1225 | |
1226 | err: | |
5bf13a5a | 1227 | if (td->error) |
4ceb30d4 | 1228 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1229 | td->verror); |
b4707ed7 JA |
1230 | |
1231 | if (td->o.verify_async) | |
1232 | verify_async_exit(td); | |
1233 | ||
24ffd2c2 | 1234 | close_and_free_files(td); |
2866c82d | 1235 | close_ioengine(td); |
ebac4655 | 1236 | cleanup_io_u(td); |
6adb38a1 | 1237 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1238 | |
d2ce18b5 JA |
1239 | if (td->o.cpumask_set) { |
1240 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1241 | ||
1242 | td_verror(td, ret, "fio_cpuset_exit"); | |
1243 | } | |
4e78e405 | 1244 | |
f29b25a3 JA |
1245 | /* |
1246 | * do this very late, it will log file closing as well | |
1247 | */ | |
1248 | if (td->o.write_iolog_file) | |
1249 | write_iolog_close(td); | |
1250 | ||
d23bb327 | 1251 | options_mem_free(td); |
ebac4655 | 1252 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1253 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1254 | } |
1255 | ||
906c8d75 JA |
1256 | /* |
1257 | * We cannot pass the td data into a forked process, so attach the td and | |
1258 | * pass it to the thread worker. | |
1259 | */ | |
a6418147 | 1260 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1261 | { |
1262 | struct thread_data *td; | |
a6418147 | 1263 | void *data, *ret; |
ebac4655 JA |
1264 | |
1265 | data = shmat(shmid, NULL, 0); | |
1266 | if (data == (void *) -1) { | |
a6418147 JA |
1267 | int __err = errno; |
1268 | ||
ebac4655 | 1269 | perror("shmat"); |
a6418147 | 1270 | return __err; |
ebac4655 JA |
1271 | } |
1272 | ||
1273 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1274 | ret = thread_main(td); |
ebac4655 | 1275 | shmdt(data); |
43d76807 | 1276 | return (int) (unsigned long) ret; |
ebac4655 JA |
1277 | } |
1278 | ||
906c8d75 JA |
1279 | /* |
1280 | * Run over the job map and reap the threads that have exited, if any. | |
1281 | */ | |
ebac4655 JA |
1282 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1283 | { | |
34572e28 | 1284 | struct thread_data *td; |
1f809d15 | 1285 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1286 | |
1287 | /* | |
1288 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1289 | */ | |
1f809d15 | 1290 | realthreads = pending = cputhreads = 0; |
34572e28 | 1291 | for_each_td(td, i) { |
3707f45b | 1292 | int flags = 0; |
a2f77c9f | 1293 | |
84585003 JA |
1294 | /* |
1295 | * ->io_ops is NULL for a thread that has closed its | |
1296 | * io engine | |
1297 | */ | |
ba0fbe10 | 1298 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1299 | cputhreads++; |
1f809d15 JA |
1300 | else |
1301 | realthreads++; | |
b990b5c0 | 1302 | |
a1dedc1f AC |
1303 | if (!td->pid) { |
1304 | pending++; | |
1305 | continue; | |
1306 | } | |
1307 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1308 | continue; |
2dc1bbeb | 1309 | if (td->o.use_thread) { |
3707f45b JA |
1310 | if (td->runstate == TD_EXITED) { |
1311 | td_set_runstate(td, TD_REAPED); | |
1312 | goto reaped; | |
1313 | } | |
1314 | continue; | |
1315 | } | |
a2f77c9f JA |
1316 | |
1317 | flags = WNOHANG; | |
1318 | if (td->runstate == TD_EXITED) | |
1319 | flags = 0; | |
1320 | ||
1321 | /* | |
1322 | * check if someone quit or got killed in an unusual way | |
1323 | */ | |
1324 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1325 | if (ret < 0) { |
a2f77c9f | 1326 | if (errno == ECHILD) { |
5921e80c JA |
1327 | log_err("fio: pid=%d disappeared %d\n", |
1328 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1329 | td_set_runstate(td, TD_REAPED); |
1330 | goto reaped; | |
1331 | } | |
1332 | perror("waitpid"); | |
1333 | } else if (ret == td->pid) { | |
1334 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1335 | int sig = WTERMSIG(status); |
1336 | ||
d9244941 | 1337 | if (sig != SIGQUIT) |
5ec10eaa | 1338 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1339 | (int) td->pid, sig); |
fab6aa71 JA |
1340 | td_set_runstate(td, TD_REAPED); |
1341 | goto reaped; | |
1342 | } | |
a2f77c9f JA |
1343 | if (WIFEXITED(status)) { |
1344 | if (WEXITSTATUS(status) && !td->error) | |
1345 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1346 | |
a2f77c9f JA |
1347 | td_set_runstate(td, TD_REAPED); |
1348 | goto reaped; | |
a6418147 JA |
1349 | } |
1350 | } | |
ebac4655 | 1351 | |
a2f77c9f JA |
1352 | /* |
1353 | * thread is not dead, continue | |
1354 | */ | |
e48676ba | 1355 | pending++; |
a2f77c9f | 1356 | continue; |
fab6aa71 | 1357 | reaped: |
ebac4655 | 1358 | (*nr_running)--; |
581e7141 JA |
1359 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1360 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1361 | if (!td->pid) |
1362 | pending--; | |
a2f77c9f JA |
1363 | |
1364 | if (td->error) | |
1365 | exit_value++; | |
d3eeeabc JA |
1366 | |
1367 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1368 | } |
b990b5c0 | 1369 | |
1f809d15 | 1370 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1371 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1372 | } |
1373 | ||
be4ecfdf JA |
1374 | static void *gtod_thread_main(void *data) |
1375 | { | |
1376 | fio_mutex_up(startup_mutex); | |
1377 | ||
1378 | /* | |
1379 | * As long as we have jobs around, update the clock. It would be nice | |
1380 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1381 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1382 | * it - we don't want to lose precision. | |
1383 | */ | |
1384 | while (threads) { | |
1385 | fio_gtod_update(); | |
1386 | nop; | |
1387 | } | |
1388 | ||
1389 | return NULL; | |
1390 | } | |
1391 | ||
1392 | static int fio_start_gtod_thread(void) | |
1393 | { | |
304a47c7 | 1394 | pthread_attr_t attr; |
f42c153d JA |
1395 | int ret; |
1396 | ||
304a47c7 VA |
1397 | pthread_attr_init(&attr); |
1398 | pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); | |
1399 | ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); | |
1400 | pthread_attr_destroy(&attr); | |
f42c153d JA |
1401 | if (ret) { |
1402 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1403 | return 1; |
1404 | } | |
f42c153d JA |
1405 | |
1406 | ret = pthread_detach(gtod_thread); | |
1407 | if (ret) { | |
1408 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1409 | return 1; |
1410 | } | |
1411 | ||
29adda3c | 1412 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1413 | fio_mutex_down(startup_mutex); |
29adda3c | 1414 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1415 | return 0; |
1416 | } | |
1417 | ||
906c8d75 JA |
1418 | /* |
1419 | * Main function for kicking off and reaping jobs, as needed. | |
1420 | */ | |
ebac4655 JA |
1421 | static void run_threads(void) |
1422 | { | |
ebac4655 JA |
1423 | struct thread_data *td; |
1424 | unsigned long spent; | |
1425 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1426 | |
2f9ade3c JA |
1427 | if (fio_pin_memory()) |
1428 | return; | |
ebac4655 | 1429 | |
be4ecfdf JA |
1430 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1431 | return; | |
1432 | ||
c6ae0a5b | 1433 | if (!terse_output) { |
4ceb30d4 | 1434 | log_info("Starting "); |
9cedf167 | 1435 | if (nr_thread) |
4ceb30d4 | 1436 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1437 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1438 | if (nr_process) { |
1439 | if (nr_thread) | |
1440 | printf(" and "); | |
4ceb30d4 | 1441 | log_info("%d process%s", nr_process, |
5ec10eaa | 1442 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1443 | } |
4ceb30d4 | 1444 | log_info("\n"); |
c6ae0a5b JA |
1445 | fflush(stdout); |
1446 | } | |
c04f7ec3 | 1447 | |
697a606c | 1448 | set_sig_handlers(); |
4efa970e | 1449 | |
ebac4655 JA |
1450 | todo = thread_number; |
1451 | nr_running = 0; | |
1452 | nr_started = 0; | |
1453 | m_rate = t_rate = 0; | |
1454 | ||
34572e28 | 1455 | for_each_td(td, i) { |
263e529f | 1456 | print_status_init(td->thread_number - 1); |
ebac4655 | 1457 | |
2dc1bbeb | 1458 | if (!td->o.create_serialize) { |
380cf265 | 1459 | init_disk_util(td); |
ebac4655 | 1460 | continue; |
380cf265 | 1461 | } |
ebac4655 JA |
1462 | |
1463 | /* | |
1464 | * do file setup here so it happens sequentially, | |
1465 | * we don't want X number of threads getting their | |
1466 | * client data interspersed on disk | |
1467 | */ | |
53cdc686 | 1468 | if (setup_files(td)) { |
5bf13a5a JA |
1469 | exit_value++; |
1470 | if (td->error) | |
5921e80c JA |
1471 | log_err("fio: pid=%d, err=%d/%s\n", |
1472 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1473 | td_set_runstate(td, TD_REAPED); |
1474 | todo--; | |
c69e8875 JA |
1475 | } else { |
1476 | struct fio_file *f; | |
1477 | unsigned int i; | |
1478 | ||
1479 | /* | |
1480 | * for sharing to work, each job must always open | |
1481 | * its own files. so close them, if we opened them | |
1482 | * for creation | |
1483 | */ | |
22a57ba8 JA |
1484 | for_each_file(td, f, i) { |
1485 | if (fio_file_open(f)) | |
1486 | td_io_close_file(td, f); | |
22a57ba8 | 1487 | } |
5ec10eaa | 1488 | } |
380cf265 JA |
1489 | |
1490 | init_disk_util(td); | |
ebac4655 JA |
1491 | } |
1492 | ||
a2f77c9f JA |
1493 | set_genesis_time(); |
1494 | ||
ebac4655 | 1495 | while (todo) { |
75154845 JA |
1496 | struct thread_data *map[MAX_JOBS]; |
1497 | struct timeval this_start; | |
1498 | int this_jobs = 0, left; | |
1499 | ||
ebac4655 JA |
1500 | /* |
1501 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1502 | */ | |
34572e28 | 1503 | for_each_td(td, i) { |
ebac4655 JA |
1504 | if (td->runstate != TD_NOT_CREATED) |
1505 | continue; | |
1506 | ||
1507 | /* | |
1508 | * never got a chance to start, killed by other | |
1509 | * thread for some reason | |
1510 | */ | |
1511 | if (td->terminate) { | |
1512 | todo--; | |
1513 | continue; | |
1514 | } | |
1515 | ||
2dc1bbeb | 1516 | if (td->o.start_delay) { |
263e529f | 1517 | spent = mtime_since_genesis(); |
ebac4655 | 1518 | |
2dc1bbeb | 1519 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1520 | continue; |
1521 | } | |
1522 | ||
6f88bcb0 JA |
1523 | if (td->o.stonewall && (nr_started || nr_running)) { |
1524 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1525 | td->o.name); | |
ebac4655 | 1526 | break; |
6f88bcb0 | 1527 | } |
ebac4655 | 1528 | |
75154845 JA |
1529 | /* |
1530 | * Set state to created. Thread will transition | |
1531 | * to TD_INITIALIZED when it's done setting up. | |
1532 | */ | |
ebac4655 | 1533 | td_set_runstate(td, TD_CREATED); |
75154845 | 1534 | map[this_jobs++] = td; |
ebac4655 JA |
1535 | nr_started++; |
1536 | ||
2dc1bbeb | 1537 | if (td->o.use_thread) { |
f42c153d JA |
1538 | int ret; |
1539 | ||
ee56ad50 | 1540 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1541 | ret = pthread_create(&td->thread, NULL, |
1542 | thread_main, td); | |
1543 | if (ret) { | |
1544 | log_err("pthread_create: %s\n", | |
1545 | strerror(ret)); | |
ebac4655 | 1546 | nr_started--; |
c8bb6faf | 1547 | break; |
ebac4655 | 1548 | } |
f42c153d JA |
1549 | ret = pthread_detach(td->thread); |
1550 | if (ret) | |
1551 | log_err("pthread_detach: %s", | |
1552 | strerror(ret)); | |
ebac4655 | 1553 | } else { |
5e1d306e | 1554 | pid_t pid; |
ee56ad50 | 1555 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1556 | pid = fork(); |
1557 | if (!pid) { | |
a6418147 JA |
1558 | int ret = fork_main(shm_id, i); |
1559 | ||
5e1d306e JA |
1560 | _exit(ret); |
1561 | } else if (i == fio_debug_jobno) | |
1562 | *fio_debug_jobp = pid; | |
ebac4655 | 1563 | } |
29adda3c | 1564 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1565 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1566 | log_err("fio: job startup hung? exiting.\n"); | |
1567 | terminate_threads(TERMINATE_ALL); | |
1568 | fio_abort = 1; | |
1569 | nr_started--; | |
1570 | break; | |
1571 | } | |
29adda3c | 1572 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1573 | } |
1574 | ||
1575 | /* | |
75154845 JA |
1576 | * Wait for the started threads to transition to |
1577 | * TD_INITIALIZED. | |
ebac4655 | 1578 | */ |
02bcaa8c | 1579 | fio_gettime(&this_start, NULL); |
75154845 | 1580 | left = this_jobs; |
6ce15a32 | 1581 | while (left && !fio_abort) { |
75154845 JA |
1582 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1583 | break; | |
1584 | ||
1585 | usleep(100000); | |
1586 | ||
1587 | for (i = 0; i < this_jobs; i++) { | |
1588 | td = map[i]; | |
1589 | if (!td) | |
1590 | continue; | |
b6f4d880 | 1591 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1592 | map[i] = NULL; |
1593 | left--; | |
b6f4d880 JA |
1594 | } else if (td->runstate >= TD_EXITED) { |
1595 | map[i] = NULL; | |
1596 | left--; | |
1597 | todo--; | |
1598 | nr_running++; /* work-around... */ | |
75154845 JA |
1599 | } |
1600 | } | |
1601 | } | |
1602 | ||
1603 | if (left) { | |
3b70d7e5 | 1604 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1605 | for (i = 0; i < this_jobs; i++) { |
1606 | td = map[i]; | |
1607 | if (!td) | |
1608 | continue; | |
1609 | kill(td->pid, SIGTERM); | |
1610 | } | |
1611 | break; | |
1612 | } | |
1613 | ||
1614 | /* | |
b6f4d880 | 1615 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1616 | */ |
34572e28 | 1617 | for_each_td(td, i) { |
75154845 | 1618 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1619 | continue; |
1620 | ||
b29ee5b3 JA |
1621 | if (in_ramp_time(td)) |
1622 | td_set_runstate(td, TD_RAMP); | |
1623 | else | |
1624 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1625 | nr_running++; |
1626 | nr_started--; | |
581e7141 JA |
1627 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1628 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1629 | todo--; |
cdd18ad8 | 1630 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1631 | } |
1632 | ||
1633 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1634 | ||
1635 | if (todo) | |
1636 | usleep(100000); | |
1637 | } | |
1638 | ||
1639 | while (nr_running) { | |
1640 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1641 | usleep(10000); | |
1642 | } | |
1643 | ||
1644 | update_io_ticks(); | |
2f9ade3c | 1645 | fio_unpin_memory(); |
ebac4655 JA |
1646 | } |
1647 | ||
ebac4655 JA |
1648 | int main(int argc, char *argv[]) |
1649 | { | |
29d610e1 JA |
1650 | long ps; |
1651 | ||
2e5cdb11 | 1652 | sinit(); |
d54a144d | 1653 | init_rand(&__fio_rand_state); |
2e5cdb11 | 1654 | |
dbe1125e JA |
1655 | /* |
1656 | * We need locale for number printing, if it isn't set then just | |
1657 | * go with the US format. | |
1658 | */ | |
1659 | if (!getenv("LC_NUMERIC")) | |
1660 | setlocale(LC_NUMERIC, "en_US"); | |
1661 | ||
29d610e1 JA |
1662 | ps = sysconf(_SC_PAGESIZE); |
1663 | if (ps < 0) { | |
1664 | log_err("Failed to get page size\n"); | |
1665 | return 1; | |
1666 | } | |
1667 | ||
cfc99db7 | 1668 | page_size = ps; |
29d610e1 JA |
1669 | page_mask = ps - 1; |
1670 | ||
74929ac2 JA |
1671 | fio_keywords_init(); |
1672 | ||
1673 | if (parse_options(argc, argv)) | |
1674 | return 1; | |
1675 | ||
07b3232d JA |
1676 | if (exec_profile && load_profile(exec_profile)) |
1677 | return 1; | |
1678 | ||
74929ac2 JA |
1679 | if (!thread_number) |
1680 | return 0; | |
1681 | ||
bb3884d8 JA |
1682 | if (write_bw_log) { |
1683 | setup_log(&agg_io_log[DDIR_READ]); | |
1684 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1685 | } | |
1686 | ||
cdd18ad8 | 1687 | startup_mutex = fio_mutex_init(0); |
d2bbb8d9 | 1688 | writeout_mutex = fio_mutex_init(1); |
07739b57 | 1689 | |
a2f77c9f JA |
1690 | set_genesis_time(); |
1691 | ||
ccb0fa24 | 1692 | status_timer_arm(); |
ebac4655 | 1693 | |
dae5341c JA |
1694 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1695 | INIT_FLIST_HEAD(cgroup_list); | |
1696 | ||
ebac4655 | 1697 | run_threads(); |
6ce15a32 | 1698 | |
bb3884d8 | 1699 | if (!fio_abort) { |
6ce15a32 | 1700 | show_run_stats(); |
bb3884d8 | 1701 | if (write_bw_log) { |
5ec10eaa JA |
1702 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1703 | __finish_log(agg_io_log[DDIR_WRITE], | |
1704 | "agg-write_bw.log"); | |
bb3884d8 JA |
1705 | } |
1706 | } | |
ebac4655 | 1707 | |
dae5341c JA |
1708 | cgroup_kill(cgroup_list); |
1709 | sfree(cgroup_list); | |
61eb313e | 1710 | sfree(cgroup_mnt); |
39f22027 | 1711 | |
cdd18ad8 | 1712 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1713 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1714 | return exit_value; |
ebac4655 | 1715 | } |