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" |
ebac4655 | 40 | |
cfc99db7 JA |
41 | unsigned long page_mask; |
42 | unsigned long page_size; | |
29d610e1 JA |
43 | #define ALIGN(buf) \ |
44 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
45 | |
46 | int groupid = 0; | |
47 | int thread_number = 0; | |
9cedf167 JA |
48 | int nr_process = 0; |
49 | int nr_thread = 0; | |
ebac4655 | 50 | int shm_id = 0; |
53cdc686 | 51 | int temp_stall_ts; |
d3eeeabc | 52 | unsigned long done_secs = 0; |
ebac4655 | 53 | |
cdd18ad8 | 54 | static struct fio_mutex *startup_mutex; |
6ce15a32 | 55 | static volatile int fio_abort; |
437c9b71 | 56 | static int exit_value; |
ccb0fa24 | 57 | static struct itimerval itimer; |
ebac4655 | 58 | |
bb3884d8 JA |
59 | struct io_log *agg_io_log[2]; |
60 | ||
ebac4655 | 61 | #define TERMINATE_ALL (-1) |
75154845 | 62 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 63 | |
b29ee5b3 | 64 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 65 | { |
d8fab1b6 JA |
66 | if (td->runstate == runstate) |
67 | return; | |
68 | ||
5921e80c JA |
69 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
70 | td->runstate, runstate); | |
6ce15a32 JA |
71 | td->runstate = runstate; |
72 | } | |
73 | ||
390c40e2 | 74 | static void terminate_threads(int group_id) |
ebac4655 | 75 | { |
34572e28 | 76 | struct thread_data *td; |
ebac4655 JA |
77 | int i; |
78 | ||
6f88bcb0 JA |
79 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
80 | ||
34572e28 | 81 | for_each_td(td, i) { |
ebac4655 | 82 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 83 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 84 | td->o.name, (int) td->pid); |
ad830ed7 JA |
85 | td->terminate = 1; |
86 | td->o.start_delay = 0; | |
87 | ||
7a93ab19 JA |
88 | /* |
89 | * if the thread is running, just let it exit | |
90 | */ | |
91 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 92 | kill(td->pid, SIGQUIT); |
f63f7f3b JA |
93 | else { |
94 | struct ioengine_ops *ops = td->io_ops; | |
95 | ||
96 | if (ops && (ops->flags & FIO_SIGQUIT)) | |
97 | kill(td->pid, SIGQUIT); | |
98 | } | |
ebac4655 JA |
99 | } |
100 | } | |
101 | } | |
102 | ||
ccb0fa24 JA |
103 | static void status_timer_arm(void) |
104 | { | |
105 | itimer.it_value.tv_sec = 0; | |
106 | itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; | |
107 | setitimer(ITIMER_REAL, &itimer, NULL); | |
108 | } | |
109 | ||
02444ad1 | 110 | static void sig_alrm(int fio_unused sig) |
ebac4655 | 111 | { |
697a606c | 112 | if (threads) { |
5ec10eaa | 113 | update_io_ticks(); |
5ec10eaa | 114 | print_thread_status(); |
ccb0fa24 | 115 | status_timer_arm(); |
697a606c JA |
116 | } |
117 | } | |
118 | ||
119 | static void sig_int(int sig) | |
120 | { | |
121 | if (threads) { | |
5ec10eaa JA |
122 | printf("\nfio: terminating on signal %d\n", sig); |
123 | fflush(stdout); | |
124 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
125 | } |
126 | } | |
127 | ||
02444ad1 | 128 | static void sig_ill(int fio_unused sig) |
de79c915 JA |
129 | { |
130 | if (!threads) | |
131 | return; | |
132 | ||
133 | log_err("fio: illegal instruction. your cpu does not support " | |
134 | "the sse4.2 instruction for crc32c\n"); | |
135 | terminate_threads(TERMINATE_ALL); | |
136 | exit(4); | |
137 | } | |
138 | ||
697a606c JA |
139 | static void set_sig_handlers(void) |
140 | { | |
141 | struct sigaction act; | |
142 | ||
143 | memset(&act, 0, sizeof(act)); | |
144 | act.sa_handler = sig_alrm; | |
145 | act.sa_flags = SA_RESTART; | |
146 | sigaction(SIGALRM, &act, NULL); | |
147 | ||
148 | memset(&act, 0, sizeof(act)); | |
149 | act.sa_handler = sig_int; | |
150 | act.sa_flags = SA_RESTART; | |
151 | sigaction(SIGINT, &act, NULL); | |
de79c915 JA |
152 | |
153 | memset(&act, 0, sizeof(act)); | |
154 | act.sa_handler = sig_ill; | |
155 | act.sa_flags = SA_RESTART; | |
156 | sigaction(SIGILL, &act, NULL); | |
697a606c JA |
157 | } |
158 | ||
993bf48b JA |
159 | static inline int should_check_rate(struct thread_data *td) |
160 | { | |
161 | /* | |
162 | * No minimum rate set, always ok | |
163 | */ | |
164 | if (!td->o.ratemin && !td->o.rate_iops_min) | |
165 | return 0; | |
166 | ||
167 | return 1; | |
168 | } | |
169 | ||
906c8d75 JA |
170 | /* |
171 | * Check if we are above the minimum rate given. | |
172 | */ | |
ebac4655 JA |
173 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
174 | { | |
0904200b | 175 | unsigned long long bytes = 0; |
4e991c23 | 176 | unsigned long iops = 0; |
ebac4655 JA |
177 | unsigned long spent; |
178 | unsigned long rate; | |
ebac4655 JA |
179 | |
180 | /* | |
181 | * allow a 2 second settle period in the beginning | |
182 | */ | |
183 | if (mtime_since(&td->start, now) < 2000) | |
184 | return 0; | |
185 | ||
4e991c23 JA |
186 | if (td_read(td)) { |
187 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 188 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
189 | } |
190 | if (td_write(td)) { | |
191 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 192 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 193 | } |
0904200b | 194 | |
ebac4655 JA |
195 | /* |
196 | * if rate blocks is set, sample is running | |
197 | */ | |
4e991c23 | 198 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 199 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 200 | if (spent < td->o.ratecycle) |
ebac4655 JA |
201 | return 0; |
202 | ||
2dc1bbeb | 203 | if (td->o.rate) { |
4e991c23 JA |
204 | /* |
205 | * check bandwidth specified rate | |
206 | */ | |
207 | if (bytes < td->rate_bytes) { | |
5ec10eaa JA |
208 | log_err("%s: min rate %u not met\n", td->o.name, |
209 | td->o.ratemin); | |
4e991c23 JA |
210 | return 1; |
211 | } else { | |
212 | rate = (bytes - td->rate_bytes) / spent; | |
5ec10eaa JA |
213 | if (rate < td->o.ratemin || |
214 | bytes < td->rate_bytes) { | |
215 | log_err("%s: min rate %u not met, got" | |
216 | " %luKiB/sec\n", td->o.name, | |
217 | td->o.ratemin, rate); | |
4e991c23 JA |
218 | return 1; |
219 | } | |
220 | } | |
413dd459 | 221 | } else { |
4e991c23 JA |
222 | /* |
223 | * checks iops specified rate | |
224 | */ | |
2dc1bbeb | 225 | if (iops < td->o.rate_iops) { |
5ec10eaa JA |
226 | log_err("%s: min iops rate %u not met\n", |
227 | td->o.name, td->o.rate_iops); | |
413dd459 | 228 | return 1; |
4e991c23 JA |
229 | } else { |
230 | rate = (iops - td->rate_blocks) / spent; | |
5ec10eaa JA |
231 | if (rate < td->o.rate_iops_min || |
232 | iops < td->rate_blocks) { | |
233 | log_err("%s: min iops rate %u not met," | |
234 | " got %lu\n", td->o.name, | |
235 | td->o.rate_iops_min, | |
236 | rate); | |
4e991c23 | 237 | } |
413dd459 | 238 | } |
ebac4655 JA |
239 | } |
240 | } | |
241 | ||
0904200b | 242 | td->rate_bytes = bytes; |
4e991c23 | 243 | td->rate_blocks = iops; |
ebac4655 JA |
244 | memcpy(&td->lastrate, now, sizeof(*now)); |
245 | return 0; | |
246 | } | |
247 | ||
248 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
249 | { | |
2dc1bbeb | 250 | if (!td->o.timeout) |
ebac4655 | 251 | return 0; |
2dc1bbeb | 252 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
253 | return 1; |
254 | ||
255 | return 0; | |
256 | } | |
257 | ||
906c8d75 JA |
258 | /* |
259 | * When job exits, we can cancel the in-flight IO if we are using async | |
260 | * io. Attempt to do so. | |
261 | */ | |
ebac4655 JA |
262 | static void cleanup_pending_aio(struct thread_data *td) |
263 | { | |
01743ee1 | 264 | struct flist_head *entry, *n; |
ebac4655 JA |
265 | struct io_u *io_u; |
266 | int r; | |
267 | ||
268 | /* | |
269 | * get immediately available events, if any | |
270 | */ | |
d7762cf8 | 271 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
272 | if (r < 0) |
273 | return; | |
ebac4655 JA |
274 | |
275 | /* | |
276 | * now cancel remaining active events | |
277 | */ | |
2866c82d | 278 | if (td->io_ops->cancel) { |
01743ee1 JA |
279 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
280 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 281 | |
0c6e7517 JA |
282 | /* |
283 | * if the io_u isn't in flight, then that generally | |
284 | * means someone leaked an io_u. complain but fix | |
285 | * it up, so we don't stall here. | |
286 | */ | |
287 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
288 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 289 | put_io_u(td, io_u); |
0c6e7517 JA |
290 | } else { |
291 | r = td->io_ops->cancel(td, io_u); | |
292 | if (!r) | |
293 | put_io_u(td, io_u); | |
294 | } | |
ebac4655 JA |
295 | } |
296 | } | |
297 | ||
97601024 | 298 | if (td->cur_depth) |
d7762cf8 | 299 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
300 | } |
301 | ||
858a3d47 JA |
302 | /* |
303 | * Helper to handle the final sync of a file. Works just like the normal | |
304 | * io path, just does everything sync. | |
305 | */ | |
306 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
307 | { | |
308 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
309 | int ret; |
310 | ||
311 | if (!io_u) | |
312 | return 1; | |
313 | ||
314 | io_u->ddir = DDIR_SYNC; | |
315 | io_u->file = f; | |
316 | ||
317 | if (td_io_prep(td, io_u)) { | |
318 | put_io_u(td, io_u); | |
319 | return 1; | |
320 | } | |
321 | ||
755200a3 | 322 | requeue: |
858a3d47 | 323 | ret = td_io_queue(td, io_u); |
36167d82 | 324 | if (ret < 0) { |
e1161c32 | 325 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 326 | put_io_u(td, io_u); |
858a3d47 | 327 | return 1; |
36167d82 | 328 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 329 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 330 | return 1; |
36167d82 JA |
331 | } else if (ret == FIO_Q_COMPLETED) { |
332 | if (io_u->error) { | |
e1161c32 | 333 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
334 | return 1; |
335 | } | |
858a3d47 | 336 | |
d7762cf8 | 337 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 338 | return 1; |
755200a3 JA |
339 | } else if (ret == FIO_Q_BUSY) { |
340 | if (td_io_commit(td)) | |
341 | return 1; | |
342 | goto requeue; | |
858a3d47 JA |
343 | } |
344 | ||
345 | return 0; | |
346 | } | |
347 | ||
993bf48b JA |
348 | static inline void update_tv_cache(struct thread_data *td) |
349 | { | |
350 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
351 | fio_gettime(&td->tv_cache, NULL); | |
352 | } | |
353 | ||
906c8d75 | 354 | /* |
34403fb1 | 355 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
356 | * reads the blocks back in, and checks the crc/md5 of the data. |
357 | */ | |
1e97cce9 | 358 | static void do_verify(struct thread_data *td) |
ebac4655 | 359 | { |
53cdc686 | 360 | struct fio_file *f; |
36167d82 | 361 | struct io_u *io_u; |
af52b345 JA |
362 | int ret, min_events; |
363 | unsigned int i; | |
e5b401d4 JA |
364 | |
365 | /* | |
366 | * sync io first and invalidate cache, to make sure we really | |
367 | * read from disk. | |
368 | */ | |
369 | for_each_file(td, f, i) { | |
3d7b485f JA |
370 | if (!(f->flags & FIO_FILE_OPEN)) |
371 | continue; | |
b2fdda43 JA |
372 | if (fio_io_sync(td, f)) |
373 | break; | |
374 | if (file_invalidate_cache(td, f)) | |
375 | break; | |
e5b401d4 | 376 | } |
ebac4655 | 377 | |
b2fdda43 JA |
378 | if (td->error) |
379 | return; | |
380 | ||
ebac4655 JA |
381 | td_set_runstate(td, TD_VERIFYING); |
382 | ||
36167d82 JA |
383 | io_u = NULL; |
384 | while (!td->terminate) { | |
4950421a | 385 | int ret2, full; |
5451792e | 386 | |
ebac4655 JA |
387 | io_u = __get_io_u(td); |
388 | if (!io_u) | |
389 | break; | |
390 | ||
993bf48b JA |
391 | update_tv_cache(td); |
392 | ||
393 | if (runtime_exceeded(td, &td->tv_cache)) { | |
069c2918 | 394 | put_io_u(td, io_u); |
90a87d4b | 395 | td->terminate = 1; |
02bcaa8c | 396 | break; |
069c2918 | 397 | } |
02bcaa8c | 398 | |
069c2918 JA |
399 | if (get_next_verify(td, io_u)) { |
400 | put_io_u(td, io_u); | |
ebac4655 | 401 | break; |
069c2918 | 402 | } |
ebac4655 | 403 | |
069c2918 JA |
404 | if (td_io_prep(td, io_u)) { |
405 | put_io_u(td, io_u); | |
53cdc686 | 406 | break; |
069c2918 | 407 | } |
d7762cf8 JA |
408 | |
409 | io_u->end_io = verify_io_u; | |
53cdc686 | 410 | |
11786802 | 411 | ret = td_io_queue(td, io_u); |
36167d82 JA |
412 | switch (ret) { |
413 | case FIO_Q_COMPLETED: | |
414 | if (io_u->error) | |
22819ec2 | 415 | ret = -io_u->error; |
9e9d164e | 416 | else if (io_u->resid) { |
36167d82 | 417 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 418 | struct fio_file *f = io_u->file; |
ebac4655 | 419 | |
9e9d164e JA |
420 | /* |
421 | * zero read, fail | |
422 | */ | |
423 | if (!bytes) { | |
41d9f0fd | 424 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
425 | put_io_u(td, io_u); |
426 | break; | |
427 | } | |
8400d9b2 | 428 | |
36167d82 JA |
429 | io_u->xfer_buflen = io_u->resid; |
430 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
431 | io_u->offset += bytes; |
432 | ||
30061b97 JA |
433 | td->ts.short_io_u[io_u->ddir]++; |
434 | ||
d460eb31 | 435 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
436 | goto sync_done; |
437 | ||
11786802 JA |
438 | requeue_io_u(td, &io_u); |
439 | } else { | |
8400d9b2 | 440 | sync_done: |
11786802 JA |
441 | ret = io_u_sync_complete(td, io_u); |
442 | if (ret < 0) | |
443 | break; | |
36167d82 | 444 | } |
36167d82 JA |
445 | continue; |
446 | case FIO_Q_QUEUED: | |
447 | break; | |
755200a3 JA |
448 | case FIO_Q_BUSY: |
449 | requeue_io_u(td, &io_u); | |
5451792e JA |
450 | ret2 = td_io_commit(td); |
451 | if (ret2 < 0) | |
452 | ret = ret2; | |
755200a3 | 453 | break; |
36167d82 JA |
454 | default: |
455 | assert(ret < 0); | |
e1161c32 | 456 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
457 | break; |
458 | } | |
459 | ||
99784632 | 460 | if (ret < 0 || td->error) |
3af6ef39 JA |
461 | break; |
462 | ||
ebac4655 | 463 | /* |
3af6ef39 JA |
464 | * if we can queue more, do so. but check if there are |
465 | * completed io_u's first. | |
ebac4655 | 466 | */ |
4950421a JA |
467 | full = queue_full(td) || ret == FIO_Q_BUSY; |
468 | if (full || !td->o.iodepth_batch_complete) { | |
469 | min_events = td->o.iodepth_batch_complete; | |
470 | if (full && !min_events) | |
838bc709 | 471 | min_events = 1; |
e916b390 | 472 | |
4950421a JA |
473 | do { |
474 | /* | |
475 | * Reap required number of io units, if any, | |
476 | * and do the verification on them through | |
477 | * the callback handler | |
478 | */ | |
479 | if (io_u_queued_complete(td, min_events) < 0) { | |
480 | ret = -1; | |
481 | break; | |
482 | } | |
483 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
484 | } | |
485 | if (ret < 0) | |
ebac4655 | 486 | break; |
36167d82 | 487 | } |
ebac4655 | 488 | |
c01c0395 JA |
489 | if (!td->error) { |
490 | min_events = td->cur_depth; | |
491 | ||
492 | if (min_events) | |
493 | ret = io_u_queued_complete(td, min_events); | |
494 | } else | |
ebac4655 JA |
495 | cleanup_pending_aio(td); |
496 | ||
497 | td_set_runstate(td, TD_RUNNING); | |
498 | } | |
499 | ||
32cd46a0 | 500 | /* |
906c8d75 | 501 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
502 | * and reaps them, checking for rate and errors along the way. |
503 | */ | |
ebac4655 JA |
504 | static void do_io(struct thread_data *td) |
505 | { | |
ebac4655 | 506 | unsigned long usec; |
af52b345 JA |
507 | unsigned int i; |
508 | int ret = 0; | |
ebac4655 | 509 | |
b29ee5b3 JA |
510 | if (in_ramp_time(td)) |
511 | td_set_runstate(td, TD_RAMP); | |
512 | else | |
513 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 514 | |
7bb48f84 | 515 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
516 | struct timeval comp_time; |
517 | long bytes_done = 0; | |
84585003 | 518 | int min_evts = 0; |
ebac4655 | 519 | struct io_u *io_u; |
4950421a | 520 | int ret2, full; |
ebac4655 JA |
521 | |
522 | if (td->terminate) | |
523 | break; | |
524 | ||
3d7c391d | 525 | io_u = get_io_u(td); |
ebac4655 JA |
526 | if (!io_u) |
527 | break; | |
528 | ||
993bf48b | 529 | update_tv_cache(td); |
97601024 | 530 | |
993bf48b | 531 | if (runtime_exceeded(td, &td->tv_cache)) { |
97601024 | 532 | put_io_u(td, io_u); |
90a87d4b | 533 | td->terminate = 1; |
97601024 JA |
534 | break; |
535 | } | |
36167d82 | 536 | |
b67740d3 JA |
537 | /* |
538 | * Add verification end_io handler, if asked to verify | |
539 | * a previously written file. | |
540 | */ | |
cc3fc855 | 541 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) { |
b67740d3 | 542 | io_u->end_io = verify_io_u; |
d8fab1b6 | 543 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
544 | } else if (in_ramp_time(td)) |
545 | td_set_runstate(td, TD_RAMP); | |
546 | else | |
d8fab1b6 | 547 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 548 | |
11786802 | 549 | ret = td_io_queue(td, io_u); |
36167d82 JA |
550 | switch (ret) { |
551 | case FIO_Q_COMPLETED: | |
5451792e JA |
552 | if (io_u->error) |
553 | ret = -io_u->error; | |
9e9d164e | 554 | else if (io_u->resid) { |
36167d82 | 555 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 556 | struct fio_file *f = io_u->file; |
36167d82 | 557 | |
9e9d164e JA |
558 | /* |
559 | * zero read, fail | |
560 | */ | |
561 | if (!bytes) { | |
41d9f0fd | 562 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
563 | put_io_u(td, io_u); |
564 | break; | |
565 | } | |
566 | ||
cec6b55d | 567 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 568 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
569 | io_u->offset += bytes; |
570 | ||
30061b97 JA |
571 | td->ts.short_io_u[io_u->ddir]++; |
572 | ||
d460eb31 | 573 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
574 | goto sync_done; |
575 | ||
11786802 JA |
576 | requeue_io_u(td, &io_u); |
577 | } else { | |
5a7c5680 | 578 | sync_done: |
993bf48b JA |
579 | if (should_check_rate(td)) |
580 | fio_gettime(&comp_time, NULL); | |
581 | ||
11786802 JA |
582 | bytes_done = io_u_sync_complete(td, io_u); |
583 | if (bytes_done < 0) | |
584 | ret = bytes_done; | |
cec6b55d | 585 | } |
36167d82 JA |
586 | break; |
587 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
588 | /* |
589 | * if the engine doesn't have a commit hook, | |
590 | * the io_u is really queued. if it does have such | |
591 | * a hook, it has to call io_u_queued() itself. | |
592 | */ | |
593 | if (td->io_ops->commit == NULL) | |
594 | io_u_queued(td, io_u); | |
36167d82 | 595 | break; |
755200a3 JA |
596 | case FIO_Q_BUSY: |
597 | requeue_io_u(td, &io_u); | |
5451792e JA |
598 | ret2 = td_io_commit(td); |
599 | if (ret2 < 0) | |
600 | ret = ret2; | |
755200a3 | 601 | break; |
36167d82 JA |
602 | default: |
603 | assert(ret < 0); | |
604 | put_io_u(td, io_u); | |
605 | break; | |
ebac4655 JA |
606 | } |
607 | ||
99784632 | 608 | if (ret < 0 || td->error) |
36167d82 JA |
609 | break; |
610 | ||
97601024 JA |
611 | /* |
612 | * See if we need to complete some commands | |
613 | */ | |
4950421a JA |
614 | full = queue_full(td) || ret == FIO_Q_BUSY; |
615 | if (full || !td->o.iodepth_batch_complete) { | |
616 | min_evts = td->o.iodepth_batch_complete; | |
617 | if (full && !min_evts) | |
36167d82 | 618 | min_evts = 1; |
4950421a | 619 | |
993bf48b JA |
620 | if (should_check_rate(td)) |
621 | fio_gettime(&comp_time, NULL); | |
4950421a JA |
622 | |
623 | do { | |
624 | ret = io_u_queued_complete(td, min_evts); | |
625 | if (ret <= 0) | |
626 | break; | |
627 | ||
628 | bytes_done += ret; | |
629 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
ebac4655 JA |
630 | } |
631 | ||
4950421a JA |
632 | if (ret < 0) |
633 | break; | |
97601024 JA |
634 | if (!bytes_done) |
635 | continue; | |
636 | ||
ebac4655 JA |
637 | /* |
638 | * the rate is batched for now, it should work for batches | |
639 | * of completions except the very first one which may look | |
640 | * a little bursty | |
641 | */ | |
993bf48b JA |
642 | if (!in_ramp_time(td) && should_check_rate(td)) { |
643 | usec = utime_since(&td->tv_cache, &comp_time); | |
ebac4655 | 644 | |
721938ae | 645 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 646 | |
721938ae JA |
647 | if (check_min_rate(td, &comp_time)) { |
648 | if (exitall_on_terminate) | |
649 | terminate_threads(td->groupid); | |
650 | td_verror(td, EIO, "check_min_rate"); | |
651 | break; | |
652 | } | |
ebac4655 JA |
653 | } |
654 | ||
2dc1bbeb | 655 | if (td->o.thinktime) { |
9c1f7434 JA |
656 | unsigned long long b; |
657 | ||
658 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 659 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
660 | int left; |
661 | ||
2dc1bbeb JA |
662 | if (td->o.thinktime_spin) |
663 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 664 | |
2dc1bbeb | 665 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
666 | if (left) |
667 | usec_sleep(td, left); | |
668 | } | |
9c1f7434 | 669 | } |
ebac4655 JA |
670 | } |
671 | ||
aa31f1f1 SL |
672 | if (td->o.fill_device && td->error == ENOSPC) { |
673 | td->error = 0; | |
674 | td->terminate = 1; | |
675 | } | |
4d2413c6 | 676 | if (!td->error) { |
3d7c391d JA |
677 | struct fio_file *f; |
678 | ||
c01c0395 JA |
679 | i = td->cur_depth; |
680 | if (i) | |
681 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 682 | |
2dc1bbeb | 683 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 684 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
685 | |
686 | for_each_file(td, f, i) { | |
687 | if (!(f->flags & FIO_FILE_OPEN)) | |
688 | continue; | |
858a3d47 | 689 | fio_io_sync(td, f); |
3d7b485f | 690 | } |
84585003 | 691 | } |
c01c0395 JA |
692 | } else |
693 | cleanup_pending_aio(td); | |
2ba1c290 JA |
694 | |
695 | /* | |
696 | * stop job if we failed doing any IO | |
697 | */ | |
698 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
699 | td->done = 1; | |
ebac4655 JA |
700 | } |
701 | ||
ebac4655 JA |
702 | static void cleanup_io_u(struct thread_data *td) |
703 | { | |
01743ee1 | 704 | struct flist_head *entry, *n; |
ebac4655 JA |
705 | struct io_u *io_u; |
706 | ||
01743ee1 JA |
707 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
708 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 709 | |
01743ee1 | 710 | flist_del(&io_u->list); |
ebac4655 JA |
711 | free(io_u); |
712 | } | |
713 | ||
2f9ade3c | 714 | free_io_mem(td); |
ebac4655 JA |
715 | } |
716 | ||
717 | static int init_io_u(struct thread_data *td) | |
718 | { | |
719 | struct io_u *io_u; | |
a00735e6 | 720 | unsigned int max_bs; |
ebac4655 JA |
721 | int i, max_units; |
722 | char *p; | |
723 | ||
1e3c09af | 724 | max_units = td->o.iodepth; |
2dc1bbeb | 725 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
726 | td->orig_buffer_size = (unsigned long long) max_bs |
727 | * (unsigned long long) max_units; | |
728 | ||
729 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
730 | unsigned long bs; | |
74b025b0 | 731 | |
5ec10eaa JA |
732 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
733 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
734 | } | |
ebac4655 | 735 | |
c3990645 JA |
736 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
737 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
738 | return 1; | |
739 | } | |
740 | ||
2f9ade3c JA |
741 | if (allocate_io_mem(td)) |
742 | return 1; | |
ebac4655 | 743 | |
b3f378e9 JA |
744 | if (td->o.odirect) |
745 | p = ALIGN(td->orig_buffer); | |
746 | else | |
747 | p = td->orig_buffer; | |
748 | ||
ebac4655 | 749 | for (i = 0; i < max_units; i++) { |
db1cd90b JA |
750 | if (td->terminate) |
751 | return 1; | |
ebac4655 JA |
752 | io_u = malloc(sizeof(*io_u)); |
753 | memset(io_u, 0, sizeof(*io_u)); | |
01743ee1 | 754 | INIT_FLIST_HEAD(&io_u->list); |
ebac4655 | 755 | |
b4c5e1ac JA |
756 | if (!(td->io_ops->flags & FIO_NOIO)) { |
757 | io_u->buf = p + max_bs * i; | |
e9459e5a | 758 | |
5973cafb JA |
759 | if (td_write(td) && !td->o.refill_buffers) |
760 | io_u_fill_buffer(td, io_u, max_bs); | |
b4c5e1ac | 761 | } |
6b9cea23 | 762 | |
b1ff3403 | 763 | io_u->index = i; |
0c6e7517 | 764 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 765 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
766 | } |
767 | ||
768 | return 0; | |
769 | } | |
770 | ||
da86774e JA |
771 | static int switch_ioscheduler(struct thread_data *td) |
772 | { | |
773 | char tmp[256], tmp2[128]; | |
774 | FILE *f; | |
775 | int ret; | |
776 | ||
ba0fbe10 | 777 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
778 | return 0; |
779 | ||
da86774e JA |
780 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
781 | ||
782 | f = fopen(tmp, "r+"); | |
783 | if (!f) { | |
cbf5c1d7 | 784 | if (errno == ENOENT) { |
5ec10eaa JA |
785 | log_err("fio: os or kernel doesn't support IO scheduler" |
786 | " switching\n"); | |
cbf5c1d7 JA |
787 | return 0; |
788 | } | |
789 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
790 | return 1; |
791 | } | |
792 | ||
793 | /* | |
794 | * Set io scheduler. | |
795 | */ | |
2dc1bbeb | 796 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 797 | if (ferror(f) || ret != 1) { |
e1161c32 | 798 | td_verror(td, errno, "fwrite"); |
da86774e JA |
799 | fclose(f); |
800 | return 1; | |
801 | } | |
802 | ||
803 | rewind(f); | |
804 | ||
805 | /* | |
806 | * Read back and check that the selected scheduler is now the default. | |
807 | */ | |
808 | ret = fread(tmp, 1, sizeof(tmp), f); | |
809 | if (ferror(f) || ret < 0) { | |
e1161c32 | 810 | td_verror(td, errno, "fread"); |
da86774e JA |
811 | fclose(f); |
812 | return 1; | |
813 | } | |
814 | ||
2dc1bbeb | 815 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 816 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 817 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 818 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
819 | fclose(f); |
820 | return 1; | |
821 | } | |
822 | ||
823 | fclose(f); | |
824 | return 0; | |
825 | } | |
826 | ||
492158cf JA |
827 | static int keep_running(struct thread_data *td) |
828 | { | |
829 | unsigned long long io_done; | |
830 | ||
20e354ef JA |
831 | if (td->done) |
832 | return 0; | |
492158cf JA |
833 | if (td->o.time_based) |
834 | return 1; | |
835 | if (td->o.loops) { | |
836 | td->o.loops--; | |
837 | return 1; | |
838 | } | |
839 | ||
5ec10eaa JA |
840 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
841 | + td->io_skip_bytes; | |
492158cf JA |
842 | if (io_done < td->o.size) |
843 | return 1; | |
844 | ||
845 | return 0; | |
846 | } | |
847 | ||
b29ee5b3 | 848 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 849 | { |
756867bd | 850 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 851 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 852 | td->zone_bytes = 0; |
d7d3b49b | 853 | td->rate_bytes = 0; |
4e991c23 | 854 | td->rate_blocks = 0; |
38d77cae | 855 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 856 | |
c1324df1 JA |
857 | td->last_was_sync = 0; |
858 | ||
2ba1c290 JA |
859 | /* |
860 | * reset file done count if we are to start over | |
861 | */ | |
862 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 863 | td->nr_done_files = 0; |
b29ee5b3 JA |
864 | } |
865 | ||
866 | void reset_all_stats(struct thread_data *td) | |
867 | { | |
868 | struct timeval tv; | |
869 | int i; | |
870 | ||
871 | reset_io_counters(td); | |
872 | ||
873 | for (i = 0; i < 2; i++) { | |
874 | td->io_bytes[i] = 0; | |
875 | td->io_blocks[i] = 0; | |
876 | td->io_issues[i] = 0; | |
877 | td->ts.total_io_u[i] = 0; | |
878 | } | |
879 | ||
880 | fio_gettime(&tv, NULL); | |
881 | memcpy(&td->epoch, &tv, sizeof(tv)); | |
882 | memcpy(&td->start, &tv, sizeof(tv)); | |
883 | } | |
884 | ||
885 | static int clear_io_state(struct thread_data *td) | |
886 | { | |
887 | struct fio_file *f; | |
888 | unsigned int i; | |
889 | int ret; | |
890 | ||
891 | reset_io_counters(td); | |
281f9b63 | 892 | |
24ffd2c2 | 893 | close_files(td); |
53cdc686 | 894 | |
a978ba68 JA |
895 | ret = 0; |
896 | for_each_file(td, f, i) { | |
281f9b63 | 897 | f->flags &= ~FIO_FILE_DONE; |
a978ba68 JA |
898 | ret = td_io_open_file(td, f); |
899 | if (ret) | |
900 | break; | |
53cdc686 | 901 | } |
a978ba68 JA |
902 | |
903 | return ret; | |
ebac4655 JA |
904 | } |
905 | ||
906c8d75 JA |
906 | /* |
907 | * Entry point for the thread based jobs. The process based jobs end up | |
908 | * here as well, after a little setup. | |
909 | */ | |
ebac4655 JA |
910 | static void *thread_main(void *data) |
911 | { | |
10927316 | 912 | unsigned long long runtime[2], elapsed; |
ebac4655 | 913 | struct thread_data *td = data; |
a978ba68 | 914 | int clear_state; |
ebac4655 | 915 | |
2dc1bbeb | 916 | if (!td->o.use_thread) |
ebac4655 JA |
917 | setsid(); |
918 | ||
919 | td->pid = getpid(); | |
920 | ||
5921e80c | 921 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 922 | |
01743ee1 JA |
923 | INIT_FLIST_HEAD(&td->io_u_freelist); |
924 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
925 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
926 | INIT_FLIST_HEAD(&td->io_log_list); | |
927 | INIT_FLIST_HEAD(&td->io_hist_list); | |
bb5d7d0b | 928 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 929 | |
db1cd90b | 930 | td_set_runstate(td, TD_INITIALIZED); |
cdd18ad8 JA |
931 | fio_mutex_up(startup_mutex); |
932 | fio_mutex_down(td->mutex); | |
db1cd90b JA |
933 | |
934 | /* | |
cdd18ad8 | 935 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
936 | * eating a file descriptor |
937 | */ | |
cdd18ad8 | 938 | fio_mutex_remove(td->mutex); |
db1cd90b | 939 | |
d84f8d49 JA |
940 | /* |
941 | * May alter parameters that init_io_u() will use, so we need to | |
942 | * do this first. | |
943 | */ | |
944 | if (init_iolog(td)) | |
945 | goto err; | |
946 | ||
ebac4655 | 947 | if (init_io_u(td)) |
db1cd90b | 948 | goto err; |
ebac4655 | 949 | |
375b2695 | 950 | if (td->o.cpumask_set && fio_setaffinity(td) == -1) { |
e1161c32 | 951 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 952 | goto err; |
ebac4655 JA |
953 | } |
954 | ||
ac684785 | 955 | if (td->ioprio_set) { |
ebac4655 | 956 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 957 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 958 | goto err; |
ebac4655 JA |
959 | } |
960 | } | |
961 | ||
2dc1bbeb | 962 | if (nice(td->o.nice) == -1) { |
e1161c32 | 963 | td_verror(td, errno, "nice"); |
db1cd90b | 964 | goto err; |
b6f4d880 JA |
965 | } |
966 | ||
2dc1bbeb | 967 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 968 | goto err; |
37f56873 | 969 | |
2dc1bbeb | 970 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
971 | goto err; |
972 | ||
7d6c5283 JA |
973 | if (td_io_init(td)) |
974 | goto err; | |
975 | ||
b5af8293 JA |
976 | if (open_files(td)) |
977 | goto err; | |
978 | ||
68727076 JA |
979 | if (init_random_map(td)) |
980 | goto err; | |
981 | ||
2dc1bbeb JA |
982 | if (td->o.exec_prerun) { |
983 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
984 | goto err; |
985 | } | |
4e0ba8af | 986 | |
69008999 | 987 | fio_gettime(&td->epoch, NULL); |
756867bd | 988 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
989 | |
990 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 991 | clear_state = 0; |
492158cf | 992 | while (keep_running(td)) { |
02bcaa8c | 993 | fio_gettime(&td->start, NULL); |
756867bd | 994 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
993bf48b | 995 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 996 | |
2dc1bbeb | 997 | if (td->o.ratemin) |
5ec10eaa JA |
998 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
999 | sizeof(td->lastrate)); | |
ebac4655 | 1000 | |
a978ba68 JA |
1001 | if (clear_state && clear_io_state(td)) |
1002 | break; | |
1003 | ||
ebac4655 JA |
1004 | prune_io_piece_log(td); |
1005 | ||
ba0fbe10 | 1006 | do_io(td); |
ebac4655 | 1007 | |
a978ba68 JA |
1008 | clear_state = 1; |
1009 | ||
38d77cae JA |
1010 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
1011 | if (td->rw_end_set[DDIR_READ]) | |
5ec10eaa JA |
1012 | elapsed = utime_since(&td->start, |
1013 | &td->rw_end[DDIR_READ]); | |
38d77cae JA |
1014 | else |
1015 | elapsed = utime_since_now(&td->start); | |
1016 | ||
1017 | runtime[DDIR_READ] += elapsed; | |
1018 | } | |
1019 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
1020 | if (td->rw_end_set[DDIR_WRITE]) | |
5ec10eaa JA |
1021 | elapsed = utime_since(&td->start, |
1022 | &td->rw_end[DDIR_WRITE]); | |
38d77cae JA |
1023 | else |
1024 | elapsed = utime_since_now(&td->start); | |
1025 | ||
1026 | runtime[DDIR_WRITE] += elapsed; | |
1027 | } | |
5ec10eaa | 1028 | |
ebac4655 JA |
1029 | if (td->error || td->terminate) |
1030 | break; | |
1031 | ||
e84c73a8 SL |
1032 | if (!td->o.do_verify || |
1033 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1034 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1035 | continue; |
1036 | ||
a978ba68 JA |
1037 | if (clear_io_state(td)) |
1038 | break; | |
1039 | ||
02bcaa8c | 1040 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1041 | |
1042 | do_verify(td); | |
1043 | ||
69008999 | 1044 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1045 | |
1046 | if (td->error || td->terminate) | |
1047 | break; | |
1048 | } | |
1049 | ||
36dff966 | 1050 | update_rusage_stat(td); |
47f0cc48 YHJT |
1051 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
1052 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
1053 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1054 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1055 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1056 | ||
1057 | if (td->ts.bw_log) | |
1058 | finish_log(td, td->ts.bw_log, "bw"); | |
1059 | if (td->ts.slat_log) | |
1060 | finish_log(td, td->ts.slat_log, "slat"); | |
1061 | if (td->ts.clat_log) | |
1062 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb JA |
1063 | if (td->o.exec_postrun) { |
1064 | if (system(td->o.exec_postrun) < 0) | |
1065 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 1066 | } |
ebac4655 JA |
1067 | |
1068 | if (exitall_on_terminate) | |
390c40e2 | 1069 | terminate_threads(td->groupid); |
ebac4655 JA |
1070 | |
1071 | err: | |
5bf13a5a | 1072 | if (td->error) |
5921e80c | 1073 | printf("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1074 | td->verror); |
24ffd2c2 | 1075 | close_and_free_files(td); |
2866c82d | 1076 | close_ioengine(td); |
ebac4655 | 1077 | cleanup_io_u(td); |
f29b25a3 JA |
1078 | |
1079 | /* | |
1080 | * do this very late, it will log file closing as well | |
1081 | */ | |
1082 | if (td->o.write_iolog_file) | |
1083 | write_iolog_close(td); | |
1084 | ||
d23bb327 | 1085 | options_mem_free(td); |
ebac4655 | 1086 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1087 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1088 | } |
1089 | ||
906c8d75 JA |
1090 | /* |
1091 | * We cannot pass the td data into a forked process, so attach the td and | |
1092 | * pass it to the thread worker. | |
1093 | */ | |
a6418147 | 1094 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1095 | { |
1096 | struct thread_data *td; | |
a6418147 | 1097 | void *data, *ret; |
ebac4655 JA |
1098 | |
1099 | data = shmat(shmid, NULL, 0); | |
1100 | if (data == (void *) -1) { | |
a6418147 JA |
1101 | int __err = errno; |
1102 | ||
ebac4655 | 1103 | perror("shmat"); |
a6418147 | 1104 | return __err; |
ebac4655 JA |
1105 | } |
1106 | ||
1107 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1108 | ret = thread_main(td); |
ebac4655 | 1109 | shmdt(data); |
43d76807 | 1110 | return (int) (unsigned long) ret; |
ebac4655 JA |
1111 | } |
1112 | ||
906c8d75 JA |
1113 | /* |
1114 | * Run over the job map and reap the threads that have exited, if any. | |
1115 | */ | |
ebac4655 JA |
1116 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1117 | { | |
34572e28 | 1118 | struct thread_data *td; |
1f809d15 | 1119 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1120 | |
1121 | /* | |
1122 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1123 | */ | |
1f809d15 | 1124 | realthreads = pending = cputhreads = 0; |
34572e28 | 1125 | for_each_td(td, i) { |
3707f45b | 1126 | int flags = 0; |
a2f77c9f | 1127 | |
84585003 JA |
1128 | /* |
1129 | * ->io_ops is NULL for a thread that has closed its | |
1130 | * io engine | |
1131 | */ | |
ba0fbe10 | 1132 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1133 | cputhreads++; |
1f809d15 JA |
1134 | else |
1135 | realthreads++; | |
b990b5c0 | 1136 | |
a1dedc1f AC |
1137 | if (!td->pid) { |
1138 | pending++; | |
1139 | continue; | |
1140 | } | |
1141 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1142 | continue; |
2dc1bbeb | 1143 | if (td->o.use_thread) { |
3707f45b JA |
1144 | if (td->runstate == TD_EXITED) { |
1145 | td_set_runstate(td, TD_REAPED); | |
1146 | goto reaped; | |
1147 | } | |
1148 | continue; | |
1149 | } | |
a2f77c9f JA |
1150 | |
1151 | flags = WNOHANG; | |
1152 | if (td->runstate == TD_EXITED) | |
1153 | flags = 0; | |
1154 | ||
1155 | /* | |
1156 | * check if someone quit or got killed in an unusual way | |
1157 | */ | |
1158 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1159 | if (ret < 0) { |
a2f77c9f | 1160 | if (errno == ECHILD) { |
5921e80c JA |
1161 | log_err("fio: pid=%d disappeared %d\n", |
1162 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1163 | td_set_runstate(td, TD_REAPED); |
1164 | goto reaped; | |
1165 | } | |
1166 | perror("waitpid"); | |
1167 | } else if (ret == td->pid) { | |
1168 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1169 | int sig = WTERMSIG(status); |
1170 | ||
d9244941 | 1171 | if (sig != SIGQUIT) |
5ec10eaa | 1172 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1173 | (int) td->pid, sig); |
fab6aa71 JA |
1174 | td_set_runstate(td, TD_REAPED); |
1175 | goto reaped; | |
1176 | } | |
a2f77c9f JA |
1177 | if (WIFEXITED(status)) { |
1178 | if (WEXITSTATUS(status) && !td->error) | |
1179 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1180 | |
a2f77c9f JA |
1181 | td_set_runstate(td, TD_REAPED); |
1182 | goto reaped; | |
a6418147 JA |
1183 | } |
1184 | } | |
ebac4655 | 1185 | |
a2f77c9f JA |
1186 | /* |
1187 | * thread is not dead, continue | |
1188 | */ | |
e48676ba | 1189 | pending++; |
a2f77c9f | 1190 | continue; |
fab6aa71 | 1191 | reaped: |
ebac4655 | 1192 | (*nr_running)--; |
2dc1bbeb JA |
1193 | (*m_rate) -= td->o.ratemin; |
1194 | (*t_rate) -= td->o.rate; | |
6f88bcb0 JA |
1195 | if (!td->pid) |
1196 | pending--; | |
a2f77c9f JA |
1197 | |
1198 | if (td->error) | |
1199 | exit_value++; | |
d3eeeabc JA |
1200 | |
1201 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1202 | } |
b990b5c0 | 1203 | |
1f809d15 | 1204 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1205 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1206 | } |
1207 | ||
906c8d75 JA |
1208 | /* |
1209 | * Main function for kicking off and reaping jobs, as needed. | |
1210 | */ | |
ebac4655 JA |
1211 | static void run_threads(void) |
1212 | { | |
ebac4655 JA |
1213 | struct thread_data *td; |
1214 | unsigned long spent; | |
1215 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1216 | |
2f9ade3c JA |
1217 | if (fio_pin_memory()) |
1218 | return; | |
ebac4655 | 1219 | |
c6ae0a5b | 1220 | if (!terse_output) { |
9cedf167 JA |
1221 | printf("Starting "); |
1222 | if (nr_thread) | |
5ec10eaa JA |
1223 | printf("%d thread%s", nr_thread, |
1224 | nr_thread > 1 ? "s" : ""); | |
9cedf167 JA |
1225 | if (nr_process) { |
1226 | if (nr_thread) | |
1227 | printf(" and "); | |
5ec10eaa JA |
1228 | printf("%d process%s", nr_process, |
1229 | nr_process > 1 ? "es" : ""); | |
9cedf167 JA |
1230 | } |
1231 | printf("\n"); | |
c6ae0a5b JA |
1232 | fflush(stdout); |
1233 | } | |
c04f7ec3 | 1234 | |
697a606c | 1235 | set_sig_handlers(); |
4efa970e | 1236 | |
ebac4655 JA |
1237 | todo = thread_number; |
1238 | nr_running = 0; | |
1239 | nr_started = 0; | |
1240 | m_rate = t_rate = 0; | |
1241 | ||
34572e28 | 1242 | for_each_td(td, i) { |
263e529f | 1243 | print_status_init(td->thread_number - 1); |
ebac4655 | 1244 | |
2dc1bbeb | 1245 | if (!td->o.create_serialize) { |
380cf265 | 1246 | init_disk_util(td); |
ebac4655 | 1247 | continue; |
380cf265 | 1248 | } |
ebac4655 JA |
1249 | |
1250 | /* | |
1251 | * do file setup here so it happens sequentially, | |
1252 | * we don't want X number of threads getting their | |
1253 | * client data interspersed on disk | |
1254 | */ | |
53cdc686 | 1255 | if (setup_files(td)) { |
5bf13a5a JA |
1256 | exit_value++; |
1257 | if (td->error) | |
5921e80c JA |
1258 | log_err("fio: pid=%d, err=%d/%s\n", |
1259 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1260 | td_set_runstate(td, TD_REAPED); |
1261 | todo--; | |
c69e8875 JA |
1262 | } else { |
1263 | struct fio_file *f; | |
1264 | unsigned int i; | |
1265 | ||
1266 | /* | |
1267 | * for sharing to work, each job must always open | |
1268 | * its own files. so close them, if we opened them | |
1269 | * for creation | |
1270 | */ | |
1271 | for_each_file(td, f, i) | |
1272 | td_io_close_file(td, f); | |
5ec10eaa | 1273 | } |
380cf265 JA |
1274 | |
1275 | init_disk_util(td); | |
ebac4655 JA |
1276 | } |
1277 | ||
a2f77c9f JA |
1278 | set_genesis_time(); |
1279 | ||
ebac4655 | 1280 | while (todo) { |
75154845 JA |
1281 | struct thread_data *map[MAX_JOBS]; |
1282 | struct timeval this_start; | |
1283 | int this_jobs = 0, left; | |
1284 | ||
ebac4655 JA |
1285 | /* |
1286 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1287 | */ | |
34572e28 | 1288 | for_each_td(td, i) { |
ebac4655 JA |
1289 | if (td->runstate != TD_NOT_CREATED) |
1290 | continue; | |
1291 | ||
1292 | /* | |
1293 | * never got a chance to start, killed by other | |
1294 | * thread for some reason | |
1295 | */ | |
1296 | if (td->terminate) { | |
1297 | todo--; | |
1298 | continue; | |
1299 | } | |
1300 | ||
2dc1bbeb | 1301 | if (td->o.start_delay) { |
263e529f | 1302 | spent = mtime_since_genesis(); |
ebac4655 | 1303 | |
2dc1bbeb | 1304 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1305 | continue; |
1306 | } | |
1307 | ||
6f88bcb0 JA |
1308 | if (td->o.stonewall && (nr_started || nr_running)) { |
1309 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1310 | td->o.name); | |
ebac4655 | 1311 | break; |
6f88bcb0 | 1312 | } |
ebac4655 | 1313 | |
75154845 JA |
1314 | /* |
1315 | * Set state to created. Thread will transition | |
1316 | * to TD_INITIALIZED when it's done setting up. | |
1317 | */ | |
ebac4655 | 1318 | td_set_runstate(td, TD_CREATED); |
75154845 | 1319 | map[this_jobs++] = td; |
ebac4655 JA |
1320 | nr_started++; |
1321 | ||
2dc1bbeb | 1322 | if (td->o.use_thread) { |
ee56ad50 | 1323 | dprint(FD_PROCESS, "will pthread_create\n"); |
5ec10eaa JA |
1324 | if (pthread_create(&td->thread, NULL, |
1325 | thread_main, td)) { | |
f4705a73 | 1326 | perror("pthread_create"); |
ebac4655 | 1327 | nr_started--; |
c8bb6faf | 1328 | break; |
ebac4655 | 1329 | } |
f4705a73 JA |
1330 | if (pthread_detach(td->thread) < 0) |
1331 | perror("pthread_detach"); | |
ebac4655 | 1332 | } else { |
5e1d306e | 1333 | pid_t pid; |
ee56ad50 | 1334 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1335 | pid = fork(); |
1336 | if (!pid) { | |
a6418147 JA |
1337 | int ret = fork_main(shm_id, i); |
1338 | ||
5e1d306e JA |
1339 | _exit(ret); |
1340 | } else if (i == fio_debug_jobno) | |
1341 | *fio_debug_jobp = pid; | |
ebac4655 | 1342 | } |
cdd18ad8 | 1343 | fio_mutex_down(startup_mutex); |
ebac4655 JA |
1344 | } |
1345 | ||
1346 | /* | |
75154845 JA |
1347 | * Wait for the started threads to transition to |
1348 | * TD_INITIALIZED. | |
ebac4655 | 1349 | */ |
02bcaa8c | 1350 | fio_gettime(&this_start, NULL); |
75154845 | 1351 | left = this_jobs; |
6ce15a32 | 1352 | while (left && !fio_abort) { |
75154845 JA |
1353 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1354 | break; | |
1355 | ||
1356 | usleep(100000); | |
1357 | ||
1358 | for (i = 0; i < this_jobs; i++) { | |
1359 | td = map[i]; | |
1360 | if (!td) | |
1361 | continue; | |
b6f4d880 | 1362 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1363 | map[i] = NULL; |
1364 | left--; | |
b6f4d880 JA |
1365 | } else if (td->runstate >= TD_EXITED) { |
1366 | map[i] = NULL; | |
1367 | left--; | |
1368 | todo--; | |
1369 | nr_running++; /* work-around... */ | |
75154845 JA |
1370 | } |
1371 | } | |
1372 | } | |
1373 | ||
1374 | if (left) { | |
3b70d7e5 | 1375 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1376 | for (i = 0; i < this_jobs; i++) { |
1377 | td = map[i]; | |
1378 | if (!td) | |
1379 | continue; | |
1380 | kill(td->pid, SIGTERM); | |
1381 | } | |
1382 | break; | |
1383 | } | |
1384 | ||
1385 | /* | |
b6f4d880 | 1386 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1387 | */ |
34572e28 | 1388 | for_each_td(td, i) { |
75154845 | 1389 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1390 | continue; |
1391 | ||
b29ee5b3 JA |
1392 | if (in_ramp_time(td)) |
1393 | td_set_runstate(td, TD_RAMP); | |
1394 | else | |
1395 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1396 | nr_running++; |
1397 | nr_started--; | |
2dc1bbeb JA |
1398 | m_rate += td->o.ratemin; |
1399 | t_rate += td->o.rate; | |
75154845 | 1400 | todo--; |
cdd18ad8 | 1401 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1402 | } |
1403 | ||
1404 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1405 | ||
1406 | if (todo) | |
1407 | usleep(100000); | |
1408 | } | |
1409 | ||
1410 | while (nr_running) { | |
1411 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1412 | usleep(10000); | |
1413 | } | |
1414 | ||
1415 | update_io_ticks(); | |
2f9ade3c | 1416 | fio_unpin_memory(); |
ebac4655 JA |
1417 | } |
1418 | ||
ebac4655 JA |
1419 | int main(int argc, char *argv[]) |
1420 | { | |
29d610e1 JA |
1421 | long ps; |
1422 | ||
2e5cdb11 JA |
1423 | sinit(); |
1424 | ||
dbe1125e JA |
1425 | /* |
1426 | * We need locale for number printing, if it isn't set then just | |
1427 | * go with the US format. | |
1428 | */ | |
1429 | if (!getenv("LC_NUMERIC")) | |
1430 | setlocale(LC_NUMERIC, "en_US"); | |
1431 | ||
ebac4655 JA |
1432 | if (parse_options(argc, argv)) |
1433 | return 1; | |
1434 | ||
4b472fa3 JA |
1435 | if (!thread_number) |
1436 | return 0; | |
ebac4655 | 1437 | |
29d610e1 JA |
1438 | ps = sysconf(_SC_PAGESIZE); |
1439 | if (ps < 0) { | |
1440 | log_err("Failed to get page size\n"); | |
1441 | return 1; | |
1442 | } | |
1443 | ||
cfc99db7 | 1444 | page_size = ps; |
29d610e1 JA |
1445 | page_mask = ps - 1; |
1446 | ||
bb3884d8 JA |
1447 | if (write_bw_log) { |
1448 | setup_log(&agg_io_log[DDIR_READ]); | |
1449 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1450 | } | |
1451 | ||
cdd18ad8 | 1452 | startup_mutex = fio_mutex_init(0); |
07739b57 | 1453 | |
a2f77c9f JA |
1454 | set_genesis_time(); |
1455 | ||
ccb0fa24 | 1456 | status_timer_arm(); |
ebac4655 JA |
1457 | |
1458 | run_threads(); | |
6ce15a32 | 1459 | |
bb3884d8 | 1460 | if (!fio_abort) { |
6ce15a32 | 1461 | show_run_stats(); |
bb3884d8 | 1462 | if (write_bw_log) { |
5ec10eaa JA |
1463 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1464 | __finish_log(agg_io_log[DDIR_WRITE], | |
1465 | "agg-write_bw.log"); | |
bb3884d8 JA |
1466 | } |
1467 | } | |
ebac4655 | 1468 | |
cdd18ad8 | 1469 | fio_mutex_remove(startup_mutex); |
437c9b71 | 1470 | return exit_value; |
ebac4655 | 1471 | } |