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 | ||
5921e80c JA |
68 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
69 | td->runstate, 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 | 82 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 83 | td->o.name, (int) 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 | { | |
6c28f0bb JA |
104 | if (!threads) |
105 | return; | |
106 | ||
ebac4655 | 107 | switch (sig) { |
5ec10eaa JA |
108 | case SIGALRM: |
109 | update_io_ticks(); | |
110 | disk_util_timer_arm(); | |
111 | print_thread_status(); | |
112 | break; | |
113 | default: | |
114 | printf("\nfio: terminating on signal %d\n", sig); | |
115 | fflush(stdout); | |
116 | terminate_threads(TERMINATE_ALL); | |
117 | break; | |
ebac4655 JA |
118 | } |
119 | } | |
120 | ||
906c8d75 JA |
121 | /* |
122 | * Check if we are above the minimum rate given. | |
123 | */ | |
ebac4655 JA |
124 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
125 | { | |
0904200b | 126 | unsigned long long bytes = 0; |
4e991c23 | 127 | unsigned long iops = 0; |
ebac4655 JA |
128 | unsigned long spent; |
129 | unsigned long rate; | |
ebac4655 | 130 | |
780bf1a1 JA |
131 | /* |
132 | * No minimum rate set, always ok | |
133 | */ | |
2dc1bbeb | 134 | if (!td->o.ratemin && !td->o.rate_iops_min) |
780bf1a1 JA |
135 | return 0; |
136 | ||
ebac4655 JA |
137 | /* |
138 | * allow a 2 second settle period in the beginning | |
139 | */ | |
140 | if (mtime_since(&td->start, now) < 2000) | |
141 | return 0; | |
142 | ||
4e991c23 JA |
143 | if (td_read(td)) { |
144 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 145 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
146 | } |
147 | if (td_write(td)) { | |
148 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 149 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 150 | } |
0904200b | 151 | |
ebac4655 JA |
152 | /* |
153 | * if rate blocks is set, sample is running | |
154 | */ | |
4e991c23 | 155 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 156 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 157 | if (spent < td->o.ratecycle) |
ebac4655 JA |
158 | return 0; |
159 | ||
2dc1bbeb | 160 | if (td->o.rate) { |
4e991c23 JA |
161 | /* |
162 | * check bandwidth specified rate | |
163 | */ | |
164 | if (bytes < td->rate_bytes) { | |
5ec10eaa JA |
165 | log_err("%s: min rate %u not met\n", td->o.name, |
166 | td->o.ratemin); | |
4e991c23 JA |
167 | return 1; |
168 | } else { | |
169 | rate = (bytes - td->rate_bytes) / spent; | |
5ec10eaa JA |
170 | if (rate < td->o.ratemin || |
171 | bytes < td->rate_bytes) { | |
172 | log_err("%s: min rate %u not met, got" | |
173 | " %luKiB/sec\n", td->o.name, | |
174 | td->o.ratemin, rate); | |
4e991c23 JA |
175 | return 1; |
176 | } | |
177 | } | |
413dd459 | 178 | } else { |
4e991c23 JA |
179 | /* |
180 | * checks iops specified rate | |
181 | */ | |
2dc1bbeb | 182 | if (iops < td->o.rate_iops) { |
5ec10eaa JA |
183 | log_err("%s: min iops rate %u not met\n", |
184 | td->o.name, td->o.rate_iops); | |
413dd459 | 185 | return 1; |
4e991c23 JA |
186 | } else { |
187 | rate = (iops - td->rate_blocks) / spent; | |
5ec10eaa JA |
188 | if (rate < td->o.rate_iops_min || |
189 | iops < td->rate_blocks) { | |
190 | log_err("%s: min iops rate %u not met," | |
191 | " got %lu\n", td->o.name, | |
192 | td->o.rate_iops_min, | |
193 | rate); | |
4e991c23 | 194 | } |
413dd459 | 195 | } |
ebac4655 JA |
196 | } |
197 | } | |
198 | ||
0904200b | 199 | td->rate_bytes = bytes; |
4e991c23 | 200 | td->rate_blocks = iops; |
ebac4655 JA |
201 | memcpy(&td->lastrate, now, sizeof(*now)); |
202 | return 0; | |
203 | } | |
204 | ||
205 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
206 | { | |
2dc1bbeb | 207 | if (!td->o.timeout) |
ebac4655 | 208 | return 0; |
2dc1bbeb | 209 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
210 | return 1; |
211 | ||
212 | return 0; | |
213 | } | |
214 | ||
906c8d75 JA |
215 | /* |
216 | * When job exits, we can cancel the in-flight IO if we are using async | |
217 | * io. Attempt to do so. | |
218 | */ | |
ebac4655 JA |
219 | static void cleanup_pending_aio(struct thread_data *td) |
220 | { | |
ebac4655 | 221 | struct list_head *entry, *n; |
ebac4655 JA |
222 | struct io_u *io_u; |
223 | int r; | |
224 | ||
225 | /* | |
226 | * get immediately available events, if any | |
227 | */ | |
d7762cf8 | 228 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
229 | if (r < 0) |
230 | return; | |
ebac4655 JA |
231 | |
232 | /* | |
233 | * now cancel remaining active events | |
234 | */ | |
2866c82d | 235 | if (td->io_ops->cancel) { |
ebac4655 JA |
236 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
237 | io_u = list_entry(entry, struct io_u, list); | |
238 | ||
0c6e7517 JA |
239 | /* |
240 | * if the io_u isn't in flight, then that generally | |
241 | * means someone leaked an io_u. complain but fix | |
242 | * it up, so we don't stall here. | |
243 | */ | |
244 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
245 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 246 | put_io_u(td, io_u); |
0c6e7517 JA |
247 | } else { |
248 | r = td->io_ops->cancel(td, io_u); | |
249 | if (!r) | |
250 | put_io_u(td, io_u); | |
251 | } | |
ebac4655 JA |
252 | } |
253 | } | |
254 | ||
97601024 | 255 | if (td->cur_depth) |
d7762cf8 | 256 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
257 | } |
258 | ||
858a3d47 JA |
259 | /* |
260 | * Helper to handle the final sync of a file. Works just like the normal | |
261 | * io path, just does everything sync. | |
262 | */ | |
263 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
264 | { | |
265 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
266 | int ret; |
267 | ||
268 | if (!io_u) | |
269 | return 1; | |
270 | ||
271 | io_u->ddir = DDIR_SYNC; | |
272 | io_u->file = f; | |
273 | ||
274 | if (td_io_prep(td, io_u)) { | |
275 | put_io_u(td, io_u); | |
276 | return 1; | |
277 | } | |
278 | ||
755200a3 | 279 | requeue: |
858a3d47 | 280 | ret = td_io_queue(td, io_u); |
36167d82 | 281 | if (ret < 0) { |
e1161c32 | 282 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 283 | put_io_u(td, io_u); |
858a3d47 | 284 | return 1; |
36167d82 | 285 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 286 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 287 | return 1; |
36167d82 JA |
288 | } else if (ret == FIO_Q_COMPLETED) { |
289 | if (io_u->error) { | |
e1161c32 | 290 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
291 | return 1; |
292 | } | |
858a3d47 | 293 | |
d7762cf8 | 294 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 295 | return 1; |
755200a3 JA |
296 | } else if (ret == FIO_Q_BUSY) { |
297 | if (td_io_commit(td)) | |
298 | return 1; | |
299 | goto requeue; | |
858a3d47 JA |
300 | } |
301 | ||
302 | return 0; | |
303 | } | |
304 | ||
906c8d75 | 305 | /* |
34403fb1 | 306 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
307 | * reads the blocks back in, and checks the crc/md5 of the data. |
308 | */ | |
1e97cce9 | 309 | static void do_verify(struct thread_data *td) |
ebac4655 | 310 | { |
53cdc686 | 311 | struct fio_file *f; |
36167d82 | 312 | struct io_u *io_u; |
af52b345 JA |
313 | int ret, min_events; |
314 | unsigned int i; | |
e5b401d4 JA |
315 | |
316 | /* | |
317 | * sync io first and invalidate cache, to make sure we really | |
318 | * read from disk. | |
319 | */ | |
320 | for_each_file(td, f, i) { | |
3d7b485f JA |
321 | if (!(f->flags & FIO_FILE_OPEN)) |
322 | continue; | |
b2fdda43 JA |
323 | if (fio_io_sync(td, f)) |
324 | break; | |
325 | if (file_invalidate_cache(td, f)) | |
326 | break; | |
e5b401d4 | 327 | } |
ebac4655 | 328 | |
b2fdda43 JA |
329 | if (td->error) |
330 | return; | |
331 | ||
ebac4655 JA |
332 | td_set_runstate(td, TD_VERIFYING); |
333 | ||
36167d82 JA |
334 | io_u = NULL; |
335 | while (!td->terminate) { | |
5451792e JA |
336 | int ret2; |
337 | ||
ebac4655 JA |
338 | io_u = __get_io_u(td); |
339 | if (!io_u) | |
340 | break; | |
341 | ||
069c2918 JA |
342 | if (runtime_exceeded(td, &io_u->start_time)) { |
343 | put_io_u(td, io_u); | |
90a87d4b | 344 | td->terminate = 1; |
02bcaa8c | 345 | break; |
069c2918 | 346 | } |
02bcaa8c | 347 | |
069c2918 JA |
348 | if (get_next_verify(td, io_u)) { |
349 | put_io_u(td, io_u); | |
ebac4655 | 350 | break; |
069c2918 | 351 | } |
ebac4655 | 352 | |
069c2918 JA |
353 | if (td_io_prep(td, io_u)) { |
354 | put_io_u(td, io_u); | |
53cdc686 | 355 | break; |
069c2918 | 356 | } |
d7762cf8 JA |
357 | |
358 | io_u->end_io = verify_io_u; | |
53cdc686 | 359 | |
11786802 | 360 | ret = td_io_queue(td, io_u); |
36167d82 JA |
361 | switch (ret) { |
362 | case FIO_Q_COMPLETED: | |
363 | if (io_u->error) | |
22819ec2 | 364 | ret = -io_u->error; |
9e9d164e | 365 | else if (io_u->resid) { |
36167d82 | 366 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 367 | struct fio_file *f = io_u->file; |
ebac4655 | 368 | |
9e9d164e JA |
369 | /* |
370 | * zero read, fail | |
371 | */ | |
372 | if (!bytes) { | |
373 | td_verror(td, ENODATA, "full resid"); | |
374 | put_io_u(td, io_u); | |
375 | break; | |
376 | } | |
8400d9b2 | 377 | |
36167d82 JA |
378 | io_u->xfer_buflen = io_u->resid; |
379 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
380 | io_u->offset += bytes; |
381 | ||
30061b97 JA |
382 | td->ts.short_io_u[io_u->ddir]++; |
383 | ||
d460eb31 | 384 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
385 | goto sync_done; |
386 | ||
11786802 JA |
387 | requeue_io_u(td, &io_u); |
388 | } else { | |
8400d9b2 | 389 | sync_done: |
11786802 JA |
390 | ret = io_u_sync_complete(td, io_u); |
391 | if (ret < 0) | |
392 | break; | |
36167d82 | 393 | } |
36167d82 JA |
394 | continue; |
395 | case FIO_Q_QUEUED: | |
396 | break; | |
755200a3 JA |
397 | case FIO_Q_BUSY: |
398 | requeue_io_u(td, &io_u); | |
5451792e JA |
399 | ret2 = td_io_commit(td); |
400 | if (ret2 < 0) | |
401 | ret = ret2; | |
755200a3 | 402 | break; |
36167d82 JA |
403 | default: |
404 | assert(ret < 0); | |
e1161c32 | 405 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
406 | break; |
407 | } | |
408 | ||
99784632 | 409 | if (ret < 0 || td->error) |
3af6ef39 JA |
410 | break; |
411 | ||
ebac4655 | 412 | /* |
3af6ef39 JA |
413 | * if we can queue more, do so. but check if there are |
414 | * completed io_u's first. | |
ebac4655 | 415 | */ |
97601024 | 416 | min_events = 0; |
e916b390 | 417 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
838bc709 | 418 | if (td->cur_depth >= td->o.iodepth_low) |
2dc1bbeb | 419 | min_events = td->cur_depth - td->o.iodepth_low; |
838bc709 JA |
420 | if (!min_events) |
421 | min_events = 1; | |
e916b390 JA |
422 | } |
423 | ||
3af6ef39 JA |
424 | /* |
425 | * Reap required number of io units, if any, and do the | |
426 | * verification on them through the callback handler | |
427 | */ | |
d7762cf8 | 428 | if (io_u_queued_complete(td, min_events) < 0) |
ebac4655 | 429 | break; |
36167d82 | 430 | } |
ebac4655 | 431 | |
c01c0395 JA |
432 | if (!td->error) { |
433 | min_events = td->cur_depth; | |
434 | ||
435 | if (min_events) | |
436 | ret = io_u_queued_complete(td, min_events); | |
437 | } else | |
ebac4655 JA |
438 | cleanup_pending_aio(td); |
439 | ||
440 | td_set_runstate(td, TD_RUNNING); | |
441 | } | |
442 | ||
32cd46a0 | 443 | /* |
906c8d75 | 444 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
445 | * and reaps them, checking for rate and errors along the way. |
446 | */ | |
ebac4655 JA |
447 | static void do_io(struct thread_data *td) |
448 | { | |
02bcaa8c | 449 | struct timeval s; |
ebac4655 | 450 | unsigned long usec; |
af52b345 JA |
451 | unsigned int i; |
452 | int ret = 0; | |
ebac4655 | 453 | |
5853e5a8 JA |
454 | td_set_runstate(td, TD_RUNNING); |
455 | ||
7bb48f84 | 456 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
457 | struct timeval comp_time; |
458 | long bytes_done = 0; | |
84585003 | 459 | int min_evts = 0; |
ebac4655 | 460 | struct io_u *io_u; |
5451792e | 461 | int ret2; |
ebac4655 JA |
462 | |
463 | if (td->terminate) | |
464 | break; | |
465 | ||
3d7c391d | 466 | io_u = get_io_u(td); |
ebac4655 JA |
467 | if (!io_u) |
468 | break; | |
469 | ||
470 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
471 | |
472 | if (runtime_exceeded(td, &s)) { | |
473 | put_io_u(td, io_u); | |
90a87d4b | 474 | td->terminate = 1; |
97601024 JA |
475 | break; |
476 | } | |
36167d82 | 477 | |
b67740d3 JA |
478 | /* |
479 | * Add verification end_io handler, if asked to verify | |
480 | * a previously written file. | |
481 | */ | |
cc3fc855 | 482 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) { |
b67740d3 | 483 | io_u->end_io = verify_io_u; |
d8fab1b6 JA |
484 | td_set_runstate(td, TD_VERIFYING); |
485 | } else | |
486 | td_set_runstate(td, TD_RUNNING); | |
b67740d3 | 487 | |
11786802 | 488 | ret = td_io_queue(td, io_u); |
36167d82 JA |
489 | switch (ret) { |
490 | case FIO_Q_COMPLETED: | |
5451792e JA |
491 | if (io_u->error) |
492 | ret = -io_u->error; | |
9e9d164e | 493 | else if (io_u->resid) { |
36167d82 | 494 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 495 | struct fio_file *f = io_u->file; |
36167d82 | 496 | |
9e9d164e JA |
497 | /* |
498 | * zero read, fail | |
499 | */ | |
500 | if (!bytes) { | |
501 | td_verror(td, ENODATA, "full resid"); | |
502 | put_io_u(td, io_u); | |
503 | break; | |
504 | } | |
505 | ||
cec6b55d | 506 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 507 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
508 | io_u->offset += bytes; |
509 | ||
30061b97 JA |
510 | td->ts.short_io_u[io_u->ddir]++; |
511 | ||
d460eb31 | 512 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
513 | goto sync_done; |
514 | ||
11786802 JA |
515 | requeue_io_u(td, &io_u); |
516 | } else { | |
5a7c5680 | 517 | sync_done: |
11786802 JA |
518 | fio_gettime(&comp_time, NULL); |
519 | bytes_done = io_u_sync_complete(td, io_u); | |
520 | if (bytes_done < 0) | |
521 | ret = bytes_done; | |
cec6b55d | 522 | } |
36167d82 JA |
523 | break; |
524 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
525 | /* |
526 | * if the engine doesn't have a commit hook, | |
527 | * the io_u is really queued. if it does have such | |
528 | * a hook, it has to call io_u_queued() itself. | |
529 | */ | |
530 | if (td->io_ops->commit == NULL) | |
531 | io_u_queued(td, io_u); | |
36167d82 | 532 | break; |
755200a3 JA |
533 | case FIO_Q_BUSY: |
534 | requeue_io_u(td, &io_u); | |
5451792e JA |
535 | ret2 = td_io_commit(td); |
536 | if (ret2 < 0) | |
537 | ret = ret2; | |
755200a3 | 538 | break; |
36167d82 JA |
539 | default: |
540 | assert(ret < 0); | |
541 | put_io_u(td, io_u); | |
542 | break; | |
ebac4655 JA |
543 | } |
544 | ||
99784632 | 545 | if (ret < 0 || td->error) |
36167d82 JA |
546 | break; |
547 | ||
97601024 JA |
548 | /* |
549 | * See if we need to complete some commands | |
550 | */ | |
838bc709 | 551 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
97601024 | 552 | min_evts = 0; |
838bc709 JA |
553 | if (td->cur_depth >= td->o.iodepth_low) |
554 | min_evts = td->cur_depth - td->o.iodepth_low; | |
555 | if (!min_evts) | |
36167d82 | 556 | min_evts = 1; |
97601024 | 557 | fio_gettime(&comp_time, NULL); |
d7762cf8 | 558 | bytes_done = io_u_queued_complete(td, min_evts); |
97601024 | 559 | if (bytes_done < 0) |
36167d82 | 560 | break; |
ebac4655 JA |
561 | } |
562 | ||
97601024 JA |
563 | if (!bytes_done) |
564 | continue; | |
565 | ||
ebac4655 JA |
566 | /* |
567 | * the rate is batched for now, it should work for batches | |
568 | * of completions except the very first one which may look | |
569 | * a little bursty | |
570 | */ | |
97601024 | 571 | usec = utime_since(&s, &comp_time); |
ebac4655 | 572 | |
413dd459 | 573 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 574 | |
97601024 | 575 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 576 | if (exitall_on_terminate) |
390c40e2 | 577 | terminate_threads(td->groupid); |
e1161c32 | 578 | td_verror(td, ENODATA, "check_min_rate"); |
ebac4655 JA |
579 | break; |
580 | } | |
581 | ||
2dc1bbeb | 582 | if (td->o.thinktime) { |
9c1f7434 JA |
583 | unsigned long long b; |
584 | ||
585 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 586 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
587 | int left; |
588 | ||
2dc1bbeb JA |
589 | if (td->o.thinktime_spin) |
590 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 591 | |
2dc1bbeb | 592 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
593 | if (left) |
594 | usec_sleep(td, left); | |
595 | } | |
9c1f7434 | 596 | } |
ebac4655 JA |
597 | } |
598 | ||
aa31f1f1 SL |
599 | if (td->o.fill_device && td->error == ENOSPC) { |
600 | td->error = 0; | |
601 | td->terminate = 1; | |
602 | } | |
4d2413c6 | 603 | if (!td->error) { |
3d7c391d JA |
604 | struct fio_file *f; |
605 | ||
c01c0395 JA |
606 | i = td->cur_depth; |
607 | if (i) | |
608 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 609 | |
2dc1bbeb | 610 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 611 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
612 | |
613 | for_each_file(td, f, i) { | |
614 | if (!(f->flags & FIO_FILE_OPEN)) | |
615 | continue; | |
858a3d47 | 616 | fio_io_sync(td, f); |
3d7b485f | 617 | } |
84585003 | 618 | } |
c01c0395 JA |
619 | } else |
620 | cleanup_pending_aio(td); | |
2ba1c290 JA |
621 | |
622 | /* | |
623 | * stop job if we failed doing any IO | |
624 | */ | |
625 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
626 | td->done = 1; | |
ebac4655 JA |
627 | } |
628 | ||
ebac4655 JA |
629 | static void cleanup_io_u(struct thread_data *td) |
630 | { | |
631 | struct list_head *entry, *n; | |
632 | struct io_u *io_u; | |
633 | ||
634 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
635 | io_u = list_entry(entry, struct io_u, list); | |
636 | ||
637 | list_del(&io_u->list); | |
638 | free(io_u); | |
639 | } | |
640 | ||
2f9ade3c | 641 | free_io_mem(td); |
ebac4655 JA |
642 | } |
643 | ||
644 | static int init_io_u(struct thread_data *td) | |
645 | { | |
646 | struct io_u *io_u; | |
a00735e6 | 647 | unsigned int max_bs; |
ebac4655 JA |
648 | int i, max_units; |
649 | char *p; | |
650 | ||
1e3c09af | 651 | max_units = td->o.iodepth; |
2dc1bbeb | 652 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
653 | td->orig_buffer_size = (unsigned long long) max_bs |
654 | * (unsigned long long) max_units; | |
655 | ||
656 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
657 | unsigned long bs; | |
74b025b0 | 658 | |
5ec10eaa JA |
659 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
660 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
661 | } | |
ebac4655 | 662 | |
c3990645 JA |
663 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
664 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
665 | return 1; | |
666 | } | |
667 | ||
2f9ade3c JA |
668 | if (allocate_io_mem(td)) |
669 | return 1; | |
ebac4655 | 670 | |
b3f378e9 JA |
671 | if (td->o.odirect) |
672 | p = ALIGN(td->orig_buffer); | |
673 | else | |
674 | p = td->orig_buffer; | |
675 | ||
ebac4655 | 676 | for (i = 0; i < max_units; i++) { |
db1cd90b JA |
677 | if (td->terminate) |
678 | return 1; | |
ebac4655 JA |
679 | io_u = malloc(sizeof(*io_u)); |
680 | memset(io_u, 0, sizeof(*io_u)); | |
681 | INIT_LIST_HEAD(&io_u->list); | |
682 | ||
b4c5e1ac JA |
683 | if (!(td->io_ops->flags & FIO_NOIO)) { |
684 | io_u->buf = p + max_bs * i; | |
e9459e5a | 685 | |
5973cafb JA |
686 | if (td_write(td) && !td->o.refill_buffers) |
687 | io_u_fill_buffer(td, io_u, max_bs); | |
b4c5e1ac | 688 | } |
6b9cea23 | 689 | |
b1ff3403 | 690 | io_u->index = i; |
0c6e7517 | 691 | io_u->flags = IO_U_F_FREE; |
ebac4655 JA |
692 | list_add(&io_u->list, &td->io_u_freelist); |
693 | } | |
694 | ||
433afcb4 JA |
695 | io_u_init_timeout(); |
696 | ||
ebac4655 JA |
697 | return 0; |
698 | } | |
699 | ||
da86774e JA |
700 | static int switch_ioscheduler(struct thread_data *td) |
701 | { | |
702 | char tmp[256], tmp2[128]; | |
703 | FILE *f; | |
704 | int ret; | |
705 | ||
ba0fbe10 | 706 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
707 | return 0; |
708 | ||
da86774e JA |
709 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
710 | ||
711 | f = fopen(tmp, "r+"); | |
712 | if (!f) { | |
cbf5c1d7 | 713 | if (errno == ENOENT) { |
5ec10eaa JA |
714 | log_err("fio: os or kernel doesn't support IO scheduler" |
715 | " switching\n"); | |
cbf5c1d7 JA |
716 | return 0; |
717 | } | |
718 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
719 | return 1; |
720 | } | |
721 | ||
722 | /* | |
723 | * Set io scheduler. | |
724 | */ | |
2dc1bbeb | 725 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 726 | if (ferror(f) || ret != 1) { |
e1161c32 | 727 | td_verror(td, errno, "fwrite"); |
da86774e JA |
728 | fclose(f); |
729 | return 1; | |
730 | } | |
731 | ||
732 | rewind(f); | |
733 | ||
734 | /* | |
735 | * Read back and check that the selected scheduler is now the default. | |
736 | */ | |
737 | ret = fread(tmp, 1, sizeof(tmp), f); | |
738 | if (ferror(f) || ret < 0) { | |
e1161c32 | 739 | td_verror(td, errno, "fread"); |
da86774e JA |
740 | fclose(f); |
741 | return 1; | |
742 | } | |
743 | ||
2dc1bbeb | 744 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 745 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 746 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 747 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
748 | fclose(f); |
749 | return 1; | |
750 | } | |
751 | ||
752 | fclose(f); | |
753 | return 0; | |
754 | } | |
755 | ||
492158cf JA |
756 | static int keep_running(struct thread_data *td) |
757 | { | |
758 | unsigned long long io_done; | |
759 | ||
20e354ef JA |
760 | if (td->done) |
761 | return 0; | |
492158cf JA |
762 | if (td->o.time_based) |
763 | return 1; | |
764 | if (td->o.loops) { | |
765 | td->o.loops--; | |
766 | return 1; | |
767 | } | |
768 | ||
5ec10eaa JA |
769 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
770 | + td->io_skip_bytes; | |
492158cf JA |
771 | if (io_done < td->o.size) |
772 | return 1; | |
773 | ||
774 | return 0; | |
775 | } | |
776 | ||
a978ba68 | 777 | static int clear_io_state(struct thread_data *td) |
ebac4655 | 778 | { |
53cdc686 | 779 | struct fio_file *f; |
af52b345 JA |
780 | unsigned int i; |
781 | int ret; | |
ebac4655 | 782 | |
756867bd | 783 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 784 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 785 | td->zone_bytes = 0; |
d7d3b49b | 786 | td->rate_bytes = 0; |
4e991c23 | 787 | td->rate_blocks = 0; |
38d77cae | 788 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 789 | |
c1324df1 JA |
790 | td->last_was_sync = 0; |
791 | ||
2ba1c290 JA |
792 | /* |
793 | * reset file done count if we are to start over | |
794 | */ | |
795 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 796 | td->nr_done_files = 0; |
281f9b63 | 797 | |
24ffd2c2 | 798 | close_files(td); |
53cdc686 | 799 | |
a978ba68 JA |
800 | ret = 0; |
801 | for_each_file(td, f, i) { | |
281f9b63 | 802 | f->flags &= ~FIO_FILE_DONE; |
a978ba68 JA |
803 | ret = td_io_open_file(td, f); |
804 | if (ret) | |
805 | break; | |
53cdc686 | 806 | } |
a978ba68 JA |
807 | |
808 | return ret; | |
ebac4655 JA |
809 | } |
810 | ||
906c8d75 JA |
811 | /* |
812 | * Entry point for the thread based jobs. The process based jobs end up | |
813 | * here as well, after a little setup. | |
814 | */ | |
ebac4655 JA |
815 | static void *thread_main(void *data) |
816 | { | |
10927316 | 817 | unsigned long long runtime[2], elapsed; |
ebac4655 | 818 | struct thread_data *td = data; |
a978ba68 | 819 | int clear_state; |
ebac4655 | 820 | |
2dc1bbeb | 821 | if (!td->o.use_thread) |
ebac4655 JA |
822 | setsid(); |
823 | ||
824 | td->pid = getpid(); | |
825 | ||
5921e80c | 826 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 827 | |
aea47d44 JA |
828 | INIT_LIST_HEAD(&td->io_u_freelist); |
829 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 830 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 | 831 | INIT_LIST_HEAD(&td->io_log_list); |
8de8f047 | 832 | INIT_LIST_HEAD(&td->io_hist_list); |
bb5d7d0b | 833 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 834 | |
db1cd90b | 835 | td_set_runstate(td, TD_INITIALIZED); |
cdd18ad8 JA |
836 | fio_mutex_up(startup_mutex); |
837 | fio_mutex_down(td->mutex); | |
db1cd90b JA |
838 | |
839 | /* | |
cdd18ad8 | 840 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
841 | * eating a file descriptor |
842 | */ | |
cdd18ad8 | 843 | fio_mutex_remove(td->mutex); |
db1cd90b | 844 | |
d84f8d49 JA |
845 | /* |
846 | * May alter parameters that init_io_u() will use, so we need to | |
847 | * do this first. | |
848 | */ | |
849 | if (init_iolog(td)) | |
850 | goto err; | |
851 | ||
ebac4655 | 852 | if (init_io_u(td)) |
db1cd90b | 853 | goto err; |
ebac4655 | 854 | |
375b2695 | 855 | if (td->o.cpumask_set && fio_setaffinity(td) == -1) { |
e1161c32 | 856 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 857 | goto err; |
ebac4655 JA |
858 | } |
859 | ||
ac684785 | 860 | if (td->ioprio_set) { |
ebac4655 | 861 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 862 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 863 | goto err; |
ebac4655 JA |
864 | } |
865 | } | |
866 | ||
2dc1bbeb | 867 | if (nice(td->o.nice) == -1) { |
e1161c32 | 868 | td_verror(td, errno, "nice"); |
db1cd90b | 869 | goto err; |
b6f4d880 JA |
870 | } |
871 | ||
2dc1bbeb | 872 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 873 | goto err; |
37f56873 | 874 | |
2dc1bbeb | 875 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
876 | goto err; |
877 | ||
7d6c5283 JA |
878 | if (td_io_init(td)) |
879 | goto err; | |
880 | ||
b5af8293 JA |
881 | if (open_files(td)) |
882 | goto err; | |
883 | ||
68727076 JA |
884 | if (init_random_map(td)) |
885 | goto err; | |
886 | ||
2dc1bbeb JA |
887 | if (td->o.exec_prerun) { |
888 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
889 | goto err; |
890 | } | |
4e0ba8af | 891 | |
69008999 | 892 | fio_gettime(&td->epoch, NULL); |
433afcb4 | 893 | memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); |
756867bd | 894 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
895 | |
896 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 897 | clear_state = 0; |
492158cf | 898 | while (keep_running(td)) { |
02bcaa8c | 899 | fio_gettime(&td->start, NULL); |
756867bd | 900 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 | 901 | |
2dc1bbeb | 902 | if (td->o.ratemin) |
5ec10eaa JA |
903 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
904 | sizeof(td->lastrate)); | |
ebac4655 | 905 | |
a978ba68 JA |
906 | if (clear_state && clear_io_state(td)) |
907 | break; | |
908 | ||
ebac4655 JA |
909 | prune_io_piece_log(td); |
910 | ||
ba0fbe10 | 911 | do_io(td); |
ebac4655 | 912 | |
a978ba68 JA |
913 | clear_state = 1; |
914 | ||
38d77cae JA |
915 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
916 | if (td->rw_end_set[DDIR_READ]) | |
5ec10eaa JA |
917 | elapsed = utime_since(&td->start, |
918 | &td->rw_end[DDIR_READ]); | |
38d77cae JA |
919 | else |
920 | elapsed = utime_since_now(&td->start); | |
921 | ||
922 | runtime[DDIR_READ] += elapsed; | |
923 | } | |
924 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
925 | if (td->rw_end_set[DDIR_WRITE]) | |
5ec10eaa JA |
926 | elapsed = utime_since(&td->start, |
927 | &td->rw_end[DDIR_WRITE]); | |
38d77cae JA |
928 | else |
929 | elapsed = utime_since_now(&td->start); | |
930 | ||
931 | runtime[DDIR_WRITE] += elapsed; | |
932 | } | |
5ec10eaa | 933 | |
ebac4655 JA |
934 | if (td->error || td->terminate) |
935 | break; | |
936 | ||
e84c73a8 SL |
937 | if (!td->o.do_verify || |
938 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 939 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
940 | continue; |
941 | ||
a978ba68 JA |
942 | if (clear_io_state(td)) |
943 | break; | |
944 | ||
02bcaa8c | 945 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
946 | |
947 | do_verify(td); | |
948 | ||
69008999 | 949 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
950 | |
951 | if (td->error || td->terminate) | |
952 | break; | |
953 | } | |
954 | ||
36dff966 | 955 | update_rusage_stat(td); |
47f0cc48 YHJT |
956 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
957 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
958 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
959 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
960 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
961 | ||
962 | if (td->ts.bw_log) | |
963 | finish_log(td, td->ts.bw_log, "bw"); | |
964 | if (td->ts.slat_log) | |
965 | finish_log(td, td->ts.slat_log, "slat"); | |
966 | if (td->ts.clat_log) | |
967 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb JA |
968 | if (td->o.exec_postrun) { |
969 | if (system(td->o.exec_postrun) < 0) | |
970 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 971 | } |
ebac4655 JA |
972 | |
973 | if (exitall_on_terminate) | |
390c40e2 | 974 | terminate_threads(td->groupid); |
ebac4655 JA |
975 | |
976 | err: | |
5bf13a5a | 977 | if (td->error) |
5921e80c | 978 | printf("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 979 | td->verror); |
24ffd2c2 | 980 | close_and_free_files(td); |
2866c82d | 981 | close_ioengine(td); |
ebac4655 | 982 | cleanup_io_u(td); |
f29b25a3 JA |
983 | |
984 | /* | |
985 | * do this very late, it will log file closing as well | |
986 | */ | |
987 | if (td->o.write_iolog_file) | |
988 | write_iolog_close(td); | |
989 | ||
d23bb327 | 990 | options_mem_free(td); |
ebac4655 | 991 | td_set_runstate(td, TD_EXITED); |
43d76807 | 992 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
993 | } |
994 | ||
906c8d75 JA |
995 | /* |
996 | * We cannot pass the td data into a forked process, so attach the td and | |
997 | * pass it to the thread worker. | |
998 | */ | |
a6418147 | 999 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1000 | { |
1001 | struct thread_data *td; | |
a6418147 | 1002 | void *data, *ret; |
ebac4655 JA |
1003 | |
1004 | data = shmat(shmid, NULL, 0); | |
1005 | if (data == (void *) -1) { | |
a6418147 JA |
1006 | int __err = errno; |
1007 | ||
ebac4655 | 1008 | perror("shmat"); |
a6418147 | 1009 | return __err; |
ebac4655 JA |
1010 | } |
1011 | ||
1012 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1013 | ret = thread_main(td); |
ebac4655 | 1014 | shmdt(data); |
43d76807 | 1015 | return (int) (unsigned long) ret; |
ebac4655 JA |
1016 | } |
1017 | ||
906c8d75 JA |
1018 | /* |
1019 | * Run over the job map and reap the threads that have exited, if any. | |
1020 | */ | |
ebac4655 JA |
1021 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1022 | { | |
34572e28 | 1023 | struct thread_data *td; |
1f809d15 | 1024 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1025 | |
1026 | /* | |
1027 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1028 | */ | |
1f809d15 | 1029 | realthreads = pending = cputhreads = 0; |
34572e28 | 1030 | for_each_td(td, i) { |
3707f45b | 1031 | int flags = 0; |
a2f77c9f | 1032 | |
84585003 JA |
1033 | /* |
1034 | * ->io_ops is NULL for a thread that has closed its | |
1035 | * io engine | |
1036 | */ | |
ba0fbe10 | 1037 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1038 | cputhreads++; |
1f809d15 JA |
1039 | else |
1040 | realthreads++; | |
b990b5c0 | 1041 | |
a1dedc1f AC |
1042 | if (!td->pid) { |
1043 | pending++; | |
1044 | continue; | |
1045 | } | |
1046 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1047 | continue; |
2dc1bbeb | 1048 | if (td->o.use_thread) { |
3707f45b JA |
1049 | if (td->runstate == TD_EXITED) { |
1050 | td_set_runstate(td, TD_REAPED); | |
1051 | goto reaped; | |
1052 | } | |
1053 | continue; | |
1054 | } | |
a2f77c9f JA |
1055 | |
1056 | flags = WNOHANG; | |
1057 | if (td->runstate == TD_EXITED) | |
1058 | flags = 0; | |
1059 | ||
1060 | /* | |
1061 | * check if someone quit or got killed in an unusual way | |
1062 | */ | |
1063 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1064 | if (ret < 0) { |
a2f77c9f | 1065 | if (errno == ECHILD) { |
5921e80c JA |
1066 | log_err("fio: pid=%d disappeared %d\n", |
1067 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1068 | td_set_runstate(td, TD_REAPED); |
1069 | goto reaped; | |
1070 | } | |
1071 | perror("waitpid"); | |
1072 | } else if (ret == td->pid) { | |
1073 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1074 | int sig = WTERMSIG(status); |
1075 | ||
d9244941 | 1076 | if (sig != SIGQUIT) |
5ec10eaa | 1077 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1078 | (int) td->pid, sig); |
fab6aa71 JA |
1079 | td_set_runstate(td, TD_REAPED); |
1080 | goto reaped; | |
1081 | } | |
a2f77c9f JA |
1082 | if (WIFEXITED(status)) { |
1083 | if (WEXITSTATUS(status) && !td->error) | |
1084 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1085 | |
a2f77c9f JA |
1086 | td_set_runstate(td, TD_REAPED); |
1087 | goto reaped; | |
a6418147 JA |
1088 | } |
1089 | } | |
ebac4655 | 1090 | |
a2f77c9f JA |
1091 | /* |
1092 | * thread is not dead, continue | |
1093 | */ | |
e48676ba | 1094 | pending++; |
a2f77c9f | 1095 | continue; |
fab6aa71 | 1096 | reaped: |
ebac4655 | 1097 | (*nr_running)--; |
2dc1bbeb JA |
1098 | (*m_rate) -= td->o.ratemin; |
1099 | (*t_rate) -= td->o.rate; | |
6f88bcb0 JA |
1100 | if (!td->pid) |
1101 | pending--; | |
a2f77c9f JA |
1102 | |
1103 | if (td->error) | |
1104 | exit_value++; | |
d3eeeabc JA |
1105 | |
1106 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1107 | } |
b990b5c0 | 1108 | |
1f809d15 | 1109 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1110 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1111 | } |
1112 | ||
906c8d75 JA |
1113 | /* |
1114 | * Main function for kicking off and reaping jobs, as needed. | |
1115 | */ | |
ebac4655 JA |
1116 | static void run_threads(void) |
1117 | { | |
ebac4655 JA |
1118 | struct thread_data *td; |
1119 | unsigned long spent; | |
1120 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1121 | |
2f9ade3c JA |
1122 | if (fio_pin_memory()) |
1123 | return; | |
ebac4655 | 1124 | |
c6ae0a5b | 1125 | if (!terse_output) { |
9cedf167 JA |
1126 | printf("Starting "); |
1127 | if (nr_thread) | |
5ec10eaa JA |
1128 | printf("%d thread%s", nr_thread, |
1129 | nr_thread > 1 ? "s" : ""); | |
9cedf167 JA |
1130 | if (nr_process) { |
1131 | if (nr_thread) | |
1132 | printf(" and "); | |
5ec10eaa JA |
1133 | printf("%d process%s", nr_process, |
1134 | nr_process > 1 ? "es" : ""); | |
9cedf167 JA |
1135 | } |
1136 | printf("\n"); | |
c6ae0a5b JA |
1137 | fflush(stdout); |
1138 | } | |
c04f7ec3 | 1139 | |
4efa970e JA |
1140 | signal(SIGINT, sig_handler); |
1141 | signal(SIGALRM, sig_handler); | |
1142 | ||
ebac4655 JA |
1143 | todo = thread_number; |
1144 | nr_running = 0; | |
1145 | nr_started = 0; | |
1146 | m_rate = t_rate = 0; | |
1147 | ||
34572e28 | 1148 | for_each_td(td, i) { |
263e529f | 1149 | print_status_init(td->thread_number - 1); |
ebac4655 | 1150 | |
2dc1bbeb | 1151 | if (!td->o.create_serialize) { |
380cf265 | 1152 | init_disk_util(td); |
ebac4655 | 1153 | continue; |
380cf265 | 1154 | } |
ebac4655 JA |
1155 | |
1156 | /* | |
1157 | * do file setup here so it happens sequentially, | |
1158 | * we don't want X number of threads getting their | |
1159 | * client data interspersed on disk | |
1160 | */ | |
53cdc686 | 1161 | if (setup_files(td)) { |
5bf13a5a JA |
1162 | exit_value++; |
1163 | if (td->error) | |
5921e80c JA |
1164 | log_err("fio: pid=%d, err=%d/%s\n", |
1165 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1166 | td_set_runstate(td, TD_REAPED); |
1167 | todo--; | |
c69e8875 JA |
1168 | } else { |
1169 | struct fio_file *f; | |
1170 | unsigned int i; | |
1171 | ||
1172 | /* | |
1173 | * for sharing to work, each job must always open | |
1174 | * its own files. so close them, if we opened them | |
1175 | * for creation | |
1176 | */ | |
1177 | for_each_file(td, f, i) | |
1178 | td_io_close_file(td, f); | |
5ec10eaa | 1179 | } |
380cf265 JA |
1180 | |
1181 | init_disk_util(td); | |
ebac4655 JA |
1182 | } |
1183 | ||
a2f77c9f JA |
1184 | set_genesis_time(); |
1185 | ||
ebac4655 | 1186 | while (todo) { |
75154845 JA |
1187 | struct thread_data *map[MAX_JOBS]; |
1188 | struct timeval this_start; | |
1189 | int this_jobs = 0, left; | |
1190 | ||
ebac4655 JA |
1191 | /* |
1192 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1193 | */ | |
34572e28 | 1194 | for_each_td(td, i) { |
ebac4655 JA |
1195 | if (td->runstate != TD_NOT_CREATED) |
1196 | continue; | |
1197 | ||
1198 | /* | |
1199 | * never got a chance to start, killed by other | |
1200 | * thread for some reason | |
1201 | */ | |
1202 | if (td->terminate) { | |
1203 | todo--; | |
1204 | continue; | |
1205 | } | |
1206 | ||
2dc1bbeb | 1207 | if (td->o.start_delay) { |
263e529f | 1208 | spent = mtime_since_genesis(); |
ebac4655 | 1209 | |
2dc1bbeb | 1210 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1211 | continue; |
1212 | } | |
1213 | ||
6f88bcb0 JA |
1214 | if (td->o.stonewall && (nr_started || nr_running)) { |
1215 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1216 | td->o.name); | |
ebac4655 | 1217 | break; |
6f88bcb0 | 1218 | } |
ebac4655 | 1219 | |
75154845 JA |
1220 | /* |
1221 | * Set state to created. Thread will transition | |
1222 | * to TD_INITIALIZED when it's done setting up. | |
1223 | */ | |
ebac4655 | 1224 | td_set_runstate(td, TD_CREATED); |
75154845 | 1225 | map[this_jobs++] = td; |
ebac4655 JA |
1226 | nr_started++; |
1227 | ||
2dc1bbeb | 1228 | if (td->o.use_thread) { |
ee56ad50 | 1229 | dprint(FD_PROCESS, "will pthread_create\n"); |
5ec10eaa JA |
1230 | if (pthread_create(&td->thread, NULL, |
1231 | thread_main, td)) { | |
f4705a73 | 1232 | perror("pthread_create"); |
ebac4655 | 1233 | nr_started--; |
c8bb6faf | 1234 | break; |
ebac4655 | 1235 | } |
f4705a73 JA |
1236 | if (pthread_detach(td->thread) < 0) |
1237 | perror("pthread_detach"); | |
ebac4655 | 1238 | } else { |
5e1d306e | 1239 | pid_t pid; |
ee56ad50 | 1240 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1241 | pid = fork(); |
1242 | if (!pid) { | |
a6418147 JA |
1243 | int ret = fork_main(shm_id, i); |
1244 | ||
5e1d306e JA |
1245 | _exit(ret); |
1246 | } else if (i == fio_debug_jobno) | |
1247 | *fio_debug_jobp = pid; | |
ebac4655 | 1248 | } |
cdd18ad8 | 1249 | fio_mutex_down(startup_mutex); |
ebac4655 JA |
1250 | } |
1251 | ||
1252 | /* | |
75154845 JA |
1253 | * Wait for the started threads to transition to |
1254 | * TD_INITIALIZED. | |
ebac4655 | 1255 | */ |
02bcaa8c | 1256 | fio_gettime(&this_start, NULL); |
75154845 | 1257 | left = this_jobs; |
6ce15a32 | 1258 | while (left && !fio_abort) { |
75154845 JA |
1259 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1260 | break; | |
1261 | ||
1262 | usleep(100000); | |
1263 | ||
1264 | for (i = 0; i < this_jobs; i++) { | |
1265 | td = map[i]; | |
1266 | if (!td) | |
1267 | continue; | |
b6f4d880 | 1268 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1269 | map[i] = NULL; |
1270 | left--; | |
b6f4d880 JA |
1271 | } else if (td->runstate >= TD_EXITED) { |
1272 | map[i] = NULL; | |
1273 | left--; | |
1274 | todo--; | |
1275 | nr_running++; /* work-around... */ | |
75154845 JA |
1276 | } |
1277 | } | |
1278 | } | |
1279 | ||
1280 | if (left) { | |
3b70d7e5 | 1281 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1282 | for (i = 0; i < this_jobs; i++) { |
1283 | td = map[i]; | |
1284 | if (!td) | |
1285 | continue; | |
1286 | kill(td->pid, SIGTERM); | |
1287 | } | |
1288 | break; | |
1289 | } | |
1290 | ||
1291 | /* | |
b6f4d880 | 1292 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1293 | */ |
34572e28 | 1294 | for_each_td(td, i) { |
75154845 | 1295 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1296 | continue; |
1297 | ||
1298 | td_set_runstate(td, TD_RUNNING); | |
1299 | nr_running++; | |
1300 | nr_started--; | |
2dc1bbeb JA |
1301 | m_rate += td->o.ratemin; |
1302 | t_rate += td->o.rate; | |
75154845 | 1303 | todo--; |
cdd18ad8 | 1304 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1305 | } |
1306 | ||
1307 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1308 | ||
1309 | if (todo) | |
1310 | usleep(100000); | |
1311 | } | |
1312 | ||
1313 | while (nr_running) { | |
1314 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1315 | usleep(10000); | |
1316 | } | |
1317 | ||
1318 | update_io_ticks(); | |
2f9ade3c | 1319 | fio_unpin_memory(); |
ebac4655 JA |
1320 | } |
1321 | ||
ebac4655 JA |
1322 | int main(int argc, char *argv[]) |
1323 | { | |
29d610e1 JA |
1324 | long ps; |
1325 | ||
2e5cdb11 JA |
1326 | sinit(); |
1327 | ||
dbe1125e JA |
1328 | /* |
1329 | * We need locale for number printing, if it isn't set then just | |
1330 | * go with the US format. | |
1331 | */ | |
1332 | if (!getenv("LC_NUMERIC")) | |
1333 | setlocale(LC_NUMERIC, "en_US"); | |
1334 | ||
ebac4655 JA |
1335 | if (parse_options(argc, argv)) |
1336 | return 1; | |
1337 | ||
4b472fa3 JA |
1338 | if (!thread_number) |
1339 | return 0; | |
ebac4655 | 1340 | |
29d610e1 JA |
1341 | ps = sysconf(_SC_PAGESIZE); |
1342 | if (ps < 0) { | |
1343 | log_err("Failed to get page size\n"); | |
1344 | return 1; | |
1345 | } | |
1346 | ||
cfc99db7 | 1347 | page_size = ps; |
29d610e1 JA |
1348 | page_mask = ps - 1; |
1349 | ||
bb3884d8 JA |
1350 | if (write_bw_log) { |
1351 | setup_log(&agg_io_log[DDIR_READ]); | |
1352 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1353 | } | |
1354 | ||
cdd18ad8 | 1355 | startup_mutex = fio_mutex_init(0); |
07739b57 | 1356 | |
a2f77c9f JA |
1357 | set_genesis_time(); |
1358 | ||
ebac4655 JA |
1359 | disk_util_timer_arm(); |
1360 | ||
1361 | run_threads(); | |
6ce15a32 | 1362 | |
bb3884d8 | 1363 | if (!fio_abort) { |
6ce15a32 | 1364 | show_run_stats(); |
bb3884d8 | 1365 | if (write_bw_log) { |
5ec10eaa JA |
1366 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1367 | __finish_log(agg_io_log[DDIR_WRITE], | |
1368 | "agg-write_bw.log"); | |
bb3884d8 JA |
1369 | } |
1370 | } | |
ebac4655 | 1371 | |
cdd18ad8 | 1372 | fio_mutex_remove(startup_mutex); |
437c9b71 | 1373 | return exit_value; |
ebac4655 | 1374 | } |