Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
aae22ca7 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
ebac4655 | 6 | * |
8e9fe637 JA |
7 | * The license below covers all files distributed with fio unless otherwise |
8 | * noted in the file itself. | |
9 | * | |
ebac4655 | 10 | * This program is free software; you can redistribute it and/or modify |
8e9fe637 JA |
11 | * it under the terms of the GNU General Public License version 2 as |
12 | * published by the Free Software Foundation. | |
ebac4655 JA |
13 | * |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
21 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
22 | * | |
23 | */ | |
ebac4655 JA |
24 | #include <unistd.h> |
25 | #include <fcntl.h> | |
26 | #include <string.h> | |
ebac4655 JA |
27 | #include <signal.h> |
28 | #include <time.h> | |
dbe1125e | 29 | #include <locale.h> |
36167d82 | 30 | #include <assert.h> |
ebac4655 JA |
31 | #include <sys/stat.h> |
32 | #include <sys/wait.h> | |
33 | #include <sys/ipc.h> | |
34 | #include <sys/shm.h> | |
ebac4655 JA |
35 | #include <sys/mman.h> |
36 | ||
37 | #include "fio.h" | |
210f43b3 | 38 | #include "hash.h" |
2e5cdb11 | 39 | #include "smalloc.h" |
ebac4655 | 40 | |
cfc99db7 JA |
41 | unsigned long page_mask; |
42 | unsigned long page_size; | |
29d610e1 JA |
43 | #define ALIGN(buf) \ |
44 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
45 | |
46 | int groupid = 0; | |
47 | int thread_number = 0; | |
9cedf167 JA |
48 | int nr_process = 0; |
49 | int nr_thread = 0; | |
ebac4655 | 50 | int shm_id = 0; |
53cdc686 | 51 | int temp_stall_ts; |
d3eeeabc | 52 | unsigned long done_secs = 0; |
ebac4655 | 53 | |
cdd18ad8 | 54 | static struct fio_mutex *startup_mutex; |
6ce15a32 | 55 | static volatile int fio_abort; |
437c9b71 | 56 | static int exit_value; |
ccb0fa24 | 57 | static struct itimerval itimer; |
ebac4655 | 58 | |
bb3884d8 JA |
59 | struct io_log *agg_io_log[2]; |
60 | ||
ebac4655 | 61 | #define TERMINATE_ALL (-1) |
75154845 | 62 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 63 | |
6ce15a32 JA |
64 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
65 | { | |
d8fab1b6 JA |
66 | if (td->runstate == runstate) |
67 | return; | |
68 | ||
5921e80c JA |
69 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
70 | td->runstate, runstate); | |
6ce15a32 JA |
71 | td->runstate = runstate; |
72 | } | |
73 | ||
390c40e2 | 74 | static void terminate_threads(int group_id) |
ebac4655 | 75 | { |
34572e28 | 76 | struct thread_data *td; |
ebac4655 JA |
77 | int i; |
78 | ||
6f88bcb0 JA |
79 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
80 | ||
34572e28 | 81 | for_each_td(td, i) { |
ebac4655 | 82 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 83 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 84 | td->o.name, (int) td->pid); |
ad830ed7 JA |
85 | td->terminate = 1; |
86 | td->o.start_delay = 0; | |
87 | ||
7a93ab19 JA |
88 | /* |
89 | * if the thread is running, just let it exit | |
90 | */ | |
91 | if (td->runstate < TD_RUNNING) | |
c1302d44 | 92 | kill(td->pid, SIGQUIT); |
f63f7f3b JA |
93 | else { |
94 | struct ioengine_ops *ops = td->io_ops; | |
95 | ||
96 | if (ops && (ops->flags & FIO_SIGQUIT)) | |
97 | kill(td->pid, SIGQUIT); | |
98 | } | |
ebac4655 JA |
99 | } |
100 | } | |
101 | } | |
102 | ||
ccb0fa24 JA |
103 | static void status_timer_arm(void) |
104 | { | |
105 | itimer.it_value.tv_sec = 0; | |
106 | itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; | |
107 | setitimer(ITIMER_REAL, &itimer, NULL); | |
108 | } | |
109 | ||
697a606c | 110 | static void sig_alrm(int sig) |
ebac4655 | 111 | { |
697a606c | 112 | if (threads) { |
5ec10eaa | 113 | update_io_ticks(); |
5ec10eaa | 114 | print_thread_status(); |
ccb0fa24 | 115 | status_timer_arm(); |
697a606c JA |
116 | } |
117 | } | |
118 | ||
119 | static void sig_int(int sig) | |
120 | { | |
121 | if (threads) { | |
5ec10eaa JA |
122 | printf("\nfio: terminating on signal %d\n", sig); |
123 | fflush(stdout); | |
124 | terminate_threads(TERMINATE_ALL); | |
ebac4655 JA |
125 | } |
126 | } | |
127 | ||
697a606c JA |
128 | /* |
129 | * We need to rearm on BSD/solaris. Switch this to sigaction in the future... | |
130 | */ | |
131 | static void set_sig_handlers(void) | |
132 | { | |
133 | struct sigaction act; | |
134 | ||
135 | memset(&act, 0, sizeof(act)); | |
136 | act.sa_handler = sig_alrm; | |
137 | act.sa_flags = SA_RESTART; | |
138 | sigaction(SIGALRM, &act, NULL); | |
139 | ||
140 | memset(&act, 0, sizeof(act)); | |
141 | act.sa_handler = sig_int; | |
142 | act.sa_flags = SA_RESTART; | |
143 | sigaction(SIGINT, &act, NULL); | |
144 | } | |
145 | ||
906c8d75 JA |
146 | /* |
147 | * Check if we are above the minimum rate given. | |
148 | */ | |
ebac4655 JA |
149 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
150 | { | |
0904200b | 151 | unsigned long long bytes = 0; |
4e991c23 | 152 | unsigned long iops = 0; |
ebac4655 JA |
153 | unsigned long spent; |
154 | unsigned long rate; | |
ebac4655 | 155 | |
780bf1a1 JA |
156 | /* |
157 | * No minimum rate set, always ok | |
158 | */ | |
2dc1bbeb | 159 | if (!td->o.ratemin && !td->o.rate_iops_min) |
780bf1a1 JA |
160 | return 0; |
161 | ||
ebac4655 JA |
162 | /* |
163 | * allow a 2 second settle period in the beginning | |
164 | */ | |
165 | if (mtime_since(&td->start, now) < 2000) | |
166 | return 0; | |
167 | ||
4e991c23 JA |
168 | if (td_read(td)) { |
169 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 170 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
171 | } |
172 | if (td_write(td)) { | |
173 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 174 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 175 | } |
0904200b | 176 | |
ebac4655 JA |
177 | /* |
178 | * if rate blocks is set, sample is running | |
179 | */ | |
4e991c23 | 180 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 181 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 182 | if (spent < td->o.ratecycle) |
ebac4655 JA |
183 | return 0; |
184 | ||
2dc1bbeb | 185 | if (td->o.rate) { |
4e991c23 JA |
186 | /* |
187 | * check bandwidth specified rate | |
188 | */ | |
189 | if (bytes < td->rate_bytes) { | |
5ec10eaa JA |
190 | log_err("%s: min rate %u not met\n", td->o.name, |
191 | td->o.ratemin); | |
4e991c23 JA |
192 | return 1; |
193 | } else { | |
194 | rate = (bytes - td->rate_bytes) / spent; | |
5ec10eaa JA |
195 | if (rate < td->o.ratemin || |
196 | bytes < td->rate_bytes) { | |
197 | log_err("%s: min rate %u not met, got" | |
198 | " %luKiB/sec\n", td->o.name, | |
199 | td->o.ratemin, rate); | |
4e991c23 JA |
200 | return 1; |
201 | } | |
202 | } | |
413dd459 | 203 | } else { |
4e991c23 JA |
204 | /* |
205 | * checks iops specified rate | |
206 | */ | |
2dc1bbeb | 207 | if (iops < td->o.rate_iops) { |
5ec10eaa JA |
208 | log_err("%s: min iops rate %u not met\n", |
209 | td->o.name, td->o.rate_iops); | |
413dd459 | 210 | return 1; |
4e991c23 JA |
211 | } else { |
212 | rate = (iops - td->rate_blocks) / spent; | |
5ec10eaa JA |
213 | if (rate < td->o.rate_iops_min || |
214 | iops < td->rate_blocks) { | |
215 | log_err("%s: min iops rate %u not met," | |
216 | " got %lu\n", td->o.name, | |
217 | td->o.rate_iops_min, | |
218 | rate); | |
4e991c23 | 219 | } |
413dd459 | 220 | } |
ebac4655 JA |
221 | } |
222 | } | |
223 | ||
0904200b | 224 | td->rate_bytes = bytes; |
4e991c23 | 225 | td->rate_blocks = iops; |
ebac4655 JA |
226 | memcpy(&td->lastrate, now, sizeof(*now)); |
227 | return 0; | |
228 | } | |
229 | ||
230 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
231 | { | |
2dc1bbeb | 232 | if (!td->o.timeout) |
ebac4655 | 233 | return 0; |
2dc1bbeb | 234 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
235 | return 1; |
236 | ||
237 | return 0; | |
238 | } | |
239 | ||
906c8d75 JA |
240 | /* |
241 | * When job exits, we can cancel the in-flight IO if we are using async | |
242 | * io. Attempt to do so. | |
243 | */ | |
ebac4655 JA |
244 | static void cleanup_pending_aio(struct thread_data *td) |
245 | { | |
01743ee1 | 246 | struct flist_head *entry, *n; |
ebac4655 JA |
247 | struct io_u *io_u; |
248 | int r; | |
249 | ||
250 | /* | |
251 | * get immediately available events, if any | |
252 | */ | |
d7762cf8 | 253 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
254 | if (r < 0) |
255 | return; | |
ebac4655 JA |
256 | |
257 | /* | |
258 | * now cancel remaining active events | |
259 | */ | |
2866c82d | 260 | if (td->io_ops->cancel) { |
01743ee1 JA |
261 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
262 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 263 | |
0c6e7517 JA |
264 | /* |
265 | * if the io_u isn't in flight, then that generally | |
266 | * means someone leaked an io_u. complain but fix | |
267 | * it up, so we don't stall here. | |
268 | */ | |
269 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
270 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 271 | put_io_u(td, io_u); |
0c6e7517 JA |
272 | } else { |
273 | r = td->io_ops->cancel(td, io_u); | |
274 | if (!r) | |
275 | put_io_u(td, io_u); | |
276 | } | |
ebac4655 JA |
277 | } |
278 | } | |
279 | ||
97601024 | 280 | if (td->cur_depth) |
d7762cf8 | 281 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
282 | } |
283 | ||
858a3d47 JA |
284 | /* |
285 | * Helper to handle the final sync of a file. Works just like the normal | |
286 | * io path, just does everything sync. | |
287 | */ | |
288 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
289 | { | |
290 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
291 | int ret; |
292 | ||
293 | if (!io_u) | |
294 | return 1; | |
295 | ||
296 | io_u->ddir = DDIR_SYNC; | |
297 | io_u->file = f; | |
298 | ||
299 | if (td_io_prep(td, io_u)) { | |
300 | put_io_u(td, io_u); | |
301 | return 1; | |
302 | } | |
303 | ||
755200a3 | 304 | requeue: |
858a3d47 | 305 | ret = td_io_queue(td, io_u); |
36167d82 | 306 | if (ret < 0) { |
e1161c32 | 307 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 308 | put_io_u(td, io_u); |
858a3d47 | 309 | return 1; |
36167d82 | 310 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 311 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 312 | return 1; |
36167d82 JA |
313 | } else if (ret == FIO_Q_COMPLETED) { |
314 | if (io_u->error) { | |
e1161c32 | 315 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
316 | return 1; |
317 | } | |
858a3d47 | 318 | |
d7762cf8 | 319 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 320 | return 1; |
755200a3 JA |
321 | } else if (ret == FIO_Q_BUSY) { |
322 | if (td_io_commit(td)) | |
323 | return 1; | |
324 | goto requeue; | |
858a3d47 JA |
325 | } |
326 | ||
327 | return 0; | |
328 | } | |
329 | ||
906c8d75 | 330 | /* |
34403fb1 | 331 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
332 | * reads the blocks back in, and checks the crc/md5 of the data. |
333 | */ | |
1e97cce9 | 334 | static void do_verify(struct thread_data *td) |
ebac4655 | 335 | { |
53cdc686 | 336 | struct fio_file *f; |
36167d82 | 337 | struct io_u *io_u; |
af52b345 JA |
338 | int ret, min_events; |
339 | unsigned int i; | |
e5b401d4 JA |
340 | |
341 | /* | |
342 | * sync io first and invalidate cache, to make sure we really | |
343 | * read from disk. | |
344 | */ | |
345 | for_each_file(td, f, i) { | |
3d7b485f JA |
346 | if (!(f->flags & FIO_FILE_OPEN)) |
347 | continue; | |
b2fdda43 JA |
348 | if (fio_io_sync(td, f)) |
349 | break; | |
350 | if (file_invalidate_cache(td, f)) | |
351 | break; | |
e5b401d4 | 352 | } |
ebac4655 | 353 | |
b2fdda43 JA |
354 | if (td->error) |
355 | return; | |
356 | ||
ebac4655 JA |
357 | td_set_runstate(td, TD_VERIFYING); |
358 | ||
36167d82 JA |
359 | io_u = NULL; |
360 | while (!td->terminate) { | |
5451792e JA |
361 | int ret2; |
362 | ||
ebac4655 JA |
363 | io_u = __get_io_u(td); |
364 | if (!io_u) | |
365 | break; | |
366 | ||
069c2918 JA |
367 | if (runtime_exceeded(td, &io_u->start_time)) { |
368 | put_io_u(td, io_u); | |
90a87d4b | 369 | td->terminate = 1; |
02bcaa8c | 370 | break; |
069c2918 | 371 | } |
02bcaa8c | 372 | |
069c2918 JA |
373 | if (get_next_verify(td, io_u)) { |
374 | put_io_u(td, io_u); | |
ebac4655 | 375 | break; |
069c2918 | 376 | } |
ebac4655 | 377 | |
069c2918 JA |
378 | if (td_io_prep(td, io_u)) { |
379 | put_io_u(td, io_u); | |
53cdc686 | 380 | break; |
069c2918 | 381 | } |
d7762cf8 JA |
382 | |
383 | io_u->end_io = verify_io_u; | |
53cdc686 | 384 | |
11786802 | 385 | ret = td_io_queue(td, io_u); |
36167d82 JA |
386 | switch (ret) { |
387 | case FIO_Q_COMPLETED: | |
388 | if (io_u->error) | |
22819ec2 | 389 | ret = -io_u->error; |
9e9d164e | 390 | else if (io_u->resid) { |
36167d82 | 391 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 392 | struct fio_file *f = io_u->file; |
ebac4655 | 393 | |
9e9d164e JA |
394 | /* |
395 | * zero read, fail | |
396 | */ | |
397 | if (!bytes) { | |
41d9f0fd | 398 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
399 | put_io_u(td, io_u); |
400 | break; | |
401 | } | |
8400d9b2 | 402 | |
36167d82 JA |
403 | io_u->xfer_buflen = io_u->resid; |
404 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
405 | io_u->offset += bytes; |
406 | ||
30061b97 JA |
407 | td->ts.short_io_u[io_u->ddir]++; |
408 | ||
d460eb31 | 409 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
410 | goto sync_done; |
411 | ||
11786802 JA |
412 | requeue_io_u(td, &io_u); |
413 | } else { | |
8400d9b2 | 414 | sync_done: |
11786802 JA |
415 | ret = io_u_sync_complete(td, io_u); |
416 | if (ret < 0) | |
417 | break; | |
36167d82 | 418 | } |
36167d82 JA |
419 | continue; |
420 | case FIO_Q_QUEUED: | |
421 | break; | |
755200a3 JA |
422 | case FIO_Q_BUSY: |
423 | requeue_io_u(td, &io_u); | |
5451792e JA |
424 | ret2 = td_io_commit(td); |
425 | if (ret2 < 0) | |
426 | ret = ret2; | |
755200a3 | 427 | break; |
36167d82 JA |
428 | default: |
429 | assert(ret < 0); | |
e1161c32 | 430 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
431 | break; |
432 | } | |
433 | ||
99784632 | 434 | if (ret < 0 || td->error) |
3af6ef39 JA |
435 | break; |
436 | ||
ebac4655 | 437 | /* |
3af6ef39 JA |
438 | * if we can queue more, do so. but check if there are |
439 | * completed io_u's first. | |
ebac4655 | 440 | */ |
97601024 | 441 | min_events = 0; |
e916b390 | 442 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
838bc709 | 443 | if (td->cur_depth >= td->o.iodepth_low) |
2dc1bbeb | 444 | min_events = td->cur_depth - td->o.iodepth_low; |
838bc709 JA |
445 | if (!min_events) |
446 | min_events = 1; | |
e916b390 JA |
447 | } |
448 | ||
3af6ef39 JA |
449 | /* |
450 | * Reap required number of io units, if any, and do the | |
451 | * verification on them through the callback handler | |
452 | */ | |
d7762cf8 | 453 | if (io_u_queued_complete(td, min_events) < 0) |
ebac4655 | 454 | break; |
36167d82 | 455 | } |
ebac4655 | 456 | |
c01c0395 JA |
457 | if (!td->error) { |
458 | min_events = td->cur_depth; | |
459 | ||
460 | if (min_events) | |
461 | ret = io_u_queued_complete(td, min_events); | |
462 | } else | |
ebac4655 JA |
463 | cleanup_pending_aio(td); |
464 | ||
465 | td_set_runstate(td, TD_RUNNING); | |
466 | } | |
467 | ||
32cd46a0 | 468 | /* |
906c8d75 | 469 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
470 | * and reaps them, checking for rate and errors along the way. |
471 | */ | |
ebac4655 JA |
472 | static void do_io(struct thread_data *td) |
473 | { | |
02bcaa8c | 474 | struct timeval s; |
ebac4655 | 475 | unsigned long usec; |
af52b345 JA |
476 | unsigned int i; |
477 | int ret = 0; | |
ebac4655 | 478 | |
5853e5a8 JA |
479 | td_set_runstate(td, TD_RUNNING); |
480 | ||
7bb48f84 | 481 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
482 | struct timeval comp_time; |
483 | long bytes_done = 0; | |
84585003 | 484 | int min_evts = 0; |
ebac4655 | 485 | struct io_u *io_u; |
5451792e | 486 | int ret2; |
ebac4655 JA |
487 | |
488 | if (td->terminate) | |
489 | break; | |
490 | ||
3d7c391d | 491 | io_u = get_io_u(td); |
ebac4655 JA |
492 | if (!io_u) |
493 | break; | |
494 | ||
495 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
496 | |
497 | if (runtime_exceeded(td, &s)) { | |
498 | put_io_u(td, io_u); | |
90a87d4b | 499 | td->terminate = 1; |
97601024 JA |
500 | break; |
501 | } | |
36167d82 | 502 | |
b67740d3 JA |
503 | /* |
504 | * Add verification end_io handler, if asked to verify | |
505 | * a previously written file. | |
506 | */ | |
cc3fc855 | 507 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) { |
b67740d3 | 508 | io_u->end_io = verify_io_u; |
d8fab1b6 JA |
509 | td_set_runstate(td, TD_VERIFYING); |
510 | } else | |
511 | td_set_runstate(td, TD_RUNNING); | |
b67740d3 | 512 | |
11786802 | 513 | ret = td_io_queue(td, io_u); |
36167d82 JA |
514 | switch (ret) { |
515 | case FIO_Q_COMPLETED: | |
5451792e JA |
516 | if (io_u->error) |
517 | ret = -io_u->error; | |
9e9d164e | 518 | else if (io_u->resid) { |
36167d82 | 519 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 520 | struct fio_file *f = io_u->file; |
36167d82 | 521 | |
9e9d164e JA |
522 | /* |
523 | * zero read, fail | |
524 | */ | |
525 | if (!bytes) { | |
41d9f0fd | 526 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
527 | put_io_u(td, io_u); |
528 | break; | |
529 | } | |
530 | ||
cec6b55d | 531 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 532 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
533 | io_u->offset += bytes; |
534 | ||
30061b97 JA |
535 | td->ts.short_io_u[io_u->ddir]++; |
536 | ||
d460eb31 | 537 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
538 | goto sync_done; |
539 | ||
11786802 JA |
540 | requeue_io_u(td, &io_u); |
541 | } else { | |
5a7c5680 | 542 | sync_done: |
11786802 JA |
543 | fio_gettime(&comp_time, NULL); |
544 | bytes_done = io_u_sync_complete(td, io_u); | |
545 | if (bytes_done < 0) | |
546 | ret = bytes_done; | |
cec6b55d | 547 | } |
36167d82 JA |
548 | break; |
549 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
550 | /* |
551 | * if the engine doesn't have a commit hook, | |
552 | * the io_u is really queued. if it does have such | |
553 | * a hook, it has to call io_u_queued() itself. | |
554 | */ | |
555 | if (td->io_ops->commit == NULL) | |
556 | io_u_queued(td, io_u); | |
36167d82 | 557 | break; |
755200a3 JA |
558 | case FIO_Q_BUSY: |
559 | requeue_io_u(td, &io_u); | |
5451792e JA |
560 | ret2 = td_io_commit(td); |
561 | if (ret2 < 0) | |
562 | ret = ret2; | |
755200a3 | 563 | break; |
36167d82 JA |
564 | default: |
565 | assert(ret < 0); | |
566 | put_io_u(td, io_u); | |
567 | break; | |
ebac4655 JA |
568 | } |
569 | ||
99784632 | 570 | if (ret < 0 || td->error) |
36167d82 JA |
571 | break; |
572 | ||
97601024 JA |
573 | /* |
574 | * See if we need to complete some commands | |
575 | */ | |
838bc709 | 576 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
97601024 | 577 | min_evts = 0; |
838bc709 JA |
578 | if (td->cur_depth >= td->o.iodepth_low) |
579 | min_evts = td->cur_depth - td->o.iodepth_low; | |
580 | if (!min_evts) | |
36167d82 | 581 | min_evts = 1; |
97601024 | 582 | fio_gettime(&comp_time, NULL); |
d7762cf8 | 583 | bytes_done = io_u_queued_complete(td, min_evts); |
97601024 | 584 | if (bytes_done < 0) |
36167d82 | 585 | break; |
ebac4655 JA |
586 | } |
587 | ||
97601024 JA |
588 | if (!bytes_done) |
589 | continue; | |
590 | ||
ebac4655 JA |
591 | /* |
592 | * the rate is batched for now, it should work for batches | |
593 | * of completions except the very first one which may look | |
594 | * a little bursty | |
595 | */ | |
97601024 | 596 | usec = utime_since(&s, &comp_time); |
ebac4655 | 597 | |
413dd459 | 598 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 599 | |
97601024 | 600 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 601 | if (exitall_on_terminate) |
390c40e2 | 602 | terminate_threads(td->groupid); |
41d9f0fd | 603 | td_verror(td, EIO, "check_min_rate"); |
ebac4655 JA |
604 | break; |
605 | } | |
606 | ||
2dc1bbeb | 607 | if (td->o.thinktime) { |
9c1f7434 JA |
608 | unsigned long long b; |
609 | ||
610 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 611 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
612 | int left; |
613 | ||
2dc1bbeb JA |
614 | if (td->o.thinktime_spin) |
615 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 616 | |
2dc1bbeb | 617 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
618 | if (left) |
619 | usec_sleep(td, left); | |
620 | } | |
9c1f7434 | 621 | } |
ebac4655 JA |
622 | } |
623 | ||
aa31f1f1 SL |
624 | if (td->o.fill_device && td->error == ENOSPC) { |
625 | td->error = 0; | |
626 | td->terminate = 1; | |
627 | } | |
4d2413c6 | 628 | if (!td->error) { |
3d7c391d JA |
629 | struct fio_file *f; |
630 | ||
c01c0395 JA |
631 | i = td->cur_depth; |
632 | if (i) | |
633 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 634 | |
2dc1bbeb | 635 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 636 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
637 | |
638 | for_each_file(td, f, i) { | |
639 | if (!(f->flags & FIO_FILE_OPEN)) | |
640 | continue; | |
858a3d47 | 641 | fio_io_sync(td, f); |
3d7b485f | 642 | } |
84585003 | 643 | } |
c01c0395 JA |
644 | } else |
645 | cleanup_pending_aio(td); | |
2ba1c290 JA |
646 | |
647 | /* | |
648 | * stop job if we failed doing any IO | |
649 | */ | |
650 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
651 | td->done = 1; | |
ebac4655 JA |
652 | } |
653 | ||
ebac4655 JA |
654 | static void cleanup_io_u(struct thread_data *td) |
655 | { | |
01743ee1 | 656 | struct flist_head *entry, *n; |
ebac4655 JA |
657 | struct io_u *io_u; |
658 | ||
01743ee1 JA |
659 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
660 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 661 | |
01743ee1 | 662 | flist_del(&io_u->list); |
ebac4655 JA |
663 | free(io_u); |
664 | } | |
665 | ||
2f9ade3c | 666 | free_io_mem(td); |
ebac4655 JA |
667 | } |
668 | ||
669 | static int init_io_u(struct thread_data *td) | |
670 | { | |
671 | struct io_u *io_u; | |
a00735e6 | 672 | unsigned int max_bs; |
ebac4655 JA |
673 | int i, max_units; |
674 | char *p; | |
675 | ||
1e3c09af | 676 | max_units = td->o.iodepth; |
2dc1bbeb | 677 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
678 | td->orig_buffer_size = (unsigned long long) max_bs |
679 | * (unsigned long long) max_units; | |
680 | ||
681 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
682 | unsigned long bs; | |
74b025b0 | 683 | |
5ec10eaa JA |
684 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
685 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
686 | } | |
ebac4655 | 687 | |
c3990645 JA |
688 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
689 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
690 | return 1; | |
691 | } | |
692 | ||
2f9ade3c JA |
693 | if (allocate_io_mem(td)) |
694 | return 1; | |
ebac4655 | 695 | |
b3f378e9 JA |
696 | if (td->o.odirect) |
697 | p = ALIGN(td->orig_buffer); | |
698 | else | |
699 | p = td->orig_buffer; | |
700 | ||
ebac4655 | 701 | for (i = 0; i < max_units; i++) { |
db1cd90b JA |
702 | if (td->terminate) |
703 | return 1; | |
ebac4655 JA |
704 | io_u = malloc(sizeof(*io_u)); |
705 | memset(io_u, 0, sizeof(*io_u)); | |
01743ee1 | 706 | INIT_FLIST_HEAD(&io_u->list); |
ebac4655 | 707 | |
b4c5e1ac JA |
708 | if (!(td->io_ops->flags & FIO_NOIO)) { |
709 | io_u->buf = p + max_bs * i; | |
e9459e5a | 710 | |
5973cafb JA |
711 | if (td_write(td) && !td->o.refill_buffers) |
712 | io_u_fill_buffer(td, io_u, max_bs); | |
b4c5e1ac | 713 | } |
6b9cea23 | 714 | |
b1ff3403 | 715 | io_u->index = i; |
0c6e7517 | 716 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 717 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
718 | } |
719 | ||
433afcb4 JA |
720 | io_u_init_timeout(); |
721 | ||
ebac4655 JA |
722 | return 0; |
723 | } | |
724 | ||
da86774e JA |
725 | static int switch_ioscheduler(struct thread_data *td) |
726 | { | |
727 | char tmp[256], tmp2[128]; | |
728 | FILE *f; | |
729 | int ret; | |
730 | ||
ba0fbe10 | 731 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
732 | return 0; |
733 | ||
da86774e JA |
734 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
735 | ||
736 | f = fopen(tmp, "r+"); | |
737 | if (!f) { | |
cbf5c1d7 | 738 | if (errno == ENOENT) { |
5ec10eaa JA |
739 | log_err("fio: os or kernel doesn't support IO scheduler" |
740 | " switching\n"); | |
cbf5c1d7 JA |
741 | return 0; |
742 | } | |
743 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
744 | return 1; |
745 | } | |
746 | ||
747 | /* | |
748 | * Set io scheduler. | |
749 | */ | |
2dc1bbeb | 750 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 751 | if (ferror(f) || ret != 1) { |
e1161c32 | 752 | td_verror(td, errno, "fwrite"); |
da86774e JA |
753 | fclose(f); |
754 | return 1; | |
755 | } | |
756 | ||
757 | rewind(f); | |
758 | ||
759 | /* | |
760 | * Read back and check that the selected scheduler is now the default. | |
761 | */ | |
762 | ret = fread(tmp, 1, sizeof(tmp), f); | |
763 | if (ferror(f) || ret < 0) { | |
e1161c32 | 764 | td_verror(td, errno, "fread"); |
da86774e JA |
765 | fclose(f); |
766 | return 1; | |
767 | } | |
768 | ||
2dc1bbeb | 769 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 770 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 771 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 772 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
773 | fclose(f); |
774 | return 1; | |
775 | } | |
776 | ||
777 | fclose(f); | |
778 | return 0; | |
779 | } | |
780 | ||
492158cf JA |
781 | static int keep_running(struct thread_data *td) |
782 | { | |
783 | unsigned long long io_done; | |
784 | ||
20e354ef JA |
785 | if (td->done) |
786 | return 0; | |
492158cf JA |
787 | if (td->o.time_based) |
788 | return 1; | |
789 | if (td->o.loops) { | |
790 | td->o.loops--; | |
791 | return 1; | |
792 | } | |
793 | ||
5ec10eaa JA |
794 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
795 | + td->io_skip_bytes; | |
492158cf JA |
796 | if (io_done < td->o.size) |
797 | return 1; | |
798 | ||
799 | return 0; | |
800 | } | |
801 | ||
a978ba68 | 802 | static int clear_io_state(struct thread_data *td) |
ebac4655 | 803 | { |
53cdc686 | 804 | struct fio_file *f; |
af52b345 JA |
805 | unsigned int i; |
806 | int ret; | |
ebac4655 | 807 | |
756867bd | 808 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 809 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 810 | td->zone_bytes = 0; |
d7d3b49b | 811 | td->rate_bytes = 0; |
4e991c23 | 812 | td->rate_blocks = 0; |
38d77cae | 813 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 814 | |
c1324df1 JA |
815 | td->last_was_sync = 0; |
816 | ||
2ba1c290 JA |
817 | /* |
818 | * reset file done count if we are to start over | |
819 | */ | |
820 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 821 | td->nr_done_files = 0; |
281f9b63 | 822 | |
24ffd2c2 | 823 | close_files(td); |
53cdc686 | 824 | |
a978ba68 JA |
825 | ret = 0; |
826 | for_each_file(td, f, i) { | |
281f9b63 | 827 | f->flags &= ~FIO_FILE_DONE; |
a978ba68 JA |
828 | ret = td_io_open_file(td, f); |
829 | if (ret) | |
830 | break; | |
53cdc686 | 831 | } |
a978ba68 JA |
832 | |
833 | return ret; | |
ebac4655 JA |
834 | } |
835 | ||
906c8d75 JA |
836 | /* |
837 | * Entry point for the thread based jobs. The process based jobs end up | |
838 | * here as well, after a little setup. | |
839 | */ | |
ebac4655 JA |
840 | static void *thread_main(void *data) |
841 | { | |
10927316 | 842 | unsigned long long runtime[2], elapsed; |
ebac4655 | 843 | struct thread_data *td = data; |
a978ba68 | 844 | int clear_state; |
ebac4655 | 845 | |
2dc1bbeb | 846 | if (!td->o.use_thread) |
ebac4655 JA |
847 | setsid(); |
848 | ||
849 | td->pid = getpid(); | |
850 | ||
5921e80c | 851 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 852 | |
01743ee1 JA |
853 | INIT_FLIST_HEAD(&td->io_u_freelist); |
854 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
855 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
856 | INIT_FLIST_HEAD(&td->io_log_list); | |
857 | INIT_FLIST_HEAD(&td->io_hist_list); | |
bb5d7d0b | 858 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 859 | |
db1cd90b | 860 | td_set_runstate(td, TD_INITIALIZED); |
cdd18ad8 JA |
861 | fio_mutex_up(startup_mutex); |
862 | fio_mutex_down(td->mutex); | |
db1cd90b JA |
863 | |
864 | /* | |
cdd18ad8 | 865 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
866 | * eating a file descriptor |
867 | */ | |
cdd18ad8 | 868 | fio_mutex_remove(td->mutex); |
db1cd90b | 869 | |
d84f8d49 JA |
870 | /* |
871 | * May alter parameters that init_io_u() will use, so we need to | |
872 | * do this first. | |
873 | */ | |
874 | if (init_iolog(td)) | |
875 | goto err; | |
876 | ||
ebac4655 | 877 | if (init_io_u(td)) |
db1cd90b | 878 | goto err; |
ebac4655 | 879 | |
375b2695 | 880 | if (td->o.cpumask_set && fio_setaffinity(td) == -1) { |
e1161c32 | 881 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 882 | goto err; |
ebac4655 JA |
883 | } |
884 | ||
ac684785 | 885 | if (td->ioprio_set) { |
ebac4655 | 886 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 887 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 888 | goto err; |
ebac4655 JA |
889 | } |
890 | } | |
891 | ||
2dc1bbeb | 892 | if (nice(td->o.nice) == -1) { |
e1161c32 | 893 | td_verror(td, errno, "nice"); |
db1cd90b | 894 | goto err; |
b6f4d880 JA |
895 | } |
896 | ||
2dc1bbeb | 897 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 898 | goto err; |
37f56873 | 899 | |
2dc1bbeb | 900 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
901 | goto err; |
902 | ||
7d6c5283 JA |
903 | if (td_io_init(td)) |
904 | goto err; | |
905 | ||
b5af8293 JA |
906 | if (open_files(td)) |
907 | goto err; | |
908 | ||
68727076 JA |
909 | if (init_random_map(td)) |
910 | goto err; | |
911 | ||
2dc1bbeb JA |
912 | if (td->o.exec_prerun) { |
913 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
914 | goto err; |
915 | } | |
4e0ba8af | 916 | |
69008999 | 917 | fio_gettime(&td->epoch, NULL); |
433afcb4 | 918 | memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); |
756867bd | 919 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
920 | |
921 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 922 | clear_state = 0; |
492158cf | 923 | while (keep_running(td)) { |
02bcaa8c | 924 | fio_gettime(&td->start, NULL); |
756867bd | 925 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 | 926 | |
2dc1bbeb | 927 | if (td->o.ratemin) |
5ec10eaa JA |
928 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
929 | sizeof(td->lastrate)); | |
ebac4655 | 930 | |
a978ba68 JA |
931 | if (clear_state && clear_io_state(td)) |
932 | break; | |
933 | ||
ebac4655 JA |
934 | prune_io_piece_log(td); |
935 | ||
ba0fbe10 | 936 | do_io(td); |
ebac4655 | 937 | |
a978ba68 JA |
938 | clear_state = 1; |
939 | ||
38d77cae JA |
940 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
941 | if (td->rw_end_set[DDIR_READ]) | |
5ec10eaa JA |
942 | elapsed = utime_since(&td->start, |
943 | &td->rw_end[DDIR_READ]); | |
38d77cae JA |
944 | else |
945 | elapsed = utime_since_now(&td->start); | |
946 | ||
947 | runtime[DDIR_READ] += elapsed; | |
948 | } | |
949 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
950 | if (td->rw_end_set[DDIR_WRITE]) | |
5ec10eaa JA |
951 | elapsed = utime_since(&td->start, |
952 | &td->rw_end[DDIR_WRITE]); | |
38d77cae JA |
953 | else |
954 | elapsed = utime_since_now(&td->start); | |
955 | ||
956 | runtime[DDIR_WRITE] += elapsed; | |
957 | } | |
5ec10eaa | 958 | |
ebac4655 JA |
959 | if (td->error || td->terminate) |
960 | break; | |
961 | ||
e84c73a8 SL |
962 | if (!td->o.do_verify || |
963 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 964 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
965 | continue; |
966 | ||
a978ba68 JA |
967 | if (clear_io_state(td)) |
968 | break; | |
969 | ||
02bcaa8c | 970 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
971 | |
972 | do_verify(td); | |
973 | ||
69008999 | 974 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
975 | |
976 | if (td->error || td->terminate) | |
977 | break; | |
978 | } | |
979 | ||
36dff966 | 980 | update_rusage_stat(td); |
47f0cc48 YHJT |
981 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
982 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
983 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
984 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
985 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
986 | ||
987 | if (td->ts.bw_log) | |
988 | finish_log(td, td->ts.bw_log, "bw"); | |
989 | if (td->ts.slat_log) | |
990 | finish_log(td, td->ts.slat_log, "slat"); | |
991 | if (td->ts.clat_log) | |
992 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb JA |
993 | if (td->o.exec_postrun) { |
994 | if (system(td->o.exec_postrun) < 0) | |
995 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 996 | } |
ebac4655 JA |
997 | |
998 | if (exitall_on_terminate) | |
390c40e2 | 999 | terminate_threads(td->groupid); |
ebac4655 JA |
1000 | |
1001 | err: | |
5bf13a5a | 1002 | if (td->error) |
5921e80c | 1003 | printf("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1004 | td->verror); |
24ffd2c2 | 1005 | close_and_free_files(td); |
2866c82d | 1006 | close_ioengine(td); |
ebac4655 | 1007 | cleanup_io_u(td); |
f29b25a3 JA |
1008 | |
1009 | /* | |
1010 | * do this very late, it will log file closing as well | |
1011 | */ | |
1012 | if (td->o.write_iolog_file) | |
1013 | write_iolog_close(td); | |
1014 | ||
d23bb327 | 1015 | options_mem_free(td); |
ebac4655 | 1016 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1017 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1018 | } |
1019 | ||
906c8d75 JA |
1020 | /* |
1021 | * We cannot pass the td data into a forked process, so attach the td and | |
1022 | * pass it to the thread worker. | |
1023 | */ | |
a6418147 | 1024 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1025 | { |
1026 | struct thread_data *td; | |
a6418147 | 1027 | void *data, *ret; |
ebac4655 JA |
1028 | |
1029 | data = shmat(shmid, NULL, 0); | |
1030 | if (data == (void *) -1) { | |
a6418147 JA |
1031 | int __err = errno; |
1032 | ||
ebac4655 | 1033 | perror("shmat"); |
a6418147 | 1034 | return __err; |
ebac4655 JA |
1035 | } |
1036 | ||
1037 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1038 | ret = thread_main(td); |
ebac4655 | 1039 | shmdt(data); |
43d76807 | 1040 | return (int) (unsigned long) ret; |
ebac4655 JA |
1041 | } |
1042 | ||
906c8d75 JA |
1043 | /* |
1044 | * Run over the job map and reap the threads that have exited, if any. | |
1045 | */ | |
ebac4655 JA |
1046 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1047 | { | |
34572e28 | 1048 | struct thread_data *td; |
1f809d15 | 1049 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1050 | |
1051 | /* | |
1052 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1053 | */ | |
1f809d15 | 1054 | realthreads = pending = cputhreads = 0; |
34572e28 | 1055 | for_each_td(td, i) { |
3707f45b | 1056 | int flags = 0; |
a2f77c9f | 1057 | |
84585003 JA |
1058 | /* |
1059 | * ->io_ops is NULL for a thread that has closed its | |
1060 | * io engine | |
1061 | */ | |
ba0fbe10 | 1062 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1063 | cputhreads++; |
1f809d15 JA |
1064 | else |
1065 | realthreads++; | |
b990b5c0 | 1066 | |
a1dedc1f AC |
1067 | if (!td->pid) { |
1068 | pending++; | |
1069 | continue; | |
1070 | } | |
1071 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1072 | continue; |
2dc1bbeb | 1073 | if (td->o.use_thread) { |
3707f45b JA |
1074 | if (td->runstate == TD_EXITED) { |
1075 | td_set_runstate(td, TD_REAPED); | |
1076 | goto reaped; | |
1077 | } | |
1078 | continue; | |
1079 | } | |
a2f77c9f JA |
1080 | |
1081 | flags = WNOHANG; | |
1082 | if (td->runstate == TD_EXITED) | |
1083 | flags = 0; | |
1084 | ||
1085 | /* | |
1086 | * check if someone quit or got killed in an unusual way | |
1087 | */ | |
1088 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1089 | if (ret < 0) { |
a2f77c9f | 1090 | if (errno == ECHILD) { |
5921e80c JA |
1091 | log_err("fio: pid=%d disappeared %d\n", |
1092 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1093 | td_set_runstate(td, TD_REAPED); |
1094 | goto reaped; | |
1095 | } | |
1096 | perror("waitpid"); | |
1097 | } else if (ret == td->pid) { | |
1098 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1099 | int sig = WTERMSIG(status); |
1100 | ||
d9244941 | 1101 | if (sig != SIGQUIT) |
5ec10eaa | 1102 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1103 | (int) td->pid, sig); |
fab6aa71 JA |
1104 | td_set_runstate(td, TD_REAPED); |
1105 | goto reaped; | |
1106 | } | |
a2f77c9f JA |
1107 | if (WIFEXITED(status)) { |
1108 | if (WEXITSTATUS(status) && !td->error) | |
1109 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1110 | |
a2f77c9f JA |
1111 | td_set_runstate(td, TD_REAPED); |
1112 | goto reaped; | |
a6418147 JA |
1113 | } |
1114 | } | |
ebac4655 | 1115 | |
a2f77c9f JA |
1116 | /* |
1117 | * thread is not dead, continue | |
1118 | */ | |
e48676ba | 1119 | pending++; |
a2f77c9f | 1120 | continue; |
fab6aa71 | 1121 | reaped: |
ebac4655 | 1122 | (*nr_running)--; |
2dc1bbeb JA |
1123 | (*m_rate) -= td->o.ratemin; |
1124 | (*t_rate) -= td->o.rate; | |
6f88bcb0 JA |
1125 | if (!td->pid) |
1126 | pending--; | |
a2f77c9f JA |
1127 | |
1128 | if (td->error) | |
1129 | exit_value++; | |
d3eeeabc JA |
1130 | |
1131 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1132 | } |
b990b5c0 | 1133 | |
1f809d15 | 1134 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1135 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1136 | } |
1137 | ||
906c8d75 JA |
1138 | /* |
1139 | * Main function for kicking off and reaping jobs, as needed. | |
1140 | */ | |
ebac4655 JA |
1141 | static void run_threads(void) |
1142 | { | |
ebac4655 JA |
1143 | struct thread_data *td; |
1144 | unsigned long spent; | |
1145 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1146 | |
2f9ade3c JA |
1147 | if (fio_pin_memory()) |
1148 | return; | |
ebac4655 | 1149 | |
c6ae0a5b | 1150 | if (!terse_output) { |
9cedf167 JA |
1151 | printf("Starting "); |
1152 | if (nr_thread) | |
5ec10eaa JA |
1153 | printf("%d thread%s", nr_thread, |
1154 | nr_thread > 1 ? "s" : ""); | |
9cedf167 JA |
1155 | if (nr_process) { |
1156 | if (nr_thread) | |
1157 | printf(" and "); | |
5ec10eaa JA |
1158 | printf("%d process%s", nr_process, |
1159 | nr_process > 1 ? "es" : ""); | |
9cedf167 JA |
1160 | } |
1161 | printf("\n"); | |
c6ae0a5b JA |
1162 | fflush(stdout); |
1163 | } | |
c04f7ec3 | 1164 | |
697a606c | 1165 | set_sig_handlers(); |
4efa970e | 1166 | |
ebac4655 JA |
1167 | todo = thread_number; |
1168 | nr_running = 0; | |
1169 | nr_started = 0; | |
1170 | m_rate = t_rate = 0; | |
1171 | ||
34572e28 | 1172 | for_each_td(td, i) { |
263e529f | 1173 | print_status_init(td->thread_number - 1); |
ebac4655 | 1174 | |
2dc1bbeb | 1175 | if (!td->o.create_serialize) { |
380cf265 | 1176 | init_disk_util(td); |
ebac4655 | 1177 | continue; |
380cf265 | 1178 | } |
ebac4655 JA |
1179 | |
1180 | /* | |
1181 | * do file setup here so it happens sequentially, | |
1182 | * we don't want X number of threads getting their | |
1183 | * client data interspersed on disk | |
1184 | */ | |
53cdc686 | 1185 | if (setup_files(td)) { |
5bf13a5a JA |
1186 | exit_value++; |
1187 | if (td->error) | |
5921e80c JA |
1188 | log_err("fio: pid=%d, err=%d/%s\n", |
1189 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1190 | td_set_runstate(td, TD_REAPED); |
1191 | todo--; | |
c69e8875 JA |
1192 | } else { |
1193 | struct fio_file *f; | |
1194 | unsigned int i; | |
1195 | ||
1196 | /* | |
1197 | * for sharing to work, each job must always open | |
1198 | * its own files. so close them, if we opened them | |
1199 | * for creation | |
1200 | */ | |
1201 | for_each_file(td, f, i) | |
1202 | td_io_close_file(td, f); | |
5ec10eaa | 1203 | } |
380cf265 JA |
1204 | |
1205 | init_disk_util(td); | |
ebac4655 JA |
1206 | } |
1207 | ||
a2f77c9f JA |
1208 | set_genesis_time(); |
1209 | ||
ebac4655 | 1210 | while (todo) { |
75154845 JA |
1211 | struct thread_data *map[MAX_JOBS]; |
1212 | struct timeval this_start; | |
1213 | int this_jobs = 0, left; | |
1214 | ||
ebac4655 JA |
1215 | /* |
1216 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1217 | */ | |
34572e28 | 1218 | for_each_td(td, i) { |
ebac4655 JA |
1219 | if (td->runstate != TD_NOT_CREATED) |
1220 | continue; | |
1221 | ||
1222 | /* | |
1223 | * never got a chance to start, killed by other | |
1224 | * thread for some reason | |
1225 | */ | |
1226 | if (td->terminate) { | |
1227 | todo--; | |
1228 | continue; | |
1229 | } | |
1230 | ||
2dc1bbeb | 1231 | if (td->o.start_delay) { |
263e529f | 1232 | spent = mtime_since_genesis(); |
ebac4655 | 1233 | |
2dc1bbeb | 1234 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1235 | continue; |
1236 | } | |
1237 | ||
6f88bcb0 JA |
1238 | if (td->o.stonewall && (nr_started || nr_running)) { |
1239 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1240 | td->o.name); | |
ebac4655 | 1241 | break; |
6f88bcb0 | 1242 | } |
ebac4655 | 1243 | |
75154845 JA |
1244 | /* |
1245 | * Set state to created. Thread will transition | |
1246 | * to TD_INITIALIZED when it's done setting up. | |
1247 | */ | |
ebac4655 | 1248 | td_set_runstate(td, TD_CREATED); |
75154845 | 1249 | map[this_jobs++] = td; |
ebac4655 JA |
1250 | nr_started++; |
1251 | ||
2dc1bbeb | 1252 | if (td->o.use_thread) { |
ee56ad50 | 1253 | dprint(FD_PROCESS, "will pthread_create\n"); |
5ec10eaa JA |
1254 | if (pthread_create(&td->thread, NULL, |
1255 | thread_main, td)) { | |
f4705a73 | 1256 | perror("pthread_create"); |
ebac4655 | 1257 | nr_started--; |
c8bb6faf | 1258 | break; |
ebac4655 | 1259 | } |
f4705a73 JA |
1260 | if (pthread_detach(td->thread) < 0) |
1261 | perror("pthread_detach"); | |
ebac4655 | 1262 | } else { |
5e1d306e | 1263 | pid_t pid; |
ee56ad50 | 1264 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1265 | pid = fork(); |
1266 | if (!pid) { | |
a6418147 JA |
1267 | int ret = fork_main(shm_id, i); |
1268 | ||
5e1d306e JA |
1269 | _exit(ret); |
1270 | } else if (i == fio_debug_jobno) | |
1271 | *fio_debug_jobp = pid; | |
ebac4655 | 1272 | } |
cdd18ad8 | 1273 | fio_mutex_down(startup_mutex); |
ebac4655 JA |
1274 | } |
1275 | ||
1276 | /* | |
75154845 JA |
1277 | * Wait for the started threads to transition to |
1278 | * TD_INITIALIZED. | |
ebac4655 | 1279 | */ |
02bcaa8c | 1280 | fio_gettime(&this_start, NULL); |
75154845 | 1281 | left = this_jobs; |
6ce15a32 | 1282 | while (left && !fio_abort) { |
75154845 JA |
1283 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1284 | break; | |
1285 | ||
1286 | usleep(100000); | |
1287 | ||
1288 | for (i = 0; i < this_jobs; i++) { | |
1289 | td = map[i]; | |
1290 | if (!td) | |
1291 | continue; | |
b6f4d880 | 1292 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1293 | map[i] = NULL; |
1294 | left--; | |
b6f4d880 JA |
1295 | } else if (td->runstate >= TD_EXITED) { |
1296 | map[i] = NULL; | |
1297 | left--; | |
1298 | todo--; | |
1299 | nr_running++; /* work-around... */ | |
75154845 JA |
1300 | } |
1301 | } | |
1302 | } | |
1303 | ||
1304 | if (left) { | |
3b70d7e5 | 1305 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1306 | for (i = 0; i < this_jobs; i++) { |
1307 | td = map[i]; | |
1308 | if (!td) | |
1309 | continue; | |
1310 | kill(td->pid, SIGTERM); | |
1311 | } | |
1312 | break; | |
1313 | } | |
1314 | ||
1315 | /* | |
b6f4d880 | 1316 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1317 | */ |
34572e28 | 1318 | for_each_td(td, i) { |
75154845 | 1319 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1320 | continue; |
1321 | ||
1322 | td_set_runstate(td, TD_RUNNING); | |
1323 | nr_running++; | |
1324 | nr_started--; | |
2dc1bbeb JA |
1325 | m_rate += td->o.ratemin; |
1326 | t_rate += td->o.rate; | |
75154845 | 1327 | todo--; |
cdd18ad8 | 1328 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1329 | } |
1330 | ||
1331 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1332 | ||
1333 | if (todo) | |
1334 | usleep(100000); | |
1335 | } | |
1336 | ||
1337 | while (nr_running) { | |
1338 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1339 | usleep(10000); | |
1340 | } | |
1341 | ||
1342 | update_io_ticks(); | |
2f9ade3c | 1343 | fio_unpin_memory(); |
ebac4655 JA |
1344 | } |
1345 | ||
ebac4655 JA |
1346 | int main(int argc, char *argv[]) |
1347 | { | |
29d610e1 JA |
1348 | long ps; |
1349 | ||
2e5cdb11 JA |
1350 | sinit(); |
1351 | ||
dbe1125e JA |
1352 | /* |
1353 | * We need locale for number printing, if it isn't set then just | |
1354 | * go with the US format. | |
1355 | */ | |
1356 | if (!getenv("LC_NUMERIC")) | |
1357 | setlocale(LC_NUMERIC, "en_US"); | |
1358 | ||
ebac4655 JA |
1359 | if (parse_options(argc, argv)) |
1360 | return 1; | |
1361 | ||
4b472fa3 JA |
1362 | if (!thread_number) |
1363 | return 0; | |
ebac4655 | 1364 | |
29d610e1 JA |
1365 | ps = sysconf(_SC_PAGESIZE); |
1366 | if (ps < 0) { | |
1367 | log_err("Failed to get page size\n"); | |
1368 | return 1; | |
1369 | } | |
1370 | ||
cfc99db7 | 1371 | page_size = ps; |
29d610e1 JA |
1372 | page_mask = ps - 1; |
1373 | ||
bb3884d8 JA |
1374 | if (write_bw_log) { |
1375 | setup_log(&agg_io_log[DDIR_READ]); | |
1376 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1377 | } | |
1378 | ||
cdd18ad8 | 1379 | startup_mutex = fio_mutex_init(0); |
07739b57 | 1380 | |
a2f77c9f JA |
1381 | set_genesis_time(); |
1382 | ||
ccb0fa24 | 1383 | status_timer_arm(); |
ebac4655 JA |
1384 | |
1385 | run_threads(); | |
6ce15a32 | 1386 | |
bb3884d8 | 1387 | if (!fio_abort) { |
6ce15a32 | 1388 | show_run_stats(); |
bb3884d8 | 1389 | if (write_bw_log) { |
5ec10eaa JA |
1390 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1391 | __finish_log(agg_io_log[DDIR_WRITE], | |
1392 | "agg-write_bw.log"); | |
bb3884d8 JA |
1393 | } |
1394 | } | |
ebac4655 | 1395 | |
cdd18ad8 | 1396 | fio_mutex_remove(startup_mutex); |
437c9b71 | 1397 | return exit_value; |
ebac4655 | 1398 | } |