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; |
ebac4655 | 57 | |
bb3884d8 JA |
58 | struct io_log *agg_io_log[2]; |
59 | ||
ebac4655 | 60 | #define TERMINATE_ALL (-1) |
75154845 | 61 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 62 | |
6ce15a32 JA |
63 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
64 | { | |
d8fab1b6 JA |
65 | if (td->runstate == runstate) |
66 | return; | |
67 | ||
bd6f78b2 JA |
68 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", td->pid, td->runstate, |
69 | runstate); | |
6ce15a32 JA |
70 | td->runstate = runstate; |
71 | } | |
72 | ||
390c40e2 | 73 | static void terminate_threads(int group_id) |
ebac4655 | 74 | { |
34572e28 | 75 | struct thread_data *td; |
ebac4655 JA |
76 | int i; |
77 | ||
6f88bcb0 JA |
78 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
79 | ||
34572e28 | 80 | for_each_td(td, i) { |
ebac4655 | 81 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 JA |
82 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
83 | td->o.name, td->pid); | |
ad830ed7 JA |
84 | td->terminate = 1; |
85 | td->o.start_delay = 0; | |
86 | ||
7a93ab19 JA |
87 | /* |
88 | * if the thread is running, just let it exit | |
89 | */ | |
90 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 91 | kill(td->pid, SIGQUIT); |
f63f7f3b JA |
92 | else { |
93 | struct ioengine_ops *ops = td->io_ops; | |
94 | ||
95 | if (ops && (ops->flags & FIO_SIGQUIT)) | |
96 | kill(td->pid, SIGQUIT); | |
97 | } | |
ebac4655 JA |
98 | } |
99 | } | |
100 | } | |
101 | ||
102 | static void sig_handler(int sig) | |
103 | { | |
104 | switch (sig) { | |
5ec10eaa JA |
105 | case SIGALRM: |
106 | update_io_ticks(); | |
107 | disk_util_timer_arm(); | |
108 | print_thread_status(); | |
109 | break; | |
110 | default: | |
111 | printf("\nfio: terminating on signal %d\n", sig); | |
112 | fflush(stdout); | |
113 | terminate_threads(TERMINATE_ALL); | |
114 | break; | |
ebac4655 JA |
115 | } |
116 | } | |
117 | ||
906c8d75 JA |
118 | /* |
119 | * Check if we are above the minimum rate given. | |
120 | */ | |
ebac4655 JA |
121 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
122 | { | |
0904200b | 123 | unsigned long long bytes = 0; |
4e991c23 | 124 | unsigned long iops = 0; |
ebac4655 JA |
125 | unsigned long spent; |
126 | unsigned long rate; | |
ebac4655 | 127 | |
780bf1a1 JA |
128 | /* |
129 | * No minimum rate set, always ok | |
130 | */ | |
2dc1bbeb | 131 | if (!td->o.ratemin && !td->o.rate_iops_min) |
780bf1a1 JA |
132 | return 0; |
133 | ||
ebac4655 JA |
134 | /* |
135 | * allow a 2 second settle period in the beginning | |
136 | */ | |
137 | if (mtime_since(&td->start, now) < 2000) | |
138 | return 0; | |
139 | ||
4e991c23 JA |
140 | if (td_read(td)) { |
141 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 142 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
143 | } |
144 | if (td_write(td)) { | |
145 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 146 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 147 | } |
0904200b | 148 | |
ebac4655 JA |
149 | /* |
150 | * if rate blocks is set, sample is running | |
151 | */ | |
4e991c23 | 152 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 153 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 154 | if (spent < td->o.ratecycle) |
ebac4655 JA |
155 | return 0; |
156 | ||
2dc1bbeb | 157 | if (td->o.rate) { |
4e991c23 JA |
158 | /* |
159 | * check bandwidth specified rate | |
160 | */ | |
161 | if (bytes < td->rate_bytes) { | |
5ec10eaa JA |
162 | log_err("%s: min rate %u not met\n", td->o.name, |
163 | td->o.ratemin); | |
4e991c23 JA |
164 | return 1; |
165 | } else { | |
166 | rate = (bytes - td->rate_bytes) / spent; | |
5ec10eaa JA |
167 | if (rate < td->o.ratemin || |
168 | bytes < td->rate_bytes) { | |
169 | log_err("%s: min rate %u not met, got" | |
170 | " %luKiB/sec\n", td->o.name, | |
171 | td->o.ratemin, rate); | |
4e991c23 JA |
172 | return 1; |
173 | } | |
174 | } | |
413dd459 | 175 | } else { |
4e991c23 JA |
176 | /* |
177 | * checks iops specified rate | |
178 | */ | |
2dc1bbeb | 179 | if (iops < td->o.rate_iops) { |
5ec10eaa JA |
180 | log_err("%s: min iops rate %u not met\n", |
181 | td->o.name, td->o.rate_iops); | |
413dd459 | 182 | return 1; |
4e991c23 JA |
183 | } else { |
184 | rate = (iops - td->rate_blocks) / spent; | |
5ec10eaa JA |
185 | if (rate < td->o.rate_iops_min || |
186 | iops < td->rate_blocks) { | |
187 | log_err("%s: min iops rate %u not met," | |
188 | " got %lu\n", td->o.name, | |
189 | td->o.rate_iops_min, | |
190 | rate); | |
4e991c23 | 191 | } |
413dd459 | 192 | } |
ebac4655 JA |
193 | } |
194 | } | |
195 | ||
0904200b | 196 | td->rate_bytes = bytes; |
4e991c23 | 197 | td->rate_blocks = iops; |
ebac4655 JA |
198 | memcpy(&td->lastrate, now, sizeof(*now)); |
199 | return 0; | |
200 | } | |
201 | ||
202 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
203 | { | |
2dc1bbeb | 204 | if (!td->o.timeout) |
ebac4655 | 205 | return 0; |
2dc1bbeb | 206 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
207 | return 1; |
208 | ||
209 | return 0; | |
210 | } | |
211 | ||
906c8d75 JA |
212 | /* |
213 | * When job exits, we can cancel the in-flight IO if we are using async | |
214 | * io. Attempt to do so. | |
215 | */ | |
ebac4655 JA |
216 | static void cleanup_pending_aio(struct thread_data *td) |
217 | { | |
ebac4655 | 218 | struct list_head *entry, *n; |
ebac4655 JA |
219 | struct io_u *io_u; |
220 | int r; | |
221 | ||
222 | /* | |
223 | * get immediately available events, if any | |
224 | */ | |
d7762cf8 | 225 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
226 | if (r < 0) |
227 | return; | |
ebac4655 JA |
228 | |
229 | /* | |
230 | * now cancel remaining active events | |
231 | */ | |
2866c82d | 232 | if (td->io_ops->cancel) { |
ebac4655 JA |
233 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
234 | io_u = list_entry(entry, struct io_u, list); | |
235 | ||
0c6e7517 JA |
236 | /* |
237 | * if the io_u isn't in flight, then that generally | |
238 | * means someone leaked an io_u. complain but fix | |
239 | * it up, so we don't stall here. | |
240 | */ | |
241 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
242 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 243 | put_io_u(td, io_u); |
0c6e7517 JA |
244 | } else { |
245 | r = td->io_ops->cancel(td, io_u); | |
246 | if (!r) | |
247 | put_io_u(td, io_u); | |
248 | } | |
ebac4655 JA |
249 | } |
250 | } | |
251 | ||
97601024 | 252 | if (td->cur_depth) |
d7762cf8 | 253 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
254 | } |
255 | ||
858a3d47 JA |
256 | /* |
257 | * Helper to handle the final sync of a file. Works just like the normal | |
258 | * io path, just does everything sync. | |
259 | */ | |
260 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
261 | { | |
262 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
263 | int ret; |
264 | ||
265 | if (!io_u) | |
266 | return 1; | |
267 | ||
268 | io_u->ddir = DDIR_SYNC; | |
269 | io_u->file = f; | |
270 | ||
271 | if (td_io_prep(td, io_u)) { | |
272 | put_io_u(td, io_u); | |
273 | return 1; | |
274 | } | |
275 | ||
755200a3 | 276 | requeue: |
858a3d47 | 277 | ret = td_io_queue(td, io_u); |
36167d82 | 278 | if (ret < 0) { |
e1161c32 | 279 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 280 | put_io_u(td, io_u); |
858a3d47 | 281 | return 1; |
36167d82 | 282 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 283 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 284 | return 1; |
36167d82 JA |
285 | } else if (ret == FIO_Q_COMPLETED) { |
286 | if (io_u->error) { | |
e1161c32 | 287 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
288 | return 1; |
289 | } | |
858a3d47 | 290 | |
d7762cf8 | 291 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 292 | return 1; |
755200a3 JA |
293 | } else if (ret == FIO_Q_BUSY) { |
294 | if (td_io_commit(td)) | |
295 | return 1; | |
296 | goto requeue; | |
858a3d47 JA |
297 | } |
298 | ||
299 | return 0; | |
300 | } | |
301 | ||
906c8d75 | 302 | /* |
34403fb1 | 303 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
304 | * reads the blocks back in, and checks the crc/md5 of the data. |
305 | */ | |
1e97cce9 | 306 | static void do_verify(struct thread_data *td) |
ebac4655 | 307 | { |
53cdc686 | 308 | struct fio_file *f; |
36167d82 | 309 | struct io_u *io_u; |
af52b345 JA |
310 | int ret, min_events; |
311 | unsigned int i; | |
e5b401d4 JA |
312 | |
313 | /* | |
314 | * sync io first and invalidate cache, to make sure we really | |
315 | * read from disk. | |
316 | */ | |
317 | for_each_file(td, f, i) { | |
3d7b485f JA |
318 | if (!(f->flags & FIO_FILE_OPEN)) |
319 | continue; | |
b2fdda43 JA |
320 | if (fio_io_sync(td, f)) |
321 | break; | |
322 | if (file_invalidate_cache(td, f)) | |
323 | break; | |
e5b401d4 | 324 | } |
ebac4655 | 325 | |
b2fdda43 JA |
326 | if (td->error) |
327 | return; | |
328 | ||
ebac4655 JA |
329 | td_set_runstate(td, TD_VERIFYING); |
330 | ||
36167d82 JA |
331 | io_u = NULL; |
332 | while (!td->terminate) { | |
5451792e JA |
333 | int ret2; |
334 | ||
ebac4655 JA |
335 | io_u = __get_io_u(td); |
336 | if (!io_u) | |
337 | break; | |
338 | ||
069c2918 JA |
339 | if (runtime_exceeded(td, &io_u->start_time)) { |
340 | put_io_u(td, io_u); | |
90a87d4b | 341 | td->terminate = 1; |
02bcaa8c | 342 | break; |
069c2918 | 343 | } |
02bcaa8c | 344 | |
069c2918 JA |
345 | if (get_next_verify(td, io_u)) { |
346 | put_io_u(td, io_u); | |
ebac4655 | 347 | break; |
069c2918 | 348 | } |
ebac4655 | 349 | |
069c2918 JA |
350 | if (td_io_prep(td, io_u)) { |
351 | put_io_u(td, io_u); | |
53cdc686 | 352 | break; |
069c2918 | 353 | } |
d7762cf8 JA |
354 | |
355 | io_u->end_io = verify_io_u; | |
53cdc686 | 356 | |
11786802 | 357 | ret = td_io_queue(td, io_u); |
36167d82 JA |
358 | switch (ret) { |
359 | case FIO_Q_COMPLETED: | |
360 | if (io_u->error) | |
22819ec2 | 361 | ret = -io_u->error; |
9e9d164e | 362 | else if (io_u->resid) { |
36167d82 | 363 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 364 | struct fio_file *f = io_u->file; |
ebac4655 | 365 | |
9e9d164e JA |
366 | /* |
367 | * zero read, fail | |
368 | */ | |
369 | if (!bytes) { | |
370 | td_verror(td, ENODATA, "full resid"); | |
371 | put_io_u(td, io_u); | |
372 | break; | |
373 | } | |
8400d9b2 | 374 | |
36167d82 JA |
375 | io_u->xfer_buflen = io_u->resid; |
376 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
377 | io_u->offset += bytes; |
378 | ||
30061b97 JA |
379 | td->ts.short_io_u[io_u->ddir]++; |
380 | ||
d460eb31 | 381 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
382 | goto sync_done; |
383 | ||
11786802 JA |
384 | requeue_io_u(td, &io_u); |
385 | } else { | |
8400d9b2 | 386 | sync_done: |
11786802 JA |
387 | ret = io_u_sync_complete(td, io_u); |
388 | if (ret < 0) | |
389 | break; | |
36167d82 | 390 | } |
36167d82 JA |
391 | continue; |
392 | case FIO_Q_QUEUED: | |
393 | break; | |
755200a3 JA |
394 | case FIO_Q_BUSY: |
395 | requeue_io_u(td, &io_u); | |
5451792e JA |
396 | ret2 = td_io_commit(td); |
397 | if (ret2 < 0) | |
398 | ret = ret2; | |
755200a3 | 399 | break; |
36167d82 JA |
400 | default: |
401 | assert(ret < 0); | |
e1161c32 | 402 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
403 | break; |
404 | } | |
405 | ||
99784632 | 406 | if (ret < 0 || td->error) |
3af6ef39 JA |
407 | break; |
408 | ||
ebac4655 | 409 | /* |
3af6ef39 JA |
410 | * if we can queue more, do so. but check if there are |
411 | * completed io_u's first. | |
ebac4655 | 412 | */ |
97601024 | 413 | min_events = 0; |
e916b390 | 414 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
3af6ef39 | 415 | min_events = 1; |
3af6ef39 | 416 | |
2dc1bbeb JA |
417 | if (td->cur_depth > td->o.iodepth_low) |
418 | min_events = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
419 | } |
420 | ||
3af6ef39 JA |
421 | /* |
422 | * Reap required number of io units, if any, and do the | |
423 | * verification on them through the callback handler | |
424 | */ | |
d7762cf8 | 425 | if (io_u_queued_complete(td, min_events) < 0) |
ebac4655 | 426 | break; |
36167d82 | 427 | } |
ebac4655 | 428 | |
c01c0395 JA |
429 | if (!td->error) { |
430 | min_events = td->cur_depth; | |
431 | ||
432 | if (min_events) | |
433 | ret = io_u_queued_complete(td, min_events); | |
434 | } else | |
ebac4655 JA |
435 | cleanup_pending_aio(td); |
436 | ||
437 | td_set_runstate(td, TD_RUNNING); | |
438 | } | |
439 | ||
32cd46a0 | 440 | /* |
906c8d75 | 441 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
442 | * and reaps them, checking for rate and errors along the way. |
443 | */ | |
ebac4655 JA |
444 | static void do_io(struct thread_data *td) |
445 | { | |
02bcaa8c | 446 | struct timeval s; |
ebac4655 | 447 | unsigned long usec; |
af52b345 JA |
448 | unsigned int i; |
449 | int ret = 0; | |
ebac4655 | 450 | |
5853e5a8 JA |
451 | td_set_runstate(td, TD_RUNNING); |
452 | ||
7bb48f84 | 453 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
454 | struct timeval comp_time; |
455 | long bytes_done = 0; | |
84585003 | 456 | int min_evts = 0; |
ebac4655 | 457 | struct io_u *io_u; |
5451792e | 458 | int ret2; |
ebac4655 JA |
459 | |
460 | if (td->terminate) | |
461 | break; | |
462 | ||
3d7c391d | 463 | io_u = get_io_u(td); |
ebac4655 JA |
464 | if (!io_u) |
465 | break; | |
466 | ||
467 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
468 | |
469 | if (runtime_exceeded(td, &s)) { | |
470 | put_io_u(td, io_u); | |
90a87d4b | 471 | td->terminate = 1; |
97601024 JA |
472 | break; |
473 | } | |
36167d82 | 474 | |
b67740d3 JA |
475 | /* |
476 | * Add verification end_io handler, if asked to verify | |
477 | * a previously written file. | |
478 | */ | |
cc3fc855 | 479 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) { |
b67740d3 | 480 | io_u->end_io = verify_io_u; |
d8fab1b6 JA |
481 | td_set_runstate(td, TD_VERIFYING); |
482 | } else | |
483 | td_set_runstate(td, TD_RUNNING); | |
b67740d3 | 484 | |
11786802 | 485 | ret = td_io_queue(td, io_u); |
36167d82 JA |
486 | switch (ret) { |
487 | case FIO_Q_COMPLETED: | |
5451792e JA |
488 | if (io_u->error) |
489 | ret = -io_u->error; | |
9e9d164e | 490 | else if (io_u->resid) { |
36167d82 | 491 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 492 | struct fio_file *f = io_u->file; |
36167d82 | 493 | |
9e9d164e JA |
494 | /* |
495 | * zero read, fail | |
496 | */ | |
497 | if (!bytes) { | |
498 | td_verror(td, ENODATA, "full resid"); | |
499 | put_io_u(td, io_u); | |
500 | break; | |
501 | } | |
502 | ||
cec6b55d | 503 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 504 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
505 | io_u->offset += bytes; |
506 | ||
30061b97 JA |
507 | td->ts.short_io_u[io_u->ddir]++; |
508 | ||
d460eb31 | 509 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
510 | goto sync_done; |
511 | ||
11786802 JA |
512 | requeue_io_u(td, &io_u); |
513 | } else { | |
5a7c5680 | 514 | sync_done: |
11786802 JA |
515 | fio_gettime(&comp_time, NULL); |
516 | bytes_done = io_u_sync_complete(td, io_u); | |
517 | if (bytes_done < 0) | |
518 | ret = bytes_done; | |
cec6b55d | 519 | } |
36167d82 JA |
520 | break; |
521 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
522 | /* |
523 | * if the engine doesn't have a commit hook, | |
524 | * the io_u is really queued. if it does have such | |
525 | * a hook, it has to call io_u_queued() itself. | |
526 | */ | |
527 | if (td->io_ops->commit == NULL) | |
528 | io_u_queued(td, io_u); | |
36167d82 | 529 | break; |
755200a3 JA |
530 | case FIO_Q_BUSY: |
531 | requeue_io_u(td, &io_u); | |
5451792e JA |
532 | ret2 = td_io_commit(td); |
533 | if (ret2 < 0) | |
534 | ret = ret2; | |
755200a3 | 535 | break; |
36167d82 JA |
536 | default: |
537 | assert(ret < 0); | |
538 | put_io_u(td, io_u); | |
539 | break; | |
ebac4655 JA |
540 | } |
541 | ||
99784632 | 542 | if (ret < 0 || td->error) |
36167d82 JA |
543 | break; |
544 | ||
97601024 JA |
545 | /* |
546 | * See if we need to complete some commands | |
547 | */ | |
755200a3 | 548 | if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) { |
97601024 | 549 | min_evts = 0; |
e916b390 | 550 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
36167d82 | 551 | min_evts = 1; |
ebac4655 | 552 | |
2dc1bbeb | 553 | if (td->cur_depth > td->o.iodepth_low) |
5ec10eaa JA |
554 | min_evts = td->cur_depth |
555 | - td->o.iodepth_low; | |
e916b390 JA |
556 | } |
557 | ||
97601024 | 558 | fio_gettime(&comp_time, NULL); |
d7762cf8 | 559 | bytes_done = io_u_queued_complete(td, min_evts); |
97601024 | 560 | if (bytes_done < 0) |
36167d82 | 561 | break; |
ebac4655 JA |
562 | } |
563 | ||
97601024 JA |
564 | if (!bytes_done) |
565 | continue; | |
566 | ||
ebac4655 JA |
567 | /* |
568 | * the rate is batched for now, it should work for batches | |
569 | * of completions except the very first one which may look | |
570 | * a little bursty | |
571 | */ | |
97601024 | 572 | usec = utime_since(&s, &comp_time); |
ebac4655 | 573 | |
413dd459 | 574 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 575 | |
97601024 | 576 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 577 | if (exitall_on_terminate) |
390c40e2 | 578 | terminate_threads(td->groupid); |
e1161c32 | 579 | td_verror(td, ENODATA, "check_min_rate"); |
ebac4655 JA |
580 | break; |
581 | } | |
582 | ||
2dc1bbeb | 583 | if (td->o.thinktime) { |
9c1f7434 JA |
584 | unsigned long long b; |
585 | ||
586 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 587 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
588 | int left; |
589 | ||
2dc1bbeb JA |
590 | if (td->o.thinktime_spin) |
591 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 592 | |
2dc1bbeb | 593 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
594 | if (left) |
595 | usec_sleep(td, left); | |
596 | } | |
9c1f7434 | 597 | } |
ebac4655 JA |
598 | } |
599 | ||
aa31f1f1 SL |
600 | if (td->o.fill_device && td->error == ENOSPC) { |
601 | td->error = 0; | |
602 | td->terminate = 1; | |
603 | } | |
4d2413c6 | 604 | if (!td->error) { |
3d7c391d JA |
605 | struct fio_file *f; |
606 | ||
c01c0395 JA |
607 | i = td->cur_depth; |
608 | if (i) | |
609 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 610 | |
2dc1bbeb | 611 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 612 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
613 | |
614 | for_each_file(td, f, i) { | |
615 | if (!(f->flags & FIO_FILE_OPEN)) | |
616 | continue; | |
858a3d47 | 617 | fio_io_sync(td, f); |
3d7b485f | 618 | } |
84585003 | 619 | } |
c01c0395 JA |
620 | } else |
621 | cleanup_pending_aio(td); | |
2ba1c290 JA |
622 | |
623 | /* | |
624 | * stop job if we failed doing any IO | |
625 | */ | |
626 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
627 | td->done = 1; | |
ebac4655 JA |
628 | } |
629 | ||
ebac4655 JA |
630 | static void cleanup_io_u(struct thread_data *td) |
631 | { | |
632 | struct list_head *entry, *n; | |
633 | struct io_u *io_u; | |
634 | ||
635 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
636 | io_u = list_entry(entry, struct io_u, list); | |
637 | ||
638 | list_del(&io_u->list); | |
639 | free(io_u); | |
640 | } | |
641 | ||
2f9ade3c | 642 | free_io_mem(td); |
ebac4655 JA |
643 | } |
644 | ||
6b9cea23 JA |
645 | /* |
646 | * "randomly" fill the buffer contents | |
647 | */ | |
e9459e5a | 648 | static void fill_io_buf(struct thread_data *td, struct io_u *io_u, int max_bs) |
6b9cea23 | 649 | { |
210f43b3 | 650 | long *ptr = io_u->buf; |
6b9cea23 | 651 | |
e9459e5a JA |
652 | if (!td->o.zero_buffers) { |
653 | while ((void *) ptr - io_u->buf < max_bs) { | |
210f43b3 | 654 | *ptr = rand() * GOLDEN_RATIO_PRIME; |
e9459e5a JA |
655 | ptr++; |
656 | } | |
657 | } else | |
658 | memset(ptr, 0, max_bs); | |
6b9cea23 JA |
659 | } |
660 | ||
ebac4655 JA |
661 | static int init_io_u(struct thread_data *td) |
662 | { | |
663 | struct io_u *io_u; | |
a00735e6 | 664 | unsigned int max_bs; |
ebac4655 JA |
665 | int i, max_units; |
666 | char *p; | |
667 | ||
1e3c09af | 668 | max_units = td->o.iodepth; |
2dc1bbeb | 669 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
670 | td->orig_buffer_size = (unsigned long long) max_bs |
671 | * (unsigned long long) max_units; | |
672 | ||
673 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
674 | unsigned long bs; | |
74b025b0 | 675 | |
5ec10eaa JA |
676 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
677 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
678 | } | |
ebac4655 | 679 | |
c3990645 JA |
680 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
681 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
682 | return 1; | |
683 | } | |
684 | ||
2f9ade3c JA |
685 | if (allocate_io_mem(td)) |
686 | return 1; | |
ebac4655 | 687 | |
b3f378e9 JA |
688 | if (td->o.odirect) |
689 | p = ALIGN(td->orig_buffer); | |
690 | else | |
691 | p = td->orig_buffer; | |
692 | ||
ebac4655 | 693 | for (i = 0; i < max_units; i++) { |
db1cd90b JA |
694 | if (td->terminate) |
695 | return 1; | |
ebac4655 JA |
696 | io_u = malloc(sizeof(*io_u)); |
697 | memset(io_u, 0, sizeof(*io_u)); | |
698 | INIT_LIST_HEAD(&io_u->list); | |
699 | ||
b4c5e1ac JA |
700 | if (!(td->io_ops->flags & FIO_NOIO)) { |
701 | io_u->buf = p + max_bs * i; | |
e9459e5a | 702 | |
b4c5e1ac JA |
703 | if (td_write(td)) |
704 | fill_io_buf(td, io_u, max_bs); | |
705 | } | |
6b9cea23 | 706 | |
b1ff3403 | 707 | io_u->index = i; |
0c6e7517 | 708 | io_u->flags = IO_U_F_FREE; |
ebac4655 JA |
709 | list_add(&io_u->list, &td->io_u_freelist); |
710 | } | |
711 | ||
433afcb4 JA |
712 | io_u_init_timeout(); |
713 | ||
ebac4655 JA |
714 | return 0; |
715 | } | |
716 | ||
da86774e JA |
717 | static int switch_ioscheduler(struct thread_data *td) |
718 | { | |
719 | char tmp[256], tmp2[128]; | |
720 | FILE *f; | |
721 | int ret; | |
722 | ||
ba0fbe10 | 723 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
724 | return 0; |
725 | ||
da86774e JA |
726 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
727 | ||
728 | f = fopen(tmp, "r+"); | |
729 | if (!f) { | |
cbf5c1d7 | 730 | if (errno == ENOENT) { |
5ec10eaa JA |
731 | log_err("fio: os or kernel doesn't support IO scheduler" |
732 | " switching\n"); | |
cbf5c1d7 JA |
733 | return 0; |
734 | } | |
735 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
736 | return 1; |
737 | } | |
738 | ||
739 | /* | |
740 | * Set io scheduler. | |
741 | */ | |
2dc1bbeb | 742 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 743 | if (ferror(f) || ret != 1) { |
e1161c32 | 744 | td_verror(td, errno, "fwrite"); |
da86774e JA |
745 | fclose(f); |
746 | return 1; | |
747 | } | |
748 | ||
749 | rewind(f); | |
750 | ||
751 | /* | |
752 | * Read back and check that the selected scheduler is now the default. | |
753 | */ | |
754 | ret = fread(tmp, 1, sizeof(tmp), f); | |
755 | if (ferror(f) || ret < 0) { | |
e1161c32 | 756 | td_verror(td, errno, "fread"); |
da86774e JA |
757 | fclose(f); |
758 | return 1; | |
759 | } | |
760 | ||
2dc1bbeb | 761 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 762 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 763 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 764 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
765 | fclose(f); |
766 | return 1; | |
767 | } | |
768 | ||
769 | fclose(f); | |
770 | return 0; | |
771 | } | |
772 | ||
492158cf JA |
773 | static int keep_running(struct thread_data *td) |
774 | { | |
775 | unsigned long long io_done; | |
776 | ||
20e354ef JA |
777 | if (td->done) |
778 | return 0; | |
492158cf JA |
779 | if (td->o.time_based) |
780 | return 1; | |
781 | if (td->o.loops) { | |
782 | td->o.loops--; | |
783 | return 1; | |
784 | } | |
785 | ||
5ec10eaa JA |
786 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
787 | + td->io_skip_bytes; | |
492158cf JA |
788 | if (io_done < td->o.size) |
789 | return 1; | |
790 | ||
791 | return 0; | |
792 | } | |
793 | ||
a978ba68 | 794 | static int clear_io_state(struct thread_data *td) |
ebac4655 | 795 | { |
53cdc686 | 796 | struct fio_file *f; |
af52b345 JA |
797 | unsigned int i; |
798 | int ret; | |
ebac4655 | 799 | |
756867bd | 800 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 801 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 802 | td->zone_bytes = 0; |
d7d3b49b | 803 | td->rate_bytes = 0; |
4e991c23 | 804 | td->rate_blocks = 0; |
38d77cae | 805 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 806 | |
c1324df1 JA |
807 | td->last_was_sync = 0; |
808 | ||
2ba1c290 JA |
809 | /* |
810 | * reset file done count if we are to start over | |
811 | */ | |
812 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 813 | td->nr_done_files = 0; |
281f9b63 | 814 | |
24ffd2c2 | 815 | close_files(td); |
53cdc686 | 816 | |
a978ba68 JA |
817 | ret = 0; |
818 | for_each_file(td, f, i) { | |
281f9b63 | 819 | f->flags &= ~FIO_FILE_DONE; |
a978ba68 JA |
820 | ret = td_io_open_file(td, f); |
821 | if (ret) | |
822 | break; | |
53cdc686 | 823 | } |
a978ba68 JA |
824 | |
825 | return ret; | |
ebac4655 JA |
826 | } |
827 | ||
906c8d75 JA |
828 | /* |
829 | * Entry point for the thread based jobs. The process based jobs end up | |
830 | * here as well, after a little setup. | |
831 | */ | |
ebac4655 JA |
832 | static void *thread_main(void *data) |
833 | { | |
10927316 | 834 | unsigned long long runtime[2], elapsed; |
ebac4655 | 835 | struct thread_data *td = data; |
a978ba68 | 836 | int clear_state; |
ebac4655 | 837 | |
2dc1bbeb | 838 | if (!td->o.use_thread) |
ebac4655 JA |
839 | setsid(); |
840 | ||
841 | td->pid = getpid(); | |
842 | ||
ee56ad50 JA |
843 | dprint(FD_PROCESS, "jobs pid=%d started\n", td->pid); |
844 | ||
aea47d44 JA |
845 | INIT_LIST_HEAD(&td->io_u_freelist); |
846 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 847 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 | 848 | INIT_LIST_HEAD(&td->io_log_list); |
8de8f047 | 849 | INIT_LIST_HEAD(&td->io_hist_list); |
bb5d7d0b | 850 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 851 | |
db1cd90b | 852 | td_set_runstate(td, TD_INITIALIZED); |
cdd18ad8 JA |
853 | fio_mutex_up(startup_mutex); |
854 | fio_mutex_down(td->mutex); | |
db1cd90b JA |
855 | |
856 | /* | |
cdd18ad8 | 857 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
858 | * eating a file descriptor |
859 | */ | |
cdd18ad8 | 860 | fio_mutex_remove(td->mutex); |
db1cd90b | 861 | |
d84f8d49 JA |
862 | /* |
863 | * May alter parameters that init_io_u() will use, so we need to | |
864 | * do this first. | |
865 | */ | |
866 | if (init_iolog(td)) | |
867 | goto err; | |
868 | ||
ebac4655 | 869 | if (init_io_u(td)) |
db1cd90b | 870 | goto err; |
ebac4655 | 871 | |
375b2695 | 872 | if (td->o.cpumask_set && fio_setaffinity(td) == -1) { |
e1161c32 | 873 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 874 | goto err; |
ebac4655 JA |
875 | } |
876 | ||
ac684785 | 877 | if (td->ioprio_set) { |
ebac4655 | 878 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 879 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 880 | goto err; |
ebac4655 JA |
881 | } |
882 | } | |
883 | ||
2dc1bbeb | 884 | if (nice(td->o.nice) == -1) { |
e1161c32 | 885 | td_verror(td, errno, "nice"); |
db1cd90b | 886 | goto err; |
b6f4d880 JA |
887 | } |
888 | ||
2dc1bbeb | 889 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 890 | goto err; |
37f56873 | 891 | |
2dc1bbeb | 892 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
893 | goto err; |
894 | ||
7d6c5283 JA |
895 | if (td_io_init(td)) |
896 | goto err; | |
897 | ||
b5af8293 JA |
898 | if (open_files(td)) |
899 | goto err; | |
900 | ||
68727076 JA |
901 | if (init_random_map(td)) |
902 | goto err; | |
903 | ||
2dc1bbeb JA |
904 | if (td->o.exec_prerun) { |
905 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
906 | goto err; |
907 | } | |
4e0ba8af | 908 | |
69008999 | 909 | fio_gettime(&td->epoch, NULL); |
433afcb4 | 910 | memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); |
756867bd | 911 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
912 | |
913 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 914 | clear_state = 0; |
492158cf | 915 | while (keep_running(td)) { |
02bcaa8c | 916 | fio_gettime(&td->start, NULL); |
756867bd | 917 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 | 918 | |
2dc1bbeb | 919 | if (td->o.ratemin) |
5ec10eaa JA |
920 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
921 | sizeof(td->lastrate)); | |
ebac4655 | 922 | |
a978ba68 JA |
923 | if (clear_state && clear_io_state(td)) |
924 | break; | |
925 | ||
ebac4655 JA |
926 | prune_io_piece_log(td); |
927 | ||
ba0fbe10 | 928 | do_io(td); |
ebac4655 | 929 | |
a978ba68 JA |
930 | clear_state = 1; |
931 | ||
38d77cae JA |
932 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
933 | if (td->rw_end_set[DDIR_READ]) | |
5ec10eaa JA |
934 | elapsed = utime_since(&td->start, |
935 | &td->rw_end[DDIR_READ]); | |
38d77cae JA |
936 | else |
937 | elapsed = utime_since_now(&td->start); | |
938 | ||
939 | runtime[DDIR_READ] += elapsed; | |
940 | } | |
941 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
942 | if (td->rw_end_set[DDIR_WRITE]) | |
5ec10eaa JA |
943 | elapsed = utime_since(&td->start, |
944 | &td->rw_end[DDIR_WRITE]); | |
38d77cae JA |
945 | else |
946 | elapsed = utime_since_now(&td->start); | |
947 | ||
948 | runtime[DDIR_WRITE] += elapsed; | |
949 | } | |
5ec10eaa | 950 | |
ebac4655 JA |
951 | if (td->error || td->terminate) |
952 | break; | |
953 | ||
e84c73a8 SL |
954 | if (!td->o.do_verify || |
955 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 956 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
957 | continue; |
958 | ||
a978ba68 JA |
959 | if (clear_io_state(td)) |
960 | break; | |
961 | ||
02bcaa8c | 962 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
963 | |
964 | do_verify(td); | |
965 | ||
69008999 | 966 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
967 | |
968 | if (td->error || td->terminate) | |
969 | break; | |
970 | } | |
971 | ||
36dff966 | 972 | update_rusage_stat(td); |
47f0cc48 YHJT |
973 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
974 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
975 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
976 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
977 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
978 | ||
979 | if (td->ts.bw_log) | |
980 | finish_log(td, td->ts.bw_log, "bw"); | |
981 | if (td->ts.slat_log) | |
982 | finish_log(td, td->ts.slat_log, "slat"); | |
983 | if (td->ts.clat_log) | |
984 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb JA |
985 | if (td->o.exec_postrun) { |
986 | if (system(td->o.exec_postrun) < 0) | |
987 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 988 | } |
ebac4655 JA |
989 | |
990 | if (exitall_on_terminate) | |
390c40e2 | 991 | terminate_threads(td->groupid); |
ebac4655 JA |
992 | |
993 | err: | |
5bf13a5a | 994 | if (td->error) |
5ec10eaa JA |
995 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, |
996 | td->verror); | |
24ffd2c2 | 997 | close_and_free_files(td); |
2866c82d | 998 | close_ioengine(td); |
ebac4655 | 999 | cleanup_io_u(td); |
f29b25a3 JA |
1000 | |
1001 | /* | |
1002 | * do this very late, it will log file closing as well | |
1003 | */ | |
1004 | if (td->o.write_iolog_file) | |
1005 | write_iolog_close(td); | |
1006 | ||
d23bb327 | 1007 | options_mem_free(td); |
ebac4655 | 1008 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1009 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1010 | } |
1011 | ||
906c8d75 JA |
1012 | /* |
1013 | * We cannot pass the td data into a forked process, so attach the td and | |
1014 | * pass it to the thread worker. | |
1015 | */ | |
a6418147 | 1016 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1017 | { |
1018 | struct thread_data *td; | |
a6418147 | 1019 | void *data, *ret; |
ebac4655 JA |
1020 | |
1021 | data = shmat(shmid, NULL, 0); | |
1022 | if (data == (void *) -1) { | |
a6418147 JA |
1023 | int __err = errno; |
1024 | ||
ebac4655 | 1025 | perror("shmat"); |
a6418147 | 1026 | return __err; |
ebac4655 JA |
1027 | } |
1028 | ||
1029 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1030 | ret = thread_main(td); |
ebac4655 | 1031 | shmdt(data); |
43d76807 | 1032 | return (int) (unsigned long) ret; |
ebac4655 JA |
1033 | } |
1034 | ||
906c8d75 JA |
1035 | /* |
1036 | * Run over the job map and reap the threads that have exited, if any. | |
1037 | */ | |
ebac4655 JA |
1038 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1039 | { | |
34572e28 | 1040 | struct thread_data *td; |
1f809d15 | 1041 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1042 | |
1043 | /* | |
1044 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1045 | */ | |
1f809d15 | 1046 | realthreads = pending = cputhreads = 0; |
34572e28 | 1047 | for_each_td(td, i) { |
3707f45b | 1048 | int flags = 0; |
a2f77c9f | 1049 | |
84585003 JA |
1050 | /* |
1051 | * ->io_ops is NULL for a thread that has closed its | |
1052 | * io engine | |
1053 | */ | |
ba0fbe10 | 1054 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1055 | cputhreads++; |
1f809d15 JA |
1056 | else |
1057 | realthreads++; | |
b990b5c0 | 1058 | |
a1dedc1f AC |
1059 | if (!td->pid) { |
1060 | pending++; | |
1061 | continue; | |
1062 | } | |
1063 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1064 | continue; |
2dc1bbeb | 1065 | if (td->o.use_thread) { |
3707f45b JA |
1066 | if (td->runstate == TD_EXITED) { |
1067 | td_set_runstate(td, TD_REAPED); | |
1068 | goto reaped; | |
1069 | } | |
1070 | continue; | |
1071 | } | |
a2f77c9f JA |
1072 | |
1073 | flags = WNOHANG; | |
1074 | if (td->runstate == TD_EXITED) | |
1075 | flags = 0; | |
1076 | ||
1077 | /* | |
1078 | * check if someone quit or got killed in an unusual way | |
1079 | */ | |
1080 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1081 | if (ret < 0) { |
a2f77c9f | 1082 | if (errno == ECHILD) { |
5ec10eaa JA |
1083 | log_err("fio: pid=%d disappeared %d\n", td->pid, |
1084 | td->runstate); | |
a2f77c9f JA |
1085 | td_set_runstate(td, TD_REAPED); |
1086 | goto reaped; | |
1087 | } | |
1088 | perror("waitpid"); | |
1089 | } else if (ret == td->pid) { | |
1090 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1091 | int sig = WTERMSIG(status); |
1092 | ||
d9244941 | 1093 | if (sig != SIGQUIT) |
5ec10eaa JA |
1094 | log_err("fio: pid=%d, got signal=%d\n", |
1095 | td->pid, sig); | |
fab6aa71 JA |
1096 | td_set_runstate(td, TD_REAPED); |
1097 | goto reaped; | |
1098 | } | |
a2f77c9f JA |
1099 | if (WIFEXITED(status)) { |
1100 | if (WEXITSTATUS(status) && !td->error) | |
1101 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1102 | |
a2f77c9f JA |
1103 | td_set_runstate(td, TD_REAPED); |
1104 | goto reaped; | |
a6418147 JA |
1105 | } |
1106 | } | |
ebac4655 | 1107 | |
a2f77c9f JA |
1108 | /* |
1109 | * thread is not dead, continue | |
1110 | */ | |
e48676ba | 1111 | pending++; |
a2f77c9f | 1112 | continue; |
fab6aa71 | 1113 | reaped: |
ebac4655 | 1114 | (*nr_running)--; |
2dc1bbeb JA |
1115 | (*m_rate) -= td->o.ratemin; |
1116 | (*t_rate) -= td->o.rate; | |
6f88bcb0 JA |
1117 | if (!td->pid) |
1118 | pending--; | |
a2f77c9f JA |
1119 | |
1120 | if (td->error) | |
1121 | exit_value++; | |
d3eeeabc JA |
1122 | |
1123 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1124 | } |
b990b5c0 | 1125 | |
1f809d15 | 1126 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1127 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1128 | } |
1129 | ||
906c8d75 JA |
1130 | /* |
1131 | * Main function for kicking off and reaping jobs, as needed. | |
1132 | */ | |
ebac4655 JA |
1133 | static void run_threads(void) |
1134 | { | |
ebac4655 JA |
1135 | struct thread_data *td; |
1136 | unsigned long spent; | |
1137 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1138 | |
2f9ade3c JA |
1139 | if (fio_pin_memory()) |
1140 | return; | |
ebac4655 | 1141 | |
c6ae0a5b | 1142 | if (!terse_output) { |
9cedf167 JA |
1143 | printf("Starting "); |
1144 | if (nr_thread) | |
5ec10eaa JA |
1145 | printf("%d thread%s", nr_thread, |
1146 | nr_thread > 1 ? "s" : ""); | |
9cedf167 JA |
1147 | if (nr_process) { |
1148 | if (nr_thread) | |
1149 | printf(" and "); | |
5ec10eaa JA |
1150 | printf("%d process%s", nr_process, |
1151 | nr_process > 1 ? "es" : ""); | |
9cedf167 JA |
1152 | } |
1153 | printf("\n"); | |
c6ae0a5b JA |
1154 | fflush(stdout); |
1155 | } | |
c04f7ec3 | 1156 | |
4efa970e JA |
1157 | signal(SIGINT, sig_handler); |
1158 | signal(SIGALRM, sig_handler); | |
1159 | ||
ebac4655 JA |
1160 | todo = thread_number; |
1161 | nr_running = 0; | |
1162 | nr_started = 0; | |
1163 | m_rate = t_rate = 0; | |
1164 | ||
34572e28 | 1165 | for_each_td(td, i) { |
263e529f | 1166 | print_status_init(td->thread_number - 1); |
ebac4655 | 1167 | |
2dc1bbeb | 1168 | if (!td->o.create_serialize) { |
380cf265 | 1169 | init_disk_util(td); |
ebac4655 | 1170 | continue; |
380cf265 | 1171 | } |
ebac4655 JA |
1172 | |
1173 | /* | |
1174 | * do file setup here so it happens sequentially, | |
1175 | * we don't want X number of threads getting their | |
1176 | * client data interspersed on disk | |
1177 | */ | |
53cdc686 | 1178 | if (setup_files(td)) { |
5bf13a5a JA |
1179 | exit_value++; |
1180 | if (td->error) | |
5ec10eaa JA |
1181 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, |
1182 | td->error, td->verror); | |
ebac4655 JA |
1183 | td_set_runstate(td, TD_REAPED); |
1184 | todo--; | |
c69e8875 JA |
1185 | } else { |
1186 | struct fio_file *f; | |
1187 | unsigned int i; | |
1188 | ||
1189 | /* | |
1190 | * for sharing to work, each job must always open | |
1191 | * its own files. so close them, if we opened them | |
1192 | * for creation | |
1193 | */ | |
1194 | for_each_file(td, f, i) | |
1195 | td_io_close_file(td, f); | |
5ec10eaa | 1196 | } |
380cf265 JA |
1197 | |
1198 | init_disk_util(td); | |
ebac4655 JA |
1199 | } |
1200 | ||
a2f77c9f JA |
1201 | set_genesis_time(); |
1202 | ||
ebac4655 | 1203 | while (todo) { |
75154845 JA |
1204 | struct thread_data *map[MAX_JOBS]; |
1205 | struct timeval this_start; | |
1206 | int this_jobs = 0, left; | |
1207 | ||
ebac4655 JA |
1208 | /* |
1209 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1210 | */ | |
34572e28 | 1211 | for_each_td(td, i) { |
ebac4655 JA |
1212 | if (td->runstate != TD_NOT_CREATED) |
1213 | continue; | |
1214 | ||
1215 | /* | |
1216 | * never got a chance to start, killed by other | |
1217 | * thread for some reason | |
1218 | */ | |
1219 | if (td->terminate) { | |
1220 | todo--; | |
1221 | continue; | |
1222 | } | |
1223 | ||
2dc1bbeb | 1224 | if (td->o.start_delay) { |
263e529f | 1225 | spent = mtime_since_genesis(); |
ebac4655 | 1226 | |
2dc1bbeb | 1227 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1228 | continue; |
1229 | } | |
1230 | ||
6f88bcb0 JA |
1231 | if (td->o.stonewall && (nr_started || nr_running)) { |
1232 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1233 | td->o.name); | |
ebac4655 | 1234 | break; |
6f88bcb0 | 1235 | } |
ebac4655 | 1236 | |
75154845 JA |
1237 | /* |
1238 | * Set state to created. Thread will transition | |
1239 | * to TD_INITIALIZED when it's done setting up. | |
1240 | */ | |
ebac4655 | 1241 | td_set_runstate(td, TD_CREATED); |
75154845 | 1242 | map[this_jobs++] = td; |
ebac4655 JA |
1243 | nr_started++; |
1244 | ||
2dc1bbeb | 1245 | if (td->o.use_thread) { |
ee56ad50 | 1246 | dprint(FD_PROCESS, "will pthread_create\n"); |
5ec10eaa JA |
1247 | if (pthread_create(&td->thread, NULL, |
1248 | thread_main, td)) { | |
f4705a73 | 1249 | perror("pthread_create"); |
ebac4655 | 1250 | nr_started--; |
c8bb6faf | 1251 | break; |
ebac4655 | 1252 | } |
f4705a73 JA |
1253 | if (pthread_detach(td->thread) < 0) |
1254 | perror("pthread_detach"); | |
ebac4655 | 1255 | } else { |
ee56ad50 | 1256 | dprint(FD_PROCESS, "will fork\n"); |
07739b57 | 1257 | if (!fork()) { |
a6418147 JA |
1258 | int ret = fork_main(shm_id, i); |
1259 | ||
1260 | exit(ret); | |
ebac4655 JA |
1261 | } |
1262 | } | |
cdd18ad8 | 1263 | fio_mutex_down(startup_mutex); |
ebac4655 JA |
1264 | } |
1265 | ||
1266 | /* | |
75154845 JA |
1267 | * Wait for the started threads to transition to |
1268 | * TD_INITIALIZED. | |
ebac4655 | 1269 | */ |
02bcaa8c | 1270 | fio_gettime(&this_start, NULL); |
75154845 | 1271 | left = this_jobs; |
6ce15a32 | 1272 | while (left && !fio_abort) { |
75154845 JA |
1273 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1274 | break; | |
1275 | ||
1276 | usleep(100000); | |
1277 | ||
1278 | for (i = 0; i < this_jobs; i++) { | |
1279 | td = map[i]; | |
1280 | if (!td) | |
1281 | continue; | |
b6f4d880 | 1282 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1283 | map[i] = NULL; |
1284 | left--; | |
b6f4d880 JA |
1285 | } else if (td->runstate >= TD_EXITED) { |
1286 | map[i] = NULL; | |
1287 | left--; | |
1288 | todo--; | |
1289 | nr_running++; /* work-around... */ | |
75154845 JA |
1290 | } |
1291 | } | |
1292 | } | |
1293 | ||
1294 | if (left) { | |
3b70d7e5 | 1295 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1296 | for (i = 0; i < this_jobs; i++) { |
1297 | td = map[i]; | |
1298 | if (!td) | |
1299 | continue; | |
1300 | kill(td->pid, SIGTERM); | |
1301 | } | |
1302 | break; | |
1303 | } | |
1304 | ||
1305 | /* | |
b6f4d880 | 1306 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1307 | */ |
34572e28 | 1308 | for_each_td(td, i) { |
75154845 | 1309 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1310 | continue; |
1311 | ||
1312 | td_set_runstate(td, TD_RUNNING); | |
1313 | nr_running++; | |
1314 | nr_started--; | |
2dc1bbeb JA |
1315 | m_rate += td->o.ratemin; |
1316 | t_rate += td->o.rate; | |
75154845 | 1317 | todo--; |
cdd18ad8 | 1318 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1319 | } |
1320 | ||
1321 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1322 | ||
1323 | if (todo) | |
1324 | usleep(100000); | |
1325 | } | |
1326 | ||
1327 | while (nr_running) { | |
1328 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1329 | usleep(10000); | |
1330 | } | |
1331 | ||
1332 | update_io_ticks(); | |
2f9ade3c | 1333 | fio_unpin_memory(); |
ebac4655 JA |
1334 | } |
1335 | ||
ebac4655 JA |
1336 | int main(int argc, char *argv[]) |
1337 | { | |
29d610e1 JA |
1338 | long ps; |
1339 | ||
2e5cdb11 JA |
1340 | sinit(); |
1341 | ||
dbe1125e JA |
1342 | /* |
1343 | * We need locale for number printing, if it isn't set then just | |
1344 | * go with the US format. | |
1345 | */ | |
1346 | if (!getenv("LC_NUMERIC")) | |
1347 | setlocale(LC_NUMERIC, "en_US"); | |
1348 | ||
ebac4655 JA |
1349 | if (parse_options(argc, argv)) |
1350 | return 1; | |
1351 | ||
4b472fa3 JA |
1352 | if (!thread_number) |
1353 | return 0; | |
ebac4655 | 1354 | |
29d610e1 JA |
1355 | ps = sysconf(_SC_PAGESIZE); |
1356 | if (ps < 0) { | |
1357 | log_err("Failed to get page size\n"); | |
1358 | return 1; | |
1359 | } | |
1360 | ||
cfc99db7 | 1361 | page_size = ps; |
29d610e1 JA |
1362 | page_mask = ps - 1; |
1363 | ||
bb3884d8 JA |
1364 | if (write_bw_log) { |
1365 | setup_log(&agg_io_log[DDIR_READ]); | |
1366 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1367 | } | |
1368 | ||
cdd18ad8 | 1369 | startup_mutex = fio_mutex_init(0); |
07739b57 | 1370 | |
a2f77c9f JA |
1371 | set_genesis_time(); |
1372 | ||
ebac4655 JA |
1373 | disk_util_timer_arm(); |
1374 | ||
1375 | run_threads(); | |
6ce15a32 | 1376 | |
bb3884d8 | 1377 | if (!fio_abort) { |
6ce15a32 | 1378 | show_run_stats(); |
bb3884d8 | 1379 | if (write_bw_log) { |
5ec10eaa JA |
1380 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1381 | __finish_log(agg_io_log[DDIR_WRITE], | |
1382 | "agg-write_bw.log"); | |
bb3884d8 JA |
1383 | } |
1384 | } | |
ebac4655 | 1385 | |
cdd18ad8 | 1386 | fio_mutex_remove(startup_mutex); |
437c9b71 | 1387 | return exit_value; |
ebac4655 | 1388 | } |