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 | |
cdd18ad8 | 52 | static struct fio_mutex *startup_mutex; |
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 | { | |
bd6f78b2 JA |
63 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", td->pid, td->runstate, |
64 | runstate); | |
6ce15a32 JA |
65 | td->runstate = runstate; |
66 | } | |
67 | ||
390c40e2 | 68 | static void terminate_threads(int group_id) |
ebac4655 | 69 | { |
34572e28 | 70 | struct thread_data *td; |
ebac4655 JA |
71 | int i; |
72 | ||
34572e28 | 73 | for_each_td(td, i) { |
ebac4655 | 74 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
ee56ad50 | 75 | dprint(FD_PROCESS, "setting terminate on %d\n",td->pid); |
ad830ed7 JA |
76 | |
77 | td->terminate = 1; | |
78 | td->o.start_delay = 0; | |
79 | ||
7a93ab19 JA |
80 | /* |
81 | * if the thread is running, just let it exit | |
82 | */ | |
83 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 84 | kill(td->pid, SIGQUIT); |
f63f7f3b JA |
85 | else { |
86 | struct ioengine_ops *ops = td->io_ops; | |
87 | ||
88 | if (ops && (ops->flags & FIO_SIGQUIT)) | |
89 | kill(td->pid, SIGQUIT); | |
90 | } | |
ebac4655 JA |
91 | } |
92 | } | |
93 | } | |
94 | ||
95 | static void sig_handler(int sig) | |
96 | { | |
97 | switch (sig) { | |
98 | case SIGALRM: | |
99 | update_io_ticks(); | |
100 | disk_util_timer_arm(); | |
101 | print_thread_status(); | |
102 | break; | |
103 | default: | |
6ce15a32 | 104 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 105 | fflush(stdout); |
390c40e2 | 106 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
107 | break; |
108 | } | |
109 | } | |
110 | ||
906c8d75 JA |
111 | /* |
112 | * Check if we are above the minimum rate given. | |
113 | */ | |
ebac4655 JA |
114 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
115 | { | |
0904200b | 116 | unsigned long long bytes = 0; |
4e991c23 | 117 | unsigned long iops = 0; |
ebac4655 JA |
118 | unsigned long spent; |
119 | unsigned long rate; | |
ebac4655 | 120 | |
780bf1a1 JA |
121 | /* |
122 | * No minimum rate set, always ok | |
123 | */ | |
2dc1bbeb | 124 | if (!td->o.ratemin && !td->o.rate_iops_min) |
780bf1a1 JA |
125 | return 0; |
126 | ||
ebac4655 JA |
127 | /* |
128 | * allow a 2 second settle period in the beginning | |
129 | */ | |
130 | if (mtime_since(&td->start, now) < 2000) | |
131 | return 0; | |
132 | ||
4e991c23 JA |
133 | if (td_read(td)) { |
134 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 135 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
136 | } |
137 | if (td_write(td)) { | |
138 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 139 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 140 | } |
0904200b | 141 | |
ebac4655 JA |
142 | /* |
143 | * if rate blocks is set, sample is running | |
144 | */ | |
4e991c23 | 145 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 146 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 147 | if (spent < td->o.ratecycle) |
ebac4655 JA |
148 | return 0; |
149 | ||
2dc1bbeb | 150 | if (td->o.rate) { |
4e991c23 JA |
151 | /* |
152 | * check bandwidth specified rate | |
153 | */ | |
154 | if (bytes < td->rate_bytes) { | |
2dc1bbeb | 155 | log_err("%s: min rate %u not met\n", td->o.name, td->o.ratemin); |
4e991c23 JA |
156 | return 1; |
157 | } else { | |
158 | rate = (bytes - td->rate_bytes) / spent; | |
2dc1bbeb JA |
159 | if (rate < td->o.ratemin || bytes < td->rate_bytes) { |
160 | log_err("%s: min rate %u not met, got %luKiB/sec\n", td->o.name, td->o.ratemin, rate); | |
4e991c23 JA |
161 | return 1; |
162 | } | |
163 | } | |
413dd459 | 164 | } else { |
4e991c23 JA |
165 | /* |
166 | * checks iops specified rate | |
167 | */ | |
2dc1bbeb JA |
168 | if (iops < td->o.rate_iops) { |
169 | log_err("%s: min iops rate %u not met\n", td->o.name, td->o.rate_iops); | |
413dd459 | 170 | return 1; |
4e991c23 JA |
171 | } else { |
172 | rate = (iops - td->rate_blocks) / spent; | |
2dc1bbeb JA |
173 | if (rate < td->o.rate_iops_min || iops < td->rate_blocks) { |
174 | log_err("%s: min iops rate %u not met, got %lu\n", td->o.name, td->o.rate_iops_min, rate); | |
4e991c23 | 175 | } |
413dd459 | 176 | } |
ebac4655 JA |
177 | } |
178 | } | |
179 | ||
0904200b | 180 | td->rate_bytes = bytes; |
4e991c23 | 181 | td->rate_blocks = iops; |
ebac4655 JA |
182 | memcpy(&td->lastrate, now, sizeof(*now)); |
183 | return 0; | |
184 | } | |
185 | ||
186 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
187 | { | |
2dc1bbeb | 188 | if (!td->o.timeout) |
ebac4655 | 189 | return 0; |
2dc1bbeb | 190 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
191 | return 1; |
192 | ||
193 | return 0; | |
194 | } | |
195 | ||
906c8d75 JA |
196 | /* |
197 | * When job exits, we can cancel the in-flight IO if we are using async | |
198 | * io. Attempt to do so. | |
199 | */ | |
ebac4655 JA |
200 | static void cleanup_pending_aio(struct thread_data *td) |
201 | { | |
ebac4655 | 202 | struct list_head *entry, *n; |
ebac4655 JA |
203 | struct io_u *io_u; |
204 | int r; | |
205 | ||
206 | /* | |
207 | * get immediately available events, if any | |
208 | */ | |
d7762cf8 | 209 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
210 | if (r < 0) |
211 | return; | |
ebac4655 JA |
212 | |
213 | /* | |
214 | * now cancel remaining active events | |
215 | */ | |
2866c82d | 216 | if (td->io_ops->cancel) { |
ebac4655 JA |
217 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
218 | io_u = list_entry(entry, struct io_u, list); | |
219 | ||
0c6e7517 JA |
220 | /* |
221 | * if the io_u isn't in flight, then that generally | |
222 | * means someone leaked an io_u. complain but fix | |
223 | * it up, so we don't stall here. | |
224 | */ | |
225 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
226 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 227 | put_io_u(td, io_u); |
0c6e7517 JA |
228 | } else { |
229 | r = td->io_ops->cancel(td, io_u); | |
230 | if (!r) | |
231 | put_io_u(td, io_u); | |
232 | } | |
ebac4655 JA |
233 | } |
234 | } | |
235 | ||
97601024 | 236 | if (td->cur_depth) |
d7762cf8 | 237 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
238 | } |
239 | ||
858a3d47 JA |
240 | /* |
241 | * Helper to handle the final sync of a file. Works just like the normal | |
242 | * io path, just does everything sync. | |
243 | */ | |
244 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
245 | { | |
246 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
247 | int ret; |
248 | ||
249 | if (!io_u) | |
250 | return 1; | |
251 | ||
252 | io_u->ddir = DDIR_SYNC; | |
253 | io_u->file = f; | |
254 | ||
255 | if (td_io_prep(td, io_u)) { | |
256 | put_io_u(td, io_u); | |
257 | return 1; | |
258 | } | |
259 | ||
755200a3 | 260 | requeue: |
858a3d47 | 261 | ret = td_io_queue(td, io_u); |
36167d82 | 262 | if (ret < 0) { |
e1161c32 | 263 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 264 | put_io_u(td, io_u); |
858a3d47 | 265 | return 1; |
36167d82 | 266 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 267 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 268 | return 1; |
36167d82 JA |
269 | } else if (ret == FIO_Q_COMPLETED) { |
270 | if (io_u->error) { | |
e1161c32 | 271 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
272 | return 1; |
273 | } | |
858a3d47 | 274 | |
d7762cf8 | 275 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 276 | return 1; |
755200a3 JA |
277 | } else if (ret == FIO_Q_BUSY) { |
278 | if (td_io_commit(td)) | |
279 | return 1; | |
280 | goto requeue; | |
858a3d47 JA |
281 | } |
282 | ||
283 | return 0; | |
284 | } | |
285 | ||
906c8d75 | 286 | /* |
34403fb1 | 287 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
288 | * reads the blocks back in, and checks the crc/md5 of the data. |
289 | */ | |
1e97cce9 | 290 | static void do_verify(struct thread_data *td) |
ebac4655 | 291 | { |
53cdc686 | 292 | struct fio_file *f; |
36167d82 | 293 | struct io_u *io_u; |
af52b345 JA |
294 | int ret, min_events; |
295 | unsigned int i; | |
e5b401d4 JA |
296 | |
297 | /* | |
298 | * sync io first and invalidate cache, to make sure we really | |
299 | * read from disk. | |
300 | */ | |
301 | for_each_file(td, f, i) { | |
3d7b485f JA |
302 | if (!(f->flags & FIO_FILE_OPEN)) |
303 | continue; | |
b2fdda43 JA |
304 | if (fio_io_sync(td, f)) |
305 | break; | |
306 | if (file_invalidate_cache(td, f)) | |
307 | break; | |
e5b401d4 | 308 | } |
ebac4655 | 309 | |
b2fdda43 JA |
310 | if (td->error) |
311 | return; | |
312 | ||
ebac4655 JA |
313 | td_set_runstate(td, TD_VERIFYING); |
314 | ||
36167d82 JA |
315 | io_u = NULL; |
316 | while (!td->terminate) { | |
5451792e JA |
317 | int ret2; |
318 | ||
ebac4655 JA |
319 | io_u = __get_io_u(td); |
320 | if (!io_u) | |
321 | break; | |
322 | ||
069c2918 JA |
323 | if (runtime_exceeded(td, &io_u->start_time)) { |
324 | put_io_u(td, io_u); | |
90a87d4b | 325 | td->terminate = 1; |
02bcaa8c | 326 | break; |
069c2918 | 327 | } |
02bcaa8c | 328 | |
069c2918 JA |
329 | if (get_next_verify(td, io_u)) { |
330 | put_io_u(td, io_u); | |
ebac4655 | 331 | break; |
069c2918 | 332 | } |
ebac4655 | 333 | |
069c2918 JA |
334 | if (td_io_prep(td, io_u)) { |
335 | put_io_u(td, io_u); | |
53cdc686 | 336 | break; |
069c2918 | 337 | } |
d7762cf8 JA |
338 | |
339 | io_u->end_io = verify_io_u; | |
53cdc686 | 340 | |
11786802 | 341 | ret = td_io_queue(td, io_u); |
36167d82 JA |
342 | switch (ret) { |
343 | case FIO_Q_COMPLETED: | |
344 | if (io_u->error) | |
22819ec2 | 345 | ret = -io_u->error; |
9e9d164e | 346 | else if (io_u->resid) { |
36167d82 | 347 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 348 | struct fio_file *f = io_u->file; |
ebac4655 | 349 | |
9e9d164e JA |
350 | /* |
351 | * zero read, fail | |
352 | */ | |
353 | if (!bytes) { | |
354 | td_verror(td, ENODATA, "full resid"); | |
355 | put_io_u(td, io_u); | |
356 | break; | |
357 | } | |
8400d9b2 | 358 | |
36167d82 JA |
359 | io_u->xfer_buflen = io_u->resid; |
360 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
361 | io_u->offset += bytes; |
362 | ||
30061b97 JA |
363 | td->ts.short_io_u[io_u->ddir]++; |
364 | ||
d460eb31 | 365 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
366 | goto sync_done; |
367 | ||
11786802 JA |
368 | requeue_io_u(td, &io_u); |
369 | } else { | |
8400d9b2 | 370 | sync_done: |
11786802 JA |
371 | ret = io_u_sync_complete(td, io_u); |
372 | if (ret < 0) | |
373 | break; | |
36167d82 | 374 | } |
36167d82 JA |
375 | continue; |
376 | case FIO_Q_QUEUED: | |
377 | break; | |
755200a3 JA |
378 | case FIO_Q_BUSY: |
379 | requeue_io_u(td, &io_u); | |
5451792e JA |
380 | ret2 = td_io_commit(td); |
381 | if (ret2 < 0) | |
382 | ret = ret2; | |
755200a3 | 383 | break; |
36167d82 JA |
384 | default: |
385 | assert(ret < 0); | |
e1161c32 | 386 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
387 | break; |
388 | } | |
389 | ||
99784632 | 390 | if (ret < 0 || td->error) |
3af6ef39 JA |
391 | break; |
392 | ||
ebac4655 | 393 | /* |
3af6ef39 JA |
394 | * if we can queue more, do so. but check if there are |
395 | * completed io_u's first. | |
ebac4655 | 396 | */ |
97601024 | 397 | min_events = 0; |
e916b390 | 398 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
3af6ef39 | 399 | min_events = 1; |
3af6ef39 | 400 | |
2dc1bbeb JA |
401 | if (td->cur_depth > td->o.iodepth_low) |
402 | min_events = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
403 | } |
404 | ||
3af6ef39 JA |
405 | /* |
406 | * Reap required number of io units, if any, and do the | |
407 | * verification on them through the callback handler | |
408 | */ | |
d7762cf8 | 409 | if (io_u_queued_complete(td, min_events) < 0) |
ebac4655 | 410 | break; |
36167d82 | 411 | } |
ebac4655 | 412 | |
c01c0395 JA |
413 | if (!td->error) { |
414 | min_events = td->cur_depth; | |
415 | ||
416 | if (min_events) | |
417 | ret = io_u_queued_complete(td, min_events); | |
418 | } else | |
ebac4655 JA |
419 | cleanup_pending_aio(td); |
420 | ||
421 | td_set_runstate(td, TD_RUNNING); | |
422 | } | |
423 | ||
32cd46a0 | 424 | /* |
906c8d75 | 425 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
426 | * and reaps them, checking for rate and errors along the way. |
427 | */ | |
ebac4655 JA |
428 | static void do_io(struct thread_data *td) |
429 | { | |
02bcaa8c | 430 | struct timeval s; |
ebac4655 | 431 | unsigned long usec; |
af52b345 JA |
432 | unsigned int i; |
433 | int ret = 0; | |
ebac4655 | 434 | |
5853e5a8 JA |
435 | td_set_runstate(td, TD_RUNNING); |
436 | ||
7bb48f84 | 437 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
438 | struct timeval comp_time; |
439 | long bytes_done = 0; | |
84585003 | 440 | int min_evts = 0; |
ebac4655 | 441 | struct io_u *io_u; |
5451792e | 442 | int ret2; |
ebac4655 JA |
443 | |
444 | if (td->terminate) | |
445 | break; | |
446 | ||
3d7c391d | 447 | io_u = get_io_u(td); |
ebac4655 JA |
448 | if (!io_u) |
449 | break; | |
450 | ||
451 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
452 | |
453 | if (runtime_exceeded(td, &s)) { | |
454 | put_io_u(td, io_u); | |
90a87d4b | 455 | td->terminate = 1; |
97601024 JA |
456 | break; |
457 | } | |
36167d82 | 458 | |
b67740d3 JA |
459 | /* |
460 | * Add verification end_io handler, if asked to verify | |
461 | * a previously written file. | |
462 | */ | |
463 | if (td->o.verify != VERIFY_NONE) | |
464 | io_u->end_io = verify_io_u; | |
465 | ||
11786802 | 466 | ret = td_io_queue(td, io_u); |
36167d82 JA |
467 | switch (ret) { |
468 | case FIO_Q_COMPLETED: | |
5451792e JA |
469 | if (io_u->error) |
470 | ret = -io_u->error; | |
9e9d164e | 471 | else if (io_u->resid) { |
36167d82 | 472 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 473 | struct fio_file *f = io_u->file; |
36167d82 | 474 | |
9e9d164e JA |
475 | /* |
476 | * zero read, fail | |
477 | */ | |
478 | if (!bytes) { | |
479 | td_verror(td, ENODATA, "full resid"); | |
480 | put_io_u(td, io_u); | |
481 | break; | |
482 | } | |
483 | ||
cec6b55d | 484 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 485 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
486 | io_u->offset += bytes; |
487 | ||
30061b97 JA |
488 | td->ts.short_io_u[io_u->ddir]++; |
489 | ||
d460eb31 | 490 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
491 | goto sync_done; |
492 | ||
11786802 JA |
493 | requeue_io_u(td, &io_u); |
494 | } else { | |
5a7c5680 | 495 | sync_done: |
11786802 JA |
496 | fio_gettime(&comp_time, NULL); |
497 | bytes_done = io_u_sync_complete(td, io_u); | |
498 | if (bytes_done < 0) | |
499 | ret = bytes_done; | |
cec6b55d | 500 | } |
36167d82 JA |
501 | break; |
502 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
503 | /* |
504 | * if the engine doesn't have a commit hook, | |
505 | * the io_u is really queued. if it does have such | |
506 | * a hook, it has to call io_u_queued() itself. | |
507 | */ | |
508 | if (td->io_ops->commit == NULL) | |
509 | io_u_queued(td, io_u); | |
36167d82 | 510 | break; |
755200a3 JA |
511 | case FIO_Q_BUSY: |
512 | requeue_io_u(td, &io_u); | |
5451792e JA |
513 | ret2 = td_io_commit(td); |
514 | if (ret2 < 0) | |
515 | ret = ret2; | |
755200a3 | 516 | break; |
36167d82 JA |
517 | default: |
518 | assert(ret < 0); | |
519 | put_io_u(td, io_u); | |
520 | break; | |
ebac4655 JA |
521 | } |
522 | ||
99784632 | 523 | if (ret < 0 || td->error) |
36167d82 JA |
524 | break; |
525 | ||
97601024 JA |
526 | /* |
527 | * See if we need to complete some commands | |
528 | */ | |
755200a3 | 529 | if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) { |
97601024 | 530 | min_evts = 0; |
e916b390 | 531 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
36167d82 | 532 | min_evts = 1; |
ebac4655 | 533 | |
2dc1bbeb JA |
534 | if (td->cur_depth > td->o.iodepth_low) |
535 | min_evts = td->cur_depth - td->o.iodepth_low; | |
e916b390 JA |
536 | } |
537 | ||
97601024 | 538 | fio_gettime(&comp_time, NULL); |
d7762cf8 | 539 | bytes_done = io_u_queued_complete(td, min_evts); |
97601024 | 540 | if (bytes_done < 0) |
36167d82 | 541 | break; |
ebac4655 JA |
542 | } |
543 | ||
97601024 JA |
544 | if (!bytes_done) |
545 | continue; | |
546 | ||
ebac4655 JA |
547 | /* |
548 | * the rate is batched for now, it should work for batches | |
549 | * of completions except the very first one which may look | |
550 | * a little bursty | |
551 | */ | |
97601024 | 552 | usec = utime_since(&s, &comp_time); |
ebac4655 | 553 | |
413dd459 | 554 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 555 | |
97601024 | 556 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 557 | if (exitall_on_terminate) |
390c40e2 | 558 | terminate_threads(td->groupid); |
e1161c32 | 559 | td_verror(td, ENODATA, "check_min_rate"); |
ebac4655 JA |
560 | break; |
561 | } | |
562 | ||
2dc1bbeb | 563 | if (td->o.thinktime) { |
9c1f7434 JA |
564 | unsigned long long b; |
565 | ||
566 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 567 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
568 | int left; |
569 | ||
2dc1bbeb JA |
570 | if (td->o.thinktime_spin) |
571 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 572 | |
2dc1bbeb | 573 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
574 | if (left) |
575 | usec_sleep(td, left); | |
576 | } | |
9c1f7434 | 577 | } |
ebac4655 JA |
578 | } |
579 | ||
aa31f1f1 SL |
580 | if (td->o.fill_device && td->error == ENOSPC) { |
581 | td->error = 0; | |
582 | td->terminate = 1; | |
583 | } | |
4d2413c6 | 584 | if (!td->error) { |
3d7c391d JA |
585 | struct fio_file *f; |
586 | ||
c01c0395 JA |
587 | i = td->cur_depth; |
588 | if (i) | |
589 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 590 | |
2dc1bbeb | 591 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 592 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
593 | |
594 | for_each_file(td, f, i) { | |
595 | if (!(f->flags & FIO_FILE_OPEN)) | |
596 | continue; | |
858a3d47 | 597 | fio_io_sync(td, f); |
3d7b485f | 598 | } |
84585003 | 599 | } |
c01c0395 JA |
600 | } else |
601 | cleanup_pending_aio(td); | |
2ba1c290 JA |
602 | |
603 | /* | |
604 | * stop job if we failed doing any IO | |
605 | */ | |
606 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
607 | td->done = 1; | |
ebac4655 JA |
608 | } |
609 | ||
ebac4655 JA |
610 | static void cleanup_io_u(struct thread_data *td) |
611 | { | |
612 | struct list_head *entry, *n; | |
613 | struct io_u *io_u; | |
614 | ||
615 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
616 | io_u = list_entry(entry, struct io_u, list); | |
617 | ||
618 | list_del(&io_u->list); | |
619 | free(io_u); | |
620 | } | |
621 | ||
2f9ade3c | 622 | free_io_mem(td); |
ebac4655 JA |
623 | } |
624 | ||
6b9cea23 JA |
625 | /* |
626 | * "randomly" fill the buffer contents | |
627 | */ | |
e9459e5a | 628 | static void fill_io_buf(struct thread_data *td, struct io_u *io_u, int max_bs) |
6b9cea23 | 629 | { |
210f43b3 | 630 | long *ptr = io_u->buf; |
6b9cea23 | 631 | |
e9459e5a JA |
632 | if (!td->o.zero_buffers) { |
633 | while ((void *) ptr - io_u->buf < max_bs) { | |
210f43b3 | 634 | *ptr = rand() * GOLDEN_RATIO_PRIME; |
e9459e5a JA |
635 | ptr++; |
636 | } | |
637 | } else | |
638 | memset(ptr, 0, max_bs); | |
6b9cea23 JA |
639 | } |
640 | ||
ebac4655 JA |
641 | static int init_io_u(struct thread_data *td) |
642 | { | |
643 | struct io_u *io_u; | |
a00735e6 | 644 | unsigned int max_bs; |
ebac4655 JA |
645 | int i, max_units; |
646 | char *p; | |
647 | ||
1e3c09af | 648 | max_units = td->o.iodepth; |
2dc1bbeb | 649 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
c3990645 | 650 | td->orig_buffer_size = (unsigned long long) max_bs * (unsigned long long) max_units; |
74b025b0 | 651 | |
2dc1bbeb JA |
652 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) |
653 | td->orig_buffer_size = (td->orig_buffer_size + td->o.hugepage_size - 1) & ~(td->o.hugepage_size - 1); | |
ebac4655 | 654 | |
c3990645 JA |
655 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
656 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
657 | return 1; | |
658 | } | |
659 | ||
2f9ade3c JA |
660 | if (allocate_io_mem(td)) |
661 | return 1; | |
ebac4655 | 662 | |
b3f378e9 JA |
663 | if (td->o.odirect) |
664 | p = ALIGN(td->orig_buffer); | |
665 | else | |
666 | p = td->orig_buffer; | |
667 | ||
ebac4655 | 668 | for (i = 0; i < max_units; i++) { |
db1cd90b JA |
669 | if (td->terminate) |
670 | return 1; | |
ebac4655 JA |
671 | io_u = malloc(sizeof(*io_u)); |
672 | memset(io_u, 0, sizeof(*io_u)); | |
673 | INIT_LIST_HEAD(&io_u->list); | |
674 | ||
b4c5e1ac JA |
675 | if (!(td->io_ops->flags & FIO_NOIO)) { |
676 | io_u->buf = p + max_bs * i; | |
e9459e5a | 677 | |
b4c5e1ac JA |
678 | if (td_write(td)) |
679 | fill_io_buf(td, io_u, max_bs); | |
680 | } | |
6b9cea23 | 681 | |
b1ff3403 | 682 | io_u->index = i; |
0c6e7517 | 683 | io_u->flags = IO_U_F_FREE; |
ebac4655 JA |
684 | list_add(&io_u->list, &td->io_u_freelist); |
685 | } | |
686 | ||
433afcb4 JA |
687 | io_u_init_timeout(); |
688 | ||
ebac4655 JA |
689 | return 0; |
690 | } | |
691 | ||
da86774e JA |
692 | static int switch_ioscheduler(struct thread_data *td) |
693 | { | |
694 | char tmp[256], tmp2[128]; | |
695 | FILE *f; | |
696 | int ret; | |
697 | ||
ba0fbe10 | 698 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
699 | return 0; |
700 | ||
da86774e JA |
701 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
702 | ||
703 | f = fopen(tmp, "r+"); | |
704 | if (!f) { | |
cbf5c1d7 JA |
705 | if (errno == ENOENT) { |
706 | log_err("fio: os or kernel doesn't support IO scheduler switching\n"); | |
707 | return 0; | |
708 | } | |
709 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
710 | return 1; |
711 | } | |
712 | ||
713 | /* | |
714 | * Set io scheduler. | |
715 | */ | |
2dc1bbeb | 716 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 717 | if (ferror(f) || ret != 1) { |
e1161c32 | 718 | td_verror(td, errno, "fwrite"); |
da86774e JA |
719 | fclose(f); |
720 | return 1; | |
721 | } | |
722 | ||
723 | rewind(f); | |
724 | ||
725 | /* | |
726 | * Read back and check that the selected scheduler is now the default. | |
727 | */ | |
728 | ret = fread(tmp, 1, sizeof(tmp), f); | |
729 | if (ferror(f) || ret < 0) { | |
e1161c32 | 730 | td_verror(td, errno, "fread"); |
da86774e JA |
731 | fclose(f); |
732 | return 1; | |
733 | } | |
734 | ||
2dc1bbeb | 735 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 736 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 737 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 738 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
739 | fclose(f); |
740 | return 1; | |
741 | } | |
742 | ||
743 | fclose(f); | |
744 | return 0; | |
745 | } | |
746 | ||
492158cf JA |
747 | static int keep_running(struct thread_data *td) |
748 | { | |
749 | unsigned long long io_done; | |
750 | ||
20e354ef JA |
751 | if (td->done) |
752 | return 0; | |
492158cf JA |
753 | if (td->o.time_based) |
754 | return 1; | |
755 | if (td->o.loops) { | |
756 | td->o.loops--; | |
757 | return 1; | |
758 | } | |
759 | ||
48f5abd3 | 760 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] + td->io_skip_bytes; |
492158cf JA |
761 | if (io_done < td->o.size) |
762 | return 1; | |
763 | ||
764 | return 0; | |
765 | } | |
766 | ||
a978ba68 | 767 | static int clear_io_state(struct thread_data *td) |
ebac4655 | 768 | { |
53cdc686 | 769 | struct fio_file *f; |
af52b345 JA |
770 | unsigned int i; |
771 | int ret; | |
ebac4655 | 772 | |
756867bd | 773 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 774 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 775 | td->zone_bytes = 0; |
d7d3b49b | 776 | td->rate_bytes = 0; |
4e991c23 | 777 | td->rate_blocks = 0; |
38d77cae | 778 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 779 | |
c1324df1 JA |
780 | td->last_was_sync = 0; |
781 | ||
2ba1c290 JA |
782 | /* |
783 | * reset file done count if we are to start over | |
784 | */ | |
785 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 786 | td->nr_done_files = 0; |
281f9b63 | 787 | |
24ffd2c2 | 788 | close_files(td); |
53cdc686 | 789 | |
a978ba68 JA |
790 | ret = 0; |
791 | for_each_file(td, f, i) { | |
281f9b63 | 792 | f->flags &= ~FIO_FILE_DONE; |
a978ba68 JA |
793 | ret = td_io_open_file(td, f); |
794 | if (ret) | |
795 | break; | |
53cdc686 | 796 | } |
a978ba68 JA |
797 | |
798 | return ret; | |
ebac4655 JA |
799 | } |
800 | ||
906c8d75 JA |
801 | /* |
802 | * Entry point for the thread based jobs. The process based jobs end up | |
803 | * here as well, after a little setup. | |
804 | */ | |
ebac4655 JA |
805 | static void *thread_main(void *data) |
806 | { | |
10927316 | 807 | unsigned long long runtime[2], elapsed; |
ebac4655 | 808 | struct thread_data *td = data; |
a978ba68 | 809 | int clear_state; |
ebac4655 | 810 | |
2dc1bbeb | 811 | if (!td->o.use_thread) |
ebac4655 JA |
812 | setsid(); |
813 | ||
814 | td->pid = getpid(); | |
815 | ||
ee56ad50 JA |
816 | dprint(FD_PROCESS, "jobs pid=%d started\n", td->pid); |
817 | ||
aea47d44 JA |
818 | INIT_LIST_HEAD(&td->io_u_freelist); |
819 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 820 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 | 821 | INIT_LIST_HEAD(&td->io_log_list); |
8de8f047 | 822 | INIT_LIST_HEAD(&td->io_hist_list); |
bb5d7d0b | 823 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 824 | |
db1cd90b | 825 | td_set_runstate(td, TD_INITIALIZED); |
cdd18ad8 JA |
826 | fio_mutex_up(startup_mutex); |
827 | fio_mutex_down(td->mutex); | |
db1cd90b JA |
828 | |
829 | /* | |
cdd18ad8 | 830 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
831 | * eating a file descriptor |
832 | */ | |
cdd18ad8 | 833 | fio_mutex_remove(td->mutex); |
db1cd90b | 834 | |
d84f8d49 JA |
835 | /* |
836 | * May alter parameters that init_io_u() will use, so we need to | |
837 | * do this first. | |
838 | */ | |
839 | if (init_iolog(td)) | |
840 | goto err; | |
841 | ||
ebac4655 | 842 | if (init_io_u(td)) |
db1cd90b | 843 | goto err; |
ebac4655 | 844 | |
375b2695 | 845 | if (td->o.cpumask_set && fio_setaffinity(td) == -1) { |
e1161c32 | 846 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 847 | goto err; |
ebac4655 JA |
848 | } |
849 | ||
ac684785 | 850 | if (td->ioprio_set) { |
ebac4655 | 851 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 852 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 853 | goto err; |
ebac4655 JA |
854 | } |
855 | } | |
856 | ||
2dc1bbeb | 857 | if (nice(td->o.nice) == -1) { |
e1161c32 | 858 | td_verror(td, errno, "nice"); |
db1cd90b | 859 | goto err; |
b6f4d880 JA |
860 | } |
861 | ||
2dc1bbeb | 862 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 863 | goto err; |
37f56873 | 864 | |
2dc1bbeb | 865 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
866 | goto err; |
867 | ||
7d6c5283 JA |
868 | if (td_io_init(td)) |
869 | goto err; | |
870 | ||
b5af8293 JA |
871 | if (open_files(td)) |
872 | goto err; | |
873 | ||
68727076 JA |
874 | if (init_random_map(td)) |
875 | goto err; | |
876 | ||
2dc1bbeb JA |
877 | if (td->o.exec_prerun) { |
878 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
879 | goto err; |
880 | } | |
4e0ba8af | 881 | |
69008999 | 882 | fio_gettime(&td->epoch, NULL); |
433afcb4 | 883 | memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); |
756867bd | 884 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
885 | |
886 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 887 | clear_state = 0; |
492158cf | 888 | while (keep_running(td)) { |
02bcaa8c | 889 | fio_gettime(&td->start, NULL); |
756867bd | 890 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 | 891 | |
2dc1bbeb | 892 | if (td->o.ratemin) |
756867bd | 893 | memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate)); |
ebac4655 | 894 | |
a978ba68 JA |
895 | if (clear_state && clear_io_state(td)) |
896 | break; | |
897 | ||
ebac4655 JA |
898 | prune_io_piece_log(td); |
899 | ||
ba0fbe10 | 900 | do_io(td); |
ebac4655 | 901 | |
a978ba68 JA |
902 | clear_state = 1; |
903 | ||
38d77cae JA |
904 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
905 | if (td->rw_end_set[DDIR_READ]) | |
906 | elapsed = utime_since(&td->start, &td->rw_end[DDIR_READ]); | |
907 | else | |
908 | elapsed = utime_since_now(&td->start); | |
909 | ||
910 | runtime[DDIR_READ] += elapsed; | |
911 | } | |
912 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
913 | if (td->rw_end_set[DDIR_WRITE]) | |
914 | elapsed = utime_since(&td->start, &td->rw_end[DDIR_WRITE]); | |
915 | else | |
916 | elapsed = utime_since_now(&td->start); | |
917 | ||
918 | runtime[DDIR_WRITE] += elapsed; | |
919 | } | |
413dd459 | 920 | |
ebac4655 JA |
921 | if (td->error || td->terminate) |
922 | break; | |
923 | ||
e84c73a8 SL |
924 | if (!td->o.do_verify || |
925 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 926 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
927 | continue; |
928 | ||
a978ba68 JA |
929 | if (clear_io_state(td)) |
930 | break; | |
931 | ||
02bcaa8c | 932 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
933 | |
934 | do_verify(td); | |
935 | ||
69008999 | 936 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
937 | |
938 | if (td->error || td->terminate) | |
939 | break; | |
940 | } | |
941 | ||
36dff966 | 942 | update_rusage_stat(td); |
47f0cc48 YHJT |
943 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
944 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
945 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
946 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
947 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
948 | ||
949 | if (td->ts.bw_log) | |
950 | finish_log(td, td->ts.bw_log, "bw"); | |
951 | if (td->ts.slat_log) | |
952 | finish_log(td, td->ts.slat_log, "slat"); | |
953 | if (td->ts.clat_log) | |
954 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb JA |
955 | if (td->o.exec_postrun) { |
956 | if (system(td->o.exec_postrun) < 0) | |
957 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 958 | } |
ebac4655 JA |
959 | |
960 | if (exitall_on_terminate) | |
390c40e2 | 961 | terminate_threads(td->groupid); |
ebac4655 JA |
962 | |
963 | err: | |
5bf13a5a JA |
964 | if (td->error) |
965 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
24ffd2c2 | 966 | close_and_free_files(td); |
2866c82d | 967 | close_ioengine(td); |
ebac4655 | 968 | cleanup_io_u(td); |
f29b25a3 JA |
969 | |
970 | /* | |
971 | * do this very late, it will log file closing as well | |
972 | */ | |
973 | if (td->o.write_iolog_file) | |
974 | write_iolog_close(td); | |
975 | ||
d23bb327 | 976 | options_mem_free(td); |
ebac4655 | 977 | td_set_runstate(td, TD_EXITED); |
43d76807 | 978 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
979 | } |
980 | ||
906c8d75 JA |
981 | /* |
982 | * We cannot pass the td data into a forked process, so attach the td and | |
983 | * pass it to the thread worker. | |
984 | */ | |
a6418147 | 985 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
986 | { |
987 | struct thread_data *td; | |
a6418147 | 988 | void *data, *ret; |
ebac4655 JA |
989 | |
990 | data = shmat(shmid, NULL, 0); | |
991 | if (data == (void *) -1) { | |
a6418147 JA |
992 | int __err = errno; |
993 | ||
ebac4655 | 994 | perror("shmat"); |
a6418147 | 995 | return __err; |
ebac4655 JA |
996 | } |
997 | ||
998 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 999 | ret = thread_main(td); |
ebac4655 | 1000 | shmdt(data); |
43d76807 | 1001 | return (int) (unsigned long) ret; |
ebac4655 JA |
1002 | } |
1003 | ||
906c8d75 JA |
1004 | /* |
1005 | * Run over the job map and reap the threads that have exited, if any. | |
1006 | */ | |
ebac4655 JA |
1007 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1008 | { | |
34572e28 | 1009 | struct thread_data *td; |
1f809d15 | 1010 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1011 | |
1012 | /* | |
1013 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1014 | */ | |
1f809d15 | 1015 | realthreads = pending = cputhreads = 0; |
34572e28 | 1016 | for_each_td(td, i) { |
3707f45b | 1017 | int flags = 0; |
a2f77c9f | 1018 | |
84585003 JA |
1019 | /* |
1020 | * ->io_ops is NULL for a thread that has closed its | |
1021 | * io engine | |
1022 | */ | |
ba0fbe10 | 1023 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1024 | cputhreads++; |
1f809d15 JA |
1025 | else |
1026 | realthreads++; | |
b990b5c0 | 1027 | |
a1dedc1f AC |
1028 | if (!td->pid) { |
1029 | pending++; | |
1030 | continue; | |
1031 | } | |
1032 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1033 | continue; |
2dc1bbeb | 1034 | if (td->o.use_thread) { |
3707f45b JA |
1035 | if (td->runstate == TD_EXITED) { |
1036 | td_set_runstate(td, TD_REAPED); | |
1037 | goto reaped; | |
1038 | } | |
1039 | continue; | |
1040 | } | |
a2f77c9f JA |
1041 | |
1042 | flags = WNOHANG; | |
1043 | if (td->runstate == TD_EXITED) | |
1044 | flags = 0; | |
1045 | ||
1046 | /* | |
1047 | * check if someone quit or got killed in an unusual way | |
1048 | */ | |
1049 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1050 | if (ret < 0) { |
a2f77c9f JA |
1051 | if (errno == ECHILD) { |
1052 | log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate); | |
1053 | td_set_runstate(td, TD_REAPED); | |
1054 | goto reaped; | |
1055 | } | |
1056 | perror("waitpid"); | |
1057 | } else if (ret == td->pid) { | |
1058 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1059 | int sig = WTERMSIG(status); |
1060 | ||
d9244941 JA |
1061 | if (sig != SIGQUIT) |
1062 | log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); | |
fab6aa71 JA |
1063 | td_set_runstate(td, TD_REAPED); |
1064 | goto reaped; | |
1065 | } | |
a2f77c9f JA |
1066 | if (WIFEXITED(status)) { |
1067 | if (WEXITSTATUS(status) && !td->error) | |
1068 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1069 | |
a2f77c9f JA |
1070 | td_set_runstate(td, TD_REAPED); |
1071 | goto reaped; | |
a6418147 JA |
1072 | } |
1073 | } | |
ebac4655 | 1074 | |
a2f77c9f JA |
1075 | /* |
1076 | * thread is not dead, continue | |
1077 | */ | |
e48676ba | 1078 | pending++; |
a2f77c9f | 1079 | continue; |
fab6aa71 | 1080 | reaped: |
2dc1bbeb | 1081 | if (td->o.use_thread) { |
3707f45b JA |
1082 | long ret; |
1083 | ||
ee56ad50 JA |
1084 | dprint(FD_PROCESS, "joining tread %d\n", td->pid); |
1085 | if (pthread_join(td->thread, (void *) &ret)) { | |
1086 | dprint(FD_PROCESS, "join failed %ld\n", ret); | |
3707f45b | 1087 | perror("pthread_join"); |
ee56ad50 | 1088 | } |
3707f45b JA |
1089 | } |
1090 | ||
ebac4655 | 1091 | (*nr_running)--; |
2dc1bbeb JA |
1092 | (*m_rate) -= td->o.ratemin; |
1093 | (*t_rate) -= td->o.rate; | |
e48676ba | 1094 | pending--; |
a2f77c9f JA |
1095 | |
1096 | if (td->error) | |
1097 | exit_value++; | |
ebac4655 | 1098 | } |
b990b5c0 | 1099 | |
1f809d15 | 1100 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1101 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1102 | } |
1103 | ||
906c8d75 JA |
1104 | /* |
1105 | * Main function for kicking off and reaping jobs, as needed. | |
1106 | */ | |
ebac4655 JA |
1107 | static void run_threads(void) |
1108 | { | |
ebac4655 JA |
1109 | struct thread_data *td; |
1110 | unsigned long spent; | |
1111 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1112 | |
2f9ade3c JA |
1113 | if (fio_pin_memory()) |
1114 | return; | |
ebac4655 | 1115 | |
c6ae0a5b | 1116 | if (!terse_output) { |
9cedf167 JA |
1117 | printf("Starting "); |
1118 | if (nr_thread) | |
1119 | printf("%d thread%s", nr_thread, nr_thread > 1 ? "s" : ""); | |
1120 | if (nr_process) { | |
1121 | if (nr_thread) | |
1122 | printf(" and "); | |
1123 | printf("%d process%s", nr_process, nr_process > 1 ? "es" : ""); | |
1124 | } | |
1125 | printf("\n"); | |
c6ae0a5b JA |
1126 | fflush(stdout); |
1127 | } | |
c04f7ec3 | 1128 | |
4efa970e JA |
1129 | signal(SIGINT, sig_handler); |
1130 | signal(SIGALRM, sig_handler); | |
1131 | ||
ebac4655 JA |
1132 | todo = thread_number; |
1133 | nr_running = 0; | |
1134 | nr_started = 0; | |
1135 | m_rate = t_rate = 0; | |
1136 | ||
34572e28 | 1137 | for_each_td(td, i) { |
263e529f | 1138 | print_status_init(td->thread_number - 1); |
ebac4655 | 1139 | |
2dc1bbeb | 1140 | if (!td->o.create_serialize) { |
380cf265 | 1141 | init_disk_util(td); |
ebac4655 | 1142 | continue; |
380cf265 | 1143 | } |
ebac4655 JA |
1144 | |
1145 | /* | |
1146 | * do file setup here so it happens sequentially, | |
1147 | * we don't want X number of threads getting their | |
1148 | * client data interspersed on disk | |
1149 | */ | |
53cdc686 | 1150 | if (setup_files(td)) { |
5bf13a5a JA |
1151 | exit_value++; |
1152 | if (td->error) | |
1153 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
ebac4655 JA |
1154 | td_set_runstate(td, TD_REAPED); |
1155 | todo--; | |
1156 | } | |
380cf265 JA |
1157 | |
1158 | init_disk_util(td); | |
ebac4655 JA |
1159 | } |
1160 | ||
a2f77c9f JA |
1161 | set_genesis_time(); |
1162 | ||
ebac4655 | 1163 | while (todo) { |
75154845 JA |
1164 | struct thread_data *map[MAX_JOBS]; |
1165 | struct timeval this_start; | |
1166 | int this_jobs = 0, left; | |
1167 | ||
ebac4655 JA |
1168 | /* |
1169 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1170 | */ | |
34572e28 | 1171 | for_each_td(td, i) { |
ebac4655 JA |
1172 | if (td->runstate != TD_NOT_CREATED) |
1173 | continue; | |
1174 | ||
1175 | /* | |
1176 | * never got a chance to start, killed by other | |
1177 | * thread for some reason | |
1178 | */ | |
1179 | if (td->terminate) { | |
1180 | todo--; | |
1181 | continue; | |
1182 | } | |
1183 | ||
2dc1bbeb | 1184 | if (td->o.start_delay) { |
263e529f | 1185 | spent = mtime_since_genesis(); |
ebac4655 | 1186 | |
2dc1bbeb | 1187 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1188 | continue; |
1189 | } | |
1190 | ||
2dc1bbeb | 1191 | if (td->o.stonewall && (nr_started || nr_running)) |
ebac4655 JA |
1192 | break; |
1193 | ||
75154845 JA |
1194 | /* |
1195 | * Set state to created. Thread will transition | |
1196 | * to TD_INITIALIZED when it's done setting up. | |
1197 | */ | |
ebac4655 | 1198 | td_set_runstate(td, TD_CREATED); |
75154845 | 1199 | map[this_jobs++] = td; |
ebac4655 JA |
1200 | nr_started++; |
1201 | ||
2dc1bbeb | 1202 | if (td->o.use_thread) { |
ee56ad50 | 1203 | dprint(FD_PROCESS, "will pthread_create\n"); |
ebac4655 JA |
1204 | if (pthread_create(&td->thread, NULL, thread_main, td)) { |
1205 | perror("thread_create"); | |
1206 | nr_started--; | |
c8bb6faf | 1207 | break; |
ebac4655 JA |
1208 | } |
1209 | } else { | |
ee56ad50 | 1210 | dprint(FD_PROCESS, "will fork\n"); |
07739b57 | 1211 | if (!fork()) { |
a6418147 JA |
1212 | int ret = fork_main(shm_id, i); |
1213 | ||
1214 | exit(ret); | |
ebac4655 JA |
1215 | } |
1216 | } | |
cdd18ad8 | 1217 | fio_mutex_down(startup_mutex); |
ebac4655 JA |
1218 | } |
1219 | ||
1220 | /* | |
75154845 JA |
1221 | * Wait for the started threads to transition to |
1222 | * TD_INITIALIZED. | |
ebac4655 | 1223 | */ |
02bcaa8c | 1224 | fio_gettime(&this_start, NULL); |
75154845 | 1225 | left = this_jobs; |
6ce15a32 | 1226 | while (left && !fio_abort) { |
75154845 JA |
1227 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1228 | break; | |
1229 | ||
1230 | usleep(100000); | |
1231 | ||
1232 | for (i = 0; i < this_jobs; i++) { | |
1233 | td = map[i]; | |
1234 | if (!td) | |
1235 | continue; | |
b6f4d880 | 1236 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1237 | map[i] = NULL; |
1238 | left--; | |
b6f4d880 JA |
1239 | } else if (td->runstate >= TD_EXITED) { |
1240 | map[i] = NULL; | |
1241 | left--; | |
1242 | todo--; | |
1243 | nr_running++; /* work-around... */ | |
75154845 JA |
1244 | } |
1245 | } | |
1246 | } | |
1247 | ||
1248 | if (left) { | |
3b70d7e5 | 1249 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1250 | for (i = 0; i < this_jobs; i++) { |
1251 | td = map[i]; | |
1252 | if (!td) | |
1253 | continue; | |
1254 | kill(td->pid, SIGTERM); | |
1255 | } | |
1256 | break; | |
1257 | } | |
1258 | ||
1259 | /* | |
b6f4d880 | 1260 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1261 | */ |
34572e28 | 1262 | for_each_td(td, i) { |
75154845 | 1263 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1264 | continue; |
1265 | ||
1266 | td_set_runstate(td, TD_RUNNING); | |
1267 | nr_running++; | |
1268 | nr_started--; | |
2dc1bbeb JA |
1269 | m_rate += td->o.ratemin; |
1270 | t_rate += td->o.rate; | |
75154845 | 1271 | todo--; |
cdd18ad8 | 1272 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1273 | } |
1274 | ||
1275 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1276 | ||
1277 | if (todo) | |
1278 | usleep(100000); | |
1279 | } | |
1280 | ||
1281 | while (nr_running) { | |
1282 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1283 | usleep(10000); | |
1284 | } | |
1285 | ||
1286 | update_io_ticks(); | |
2f9ade3c | 1287 | fio_unpin_memory(); |
ebac4655 JA |
1288 | } |
1289 | ||
ebac4655 JA |
1290 | int main(int argc, char *argv[]) |
1291 | { | |
29d610e1 JA |
1292 | long ps; |
1293 | ||
dbe1125e JA |
1294 | /* |
1295 | * We need locale for number printing, if it isn't set then just | |
1296 | * go with the US format. | |
1297 | */ | |
1298 | if (!getenv("LC_NUMERIC")) | |
1299 | setlocale(LC_NUMERIC, "en_US"); | |
1300 | ||
ebac4655 JA |
1301 | if (parse_options(argc, argv)) |
1302 | return 1; | |
1303 | ||
4b472fa3 JA |
1304 | if (!thread_number) |
1305 | return 0; | |
ebac4655 | 1306 | |
29d610e1 JA |
1307 | ps = sysconf(_SC_PAGESIZE); |
1308 | if (ps < 0) { | |
1309 | log_err("Failed to get page size\n"); | |
1310 | return 1; | |
1311 | } | |
1312 | ||
cfc99db7 | 1313 | page_size = ps; |
29d610e1 JA |
1314 | page_mask = ps - 1; |
1315 | ||
bb3884d8 JA |
1316 | if (write_bw_log) { |
1317 | setup_log(&agg_io_log[DDIR_READ]); | |
1318 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1319 | } | |
1320 | ||
cdd18ad8 | 1321 | startup_mutex = fio_mutex_init(0); |
07739b57 | 1322 | |
a2f77c9f JA |
1323 | set_genesis_time(); |
1324 | ||
ebac4655 JA |
1325 | disk_util_timer_arm(); |
1326 | ||
1327 | run_threads(); | |
6ce15a32 | 1328 | |
bb3884d8 | 1329 | if (!fio_abort) { |
6ce15a32 | 1330 | show_run_stats(); |
bb3884d8 JA |
1331 | if (write_bw_log) { |
1332 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1333 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1334 | } | |
1335 | } | |
ebac4655 | 1336 | |
cdd18ad8 | 1337 | fio_mutex_remove(startup_mutex); |
437c9b71 | 1338 | return exit_value; |
ebac4655 | 1339 | } |