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; |
ebac4655 | 50 | |
bb3884d8 JA |
51 | struct io_log *agg_io_log[2]; |
52 | ||
ebac4655 | 53 | #define TERMINATE_ALL (-1) |
75154845 | 54 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 55 | |
6ce15a32 JA |
56 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
57 | { | |
58 | td->runstate = runstate; | |
59 | } | |
60 | ||
61 | static void terminate_threads(int group_id, int forced_kill) | |
ebac4655 | 62 | { |
34572e28 | 63 | struct thread_data *td; |
ebac4655 JA |
64 | int i; |
65 | ||
34572e28 | 66 | for_each_td(td, i) { |
ebac4655 JA |
67 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
68 | td->terminate = 1; | |
69 | td->start_delay = 0; | |
6ce15a32 JA |
70 | if (forced_kill) |
71 | td_set_runstate(td, TD_EXITED); | |
ebac4655 JA |
72 | } |
73 | } | |
74 | } | |
75 | ||
76 | static void sig_handler(int sig) | |
77 | { | |
78 | switch (sig) { | |
79 | case SIGALRM: | |
80 | update_io_ticks(); | |
81 | disk_util_timer_arm(); | |
82 | print_thread_status(); | |
83 | break; | |
84 | default: | |
6ce15a32 | 85 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 86 | fflush(stdout); |
6ce15a32 | 87 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
88 | break; |
89 | } | |
90 | } | |
91 | ||
906c8d75 JA |
92 | /* |
93 | * Check if we are above the minimum rate given. | |
94 | */ | |
ebac4655 JA |
95 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
96 | { | |
97 | unsigned long spent; | |
98 | unsigned long rate; | |
99 | int ddir = td->ddir; | |
100 | ||
101 | /* | |
102 | * allow a 2 second settle period in the beginning | |
103 | */ | |
104 | if (mtime_since(&td->start, now) < 2000) | |
105 | return 0; | |
106 | ||
107 | /* | |
108 | * if rate blocks is set, sample is running | |
109 | */ | |
110 | if (td->rate_bytes) { | |
111 | spent = mtime_since(&td->lastrate, now); | |
112 | if (spent < td->ratecycle) | |
113 | return 0; | |
114 | ||
115 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
116 | if (rate < td->ratemin) { | |
1e97cce9 | 117 | fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
118 | return 1; |
119 | } | |
120 | } | |
121 | ||
122 | td->rate_bytes = td->this_io_bytes[ddir]; | |
123 | memcpy(&td->lastrate, now, sizeof(*now)); | |
124 | return 0; | |
125 | } | |
126 | ||
127 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
128 | { | |
129 | if (!td->timeout) | |
130 | return 0; | |
131 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
132 | return 1; | |
133 | ||
134 | return 0; | |
135 | } | |
136 | ||
53cdc686 JA |
137 | static struct fio_file *get_next_file(struct thread_data *td) |
138 | { | |
0ab8db89 | 139 | unsigned int old_next_file = td->next_file; |
b2a15192 JA |
140 | struct fio_file *f; |
141 | ||
142 | do { | |
143 | f = &td->files[td->next_file]; | |
144 | ||
145 | td->next_file++; | |
146 | if (td->next_file >= td->nr_files) | |
147 | td->next_file = 0; | |
148 | ||
149 | if (f->fd != -1) | |
150 | break; | |
53cdc686 | 151 | |
b2a15192 JA |
152 | f = NULL; |
153 | } while (td->next_file != old_next_file); | |
53cdc686 JA |
154 | |
155 | return f; | |
156 | } | |
157 | ||
906c8d75 JA |
158 | /* |
159 | * When job exits, we can cancel the in-flight IO if we are using async | |
160 | * io. Attempt to do so. | |
161 | */ | |
ebac4655 JA |
162 | static void cleanup_pending_aio(struct thread_data *td) |
163 | { | |
164 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
165 | struct list_head *entry, *n; | |
166 | struct io_completion_data icd; | |
167 | struct io_u *io_u; | |
168 | int r; | |
169 | ||
170 | /* | |
171 | * get immediately available events, if any | |
172 | */ | |
45bee283 | 173 | r = td_io_getevents(td, 0, td->cur_depth, &ts); |
ebac4655 JA |
174 | if (r > 0) { |
175 | icd.nr = r; | |
176 | ios_completed(td, &icd); | |
177 | } | |
178 | ||
179 | /* | |
180 | * now cancel remaining active events | |
181 | */ | |
2866c82d | 182 | if (td->io_ops->cancel) { |
ebac4655 JA |
183 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
184 | io_u = list_entry(entry, struct io_u, list); | |
185 | ||
2866c82d | 186 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
187 | if (!r) |
188 | put_io_u(td, io_u); | |
189 | } | |
190 | } | |
191 | ||
192 | if (td->cur_depth) { | |
45bee283 | 193 | r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL); |
ebac4655 JA |
194 | if (r > 0) { |
195 | icd.nr = r; | |
196 | ios_completed(td, &icd); | |
197 | } | |
198 | } | |
199 | } | |
200 | ||
858a3d47 JA |
201 | /* |
202 | * Helper to handle the final sync of a file. Works just like the normal | |
203 | * io path, just does everything sync. | |
204 | */ | |
205 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
206 | { | |
207 | struct io_u *io_u = __get_io_u(td); | |
208 | struct io_completion_data icd; | |
209 | int ret; | |
210 | ||
211 | if (!io_u) | |
212 | return 1; | |
213 | ||
214 | io_u->ddir = DDIR_SYNC; | |
215 | io_u->file = f; | |
216 | ||
217 | if (td_io_prep(td, io_u)) { | |
218 | put_io_u(td, io_u); | |
219 | return 1; | |
220 | } | |
221 | ||
222 | ret = td_io_queue(td, io_u); | |
223 | if (ret) { | |
353a7e0e | 224 | td_verror(td, io_u->error); |
858a3d47 | 225 | put_io_u(td, io_u); |
858a3d47 JA |
226 | return 1; |
227 | } | |
228 | ||
229 | ret = td_io_getevents(td, 1, td->cur_depth, NULL); | |
230 | if (ret < 0) { | |
353a7e0e | 231 | td_verror(td, ret); |
858a3d47 JA |
232 | return 1; |
233 | } | |
234 | ||
235 | icd.nr = ret; | |
236 | ios_completed(td, &icd); | |
237 | if (icd.error) { | |
238 | td_verror(td, icd.error); | |
239 | return 1; | |
240 | } | |
241 | ||
242 | return 0; | |
243 | } | |
244 | ||
906c8d75 JA |
245 | /* |
246 | * The main verify engine. Runs over the writes we previusly submitted, | |
247 | * reads the blocks back in, and checks the crc/md5 of the data. | |
248 | */ | |
1e97cce9 | 249 | static void do_verify(struct thread_data *td) |
ebac4655 | 250 | { |
ebac4655 JA |
251 | struct io_u *io_u, *v_io_u = NULL; |
252 | struct io_completion_data icd; | |
53cdc686 | 253 | struct fio_file *f; |
e5b401d4 JA |
254 | int ret, i; |
255 | ||
256 | /* | |
257 | * sync io first and invalidate cache, to make sure we really | |
258 | * read from disk. | |
259 | */ | |
260 | for_each_file(td, f, i) { | |
858a3d47 | 261 | fio_io_sync(td, f); |
e5b401d4 JA |
262 | file_invalidate_cache(td, f); |
263 | } | |
ebac4655 JA |
264 | |
265 | td_set_runstate(td, TD_VERIFYING); | |
266 | ||
267 | do { | |
268 | if (td->terminate) | |
269 | break; | |
270 | ||
ebac4655 JA |
271 | io_u = __get_io_u(td); |
272 | if (!io_u) | |
273 | break; | |
274 | ||
02bcaa8c JA |
275 | if (runtime_exceeded(td, &io_u->start_time)) { |
276 | put_io_u(td, io_u); | |
277 | break; | |
278 | } | |
279 | ||
aea47d44 | 280 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
281 | put_io_u(td, io_u); |
282 | break; | |
283 | } | |
284 | ||
53cdc686 JA |
285 | f = get_next_file(td); |
286 | if (!f) | |
287 | break; | |
288 | ||
289 | io_u->file = f; | |
290 | ||
aea47d44 | 291 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
292 | put_io_u(td, io_u); |
293 | break; | |
294 | } | |
295 | ||
45bee283 | 296 | ret = td_io_queue(td, io_u); |
ebac4655 | 297 | if (ret) { |
353a7e0e | 298 | td_verror(td, io_u->error); |
ebac4655 | 299 | put_io_u(td, io_u); |
ebac4655 JA |
300 | break; |
301 | } | |
302 | ||
303 | /* | |
304 | * we have one pending to verify, do that while | |
305 | * we are doing io on the next one | |
306 | */ | |
307 | if (do_io_u_verify(td, &v_io_u)) | |
308 | break; | |
309 | ||
45bee283 | 310 | ret = td_io_getevents(td, 1, 1, NULL); |
ebac4655 JA |
311 | if (ret != 1) { |
312 | if (ret < 0) | |
313 | td_verror(td, ret); | |
314 | break; | |
315 | } | |
316 | ||
2866c82d | 317 | v_io_u = td->io_ops->event(td, 0); |
ebac4655 JA |
318 | icd.nr = 1; |
319 | icd.error = 0; | |
02bcaa8c | 320 | fio_gettime(&icd.time, NULL); |
ebac4655 JA |
321 | io_completed(td, v_io_u, &icd); |
322 | ||
323 | if (icd.error) { | |
324 | td_verror(td, icd.error); | |
325 | put_io_u(td, v_io_u); | |
326 | v_io_u = NULL; | |
327 | break; | |
328 | } | |
329 | ||
ebac4655 JA |
330 | /* |
331 | * if we can't submit more io, we need to verify now | |
332 | */ | |
333 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
334 | break; | |
335 | ||
336 | } while (1); | |
337 | ||
338 | do_io_u_verify(td, &v_io_u); | |
339 | ||
340 | if (td->cur_depth) | |
341 | cleanup_pending_aio(td); | |
342 | ||
343 | td_set_runstate(td, TD_RUNNING); | |
344 | } | |
345 | ||
b990b5c0 JA |
346 | /* |
347 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
348 | * manner. | |
349 | */ | |
350 | static void do_cpuio(struct thread_data *td) | |
351 | { | |
352 | struct timeval e; | |
353 | int split = 100 / td->cpuload; | |
354 | int i = 0; | |
355 | ||
356 | while (!td->terminate) { | |
02bcaa8c | 357 | fio_gettime(&e, NULL); |
b990b5c0 JA |
358 | |
359 | if (runtime_exceeded(td, &e)) | |
360 | break; | |
361 | ||
362 | if (!(i % split)) | |
363 | __usec_sleep(10000); | |
364 | else | |
365 | usec_sleep(td, 10000); | |
366 | ||
367 | i++; | |
368 | } | |
369 | } | |
370 | ||
32cd46a0 | 371 | /* |
906c8d75 | 372 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
373 | * and reaps them, checking for rate and errors along the way. |
374 | */ | |
ebac4655 JA |
375 | static void do_io(struct thread_data *td) |
376 | { | |
377 | struct io_completion_data icd; | |
02bcaa8c | 378 | struct timeval s; |
ebac4655 | 379 | unsigned long usec; |
53cdc686 | 380 | struct fio_file *f; |
84585003 | 381 | int i, ret = 0; |
ebac4655 | 382 | |
5853e5a8 JA |
383 | td_set_runstate(td, TD_RUNNING); |
384 | ||
ebac4655 | 385 | while (td->this_io_bytes[td->ddir] < td->io_size) { |
ebac4655 | 386 | struct timespec *timeout; |
84585003 | 387 | int min_evts = 0; |
ebac4655 JA |
388 | struct io_u *io_u; |
389 | ||
390 | if (td->terminate) | |
391 | break; | |
392 | ||
53cdc686 JA |
393 | f = get_next_file(td); |
394 | if (!f) | |
395 | break; | |
396 | ||
397 | io_u = get_io_u(td, f); | |
ebac4655 JA |
398 | if (!io_u) |
399 | break; | |
400 | ||
401 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
402 | ||
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 { | |
415 | td_verror(td, io_u->error); | |
416 | put_io_u(td, io_u); | |
417 | break; | |
418 | } | |
ebac4655 JA |
419 | } |
420 | ||
421 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
422 | ||
423 | if (td->cur_depth < td->iodepth) { | |
8d7ad8d9 JA |
424 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; |
425 | ||
ebac4655 JA |
426 | timeout = &ts; |
427 | min_evts = 0; | |
428 | } else { | |
429 | timeout = NULL; | |
430 | min_evts = 1; | |
431 | } | |
432 | ||
45bee283 | 433 | ret = td_io_getevents(td, min_evts, td->cur_depth, timeout); |
ebac4655 | 434 | if (ret < 0) { |
87dc1ab1 | 435 | td_verror(td, ret); |
ebac4655 JA |
436 | break; |
437 | } else if (!ret) | |
438 | continue; | |
439 | ||
440 | icd.nr = ret; | |
441 | ios_completed(td, &icd); | |
442 | if (icd.error) { | |
443 | td_verror(td, icd.error); | |
444 | break; | |
445 | } | |
446 | ||
447 | /* | |
448 | * the rate is batched for now, it should work for batches | |
449 | * of completions except the very first one which may look | |
450 | * a little bursty | |
451 | */ | |
02bcaa8c | 452 | usec = utime_since(&s, &icd.time); |
ebac4655 | 453 | |
a00735e6 | 454 | rate_throttle(td, usec, icd.bytes_done[td->ddir], td->ddir); |
ebac4655 | 455 | |
02bcaa8c | 456 | if (check_min_rate(td, &icd.time)) { |
98aa62d8 | 457 | if (exitall_on_terminate) |
6ce15a32 | 458 | terminate_threads(td->groupid, 0); |
fd841467 | 459 | td_verror(td, ENODATA); |
ebac4655 JA |
460 | break; |
461 | } | |
462 | ||
02bcaa8c | 463 | if (runtime_exceeded(td, &icd.time)) |
ebac4655 JA |
464 | break; |
465 | ||
9c1f7434 JA |
466 | if (td->thinktime) { |
467 | unsigned long long b; | |
468 | ||
469 | b = td->io_blocks[0] + td->io_blocks[1]; | |
25309a33 | 470 | if (!(b % td->thinktime_blocks)) |
9c1f7434 JA |
471 | usec_sleep(td, td->thinktime); |
472 | } | |
ebac4655 JA |
473 | } |
474 | ||
4d2413c6 | 475 | if (!td->error) { |
84585003 JA |
476 | if (td->cur_depth) |
477 | cleanup_pending_aio(td); | |
ebac4655 | 478 | |
84585003 JA |
479 | if (should_fsync(td) && td->end_fsync) { |
480 | td_set_runstate(td, TD_FSYNCING); | |
481 | for_each_file(td, f, i) | |
858a3d47 | 482 | fio_io_sync(td, f); |
84585003 | 483 | } |
5853e5a8 | 484 | } |
ebac4655 JA |
485 | } |
486 | ||
ebac4655 JA |
487 | static void cleanup_io_u(struct thread_data *td) |
488 | { | |
489 | struct list_head *entry, *n; | |
490 | struct io_u *io_u; | |
491 | ||
492 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
493 | io_u = list_entry(entry, struct io_u, list); | |
494 | ||
495 | list_del(&io_u->list); | |
496 | free(io_u); | |
497 | } | |
498 | ||
2f9ade3c | 499 | free_io_mem(td); |
ebac4655 JA |
500 | } |
501 | ||
6b9cea23 JA |
502 | /* |
503 | * "randomly" fill the buffer contents | |
504 | */ | |
66eeb296 | 505 | static void fill_rand_buf(struct io_u *io_u, int max_bs) |
6b9cea23 | 506 | { |
66eeb296 | 507 | int *ptr = io_u->buf; |
6b9cea23 JA |
508 | |
509 | while ((void *) ptr - io_u->buf < max_bs) { | |
510 | *ptr = rand() * 0x9e370001; | |
511 | ptr++; | |
512 | } | |
513 | } | |
514 | ||
ebac4655 JA |
515 | static int init_io_u(struct thread_data *td) |
516 | { | |
517 | struct io_u *io_u; | |
a00735e6 | 518 | unsigned int max_bs; |
ebac4655 JA |
519 | int i, max_units; |
520 | char *p; | |
521 | ||
2866c82d | 522 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
523 | return 0; |
524 | ||
2866c82d | 525 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
526 | max_units = 1; |
527 | else | |
528 | max_units = td->iodepth; | |
529 | ||
a00735e6 | 530 | max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]); |
74b025b0 JA |
531 | td->orig_buffer_size = max_bs * max_units; |
532 | ||
d0bdaf49 | 533 | if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE) |
56bb17f2 | 534 | td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1); |
74b025b0 | 535 | else |
29d610e1 | 536 | td->orig_buffer_size += page_mask; |
ebac4655 | 537 | |
2f9ade3c JA |
538 | if (allocate_io_mem(td)) |
539 | return 1; | |
ebac4655 | 540 | |
ebac4655 JA |
541 | p = ALIGN(td->orig_buffer); |
542 | for (i = 0; i < max_units; i++) { | |
543 | io_u = malloc(sizeof(*io_u)); | |
544 | memset(io_u, 0, sizeof(*io_u)); | |
545 | INIT_LIST_HEAD(&io_u->list); | |
546 | ||
a00735e6 | 547 | io_u->buf = p + max_bs * i; |
6b9cea23 | 548 | if (td_write(td) || td_rw(td)) |
a00735e6 | 549 | fill_rand_buf(io_u, max_bs); |
6b9cea23 | 550 | |
b1ff3403 | 551 | io_u->index = i; |
ebac4655 JA |
552 | list_add(&io_u->list, &td->io_u_freelist); |
553 | } | |
554 | ||
555 | return 0; | |
556 | } | |
557 | ||
da86774e JA |
558 | static int switch_ioscheduler(struct thread_data *td) |
559 | { | |
560 | char tmp[256], tmp2[128]; | |
561 | FILE *f; | |
562 | int ret; | |
563 | ||
f48b467c JA |
564 | if (td->io_ops->flags & FIO_CPUIO) |
565 | return 0; | |
566 | ||
da86774e JA |
567 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
568 | ||
569 | f = fopen(tmp, "r+"); | |
570 | if (!f) { | |
571 | td_verror(td, errno); | |
572 | return 1; | |
573 | } | |
574 | ||
575 | /* | |
576 | * Set io scheduler. | |
577 | */ | |
578 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
579 | if (ferror(f) || ret != 1) { | |
580 | td_verror(td, errno); | |
581 | fclose(f); | |
582 | return 1; | |
583 | } | |
584 | ||
585 | rewind(f); | |
586 | ||
587 | /* | |
588 | * Read back and check that the selected scheduler is now the default. | |
589 | */ | |
590 | ret = fread(tmp, 1, sizeof(tmp), f); | |
591 | if (ferror(f) || ret < 0) { | |
592 | td_verror(td, errno); | |
593 | fclose(f); | |
594 | return 1; | |
595 | } | |
596 | ||
597 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
598 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 599 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
600 | td_verror(td, EINVAL); |
601 | fclose(f); | |
602 | return 1; | |
603 | } | |
604 | ||
605 | fclose(f); | |
606 | return 0; | |
607 | } | |
608 | ||
ebac4655 JA |
609 | static void clear_io_state(struct thread_data *td) |
610 | { | |
53cdc686 JA |
611 | struct fio_file *f; |
612 | int i; | |
ebac4655 | 613 | |
ebac4655 JA |
614 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
615 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 616 | td->zone_bytes = 0; |
ebac4655 | 617 | |
53cdc686 JA |
618 | for_each_file(td, f, i) { |
619 | f->last_pos = 0; | |
620 | if (td->io_ops->flags & FIO_SYNCIO) | |
621 | lseek(f->fd, SEEK_SET, 0); | |
622 | ||
623 | if (f->file_map) | |
624 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
625 | } | |
ebac4655 JA |
626 | } |
627 | ||
906c8d75 JA |
628 | /* |
629 | * Entry point for the thread based jobs. The process based jobs end up | |
630 | * here as well, after a little setup. | |
631 | */ | |
ebac4655 JA |
632 | static void *thread_main(void *data) |
633 | { | |
69008999 | 634 | unsigned long long runtime[2]; |
ebac4655 | 635 | struct thread_data *td = data; |
ebac4655 JA |
636 | |
637 | if (!td->use_thread) | |
638 | setsid(); | |
639 | ||
640 | td->pid = getpid(); | |
641 | ||
aea47d44 JA |
642 | INIT_LIST_HEAD(&td->io_u_freelist); |
643 | INIT_LIST_HEAD(&td->io_u_busylist); | |
644 | INIT_LIST_HEAD(&td->io_hist_list); | |
645 | INIT_LIST_HEAD(&td->io_log_list); | |
646 | ||
ebac4655 JA |
647 | if (init_io_u(td)) |
648 | goto err; | |
649 | ||
650 | if (fio_setaffinity(td) == -1) { | |
651 | td_verror(td, errno); | |
652 | goto err; | |
653 | } | |
654 | ||
aea47d44 JA |
655 | if (init_iolog(td)) |
656 | goto err; | |
657 | ||
ebac4655 JA |
658 | if (td->ioprio) { |
659 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
660 | td_verror(td, errno); | |
661 | goto err; | |
662 | } | |
663 | } | |
664 | ||
1056eaad | 665 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
666 | td_verror(td, errno); |
667 | goto err; | |
668 | } | |
669 | ||
75154845 JA |
670 | if (init_random_state(td)) |
671 | goto err; | |
672 | ||
56f9498d JA |
673 | if (td->ioscheduler && switch_ioscheduler(td)) |
674 | goto err; | |
da86774e | 675 | |
75154845 | 676 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
677 | fio_sem_up(&startup_sem); |
678 | fio_sem_down(&td->mutex); | |
ebac4655 | 679 | |
53cdc686 | 680 | if (!td->create_serialize && setup_files(td)) |
ebac4655 | 681 | goto err; |
21972cde JA |
682 | if (open_files(td)) |
683 | goto err; | |
ebac4655 | 684 | |
7d6c5283 JA |
685 | /* |
686 | * Do this late, as some IO engines would like to have the | |
687 | * files setup prior to initializing structures. | |
688 | */ | |
689 | if (td_io_init(td)) | |
690 | goto err; | |
691 | ||
69cfd7e0 JA |
692 | if (td->exec_prerun) { |
693 | if (system(td->exec_prerun) < 0) | |
694 | goto err; | |
695 | } | |
4e0ba8af | 696 | |
69008999 | 697 | fio_gettime(&td->epoch, NULL); |
36dff966 | 698 | getrusage(RUSAGE_SELF, &td->ru_start); |
69008999 JA |
699 | |
700 | runtime[0] = runtime[1] = 0; | |
ebac4655 | 701 | while (td->loops--) { |
02bcaa8c | 702 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
703 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); |
704 | ||
705 | if (td->ratemin) | |
706 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
707 | ||
708 | clear_io_state(td); | |
709 | prune_io_piece_log(td); | |
710 | ||
2866c82d | 711 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
712 | do_cpuio(td); |
713 | else | |
714 | do_io(td); | |
ebac4655 | 715 | |
69008999 | 716 | runtime[td->ddir] += utime_since_now(&td->start); |
aea47d44 | 717 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
69008999 | 718 | runtime[td->ddir ^ 1] = runtime[td->ddir]; |
3d60d1ed | 719 | |
ebac4655 JA |
720 | if (td->error || td->terminate) |
721 | break; | |
722 | ||
723 | if (td->verify == VERIFY_NONE) | |
724 | continue; | |
725 | ||
726 | clear_io_state(td); | |
02bcaa8c | 727 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
728 | |
729 | do_verify(td); | |
730 | ||
69008999 | 731 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
732 | |
733 | if (td->error || td->terminate) | |
734 | break; | |
735 | } | |
736 | ||
36dff966 | 737 | update_rusage_stat(td); |
69008999 JA |
738 | fio_gettime(&td->end_time, NULL); |
739 | td->runtime[0] = runtime[0] / 1000; | |
740 | td->runtime[1] = runtime[1] / 1000; | |
741 | ||
ebac4655 JA |
742 | if (td->bw_log) |
743 | finish_log(td, td->bw_log, "bw"); | |
744 | if (td->slat_log) | |
745 | finish_log(td, td->slat_log, "slat"); | |
746 | if (td->clat_log) | |
747 | finish_log(td, td->clat_log, "clat"); | |
076efc7c | 748 | if (td->write_iolog_file) |
843a7413 | 749 | write_iolog_close(td); |
69cfd7e0 JA |
750 | if (td->exec_postrun) { |
751 | if (system(td->exec_postrun) < 0) | |
752 | log_err("fio: postrun %s failed\n", td->exec_postrun); | |
753 | } | |
ebac4655 JA |
754 | |
755 | if (exitall_on_terminate) | |
6ce15a32 | 756 | terminate_threads(td->groupid, 0); |
ebac4655 JA |
757 | |
758 | err: | |
53cdc686 | 759 | close_files(td); |
2866c82d | 760 | close_ioengine(td); |
ebac4655 | 761 | cleanup_io_u(td); |
ebac4655 JA |
762 | td_set_runstate(td, TD_EXITED); |
763 | return NULL; | |
764 | ||
765 | } | |
766 | ||
906c8d75 JA |
767 | /* |
768 | * We cannot pass the td data into a forked process, so attach the td and | |
769 | * pass it to the thread worker. | |
770 | */ | |
ebac4655 JA |
771 | static void *fork_main(int shmid, int offset) |
772 | { | |
773 | struct thread_data *td; | |
774 | void *data; | |
775 | ||
776 | data = shmat(shmid, NULL, 0); | |
777 | if (data == (void *) -1) { | |
778 | perror("shmat"); | |
779 | return NULL; | |
780 | } | |
781 | ||
782 | td = data + offset * sizeof(struct thread_data); | |
783 | thread_main(td); | |
784 | shmdt(data); | |
785 | return NULL; | |
786 | } | |
787 | ||
906c8d75 JA |
788 | /* |
789 | * Run over the job map and reap the threads that have exited, if any. | |
790 | */ | |
ebac4655 JA |
791 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
792 | { | |
34572e28 | 793 | struct thread_data *td; |
4d2413c6 | 794 | int i, cputhreads, pending; |
ebac4655 JA |
795 | |
796 | /* | |
797 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
798 | */ | |
4d2413c6 | 799 | pending = cputhreads = 0; |
34572e28 | 800 | for_each_td(td, i) { |
84585003 JA |
801 | /* |
802 | * ->io_ops is NULL for a thread that has closed its | |
803 | * io engine | |
804 | */ | |
805 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
806 | cputhreads++; |
807 | ||
4d2413c6 JA |
808 | if (td->runstate != TD_EXITED) { |
809 | if (td->runstate < TD_RUNNING) | |
810 | pending++; | |
811 | ||
ebac4655 | 812 | continue; |
4d2413c6 | 813 | } |
ebac4655 JA |
814 | |
815 | td_set_runstate(td, TD_REAPED); | |
816 | ||
817 | if (td->use_thread) { | |
818 | long ret; | |
819 | ||
820 | if (pthread_join(td->thread, (void *) &ret)) | |
821 | perror("thread_join"); | |
822 | } else | |
823 | waitpid(td->pid, NULL, 0); | |
824 | ||
825 | (*nr_running)--; | |
826 | (*m_rate) -= td->ratemin; | |
827 | (*t_rate) -= td->rate; | |
828 | } | |
b990b5c0 | 829 | |
4d2413c6 | 830 | if (*nr_running == cputhreads && !pending) |
6ce15a32 | 831 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
832 | } |
833 | ||
906c8d75 JA |
834 | /* |
835 | * Main function for kicking off and reaping jobs, as needed. | |
836 | */ | |
ebac4655 JA |
837 | static void run_threads(void) |
838 | { | |
ebac4655 JA |
839 | struct thread_data *td; |
840 | unsigned long spent; | |
841 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 842 | |
2f9ade3c JA |
843 | if (fio_pin_memory()) |
844 | return; | |
ebac4655 | 845 | |
c6ae0a5b JA |
846 | if (!terse_output) { |
847 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
848 | fflush(stdout); | |
849 | } | |
c04f7ec3 | 850 | |
4efa970e JA |
851 | signal(SIGINT, sig_handler); |
852 | signal(SIGALRM, sig_handler); | |
853 | ||
ebac4655 JA |
854 | todo = thread_number; |
855 | nr_running = 0; | |
856 | nr_started = 0; | |
857 | m_rate = t_rate = 0; | |
858 | ||
34572e28 | 859 | for_each_td(td, i) { |
263e529f | 860 | print_status_init(td->thread_number - 1); |
ebac4655 | 861 | |
380cf265 JA |
862 | if (!td->create_serialize) { |
863 | init_disk_util(td); | |
ebac4655 | 864 | continue; |
380cf265 | 865 | } |
ebac4655 JA |
866 | |
867 | /* | |
868 | * do file setup here so it happens sequentially, | |
869 | * we don't want X number of threads getting their | |
870 | * client data interspersed on disk | |
871 | */ | |
53cdc686 | 872 | if (setup_files(td)) { |
ebac4655 JA |
873 | td_set_runstate(td, TD_REAPED); |
874 | todo--; | |
875 | } | |
380cf265 JA |
876 | |
877 | init_disk_util(td); | |
ebac4655 JA |
878 | } |
879 | ||
ebac4655 | 880 | while (todo) { |
75154845 JA |
881 | struct thread_data *map[MAX_JOBS]; |
882 | struct timeval this_start; | |
883 | int this_jobs = 0, left; | |
884 | ||
ebac4655 JA |
885 | /* |
886 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
887 | */ | |
34572e28 | 888 | for_each_td(td, i) { |
ebac4655 JA |
889 | if (td->runstate != TD_NOT_CREATED) |
890 | continue; | |
891 | ||
892 | /* | |
893 | * never got a chance to start, killed by other | |
894 | * thread for some reason | |
895 | */ | |
896 | if (td->terminate) { | |
897 | todo--; | |
898 | continue; | |
899 | } | |
900 | ||
901 | if (td->start_delay) { | |
263e529f | 902 | spent = mtime_since_genesis(); |
ebac4655 JA |
903 | |
904 | if (td->start_delay * 1000 > spent) | |
905 | continue; | |
906 | } | |
907 | ||
908 | if (td->stonewall && (nr_started || nr_running)) | |
909 | break; | |
910 | ||
75154845 JA |
911 | /* |
912 | * Set state to created. Thread will transition | |
913 | * to TD_INITIALIZED when it's done setting up. | |
914 | */ | |
ebac4655 | 915 | td_set_runstate(td, TD_CREATED); |
75154845 | 916 | map[this_jobs++] = td; |
bbfd6b00 | 917 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
918 | nr_started++; |
919 | ||
920 | if (td->use_thread) { | |
921 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
922 | perror("thread_create"); | |
923 | nr_started--; | |
924 | } | |
925 | } else { | |
926 | if (fork()) | |
bbfd6b00 | 927 | fio_sem_down(&startup_sem); |
ebac4655 JA |
928 | else { |
929 | fork_main(shm_id, i); | |
930 | exit(0); | |
931 | } | |
932 | } | |
933 | } | |
934 | ||
935 | /* | |
75154845 JA |
936 | * Wait for the started threads to transition to |
937 | * TD_INITIALIZED. | |
ebac4655 | 938 | */ |
02bcaa8c | 939 | fio_gettime(&this_start, NULL); |
75154845 | 940 | left = this_jobs; |
6ce15a32 | 941 | while (left && !fio_abort) { |
75154845 JA |
942 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
943 | break; | |
944 | ||
945 | usleep(100000); | |
946 | ||
947 | for (i = 0; i < this_jobs; i++) { | |
948 | td = map[i]; | |
949 | if (!td) | |
950 | continue; | |
b6f4d880 | 951 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
952 | map[i] = NULL; |
953 | left--; | |
b6f4d880 JA |
954 | } else if (td->runstate >= TD_EXITED) { |
955 | map[i] = NULL; | |
956 | left--; | |
957 | todo--; | |
958 | nr_running++; /* work-around... */ | |
75154845 JA |
959 | } |
960 | } | |
961 | } | |
962 | ||
963 | if (left) { | |
3b70d7e5 | 964 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
965 | for (i = 0; i < this_jobs; i++) { |
966 | td = map[i]; | |
967 | if (!td) | |
968 | continue; | |
969 | kill(td->pid, SIGTERM); | |
970 | } | |
971 | break; | |
972 | } | |
973 | ||
974 | /* | |
b6f4d880 | 975 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 976 | */ |
34572e28 | 977 | for_each_td(td, i) { |
75154845 | 978 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
979 | continue; |
980 | ||
981 | td_set_runstate(td, TD_RUNNING); | |
982 | nr_running++; | |
983 | nr_started--; | |
984 | m_rate += td->ratemin; | |
985 | t_rate += td->rate; | |
75154845 | 986 | todo--; |
bbfd6b00 | 987 | fio_sem_up(&td->mutex); |
ebac4655 JA |
988 | } |
989 | ||
990 | reap_threads(&nr_running, &t_rate, &m_rate); | |
991 | ||
992 | if (todo) | |
993 | usleep(100000); | |
994 | } | |
995 | ||
996 | while (nr_running) { | |
997 | reap_threads(&nr_running, &t_rate, &m_rate); | |
998 | usleep(10000); | |
999 | } | |
1000 | ||
1001 | update_io_ticks(); | |
2f9ade3c | 1002 | fio_unpin_memory(); |
ebac4655 JA |
1003 | } |
1004 | ||
ebac4655 JA |
1005 | int main(int argc, char *argv[]) |
1006 | { | |
29d610e1 JA |
1007 | long ps; |
1008 | ||
dbe1125e JA |
1009 | /* |
1010 | * We need locale for number printing, if it isn't set then just | |
1011 | * go with the US format. | |
1012 | */ | |
1013 | if (!getenv("LC_NUMERIC")) | |
1014 | setlocale(LC_NUMERIC, "en_US"); | |
1015 | ||
ebac4655 JA |
1016 | if (parse_options(argc, argv)) |
1017 | return 1; | |
1018 | ||
1019 | if (!thread_number) { | |
3b70d7e5 | 1020 | log_err("Nothing to do\n"); |
ebac4655 JA |
1021 | return 1; |
1022 | } | |
1023 | ||
29d610e1 JA |
1024 | ps = sysconf(_SC_PAGESIZE); |
1025 | if (ps < 0) { | |
1026 | log_err("Failed to get page size\n"); | |
1027 | return 1; | |
1028 | } | |
1029 | ||
1030 | page_mask = ps - 1; | |
1031 | ||
bb3884d8 JA |
1032 | if (write_bw_log) { |
1033 | setup_log(&agg_io_log[DDIR_READ]); | |
1034 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1035 | } | |
1036 | ||
ebac4655 JA |
1037 | disk_util_timer_arm(); |
1038 | ||
1039 | run_threads(); | |
6ce15a32 | 1040 | |
bb3884d8 | 1041 | if (!fio_abort) { |
6ce15a32 | 1042 | show_run_stats(); |
bb3884d8 JA |
1043 | if (write_bw_log) { |
1044 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1045 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1046 | } | |
1047 | } | |
ebac4655 JA |
1048 | |
1049 | return 0; | |
1050 | } |