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 | static void set_sig_handlers(void) |
129 | { | |
130 | struct sigaction act; | |
131 | ||
132 | memset(&act, 0, sizeof(act)); | |
133 | act.sa_handler = sig_alrm; | |
134 | act.sa_flags = SA_RESTART; | |
135 | sigaction(SIGALRM, &act, NULL); | |
136 | ||
137 | memset(&act, 0, sizeof(act)); | |
138 | act.sa_handler = sig_int; | |
139 | act.sa_flags = SA_RESTART; | |
140 | sigaction(SIGINT, &act, NULL); | |
141 | } | |
142 | ||
906c8d75 JA |
143 | /* |
144 | * Check if we are above the minimum rate given. | |
145 | */ | |
ebac4655 JA |
146 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
147 | { | |
0904200b | 148 | unsigned long long bytes = 0; |
4e991c23 | 149 | unsigned long iops = 0; |
ebac4655 JA |
150 | unsigned long spent; |
151 | unsigned long rate; | |
ebac4655 | 152 | |
780bf1a1 JA |
153 | /* |
154 | * No minimum rate set, always ok | |
155 | */ | |
2dc1bbeb | 156 | if (!td->o.ratemin && !td->o.rate_iops_min) |
780bf1a1 JA |
157 | return 0; |
158 | ||
ebac4655 JA |
159 | /* |
160 | * allow a 2 second settle period in the beginning | |
161 | */ | |
162 | if (mtime_since(&td->start, now) < 2000) | |
163 | return 0; | |
164 | ||
4e991c23 JA |
165 | if (td_read(td)) { |
166 | iops += td->io_blocks[DDIR_READ]; | |
0904200b | 167 | bytes += td->this_io_bytes[DDIR_READ]; |
4e991c23 JA |
168 | } |
169 | if (td_write(td)) { | |
170 | iops += td->io_blocks[DDIR_WRITE]; | |
0904200b | 171 | bytes += td->this_io_bytes[DDIR_WRITE]; |
4e991c23 | 172 | } |
0904200b | 173 | |
ebac4655 JA |
174 | /* |
175 | * if rate blocks is set, sample is running | |
176 | */ | |
4e991c23 | 177 | if (td->rate_bytes || td->rate_blocks) { |
ebac4655 | 178 | spent = mtime_since(&td->lastrate, now); |
2dc1bbeb | 179 | if (spent < td->o.ratecycle) |
ebac4655 JA |
180 | return 0; |
181 | ||
2dc1bbeb | 182 | if (td->o.rate) { |
4e991c23 JA |
183 | /* |
184 | * check bandwidth specified rate | |
185 | */ | |
186 | if (bytes < td->rate_bytes) { | |
5ec10eaa JA |
187 | log_err("%s: min rate %u not met\n", td->o.name, |
188 | td->o.ratemin); | |
4e991c23 JA |
189 | return 1; |
190 | } else { | |
191 | rate = (bytes - td->rate_bytes) / spent; | |
5ec10eaa JA |
192 | if (rate < td->o.ratemin || |
193 | bytes < td->rate_bytes) { | |
194 | log_err("%s: min rate %u not met, got" | |
195 | " %luKiB/sec\n", td->o.name, | |
196 | td->o.ratemin, rate); | |
4e991c23 JA |
197 | return 1; |
198 | } | |
199 | } | |
413dd459 | 200 | } else { |
4e991c23 JA |
201 | /* |
202 | * checks iops specified rate | |
203 | */ | |
2dc1bbeb | 204 | if (iops < td->o.rate_iops) { |
5ec10eaa JA |
205 | log_err("%s: min iops rate %u not met\n", |
206 | td->o.name, td->o.rate_iops); | |
413dd459 | 207 | return 1; |
4e991c23 JA |
208 | } else { |
209 | rate = (iops - td->rate_blocks) / spent; | |
5ec10eaa JA |
210 | if (rate < td->o.rate_iops_min || |
211 | iops < td->rate_blocks) { | |
212 | log_err("%s: min iops rate %u not met," | |
213 | " got %lu\n", td->o.name, | |
214 | td->o.rate_iops_min, | |
215 | rate); | |
4e991c23 | 216 | } |
413dd459 | 217 | } |
ebac4655 JA |
218 | } |
219 | } | |
220 | ||
0904200b | 221 | td->rate_bytes = bytes; |
4e991c23 | 222 | td->rate_blocks = iops; |
ebac4655 JA |
223 | memcpy(&td->lastrate, now, sizeof(*now)); |
224 | return 0; | |
225 | } | |
226 | ||
227 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
228 | { | |
2dc1bbeb | 229 | if (!td->o.timeout) |
ebac4655 | 230 | return 0; |
2dc1bbeb | 231 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
232 | return 1; |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
906c8d75 JA |
237 | /* |
238 | * When job exits, we can cancel the in-flight IO if we are using async | |
239 | * io. Attempt to do so. | |
240 | */ | |
ebac4655 JA |
241 | static void cleanup_pending_aio(struct thread_data *td) |
242 | { | |
01743ee1 | 243 | struct flist_head *entry, *n; |
ebac4655 JA |
244 | struct io_u *io_u; |
245 | int r; | |
246 | ||
247 | /* | |
248 | * get immediately available events, if any | |
249 | */ | |
d7762cf8 | 250 | r = io_u_queued_complete(td, 0); |
b2fdda43 JA |
251 | if (r < 0) |
252 | return; | |
ebac4655 JA |
253 | |
254 | /* | |
255 | * now cancel remaining active events | |
256 | */ | |
2866c82d | 257 | if (td->io_ops->cancel) { |
01743ee1 JA |
258 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
259 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 260 | |
0c6e7517 JA |
261 | /* |
262 | * if the io_u isn't in flight, then that generally | |
263 | * means someone leaked an io_u. complain but fix | |
264 | * it up, so we don't stall here. | |
265 | */ | |
266 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
267 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 268 | put_io_u(td, io_u); |
0c6e7517 JA |
269 | } else { |
270 | r = td->io_ops->cancel(td, io_u); | |
271 | if (!r) | |
272 | put_io_u(td, io_u); | |
273 | } | |
ebac4655 JA |
274 | } |
275 | } | |
276 | ||
97601024 | 277 | if (td->cur_depth) |
d7762cf8 | 278 | r = io_u_queued_complete(td, td->cur_depth); |
ebac4655 JA |
279 | } |
280 | ||
858a3d47 JA |
281 | /* |
282 | * Helper to handle the final sync of a file. Works just like the normal | |
283 | * io path, just does everything sync. | |
284 | */ | |
285 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
286 | { | |
287 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
288 | int ret; |
289 | ||
290 | if (!io_u) | |
291 | return 1; | |
292 | ||
293 | io_u->ddir = DDIR_SYNC; | |
294 | io_u->file = f; | |
295 | ||
296 | if (td_io_prep(td, io_u)) { | |
297 | put_io_u(td, io_u); | |
298 | return 1; | |
299 | } | |
300 | ||
755200a3 | 301 | requeue: |
858a3d47 | 302 | ret = td_io_queue(td, io_u); |
36167d82 | 303 | if (ret < 0) { |
e1161c32 | 304 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 305 | put_io_u(td, io_u); |
858a3d47 | 306 | return 1; |
36167d82 | 307 | } else if (ret == FIO_Q_QUEUED) { |
d7762cf8 | 308 | if (io_u_queued_complete(td, 1) < 0) |
36167d82 | 309 | return 1; |
36167d82 JA |
310 | } else if (ret == FIO_Q_COMPLETED) { |
311 | if (io_u->error) { | |
e1161c32 | 312 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
313 | return 1; |
314 | } | |
858a3d47 | 315 | |
d7762cf8 | 316 | if (io_u_sync_complete(td, io_u) < 0) |
b2fdda43 | 317 | return 1; |
755200a3 JA |
318 | } else if (ret == FIO_Q_BUSY) { |
319 | if (td_io_commit(td)) | |
320 | return 1; | |
321 | goto requeue; | |
858a3d47 JA |
322 | } |
323 | ||
324 | return 0; | |
325 | } | |
326 | ||
906c8d75 | 327 | /* |
34403fb1 | 328 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
329 | * reads the blocks back in, and checks the crc/md5 of the data. |
330 | */ | |
1e97cce9 | 331 | static void do_verify(struct thread_data *td) |
ebac4655 | 332 | { |
53cdc686 | 333 | struct fio_file *f; |
36167d82 | 334 | struct io_u *io_u; |
af52b345 JA |
335 | int ret, min_events; |
336 | unsigned int i; | |
e5b401d4 JA |
337 | |
338 | /* | |
339 | * sync io first and invalidate cache, to make sure we really | |
340 | * read from disk. | |
341 | */ | |
342 | for_each_file(td, f, i) { | |
3d7b485f JA |
343 | if (!(f->flags & FIO_FILE_OPEN)) |
344 | continue; | |
b2fdda43 JA |
345 | if (fio_io_sync(td, f)) |
346 | break; | |
347 | if (file_invalidate_cache(td, f)) | |
348 | break; | |
e5b401d4 | 349 | } |
ebac4655 | 350 | |
b2fdda43 JA |
351 | if (td->error) |
352 | return; | |
353 | ||
ebac4655 JA |
354 | td_set_runstate(td, TD_VERIFYING); |
355 | ||
36167d82 JA |
356 | io_u = NULL; |
357 | while (!td->terminate) { | |
4950421a | 358 | int ret2, full; |
5451792e | 359 | |
ebac4655 JA |
360 | io_u = __get_io_u(td); |
361 | if (!io_u) | |
362 | break; | |
363 | ||
069c2918 JA |
364 | if (runtime_exceeded(td, &io_u->start_time)) { |
365 | put_io_u(td, io_u); | |
90a87d4b | 366 | td->terminate = 1; |
02bcaa8c | 367 | break; |
069c2918 | 368 | } |
02bcaa8c | 369 | |
069c2918 JA |
370 | if (get_next_verify(td, io_u)) { |
371 | put_io_u(td, io_u); | |
ebac4655 | 372 | break; |
069c2918 | 373 | } |
ebac4655 | 374 | |
069c2918 JA |
375 | if (td_io_prep(td, io_u)) { |
376 | put_io_u(td, io_u); | |
53cdc686 | 377 | break; |
069c2918 | 378 | } |
d7762cf8 JA |
379 | |
380 | io_u->end_io = verify_io_u; | |
53cdc686 | 381 | |
11786802 | 382 | ret = td_io_queue(td, io_u); |
36167d82 JA |
383 | switch (ret) { |
384 | case FIO_Q_COMPLETED: | |
385 | if (io_u->error) | |
22819ec2 | 386 | ret = -io_u->error; |
9e9d164e | 387 | else if (io_u->resid) { |
36167d82 | 388 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 389 | struct fio_file *f = io_u->file; |
ebac4655 | 390 | |
9e9d164e JA |
391 | /* |
392 | * zero read, fail | |
393 | */ | |
394 | if (!bytes) { | |
41d9f0fd | 395 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
396 | put_io_u(td, io_u); |
397 | break; | |
398 | } | |
8400d9b2 | 399 | |
36167d82 JA |
400 | io_u->xfer_buflen = io_u->resid; |
401 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
402 | io_u->offset += bytes; |
403 | ||
30061b97 JA |
404 | td->ts.short_io_u[io_u->ddir]++; |
405 | ||
d460eb31 | 406 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
407 | goto sync_done; |
408 | ||
11786802 JA |
409 | requeue_io_u(td, &io_u); |
410 | } else { | |
8400d9b2 | 411 | sync_done: |
11786802 JA |
412 | ret = io_u_sync_complete(td, io_u); |
413 | if (ret < 0) | |
414 | break; | |
36167d82 | 415 | } |
36167d82 JA |
416 | continue; |
417 | case FIO_Q_QUEUED: | |
418 | break; | |
755200a3 JA |
419 | case FIO_Q_BUSY: |
420 | requeue_io_u(td, &io_u); | |
5451792e JA |
421 | ret2 = td_io_commit(td); |
422 | if (ret2 < 0) | |
423 | ret = ret2; | |
755200a3 | 424 | break; |
36167d82 JA |
425 | default: |
426 | assert(ret < 0); | |
e1161c32 | 427 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
428 | break; |
429 | } | |
430 | ||
99784632 | 431 | if (ret < 0 || td->error) |
3af6ef39 JA |
432 | break; |
433 | ||
ebac4655 | 434 | /* |
3af6ef39 JA |
435 | * if we can queue more, do so. but check if there are |
436 | * completed io_u's first. | |
ebac4655 | 437 | */ |
4950421a JA |
438 | full = queue_full(td) || ret == FIO_Q_BUSY; |
439 | if (full || !td->o.iodepth_batch_complete) { | |
440 | min_events = td->o.iodepth_batch_complete; | |
441 | if (full && !min_events) | |
838bc709 | 442 | min_events = 1; |
e916b390 | 443 | |
4950421a JA |
444 | do { |
445 | /* | |
446 | * Reap required number of io units, if any, | |
447 | * and do the verification on them through | |
448 | * the callback handler | |
449 | */ | |
450 | if (io_u_queued_complete(td, min_events) < 0) { | |
451 | ret = -1; | |
452 | break; | |
453 | } | |
454 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
455 | } | |
456 | if (ret < 0) | |
ebac4655 | 457 | break; |
36167d82 | 458 | } |
ebac4655 | 459 | |
c01c0395 JA |
460 | if (!td->error) { |
461 | min_events = td->cur_depth; | |
462 | ||
463 | if (min_events) | |
464 | ret = io_u_queued_complete(td, min_events); | |
465 | } else | |
ebac4655 JA |
466 | cleanup_pending_aio(td); |
467 | ||
468 | td_set_runstate(td, TD_RUNNING); | |
469 | } | |
470 | ||
32cd46a0 | 471 | /* |
906c8d75 | 472 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
473 | * and reaps them, checking for rate and errors along the way. |
474 | */ | |
ebac4655 JA |
475 | static void do_io(struct thread_data *td) |
476 | { | |
02bcaa8c | 477 | struct timeval s; |
ebac4655 | 478 | unsigned long usec; |
af52b345 JA |
479 | unsigned int i; |
480 | int ret = 0; | |
ebac4655 | 481 | |
5853e5a8 JA |
482 | td_set_runstate(td, TD_RUNNING); |
483 | ||
7bb48f84 | 484 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) { |
97601024 JA |
485 | struct timeval comp_time; |
486 | long bytes_done = 0; | |
84585003 | 487 | int min_evts = 0; |
ebac4655 | 488 | struct io_u *io_u; |
4950421a | 489 | int ret2, full; |
ebac4655 JA |
490 | |
491 | if (td->terminate) | |
492 | break; | |
493 | ||
3d7c391d | 494 | io_u = get_io_u(td); |
ebac4655 JA |
495 | if (!io_u) |
496 | break; | |
497 | ||
498 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
499 | |
500 | if (runtime_exceeded(td, &s)) { | |
501 | put_io_u(td, io_u); | |
90a87d4b | 502 | td->terminate = 1; |
97601024 JA |
503 | break; |
504 | } | |
36167d82 | 505 | |
b67740d3 JA |
506 | /* |
507 | * Add verification end_io handler, if asked to verify | |
508 | * a previously written file. | |
509 | */ | |
cc3fc855 | 510 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) { |
b67740d3 | 511 | io_u->end_io = verify_io_u; |
d8fab1b6 JA |
512 | td_set_runstate(td, TD_VERIFYING); |
513 | } else | |
514 | td_set_runstate(td, TD_RUNNING); | |
b67740d3 | 515 | |
11786802 | 516 | ret = td_io_queue(td, io_u); |
36167d82 JA |
517 | switch (ret) { |
518 | case FIO_Q_COMPLETED: | |
5451792e JA |
519 | if (io_u->error) |
520 | ret = -io_u->error; | |
9e9d164e | 521 | else if (io_u->resid) { |
36167d82 | 522 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 523 | struct fio_file *f = io_u->file; |
36167d82 | 524 | |
9e9d164e JA |
525 | /* |
526 | * zero read, fail | |
527 | */ | |
528 | if (!bytes) { | |
41d9f0fd | 529 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
530 | put_io_u(td, io_u); |
531 | break; | |
532 | } | |
533 | ||
cec6b55d | 534 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 535 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
536 | io_u->offset += bytes; |
537 | ||
30061b97 JA |
538 | td->ts.short_io_u[io_u->ddir]++; |
539 | ||
d460eb31 | 540 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
541 | goto sync_done; |
542 | ||
11786802 JA |
543 | requeue_io_u(td, &io_u); |
544 | } else { | |
5a7c5680 | 545 | sync_done: |
11786802 JA |
546 | fio_gettime(&comp_time, NULL); |
547 | bytes_done = io_u_sync_complete(td, io_u); | |
548 | if (bytes_done < 0) | |
549 | ret = bytes_done; | |
cec6b55d | 550 | } |
36167d82 JA |
551 | break; |
552 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
553 | /* |
554 | * if the engine doesn't have a commit hook, | |
555 | * the io_u is really queued. if it does have such | |
556 | * a hook, it has to call io_u_queued() itself. | |
557 | */ | |
558 | if (td->io_ops->commit == NULL) | |
559 | io_u_queued(td, io_u); | |
36167d82 | 560 | break; |
755200a3 JA |
561 | case FIO_Q_BUSY: |
562 | requeue_io_u(td, &io_u); | |
5451792e JA |
563 | ret2 = td_io_commit(td); |
564 | if (ret2 < 0) | |
565 | ret = ret2; | |
755200a3 | 566 | break; |
36167d82 JA |
567 | default: |
568 | assert(ret < 0); | |
569 | put_io_u(td, io_u); | |
570 | break; | |
ebac4655 JA |
571 | } |
572 | ||
99784632 | 573 | if (ret < 0 || td->error) |
36167d82 JA |
574 | break; |
575 | ||
97601024 JA |
576 | /* |
577 | * See if we need to complete some commands | |
578 | */ | |
4950421a JA |
579 | full = queue_full(td) || ret == FIO_Q_BUSY; |
580 | if (full || !td->o.iodepth_batch_complete) { | |
581 | min_evts = td->o.iodepth_batch_complete; | |
582 | if (full && !min_evts) | |
36167d82 | 583 | min_evts = 1; |
4950421a | 584 | |
97601024 | 585 | fio_gettime(&comp_time, NULL); |
4950421a JA |
586 | |
587 | do { | |
588 | ret = io_u_queued_complete(td, min_evts); | |
589 | if (ret <= 0) | |
590 | break; | |
591 | ||
592 | bytes_done += ret; | |
593 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
ebac4655 JA |
594 | } |
595 | ||
4950421a JA |
596 | if (ret < 0) |
597 | break; | |
97601024 JA |
598 | if (!bytes_done) |
599 | continue; | |
600 | ||
ebac4655 JA |
601 | /* |
602 | * the rate is batched for now, it should work for batches | |
603 | * of completions except the very first one which may look | |
604 | * a little bursty | |
605 | */ | |
97601024 | 606 | usec = utime_since(&s, &comp_time); |
ebac4655 | 607 | |
413dd459 | 608 | rate_throttle(td, usec, bytes_done); |
ebac4655 | 609 | |
97601024 | 610 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 611 | if (exitall_on_terminate) |
390c40e2 | 612 | terminate_threads(td->groupid); |
41d9f0fd | 613 | td_verror(td, EIO, "check_min_rate"); |
ebac4655 JA |
614 | break; |
615 | } | |
616 | ||
2dc1bbeb | 617 | if (td->o.thinktime) { |
9c1f7434 JA |
618 | unsigned long long b; |
619 | ||
620 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 621 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
622 | int left; |
623 | ||
2dc1bbeb JA |
624 | if (td->o.thinktime_spin) |
625 | __usec_sleep(td->o.thinktime_spin); | |
48097d5c | 626 | |
2dc1bbeb | 627 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
628 | if (left) |
629 | usec_sleep(td, left); | |
630 | } | |
9c1f7434 | 631 | } |
ebac4655 JA |
632 | } |
633 | ||
aa31f1f1 SL |
634 | if (td->o.fill_device && td->error == ENOSPC) { |
635 | td->error = 0; | |
636 | td->terminate = 1; | |
637 | } | |
4d2413c6 | 638 | if (!td->error) { |
3d7c391d JA |
639 | struct fio_file *f; |
640 | ||
c01c0395 JA |
641 | i = td->cur_depth; |
642 | if (i) | |
643 | ret = io_u_queued_complete(td, i); | |
ebac4655 | 644 | |
2dc1bbeb | 645 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 646 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
647 | |
648 | for_each_file(td, f, i) { | |
649 | if (!(f->flags & FIO_FILE_OPEN)) | |
650 | continue; | |
858a3d47 | 651 | fio_io_sync(td, f); |
3d7b485f | 652 | } |
84585003 | 653 | } |
c01c0395 JA |
654 | } else |
655 | cleanup_pending_aio(td); | |
2ba1c290 JA |
656 | |
657 | /* | |
658 | * stop job if we failed doing any IO | |
659 | */ | |
660 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
661 | td->done = 1; | |
ebac4655 JA |
662 | } |
663 | ||
ebac4655 JA |
664 | static void cleanup_io_u(struct thread_data *td) |
665 | { | |
01743ee1 | 666 | struct flist_head *entry, *n; |
ebac4655 JA |
667 | struct io_u *io_u; |
668 | ||
01743ee1 JA |
669 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
670 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 671 | |
01743ee1 | 672 | flist_del(&io_u->list); |
ebac4655 JA |
673 | free(io_u); |
674 | } | |
675 | ||
2f9ade3c | 676 | free_io_mem(td); |
ebac4655 JA |
677 | } |
678 | ||
679 | static int init_io_u(struct thread_data *td) | |
680 | { | |
681 | struct io_u *io_u; | |
a00735e6 | 682 | unsigned int max_bs; |
ebac4655 JA |
683 | int i, max_units; |
684 | char *p; | |
685 | ||
1e3c09af | 686 | max_units = td->o.iodepth; |
2dc1bbeb | 687 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
688 | td->orig_buffer_size = (unsigned long long) max_bs |
689 | * (unsigned long long) max_units; | |
690 | ||
691 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
692 | unsigned long bs; | |
74b025b0 | 693 | |
5ec10eaa JA |
694 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
695 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
696 | } | |
ebac4655 | 697 | |
c3990645 JA |
698 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
699 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
700 | return 1; | |
701 | } | |
702 | ||
2f9ade3c JA |
703 | if (allocate_io_mem(td)) |
704 | return 1; | |
ebac4655 | 705 | |
b3f378e9 JA |
706 | if (td->o.odirect) |
707 | p = ALIGN(td->orig_buffer); | |
708 | else | |
709 | p = td->orig_buffer; | |
710 | ||
ebac4655 | 711 | for (i = 0; i < max_units; i++) { |
db1cd90b JA |
712 | if (td->terminate) |
713 | return 1; | |
ebac4655 JA |
714 | io_u = malloc(sizeof(*io_u)); |
715 | memset(io_u, 0, sizeof(*io_u)); | |
01743ee1 | 716 | INIT_FLIST_HEAD(&io_u->list); |
ebac4655 | 717 | |
b4c5e1ac JA |
718 | if (!(td->io_ops->flags & FIO_NOIO)) { |
719 | io_u->buf = p + max_bs * i; | |
e9459e5a | 720 | |
5973cafb JA |
721 | if (td_write(td) && !td->o.refill_buffers) |
722 | io_u_fill_buffer(td, io_u, max_bs); | |
b4c5e1ac | 723 | } |
6b9cea23 | 724 | |
b1ff3403 | 725 | io_u->index = i; |
0c6e7517 | 726 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 727 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
728 | } |
729 | ||
433afcb4 JA |
730 | io_u_init_timeout(); |
731 | ||
ebac4655 JA |
732 | return 0; |
733 | } | |
734 | ||
da86774e JA |
735 | static int switch_ioscheduler(struct thread_data *td) |
736 | { | |
737 | char tmp[256], tmp2[128]; | |
738 | FILE *f; | |
739 | int ret; | |
740 | ||
ba0fbe10 | 741 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
742 | return 0; |
743 | ||
da86774e JA |
744 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
745 | ||
746 | f = fopen(tmp, "r+"); | |
747 | if (!f) { | |
cbf5c1d7 | 748 | if (errno == ENOENT) { |
5ec10eaa JA |
749 | log_err("fio: os or kernel doesn't support IO scheduler" |
750 | " switching\n"); | |
cbf5c1d7 JA |
751 | return 0; |
752 | } | |
753 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
754 | return 1; |
755 | } | |
756 | ||
757 | /* | |
758 | * Set io scheduler. | |
759 | */ | |
2dc1bbeb | 760 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 761 | if (ferror(f) || ret != 1) { |
e1161c32 | 762 | td_verror(td, errno, "fwrite"); |
da86774e JA |
763 | fclose(f); |
764 | return 1; | |
765 | } | |
766 | ||
767 | rewind(f); | |
768 | ||
769 | /* | |
770 | * Read back and check that the selected scheduler is now the default. | |
771 | */ | |
772 | ret = fread(tmp, 1, sizeof(tmp), f); | |
773 | if (ferror(f) || ret < 0) { | |
e1161c32 | 774 | td_verror(td, errno, "fread"); |
da86774e JA |
775 | fclose(f); |
776 | return 1; | |
777 | } | |
778 | ||
2dc1bbeb | 779 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 780 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 781 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 782 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
783 | fclose(f); |
784 | return 1; | |
785 | } | |
786 | ||
787 | fclose(f); | |
788 | return 0; | |
789 | } | |
790 | ||
492158cf JA |
791 | static int keep_running(struct thread_data *td) |
792 | { | |
793 | unsigned long long io_done; | |
794 | ||
20e354ef JA |
795 | if (td->done) |
796 | return 0; | |
492158cf JA |
797 | if (td->o.time_based) |
798 | return 1; | |
799 | if (td->o.loops) { | |
800 | td->o.loops--; | |
801 | return 1; | |
802 | } | |
803 | ||
5ec10eaa JA |
804 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
805 | + td->io_skip_bytes; | |
492158cf JA |
806 | if (io_done < td->o.size) |
807 | return 1; | |
808 | ||
809 | return 0; | |
810 | } | |
811 | ||
a978ba68 | 812 | static int clear_io_state(struct thread_data *td) |
ebac4655 | 813 | { |
53cdc686 | 814 | struct fio_file *f; |
af52b345 JA |
815 | unsigned int i; |
816 | int ret; | |
ebac4655 | 817 | |
756867bd | 818 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 819 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 820 | td->zone_bytes = 0; |
d7d3b49b | 821 | td->rate_bytes = 0; |
4e991c23 | 822 | td->rate_blocks = 0; |
38d77cae | 823 | td->rw_end_set[0] = td->rw_end_set[1] = 0; |
ebac4655 | 824 | |
c1324df1 JA |
825 | td->last_was_sync = 0; |
826 | ||
2ba1c290 JA |
827 | /* |
828 | * reset file done count if we are to start over | |
829 | */ | |
830 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 831 | td->nr_done_files = 0; |
281f9b63 | 832 | |
24ffd2c2 | 833 | close_files(td); |
53cdc686 | 834 | |
a978ba68 JA |
835 | ret = 0; |
836 | for_each_file(td, f, i) { | |
281f9b63 | 837 | f->flags &= ~FIO_FILE_DONE; |
a978ba68 JA |
838 | ret = td_io_open_file(td, f); |
839 | if (ret) | |
840 | break; | |
53cdc686 | 841 | } |
a978ba68 JA |
842 | |
843 | return ret; | |
ebac4655 JA |
844 | } |
845 | ||
906c8d75 JA |
846 | /* |
847 | * Entry point for the thread based jobs. The process based jobs end up | |
848 | * here as well, after a little setup. | |
849 | */ | |
ebac4655 JA |
850 | static void *thread_main(void *data) |
851 | { | |
10927316 | 852 | unsigned long long runtime[2], elapsed; |
ebac4655 | 853 | struct thread_data *td = data; |
a978ba68 | 854 | int clear_state; |
ebac4655 | 855 | |
2dc1bbeb | 856 | if (!td->o.use_thread) |
ebac4655 JA |
857 | setsid(); |
858 | ||
859 | td->pid = getpid(); | |
860 | ||
5921e80c | 861 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 862 | |
01743ee1 JA |
863 | INIT_FLIST_HEAD(&td->io_u_freelist); |
864 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
865 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
866 | INIT_FLIST_HEAD(&td->io_log_list); | |
867 | INIT_FLIST_HEAD(&td->io_hist_list); | |
bb5d7d0b | 868 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 869 | |
db1cd90b | 870 | td_set_runstate(td, TD_INITIALIZED); |
cdd18ad8 JA |
871 | fio_mutex_up(startup_mutex); |
872 | fio_mutex_down(td->mutex); | |
db1cd90b JA |
873 | |
874 | /* | |
cdd18ad8 | 875 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
876 | * eating a file descriptor |
877 | */ | |
cdd18ad8 | 878 | fio_mutex_remove(td->mutex); |
db1cd90b | 879 | |
d84f8d49 JA |
880 | /* |
881 | * May alter parameters that init_io_u() will use, so we need to | |
882 | * do this first. | |
883 | */ | |
884 | if (init_iolog(td)) | |
885 | goto err; | |
886 | ||
ebac4655 | 887 | if (init_io_u(td)) |
db1cd90b | 888 | goto err; |
ebac4655 | 889 | |
375b2695 | 890 | if (td->o.cpumask_set && fio_setaffinity(td) == -1) { |
e1161c32 | 891 | td_verror(td, errno, "cpu_set_affinity"); |
db1cd90b | 892 | goto err; |
ebac4655 JA |
893 | } |
894 | ||
ac684785 | 895 | if (td->ioprio_set) { |
ebac4655 | 896 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 897 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 898 | goto err; |
ebac4655 JA |
899 | } |
900 | } | |
901 | ||
2dc1bbeb | 902 | if (nice(td->o.nice) == -1) { |
e1161c32 | 903 | td_verror(td, errno, "nice"); |
db1cd90b | 904 | goto err; |
b6f4d880 JA |
905 | } |
906 | ||
2dc1bbeb | 907 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 908 | goto err; |
37f56873 | 909 | |
2dc1bbeb | 910 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
911 | goto err; |
912 | ||
7d6c5283 JA |
913 | if (td_io_init(td)) |
914 | goto err; | |
915 | ||
b5af8293 JA |
916 | if (open_files(td)) |
917 | goto err; | |
918 | ||
68727076 JA |
919 | if (init_random_map(td)) |
920 | goto err; | |
921 | ||
2dc1bbeb JA |
922 | if (td->o.exec_prerun) { |
923 | if (system(td->o.exec_prerun) < 0) | |
69cfd7e0 JA |
924 | goto err; |
925 | } | |
4e0ba8af | 926 | |
69008999 | 927 | fio_gettime(&td->epoch, NULL); |
433afcb4 | 928 | memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch)); |
756867bd | 929 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
930 | |
931 | runtime[0] = runtime[1] = 0; | |
a978ba68 | 932 | clear_state = 0; |
492158cf | 933 | while (keep_running(td)) { |
02bcaa8c | 934 | fio_gettime(&td->start, NULL); |
756867bd | 935 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 | 936 | |
2dc1bbeb | 937 | if (td->o.ratemin) |
5ec10eaa JA |
938 | memcpy(&td->lastrate, &td->ts.stat_sample_time, |
939 | sizeof(td->lastrate)); | |
ebac4655 | 940 | |
a978ba68 JA |
941 | if (clear_state && clear_io_state(td)) |
942 | break; | |
943 | ||
ebac4655 JA |
944 | prune_io_piece_log(td); |
945 | ||
ba0fbe10 | 946 | do_io(td); |
ebac4655 | 947 | |
a978ba68 JA |
948 | clear_state = 1; |
949 | ||
38d77cae JA |
950 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
951 | if (td->rw_end_set[DDIR_READ]) | |
5ec10eaa JA |
952 | elapsed = utime_since(&td->start, |
953 | &td->rw_end[DDIR_READ]); | |
38d77cae JA |
954 | else |
955 | elapsed = utime_since_now(&td->start); | |
956 | ||
957 | runtime[DDIR_READ] += elapsed; | |
958 | } | |
959 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
960 | if (td->rw_end_set[DDIR_WRITE]) | |
5ec10eaa JA |
961 | elapsed = utime_since(&td->start, |
962 | &td->rw_end[DDIR_WRITE]); | |
38d77cae JA |
963 | else |
964 | elapsed = utime_since_now(&td->start); | |
965 | ||
966 | runtime[DDIR_WRITE] += elapsed; | |
967 | } | |
5ec10eaa | 968 | |
ebac4655 JA |
969 | if (td->error || td->terminate) |
970 | break; | |
971 | ||
e84c73a8 SL |
972 | if (!td->o.do_verify || |
973 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 974 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
975 | continue; |
976 | ||
a978ba68 JA |
977 | if (clear_io_state(td)) |
978 | break; | |
979 | ||
02bcaa8c | 980 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
981 | |
982 | do_verify(td); | |
983 | ||
69008999 | 984 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
985 | |
986 | if (td->error || td->terminate) | |
987 | break; | |
988 | } | |
989 | ||
36dff966 | 990 | update_rusage_stat(td); |
47f0cc48 YHJT |
991 | td->ts.runtime[0] = (runtime[0] + 999) / 1000; |
992 | td->ts.runtime[1] = (runtime[1] + 999) / 1000; | |
756867bd JA |
993 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
994 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
995 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
996 | ||
997 | if (td->ts.bw_log) | |
998 | finish_log(td, td->ts.bw_log, "bw"); | |
999 | if (td->ts.slat_log) | |
1000 | finish_log(td, td->ts.slat_log, "slat"); | |
1001 | if (td->ts.clat_log) | |
1002 | finish_log(td, td->ts.clat_log, "clat"); | |
2dc1bbeb JA |
1003 | if (td->o.exec_postrun) { |
1004 | if (system(td->o.exec_postrun) < 0) | |
1005 | log_err("fio: postrun %s failed\n", td->o.exec_postrun); | |
69cfd7e0 | 1006 | } |
ebac4655 JA |
1007 | |
1008 | if (exitall_on_terminate) | |
390c40e2 | 1009 | terminate_threads(td->groupid); |
ebac4655 JA |
1010 | |
1011 | err: | |
5bf13a5a | 1012 | if (td->error) |
5921e80c | 1013 | printf("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1014 | td->verror); |
24ffd2c2 | 1015 | close_and_free_files(td); |
2866c82d | 1016 | close_ioengine(td); |
ebac4655 | 1017 | cleanup_io_u(td); |
f29b25a3 JA |
1018 | |
1019 | /* | |
1020 | * do this very late, it will log file closing as well | |
1021 | */ | |
1022 | if (td->o.write_iolog_file) | |
1023 | write_iolog_close(td); | |
1024 | ||
d23bb327 | 1025 | options_mem_free(td); |
ebac4655 | 1026 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1027 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1028 | } |
1029 | ||
906c8d75 JA |
1030 | /* |
1031 | * We cannot pass the td data into a forked process, so attach the td and | |
1032 | * pass it to the thread worker. | |
1033 | */ | |
a6418147 | 1034 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1035 | { |
1036 | struct thread_data *td; | |
a6418147 | 1037 | void *data, *ret; |
ebac4655 JA |
1038 | |
1039 | data = shmat(shmid, NULL, 0); | |
1040 | if (data == (void *) -1) { | |
a6418147 JA |
1041 | int __err = errno; |
1042 | ||
ebac4655 | 1043 | perror("shmat"); |
a6418147 | 1044 | return __err; |
ebac4655 JA |
1045 | } |
1046 | ||
1047 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1048 | ret = thread_main(td); |
ebac4655 | 1049 | shmdt(data); |
43d76807 | 1050 | return (int) (unsigned long) ret; |
ebac4655 JA |
1051 | } |
1052 | ||
906c8d75 JA |
1053 | /* |
1054 | * Run over the job map and reap the threads that have exited, if any. | |
1055 | */ | |
ebac4655 JA |
1056 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1057 | { | |
34572e28 | 1058 | struct thread_data *td; |
1f809d15 | 1059 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1060 | |
1061 | /* | |
1062 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1063 | */ | |
1f809d15 | 1064 | realthreads = pending = cputhreads = 0; |
34572e28 | 1065 | for_each_td(td, i) { |
3707f45b | 1066 | int flags = 0; |
a2f77c9f | 1067 | |
84585003 JA |
1068 | /* |
1069 | * ->io_ops is NULL for a thread that has closed its | |
1070 | * io engine | |
1071 | */ | |
ba0fbe10 | 1072 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1073 | cputhreads++; |
1f809d15 JA |
1074 | else |
1075 | realthreads++; | |
b990b5c0 | 1076 | |
a1dedc1f AC |
1077 | if (!td->pid) { |
1078 | pending++; | |
1079 | continue; | |
1080 | } | |
1081 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1082 | continue; |
2dc1bbeb | 1083 | if (td->o.use_thread) { |
3707f45b JA |
1084 | if (td->runstate == TD_EXITED) { |
1085 | td_set_runstate(td, TD_REAPED); | |
1086 | goto reaped; | |
1087 | } | |
1088 | continue; | |
1089 | } | |
a2f77c9f JA |
1090 | |
1091 | flags = WNOHANG; | |
1092 | if (td->runstate == TD_EXITED) | |
1093 | flags = 0; | |
1094 | ||
1095 | /* | |
1096 | * check if someone quit or got killed in an unusual way | |
1097 | */ | |
1098 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1099 | if (ret < 0) { |
a2f77c9f | 1100 | if (errno == ECHILD) { |
5921e80c JA |
1101 | log_err("fio: pid=%d disappeared %d\n", |
1102 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1103 | td_set_runstate(td, TD_REAPED); |
1104 | goto reaped; | |
1105 | } | |
1106 | perror("waitpid"); | |
1107 | } else if (ret == td->pid) { | |
1108 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1109 | int sig = WTERMSIG(status); |
1110 | ||
d9244941 | 1111 | if (sig != SIGQUIT) |
5ec10eaa | 1112 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1113 | (int) td->pid, sig); |
fab6aa71 JA |
1114 | td_set_runstate(td, TD_REAPED); |
1115 | goto reaped; | |
1116 | } | |
a2f77c9f JA |
1117 | if (WIFEXITED(status)) { |
1118 | if (WEXITSTATUS(status) && !td->error) | |
1119 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1120 | |
a2f77c9f JA |
1121 | td_set_runstate(td, TD_REAPED); |
1122 | goto reaped; | |
a6418147 JA |
1123 | } |
1124 | } | |
ebac4655 | 1125 | |
a2f77c9f JA |
1126 | /* |
1127 | * thread is not dead, continue | |
1128 | */ | |
e48676ba | 1129 | pending++; |
a2f77c9f | 1130 | continue; |
fab6aa71 | 1131 | reaped: |
ebac4655 | 1132 | (*nr_running)--; |
2dc1bbeb JA |
1133 | (*m_rate) -= td->o.ratemin; |
1134 | (*t_rate) -= td->o.rate; | |
6f88bcb0 JA |
1135 | if (!td->pid) |
1136 | pending--; | |
a2f77c9f JA |
1137 | |
1138 | if (td->error) | |
1139 | exit_value++; | |
d3eeeabc JA |
1140 | |
1141 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1142 | } |
b990b5c0 | 1143 | |
1f809d15 | 1144 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1145 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1146 | } |
1147 | ||
906c8d75 JA |
1148 | /* |
1149 | * Main function for kicking off and reaping jobs, as needed. | |
1150 | */ | |
ebac4655 JA |
1151 | static void run_threads(void) |
1152 | { | |
ebac4655 JA |
1153 | struct thread_data *td; |
1154 | unsigned long spent; | |
1155 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1156 | |
2f9ade3c JA |
1157 | if (fio_pin_memory()) |
1158 | return; | |
ebac4655 | 1159 | |
c6ae0a5b | 1160 | if (!terse_output) { |
9cedf167 JA |
1161 | printf("Starting "); |
1162 | if (nr_thread) | |
5ec10eaa JA |
1163 | printf("%d thread%s", nr_thread, |
1164 | nr_thread > 1 ? "s" : ""); | |
9cedf167 JA |
1165 | if (nr_process) { |
1166 | if (nr_thread) | |
1167 | printf(" and "); | |
5ec10eaa JA |
1168 | printf("%d process%s", nr_process, |
1169 | nr_process > 1 ? "es" : ""); | |
9cedf167 JA |
1170 | } |
1171 | printf("\n"); | |
c6ae0a5b JA |
1172 | fflush(stdout); |
1173 | } | |
c04f7ec3 | 1174 | |
697a606c | 1175 | set_sig_handlers(); |
4efa970e | 1176 | |
ebac4655 JA |
1177 | todo = thread_number; |
1178 | nr_running = 0; | |
1179 | nr_started = 0; | |
1180 | m_rate = t_rate = 0; | |
1181 | ||
34572e28 | 1182 | for_each_td(td, i) { |
263e529f | 1183 | print_status_init(td->thread_number - 1); |
ebac4655 | 1184 | |
2dc1bbeb | 1185 | if (!td->o.create_serialize) { |
380cf265 | 1186 | init_disk_util(td); |
ebac4655 | 1187 | continue; |
380cf265 | 1188 | } |
ebac4655 JA |
1189 | |
1190 | /* | |
1191 | * do file setup here so it happens sequentially, | |
1192 | * we don't want X number of threads getting their | |
1193 | * client data interspersed on disk | |
1194 | */ | |
53cdc686 | 1195 | if (setup_files(td)) { |
5bf13a5a JA |
1196 | exit_value++; |
1197 | if (td->error) | |
5921e80c JA |
1198 | log_err("fio: pid=%d, err=%d/%s\n", |
1199 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1200 | td_set_runstate(td, TD_REAPED); |
1201 | todo--; | |
c69e8875 JA |
1202 | } else { |
1203 | struct fio_file *f; | |
1204 | unsigned int i; | |
1205 | ||
1206 | /* | |
1207 | * for sharing to work, each job must always open | |
1208 | * its own files. so close them, if we opened them | |
1209 | * for creation | |
1210 | */ | |
1211 | for_each_file(td, f, i) | |
1212 | td_io_close_file(td, f); | |
5ec10eaa | 1213 | } |
380cf265 JA |
1214 | |
1215 | init_disk_util(td); | |
ebac4655 JA |
1216 | } |
1217 | ||
a2f77c9f JA |
1218 | set_genesis_time(); |
1219 | ||
ebac4655 | 1220 | while (todo) { |
75154845 JA |
1221 | struct thread_data *map[MAX_JOBS]; |
1222 | struct timeval this_start; | |
1223 | int this_jobs = 0, left; | |
1224 | ||
ebac4655 JA |
1225 | /* |
1226 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1227 | */ | |
34572e28 | 1228 | for_each_td(td, i) { |
ebac4655 JA |
1229 | if (td->runstate != TD_NOT_CREATED) |
1230 | continue; | |
1231 | ||
1232 | /* | |
1233 | * never got a chance to start, killed by other | |
1234 | * thread for some reason | |
1235 | */ | |
1236 | if (td->terminate) { | |
1237 | todo--; | |
1238 | continue; | |
1239 | } | |
1240 | ||
2dc1bbeb | 1241 | if (td->o.start_delay) { |
263e529f | 1242 | spent = mtime_since_genesis(); |
ebac4655 | 1243 | |
2dc1bbeb | 1244 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1245 | continue; |
1246 | } | |
1247 | ||
6f88bcb0 JA |
1248 | if (td->o.stonewall && (nr_started || nr_running)) { |
1249 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1250 | td->o.name); | |
ebac4655 | 1251 | break; |
6f88bcb0 | 1252 | } |
ebac4655 | 1253 | |
75154845 JA |
1254 | /* |
1255 | * Set state to created. Thread will transition | |
1256 | * to TD_INITIALIZED when it's done setting up. | |
1257 | */ | |
ebac4655 | 1258 | td_set_runstate(td, TD_CREATED); |
75154845 | 1259 | map[this_jobs++] = td; |
ebac4655 JA |
1260 | nr_started++; |
1261 | ||
2dc1bbeb | 1262 | if (td->o.use_thread) { |
ee56ad50 | 1263 | dprint(FD_PROCESS, "will pthread_create\n"); |
5ec10eaa JA |
1264 | if (pthread_create(&td->thread, NULL, |
1265 | thread_main, td)) { | |
f4705a73 | 1266 | perror("pthread_create"); |
ebac4655 | 1267 | nr_started--; |
c8bb6faf | 1268 | break; |
ebac4655 | 1269 | } |
f4705a73 JA |
1270 | if (pthread_detach(td->thread) < 0) |
1271 | perror("pthread_detach"); | |
ebac4655 | 1272 | } else { |
5e1d306e | 1273 | pid_t pid; |
ee56ad50 | 1274 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1275 | pid = fork(); |
1276 | if (!pid) { | |
a6418147 JA |
1277 | int ret = fork_main(shm_id, i); |
1278 | ||
5e1d306e JA |
1279 | _exit(ret); |
1280 | } else if (i == fio_debug_jobno) | |
1281 | *fio_debug_jobp = pid; | |
ebac4655 | 1282 | } |
cdd18ad8 | 1283 | fio_mutex_down(startup_mutex); |
ebac4655 JA |
1284 | } |
1285 | ||
1286 | /* | |
75154845 JA |
1287 | * Wait for the started threads to transition to |
1288 | * TD_INITIALIZED. | |
ebac4655 | 1289 | */ |
02bcaa8c | 1290 | fio_gettime(&this_start, NULL); |
75154845 | 1291 | left = this_jobs; |
6ce15a32 | 1292 | while (left && !fio_abort) { |
75154845 JA |
1293 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1294 | break; | |
1295 | ||
1296 | usleep(100000); | |
1297 | ||
1298 | for (i = 0; i < this_jobs; i++) { | |
1299 | td = map[i]; | |
1300 | if (!td) | |
1301 | continue; | |
b6f4d880 | 1302 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1303 | map[i] = NULL; |
1304 | left--; | |
b6f4d880 JA |
1305 | } else if (td->runstate >= TD_EXITED) { |
1306 | map[i] = NULL; | |
1307 | left--; | |
1308 | todo--; | |
1309 | nr_running++; /* work-around... */ | |
75154845 JA |
1310 | } |
1311 | } | |
1312 | } | |
1313 | ||
1314 | if (left) { | |
3b70d7e5 | 1315 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1316 | for (i = 0; i < this_jobs; i++) { |
1317 | td = map[i]; | |
1318 | if (!td) | |
1319 | continue; | |
1320 | kill(td->pid, SIGTERM); | |
1321 | } | |
1322 | break; | |
1323 | } | |
1324 | ||
1325 | /* | |
b6f4d880 | 1326 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1327 | */ |
34572e28 | 1328 | for_each_td(td, i) { |
75154845 | 1329 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1330 | continue; |
1331 | ||
1332 | td_set_runstate(td, TD_RUNNING); | |
1333 | nr_running++; | |
1334 | nr_started--; | |
2dc1bbeb JA |
1335 | m_rate += td->o.ratemin; |
1336 | t_rate += td->o.rate; | |
75154845 | 1337 | todo--; |
cdd18ad8 | 1338 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1339 | } |
1340 | ||
1341 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1342 | ||
1343 | if (todo) | |
1344 | usleep(100000); | |
1345 | } | |
1346 | ||
1347 | while (nr_running) { | |
1348 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1349 | usleep(10000); | |
1350 | } | |
1351 | ||
1352 | update_io_ticks(); | |
2f9ade3c | 1353 | fio_unpin_memory(); |
ebac4655 JA |
1354 | } |
1355 | ||
ebac4655 JA |
1356 | int main(int argc, char *argv[]) |
1357 | { | |
29d610e1 JA |
1358 | long ps; |
1359 | ||
2e5cdb11 JA |
1360 | sinit(); |
1361 | ||
dbe1125e JA |
1362 | /* |
1363 | * We need locale for number printing, if it isn't set then just | |
1364 | * go with the US format. | |
1365 | */ | |
1366 | if (!getenv("LC_NUMERIC")) | |
1367 | setlocale(LC_NUMERIC, "en_US"); | |
1368 | ||
ebac4655 JA |
1369 | if (parse_options(argc, argv)) |
1370 | return 1; | |
1371 | ||
4b472fa3 JA |
1372 | if (!thread_number) |
1373 | return 0; | |
ebac4655 | 1374 | |
29d610e1 JA |
1375 | ps = sysconf(_SC_PAGESIZE); |
1376 | if (ps < 0) { | |
1377 | log_err("Failed to get page size\n"); | |
1378 | return 1; | |
1379 | } | |
1380 | ||
cfc99db7 | 1381 | page_size = ps; |
29d610e1 JA |
1382 | page_mask = ps - 1; |
1383 | ||
bb3884d8 JA |
1384 | if (write_bw_log) { |
1385 | setup_log(&agg_io_log[DDIR_READ]); | |
1386 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1387 | } | |
1388 | ||
cdd18ad8 | 1389 | startup_mutex = fio_mutex_init(0); |
07739b57 | 1390 | |
a2f77c9f JA |
1391 | set_genesis_time(); |
1392 | ||
ccb0fa24 | 1393 | status_timer_arm(); |
ebac4655 JA |
1394 | |
1395 | run_threads(); | |
6ce15a32 | 1396 | |
bb3884d8 | 1397 | if (!fio_abort) { |
6ce15a32 | 1398 | show_run_stats(); |
bb3884d8 | 1399 | if (write_bw_log) { |
5ec10eaa JA |
1400 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1401 | __finish_log(agg_io_log[DDIR_WRITE], | |
1402 | "agg-write_bw.log"); | |
bb3884d8 JA |
1403 | } |
1404 | } | |
ebac4655 | 1405 | |
cdd18ad8 | 1406 | fio_mutex_remove(startup_mutex); |
437c9b71 | 1407 | return exit_value; |
ebac4655 | 1408 | } |