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" | |
38 | #include "os.h" | |
39 | ||
29d610e1 JA |
40 | static unsigned long page_mask; |
41 | #define ALIGN(buf) \ | |
42 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
43 | |
44 | int groupid = 0; | |
45 | int thread_number = 0; | |
ebac4655 | 46 | int shm_id = 0; |
53cdc686 | 47 | int temp_stall_ts; |
ebac4655 | 48 | |
bbfd6b00 | 49 | static volatile int startup_sem; |
6ce15a32 | 50 | static volatile int fio_abort; |
437c9b71 | 51 | static int exit_value; |
ebac4655 | 52 | |
bb3884d8 JA |
53 | struct io_log *agg_io_log[2]; |
54 | ||
ebac4655 | 55 | #define TERMINATE_ALL (-1) |
75154845 | 56 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 57 | |
6ce15a32 JA |
58 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
59 | { | |
60 | td->runstate = runstate; | |
61 | } | |
62 | ||
63 | static void terminate_threads(int group_id, int forced_kill) | |
ebac4655 | 64 | { |
34572e28 | 65 | struct thread_data *td; |
ebac4655 JA |
66 | int i; |
67 | ||
34572e28 | 68 | for_each_td(td, i) { |
ebac4655 JA |
69 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
70 | td->terminate = 1; | |
71 | td->start_delay = 0; | |
6ce15a32 JA |
72 | if (forced_kill) |
73 | td_set_runstate(td, TD_EXITED); | |
ebac4655 JA |
74 | } |
75 | } | |
76 | } | |
77 | ||
78 | static void sig_handler(int sig) | |
79 | { | |
80 | switch (sig) { | |
81 | case SIGALRM: | |
82 | update_io_ticks(); | |
83 | disk_util_timer_arm(); | |
84 | print_thread_status(); | |
85 | break; | |
86 | default: | |
6ce15a32 | 87 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 88 | fflush(stdout); |
6ce15a32 | 89 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
90 | break; |
91 | } | |
92 | } | |
93 | ||
906c8d75 JA |
94 | /* |
95 | * Check if we are above the minimum rate given. | |
96 | */ | |
ebac4655 JA |
97 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
98 | { | |
99 | unsigned long spent; | |
100 | unsigned long rate; | |
101 | int ddir = td->ddir; | |
102 | ||
103 | /* | |
104 | * allow a 2 second settle period in the beginning | |
105 | */ | |
106 | if (mtime_since(&td->start, now) < 2000) | |
107 | return 0; | |
108 | ||
109 | /* | |
110 | * if rate blocks is set, sample is running | |
111 | */ | |
112 | if (td->rate_bytes) { | |
113 | spent = mtime_since(&td->lastrate, now); | |
114 | if (spent < td->ratecycle) | |
115 | return 0; | |
116 | ||
117 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
118 | if (rate < td->ratemin) { | |
1e97cce9 | 119 | fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
120 | return 1; |
121 | } | |
122 | } | |
123 | ||
124 | td->rate_bytes = td->this_io_bytes[ddir]; | |
125 | memcpy(&td->lastrate, now, sizeof(*now)); | |
126 | return 0; | |
127 | } | |
128 | ||
129 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
130 | { | |
131 | if (!td->timeout) | |
132 | return 0; | |
133 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
134 | return 1; | |
135 | ||
136 | return 0; | |
137 | } | |
138 | ||
906c8d75 JA |
139 | /* |
140 | * When job exits, we can cancel the in-flight IO if we are using async | |
141 | * io. Attempt to do so. | |
142 | */ | |
ebac4655 JA |
143 | static void cleanup_pending_aio(struct thread_data *td) |
144 | { | |
ebac4655 | 145 | struct list_head *entry, *n; |
ebac4655 JA |
146 | struct io_u *io_u; |
147 | int r; | |
148 | ||
149 | /* | |
150 | * get immediately available events, if any | |
151 | */ | |
b2fdda43 JA |
152 | r = io_u_queued_complete(td, 0, NULL); |
153 | if (r < 0) | |
154 | return; | |
ebac4655 JA |
155 | |
156 | /* | |
157 | * now cancel remaining active events | |
158 | */ | |
2866c82d | 159 | if (td->io_ops->cancel) { |
ebac4655 JA |
160 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
161 | io_u = list_entry(entry, struct io_u, list); | |
162 | ||
2866c82d | 163 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
164 | if (!r) |
165 | put_io_u(td, io_u); | |
166 | } | |
167 | } | |
168 | ||
97601024 | 169 | if (td->cur_depth) |
b2fdda43 | 170 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
171 | } |
172 | ||
858a3d47 JA |
173 | /* |
174 | * Helper to handle the final sync of a file. Works just like the normal | |
175 | * io path, just does everything sync. | |
176 | */ | |
177 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
178 | { | |
179 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
180 | int ret; |
181 | ||
182 | if (!io_u) | |
183 | return 1; | |
184 | ||
185 | io_u->ddir = DDIR_SYNC; | |
186 | io_u->file = f; | |
187 | ||
188 | if (td_io_prep(td, io_u)) { | |
189 | put_io_u(td, io_u); | |
190 | return 1; | |
191 | } | |
192 | ||
755200a3 | 193 | requeue: |
858a3d47 | 194 | ret = td_io_queue(td, io_u); |
36167d82 | 195 | if (ret < 0) { |
353a7e0e | 196 | td_verror(td, io_u->error); |
858a3d47 | 197 | put_io_u(td, io_u); |
858a3d47 | 198 | return 1; |
36167d82 | 199 | } else if (ret == FIO_Q_QUEUED) { |
97601024 | 200 | if (io_u_queued_complete(td, 1, NULL)) |
36167d82 | 201 | return 1; |
36167d82 JA |
202 | } else if (ret == FIO_Q_COMPLETED) { |
203 | if (io_u->error) { | |
204 | td_verror(td, io_u->error); | |
205 | return 1; | |
206 | } | |
858a3d47 | 207 | |
b2fdda43 JA |
208 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
209 | return 1; | |
755200a3 JA |
210 | } else if (ret == FIO_Q_BUSY) { |
211 | if (td_io_commit(td)) | |
212 | return 1; | |
213 | goto requeue; | |
858a3d47 JA |
214 | } |
215 | ||
216 | return 0; | |
217 | } | |
218 | ||
906c8d75 JA |
219 | /* |
220 | * The main verify engine. Runs over the writes we previusly submitted, | |
221 | * reads the blocks back in, and checks the crc/md5 of the data. | |
222 | */ | |
1e97cce9 | 223 | static void do_verify(struct thread_data *td) |
ebac4655 | 224 | { |
53cdc686 | 225 | struct fio_file *f; |
36167d82 | 226 | struct io_u *io_u; |
3af6ef39 | 227 | int ret, i, min_events; |
e5b401d4 JA |
228 | |
229 | /* | |
230 | * sync io first and invalidate cache, to make sure we really | |
231 | * read from disk. | |
232 | */ | |
233 | for_each_file(td, f, i) { | |
b2fdda43 JA |
234 | if (fio_io_sync(td, f)) |
235 | break; | |
236 | if (file_invalidate_cache(td, f)) | |
237 | break; | |
e5b401d4 | 238 | } |
ebac4655 | 239 | |
b2fdda43 JA |
240 | if (td->error) |
241 | return; | |
242 | ||
ebac4655 JA |
243 | td_set_runstate(td, TD_VERIFYING); |
244 | ||
36167d82 JA |
245 | io_u = NULL; |
246 | while (!td->terminate) { | |
ebac4655 JA |
247 | io_u = __get_io_u(td); |
248 | if (!io_u) | |
249 | break; | |
250 | ||
36167d82 | 251 | if (runtime_exceeded(td, &io_u->start_time)) |
02bcaa8c | 252 | break; |
02bcaa8c | 253 | |
36167d82 | 254 | if (get_next_verify(td, io_u)) |
ebac4655 | 255 | break; |
ebac4655 | 256 | |
36167d82 | 257 | if (td_io_prep(td, io_u)) |
53cdc686 JA |
258 | break; |
259 | ||
36167d82 JA |
260 | requeue: |
261 | ret = td_io_queue(td, io_u); | |
53cdc686 | 262 | |
36167d82 JA |
263 | switch (ret) { |
264 | case FIO_Q_COMPLETED: | |
265 | if (io_u->error) | |
22819ec2 | 266 | ret = -io_u->error; |
36167d82 JA |
267 | if (io_u->xfer_buflen != io_u->resid && io_u->resid) { |
268 | int bytes = io_u->xfer_buflen - io_u->resid; | |
ebac4655 | 269 | |
36167d82 JA |
270 | io_u->xfer_buflen = io_u->resid; |
271 | io_u->xfer_buf += bytes; | |
272 | goto requeue; | |
273 | } | |
99784632 JA |
274 | ret = io_u_sync_complete(td, io_u, verify_io_u); |
275 | if (ret) | |
36167d82 | 276 | break; |
36167d82 JA |
277 | continue; |
278 | case FIO_Q_QUEUED: | |
279 | break; | |
755200a3 JA |
280 | case FIO_Q_BUSY: |
281 | requeue_io_u(td, &io_u); | |
282 | ret = td_io_commit(td); | |
283 | break; | |
36167d82 JA |
284 | default: |
285 | assert(ret < 0); | |
22819ec2 | 286 | td_verror(td, -ret); |
ebac4655 JA |
287 | break; |
288 | } | |
289 | ||
99784632 | 290 | if (ret < 0 || td->error) |
3af6ef39 JA |
291 | break; |
292 | ||
ebac4655 | 293 | /* |
3af6ef39 JA |
294 | * if we can queue more, do so. but check if there are |
295 | * completed io_u's first. | |
ebac4655 | 296 | */ |
97601024 | 297 | min_events = 0; |
e916b390 | 298 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
3af6ef39 | 299 | min_events = 1; |
3af6ef39 | 300 | |
e916b390 JA |
301 | if (td->cur_depth > td->iodepth_low) |
302 | min_events = td->cur_depth - td->iodepth_low; | |
303 | } | |
304 | ||
3af6ef39 JA |
305 | /* |
306 | * Reap required number of io units, if any, and do the | |
307 | * verification on them through the callback handler | |
308 | */ | |
97601024 | 309 | if (io_u_queued_complete(td, min_events, verify_io_u)) |
ebac4655 | 310 | break; |
36167d82 | 311 | } |
ebac4655 | 312 | |
36167d82 JA |
313 | if (io_u) |
314 | put_io_u(td, io_u); | |
ebac4655 JA |
315 | |
316 | if (td->cur_depth) | |
317 | cleanup_pending_aio(td); | |
318 | ||
319 | td_set_runstate(td, TD_RUNNING); | |
320 | } | |
321 | ||
b990b5c0 JA |
322 | /* |
323 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
324 | * manner. | |
325 | */ | |
326 | static void do_cpuio(struct thread_data *td) | |
327 | { | |
328 | struct timeval e; | |
329 | int split = 100 / td->cpuload; | |
330 | int i = 0; | |
331 | ||
332 | while (!td->terminate) { | |
02bcaa8c | 333 | fio_gettime(&e, NULL); |
b990b5c0 JA |
334 | |
335 | if (runtime_exceeded(td, &e)) | |
336 | break; | |
337 | ||
338 | if (!(i % split)) | |
339 | __usec_sleep(10000); | |
340 | else | |
341 | usec_sleep(td, 10000); | |
342 | ||
343 | i++; | |
344 | } | |
345 | } | |
346 | ||
32cd46a0 | 347 | /* |
906c8d75 | 348 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
349 | * and reaps them, checking for rate and errors along the way. |
350 | */ | |
ebac4655 JA |
351 | static void do_io(struct thread_data *td) |
352 | { | |
02bcaa8c | 353 | struct timeval s; |
ebac4655 | 354 | unsigned long usec; |
84585003 | 355 | int i, ret = 0; |
ebac4655 | 356 | |
5853e5a8 JA |
357 | td_set_runstate(td, TD_RUNNING); |
358 | ||
3951414d | 359 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) { |
97601024 JA |
360 | struct timeval comp_time; |
361 | long bytes_done = 0; | |
84585003 | 362 | int min_evts = 0; |
ebac4655 JA |
363 | struct io_u *io_u; |
364 | ||
365 | if (td->terminate) | |
366 | break; | |
367 | ||
3d7c391d | 368 | io_u = get_io_u(td); |
ebac4655 JA |
369 | if (!io_u) |
370 | break; | |
371 | ||
372 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
373 | |
374 | if (runtime_exceeded(td, &s)) { | |
375 | put_io_u(td, io_u); | |
376 | break; | |
377 | } | |
cec6b55d | 378 | requeue: |
45bee283 | 379 | ret = td_io_queue(td, io_u); |
36167d82 JA |
380 | |
381 | switch (ret) { | |
382 | case FIO_Q_COMPLETED: | |
383 | if (io_u->error) { | |
384 | ret = io_u->error; | |
385 | break; | |
386 | } | |
387 | if (io_u->xfer_buflen != io_u->resid && io_u->resid) { | |
388 | int bytes = io_u->xfer_buflen - io_u->resid; | |
389 | ||
cec6b55d | 390 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 391 | io_u->xfer_buf += bytes; |
cec6b55d | 392 | goto requeue; |
cec6b55d | 393 | } |
97601024 JA |
394 | fio_gettime(&comp_time, NULL); |
395 | bytes_done = io_u_sync_complete(td, io_u, NULL); | |
99784632 JA |
396 | if (bytes_done < 0) |
397 | ret = bytes_done; | |
36167d82 JA |
398 | break; |
399 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
400 | /* |
401 | * if the engine doesn't have a commit hook, | |
402 | * the io_u is really queued. if it does have such | |
403 | * a hook, it has to call io_u_queued() itself. | |
404 | */ | |
405 | if (td->io_ops->commit == NULL) | |
406 | io_u_queued(td, io_u); | |
36167d82 | 407 | break; |
755200a3 JA |
408 | case FIO_Q_BUSY: |
409 | requeue_io_u(td, &io_u); | |
410 | ret = td_io_commit(td); | |
411 | break; | |
36167d82 JA |
412 | default: |
413 | assert(ret < 0); | |
414 | put_io_u(td, io_u); | |
415 | break; | |
ebac4655 JA |
416 | } |
417 | ||
99784632 | 418 | if (ret < 0 || td->error) |
36167d82 JA |
419 | break; |
420 | ||
97601024 JA |
421 | /* |
422 | * See if we need to complete some commands | |
423 | */ | |
755200a3 | 424 | if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) { |
97601024 | 425 | min_evts = 0; |
e916b390 | 426 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
36167d82 | 427 | min_evts = 1; |
ebac4655 | 428 | |
e916b390 JA |
429 | if (td->cur_depth > td->iodepth_low) |
430 | min_evts = td->cur_depth - td->iodepth_low; | |
431 | } | |
432 | ||
97601024 JA |
433 | fio_gettime(&comp_time, NULL); |
434 | bytes_done = io_u_queued_complete(td, min_evts, NULL); | |
435 | if (bytes_done < 0) | |
36167d82 | 436 | break; |
ebac4655 JA |
437 | } |
438 | ||
97601024 JA |
439 | if (!bytes_done) |
440 | continue; | |
441 | ||
ebac4655 JA |
442 | /* |
443 | * the rate is batched for now, it should work for batches | |
444 | * of completions except the very first one which may look | |
445 | * a little bursty | |
446 | */ | |
97601024 | 447 | usec = utime_since(&s, &comp_time); |
ebac4655 | 448 | |
97601024 | 449 | rate_throttle(td, usec, bytes_done, td->ddir); |
ebac4655 | 450 | |
97601024 | 451 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 452 | if (exitall_on_terminate) |
6ce15a32 | 453 | terminate_threads(td->groupid, 0); |
fd841467 | 454 | td_verror(td, ENODATA); |
ebac4655 JA |
455 | break; |
456 | } | |
457 | ||
9c1f7434 JA |
458 | if (td->thinktime) { |
459 | unsigned long long b; | |
460 | ||
461 | b = td->io_blocks[0] + td->io_blocks[1]; | |
48097d5c JA |
462 | if (!(b % td->thinktime_blocks)) { |
463 | int left; | |
464 | ||
465 | if (td->thinktime_spin) | |
466 | __usec_sleep(td->thinktime_spin); | |
467 | ||
468 | left = td->thinktime - td->thinktime_spin; | |
469 | if (left) | |
470 | usec_sleep(td, left); | |
471 | } | |
9c1f7434 | 472 | } |
ebac4655 JA |
473 | } |
474 | ||
4d2413c6 | 475 | if (!td->error) { |
3d7c391d JA |
476 | struct fio_file *f; |
477 | ||
84585003 JA |
478 | if (td->cur_depth) |
479 | cleanup_pending_aio(td); | |
ebac4655 | 480 | |
84585003 JA |
481 | if (should_fsync(td) && td->end_fsync) { |
482 | td_set_runstate(td, TD_FSYNCING); | |
483 | for_each_file(td, f, i) | |
858a3d47 | 484 | fio_io_sync(td, f); |
84585003 | 485 | } |
5853e5a8 | 486 | } |
ebac4655 JA |
487 | } |
488 | ||
ebac4655 JA |
489 | static void cleanup_io_u(struct thread_data *td) |
490 | { | |
491 | struct list_head *entry, *n; | |
492 | struct io_u *io_u; | |
493 | ||
494 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
495 | io_u = list_entry(entry, struct io_u, list); | |
496 | ||
497 | list_del(&io_u->list); | |
498 | free(io_u); | |
499 | } | |
500 | ||
2f9ade3c | 501 | free_io_mem(td); |
ebac4655 JA |
502 | } |
503 | ||
6b9cea23 JA |
504 | /* |
505 | * "randomly" fill the buffer contents | |
506 | */ | |
66eeb296 | 507 | static void fill_rand_buf(struct io_u *io_u, int max_bs) |
6b9cea23 | 508 | { |
66eeb296 | 509 | int *ptr = io_u->buf; |
6b9cea23 JA |
510 | |
511 | while ((void *) ptr - io_u->buf < max_bs) { | |
512 | *ptr = rand() * 0x9e370001; | |
513 | ptr++; | |
514 | } | |
515 | } | |
516 | ||
ebac4655 JA |
517 | static int init_io_u(struct thread_data *td) |
518 | { | |
519 | struct io_u *io_u; | |
a00735e6 | 520 | unsigned int max_bs; |
ebac4655 JA |
521 | int i, max_units; |
522 | char *p; | |
523 | ||
2866c82d | 524 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
525 | return 0; |
526 | ||
2866c82d | 527 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
528 | max_units = 1; |
529 | else | |
530 | max_units = td->iodepth; | |
531 | ||
a00735e6 | 532 | max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]); |
74b025b0 JA |
533 | td->orig_buffer_size = max_bs * max_units; |
534 | ||
d0bdaf49 | 535 | if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE) |
56bb17f2 | 536 | td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1); |
74b025b0 | 537 | else |
29d610e1 | 538 | td->orig_buffer_size += page_mask; |
ebac4655 | 539 | |
2f9ade3c JA |
540 | if (allocate_io_mem(td)) |
541 | return 1; | |
ebac4655 | 542 | |
ebac4655 JA |
543 | p = ALIGN(td->orig_buffer); |
544 | for (i = 0; i < max_units; i++) { | |
545 | io_u = malloc(sizeof(*io_u)); | |
546 | memset(io_u, 0, sizeof(*io_u)); | |
547 | INIT_LIST_HEAD(&io_u->list); | |
548 | ||
a00735e6 | 549 | io_u->buf = p + max_bs * i; |
6b9cea23 | 550 | if (td_write(td) || td_rw(td)) |
a00735e6 | 551 | fill_rand_buf(io_u, max_bs); |
6b9cea23 | 552 | |
b1ff3403 | 553 | io_u->index = i; |
ebac4655 JA |
554 | list_add(&io_u->list, &td->io_u_freelist); |
555 | } | |
556 | ||
557 | return 0; | |
558 | } | |
559 | ||
da86774e JA |
560 | static int switch_ioscheduler(struct thread_data *td) |
561 | { | |
562 | char tmp[256], tmp2[128]; | |
563 | FILE *f; | |
564 | int ret; | |
565 | ||
f48b467c JA |
566 | if (td->io_ops->flags & FIO_CPUIO) |
567 | return 0; | |
568 | ||
da86774e JA |
569 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
570 | ||
571 | f = fopen(tmp, "r+"); | |
572 | if (!f) { | |
573 | td_verror(td, errno); | |
574 | return 1; | |
575 | } | |
576 | ||
577 | /* | |
578 | * Set io scheduler. | |
579 | */ | |
580 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
581 | if (ferror(f) || ret != 1) { | |
582 | td_verror(td, errno); | |
583 | fclose(f); | |
584 | return 1; | |
585 | } | |
586 | ||
587 | rewind(f); | |
588 | ||
589 | /* | |
590 | * Read back and check that the selected scheduler is now the default. | |
591 | */ | |
592 | ret = fread(tmp, 1, sizeof(tmp), f); | |
593 | if (ferror(f) || ret < 0) { | |
594 | td_verror(td, errno); | |
595 | fclose(f); | |
596 | return 1; | |
597 | } | |
598 | ||
599 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
600 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 601 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
602 | td_verror(td, EINVAL); |
603 | fclose(f); | |
604 | return 1; | |
605 | } | |
606 | ||
607 | fclose(f); | |
608 | return 0; | |
609 | } | |
610 | ||
ebac4655 JA |
611 | static void clear_io_state(struct thread_data *td) |
612 | { | |
53cdc686 JA |
613 | struct fio_file *f; |
614 | int i; | |
ebac4655 | 615 | |
079ad09b | 616 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 617 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 618 | td->zone_bytes = 0; |
ebac4655 | 619 | |
c1324df1 JA |
620 | td->last_was_sync = 0; |
621 | ||
53cdc686 | 622 | for_each_file(td, f, i) { |
c1324df1 JA |
623 | f->last_completed_pos = 0; |
624 | ||
53cdc686 JA |
625 | f->last_pos = 0; |
626 | if (td->io_ops->flags & FIO_SYNCIO) | |
627 | lseek(f->fd, SEEK_SET, 0); | |
628 | ||
629 | if (f->file_map) | |
630 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
631 | } | |
ebac4655 JA |
632 | } |
633 | ||
906c8d75 JA |
634 | /* |
635 | * Entry point for the thread based jobs. The process based jobs end up | |
636 | * here as well, after a little setup. | |
637 | */ | |
ebac4655 JA |
638 | static void *thread_main(void *data) |
639 | { | |
69008999 | 640 | unsigned long long runtime[2]; |
ebac4655 | 641 | struct thread_data *td = data; |
ebac4655 JA |
642 | |
643 | if (!td->use_thread) | |
644 | setsid(); | |
645 | ||
646 | td->pid = getpid(); | |
647 | ||
aea47d44 JA |
648 | INIT_LIST_HEAD(&td->io_u_freelist); |
649 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 650 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 JA |
651 | INIT_LIST_HEAD(&td->io_hist_list); |
652 | INIT_LIST_HEAD(&td->io_log_list); | |
653 | ||
ebac4655 JA |
654 | if (init_io_u(td)) |
655 | goto err; | |
656 | ||
657 | if (fio_setaffinity(td) == -1) { | |
658 | td_verror(td, errno); | |
659 | goto err; | |
660 | } | |
661 | ||
aea47d44 JA |
662 | if (init_iolog(td)) |
663 | goto err; | |
664 | ||
ebac4655 JA |
665 | if (td->ioprio) { |
666 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
667 | td_verror(td, errno); | |
668 | goto err; | |
669 | } | |
670 | } | |
671 | ||
1056eaad | 672 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
673 | td_verror(td, errno); |
674 | goto err; | |
675 | } | |
676 | ||
75154845 JA |
677 | if (init_random_state(td)) |
678 | goto err; | |
679 | ||
56f9498d JA |
680 | if (td->ioscheduler && switch_ioscheduler(td)) |
681 | goto err; | |
da86774e | 682 | |
75154845 | 683 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
684 | fio_sem_up(&startup_sem); |
685 | fio_sem_down(&td->mutex); | |
ebac4655 | 686 | |
53cdc686 | 687 | if (!td->create_serialize && setup_files(td)) |
ebac4655 | 688 | goto err; |
21972cde JA |
689 | if (open_files(td)) |
690 | goto err; | |
ebac4655 | 691 | |
7d6c5283 JA |
692 | /* |
693 | * Do this late, as some IO engines would like to have the | |
694 | * files setup prior to initializing structures. | |
695 | */ | |
696 | if (td_io_init(td)) | |
697 | goto err; | |
698 | ||
69cfd7e0 JA |
699 | if (td->exec_prerun) { |
700 | if (system(td->exec_prerun) < 0) | |
701 | goto err; | |
702 | } | |
4e0ba8af | 703 | |
69008999 | 704 | fio_gettime(&td->epoch, NULL); |
079ad09b | 705 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
706 | |
707 | runtime[0] = runtime[1] = 0; | |
ebac4655 | 708 | while (td->loops--) { |
02bcaa8c | 709 | fio_gettime(&td->start, NULL); |
079ad09b | 710 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 JA |
711 | |
712 | if (td->ratemin) | |
079ad09b | 713 | memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate)); |
ebac4655 JA |
714 | |
715 | clear_io_state(td); | |
716 | prune_io_piece_log(td); | |
717 | ||
2866c82d | 718 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
719 | do_cpuio(td); |
720 | else | |
721 | do_io(td); | |
ebac4655 | 722 | |
69008999 | 723 | runtime[td->ddir] += utime_since_now(&td->start); |
aea47d44 | 724 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
69008999 | 725 | runtime[td->ddir ^ 1] = runtime[td->ddir]; |
3d60d1ed | 726 | |
ebac4655 JA |
727 | if (td->error || td->terminate) |
728 | break; | |
729 | ||
730 | if (td->verify == VERIFY_NONE) | |
731 | continue; | |
732 | ||
733 | clear_io_state(td); | |
02bcaa8c | 734 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
735 | |
736 | do_verify(td); | |
737 | ||
69008999 | 738 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
739 | |
740 | if (td->error || td->terminate) | |
741 | break; | |
742 | } | |
743 | ||
36dff966 | 744 | update_rusage_stat(td); |
69008999 JA |
745 | fio_gettime(&td->end_time, NULL); |
746 | td->runtime[0] = runtime[0] / 1000; | |
747 | td->runtime[1] = runtime[1] / 1000; | |
748 | ||
079ad09b JA |
749 | if (td->ts.bw_log) |
750 | finish_log(td, td->ts.bw_log, "bw"); | |
751 | if (td->ts.slat_log) | |
752 | finish_log(td, td->ts.slat_log, "slat"); | |
753 | if (td->ts.clat_log) | |
754 | finish_log(td, td->ts.clat_log, "clat"); | |
076efc7c | 755 | if (td->write_iolog_file) |
843a7413 | 756 | write_iolog_close(td); |
69cfd7e0 JA |
757 | if (td->exec_postrun) { |
758 | if (system(td->exec_postrun) < 0) | |
759 | log_err("fio: postrun %s failed\n", td->exec_postrun); | |
760 | } | |
ebac4655 JA |
761 | |
762 | if (exitall_on_terminate) | |
6ce15a32 | 763 | terminate_threads(td->groupid, 0); |
ebac4655 JA |
764 | |
765 | err: | |
5bf13a5a JA |
766 | if (td->error) |
767 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
53cdc686 | 768 | close_files(td); |
2866c82d | 769 | close_ioengine(td); |
ebac4655 | 770 | cleanup_io_u(td); |
ebac4655 | 771 | td_set_runstate(td, TD_EXITED); |
43d76807 | 772 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
773 | } |
774 | ||
906c8d75 JA |
775 | /* |
776 | * We cannot pass the td data into a forked process, so attach the td and | |
777 | * pass it to the thread worker. | |
778 | */ | |
a6418147 | 779 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
780 | { |
781 | struct thread_data *td; | |
a6418147 | 782 | void *data, *ret; |
ebac4655 JA |
783 | |
784 | data = shmat(shmid, NULL, 0); | |
785 | if (data == (void *) -1) { | |
a6418147 JA |
786 | int __err = errno; |
787 | ||
ebac4655 | 788 | perror("shmat"); |
a6418147 | 789 | return __err; |
ebac4655 JA |
790 | } |
791 | ||
792 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 793 | ret = thread_main(td); |
ebac4655 | 794 | shmdt(data); |
43d76807 | 795 | return (int) (unsigned long) ret; |
ebac4655 JA |
796 | } |
797 | ||
906c8d75 JA |
798 | /* |
799 | * Run over the job map and reap the threads that have exited, if any. | |
800 | */ | |
ebac4655 JA |
801 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
802 | { | |
34572e28 | 803 | struct thread_data *td; |
fab6aa71 | 804 | int i, cputhreads, pending, status, ret; |
ebac4655 JA |
805 | |
806 | /* | |
807 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
808 | */ | |
4d2413c6 | 809 | pending = cputhreads = 0; |
34572e28 | 810 | for_each_td(td, i) { |
84585003 JA |
811 | /* |
812 | * ->io_ops is NULL for a thread that has closed its | |
813 | * io engine | |
814 | */ | |
815 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
816 | cputhreads++; |
817 | ||
fab6aa71 JA |
818 | if (td->runstate < TD_EXITED) { |
819 | /* | |
820 | * check if someone quit or got killed in an unusual way | |
821 | */ | |
822 | ret = waitpid(td->pid, &status, WNOHANG); | |
823 | if (ret < 0) | |
824 | perror("waitpid"); | |
825 | else if ((ret == td->pid) && WIFSIGNALED(status)) { | |
826 | int sig = WTERMSIG(status); | |
827 | ||
828 | log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); | |
829 | td_set_runstate(td, TD_REAPED); | |
830 | goto reaped; | |
831 | } | |
832 | } | |
833 | ||
4d2413c6 JA |
834 | if (td->runstate != TD_EXITED) { |
835 | if (td->runstate < TD_RUNNING) | |
836 | pending++; | |
837 | ||
ebac4655 | 838 | continue; |
4d2413c6 | 839 | } |
ebac4655 | 840 | |
437c9b71 JA |
841 | if (td->error) |
842 | exit_value++; | |
843 | ||
ebac4655 JA |
844 | td_set_runstate(td, TD_REAPED); |
845 | ||
846 | if (td->use_thread) { | |
847 | long ret; | |
848 | ||
849 | if (pthread_join(td->thread, (void *) &ret)) | |
850 | perror("thread_join"); | |
a6418147 JA |
851 | } else { |
852 | int status; | |
853 | ||
fab6aa71 JA |
854 | ret = waitpid(td->pid, &status, 0); |
855 | if (ret < 0) | |
856 | perror("waitpid"); | |
73170f19 | 857 | else if (WIFEXITED(status) && WEXITSTATUS(status)) { |
a6418147 JA |
858 | if (!exit_value) |
859 | exit_value++; | |
860 | } | |
861 | } | |
ebac4655 | 862 | |
fab6aa71 | 863 | reaped: |
ebac4655 JA |
864 | (*nr_running)--; |
865 | (*m_rate) -= td->ratemin; | |
866 | (*t_rate) -= td->rate; | |
867 | } | |
b990b5c0 | 868 | |
4d2413c6 | 869 | if (*nr_running == cputhreads && !pending) |
6ce15a32 | 870 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
871 | } |
872 | ||
906c8d75 JA |
873 | /* |
874 | * Main function for kicking off and reaping jobs, as needed. | |
875 | */ | |
ebac4655 JA |
876 | static void run_threads(void) |
877 | { | |
ebac4655 JA |
878 | struct thread_data *td; |
879 | unsigned long spent; | |
880 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 881 | |
2f9ade3c JA |
882 | if (fio_pin_memory()) |
883 | return; | |
ebac4655 | 884 | |
c6ae0a5b JA |
885 | if (!terse_output) { |
886 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
887 | fflush(stdout); | |
888 | } | |
c04f7ec3 | 889 | |
4efa970e JA |
890 | signal(SIGINT, sig_handler); |
891 | signal(SIGALRM, sig_handler); | |
892 | ||
ebac4655 JA |
893 | todo = thread_number; |
894 | nr_running = 0; | |
895 | nr_started = 0; | |
896 | m_rate = t_rate = 0; | |
897 | ||
34572e28 | 898 | for_each_td(td, i) { |
263e529f | 899 | print_status_init(td->thread_number - 1); |
ebac4655 | 900 | |
380cf265 JA |
901 | if (!td->create_serialize) { |
902 | init_disk_util(td); | |
ebac4655 | 903 | continue; |
380cf265 | 904 | } |
ebac4655 JA |
905 | |
906 | /* | |
907 | * do file setup here so it happens sequentially, | |
908 | * we don't want X number of threads getting their | |
909 | * client data interspersed on disk | |
910 | */ | |
53cdc686 | 911 | if (setup_files(td)) { |
5bf13a5a JA |
912 | exit_value++; |
913 | if (td->error) | |
914 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
ebac4655 JA |
915 | td_set_runstate(td, TD_REAPED); |
916 | todo--; | |
917 | } | |
380cf265 JA |
918 | |
919 | init_disk_util(td); | |
ebac4655 JA |
920 | } |
921 | ||
ebac4655 | 922 | while (todo) { |
75154845 JA |
923 | struct thread_data *map[MAX_JOBS]; |
924 | struct timeval this_start; | |
925 | int this_jobs = 0, left; | |
926 | ||
ebac4655 JA |
927 | /* |
928 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
929 | */ | |
34572e28 | 930 | for_each_td(td, i) { |
ebac4655 JA |
931 | if (td->runstate != TD_NOT_CREATED) |
932 | continue; | |
933 | ||
934 | /* | |
935 | * never got a chance to start, killed by other | |
936 | * thread for some reason | |
937 | */ | |
938 | if (td->terminate) { | |
939 | todo--; | |
940 | continue; | |
941 | } | |
942 | ||
943 | if (td->start_delay) { | |
263e529f | 944 | spent = mtime_since_genesis(); |
ebac4655 JA |
945 | |
946 | if (td->start_delay * 1000 > spent) | |
947 | continue; | |
948 | } | |
949 | ||
950 | if (td->stonewall && (nr_started || nr_running)) | |
951 | break; | |
952 | ||
75154845 JA |
953 | /* |
954 | * Set state to created. Thread will transition | |
955 | * to TD_INITIALIZED when it's done setting up. | |
956 | */ | |
ebac4655 | 957 | td_set_runstate(td, TD_CREATED); |
75154845 | 958 | map[this_jobs++] = td; |
bbfd6b00 | 959 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
960 | nr_started++; |
961 | ||
962 | if (td->use_thread) { | |
963 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
964 | perror("thread_create"); | |
965 | nr_started--; | |
966 | } | |
967 | } else { | |
968 | if (fork()) | |
bbfd6b00 | 969 | fio_sem_down(&startup_sem); |
ebac4655 | 970 | else { |
a6418147 JA |
971 | int ret = fork_main(shm_id, i); |
972 | ||
973 | exit(ret); | |
ebac4655 JA |
974 | } |
975 | } | |
976 | } | |
977 | ||
978 | /* | |
75154845 JA |
979 | * Wait for the started threads to transition to |
980 | * TD_INITIALIZED. | |
ebac4655 | 981 | */ |
02bcaa8c | 982 | fio_gettime(&this_start, NULL); |
75154845 | 983 | left = this_jobs; |
6ce15a32 | 984 | while (left && !fio_abort) { |
75154845 JA |
985 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
986 | break; | |
987 | ||
988 | usleep(100000); | |
989 | ||
990 | for (i = 0; i < this_jobs; i++) { | |
991 | td = map[i]; | |
992 | if (!td) | |
993 | continue; | |
b6f4d880 | 994 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
995 | map[i] = NULL; |
996 | left--; | |
b6f4d880 JA |
997 | } else if (td->runstate >= TD_EXITED) { |
998 | map[i] = NULL; | |
999 | left--; | |
1000 | todo--; | |
1001 | nr_running++; /* work-around... */ | |
75154845 JA |
1002 | } |
1003 | } | |
1004 | } | |
1005 | ||
1006 | if (left) { | |
3b70d7e5 | 1007 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1008 | for (i = 0; i < this_jobs; i++) { |
1009 | td = map[i]; | |
1010 | if (!td) | |
1011 | continue; | |
1012 | kill(td->pid, SIGTERM); | |
1013 | } | |
1014 | break; | |
1015 | } | |
1016 | ||
1017 | /* | |
b6f4d880 | 1018 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1019 | */ |
34572e28 | 1020 | for_each_td(td, i) { |
75154845 | 1021 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1022 | continue; |
1023 | ||
1024 | td_set_runstate(td, TD_RUNNING); | |
1025 | nr_running++; | |
1026 | nr_started--; | |
1027 | m_rate += td->ratemin; | |
1028 | t_rate += td->rate; | |
75154845 | 1029 | todo--; |
bbfd6b00 | 1030 | fio_sem_up(&td->mutex); |
ebac4655 JA |
1031 | } |
1032 | ||
1033 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1034 | ||
1035 | if (todo) | |
1036 | usleep(100000); | |
1037 | } | |
1038 | ||
1039 | while (nr_running) { | |
1040 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1041 | usleep(10000); | |
1042 | } | |
1043 | ||
1044 | update_io_ticks(); | |
2f9ade3c | 1045 | fio_unpin_memory(); |
ebac4655 JA |
1046 | } |
1047 | ||
ebac4655 JA |
1048 | int main(int argc, char *argv[]) |
1049 | { | |
29d610e1 JA |
1050 | long ps; |
1051 | ||
dbe1125e JA |
1052 | /* |
1053 | * We need locale for number printing, if it isn't set then just | |
1054 | * go with the US format. | |
1055 | */ | |
1056 | if (!getenv("LC_NUMERIC")) | |
1057 | setlocale(LC_NUMERIC, "en_US"); | |
1058 | ||
ebac4655 JA |
1059 | if (parse_options(argc, argv)) |
1060 | return 1; | |
1061 | ||
1062 | if (!thread_number) { | |
3b70d7e5 | 1063 | log_err("Nothing to do\n"); |
ebac4655 JA |
1064 | return 1; |
1065 | } | |
1066 | ||
29d610e1 JA |
1067 | ps = sysconf(_SC_PAGESIZE); |
1068 | if (ps < 0) { | |
1069 | log_err("Failed to get page size\n"); | |
1070 | return 1; | |
1071 | } | |
1072 | ||
1073 | page_mask = ps - 1; | |
1074 | ||
bb3884d8 JA |
1075 | if (write_bw_log) { |
1076 | setup_log(&agg_io_log[DDIR_READ]); | |
1077 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1078 | } | |
1079 | ||
ebac4655 JA |
1080 | disk_util_timer_arm(); |
1081 | ||
1082 | run_threads(); | |
6ce15a32 | 1083 | |
bb3884d8 | 1084 | if (!fio_abort) { |
6ce15a32 | 1085 | show_run_stats(); |
bb3884d8 JA |
1086 | if (write_bw_log) { |
1087 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1088 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1089 | } | |
1090 | } | |
ebac4655 | 1091 | |
437c9b71 | 1092 | return exit_value; |
ebac4655 | 1093 | } |