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" |
ebac4655 | 39 | |
cfc99db7 JA |
40 | unsigned long page_mask; |
41 | unsigned long page_size; | |
29d610e1 JA |
42 | #define ALIGN(buf) \ |
43 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
44 | |
45 | int groupid = 0; | |
46 | int thread_number = 0; | |
9cedf167 JA |
47 | int nr_process = 0; |
48 | int nr_thread = 0; | |
ebac4655 | 49 | int shm_id = 0; |
53cdc686 | 50 | int temp_stall_ts; |
ebac4655 | 51 | |
07739b57 | 52 | static struct fio_sem *startup_sem; |
6ce15a32 | 53 | static volatile int fio_abort; |
437c9b71 | 54 | static int exit_value; |
ebac4655 | 55 | |
bb3884d8 JA |
56 | struct io_log *agg_io_log[2]; |
57 | ||
ebac4655 | 58 | #define TERMINATE_ALL (-1) |
75154845 | 59 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 60 | |
6ce15a32 JA |
61 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
62 | { | |
63 | td->runstate = runstate; | |
64 | } | |
65 | ||
390c40e2 | 66 | static void terminate_threads(int group_id) |
ebac4655 | 67 | { |
34572e28 | 68 | struct thread_data *td; |
ebac4655 JA |
69 | int i; |
70 | ||
34572e28 | 71 | for_each_td(td, i) { |
ebac4655 | 72 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
7a93ab19 JA |
73 | /* |
74 | * if the thread is running, just let it exit | |
75 | */ | |
76 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 77 | kill(td->pid, SIGQUIT); |
ebac4655 | 78 | td->terminate = 1; |
2dc1bbeb | 79 | td->o.start_delay = 0; |
ebac4655 JA |
80 | } |
81 | } | |
82 | } | |
83 | ||
84 | static void sig_handler(int sig) | |
85 | { | |
86 | switch (sig) { | |
87 | case SIGALRM: | |
88 | update_io_ticks(); | |
89 | disk_util_timer_arm(); | |
90 | print_thread_status(); | |
91 | break; | |
92 | default: | |
6ce15a32 | 93 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 94 | fflush(stdout); |
390c40e2 | 95 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
96 | break; |
97 | } | |
98 | } | |
99 | ||
906c8d75 JA |
100 | /* |
101 | * Check if we are above the minimum rate given. | |
102 | */ | |
ebac4655 JA |
103 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
104 | { | |
0904200b | 105 | unsigned long long bytes = 0; |
4e991c23 | 106 | unsigned long iops = 0; |
ebac4655 JA |
107 | unsigned long spent; |
108 | unsigned long rate; | |
ebac4655 | 109 | |
780bf1a1 JA |
110 | /* |
111 | * No minimum rate set, always ok | |
112 | */ | |
2dc1bbeb | 113 | if (!td->o.ratemin && !td->o.rate_iops_min) |
780bf1a1 JA |
114 | return 0; |
115 | ||
ebac4655 JA |
116 | /* |
117 | * allow a 2 second settle period in the beginning | |
118 | */ | |
119 | if (mtime_since(&td->start, now) < 2000) | |
120 | return 0; | |
121 | ||
4e991c23 JA |
122 | if (td_read(td)) { |
123 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 124 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
125 | } |
126 | if (td_write(td)) { | |
127 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 128 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 129 | } |
0904200b | 130 | |
ebac4655 JA |
131 | /* |
132 | * if rate blocks is set, sample is running | |
133 | */ | |
4e991c23 | 134 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 135 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 136 | if (spent < td->o.ratecycle) |
ebac4655 JA |
137 | return 0; |
138 | ||
2dc1bbeb | 139 | if (td->o.rate) { |
4e991c23 JA |
140 | /* |
141 | * check bandwidth specified rate | |
142 | */ | |
143 | if (bytes < td->rate_bytes) { | |
2dc1bbeb | 144 | log_err("%s: min rate %u not met\n", td->o.name, td->o.ratemin); |
4e991c23 JA |
145 | return 1; |
146 | } else { | |
147 | rate = (bytes - td->rate_bytes) / spent; | |
2dc1bbeb JA |
148 | if (rate < td->o.ratemin || bytes < td->rate_bytes) { |
149 | log_err("%s: min rate %u not met, got %luKiB/sec\n", td->o.name, td->o.ratemin, rate); | |
4e991c23 JA |
150 | return 1; |
151 | } | |
152 | } | |
413dd459 | 153 | } else { |
4e991c23 JA |
154 | /* |
155 | * checks iops specified rate | |
156 | */ | |
2dc1bbeb JA |
157 | if (iops < td->o.rate_iops) { |
158 | log_err("%s: min iops rate %u not met\n", td->o.name, td->o.rate_iops); | |
413dd459 | 159 | return 1; |
4e991c23 JA |
160 | } else { |
161 | rate = (iops - td->rate_blocks) / spent; | |
2dc1bbeb JA |
162 | if (rate < td->o.rate_iops_min || iops < td->rate_blocks) { |
163 | log_err("%s: min iops rate %u not met, got %lu\n", td->o.name, td->o.rate_iops_min, rate); | |
4e991c23 | 164 | } |
413dd459 | 165 | } |
ebac4655 JA |
166 | } |
167 | } | |
168 | ||
0904200b | 169 | td->rate_bytes = bytes; |
4e991c23 | 170 | td->rate_blocks = iops; |
ebac4655 JA |
171 | memcpy(&td->lastrate, now, sizeof(*now)); |
172 | return 0; | |
173 | } | |
174 | ||
175 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
176 | { | |
2dc1bbeb | 177 | if (!td->o.timeout) |
ebac4655 | 178 | return 0; |
2dc1bbeb | 179 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
180 | return 1; |
181 | ||
182 | return 0; | |
183 | } | |
184 | ||
906c8d75 JA |
185 | /* |
186 | * When job exits, we can cancel the in-flight IO if we are using async | |
187 | * io. Attempt to do so. | |
188 | */ | |
ebac4655 JA |
189 | static void cleanup_pending_aio(struct thread_data *td) |
190 | { | |
ebac4655 | 191 | struct list_head *entry, *n; |
ebac4655 JA |
192 | struct io_u *io_u; |
193 | int r; | |
194 | ||
195 | /* | |
196 | * get immediately available events, if any | |
197 | */ | |
d7762cf8 | 198 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
199 | if (r < 0) |
200 | return; | |
ebac4655 JA |
201 | |
202 | /* | |
203 | * now cancel remaining active events | |
204 | */ | |
2866c82d | 205 | if (td->io_ops->cancel) { |
ebac4655 JA |
206 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
207 | io_u = list_entry(entry, struct io_u, list); | |
208 | ||
0c6e7517 JA |
209 | /* |
210 | * if the io_u isn't in flight, then that generally | |
211 | * means someone leaked an io_u. complain but fix | |
212 | * it up, so we don't stall here. | |
213 | */ | |
214 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
215 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 216 | put_io_u(td, io_u); |
0c6e7517 JA |
217 | } else { |
218 | r = td->io_ops->cancel(td, io_u); | |
219 | if (!r) | |
220 | put_io_u(td, io_u); | |
221 | } | |
ebac4655 JA |
222 | } |
223 | } | |
224 | ||
97601024 | 225 | if (td->cur_depth) |
d7762cf8 | 226 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
227 | } |
228 | ||
858a3d47 JA |
229 | /* |
230 | * Helper to handle the final sync of a file. Works just like the normal | |
231 | * io path, just does everything sync. | |
232 | */ | |
233 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
234 | { | |
235 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
236 | int ret; |
237 | ||
238 | if (!io_u) | |
239 | return 1; | |
240 | ||
241 | io_u->ddir = DDIR_SYNC; | |
242 | io_u->file = f; | |
243 | ||
244 | if (td_io_prep(td, io_u)) { | |
245 | put_io_u(td, io_u); | |
246 | return 1; | |
247 | } | |
248 | ||
755200a3 | 249 | requeue: |
858a3d47 | 250 | ret = td_io_queue(td, io_u); |
36167d82 | 251 | if (ret < 0) { |
e1161c32 | 252 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 253 | put_io_u(td, io_u); |
858a3d47 | 254 | return 1; |
36167d82 | 255 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 256 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 257 | return 1; |
36167d82 JA |
258 | } else if (ret == FIO_Q_COMPLETED) { |
259 | if (io_u->error) { | |
e1161c32 | 260 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
261 | return 1; |
262 | } | |
858a3d47 | 263 | |
d7762cf8 | 264 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 265 | return 1; |
755200a3 JA |
266 | } else if (ret == FIO_Q_BUSY) { |
267 | if (td_io_commit(td)) | |
268 | return 1; | |
269 | goto requeue; | |
858a3d47 JA |
270 | } |
271 | ||
272 | return 0; | |
273 | } | |
274 | ||
906c8d75 | 275 | /* |
34403fb1 | 276 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
277 | * reads the blocks back in, and checks the crc/md5 of the data. |
278 | */ | |
1e97cce9 | 279 | static void do_verify(struct thread_data *td) |
ebac4655 | 280 | { |
53cdc686 | 281 | struct fio_file *f; |
36167d82 | 282 | struct io_u *io_u; |
af52b345 JA |
283 | int ret, min_events; |
284 | unsigned int i; | |
e5b401d4 JA |
285 | |
286 | /* | |
287 | * sync io first and invalidate cache, to make sure we really | |
288 | * read from disk. | |
289 | */ | |
290 | for_each_file(td, f, i) { | |
3d7b485f JA |
291 | if (!(f->flags & FIO_FILE_OPEN)) |
292 | continue; | |
b2fdda43 JA |
293 | if (fio_io_sync(td, f)) |
294 | break; | |
295 | if (file_invalidate_cache(td, f)) | |
296 | break; | |
e5b401d4 | 297 | } |
ebac4655 | 298 | |
b2fdda43 JA |
299 | if (td->error) |
300 | return; | |
301 | ||
ebac4655 JA |
302 | td_set_runstate(td, TD_VERIFYING); |
303 | ||
36167d82 JA |
304 | io_u = NULL; |
305 | while (!td->terminate) { | |
5451792e JA |
306 | int ret2; |
307 | ||
ebac4655 JA |
308 | io_u = __get_io_u(td); |
309 | if (!io_u) | |
310 | break; | |
311 | ||
069c2918 JA |
312 | if (runtime_exceeded(td, &io_u->start_time)) { |
313 | put_io_u(td, io_u); | |
90a87d4b | 314 | td->terminate = 1; |
02bcaa8c | 315 | break; |
069c2918 | 316 | } |
02bcaa8c | 317 | |
069c2918 JA |
318 | if (get_next_verify(td, io_u)) { |
319 | put_io_u(td, io_u); | |
ebac4655 | 320 | break; |
069c2918 | 321 | } |
ebac4655 | 322 | |
069c2918 JA |
323 | if (td_io_prep(td, io_u)) { |
324 | put_io_u(td, io_u); | |
53cdc686 | 325 | break; |
069c2918 | 326 | } |
d7762cf8 JA |
327 | |
328 | io_u->end_io = verify_io_u; | |
53cdc686 | 329 | |
11786802 | 330 | ret = td_io_queue(td, io_u); |
36167d82 JA |
331 | switch (ret) { |
332 | case FIO_Q_COMPLETED: | |
333 | if (io_u->error) | |
22819ec2 | 334 | ret = -io_u->error; |
9e9d164e | 335 | else if (io_u->resid) { |
36167d82 | 336 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 337 | struct fio_file *f = io_u->file; |
ebac4655 | 338 | |
9e9d164e JA |
339 | /* |
340 | * zero read, fail | |
341 | */ | |
342 | if (!bytes) { | |
343 | td_verror(td, ENODATA, "full resid"); | |
344 | put_io_u(td, io_u); | |
345 | break; | |
346 | } | |
8400d9b2 | 347 | |
36167d82 JA |
348 | io_u->xfer_buflen = io_u->resid; |
349 | io_u->xfer_buf += bytes; | |
8400d9b2 | 350 | io_u->offset += bytes; |
d460eb31 | 351 | f->last_completed_pos = io_u->offset; |
8400d9b2 | 352 | |
30061b97 JA |
353 | td->ts.short_io_u[io_u->ddir]++; |
354 | ||
d460eb31 | 355 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
356 | goto sync_done; |
357 | ||
11786802 JA |
358 | requeue_io_u(td, &io_u); |
359 | } else { | |
8400d9b2 | 360 | sync_done: |
11786802 JA |
361 | ret = io_u_sync_complete(td, io_u); |
362 | if (ret < 0) | |
363 | break; | |
36167d82 | 364 | } |
36167d82 JA |
365 | continue; |
366 | case FIO_Q_QUEUED: | |
367 | break; | |
755200a3 JA |
368 | case FIO_Q_BUSY: |
369 | requeue_io_u(td, &io_u); | |
5451792e JA |
370 | ret2 = td_io_commit(td); |
371 | if (ret2 < 0) | |
372 | ret = ret2; | |
755200a3 | 373 | break; |
36167d82 JA |
374 | default: |
375 | assert(ret < 0); | |
e1161c32 | 376 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
377 | break; |
378 | } | |
379 | ||
99784632 | 380 | if (ret < 0 || td->error) |
3af6ef39 JA |
381 | break; |
382 | ||
ebac4655 | 383 | /* |
3af6ef39 JA |
384 | * if we can queue more, do so. but check if there are |
385 | * completed io_u's first. | |
ebac4655 | 386 | */ |
97601024 | 387 | min_events = 0; |
e916b390 | 388 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
3af6ef39 | 389 | min_events = 1; |
3af6ef39 | 390 | |
2dc1bbeb JA |
391 | if (td->cur_depth > td->o.iodepth_low) |
392 | min_events = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
393 | } |
394 | ||
3af6ef39 JA |
395 | /* |
396 | * Reap required number of io units, if any, and do the | |
397 | * verification on them through the callback handler | |
398 | */ | |
d7762cf8 | 399 | if (io_u_queued_complete(td, min_events) < 0) |
ebac4655 | 400 | break; |
36167d82 | 401 | } |
ebac4655 | 402 | |
c01c0395 JA |
403 | if (!td->error) { |
404 | min_events = td->cur_depth; | |
405 | ||
406 | if (min_events) | |
407 | ret = io_u_queued_complete(td, min_events); | |
408 | } else | |
ebac4655 JA |
409 | cleanup_pending_aio(td); |
410 | ||
411 | td_set_runstate(td, TD_RUNNING); | |
412 | } | |
413 | ||
32cd46a0 | 414 | /* |
906c8d75 | 415 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
416 | * and reaps them, checking for rate and errors along the way. |
417 | */ | |
ebac4655 JA |
418 | static void do_io(struct thread_data *td) |
419 | { | |
02bcaa8c | 420 | struct timeval s; |
ebac4655 | 421 | unsigned long usec; |
af52b345 JA |
422 | unsigned int i; |
423 | int ret = 0; | |
ebac4655 | 424 | |
5853e5a8 JA |
425 | td_set_runstate(td, TD_RUNNING); |
426 | ||
7bb48f84 | 427 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
428 | struct timeval comp_time; |
429 | long bytes_done = 0; | |
84585003 | 430 | int min_evts = 0; |
ebac4655 | 431 | struct io_u *io_u; |
5451792e | 432 | int ret2; |
ebac4655 JA |
433 | |
434 | if (td->terminate) | |
435 | break; | |
436 | ||
3d7c391d | 437 | io_u = get_io_u(td); |
ebac4655 JA |
438 | if (!io_u) |
439 | break; | |
440 | ||
441 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
442 | |
443 | if (runtime_exceeded(td, &s)) { | |
444 | put_io_u(td, io_u); | |
90a87d4b | 445 | td->terminate = 1; |
97601024 JA |
446 | break; |
447 | } | |
36167d82 | 448 | |
b67740d3 JA |
449 | /* |
450 | * Add verification end_io handler, if asked to verify | |
451 | * a previously written file. | |
452 | */ | |
453 | if (td->o.verify != VERIFY_NONE) | |
454 | io_u->end_io = verify_io_u; | |
455 | ||
11786802 | 456 | ret = td_io_queue(td, io_u); |
36167d82 JA |
457 | switch (ret) { |
458 | case FIO_Q_COMPLETED: | |
5451792e JA |
459 | if (io_u->error) |
460 | ret = -io_u->error; | |
9e9d164e | 461 | else if (io_u->resid) { |
36167d82 | 462 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 463 | struct fio_file *f = io_u->file; |
36167d82 | 464 | |
9e9d164e JA |
465 | /* |
466 | * zero read, fail | |
467 | */ | |
468 | if (!bytes) { | |
469 | td_verror(td, ENODATA, "full resid"); | |
470 | put_io_u(td, io_u); | |
471 | break; | |
472 | } | |
473 | ||
cec6b55d | 474 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 475 | io_u->xfer_buf += bytes; |
5a7c5680 | 476 | io_u->offset += bytes; |
d460eb31 | 477 | f->last_completed_pos = io_u->offset; |
5a7c5680 | 478 | |
30061b97 JA |
479 | td->ts.short_io_u[io_u->ddir]++; |
480 | ||
d460eb31 | 481 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
482 | goto sync_done; |
483 | ||
11786802 JA |
484 | requeue_io_u(td, &io_u); |
485 | } else { | |
5a7c5680 | 486 | sync_done: |
11786802 JA |
487 | fio_gettime(&comp_time, NULL); |
488 | bytes_done = io_u_sync_complete(td, io_u); | |
489 | if (bytes_done < 0) | |
490 | ret = bytes_done; | |
cec6b55d | 491 | } |
36167d82 JA |
492 | break; |
493 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
494 | /* |
495 | * if the engine doesn't have a commit hook, | |
496 | * the io_u is really queued. if it does have such | |
497 | * a hook, it has to call io_u_queued() itself. | |
498 | */ | |
499 | if (td->io_ops->commit == NULL) | |
500 | io_u_queued(td, io_u); | |
36167d82 | 501 | break; |
755200a3 JA |
502 | case FIO_Q_BUSY: |
503 | requeue_io_u(td, &io_u); | |
5451792e JA |
504 | ret2 = td_io_commit(td); |
505 | if (ret2 < 0) | |
506 | ret = ret2; | |
755200a3 | 507 | break; |
36167d82 JA |
508 | default: |
509 | assert(ret < 0); | |
510 | put_io_u(td, io_u); | |
511 | break; | |
ebac4655 JA |
512 | } |
513 | ||
99784632 | 514 | if (ret < 0 || td->error) |
36167d82 JA |
515 | break; |
516 | ||
97601024 JA |
517 | /* |
518 | * See if we need to complete some commands | |
519 | */ | |
755200a3 | 520 | if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) { |
97601024 | 521 | min_evts = 0; |
e916b390 | 522 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
36167d82 | 523 | min_evts = 1; |
ebac4655 | 524 | |
2dc1bbeb JA |
525 | if (td->cur_depth > td->o.iodepth_low) |
526 | min_evts = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
527 | } |
528 | ||
97601024 | 529 | fio_gettime(&comp_time, NULL); |
d7762cf8 | 530 | bytes_done = io_u_queued_complete(td, min_evts); |
97601024 | 531 | if (bytes_done < 0) |
36167d82 | 532 | break; |
ebac4655 JA |
533 | } |
534 | ||
97601024 JA |
535 | if (!bytes_done) |
536 | continue; | |
537 | ||
ebac4655 JA |
538 | /* |
539 | * the rate is batched for now, it should work for batches | |
540 | * of completions except the very first one which may look | |
541 | * a little bursty | |
542 | */ | |
97601024 | 543 | usec = utime_since(&s, &comp_time); |
ebac4655 | 544 | |
413dd459 | 545 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 546 | |
97601024 | 547 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 548 | if (exitall_on_terminate) |
390c40e2 | 549 | terminate_threads(td->groupid); |
e1161c32 | 550 | td_verror(td, ENODATA, "check_min_rate"); |
ebac4655 JA |
551 | break; |
552 | } | |
553 | ||
2dc1bbeb | 554 | if (td->o.thinktime) { |
9c1f7434 JA |
555 | unsigned long long b; |
556 | ||
557 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 558 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
559 | int left; |
560 | ||
2dc1bbeb JA |
561 | if (td->o.thinktime_spin) |
562 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 563 | |
2dc1bbeb | 564 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
565 | if (left) |
566 | usec_sleep(td, left); | |
567 | } | |
9c1f7434 | 568 | } |
ebac4655 JA |
569 | } |
570 | ||
886a2b5b JA |
571 | /* |
572 | * stop job if we failed doing any IO | |
573 | */ | |
574 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
575 | td->done = 1; | |
576 | ||
aa31f1f1 SL |
577 | if (td->o.fill_device && td->error == ENOSPC) { |
578 | td->error = 0; | |
579 | td->terminate = 1; | |
580 | } | |
4d2413c6 | 581 | if (!td->error) { |
3d7c391d JA |
582 | struct fio_file *f; |
583 | ||
c01c0395 JA |
584 | i = td->cur_depth; |
585 | if (i) | |
586 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 587 | |
2dc1bbeb | 588 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 589 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
590 | |
591 | for_each_file(td, f, i) { | |
592 | if (!(f->flags & FIO_FILE_OPEN)) | |
593 | continue; | |
858a3d47 | 594 | fio_io_sync(td, f); |
3d7b485f | 595 | } |
84585003 | 596 | } |
c01c0395 JA |
597 | } else |
598 | cleanup_pending_aio(td); | |
ebac4655 JA |
599 | } |
600 | ||
ebac4655 JA |
601 | static void cleanup_io_u(struct thread_data *td) |
602 | { | |
603 | struct list_head *entry, *n; | |
604 | struct io_u *io_u; | |
605 | ||
606 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
607 | io_u = list_entry(entry, struct io_u, list); | |
608 | ||
609 | list_del(&io_u->list); | |
610 | free(io_u); | |
611 | } | |
612 | ||
2f9ade3c | 613 | free_io_mem(td); |
ebac4655 JA |
614 | } |
615 | ||
6b9cea23 JA |
616 | /* |
617 | * "randomly" fill the buffer contents | |
618 | */ | |
e9459e5a | 619 | static void fill_io_buf(struct thread_data *td, struct io_u *io_u, int max_bs) |
6b9cea23 | 620 | { |
210f43b3 | 621 | long *ptr = io_u->buf; |
6b9cea23 | 622 | |
e9459e5a JA |
623 | if (!td->o.zero_buffers) { |
624 | while ((void *) ptr - io_u->buf < max_bs) { | |
210f43b3 | 625 | *ptr = rand() * GOLDEN_RATIO_PRIME; |
e9459e5a JA |
626 | ptr++; |
627 | } | |
628 | } else | |
629 | memset(ptr, 0, max_bs); | |
6b9cea23 JA |
630 | } |
631 | ||
ebac4655 JA |
632 | static int init_io_u(struct thread_data *td) |
633 | { | |
634 | struct io_u *io_u; | |
a00735e6 | 635 | unsigned int max_bs; |
ebac4655 JA |
636 | int i, max_units; |
637 | char *p; | |
638 | ||
2866c82d | 639 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
640 | max_units = 1; |
641 | else | |
2dc1bbeb | 642 | max_units = td->o.iodepth; |
ebac4655 | 643 | |
2dc1bbeb | 644 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
c3990645 | 645 | td->orig_buffer_size = (unsigned long long) max_bs * (unsigned long long) max_units; |
74b025b0 | 646 | |
2dc1bbeb JA |
647 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) |
648 | td->orig_buffer_size = (td->orig_buffer_size + td->o.hugepage_size - 1) & ~(td->o.hugepage_size - 1); | |
ebac4655 | 649 | |
c3990645 JA |
650 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
651 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
652 | return 1; | |
653 | } | |
654 | ||
2f9ade3c JA |
655 | if (allocate_io_mem(td)) |
656 | return 1; | |
ebac4655 | 657 | |
b3f378e9 JA |
658 | if (td->o.odirect) |
659 | p = ALIGN(td->orig_buffer); | |
660 | else | |
661 | p = td->orig_buffer; | |
662 | ||
ebac4655 | 663 | for (i = 0; i < max_units; i++) { |
db1cd90b JA |
664 | if (td->terminate) |
665 | return 1; | |
ebac4655 JA |
666 | io_u = malloc(sizeof(*io_u)); |
667 | memset(io_u, 0, sizeof(*io_u)); | |
668 | INIT_LIST_HEAD(&io_u->list); | |
669 | ||
b4c5e1ac JA |
670 | if (!(td->io_ops->flags & FIO_NOIO)) { |
671 | io_u->buf = p + max_bs * i; | |
e9459e5a | 672 | |
b4c5e1ac JA |
673 | if (td_write(td)) |
674 | fill_io_buf(td, io_u, max_bs); | |
675 | } | |
6b9cea23 | 676 | |
b1ff3403 | 677 | io_u->index = i; |
0c6e7517 | 678 | io_u->flags = IO_U_F_FREE; |
ebac4655 JA |
679 | list_add(&io_u->list, &td->io_u_freelist); |
680 | } | |
681 | ||
433afcb4 JA |
682 | io_u_init_timeout(); |
683 | ||
ebac4655 JA |
684 | return 0; |
685 | } | |
686 | ||
da86774e JA |
687 | static int switch_ioscheduler(struct thread_data *td) |
688 | { | |
689 | char tmp[256], tmp2[128]; | |
690 | FILE *f; | |
691 | int ret; | |
692 | ||
ba0fbe10 | 693 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
694 | return 0; |
695 | ||
da86774e JA |
696 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
697 | ||
698 | f = fopen(tmp, "r+"); | |
699 | if (!f) { | |
cbf5c1d7 JA |
700 | if (errno == ENOENT) { |
701 | log_err("fio: os or kernel doesn't support IO scheduler switching\n"); | |
702 | return 0; | |
703 | } | |
704 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
705 | return 1; |
706 | } | |
707 | ||
708 | /* | |
709 | * Set io scheduler. | |
710 | */ | |
2dc1bbeb | 711 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 712 | if (ferror(f) || ret != 1) { |
e1161c32 | 713 | td_verror(td, errno, "fwrite"); |
da86774e JA |
714 | fclose(f); |
715 | return 1; | |
716 | } | |
717 | ||
718 | rewind(f); | |
719 | ||
720 | /* | |
721 | * Read back and check that the selected scheduler is now the default. | |
722 | */ | |
723 | ret = fread(tmp, 1, sizeof(tmp), f); | |
724 | if (ferror(f) || ret < 0) { | |
e1161c32 | 725 | td_verror(td, errno, "fread"); |
da86774e JA |
726 | fclose(f); |
727 | return 1; | |
728 | } | |
729 | ||
2dc1bbeb | 730 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 731 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 732 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 733 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
734 | fclose(f); |
735 | return 1; | |
736 | } | |
737 | ||
738 | fclose(f); | |
739 | return 0; | |
740 | } | |
741 | ||
492158cf JA |
742 | static int keep_running(struct thread_data *td) |
743 | { | |
744 | unsigned long long io_done; | |
745 | ||
20e354ef JA |
746 | if (td->done) |
747 | return 0; | |
492158cf JA |
748 | if (td->o.time_based) |
749 | return 1; | |
750 | if (td->o.loops) { | |
751 | td->o.loops--; | |
752 | return 1; | |
753 | } | |
754 | ||
48f5abd3 | 755 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] + td->io_skip_bytes; |
492158cf JA |
756 | if (io_done < td->o.size) |
757 | return 1; | |
758 | ||
759 | return 0; | |
760 | } | |
761 | ||
a978ba68 | 762 | static int clear_io_state(struct thread_data *td) |
ebac4655 | 763 | { |
53cdc686 | 764 | struct fio_file *f; |
af52b345 JA |
765 | unsigned int i; |
766 | int ret; | |
ebac4655 | 767 | |
756867bd | 768 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 769 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 770 | td->zone_bytes = 0; |
d7d3b49b | 771 | td->rate_bytes = 0; |
4e991c23 | 772 | td->rate_blocks = 0; |
38d77cae | 773 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 774 | |
c1324df1 JA |
775 | td->last_was_sync = 0; |
776 | ||
48f5abd3 JA |
777 | if (td->o.time_based) |
778 | td->nr_done_files = 0; | |
281f9b63 | 779 | |
a978ba68 JA |
780 | for_each_file(td, f, i) |
781 | td_io_close_file(td, f); | |
53cdc686 | 782 | |
a978ba68 JA |
783 | ret = 0; |
784 | for_each_file(td, f, i) { | |
281f9b63 | 785 | f->flags &= ~FIO_FILE_DONE; |
a978ba68 JA |
786 | ret = td_io_open_file(td, f); |
787 | if (ret) | |
788 | break; | |
53cdc686 | 789 | } |
a978ba68 JA |
790 | |
791 | return ret; | |
ebac4655 JA |
792 | } |
793 | ||
906c8d75 JA |
794 | /* |
795 | * Entry point for the thread based jobs. The process based jobs end up | |
796 | * here as well, after a little setup. | |
797 | */ | |
ebac4655 JA |
798 | static void *thread_main(void *data) |
799 | { | |
10927316 | 800 | unsigned long long runtime[2], elapsed; |
ebac4655 | 801 | struct thread_data *td = data; |
a978ba68 | 802 | int clear_state; |
ebac4655 | 803 | |
2dc1bbeb | 804 | if (!td->o.use_thread) |
ebac4655 JA |
805 | setsid(); |
806 | ||
807 | td->pid = getpid(); | |
808 | ||
aea47d44 JA |
809 | INIT_LIST_HEAD(&td->io_u_freelist); |
810 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 811 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 | 812 | INIT_LIST_HEAD(&td->io_log_list); |
8de8f047 | 813 | INIT_LIST_HEAD(&td->io_hist_list); |
bb5d7d0b | 814 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 815 | |
db1cd90b JA |
816 | td_set_runstate(td, TD_INITIALIZED); |
817 | fio_sem_up(startup_sem); | |
818 | fio_sem_down(td->mutex); | |
819 | ||
820 | /* | |
821 | * the ->mutex semaphore is now no longer used, close it to avoid | |
822 | * eating a file descriptor | |
823 | */ | |
824 | fio_sem_remove(td->mutex); | |
825 | ||
d84f8d49 JA |
826 | /* |
827 | * May alter parameters that init_io_u() will use, so we need to | |
828 | * do this first. | |
829 | */ | |
830 | if (init_iolog(td)) | |
831 | goto err; | |
832 | ||
ebac4655 | 833 | if (init_io_u(td)) |
db1cd90b | 834 | goto err; |
ebac4655 | 835 | |
375b2695 | 836 | if (td->o.cpumask_set && fio_setaffinity(td) == -1) { |
e1161c32 | 837 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 838 | goto err; |
ebac4655 JA |
839 | } |
840 | ||
ac684785 | 841 | if (td->ioprio_set) { |
ebac4655 | 842 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 843 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 844 | goto err; |
ebac4655 JA |
845 | } |
846 | } | |
847 | ||
2dc1bbeb | 848 | if (nice(td->o.nice) == -1) { |
e1161c32 | 849 | td_verror(td, errno, "nice"); |
db1cd90b | 850 | goto err; |
b6f4d880 JA |
851 | } |
852 | ||
2dc1bbeb | 853 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 854 | goto err; |
37f56873 | 855 | |
2dc1bbeb | 856 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
857 | goto err; |
858 | ||
7d6c5283 JA |
859 | if (td_io_init(td)) |
860 | goto err; | |
861 | ||
b5af8293 JA |
862 | if (open_files(td)) |
863 | goto err; | |
864 | ||
68727076 JA |
865 | if (init_random_map(td)) |
866 | goto err; | |
867 | ||
2dc1bbeb JA |
868 | if (td->o.exec_prerun) { |
869 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
870 | goto err; |
871 | } | |
4e0ba8af | 872 | |
69008999 | 873 | fio_gettime(&td->epoch, NULL); |
433afcb4 | 874 | memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); |
756867bd | 875 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
876 | |
877 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 878 | clear_state = 0; |
492158cf | 879 | while (keep_running(td)) { |
02bcaa8c | 880 | fio_gettime(&td->start, NULL); |
756867bd | 881 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 | 882 | |
2dc1bbeb | 883 | if (td->o.ratemin) |
756867bd | 884 | memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate)); |
ebac4655 | 885 | |
a978ba68 JA |
886 | if (clear_state && clear_io_state(td)) |
887 | break; | |
888 | ||
ebac4655 JA |
889 | prune_io_piece_log(td); |
890 | ||
ba0fbe10 | 891 | do_io(td); |
ebac4655 | 892 | |
a978ba68 JA |
893 | clear_state = 1; |
894 | ||
38d77cae JA |
895 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
896 | if (td->rw_end_set[DDIR_READ]) | |
897 | elapsed = utime_since(&td->start, &td->rw_end[DDIR_READ]); | |
898 | else | |
899 | elapsed = utime_since_now(&td->start); | |
900 | ||
901 | runtime[DDIR_READ] += elapsed; | |
902 | } | |
903 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
904 | if (td->rw_end_set[DDIR_WRITE]) | |
905 | elapsed = utime_since(&td->start, &td->rw_end[DDIR_WRITE]); | |
906 | else | |
907 | elapsed = utime_since_now(&td->start); | |
908 | ||
909 | runtime[DDIR_WRITE] += elapsed; | |
910 | } | |
413dd459 | 911 | |
ebac4655 JA |
912 | if (td->error || td->terminate) |
913 | break; | |
914 | ||
e84c73a8 SL |
915 | if (!td->o.do_verify || |
916 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 917 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
918 | continue; |
919 | ||
a978ba68 JA |
920 | if (clear_io_state(td)) |
921 | break; | |
922 | ||
02bcaa8c | 923 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
924 | |
925 | do_verify(td); | |
926 | ||
69008999 | 927 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
928 | |
929 | if (td->error || td->terminate) | |
930 | break; | |
931 | } | |
932 | ||
36dff966 | 933 | update_rusage_stat(td); |
47f0cc48 YHJT |
934 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
935 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
936 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
937 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
938 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
939 | ||
940 | if (td->ts.bw_log) | |
941 | finish_log(td, td->ts.bw_log, "bw"); | |
942 | if (td->ts.slat_log) | |
943 | finish_log(td, td->ts.slat_log, "slat"); | |
944 | if (td->ts.clat_log) | |
945 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb JA |
946 | if (td->o.exec_postrun) { |
947 | if (system(td->o.exec_postrun) < 0) | |
948 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 949 | } |
ebac4655 JA |
950 | |
951 | if (exitall_on_terminate) | |
390c40e2 | 952 | terminate_threads(td->groupid); |
ebac4655 JA |
953 | |
954 | err: | |
5bf13a5a JA |
955 | if (td->error) |
956 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
53cdc686 | 957 | close_files(td); |
2866c82d | 958 | close_ioengine(td); |
ebac4655 | 959 | cleanup_io_u(td); |
f29b25a3 JA |
960 | |
961 | /* | |
962 | * do this very late, it will log file closing as well | |
963 | */ | |
964 | if (td->o.write_iolog_file) | |
965 | write_iolog_close(td); | |
966 | ||
d23bb327 | 967 | options_mem_free(td); |
ebac4655 | 968 | td_set_runstate(td, TD_EXITED); |
43d76807 | 969 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
970 | } |
971 | ||
906c8d75 JA |
972 | /* |
973 | * We cannot pass the td data into a forked process, so attach the td and | |
974 | * pass it to the thread worker. | |
975 | */ | |
a6418147 | 976 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
977 | { |
978 | struct thread_data *td; | |
a6418147 | 979 | void *data, *ret; |
ebac4655 JA |
980 | |
981 | data = shmat(shmid, NULL, 0); | |
982 | if (data == (void *) -1) { | |
a6418147 JA |
983 | int __err = errno; |
984 | ||
ebac4655 | 985 | perror("shmat"); |
a6418147 | 986 | return __err; |
ebac4655 JA |
987 | } |
988 | ||
989 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 990 | ret = thread_main(td); |
ebac4655 | 991 | shmdt(data); |
43d76807 | 992 | return (int) (unsigned long) ret; |
ebac4655 JA |
993 | } |
994 | ||
906c8d75 JA |
995 | /* |
996 | * Run over the job map and reap the threads that have exited, if any. | |
997 | */ | |
ebac4655 JA |
998 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
999 | { | |
34572e28 | 1000 | struct thread_data *td; |
1f809d15 | 1001 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1002 | |
1003 | /* | |
1004 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1005 | */ | |
1f809d15 | 1006 | realthreads = pending = cputhreads = 0; |
34572e28 | 1007 | for_each_td(td, i) { |
3707f45b | 1008 | int flags = 0; |
a2f77c9f | 1009 | |
84585003 JA |
1010 | /* |
1011 | * ->io_ops is NULL for a thread that has closed its | |
1012 | * io engine | |
1013 | */ | |
ba0fbe10 | 1014 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1015 | cputhreads++; |
1f809d15 JA |
1016 | else |
1017 | realthreads++; | |
b990b5c0 | 1018 | |
a2f77c9f JA |
1019 | if (!td->pid || td->runstate == TD_REAPED) |
1020 | continue; | |
2dc1bbeb | 1021 | if (td->o.use_thread) { |
3707f45b JA |
1022 | if (td->runstate == TD_EXITED) { |
1023 | td_set_runstate(td, TD_REAPED); | |
1024 | goto reaped; | |
1025 | } | |
1026 | continue; | |
1027 | } | |
a2f77c9f JA |
1028 | |
1029 | flags = WNOHANG; | |
1030 | if (td->runstate == TD_EXITED) | |
1031 | flags = 0; | |
1032 | ||
1033 | /* | |
1034 | * check if someone quit or got killed in an unusual way | |
1035 | */ | |
1036 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1037 | if (ret < 0) { |
a2f77c9f JA |
1038 | if (errno == ECHILD) { |
1039 | log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate); | |
1040 | td_set_runstate(td, TD_REAPED); | |
1041 | goto reaped; | |
1042 | } | |
1043 | perror("waitpid"); | |
1044 | } else if (ret == td->pid) { | |
1045 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1046 | int sig = WTERMSIG(status); |
1047 | ||
d9244941 JA |
1048 | if (sig != SIGQUIT) |
1049 | log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); | |
fab6aa71 JA |
1050 | td_set_runstate(td, TD_REAPED); |
1051 | goto reaped; | |
1052 | } | |
a2f77c9f JA |
1053 | if (WIFEXITED(status)) { |
1054 | if (WEXITSTATUS(status) && !td->error) | |
1055 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1056 | |
a2f77c9f JA |
1057 | td_set_runstate(td, TD_REAPED); |
1058 | goto reaped; | |
a6418147 JA |
1059 | } |
1060 | } | |
ebac4655 | 1061 | |
a2f77c9f JA |
1062 | /* |
1063 | * thread is not dead, continue | |
1064 | */ | |
e48676ba | 1065 | pending++; |
a2f77c9f | 1066 | continue; |
fab6aa71 | 1067 | reaped: |
2dc1bbeb | 1068 | if (td->o.use_thread) { |
3707f45b JA |
1069 | long ret; |
1070 | ||
1071 | if (pthread_join(td->thread, (void *) &ret)) | |
1072 | perror("pthread_join"); | |
1073 | } | |
1074 | ||
ebac4655 | 1075 | (*nr_running)--; |
2dc1bbeb JA |
1076 | (*m_rate) -= td->o.ratemin; |
1077 | (*t_rate) -= td->o.rate; | |
e48676ba | 1078 | pending--; |
a2f77c9f JA |
1079 | |
1080 | if (td->error) | |
1081 | exit_value++; | |
ebac4655 | 1082 | } |
b990b5c0 | 1083 | |
1f809d15 | 1084 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1085 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1086 | } |
1087 | ||
906c8d75 JA |
1088 | /* |
1089 | * Main function for kicking off and reaping jobs, as needed. | |
1090 | */ | |
ebac4655 JA |
1091 | static void run_threads(void) |
1092 | { | |
ebac4655 JA |
1093 | struct thread_data *td; |
1094 | unsigned long spent; | |
1095 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1096 | |
2f9ade3c JA |
1097 | if (fio_pin_memory()) |
1098 | return; | |
ebac4655 | 1099 | |
c6ae0a5b | 1100 | if (!terse_output) { |
9cedf167 JA |
1101 | printf("Starting "); |
1102 | if (nr_thread) | |
1103 | printf("%d thread%s", nr_thread, nr_thread > 1 ? "s" : ""); | |
1104 | if (nr_process) { | |
1105 | if (nr_thread) | |
1106 | printf(" and "); | |
1107 | printf("%d process%s", nr_process, nr_process > 1 ? "es" : ""); | |
1108 | } | |
1109 | printf("\n"); | |
c6ae0a5b JA |
1110 | fflush(stdout); |
1111 | } | |
c04f7ec3 | 1112 | |
4efa970e JA |
1113 | signal(SIGINT, sig_handler); |
1114 | signal(SIGALRM, sig_handler); | |
1115 | ||
ebac4655 JA |
1116 | todo = thread_number; |
1117 | nr_running = 0; | |
1118 | nr_started = 0; | |
1119 | m_rate = t_rate = 0; | |
1120 | ||
34572e28 | 1121 | for_each_td(td, i) { |
263e529f | 1122 | print_status_init(td->thread_number - 1); |
ebac4655 | 1123 | |
2dc1bbeb | 1124 | if (!td->o.create_serialize) { |
380cf265 | 1125 | init_disk_util(td); |
ebac4655 | 1126 | continue; |
380cf265 | 1127 | } |
ebac4655 JA |
1128 | |
1129 | /* | |
1130 | * do file setup here so it happens sequentially, | |
1131 | * we don't want X number of threads getting their | |
1132 | * client data interspersed on disk | |
1133 | */ | |
53cdc686 | 1134 | if (setup_files(td)) { |
5bf13a5a JA |
1135 | exit_value++; |
1136 | if (td->error) | |
1137 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
ebac4655 JA |
1138 | td_set_runstate(td, TD_REAPED); |
1139 | todo--; | |
1140 | } | |
380cf265 JA |
1141 | |
1142 | init_disk_util(td); | |
ebac4655 JA |
1143 | } |
1144 | ||
a2f77c9f JA |
1145 | set_genesis_time(); |
1146 | ||
ebac4655 | 1147 | while (todo) { |
75154845 JA |
1148 | struct thread_data *map[MAX_JOBS]; |
1149 | struct timeval this_start; | |
1150 | int this_jobs = 0, left; | |
1151 | ||
ebac4655 JA |
1152 | /* |
1153 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1154 | */ | |
34572e28 | 1155 | for_each_td(td, i) { |
ebac4655 JA |
1156 | if (td->runstate != TD_NOT_CREATED) |
1157 | continue; | |
1158 | ||
1159 | /* | |
1160 | * never got a chance to start, killed by other | |
1161 | * thread for some reason | |
1162 | */ | |
1163 | if (td->terminate) { | |
1164 | todo--; | |
1165 | continue; | |
1166 | } | |
1167 | ||
2dc1bbeb | 1168 | if (td->o.start_delay) { |
263e529f | 1169 | spent = mtime_since_genesis(); |
ebac4655 | 1170 | |
2dc1bbeb | 1171 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1172 | continue; |
1173 | } | |
1174 | ||
2dc1bbeb | 1175 | if (td->o.stonewall && (nr_started || nr_running)) |
ebac4655 JA |
1176 | break; |
1177 | ||
75154845 JA |
1178 | /* |
1179 | * Set state to created. Thread will transition | |
1180 | * to TD_INITIALIZED when it's done setting up. | |
1181 | */ | |
ebac4655 | 1182 | td_set_runstate(td, TD_CREATED); |
75154845 | 1183 | map[this_jobs++] = td; |
ebac4655 JA |
1184 | nr_started++; |
1185 | ||
2dc1bbeb | 1186 | if (td->o.use_thread) { |
ebac4655 JA |
1187 | if (pthread_create(&td->thread, NULL, thread_main, td)) { |
1188 | perror("thread_create"); | |
1189 | nr_started--; | |
c8bb6faf | 1190 | break; |
ebac4655 JA |
1191 | } |
1192 | } else { | |
07739b57 | 1193 | if (!fork()) { |
a6418147 JA |
1194 | int ret = fork_main(shm_id, i); |
1195 | ||
1196 | exit(ret); | |
ebac4655 JA |
1197 | } |
1198 | } | |
07739b57 | 1199 | fio_sem_down(startup_sem); |
ebac4655 JA |
1200 | } |
1201 | ||
1202 | /* | |
75154845 JA |
1203 | * Wait for the started threads to transition to |
1204 | * TD_INITIALIZED. | |
ebac4655 | 1205 | */ |
02bcaa8c | 1206 | fio_gettime(&this_start, NULL); |
75154845 | 1207 | left = this_jobs; |
6ce15a32 | 1208 | while (left && !fio_abort) { |
75154845 JA |
1209 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1210 | break; | |
1211 | ||
1212 | usleep(100000); | |
1213 | ||
1214 | for (i = 0; i < this_jobs; i++) { | |
1215 | td = map[i]; | |
1216 | if (!td) | |
1217 | continue; | |
b6f4d880 | 1218 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1219 | map[i] = NULL; |
1220 | left--; | |
b6f4d880 JA |
1221 | } else if (td->runstate >= TD_EXITED) { |
1222 | map[i] = NULL; | |
1223 | left--; | |
1224 | todo--; | |
1225 | nr_running++; /* work-around... */ | |
75154845 JA |
1226 | } |
1227 | } | |
1228 | } | |
1229 | ||
1230 | if (left) { | |
3b70d7e5 | 1231 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1232 | for (i = 0; i < this_jobs; i++) { |
1233 | td = map[i]; | |
1234 | if (!td) | |
1235 | continue; | |
1236 | kill(td->pid, SIGTERM); | |
1237 | } | |
1238 | break; | |
1239 | } | |
1240 | ||
1241 | /* | |
b6f4d880 | 1242 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1243 | */ |
34572e28 | 1244 | for_each_td(td, i) { |
75154845 | 1245 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1246 | continue; |
1247 | ||
1248 | td_set_runstate(td, TD_RUNNING); | |
1249 | nr_running++; | |
1250 | nr_started--; | |
2dc1bbeb JA |
1251 | m_rate += td->o.ratemin; |
1252 | t_rate += td->o.rate; | |
75154845 | 1253 | todo--; |
07739b57 | 1254 | fio_sem_up(td->mutex); |
ebac4655 JA |
1255 | } |
1256 | ||
1257 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1258 | ||
1259 | if (todo) | |
1260 | usleep(100000); | |
1261 | } | |
1262 | ||
1263 | while (nr_running) { | |
1264 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1265 | usleep(10000); | |
1266 | } | |
1267 | ||
1268 | update_io_ticks(); | |
2f9ade3c | 1269 | fio_unpin_memory(); |
ebac4655 JA |
1270 | } |
1271 | ||
ebac4655 JA |
1272 | int main(int argc, char *argv[]) |
1273 | { | |
29d610e1 JA |
1274 | long ps; |
1275 | ||
dbe1125e JA |
1276 | /* |
1277 | * We need locale for number printing, if it isn't set then just | |
1278 | * go with the US format. | |
1279 | */ | |
1280 | if (!getenv("LC_NUMERIC")) | |
1281 | setlocale(LC_NUMERIC, "en_US"); | |
1282 | ||
ebac4655 JA |
1283 | if (parse_options(argc, argv)) |
1284 | return 1; | |
1285 | ||
4b472fa3 JA |
1286 | if (!thread_number) |
1287 | return 0; | |
ebac4655 | 1288 | |
29d610e1 JA |
1289 | ps = sysconf(_SC_PAGESIZE); |
1290 | if (ps < 0) { | |
1291 | log_err("Failed to get page size\n"); | |
1292 | return 1; | |
1293 | } | |
1294 | ||
cfc99db7 | 1295 | page_size = ps; |
29d610e1 JA |
1296 | page_mask = ps - 1; |
1297 | ||
bb3884d8 JA |
1298 | if (write_bw_log) { |
1299 | setup_log(&agg_io_log[DDIR_READ]); | |
1300 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1301 | } | |
1302 | ||
07739b57 JA |
1303 | startup_sem = fio_sem_init(0); |
1304 | ||
a2f77c9f JA |
1305 | set_genesis_time(); |
1306 | ||
ebac4655 JA |
1307 | disk_util_timer_arm(); |
1308 | ||
1309 | run_threads(); | |
6ce15a32 | 1310 | |
bb3884d8 | 1311 | if (!fio_abort) { |
6ce15a32 | 1312 | show_run_stats(); |
bb3884d8 JA |
1313 | if (write_bw_log) { |
1314 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1315 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1316 | } | |
1317 | } | |
ebac4655 | 1318 | |
07739b57 | 1319 | fio_sem_remove(startup_sem); |
437c9b71 | 1320 | return exit_value; |
ebac4655 | 1321 | } |