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> |
ebac4655 JA |
30 | #include <sys/stat.h> |
31 | #include <sys/wait.h> | |
32 | #include <sys/ipc.h> | |
33 | #include <sys/shm.h> | |
ebac4655 JA |
34 | #include <sys/mman.h> |
35 | ||
36 | #include "fio.h" | |
37 | #include "os.h" | |
38 | ||
29d610e1 JA |
39 | static unsigned long page_mask; |
40 | #define ALIGN(buf) \ | |
41 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
42 | |
43 | int groupid = 0; | |
44 | int thread_number = 0; | |
ebac4655 | 45 | int shm_id = 0; |
53cdc686 | 46 | int temp_stall_ts; |
ebac4655 | 47 | |
bbfd6b00 | 48 | static volatile int startup_sem; |
6ce15a32 | 49 | static volatile int fio_abort; |
437c9b71 | 50 | static int exit_value; |
ebac4655 | 51 | |
bb3884d8 JA |
52 | struct io_log *agg_io_log[2]; |
53 | ||
ebac4655 | 54 | #define TERMINATE_ALL (-1) |
75154845 | 55 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 56 | |
6ce15a32 JA |
57 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
58 | { | |
59 | td->runstate = runstate; | |
60 | } | |
61 | ||
62 | static void terminate_threads(int group_id, int forced_kill) | |
ebac4655 | 63 | { |
34572e28 | 64 | struct thread_data *td; |
ebac4655 JA |
65 | int i; |
66 | ||
34572e28 | 67 | for_each_td(td, i) { |
ebac4655 JA |
68 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
69 | td->terminate = 1; | |
70 | td->start_delay = 0; | |
6ce15a32 JA |
71 | if (forced_kill) |
72 | td_set_runstate(td, TD_EXITED); | |
ebac4655 JA |
73 | } |
74 | } | |
75 | } | |
76 | ||
77 | static void sig_handler(int sig) | |
78 | { | |
79 | switch (sig) { | |
80 | case SIGALRM: | |
81 | update_io_ticks(); | |
82 | disk_util_timer_arm(); | |
83 | print_thread_status(); | |
84 | break; | |
85 | default: | |
6ce15a32 | 86 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 87 | fflush(stdout); |
6ce15a32 | 88 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
89 | break; |
90 | } | |
91 | } | |
92 | ||
906c8d75 JA |
93 | /* |
94 | * Check if we are above the minimum rate given. | |
95 | */ | |
ebac4655 JA |
96 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
97 | { | |
98 | unsigned long spent; | |
99 | unsigned long rate; | |
100 | int ddir = td->ddir; | |
101 | ||
102 | /* | |
103 | * allow a 2 second settle period in the beginning | |
104 | */ | |
105 | if (mtime_since(&td->start, now) < 2000) | |
106 | return 0; | |
107 | ||
108 | /* | |
109 | * if rate blocks is set, sample is running | |
110 | */ | |
111 | if (td->rate_bytes) { | |
112 | spent = mtime_since(&td->lastrate, now); | |
113 | if (spent < td->ratecycle) | |
114 | return 0; | |
115 | ||
116 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
117 | if (rate < td->ratemin) { | |
1e97cce9 | 118 | fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
119 | return 1; |
120 | } | |
121 | } | |
122 | ||
123 | td->rate_bytes = td->this_io_bytes[ddir]; | |
124 | memcpy(&td->lastrate, now, sizeof(*now)); | |
125 | return 0; | |
126 | } | |
127 | ||
128 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
129 | { | |
130 | if (!td->timeout) | |
131 | return 0; | |
132 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
133 | return 1; | |
134 | ||
135 | return 0; | |
136 | } | |
137 | ||
53cdc686 JA |
138 | static struct fio_file *get_next_file(struct thread_data *td) |
139 | { | |
0ab8db89 | 140 | unsigned int old_next_file = td->next_file; |
b2a15192 JA |
141 | struct fio_file *f; |
142 | ||
143 | do { | |
144 | f = &td->files[td->next_file]; | |
145 | ||
146 | td->next_file++; | |
147 | if (td->next_file >= td->nr_files) | |
148 | td->next_file = 0; | |
149 | ||
150 | if (f->fd != -1) | |
151 | break; | |
53cdc686 | 152 | |
b2a15192 JA |
153 | f = NULL; |
154 | } while (td->next_file != old_next_file); | |
53cdc686 JA |
155 | |
156 | return f; | |
157 | } | |
158 | ||
906c8d75 JA |
159 | /* |
160 | * When job exits, we can cancel the in-flight IO if we are using async | |
161 | * io. Attempt to do so. | |
162 | */ | |
ebac4655 JA |
163 | static void cleanup_pending_aio(struct thread_data *td) |
164 | { | |
165 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
166 | struct list_head *entry, *n; | |
167 | struct io_completion_data icd; | |
168 | struct io_u *io_u; | |
169 | int r; | |
170 | ||
171 | /* | |
172 | * get immediately available events, if any | |
173 | */ | |
45bee283 | 174 | r = td_io_getevents(td, 0, td->cur_depth, &ts); |
ebac4655 JA |
175 | if (r > 0) { |
176 | icd.nr = r; | |
177 | ios_completed(td, &icd); | |
178 | } | |
179 | ||
180 | /* | |
181 | * now cancel remaining active events | |
182 | */ | |
2866c82d | 183 | if (td->io_ops->cancel) { |
ebac4655 JA |
184 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
185 | io_u = list_entry(entry, struct io_u, list); | |
186 | ||
2866c82d | 187 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
188 | if (!r) |
189 | put_io_u(td, io_u); | |
190 | } | |
191 | } | |
192 | ||
193 | if (td->cur_depth) { | |
45bee283 | 194 | r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL); |
ebac4655 JA |
195 | if (r > 0) { |
196 | icd.nr = r; | |
197 | ios_completed(td, &icd); | |
198 | } | |
199 | } | |
200 | } | |
201 | ||
858a3d47 JA |
202 | /* |
203 | * Helper to handle the final sync of a file. Works just like the normal | |
204 | * io path, just does everything sync. | |
205 | */ | |
206 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
207 | { | |
208 | struct io_u *io_u = __get_io_u(td); | |
209 | struct io_completion_data icd; | |
210 | int ret; | |
211 | ||
212 | if (!io_u) | |
213 | return 1; | |
214 | ||
215 | io_u->ddir = DDIR_SYNC; | |
216 | io_u->file = f; | |
217 | ||
218 | if (td_io_prep(td, io_u)) { | |
219 | put_io_u(td, io_u); | |
220 | return 1; | |
221 | } | |
222 | ||
223 | ret = td_io_queue(td, io_u); | |
224 | if (ret) { | |
353a7e0e | 225 | td_verror(td, io_u->error); |
858a3d47 | 226 | put_io_u(td, io_u); |
858a3d47 JA |
227 | return 1; |
228 | } | |
229 | ||
230 | ret = td_io_getevents(td, 1, td->cur_depth, NULL); | |
231 | if (ret < 0) { | |
353a7e0e | 232 | td_verror(td, ret); |
858a3d47 JA |
233 | return 1; |
234 | } | |
235 | ||
236 | icd.nr = ret; | |
237 | ios_completed(td, &icd); | |
238 | if (icd.error) { | |
239 | td_verror(td, icd.error); | |
240 | return 1; | |
241 | } | |
242 | ||
243 | return 0; | |
244 | } | |
245 | ||
906c8d75 JA |
246 | /* |
247 | * The main verify engine. Runs over the writes we previusly submitted, | |
248 | * reads the blocks back in, and checks the crc/md5 of the data. | |
249 | */ | |
1e97cce9 | 250 | static void do_verify(struct thread_data *td) |
ebac4655 | 251 | { |
ebac4655 JA |
252 | struct io_u *io_u, *v_io_u = NULL; |
253 | struct io_completion_data icd; | |
53cdc686 | 254 | struct fio_file *f; |
e5b401d4 JA |
255 | int ret, i; |
256 | ||
257 | /* | |
258 | * sync io first and invalidate cache, to make sure we really | |
259 | * read from disk. | |
260 | */ | |
261 | for_each_file(td, f, i) { | |
858a3d47 | 262 | fio_io_sync(td, f); |
e5b401d4 JA |
263 | file_invalidate_cache(td, f); |
264 | } | |
ebac4655 JA |
265 | |
266 | td_set_runstate(td, TD_VERIFYING); | |
267 | ||
268 | do { | |
269 | if (td->terminate) | |
270 | break; | |
271 | ||
ebac4655 JA |
272 | io_u = __get_io_u(td); |
273 | if (!io_u) | |
274 | break; | |
275 | ||
02bcaa8c JA |
276 | if (runtime_exceeded(td, &io_u->start_time)) { |
277 | put_io_u(td, io_u); | |
278 | break; | |
279 | } | |
280 | ||
aea47d44 | 281 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
282 | put_io_u(td, io_u); |
283 | break; | |
284 | } | |
285 | ||
53cdc686 JA |
286 | f = get_next_file(td); |
287 | if (!f) | |
288 | break; | |
289 | ||
290 | io_u->file = f; | |
291 | ||
aea47d44 | 292 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
293 | put_io_u(td, io_u); |
294 | break; | |
295 | } | |
296 | ||
45bee283 | 297 | ret = td_io_queue(td, io_u); |
ebac4655 | 298 | if (ret) { |
353a7e0e | 299 | td_verror(td, io_u->error); |
ebac4655 | 300 | put_io_u(td, io_u); |
ebac4655 JA |
301 | break; |
302 | } | |
303 | ||
304 | /* | |
305 | * we have one pending to verify, do that while | |
306 | * we are doing io on the next one | |
307 | */ | |
308 | if (do_io_u_verify(td, &v_io_u)) | |
309 | break; | |
310 | ||
45bee283 | 311 | ret = td_io_getevents(td, 1, 1, NULL); |
ebac4655 JA |
312 | if (ret != 1) { |
313 | if (ret < 0) | |
314 | td_verror(td, ret); | |
315 | break; | |
316 | } | |
317 | ||
2866c82d | 318 | v_io_u = td->io_ops->event(td, 0); |
ebac4655 JA |
319 | icd.nr = 1; |
320 | icd.error = 0; | |
02bcaa8c | 321 | fio_gettime(&icd.time, NULL); |
ebac4655 JA |
322 | io_completed(td, v_io_u, &icd); |
323 | ||
324 | if (icd.error) { | |
325 | td_verror(td, icd.error); | |
326 | put_io_u(td, v_io_u); | |
327 | v_io_u = NULL; | |
328 | break; | |
329 | } | |
330 | ||
ebac4655 JA |
331 | /* |
332 | * if we can't submit more io, we need to verify now | |
333 | */ | |
334 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
335 | break; | |
336 | ||
337 | } while (1); | |
338 | ||
339 | do_io_u_verify(td, &v_io_u); | |
340 | ||
341 | if (td->cur_depth) | |
342 | cleanup_pending_aio(td); | |
343 | ||
344 | td_set_runstate(td, TD_RUNNING); | |
345 | } | |
346 | ||
b990b5c0 JA |
347 | /* |
348 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
349 | * manner. | |
350 | */ | |
351 | static void do_cpuio(struct thread_data *td) | |
352 | { | |
353 | struct timeval e; | |
354 | int split = 100 / td->cpuload; | |
355 | int i = 0; | |
356 | ||
357 | while (!td->terminate) { | |
02bcaa8c | 358 | fio_gettime(&e, NULL); |
b990b5c0 JA |
359 | |
360 | if (runtime_exceeded(td, &e)) | |
361 | break; | |
362 | ||
363 | if (!(i % split)) | |
364 | __usec_sleep(10000); | |
365 | else | |
366 | usec_sleep(td, 10000); | |
367 | ||
368 | i++; | |
369 | } | |
370 | } | |
371 | ||
32cd46a0 | 372 | /* |
906c8d75 | 373 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
374 | * and reaps them, checking for rate and errors along the way. |
375 | */ | |
ebac4655 JA |
376 | static void do_io(struct thread_data *td) |
377 | { | |
378 | struct io_completion_data icd; | |
02bcaa8c | 379 | struct timeval s; |
ebac4655 | 380 | unsigned long usec; |
53cdc686 | 381 | struct fio_file *f; |
84585003 | 382 | int i, ret = 0; |
ebac4655 | 383 | |
5853e5a8 JA |
384 | td_set_runstate(td, TD_RUNNING); |
385 | ||
3951414d | 386 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) { |
ebac4655 | 387 | struct timespec *timeout; |
84585003 | 388 | int min_evts = 0; |
ebac4655 JA |
389 | struct io_u *io_u; |
390 | ||
391 | if (td->terminate) | |
392 | break; | |
393 | ||
53cdc686 JA |
394 | f = get_next_file(td); |
395 | if (!f) | |
396 | break; | |
397 | ||
398 | io_u = get_io_u(td, f); | |
ebac4655 JA |
399 | if (!io_u) |
400 | break; | |
401 | ||
402 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
cec6b55d | 403 | requeue: |
45bee283 | 404 | ret = td_io_queue(td, io_u); |
ebac4655 | 405 | if (ret) { |
cec6b55d JA |
406 | if (ret > 0 && (io_u->xfer_buflen != io_u->resid) && |
407 | io_u->resid) { | |
408 | /* | |
409 | * short read/write. requeue. | |
410 | */ | |
411 | io_u->xfer_buflen = io_u->resid; | |
412 | io_u->xfer_buf += ret; | |
413 | goto requeue; | |
414 | } else { | |
cec6b55d JA |
415 | put_io_u(td, io_u); |
416 | break; | |
417 | } | |
ebac4655 JA |
418 | } |
419 | ||
420 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
421 | ||
422 | if (td->cur_depth < td->iodepth) { | |
8d7ad8d9 JA |
423 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; |
424 | ||
ebac4655 JA |
425 | timeout = &ts; |
426 | min_evts = 0; | |
427 | } else { | |
428 | timeout = NULL; | |
429 | min_evts = 1; | |
430 | } | |
431 | ||
45bee283 | 432 | ret = td_io_getevents(td, min_evts, td->cur_depth, timeout); |
ebac4655 | 433 | if (ret < 0) { |
87dc1ab1 | 434 | td_verror(td, ret); |
ebac4655 JA |
435 | break; |
436 | } else if (!ret) | |
437 | continue; | |
438 | ||
439 | icd.nr = ret; | |
440 | ios_completed(td, &icd); | |
441 | if (icd.error) { | |
442 | td_verror(td, icd.error); | |
443 | break; | |
444 | } | |
445 | ||
446 | /* | |
447 | * the rate is batched for now, it should work for batches | |
448 | * of completions except the very first one which may look | |
449 | * a little bursty | |
450 | */ | |
02bcaa8c | 451 | usec = utime_since(&s, &icd.time); |
ebac4655 | 452 | |
a00735e6 | 453 | rate_throttle(td, usec, icd.bytes_done[td->ddir], td->ddir); |
ebac4655 | 454 | |
02bcaa8c | 455 | if (check_min_rate(td, &icd.time)) { |
98aa62d8 | 456 | if (exitall_on_terminate) |
6ce15a32 | 457 | terminate_threads(td->groupid, 0); |
fd841467 | 458 | td_verror(td, ENODATA); |
ebac4655 JA |
459 | break; |
460 | } | |
461 | ||
02bcaa8c | 462 | if (runtime_exceeded(td, &icd.time)) |
ebac4655 JA |
463 | break; |
464 | ||
9c1f7434 JA |
465 | if (td->thinktime) { |
466 | unsigned long long b; | |
467 | ||
468 | b = td->io_blocks[0] + td->io_blocks[1]; | |
48097d5c JA |
469 | if (!(b % td->thinktime_blocks)) { |
470 | int left; | |
471 | ||
472 | if (td->thinktime_spin) | |
473 | __usec_sleep(td->thinktime_spin); | |
474 | ||
475 | left = td->thinktime - td->thinktime_spin; | |
476 | if (left) | |
477 | usec_sleep(td, left); | |
478 | } | |
9c1f7434 | 479 | } |
ebac4655 JA |
480 | } |
481 | ||
4d2413c6 | 482 | if (!td->error) { |
84585003 JA |
483 | if (td->cur_depth) |
484 | cleanup_pending_aio(td); | |
ebac4655 | 485 | |
84585003 JA |
486 | if (should_fsync(td) && td->end_fsync) { |
487 | td_set_runstate(td, TD_FSYNCING); | |
488 | for_each_file(td, f, i) | |
858a3d47 | 489 | fio_io_sync(td, f); |
84585003 | 490 | } |
5853e5a8 | 491 | } |
ebac4655 JA |
492 | } |
493 | ||
ebac4655 JA |
494 | static void cleanup_io_u(struct thread_data *td) |
495 | { | |
496 | struct list_head *entry, *n; | |
497 | struct io_u *io_u; | |
498 | ||
499 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
500 | io_u = list_entry(entry, struct io_u, list); | |
501 | ||
502 | list_del(&io_u->list); | |
503 | free(io_u); | |
504 | } | |
505 | ||
2f9ade3c | 506 | free_io_mem(td); |
ebac4655 JA |
507 | } |
508 | ||
6b9cea23 JA |
509 | /* |
510 | * "randomly" fill the buffer contents | |
511 | */ | |
66eeb296 | 512 | static void fill_rand_buf(struct io_u *io_u, int max_bs) |
6b9cea23 | 513 | { |
66eeb296 | 514 | int *ptr = io_u->buf; |
6b9cea23 JA |
515 | |
516 | while ((void *) ptr - io_u->buf < max_bs) { | |
517 | *ptr = rand() * 0x9e370001; | |
518 | ptr++; | |
519 | } | |
520 | } | |
521 | ||
ebac4655 JA |
522 | static int init_io_u(struct thread_data *td) |
523 | { | |
524 | struct io_u *io_u; | |
a00735e6 | 525 | unsigned int max_bs; |
ebac4655 JA |
526 | int i, max_units; |
527 | char *p; | |
528 | ||
2866c82d | 529 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
530 | return 0; |
531 | ||
2866c82d | 532 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
533 | max_units = 1; |
534 | else | |
535 | max_units = td->iodepth; | |
536 | ||
a00735e6 | 537 | max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]); |
74b025b0 JA |
538 | td->orig_buffer_size = max_bs * max_units; |
539 | ||
d0bdaf49 | 540 | if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE) |
56bb17f2 | 541 | td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1); |
74b025b0 | 542 | else |
29d610e1 | 543 | td->orig_buffer_size += page_mask; |
ebac4655 | 544 | |
2f9ade3c JA |
545 | if (allocate_io_mem(td)) |
546 | return 1; | |
ebac4655 | 547 | |
ebac4655 JA |
548 | p = ALIGN(td->orig_buffer); |
549 | for (i = 0; i < max_units; i++) { | |
550 | io_u = malloc(sizeof(*io_u)); | |
551 | memset(io_u, 0, sizeof(*io_u)); | |
552 | INIT_LIST_HEAD(&io_u->list); | |
553 | ||
a00735e6 | 554 | io_u->buf = p + max_bs * i; |
6b9cea23 | 555 | if (td_write(td) || td_rw(td)) |
a00735e6 | 556 | fill_rand_buf(io_u, max_bs); |
6b9cea23 | 557 | |
b1ff3403 | 558 | io_u->index = i; |
ebac4655 JA |
559 | list_add(&io_u->list, &td->io_u_freelist); |
560 | } | |
561 | ||
562 | return 0; | |
563 | } | |
564 | ||
da86774e JA |
565 | static int switch_ioscheduler(struct thread_data *td) |
566 | { | |
567 | char tmp[256], tmp2[128]; | |
568 | FILE *f; | |
569 | int ret; | |
570 | ||
f48b467c JA |
571 | if (td->io_ops->flags & FIO_CPUIO) |
572 | return 0; | |
573 | ||
da86774e JA |
574 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
575 | ||
576 | f = fopen(tmp, "r+"); | |
577 | if (!f) { | |
578 | td_verror(td, errno); | |
579 | return 1; | |
580 | } | |
581 | ||
582 | /* | |
583 | * Set io scheduler. | |
584 | */ | |
585 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
586 | if (ferror(f) || ret != 1) { | |
587 | td_verror(td, errno); | |
588 | fclose(f); | |
589 | return 1; | |
590 | } | |
591 | ||
592 | rewind(f); | |
593 | ||
594 | /* | |
595 | * Read back and check that the selected scheduler is now the default. | |
596 | */ | |
597 | ret = fread(tmp, 1, sizeof(tmp), f); | |
598 | if (ferror(f) || ret < 0) { | |
599 | td_verror(td, errno); | |
600 | fclose(f); | |
601 | return 1; | |
602 | } | |
603 | ||
604 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
605 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 606 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
607 | td_verror(td, EINVAL); |
608 | fclose(f); | |
609 | return 1; | |
610 | } | |
611 | ||
612 | fclose(f); | |
613 | return 0; | |
614 | } | |
615 | ||
ebac4655 JA |
616 | static void clear_io_state(struct thread_data *td) |
617 | { | |
53cdc686 JA |
618 | struct fio_file *f; |
619 | int i; | |
ebac4655 | 620 | |
ebac4655 JA |
621 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
622 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 623 | td->zone_bytes = 0; |
ebac4655 | 624 | |
53cdc686 JA |
625 | for_each_file(td, f, i) { |
626 | f->last_pos = 0; | |
627 | if (td->io_ops->flags & FIO_SYNCIO) | |
628 | lseek(f->fd, SEEK_SET, 0); | |
629 | ||
630 | if (f->file_map) | |
631 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
632 | } | |
ebac4655 JA |
633 | } |
634 | ||
906c8d75 JA |
635 | /* |
636 | * Entry point for the thread based jobs. The process based jobs end up | |
637 | * here as well, after a little setup. | |
638 | */ | |
ebac4655 JA |
639 | static void *thread_main(void *data) |
640 | { | |
69008999 | 641 | unsigned long long runtime[2]; |
ebac4655 | 642 | struct thread_data *td = data; |
ebac4655 JA |
643 | |
644 | if (!td->use_thread) | |
645 | setsid(); | |
646 | ||
647 | td->pid = getpid(); | |
648 | ||
aea47d44 JA |
649 | INIT_LIST_HEAD(&td->io_u_freelist); |
650 | INIT_LIST_HEAD(&td->io_u_busylist); | |
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); |
36dff966 | 705 | getrusage(RUSAGE_SELF, &td->ru_start); |
69008999 JA |
706 | |
707 | runtime[0] = runtime[1] = 0; | |
ebac4655 | 708 | while (td->loops--) { |
02bcaa8c | 709 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
710 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); |
711 | ||
712 | if (td->ratemin) | |
713 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
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 | ||
ebac4655 JA |
749 | if (td->bw_log) |
750 | finish_log(td, td->bw_log, "bw"); | |
751 | if (td->slat_log) | |
752 | finish_log(td, td->slat_log, "slat"); | |
753 | if (td->clat_log) | |
754 | finish_log(td, td->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); |
a6418147 | 772 | return (void *) 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); |
a6418147 | 795 | return (int) 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 | } |