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) { |
49db69aa | 200 | if (io_u_queued_complete(td, 1, NULL) < 0) |
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 | ||
069c2918 JA |
251 | if (runtime_exceeded(td, &io_u->start_time)) { |
252 | put_io_u(td, io_u); | |
02bcaa8c | 253 | break; |
069c2918 | 254 | } |
02bcaa8c | 255 | |
069c2918 JA |
256 | if (get_next_verify(td, io_u)) { |
257 | put_io_u(td, io_u); | |
ebac4655 | 258 | break; |
069c2918 | 259 | } |
ebac4655 | 260 | |
069c2918 JA |
261 | if (td_io_prep(td, io_u)) { |
262 | put_io_u(td, io_u); | |
53cdc686 | 263 | break; |
069c2918 | 264 | } |
36167d82 JA |
265 | requeue: |
266 | ret = td_io_queue(td, io_u); | |
53cdc686 | 267 | |
36167d82 JA |
268 | switch (ret) { |
269 | case FIO_Q_COMPLETED: | |
270 | if (io_u->error) | |
22819ec2 | 271 | ret = -io_u->error; |
36167d82 JA |
272 | if (io_u->xfer_buflen != io_u->resid && io_u->resid) { |
273 | int bytes = io_u->xfer_buflen - io_u->resid; | |
ebac4655 | 274 | |
36167d82 JA |
275 | io_u->xfer_buflen = io_u->resid; |
276 | io_u->xfer_buf += bytes; | |
277 | goto requeue; | |
278 | } | |
99784632 | 279 | ret = io_u_sync_complete(td, io_u, verify_io_u); |
49db69aa | 280 | if (ret < 0) |
36167d82 | 281 | break; |
36167d82 JA |
282 | continue; |
283 | case FIO_Q_QUEUED: | |
284 | break; | |
755200a3 JA |
285 | case FIO_Q_BUSY: |
286 | requeue_io_u(td, &io_u); | |
287 | ret = td_io_commit(td); | |
288 | break; | |
36167d82 JA |
289 | default: |
290 | assert(ret < 0); | |
22819ec2 | 291 | td_verror(td, -ret); |
ebac4655 JA |
292 | break; |
293 | } | |
294 | ||
99784632 | 295 | if (ret < 0 || td->error) |
3af6ef39 JA |
296 | break; |
297 | ||
ebac4655 | 298 | /* |
3af6ef39 JA |
299 | * if we can queue more, do so. but check if there are |
300 | * completed io_u's first. | |
ebac4655 | 301 | */ |
97601024 | 302 | min_events = 0; |
e916b390 | 303 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
3af6ef39 | 304 | min_events = 1; |
3af6ef39 | 305 | |
e916b390 JA |
306 | if (td->cur_depth > td->iodepth_low) |
307 | min_events = td->cur_depth - td->iodepth_low; | |
308 | } | |
309 | ||
3af6ef39 JA |
310 | /* |
311 | * Reap required number of io units, if any, and do the | |
312 | * verification on them through the callback handler | |
313 | */ | |
49db69aa | 314 | if (io_u_queued_complete(td, min_events, verify_io_u) < 0) |
ebac4655 | 315 | break; |
36167d82 | 316 | } |
ebac4655 | 317 | |
ebac4655 JA |
318 | if (td->cur_depth) |
319 | cleanup_pending_aio(td); | |
320 | ||
321 | td_set_runstate(td, TD_RUNNING); | |
322 | } | |
323 | ||
b990b5c0 JA |
324 | /* |
325 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
326 | * manner. | |
327 | */ | |
328 | static void do_cpuio(struct thread_data *td) | |
329 | { | |
330 | struct timeval e; | |
331 | int split = 100 / td->cpuload; | |
332 | int i = 0; | |
333 | ||
334 | while (!td->terminate) { | |
02bcaa8c | 335 | fio_gettime(&e, NULL); |
b990b5c0 JA |
336 | |
337 | if (runtime_exceeded(td, &e)) | |
338 | break; | |
339 | ||
340 | if (!(i % split)) | |
341 | __usec_sleep(10000); | |
342 | else | |
343 | usec_sleep(td, 10000); | |
344 | ||
345 | i++; | |
346 | } | |
347 | } | |
348 | ||
32cd46a0 | 349 | /* |
906c8d75 | 350 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
351 | * and reaps them, checking for rate and errors along the way. |
352 | */ | |
ebac4655 JA |
353 | static void do_io(struct thread_data *td) |
354 | { | |
02bcaa8c | 355 | struct timeval s; |
ebac4655 | 356 | unsigned long usec; |
84585003 | 357 | int i, ret = 0; |
ebac4655 | 358 | |
5853e5a8 JA |
359 | td_set_runstate(td, TD_RUNNING); |
360 | ||
3951414d | 361 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) { |
97601024 JA |
362 | struct timeval comp_time; |
363 | long bytes_done = 0; | |
84585003 | 364 | int min_evts = 0; |
ebac4655 JA |
365 | struct io_u *io_u; |
366 | ||
367 | if (td->terminate) | |
368 | break; | |
369 | ||
3d7c391d | 370 | io_u = get_io_u(td); |
ebac4655 JA |
371 | if (!io_u) |
372 | break; | |
373 | ||
374 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
375 | |
376 | if (runtime_exceeded(td, &s)) { | |
377 | put_io_u(td, io_u); | |
378 | break; | |
379 | } | |
cec6b55d | 380 | requeue: |
45bee283 | 381 | ret = td_io_queue(td, io_u); |
36167d82 JA |
382 | |
383 | switch (ret) { | |
384 | case FIO_Q_COMPLETED: | |
385 | if (io_u->error) { | |
386 | ret = io_u->error; | |
387 | break; | |
388 | } | |
389 | if (io_u->xfer_buflen != io_u->resid && io_u->resid) { | |
390 | int bytes = io_u->xfer_buflen - io_u->resid; | |
391 | ||
cec6b55d | 392 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 393 | io_u->xfer_buf += bytes; |
cec6b55d | 394 | goto requeue; |
cec6b55d | 395 | } |
97601024 JA |
396 | fio_gettime(&comp_time, NULL); |
397 | bytes_done = io_u_sync_complete(td, io_u, NULL); | |
99784632 JA |
398 | if (bytes_done < 0) |
399 | ret = bytes_done; | |
36167d82 JA |
400 | break; |
401 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
402 | /* |
403 | * if the engine doesn't have a commit hook, | |
404 | * the io_u is really queued. if it does have such | |
405 | * a hook, it has to call io_u_queued() itself. | |
406 | */ | |
407 | if (td->io_ops->commit == NULL) | |
408 | io_u_queued(td, io_u); | |
36167d82 | 409 | break; |
755200a3 JA |
410 | case FIO_Q_BUSY: |
411 | requeue_io_u(td, &io_u); | |
412 | ret = td_io_commit(td); | |
413 | break; | |
36167d82 JA |
414 | default: |
415 | assert(ret < 0); | |
416 | put_io_u(td, io_u); | |
417 | break; | |
ebac4655 JA |
418 | } |
419 | ||
99784632 | 420 | if (ret < 0 || td->error) |
36167d82 JA |
421 | break; |
422 | ||
97601024 JA |
423 | /* |
424 | * See if we need to complete some commands | |
425 | */ | |
755200a3 | 426 | if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) { |
97601024 | 427 | min_evts = 0; |
e916b390 | 428 | if (queue_full(td) || ret == FIO_Q_BUSY) { |
36167d82 | 429 | min_evts = 1; |
ebac4655 | 430 | |
e916b390 JA |
431 | if (td->cur_depth > td->iodepth_low) |
432 | min_evts = td->cur_depth - td->iodepth_low; | |
433 | } | |
434 | ||
97601024 JA |
435 | fio_gettime(&comp_time, NULL); |
436 | bytes_done = io_u_queued_complete(td, min_evts, NULL); | |
437 | if (bytes_done < 0) | |
36167d82 | 438 | break; |
ebac4655 JA |
439 | } |
440 | ||
97601024 JA |
441 | if (!bytes_done) |
442 | continue; | |
443 | ||
ebac4655 JA |
444 | /* |
445 | * the rate is batched for now, it should work for batches | |
446 | * of completions except the very first one which may look | |
447 | * a little bursty | |
448 | */ | |
97601024 | 449 | usec = utime_since(&s, &comp_time); |
ebac4655 | 450 | |
97601024 | 451 | rate_throttle(td, usec, bytes_done, td->ddir); |
ebac4655 | 452 | |
97601024 | 453 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 454 | if (exitall_on_terminate) |
6ce15a32 | 455 | terminate_threads(td->groupid, 0); |
fd841467 | 456 | td_verror(td, ENODATA); |
ebac4655 JA |
457 | break; |
458 | } | |
459 | ||
9c1f7434 JA |
460 | if (td->thinktime) { |
461 | unsigned long long b; | |
462 | ||
463 | b = td->io_blocks[0] + td->io_blocks[1]; | |
48097d5c JA |
464 | if (!(b % td->thinktime_blocks)) { |
465 | int left; | |
466 | ||
467 | if (td->thinktime_spin) | |
468 | __usec_sleep(td->thinktime_spin); | |
469 | ||
470 | left = td->thinktime - td->thinktime_spin; | |
471 | if (left) | |
472 | usec_sleep(td, left); | |
473 | } | |
9c1f7434 | 474 | } |
ebac4655 JA |
475 | } |
476 | ||
4d2413c6 | 477 | if (!td->error) { |
3d7c391d JA |
478 | struct fio_file *f; |
479 | ||
84585003 JA |
480 | if (td->cur_depth) |
481 | cleanup_pending_aio(td); | |
ebac4655 | 482 | |
84585003 JA |
483 | if (should_fsync(td) && td->end_fsync) { |
484 | td_set_runstate(td, TD_FSYNCING); | |
485 | for_each_file(td, f, i) | |
858a3d47 | 486 | fio_io_sync(td, f); |
84585003 | 487 | } |
5853e5a8 | 488 | } |
ebac4655 JA |
489 | } |
490 | ||
ebac4655 JA |
491 | static void cleanup_io_u(struct thread_data *td) |
492 | { | |
493 | struct list_head *entry, *n; | |
494 | struct io_u *io_u; | |
495 | ||
496 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
497 | io_u = list_entry(entry, struct io_u, list); | |
498 | ||
499 | list_del(&io_u->list); | |
500 | free(io_u); | |
501 | } | |
502 | ||
2f9ade3c | 503 | free_io_mem(td); |
ebac4655 JA |
504 | } |
505 | ||
6b9cea23 JA |
506 | /* |
507 | * "randomly" fill the buffer contents | |
508 | */ | |
66eeb296 | 509 | static void fill_rand_buf(struct io_u *io_u, int max_bs) |
6b9cea23 | 510 | { |
66eeb296 | 511 | int *ptr = io_u->buf; |
6b9cea23 JA |
512 | |
513 | while ((void *) ptr - io_u->buf < max_bs) { | |
514 | *ptr = rand() * 0x9e370001; | |
515 | ptr++; | |
516 | } | |
517 | } | |
518 | ||
ebac4655 JA |
519 | static int init_io_u(struct thread_data *td) |
520 | { | |
521 | struct io_u *io_u; | |
a00735e6 | 522 | unsigned int max_bs; |
ebac4655 JA |
523 | int i, max_units; |
524 | char *p; | |
525 | ||
2866c82d | 526 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
527 | return 0; |
528 | ||
2866c82d | 529 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
530 | max_units = 1; |
531 | else | |
532 | max_units = td->iodepth; | |
533 | ||
a00735e6 | 534 | max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]); |
74b025b0 JA |
535 | td->orig_buffer_size = max_bs * max_units; |
536 | ||
d0bdaf49 | 537 | if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE) |
56bb17f2 | 538 | td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1); |
74b025b0 | 539 | else |
29d610e1 | 540 | td->orig_buffer_size += page_mask; |
ebac4655 | 541 | |
2f9ade3c JA |
542 | if (allocate_io_mem(td)) |
543 | return 1; | |
ebac4655 | 544 | |
ebac4655 JA |
545 | p = ALIGN(td->orig_buffer); |
546 | for (i = 0; i < max_units; i++) { | |
547 | io_u = malloc(sizeof(*io_u)); | |
548 | memset(io_u, 0, sizeof(*io_u)); | |
549 | INIT_LIST_HEAD(&io_u->list); | |
550 | ||
a00735e6 | 551 | io_u->buf = p + max_bs * i; |
6b9cea23 | 552 | if (td_write(td) || td_rw(td)) |
a00735e6 | 553 | fill_rand_buf(io_u, max_bs); |
6b9cea23 | 554 | |
b1ff3403 | 555 | io_u->index = i; |
ebac4655 JA |
556 | list_add(&io_u->list, &td->io_u_freelist); |
557 | } | |
558 | ||
559 | return 0; | |
560 | } | |
561 | ||
da86774e JA |
562 | static int switch_ioscheduler(struct thread_data *td) |
563 | { | |
564 | char tmp[256], tmp2[128]; | |
565 | FILE *f; | |
566 | int ret; | |
567 | ||
f48b467c JA |
568 | if (td->io_ops->flags & FIO_CPUIO) |
569 | return 0; | |
570 | ||
da86774e JA |
571 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
572 | ||
573 | f = fopen(tmp, "r+"); | |
574 | if (!f) { | |
575 | td_verror(td, errno); | |
576 | return 1; | |
577 | } | |
578 | ||
579 | /* | |
580 | * Set io scheduler. | |
581 | */ | |
582 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
583 | if (ferror(f) || ret != 1) { | |
584 | td_verror(td, errno); | |
585 | fclose(f); | |
586 | return 1; | |
587 | } | |
588 | ||
589 | rewind(f); | |
590 | ||
591 | /* | |
592 | * Read back and check that the selected scheduler is now the default. | |
593 | */ | |
594 | ret = fread(tmp, 1, sizeof(tmp), f); | |
595 | if (ferror(f) || ret < 0) { | |
596 | td_verror(td, errno); | |
597 | fclose(f); | |
598 | return 1; | |
599 | } | |
600 | ||
601 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
602 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 603 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
604 | td_verror(td, EINVAL); |
605 | fclose(f); | |
606 | return 1; | |
607 | } | |
608 | ||
609 | fclose(f); | |
610 | return 0; | |
611 | } | |
612 | ||
ebac4655 JA |
613 | static void clear_io_state(struct thread_data *td) |
614 | { | |
53cdc686 JA |
615 | struct fio_file *f; |
616 | int i; | |
ebac4655 | 617 | |
079ad09b | 618 | td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0; |
ebac4655 | 619 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 620 | td->zone_bytes = 0; |
ebac4655 | 621 | |
c1324df1 JA |
622 | td->last_was_sync = 0; |
623 | ||
53cdc686 | 624 | for_each_file(td, f, i) { |
c1324df1 JA |
625 | f->last_completed_pos = 0; |
626 | ||
53cdc686 JA |
627 | f->last_pos = 0; |
628 | if (td->io_ops->flags & FIO_SYNCIO) | |
629 | lseek(f->fd, SEEK_SET, 0); | |
630 | ||
631 | if (f->file_map) | |
632 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
633 | } | |
ebac4655 JA |
634 | } |
635 | ||
906c8d75 JA |
636 | /* |
637 | * Entry point for the thread based jobs. The process based jobs end up | |
638 | * here as well, after a little setup. | |
639 | */ | |
ebac4655 JA |
640 | static void *thread_main(void *data) |
641 | { | |
69008999 | 642 | unsigned long long runtime[2]; |
ebac4655 | 643 | struct thread_data *td = data; |
ebac4655 JA |
644 | |
645 | if (!td->use_thread) | |
646 | setsid(); | |
647 | ||
648 | td->pid = getpid(); | |
649 | ||
aea47d44 JA |
650 | INIT_LIST_HEAD(&td->io_u_freelist); |
651 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 652 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 JA |
653 | INIT_LIST_HEAD(&td->io_hist_list); |
654 | INIT_LIST_HEAD(&td->io_log_list); | |
655 | ||
ebac4655 JA |
656 | if (init_io_u(td)) |
657 | goto err; | |
658 | ||
659 | if (fio_setaffinity(td) == -1) { | |
660 | td_verror(td, errno); | |
661 | goto err; | |
662 | } | |
663 | ||
aea47d44 JA |
664 | if (init_iolog(td)) |
665 | goto err; | |
666 | ||
ebac4655 JA |
667 | if (td->ioprio) { |
668 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
669 | td_verror(td, errno); | |
670 | goto err; | |
671 | } | |
672 | } | |
673 | ||
1056eaad | 674 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
675 | td_verror(td, errno); |
676 | goto err; | |
677 | } | |
678 | ||
75154845 JA |
679 | if (init_random_state(td)) |
680 | goto err; | |
681 | ||
56f9498d JA |
682 | if (td->ioscheduler && switch_ioscheduler(td)) |
683 | goto err; | |
da86774e | 684 | |
75154845 | 685 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
686 | fio_sem_up(&startup_sem); |
687 | fio_sem_down(&td->mutex); | |
ebac4655 | 688 | |
53cdc686 | 689 | if (!td->create_serialize && setup_files(td)) |
ebac4655 | 690 | goto err; |
21972cde JA |
691 | if (open_files(td)) |
692 | goto err; | |
ebac4655 | 693 | |
7d6c5283 JA |
694 | /* |
695 | * Do this late, as some IO engines would like to have the | |
696 | * files setup prior to initializing structures. | |
697 | */ | |
698 | if (td_io_init(td)) | |
699 | goto err; | |
700 | ||
69cfd7e0 JA |
701 | if (td->exec_prerun) { |
702 | if (system(td->exec_prerun) < 0) | |
703 | goto err; | |
704 | } | |
4e0ba8af | 705 | |
69008999 | 706 | fio_gettime(&td->epoch, NULL); |
079ad09b | 707 | getrusage(RUSAGE_SELF, &td->ts.ru_start); |
69008999 JA |
708 | |
709 | runtime[0] = runtime[1] = 0; | |
ebac4655 | 710 | while (td->loops--) { |
02bcaa8c | 711 | fio_gettime(&td->start, NULL); |
079ad09b | 712 | memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start)); |
ebac4655 JA |
713 | |
714 | if (td->ratemin) | |
079ad09b | 715 | memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate)); |
ebac4655 JA |
716 | |
717 | clear_io_state(td); | |
718 | prune_io_piece_log(td); | |
719 | ||
2866c82d | 720 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
721 | do_cpuio(td); |
722 | else | |
723 | do_io(td); | |
ebac4655 | 724 | |
69008999 | 725 | runtime[td->ddir] += utime_since_now(&td->start); |
aea47d44 | 726 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
69008999 | 727 | runtime[td->ddir ^ 1] = runtime[td->ddir]; |
3d60d1ed | 728 | |
ebac4655 JA |
729 | if (td->error || td->terminate) |
730 | break; | |
731 | ||
732 | if (td->verify == VERIFY_NONE) | |
733 | continue; | |
734 | ||
735 | clear_io_state(td); | |
02bcaa8c | 736 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
737 | |
738 | do_verify(td); | |
739 | ||
69008999 | 740 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
741 | |
742 | if (td->error || td->terminate) | |
743 | break; | |
744 | } | |
745 | ||
36dff966 | 746 | update_rusage_stat(td); |
69008999 JA |
747 | fio_gettime(&td->end_time, NULL); |
748 | td->runtime[0] = runtime[0] / 1000; | |
749 | td->runtime[1] = runtime[1] / 1000; | |
750 | ||
079ad09b JA |
751 | if (td->ts.bw_log) |
752 | finish_log(td, td->ts.bw_log, "bw"); | |
753 | if (td->ts.slat_log) | |
754 | finish_log(td, td->ts.slat_log, "slat"); | |
755 | if (td->ts.clat_log) | |
756 | finish_log(td, td->ts.clat_log, "clat"); | |
076efc7c | 757 | if (td->write_iolog_file) |
843a7413 | 758 | write_iolog_close(td); |
69cfd7e0 JA |
759 | if (td->exec_postrun) { |
760 | if (system(td->exec_postrun) < 0) | |
761 | log_err("fio: postrun %s failed\n", td->exec_postrun); | |
762 | } | |
ebac4655 JA |
763 | |
764 | if (exitall_on_terminate) | |
6ce15a32 | 765 | terminate_threads(td->groupid, 0); |
ebac4655 JA |
766 | |
767 | err: | |
5bf13a5a JA |
768 | if (td->error) |
769 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
53cdc686 | 770 | close_files(td); |
2866c82d | 771 | close_ioengine(td); |
ebac4655 | 772 | cleanup_io_u(td); |
ebac4655 | 773 | td_set_runstate(td, TD_EXITED); |
43d76807 | 774 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
775 | } |
776 | ||
906c8d75 JA |
777 | /* |
778 | * We cannot pass the td data into a forked process, so attach the td and | |
779 | * pass it to the thread worker. | |
780 | */ | |
a6418147 | 781 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
782 | { |
783 | struct thread_data *td; | |
a6418147 | 784 | void *data, *ret; |
ebac4655 JA |
785 | |
786 | data = shmat(shmid, NULL, 0); | |
787 | if (data == (void *) -1) { | |
a6418147 JA |
788 | int __err = errno; |
789 | ||
ebac4655 | 790 | perror("shmat"); |
a6418147 | 791 | return __err; |
ebac4655 JA |
792 | } |
793 | ||
794 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 795 | ret = thread_main(td); |
ebac4655 | 796 | shmdt(data); |
43d76807 | 797 | return (int) (unsigned long) ret; |
ebac4655 JA |
798 | } |
799 | ||
906c8d75 JA |
800 | /* |
801 | * Run over the job map and reap the threads that have exited, if any. | |
802 | */ | |
ebac4655 JA |
803 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
804 | { | |
34572e28 | 805 | struct thread_data *td; |
fab6aa71 | 806 | int i, cputhreads, pending, status, ret; |
ebac4655 JA |
807 | |
808 | /* | |
809 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
810 | */ | |
4d2413c6 | 811 | pending = cputhreads = 0; |
34572e28 | 812 | for_each_td(td, i) { |
84585003 JA |
813 | /* |
814 | * ->io_ops is NULL for a thread that has closed its | |
815 | * io engine | |
816 | */ | |
817 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
818 | cputhreads++; |
819 | ||
fab6aa71 JA |
820 | if (td->runstate < TD_EXITED) { |
821 | /* | |
822 | * check if someone quit or got killed in an unusual way | |
823 | */ | |
824 | ret = waitpid(td->pid, &status, WNOHANG); | |
825 | if (ret < 0) | |
826 | perror("waitpid"); | |
827 | else if ((ret == td->pid) && WIFSIGNALED(status)) { | |
828 | int sig = WTERMSIG(status); | |
829 | ||
830 | log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); | |
831 | td_set_runstate(td, TD_REAPED); | |
832 | goto reaped; | |
833 | } | |
834 | } | |
835 | ||
4d2413c6 JA |
836 | if (td->runstate != TD_EXITED) { |
837 | if (td->runstate < TD_RUNNING) | |
838 | pending++; | |
839 | ||
ebac4655 | 840 | continue; |
4d2413c6 | 841 | } |
ebac4655 | 842 | |
437c9b71 JA |
843 | if (td->error) |
844 | exit_value++; | |
845 | ||
ebac4655 JA |
846 | td_set_runstate(td, TD_REAPED); |
847 | ||
848 | if (td->use_thread) { | |
849 | long ret; | |
850 | ||
851 | if (pthread_join(td->thread, (void *) &ret)) | |
852 | perror("thread_join"); | |
a6418147 JA |
853 | } else { |
854 | int status; | |
855 | ||
fab6aa71 JA |
856 | ret = waitpid(td->pid, &status, 0); |
857 | if (ret < 0) | |
858 | perror("waitpid"); | |
73170f19 | 859 | else if (WIFEXITED(status) && WEXITSTATUS(status)) { |
a6418147 JA |
860 | if (!exit_value) |
861 | exit_value++; | |
862 | } | |
863 | } | |
ebac4655 | 864 | |
fab6aa71 | 865 | reaped: |
ebac4655 JA |
866 | (*nr_running)--; |
867 | (*m_rate) -= td->ratemin; | |
868 | (*t_rate) -= td->rate; | |
869 | } | |
b990b5c0 | 870 | |
4d2413c6 | 871 | if (*nr_running == cputhreads && !pending) |
6ce15a32 | 872 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
873 | } |
874 | ||
906c8d75 JA |
875 | /* |
876 | * Main function for kicking off and reaping jobs, as needed. | |
877 | */ | |
ebac4655 JA |
878 | static void run_threads(void) |
879 | { | |
ebac4655 JA |
880 | struct thread_data *td; |
881 | unsigned long spent; | |
882 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 883 | |
2f9ade3c JA |
884 | if (fio_pin_memory()) |
885 | return; | |
ebac4655 | 886 | |
c6ae0a5b JA |
887 | if (!terse_output) { |
888 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
889 | fflush(stdout); | |
890 | } | |
c04f7ec3 | 891 | |
4efa970e JA |
892 | signal(SIGINT, sig_handler); |
893 | signal(SIGALRM, sig_handler); | |
894 | ||
ebac4655 JA |
895 | todo = thread_number; |
896 | nr_running = 0; | |
897 | nr_started = 0; | |
898 | m_rate = t_rate = 0; | |
899 | ||
34572e28 | 900 | for_each_td(td, i) { |
263e529f | 901 | print_status_init(td->thread_number - 1); |
ebac4655 | 902 | |
380cf265 JA |
903 | if (!td->create_serialize) { |
904 | init_disk_util(td); | |
ebac4655 | 905 | continue; |
380cf265 | 906 | } |
ebac4655 JA |
907 | |
908 | /* | |
909 | * do file setup here so it happens sequentially, | |
910 | * we don't want X number of threads getting their | |
911 | * client data interspersed on disk | |
912 | */ | |
53cdc686 | 913 | if (setup_files(td)) { |
5bf13a5a JA |
914 | exit_value++; |
915 | if (td->error) | |
916 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
ebac4655 JA |
917 | td_set_runstate(td, TD_REAPED); |
918 | todo--; | |
919 | } | |
380cf265 JA |
920 | |
921 | init_disk_util(td); | |
ebac4655 JA |
922 | } |
923 | ||
ebac4655 | 924 | while (todo) { |
75154845 JA |
925 | struct thread_data *map[MAX_JOBS]; |
926 | struct timeval this_start; | |
927 | int this_jobs = 0, left; | |
928 | ||
ebac4655 JA |
929 | /* |
930 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
931 | */ | |
34572e28 | 932 | for_each_td(td, i) { |
ebac4655 JA |
933 | if (td->runstate != TD_NOT_CREATED) |
934 | continue; | |
935 | ||
936 | /* | |
937 | * never got a chance to start, killed by other | |
938 | * thread for some reason | |
939 | */ | |
940 | if (td->terminate) { | |
941 | todo--; | |
942 | continue; | |
943 | } | |
944 | ||
945 | if (td->start_delay) { | |
263e529f | 946 | spent = mtime_since_genesis(); |
ebac4655 JA |
947 | |
948 | if (td->start_delay * 1000 > spent) | |
949 | continue; | |
950 | } | |
951 | ||
952 | if (td->stonewall && (nr_started || nr_running)) | |
953 | break; | |
954 | ||
75154845 JA |
955 | /* |
956 | * Set state to created. Thread will transition | |
957 | * to TD_INITIALIZED when it's done setting up. | |
958 | */ | |
ebac4655 | 959 | td_set_runstate(td, TD_CREATED); |
75154845 | 960 | map[this_jobs++] = td; |
bbfd6b00 | 961 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
962 | nr_started++; |
963 | ||
964 | if (td->use_thread) { | |
965 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
966 | perror("thread_create"); | |
967 | nr_started--; | |
968 | } | |
969 | } else { | |
970 | if (fork()) | |
bbfd6b00 | 971 | fio_sem_down(&startup_sem); |
ebac4655 | 972 | else { |
a6418147 JA |
973 | int ret = fork_main(shm_id, i); |
974 | ||
975 | exit(ret); | |
ebac4655 JA |
976 | } |
977 | } | |
978 | } | |
979 | ||
980 | /* | |
75154845 JA |
981 | * Wait for the started threads to transition to |
982 | * TD_INITIALIZED. | |
ebac4655 | 983 | */ |
02bcaa8c | 984 | fio_gettime(&this_start, NULL); |
75154845 | 985 | left = this_jobs; |
6ce15a32 | 986 | while (left && !fio_abort) { |
75154845 JA |
987 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
988 | break; | |
989 | ||
990 | usleep(100000); | |
991 | ||
992 | for (i = 0; i < this_jobs; i++) { | |
993 | td = map[i]; | |
994 | if (!td) | |
995 | continue; | |
b6f4d880 | 996 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
997 | map[i] = NULL; |
998 | left--; | |
b6f4d880 JA |
999 | } else if (td->runstate >= TD_EXITED) { |
1000 | map[i] = NULL; | |
1001 | left--; | |
1002 | todo--; | |
1003 | nr_running++; /* work-around... */ | |
75154845 JA |
1004 | } |
1005 | } | |
1006 | } | |
1007 | ||
1008 | if (left) { | |
3b70d7e5 | 1009 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1010 | for (i = 0; i < this_jobs; i++) { |
1011 | td = map[i]; | |
1012 | if (!td) | |
1013 | continue; | |
1014 | kill(td->pid, SIGTERM); | |
1015 | } | |
1016 | break; | |
1017 | } | |
1018 | ||
1019 | /* | |
b6f4d880 | 1020 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1021 | */ |
34572e28 | 1022 | for_each_td(td, i) { |
75154845 | 1023 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1024 | continue; |
1025 | ||
1026 | td_set_runstate(td, TD_RUNNING); | |
1027 | nr_running++; | |
1028 | nr_started--; | |
1029 | m_rate += td->ratemin; | |
1030 | t_rate += td->rate; | |
75154845 | 1031 | todo--; |
bbfd6b00 | 1032 | fio_sem_up(&td->mutex); |
ebac4655 JA |
1033 | } |
1034 | ||
1035 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1036 | ||
1037 | if (todo) | |
1038 | usleep(100000); | |
1039 | } | |
1040 | ||
1041 | while (nr_running) { | |
1042 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1043 | usleep(10000); | |
1044 | } | |
1045 | ||
1046 | update_io_ticks(); | |
2f9ade3c | 1047 | fio_unpin_memory(); |
ebac4655 JA |
1048 | } |
1049 | ||
ebac4655 JA |
1050 | int main(int argc, char *argv[]) |
1051 | { | |
29d610e1 JA |
1052 | long ps; |
1053 | ||
dbe1125e JA |
1054 | /* |
1055 | * We need locale for number printing, if it isn't set then just | |
1056 | * go with the US format. | |
1057 | */ | |
1058 | if (!getenv("LC_NUMERIC")) | |
1059 | setlocale(LC_NUMERIC, "en_US"); | |
1060 | ||
ebac4655 JA |
1061 | if (parse_options(argc, argv)) |
1062 | return 1; | |
1063 | ||
1064 | if (!thread_number) { | |
3b70d7e5 | 1065 | log_err("Nothing to do\n"); |
ebac4655 JA |
1066 | return 1; |
1067 | } | |
1068 | ||
29d610e1 JA |
1069 | ps = sysconf(_SC_PAGESIZE); |
1070 | if (ps < 0) { | |
1071 | log_err("Failed to get page size\n"); | |
1072 | return 1; | |
1073 | } | |
1074 | ||
1075 | page_mask = ps - 1; | |
1076 | ||
bb3884d8 JA |
1077 | if (write_bw_log) { |
1078 | setup_log(&agg_io_log[DDIR_READ]); | |
1079 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1080 | } | |
1081 | ||
ebac4655 JA |
1082 | disk_util_timer_arm(); |
1083 | ||
1084 | run_threads(); | |
6ce15a32 | 1085 | |
bb3884d8 | 1086 | if (!fio_abort) { |
6ce15a32 | 1087 | show_run_stats(); |
bb3884d8 JA |
1088 | if (write_bw_log) { |
1089 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1090 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1091 | } | |
1092 | } | |
ebac4655 | 1093 | |
437c9b71 | 1094 | return exit_value; |
ebac4655 | 1095 | } |