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 | ||
3951414d | 385 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < 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 { | |
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]; | |
25309a33 | 469 | if (!(b % td->thinktime_blocks)) |
9c1f7434 JA |
470 | usec_sleep(td, td->thinktime); |
471 | } | |
ebac4655 JA |
472 | } |
473 | ||
4d2413c6 | 474 | if (!td->error) { |
84585003 JA |
475 | if (td->cur_depth) |
476 | cleanup_pending_aio(td); | |
ebac4655 | 477 | |
84585003 JA |
478 | if (should_fsync(td) && td->end_fsync) { |
479 | td_set_runstate(td, TD_FSYNCING); | |
480 | for_each_file(td, f, i) | |
858a3d47 | 481 | fio_io_sync(td, f); |
84585003 | 482 | } |
5853e5a8 | 483 | } |
ebac4655 JA |
484 | } |
485 | ||
ebac4655 JA |
486 | static void cleanup_io_u(struct thread_data *td) |
487 | { | |
488 | struct list_head *entry, *n; | |
489 | struct io_u *io_u; | |
490 | ||
491 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
492 | io_u = list_entry(entry, struct io_u, list); | |
493 | ||
494 | list_del(&io_u->list); | |
495 | free(io_u); | |
496 | } | |
497 | ||
2f9ade3c | 498 | free_io_mem(td); |
ebac4655 JA |
499 | } |
500 | ||
6b9cea23 JA |
501 | /* |
502 | * "randomly" fill the buffer contents | |
503 | */ | |
66eeb296 | 504 | static void fill_rand_buf(struct io_u *io_u, int max_bs) |
6b9cea23 | 505 | { |
66eeb296 | 506 | int *ptr = io_u->buf; |
6b9cea23 JA |
507 | |
508 | while ((void *) ptr - io_u->buf < max_bs) { | |
509 | *ptr = rand() * 0x9e370001; | |
510 | ptr++; | |
511 | } | |
512 | } | |
513 | ||
ebac4655 JA |
514 | static int init_io_u(struct thread_data *td) |
515 | { | |
516 | struct io_u *io_u; | |
a00735e6 | 517 | unsigned int max_bs; |
ebac4655 JA |
518 | int i, max_units; |
519 | char *p; | |
520 | ||
2866c82d | 521 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
522 | return 0; |
523 | ||
2866c82d | 524 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
525 | max_units = 1; |
526 | else | |
527 | max_units = td->iodepth; | |
528 | ||
a00735e6 | 529 | max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]); |
74b025b0 JA |
530 | td->orig_buffer_size = max_bs * max_units; |
531 | ||
d0bdaf49 | 532 | if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE) |
56bb17f2 | 533 | td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1); |
74b025b0 | 534 | else |
29d610e1 | 535 | td->orig_buffer_size += page_mask; |
ebac4655 | 536 | |
2f9ade3c JA |
537 | if (allocate_io_mem(td)) |
538 | return 1; | |
ebac4655 | 539 | |
ebac4655 JA |
540 | p = ALIGN(td->orig_buffer); |
541 | for (i = 0; i < max_units; i++) { | |
542 | io_u = malloc(sizeof(*io_u)); | |
543 | memset(io_u, 0, sizeof(*io_u)); | |
544 | INIT_LIST_HEAD(&io_u->list); | |
545 | ||
a00735e6 | 546 | io_u->buf = p + max_bs * i; |
6b9cea23 | 547 | if (td_write(td) || td_rw(td)) |
a00735e6 | 548 | fill_rand_buf(io_u, max_bs); |
6b9cea23 | 549 | |
b1ff3403 | 550 | io_u->index = i; |
ebac4655 JA |
551 | list_add(&io_u->list, &td->io_u_freelist); |
552 | } | |
553 | ||
554 | return 0; | |
555 | } | |
556 | ||
da86774e JA |
557 | static int switch_ioscheduler(struct thread_data *td) |
558 | { | |
559 | char tmp[256], tmp2[128]; | |
560 | FILE *f; | |
561 | int ret; | |
562 | ||
f48b467c JA |
563 | if (td->io_ops->flags & FIO_CPUIO) |
564 | return 0; | |
565 | ||
da86774e JA |
566 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
567 | ||
568 | f = fopen(tmp, "r+"); | |
569 | if (!f) { | |
570 | td_verror(td, errno); | |
571 | return 1; | |
572 | } | |
573 | ||
574 | /* | |
575 | * Set io scheduler. | |
576 | */ | |
577 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
578 | if (ferror(f) || ret != 1) { | |
579 | td_verror(td, errno); | |
580 | fclose(f); | |
581 | return 1; | |
582 | } | |
583 | ||
584 | rewind(f); | |
585 | ||
586 | /* | |
587 | * Read back and check that the selected scheduler is now the default. | |
588 | */ | |
589 | ret = fread(tmp, 1, sizeof(tmp), f); | |
590 | if (ferror(f) || ret < 0) { | |
591 | td_verror(td, errno); | |
592 | fclose(f); | |
593 | return 1; | |
594 | } | |
595 | ||
596 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
597 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 598 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
599 | td_verror(td, EINVAL); |
600 | fclose(f); | |
601 | return 1; | |
602 | } | |
603 | ||
604 | fclose(f); | |
605 | return 0; | |
606 | } | |
607 | ||
ebac4655 JA |
608 | static void clear_io_state(struct thread_data *td) |
609 | { | |
53cdc686 JA |
610 | struct fio_file *f; |
611 | int i; | |
ebac4655 | 612 | |
ebac4655 JA |
613 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
614 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 615 | td->zone_bytes = 0; |
ebac4655 | 616 | |
53cdc686 JA |
617 | for_each_file(td, f, i) { |
618 | f->last_pos = 0; | |
619 | if (td->io_ops->flags & FIO_SYNCIO) | |
620 | lseek(f->fd, SEEK_SET, 0); | |
621 | ||
622 | if (f->file_map) | |
623 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
624 | } | |
ebac4655 JA |
625 | } |
626 | ||
906c8d75 JA |
627 | /* |
628 | * Entry point for the thread based jobs. The process based jobs end up | |
629 | * here as well, after a little setup. | |
630 | */ | |
ebac4655 JA |
631 | static void *thread_main(void *data) |
632 | { | |
69008999 | 633 | unsigned long long runtime[2]; |
ebac4655 | 634 | struct thread_data *td = data; |
ebac4655 JA |
635 | |
636 | if (!td->use_thread) | |
637 | setsid(); | |
638 | ||
639 | td->pid = getpid(); | |
640 | ||
aea47d44 JA |
641 | INIT_LIST_HEAD(&td->io_u_freelist); |
642 | INIT_LIST_HEAD(&td->io_u_busylist); | |
643 | INIT_LIST_HEAD(&td->io_hist_list); | |
644 | INIT_LIST_HEAD(&td->io_log_list); | |
645 | ||
ebac4655 JA |
646 | if (init_io_u(td)) |
647 | goto err; | |
648 | ||
649 | if (fio_setaffinity(td) == -1) { | |
650 | td_verror(td, errno); | |
651 | goto err; | |
652 | } | |
653 | ||
aea47d44 JA |
654 | if (init_iolog(td)) |
655 | goto err; | |
656 | ||
ebac4655 JA |
657 | if (td->ioprio) { |
658 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
659 | td_verror(td, errno); | |
660 | goto err; | |
661 | } | |
662 | } | |
663 | ||
1056eaad | 664 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
665 | td_verror(td, errno); |
666 | goto err; | |
667 | } | |
668 | ||
75154845 JA |
669 | if (init_random_state(td)) |
670 | goto err; | |
671 | ||
56f9498d JA |
672 | if (td->ioscheduler && switch_ioscheduler(td)) |
673 | goto err; | |
da86774e | 674 | |
75154845 | 675 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
676 | fio_sem_up(&startup_sem); |
677 | fio_sem_down(&td->mutex); | |
ebac4655 | 678 | |
53cdc686 | 679 | if (!td->create_serialize && setup_files(td)) |
ebac4655 | 680 | goto err; |
21972cde JA |
681 | if (open_files(td)) |
682 | goto err; | |
ebac4655 | 683 | |
7d6c5283 JA |
684 | /* |
685 | * Do this late, as some IO engines would like to have the | |
686 | * files setup prior to initializing structures. | |
687 | */ | |
688 | if (td_io_init(td)) | |
689 | goto err; | |
690 | ||
69cfd7e0 JA |
691 | if (td->exec_prerun) { |
692 | if (system(td->exec_prerun) < 0) | |
693 | goto err; | |
694 | } | |
4e0ba8af | 695 | |
69008999 | 696 | fio_gettime(&td->epoch, NULL); |
36dff966 | 697 | getrusage(RUSAGE_SELF, &td->ru_start); |
69008999 JA |
698 | |
699 | runtime[0] = runtime[1] = 0; | |
ebac4655 | 700 | while (td->loops--) { |
02bcaa8c | 701 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
702 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); |
703 | ||
704 | if (td->ratemin) | |
705 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
706 | ||
707 | clear_io_state(td); | |
708 | prune_io_piece_log(td); | |
709 | ||
2866c82d | 710 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
711 | do_cpuio(td); |
712 | else | |
713 | do_io(td); | |
ebac4655 | 714 | |
69008999 | 715 | runtime[td->ddir] += utime_since_now(&td->start); |
aea47d44 | 716 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
69008999 | 717 | runtime[td->ddir ^ 1] = runtime[td->ddir]; |
3d60d1ed | 718 | |
ebac4655 JA |
719 | if (td->error || td->terminate) |
720 | break; | |
721 | ||
722 | if (td->verify == VERIFY_NONE) | |
723 | continue; | |
724 | ||
725 | clear_io_state(td); | |
02bcaa8c | 726 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
727 | |
728 | do_verify(td); | |
729 | ||
69008999 | 730 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
731 | |
732 | if (td->error || td->terminate) | |
733 | break; | |
734 | } | |
735 | ||
36dff966 | 736 | update_rusage_stat(td); |
69008999 JA |
737 | fio_gettime(&td->end_time, NULL); |
738 | td->runtime[0] = runtime[0] / 1000; | |
739 | td->runtime[1] = runtime[1] / 1000; | |
740 | ||
ebac4655 JA |
741 | if (td->bw_log) |
742 | finish_log(td, td->bw_log, "bw"); | |
743 | if (td->slat_log) | |
744 | finish_log(td, td->slat_log, "slat"); | |
745 | if (td->clat_log) | |
746 | finish_log(td, td->clat_log, "clat"); | |
076efc7c | 747 | if (td->write_iolog_file) |
843a7413 | 748 | write_iolog_close(td); |
69cfd7e0 JA |
749 | if (td->exec_postrun) { |
750 | if (system(td->exec_postrun) < 0) | |
751 | log_err("fio: postrun %s failed\n", td->exec_postrun); | |
752 | } | |
ebac4655 JA |
753 | |
754 | if (exitall_on_terminate) | |
6ce15a32 | 755 | terminate_threads(td->groupid, 0); |
ebac4655 JA |
756 | |
757 | err: | |
53cdc686 | 758 | close_files(td); |
2866c82d | 759 | close_ioengine(td); |
ebac4655 | 760 | cleanup_io_u(td); |
ebac4655 JA |
761 | td_set_runstate(td, TD_EXITED); |
762 | return NULL; | |
763 | ||
764 | } | |
765 | ||
906c8d75 JA |
766 | /* |
767 | * We cannot pass the td data into a forked process, so attach the td and | |
768 | * pass it to the thread worker. | |
769 | */ | |
ebac4655 JA |
770 | static void *fork_main(int shmid, int offset) |
771 | { | |
772 | struct thread_data *td; | |
773 | void *data; | |
774 | ||
775 | data = shmat(shmid, NULL, 0); | |
776 | if (data == (void *) -1) { | |
777 | perror("shmat"); | |
778 | return NULL; | |
779 | } | |
780 | ||
781 | td = data + offset * sizeof(struct thread_data); | |
782 | thread_main(td); | |
783 | shmdt(data); | |
784 | return NULL; | |
785 | } | |
786 | ||
906c8d75 JA |
787 | /* |
788 | * Run over the job map and reap the threads that have exited, if any. | |
789 | */ | |
ebac4655 JA |
790 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
791 | { | |
34572e28 | 792 | struct thread_data *td; |
4d2413c6 | 793 | int i, cputhreads, pending; |
ebac4655 JA |
794 | |
795 | /* | |
796 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
797 | */ | |
4d2413c6 | 798 | pending = cputhreads = 0; |
34572e28 | 799 | for_each_td(td, i) { |
84585003 JA |
800 | /* |
801 | * ->io_ops is NULL for a thread that has closed its | |
802 | * io engine | |
803 | */ | |
804 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
805 | cputhreads++; |
806 | ||
4d2413c6 JA |
807 | if (td->runstate != TD_EXITED) { |
808 | if (td->runstate < TD_RUNNING) | |
809 | pending++; | |
810 | ||
ebac4655 | 811 | continue; |
4d2413c6 | 812 | } |
ebac4655 JA |
813 | |
814 | td_set_runstate(td, TD_REAPED); | |
815 | ||
816 | if (td->use_thread) { | |
817 | long ret; | |
818 | ||
819 | if (pthread_join(td->thread, (void *) &ret)) | |
820 | perror("thread_join"); | |
821 | } else | |
822 | waitpid(td->pid, NULL, 0); | |
823 | ||
824 | (*nr_running)--; | |
825 | (*m_rate) -= td->ratemin; | |
826 | (*t_rate) -= td->rate; | |
827 | } | |
b990b5c0 | 828 | |
4d2413c6 | 829 | if (*nr_running == cputhreads && !pending) |
6ce15a32 | 830 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
831 | } |
832 | ||
906c8d75 JA |
833 | /* |
834 | * Main function for kicking off and reaping jobs, as needed. | |
835 | */ | |
ebac4655 JA |
836 | static void run_threads(void) |
837 | { | |
ebac4655 JA |
838 | struct thread_data *td; |
839 | unsigned long spent; | |
840 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 841 | |
2f9ade3c JA |
842 | if (fio_pin_memory()) |
843 | return; | |
ebac4655 | 844 | |
c6ae0a5b JA |
845 | if (!terse_output) { |
846 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
847 | fflush(stdout); | |
848 | } | |
c04f7ec3 | 849 | |
4efa970e JA |
850 | signal(SIGINT, sig_handler); |
851 | signal(SIGALRM, sig_handler); | |
852 | ||
ebac4655 JA |
853 | todo = thread_number; |
854 | nr_running = 0; | |
855 | nr_started = 0; | |
856 | m_rate = t_rate = 0; | |
857 | ||
34572e28 | 858 | for_each_td(td, i) { |
263e529f | 859 | print_status_init(td->thread_number - 1); |
ebac4655 | 860 | |
380cf265 JA |
861 | if (!td->create_serialize) { |
862 | init_disk_util(td); | |
ebac4655 | 863 | continue; |
380cf265 | 864 | } |
ebac4655 JA |
865 | |
866 | /* | |
867 | * do file setup here so it happens sequentially, | |
868 | * we don't want X number of threads getting their | |
869 | * client data interspersed on disk | |
870 | */ | |
53cdc686 | 871 | if (setup_files(td)) { |
ebac4655 JA |
872 | td_set_runstate(td, TD_REAPED); |
873 | todo--; | |
874 | } | |
380cf265 JA |
875 | |
876 | init_disk_util(td); | |
ebac4655 JA |
877 | } |
878 | ||
ebac4655 | 879 | while (todo) { |
75154845 JA |
880 | struct thread_data *map[MAX_JOBS]; |
881 | struct timeval this_start; | |
882 | int this_jobs = 0, left; | |
883 | ||
ebac4655 JA |
884 | /* |
885 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
886 | */ | |
34572e28 | 887 | for_each_td(td, i) { |
ebac4655 JA |
888 | if (td->runstate != TD_NOT_CREATED) |
889 | continue; | |
890 | ||
891 | /* | |
892 | * never got a chance to start, killed by other | |
893 | * thread for some reason | |
894 | */ | |
895 | if (td->terminate) { | |
896 | todo--; | |
897 | continue; | |
898 | } | |
899 | ||
900 | if (td->start_delay) { | |
263e529f | 901 | spent = mtime_since_genesis(); |
ebac4655 JA |
902 | |
903 | if (td->start_delay * 1000 > spent) | |
904 | continue; | |
905 | } | |
906 | ||
907 | if (td->stonewall && (nr_started || nr_running)) | |
908 | break; | |
909 | ||
75154845 JA |
910 | /* |
911 | * Set state to created. Thread will transition | |
912 | * to TD_INITIALIZED when it's done setting up. | |
913 | */ | |
ebac4655 | 914 | td_set_runstate(td, TD_CREATED); |
75154845 | 915 | map[this_jobs++] = td; |
bbfd6b00 | 916 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
917 | nr_started++; |
918 | ||
919 | if (td->use_thread) { | |
920 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
921 | perror("thread_create"); | |
922 | nr_started--; | |
923 | } | |
924 | } else { | |
925 | if (fork()) | |
bbfd6b00 | 926 | fio_sem_down(&startup_sem); |
ebac4655 JA |
927 | else { |
928 | fork_main(shm_id, i); | |
929 | exit(0); | |
930 | } | |
931 | } | |
932 | } | |
933 | ||
934 | /* | |
75154845 JA |
935 | * Wait for the started threads to transition to |
936 | * TD_INITIALIZED. | |
ebac4655 | 937 | */ |
02bcaa8c | 938 | fio_gettime(&this_start, NULL); |
75154845 | 939 | left = this_jobs; |
6ce15a32 | 940 | while (left && !fio_abort) { |
75154845 JA |
941 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
942 | break; | |
943 | ||
944 | usleep(100000); | |
945 | ||
946 | for (i = 0; i < this_jobs; i++) { | |
947 | td = map[i]; | |
948 | if (!td) | |
949 | continue; | |
b6f4d880 | 950 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
951 | map[i] = NULL; |
952 | left--; | |
b6f4d880 JA |
953 | } else if (td->runstate >= TD_EXITED) { |
954 | map[i] = NULL; | |
955 | left--; | |
956 | todo--; | |
957 | nr_running++; /* work-around... */ | |
75154845 JA |
958 | } |
959 | } | |
960 | } | |
961 | ||
962 | if (left) { | |
3b70d7e5 | 963 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
964 | for (i = 0; i < this_jobs; i++) { |
965 | td = map[i]; | |
966 | if (!td) | |
967 | continue; | |
968 | kill(td->pid, SIGTERM); | |
969 | } | |
970 | break; | |
971 | } | |
972 | ||
973 | /* | |
b6f4d880 | 974 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 975 | */ |
34572e28 | 976 | for_each_td(td, i) { |
75154845 | 977 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
978 | continue; |
979 | ||
980 | td_set_runstate(td, TD_RUNNING); | |
981 | nr_running++; | |
982 | nr_started--; | |
983 | m_rate += td->ratemin; | |
984 | t_rate += td->rate; | |
75154845 | 985 | todo--; |
bbfd6b00 | 986 | fio_sem_up(&td->mutex); |
ebac4655 JA |
987 | } |
988 | ||
989 | reap_threads(&nr_running, &t_rate, &m_rate); | |
990 | ||
991 | if (todo) | |
992 | usleep(100000); | |
993 | } | |
994 | ||
995 | while (nr_running) { | |
996 | reap_threads(&nr_running, &t_rate, &m_rate); | |
997 | usleep(10000); | |
998 | } | |
999 | ||
1000 | update_io_ticks(); | |
2f9ade3c | 1001 | fio_unpin_memory(); |
ebac4655 JA |
1002 | } |
1003 | ||
ebac4655 JA |
1004 | int main(int argc, char *argv[]) |
1005 | { | |
29d610e1 JA |
1006 | long ps; |
1007 | ||
dbe1125e JA |
1008 | /* |
1009 | * We need locale for number printing, if it isn't set then just | |
1010 | * go with the US format. | |
1011 | */ | |
1012 | if (!getenv("LC_NUMERIC")) | |
1013 | setlocale(LC_NUMERIC, "en_US"); | |
1014 | ||
ebac4655 JA |
1015 | if (parse_options(argc, argv)) |
1016 | return 1; | |
1017 | ||
1018 | if (!thread_number) { | |
3b70d7e5 | 1019 | log_err("Nothing to do\n"); |
ebac4655 JA |
1020 | return 1; |
1021 | } | |
1022 | ||
29d610e1 JA |
1023 | ps = sysconf(_SC_PAGESIZE); |
1024 | if (ps < 0) { | |
1025 | log_err("Failed to get page size\n"); | |
1026 | return 1; | |
1027 | } | |
1028 | ||
1029 | page_mask = ps - 1; | |
1030 | ||
bb3884d8 JA |
1031 | if (write_bw_log) { |
1032 | setup_log(&agg_io_log[DDIR_READ]); | |
1033 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1034 | } | |
1035 | ||
ebac4655 JA |
1036 | disk_util_timer_arm(); |
1037 | ||
1038 | run_threads(); | |
6ce15a32 | 1039 | |
bb3884d8 | 1040 | if (!fio_abort) { |
6ce15a32 | 1041 | show_run_stats(); |
bb3884d8 JA |
1042 | if (write_bw_log) { |
1043 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1044 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1045 | } | |
1046 | } | |
ebac4655 JA |
1047 | |
1048 | return 0; | |
1049 | } |