Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
aae22ca7 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
ebac4655 | 6 | * |
8e9fe637 JA |
7 | * The license below covers all files distributed with fio unless otherwise |
8 | * noted in the file itself. | |
9 | * | |
ebac4655 | 10 | * This program is free software; you can redistribute it and/or modify |
8e9fe637 JA |
11 | * it under the terms of the GNU General Public License version 2 as |
12 | * published by the Free Software Foundation. | |
ebac4655 JA |
13 | * |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
21 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
22 | * | |
23 | */ | |
ebac4655 JA |
24 | #include <unistd.h> |
25 | #include <fcntl.h> | |
26 | #include <string.h> | |
ebac4655 JA |
27 | #include <signal.h> |
28 | #include <time.h> | |
dbe1125e | 29 | #include <locale.h> |
36167d82 | 30 | #include <assert.h> |
ebac4655 JA |
31 | #include <sys/stat.h> |
32 | #include <sys/wait.h> | |
33 | #include <sys/ipc.h> | |
34 | #include <sys/shm.h> | |
ebac4655 JA |
35 | #include <sys/mman.h> |
36 | ||
37 | #include "fio.h" | |
38 | #include "os.h" | |
39 | ||
29d610e1 JA |
40 | static unsigned long page_mask; |
41 | #define ALIGN(buf) \ | |
42 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
43 | |
44 | int groupid = 0; | |
45 | int thread_number = 0; | |
ebac4655 | 46 | int shm_id = 0; |
53cdc686 | 47 | int temp_stall_ts; |
ebac4655 | 48 | |
bbfd6b00 | 49 | static volatile int startup_sem; |
6ce15a32 | 50 | static volatile int fio_abort; |
437c9b71 | 51 | static int exit_value; |
ebac4655 | 52 | |
bb3884d8 JA |
53 | struct io_log *agg_io_log[2]; |
54 | ||
ebac4655 | 55 | #define TERMINATE_ALL (-1) |
75154845 | 56 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 57 | |
6ce15a32 JA |
58 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
59 | { | |
60 | td->runstate = runstate; | |
61 | } | |
62 | ||
63 | static void terminate_threads(int group_id, int forced_kill) | |
ebac4655 | 64 | { |
34572e28 | 65 | struct thread_data *td; |
ebac4655 JA |
66 | int i; |
67 | ||
34572e28 | 68 | for_each_td(td, i) { |
ebac4655 JA |
69 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
70 | td->terminate = 1; | |
71 | td->start_delay = 0; | |
6ce15a32 JA |
72 | if (forced_kill) |
73 | td_set_runstate(td, TD_EXITED); | |
ebac4655 JA |
74 | } |
75 | } | |
76 | } | |
77 | ||
78 | static void sig_handler(int sig) | |
79 | { | |
80 | switch (sig) { | |
81 | case SIGALRM: | |
82 | update_io_ticks(); | |
83 | disk_util_timer_arm(); | |
84 | print_thread_status(); | |
85 | break; | |
86 | default: | |
6ce15a32 | 87 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 88 | fflush(stdout); |
6ce15a32 | 89 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
90 | break; |
91 | } | |
92 | } | |
93 | ||
906c8d75 JA |
94 | /* |
95 | * Check if we are above the minimum rate given. | |
96 | */ | |
ebac4655 JA |
97 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
98 | { | |
99 | unsigned long spent; | |
100 | unsigned long rate; | |
101 | int ddir = td->ddir; | |
102 | ||
103 | /* | |
104 | * allow a 2 second settle period in the beginning | |
105 | */ | |
106 | if (mtime_since(&td->start, now) < 2000) | |
107 | return 0; | |
108 | ||
109 | /* | |
110 | * if rate blocks is set, sample is running | |
111 | */ | |
112 | if (td->rate_bytes) { | |
113 | spent = mtime_since(&td->lastrate, now); | |
114 | if (spent < td->ratecycle) | |
115 | return 0; | |
116 | ||
117 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
118 | if (rate < td->ratemin) { | |
1e97cce9 | 119 | fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
120 | return 1; |
121 | } | |
122 | } | |
123 | ||
124 | td->rate_bytes = td->this_io_bytes[ddir]; | |
125 | memcpy(&td->lastrate, now, sizeof(*now)); | |
126 | return 0; | |
127 | } | |
128 | ||
129 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
130 | { | |
131 | if (!td->timeout) | |
132 | return 0; | |
133 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
134 | return 1; | |
135 | ||
136 | return 0; | |
137 | } | |
138 | ||
53cdc686 JA |
139 | static struct fio_file *get_next_file(struct thread_data *td) |
140 | { | |
0ab8db89 | 141 | unsigned int old_next_file = td->next_file; |
b2a15192 JA |
142 | struct fio_file *f; |
143 | ||
144 | do { | |
145 | f = &td->files[td->next_file]; | |
146 | ||
147 | td->next_file++; | |
148 | if (td->next_file >= td->nr_files) | |
149 | td->next_file = 0; | |
150 | ||
151 | if (f->fd != -1) | |
152 | break; | |
53cdc686 | 153 | |
b2a15192 JA |
154 | f = NULL; |
155 | } while (td->next_file != old_next_file); | |
53cdc686 JA |
156 | |
157 | return f; | |
158 | } | |
159 | ||
906c8d75 JA |
160 | /* |
161 | * When job exits, we can cancel the in-flight IO if we are using async | |
162 | * io. Attempt to do so. | |
163 | */ | |
ebac4655 JA |
164 | static void cleanup_pending_aio(struct thread_data *td) |
165 | { | |
ebac4655 | 166 | struct list_head *entry, *n; |
ebac4655 JA |
167 | struct io_u *io_u; |
168 | int r; | |
169 | ||
170 | /* | |
171 | * get immediately available events, if any | |
172 | */ | |
97601024 | 173 | io_u_queued_complete(td, 0, NULL); |
ebac4655 JA |
174 | |
175 | /* | |
176 | * now cancel remaining active events | |
177 | */ | |
2866c82d | 178 | if (td->io_ops->cancel) { |
ebac4655 JA |
179 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
180 | io_u = list_entry(entry, struct io_u, list); | |
181 | ||
2866c82d | 182 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
183 | if (!r) |
184 | put_io_u(td, io_u); | |
185 | } | |
186 | } | |
187 | ||
97601024 JA |
188 | if (td->cur_depth) |
189 | io_u_queued_complete(td, td->cur_depth, NULL); | |
ebac4655 JA |
190 | } |
191 | ||
858a3d47 JA |
192 | /* |
193 | * Helper to handle the final sync of a file. Works just like the normal | |
194 | * io path, just does everything sync. | |
195 | */ | |
196 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
197 | { | |
198 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
199 | int ret; |
200 | ||
201 | if (!io_u) | |
202 | return 1; | |
203 | ||
204 | io_u->ddir = DDIR_SYNC; | |
205 | io_u->file = f; | |
206 | ||
207 | if (td_io_prep(td, io_u)) { | |
208 | put_io_u(td, io_u); | |
209 | return 1; | |
210 | } | |
211 | ||
212 | ret = td_io_queue(td, io_u); | |
36167d82 | 213 | if (ret < 0) { |
353a7e0e | 214 | td_verror(td, io_u->error); |
858a3d47 | 215 | put_io_u(td, io_u); |
858a3d47 | 216 | return 1; |
36167d82 | 217 | } else if (ret == FIO_Q_QUEUED) { |
97601024 | 218 | if (io_u_queued_complete(td, 1, NULL)) |
36167d82 | 219 | return 1; |
36167d82 JA |
220 | } else if (ret == FIO_Q_COMPLETED) { |
221 | if (io_u->error) { | |
222 | td_verror(td, io_u->error); | |
223 | return 1; | |
224 | } | |
858a3d47 | 225 | |
97601024 | 226 | io_u_sync_complete(td, io_u, NULL); |
858a3d47 JA |
227 | } |
228 | ||
229 | return 0; | |
230 | } | |
231 | ||
906c8d75 JA |
232 | /* |
233 | * The main verify engine. Runs over the writes we previusly submitted, | |
234 | * reads the blocks back in, and checks the crc/md5 of the data. | |
235 | */ | |
1e97cce9 | 236 | static void do_verify(struct thread_data *td) |
ebac4655 | 237 | { |
53cdc686 | 238 | struct fio_file *f; |
36167d82 | 239 | struct io_u *io_u; |
3af6ef39 | 240 | int ret, i, min_events; |
e5b401d4 JA |
241 | |
242 | /* | |
243 | * sync io first and invalidate cache, to make sure we really | |
244 | * read from disk. | |
245 | */ | |
246 | for_each_file(td, f, i) { | |
858a3d47 | 247 | fio_io_sync(td, f); |
e5b401d4 JA |
248 | file_invalidate_cache(td, f); |
249 | } | |
ebac4655 JA |
250 | |
251 | td_set_runstate(td, TD_VERIFYING); | |
252 | ||
36167d82 JA |
253 | io_u = NULL; |
254 | while (!td->terminate) { | |
ebac4655 JA |
255 | io_u = __get_io_u(td); |
256 | if (!io_u) | |
257 | break; | |
258 | ||
36167d82 | 259 | if (runtime_exceeded(td, &io_u->start_time)) |
02bcaa8c | 260 | break; |
02bcaa8c | 261 | |
36167d82 | 262 | if (get_next_verify(td, io_u)) |
ebac4655 | 263 | break; |
ebac4655 | 264 | |
36167d82 | 265 | if (td_io_prep(td, io_u)) |
53cdc686 JA |
266 | break; |
267 | ||
36167d82 JA |
268 | requeue: |
269 | ret = td_io_queue(td, io_u); | |
53cdc686 | 270 | |
36167d82 JA |
271 | switch (ret) { |
272 | case FIO_Q_COMPLETED: | |
273 | if (io_u->error) | |
22819ec2 | 274 | ret = -io_u->error; |
36167d82 JA |
275 | if (io_u->xfer_buflen != io_u->resid && io_u->resid) { |
276 | int bytes = io_u->xfer_buflen - io_u->resid; | |
ebac4655 | 277 | |
36167d82 JA |
278 | io_u->xfer_buflen = io_u->resid; |
279 | io_u->xfer_buf += bytes; | |
280 | goto requeue; | |
281 | } | |
99784632 JA |
282 | ret = io_u_sync_complete(td, io_u, verify_io_u); |
283 | if (ret) | |
36167d82 | 284 | break; |
36167d82 JA |
285 | continue; |
286 | case FIO_Q_QUEUED: | |
287 | break; | |
288 | default: | |
289 | assert(ret < 0); | |
22819ec2 | 290 | td_verror(td, -ret); |
ebac4655 JA |
291 | break; |
292 | } | |
293 | ||
99784632 | 294 | if (ret < 0 || td->error) |
3af6ef39 JA |
295 | break; |
296 | ||
ebac4655 | 297 | /* |
3af6ef39 JA |
298 | * if we can queue more, do so. but check if there are |
299 | * completed io_u's first. | |
ebac4655 | 300 | */ |
97601024 JA |
301 | min_events = 0; |
302 | if (queue_full(td)) | |
3af6ef39 | 303 | min_events = 1; |
3af6ef39 JA |
304 | |
305 | /* | |
306 | * Reap required number of io units, if any, and do the | |
307 | * verification on them through the callback handler | |
308 | */ | |
97601024 | 309 | if (io_u_queued_complete(td, min_events, verify_io_u)) |
ebac4655 | 310 | break; |
36167d82 | 311 | } |
ebac4655 | 312 | |
36167d82 JA |
313 | if (io_u) |
314 | put_io_u(td, io_u); | |
ebac4655 JA |
315 | |
316 | if (td->cur_depth) | |
317 | cleanup_pending_aio(td); | |
318 | ||
319 | td_set_runstate(td, TD_RUNNING); | |
320 | } | |
321 | ||
b990b5c0 JA |
322 | /* |
323 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
324 | * manner. | |
325 | */ | |
326 | static void do_cpuio(struct thread_data *td) | |
327 | { | |
328 | struct timeval e; | |
329 | int split = 100 / td->cpuload; | |
330 | int i = 0; | |
331 | ||
332 | while (!td->terminate) { | |
02bcaa8c | 333 | fio_gettime(&e, NULL); |
b990b5c0 JA |
334 | |
335 | if (runtime_exceeded(td, &e)) | |
336 | break; | |
337 | ||
338 | if (!(i % split)) | |
339 | __usec_sleep(10000); | |
340 | else | |
341 | usec_sleep(td, 10000); | |
342 | ||
343 | i++; | |
344 | } | |
345 | } | |
346 | ||
32cd46a0 | 347 | /* |
906c8d75 | 348 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
349 | * and reaps them, checking for rate and errors along the way. |
350 | */ | |
ebac4655 JA |
351 | static void do_io(struct thread_data *td) |
352 | { | |
02bcaa8c | 353 | struct timeval s; |
ebac4655 | 354 | unsigned long usec; |
53cdc686 | 355 | struct fio_file *f; |
84585003 | 356 | int i, ret = 0; |
ebac4655 | 357 | |
5853e5a8 JA |
358 | td_set_runstate(td, TD_RUNNING); |
359 | ||
3951414d | 360 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) { |
97601024 JA |
361 | struct timeval comp_time; |
362 | long bytes_done = 0; | |
84585003 | 363 | int min_evts = 0; |
ebac4655 JA |
364 | struct io_u *io_u; |
365 | ||
366 | if (td->terminate) | |
367 | break; | |
368 | ||
53cdc686 JA |
369 | f = get_next_file(td); |
370 | if (!f) | |
371 | break; | |
372 | ||
373 | io_u = get_io_u(td, f); | |
ebac4655 JA |
374 | if (!io_u) |
375 | break; | |
376 | ||
377 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
378 | |
379 | if (runtime_exceeded(td, &s)) { | |
380 | put_io_u(td, io_u); | |
381 | break; | |
382 | } | |
cec6b55d | 383 | requeue: |
45bee283 | 384 | ret = td_io_queue(td, io_u); |
36167d82 JA |
385 | |
386 | switch (ret) { | |
387 | case FIO_Q_COMPLETED: | |
388 | if (io_u->error) { | |
389 | ret = io_u->error; | |
390 | break; | |
391 | } | |
392 | if (io_u->xfer_buflen != io_u->resid && io_u->resid) { | |
393 | int bytes = io_u->xfer_buflen - io_u->resid; | |
394 | ||
cec6b55d | 395 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 396 | io_u->xfer_buf += bytes; |
cec6b55d | 397 | goto requeue; |
cec6b55d | 398 | } |
97601024 JA |
399 | fio_gettime(&comp_time, NULL); |
400 | bytes_done = io_u_sync_complete(td, io_u, NULL); | |
99784632 JA |
401 | if (bytes_done < 0) |
402 | ret = bytes_done; | |
36167d82 JA |
403 | break; |
404 | case FIO_Q_QUEUED: | |
405 | break; | |
406 | default: | |
407 | assert(ret < 0); | |
408 | put_io_u(td, io_u); | |
409 | break; | |
ebac4655 JA |
410 | } |
411 | ||
99784632 | 412 | if (ret < 0 || td->error) |
36167d82 JA |
413 | break; |
414 | ||
ebac4655 JA |
415 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); |
416 | ||
97601024 JA |
417 | /* |
418 | * See if we need to complete some commands | |
419 | */ | |
36167d82 | 420 | if (ret == FIO_Q_QUEUED) { |
97601024 JA |
421 | min_evts = 0; |
422 | if (queue_full(td)) | |
36167d82 | 423 | min_evts = 1; |
ebac4655 | 424 | |
97601024 JA |
425 | fio_gettime(&comp_time, NULL); |
426 | bytes_done = io_u_queued_complete(td, min_evts, NULL); | |
427 | if (bytes_done < 0) | |
36167d82 | 428 | break; |
ebac4655 JA |
429 | } |
430 | ||
97601024 JA |
431 | if (!bytes_done) |
432 | continue; | |
433 | ||
ebac4655 JA |
434 | /* |
435 | * the rate is batched for now, it should work for batches | |
436 | * of completions except the very first one which may look | |
437 | * a little bursty | |
438 | */ | |
97601024 | 439 | usec = utime_since(&s, &comp_time); |
ebac4655 | 440 | |
97601024 | 441 | rate_throttle(td, usec, bytes_done, td->ddir); |
ebac4655 | 442 | |
97601024 | 443 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 444 | if (exitall_on_terminate) |
6ce15a32 | 445 | terminate_threads(td->groupid, 0); |
fd841467 | 446 | td_verror(td, ENODATA); |
ebac4655 JA |
447 | break; |
448 | } | |
449 | ||
9c1f7434 JA |
450 | if (td->thinktime) { |
451 | unsigned long long b; | |
452 | ||
453 | b = td->io_blocks[0] + td->io_blocks[1]; | |
48097d5c JA |
454 | if (!(b % td->thinktime_blocks)) { |
455 | int left; | |
456 | ||
457 | if (td->thinktime_spin) | |
458 | __usec_sleep(td->thinktime_spin); | |
459 | ||
460 | left = td->thinktime - td->thinktime_spin; | |
461 | if (left) | |
462 | usec_sleep(td, left); | |
463 | } | |
9c1f7434 | 464 | } |
ebac4655 JA |
465 | } |
466 | ||
4d2413c6 | 467 | if (!td->error) { |
84585003 JA |
468 | if (td->cur_depth) |
469 | cleanup_pending_aio(td); | |
ebac4655 | 470 | |
84585003 JA |
471 | if (should_fsync(td) && td->end_fsync) { |
472 | td_set_runstate(td, TD_FSYNCING); | |
473 | for_each_file(td, f, i) | |
858a3d47 | 474 | fio_io_sync(td, f); |
84585003 | 475 | } |
5853e5a8 | 476 | } |
ebac4655 JA |
477 | } |
478 | ||
ebac4655 JA |
479 | static void cleanup_io_u(struct thread_data *td) |
480 | { | |
481 | struct list_head *entry, *n; | |
482 | struct io_u *io_u; | |
483 | ||
484 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
485 | io_u = list_entry(entry, struct io_u, list); | |
486 | ||
487 | list_del(&io_u->list); | |
488 | free(io_u); | |
489 | } | |
490 | ||
2f9ade3c | 491 | free_io_mem(td); |
ebac4655 JA |
492 | } |
493 | ||
6b9cea23 JA |
494 | /* |
495 | * "randomly" fill the buffer contents | |
496 | */ | |
66eeb296 | 497 | static void fill_rand_buf(struct io_u *io_u, int max_bs) |
6b9cea23 | 498 | { |
66eeb296 | 499 | int *ptr = io_u->buf; |
6b9cea23 JA |
500 | |
501 | while ((void *) ptr - io_u->buf < max_bs) { | |
502 | *ptr = rand() * 0x9e370001; | |
503 | ptr++; | |
504 | } | |
505 | } | |
506 | ||
ebac4655 JA |
507 | static int init_io_u(struct thread_data *td) |
508 | { | |
509 | struct io_u *io_u; | |
a00735e6 | 510 | unsigned int max_bs; |
ebac4655 JA |
511 | int i, max_units; |
512 | char *p; | |
513 | ||
2866c82d | 514 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
515 | return 0; |
516 | ||
2866c82d | 517 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
518 | max_units = 1; |
519 | else | |
520 | max_units = td->iodepth; | |
521 | ||
a00735e6 | 522 | max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]); |
74b025b0 JA |
523 | td->orig_buffer_size = max_bs * max_units; |
524 | ||
d0bdaf49 | 525 | if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE) |
56bb17f2 | 526 | td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1); |
74b025b0 | 527 | else |
29d610e1 | 528 | td->orig_buffer_size += page_mask; |
ebac4655 | 529 | |
2f9ade3c JA |
530 | if (allocate_io_mem(td)) |
531 | return 1; | |
ebac4655 | 532 | |
ebac4655 JA |
533 | p = ALIGN(td->orig_buffer); |
534 | for (i = 0; i < max_units; i++) { | |
535 | io_u = malloc(sizeof(*io_u)); | |
536 | memset(io_u, 0, sizeof(*io_u)); | |
537 | INIT_LIST_HEAD(&io_u->list); | |
538 | ||
a00735e6 | 539 | io_u->buf = p + max_bs * i; |
6b9cea23 | 540 | if (td_write(td) || td_rw(td)) |
a00735e6 | 541 | fill_rand_buf(io_u, max_bs); |
6b9cea23 | 542 | |
b1ff3403 | 543 | io_u->index = i; |
ebac4655 JA |
544 | list_add(&io_u->list, &td->io_u_freelist); |
545 | } | |
546 | ||
547 | return 0; | |
548 | } | |
549 | ||
da86774e JA |
550 | static int switch_ioscheduler(struct thread_data *td) |
551 | { | |
552 | char tmp[256], tmp2[128]; | |
553 | FILE *f; | |
554 | int ret; | |
555 | ||
f48b467c JA |
556 | if (td->io_ops->flags & FIO_CPUIO) |
557 | return 0; | |
558 | ||
da86774e JA |
559 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
560 | ||
561 | f = fopen(tmp, "r+"); | |
562 | if (!f) { | |
563 | td_verror(td, errno); | |
564 | return 1; | |
565 | } | |
566 | ||
567 | /* | |
568 | * Set io scheduler. | |
569 | */ | |
570 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
571 | if (ferror(f) || ret != 1) { | |
572 | td_verror(td, errno); | |
573 | fclose(f); | |
574 | return 1; | |
575 | } | |
576 | ||
577 | rewind(f); | |
578 | ||
579 | /* | |
580 | * Read back and check that the selected scheduler is now the default. | |
581 | */ | |
582 | ret = fread(tmp, 1, sizeof(tmp), f); | |
583 | if (ferror(f) || ret < 0) { | |
584 | td_verror(td, errno); | |
585 | fclose(f); | |
586 | return 1; | |
587 | } | |
588 | ||
589 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
590 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 591 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
592 | td_verror(td, EINVAL); |
593 | fclose(f); | |
594 | return 1; | |
595 | } | |
596 | ||
597 | fclose(f); | |
598 | return 0; | |
599 | } | |
600 | ||
ebac4655 JA |
601 | static void clear_io_state(struct thread_data *td) |
602 | { | |
53cdc686 JA |
603 | struct fio_file *f; |
604 | int i; | |
ebac4655 | 605 | |
ebac4655 JA |
606 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
607 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 608 | td->zone_bytes = 0; |
ebac4655 | 609 | |
53cdc686 JA |
610 | for_each_file(td, f, i) { |
611 | f->last_pos = 0; | |
612 | if (td->io_ops->flags & FIO_SYNCIO) | |
613 | lseek(f->fd, SEEK_SET, 0); | |
614 | ||
615 | if (f->file_map) | |
616 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
617 | } | |
ebac4655 JA |
618 | } |
619 | ||
906c8d75 JA |
620 | /* |
621 | * Entry point for the thread based jobs. The process based jobs end up | |
622 | * here as well, after a little setup. | |
623 | */ | |
ebac4655 JA |
624 | static void *thread_main(void *data) |
625 | { | |
69008999 | 626 | unsigned long long runtime[2]; |
ebac4655 | 627 | struct thread_data *td = data; |
ebac4655 JA |
628 | |
629 | if (!td->use_thread) | |
630 | setsid(); | |
631 | ||
632 | td->pid = getpid(); | |
633 | ||
aea47d44 JA |
634 | INIT_LIST_HEAD(&td->io_u_freelist); |
635 | INIT_LIST_HEAD(&td->io_u_busylist); | |
636 | INIT_LIST_HEAD(&td->io_hist_list); | |
637 | INIT_LIST_HEAD(&td->io_log_list); | |
638 | ||
ebac4655 JA |
639 | if (init_io_u(td)) |
640 | goto err; | |
641 | ||
642 | if (fio_setaffinity(td) == -1) { | |
643 | td_verror(td, errno); | |
644 | goto err; | |
645 | } | |
646 | ||
aea47d44 JA |
647 | if (init_iolog(td)) |
648 | goto err; | |
649 | ||
ebac4655 JA |
650 | if (td->ioprio) { |
651 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
652 | td_verror(td, errno); | |
653 | goto err; | |
654 | } | |
655 | } | |
656 | ||
1056eaad | 657 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
658 | td_verror(td, errno); |
659 | goto err; | |
660 | } | |
661 | ||
75154845 JA |
662 | if (init_random_state(td)) |
663 | goto err; | |
664 | ||
56f9498d JA |
665 | if (td->ioscheduler && switch_ioscheduler(td)) |
666 | goto err; | |
da86774e | 667 | |
75154845 | 668 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
669 | fio_sem_up(&startup_sem); |
670 | fio_sem_down(&td->mutex); | |
ebac4655 | 671 | |
53cdc686 | 672 | if (!td->create_serialize && setup_files(td)) |
ebac4655 | 673 | goto err; |
21972cde JA |
674 | if (open_files(td)) |
675 | goto err; | |
ebac4655 | 676 | |
7d6c5283 JA |
677 | /* |
678 | * Do this late, as some IO engines would like to have the | |
679 | * files setup prior to initializing structures. | |
680 | */ | |
681 | if (td_io_init(td)) | |
682 | goto err; | |
683 | ||
69cfd7e0 JA |
684 | if (td->exec_prerun) { |
685 | if (system(td->exec_prerun) < 0) | |
686 | goto err; | |
687 | } | |
4e0ba8af | 688 | |
69008999 | 689 | fio_gettime(&td->epoch, NULL); |
36dff966 | 690 | getrusage(RUSAGE_SELF, &td->ru_start); |
69008999 JA |
691 | |
692 | runtime[0] = runtime[1] = 0; | |
ebac4655 | 693 | while (td->loops--) { |
02bcaa8c | 694 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
695 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); |
696 | ||
697 | if (td->ratemin) | |
698 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
699 | ||
700 | clear_io_state(td); | |
701 | prune_io_piece_log(td); | |
702 | ||
2866c82d | 703 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
704 | do_cpuio(td); |
705 | else | |
706 | do_io(td); | |
ebac4655 | 707 | |
69008999 | 708 | runtime[td->ddir] += utime_since_now(&td->start); |
aea47d44 | 709 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
69008999 | 710 | runtime[td->ddir ^ 1] = runtime[td->ddir]; |
3d60d1ed | 711 | |
ebac4655 JA |
712 | if (td->error || td->terminate) |
713 | break; | |
714 | ||
715 | if (td->verify == VERIFY_NONE) | |
716 | continue; | |
717 | ||
718 | clear_io_state(td); | |
02bcaa8c | 719 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
720 | |
721 | do_verify(td); | |
722 | ||
69008999 | 723 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
724 | |
725 | if (td->error || td->terminate) | |
726 | break; | |
727 | } | |
728 | ||
36dff966 | 729 | update_rusage_stat(td); |
69008999 JA |
730 | fio_gettime(&td->end_time, NULL); |
731 | td->runtime[0] = runtime[0] / 1000; | |
732 | td->runtime[1] = runtime[1] / 1000; | |
733 | ||
ebac4655 JA |
734 | if (td->bw_log) |
735 | finish_log(td, td->bw_log, "bw"); | |
736 | if (td->slat_log) | |
737 | finish_log(td, td->slat_log, "slat"); | |
738 | if (td->clat_log) | |
739 | finish_log(td, td->clat_log, "clat"); | |
076efc7c | 740 | if (td->write_iolog_file) |
843a7413 | 741 | write_iolog_close(td); |
69cfd7e0 JA |
742 | if (td->exec_postrun) { |
743 | if (system(td->exec_postrun) < 0) | |
744 | log_err("fio: postrun %s failed\n", td->exec_postrun); | |
745 | } | |
ebac4655 JA |
746 | |
747 | if (exitall_on_terminate) | |
6ce15a32 | 748 | terminate_threads(td->groupid, 0); |
ebac4655 JA |
749 | |
750 | err: | |
5bf13a5a JA |
751 | if (td->error) |
752 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
53cdc686 | 753 | close_files(td); |
2866c82d | 754 | close_ioengine(td); |
ebac4655 | 755 | cleanup_io_u(td); |
ebac4655 | 756 | td_set_runstate(td, TD_EXITED); |
a6418147 | 757 | return (void *) td->error; |
ebac4655 JA |
758 | } |
759 | ||
906c8d75 JA |
760 | /* |
761 | * We cannot pass the td data into a forked process, so attach the td and | |
762 | * pass it to the thread worker. | |
763 | */ | |
a6418147 | 764 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
765 | { |
766 | struct thread_data *td; | |
a6418147 | 767 | void *data, *ret; |
ebac4655 JA |
768 | |
769 | data = shmat(shmid, NULL, 0); | |
770 | if (data == (void *) -1) { | |
a6418147 JA |
771 | int __err = errno; |
772 | ||
ebac4655 | 773 | perror("shmat"); |
a6418147 | 774 | return __err; |
ebac4655 JA |
775 | } |
776 | ||
777 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 778 | ret = thread_main(td); |
ebac4655 | 779 | shmdt(data); |
a6418147 | 780 | return (int) ret; |
ebac4655 JA |
781 | } |
782 | ||
906c8d75 JA |
783 | /* |
784 | * Run over the job map and reap the threads that have exited, if any. | |
785 | */ | |
ebac4655 JA |
786 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
787 | { | |
34572e28 | 788 | struct thread_data *td; |
fab6aa71 | 789 | int i, cputhreads, pending, status, ret; |
ebac4655 JA |
790 | |
791 | /* | |
792 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
793 | */ | |
4d2413c6 | 794 | pending = cputhreads = 0; |
34572e28 | 795 | for_each_td(td, i) { |
84585003 JA |
796 | /* |
797 | * ->io_ops is NULL for a thread that has closed its | |
798 | * io engine | |
799 | */ | |
800 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
801 | cputhreads++; |
802 | ||
fab6aa71 JA |
803 | if (td->runstate < TD_EXITED) { |
804 | /* | |
805 | * check if someone quit or got killed in an unusual way | |
806 | */ | |
807 | ret = waitpid(td->pid, &status, WNOHANG); | |
808 | if (ret < 0) | |
809 | perror("waitpid"); | |
810 | else if ((ret == td->pid) && WIFSIGNALED(status)) { | |
811 | int sig = WTERMSIG(status); | |
812 | ||
813 | log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); | |
814 | td_set_runstate(td, TD_REAPED); | |
815 | goto reaped; | |
816 | } | |
817 | } | |
818 | ||
4d2413c6 JA |
819 | if (td->runstate != TD_EXITED) { |
820 | if (td->runstate < TD_RUNNING) | |
821 | pending++; | |
822 | ||
ebac4655 | 823 | continue; |
4d2413c6 | 824 | } |
ebac4655 | 825 | |
437c9b71 JA |
826 | if (td->error) |
827 | exit_value++; | |
828 | ||
ebac4655 JA |
829 | td_set_runstate(td, TD_REAPED); |
830 | ||
831 | if (td->use_thread) { | |
832 | long ret; | |
833 | ||
834 | if (pthread_join(td->thread, (void *) &ret)) | |
835 | perror("thread_join"); | |
a6418147 JA |
836 | } else { |
837 | int status; | |
838 | ||
fab6aa71 JA |
839 | ret = waitpid(td->pid, &status, 0); |
840 | if (ret < 0) | |
841 | perror("waitpid"); | |
73170f19 | 842 | else if (WIFEXITED(status) && WEXITSTATUS(status)) { |
a6418147 JA |
843 | if (!exit_value) |
844 | exit_value++; | |
845 | } | |
846 | } | |
ebac4655 | 847 | |
fab6aa71 | 848 | reaped: |
ebac4655 JA |
849 | (*nr_running)--; |
850 | (*m_rate) -= td->ratemin; | |
851 | (*t_rate) -= td->rate; | |
852 | } | |
b990b5c0 | 853 | |
4d2413c6 | 854 | if (*nr_running == cputhreads && !pending) |
6ce15a32 | 855 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
856 | } |
857 | ||
906c8d75 JA |
858 | /* |
859 | * Main function for kicking off and reaping jobs, as needed. | |
860 | */ | |
ebac4655 JA |
861 | static void run_threads(void) |
862 | { | |
ebac4655 JA |
863 | struct thread_data *td; |
864 | unsigned long spent; | |
865 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 866 | |
2f9ade3c JA |
867 | if (fio_pin_memory()) |
868 | return; | |
ebac4655 | 869 | |
c6ae0a5b JA |
870 | if (!terse_output) { |
871 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
872 | fflush(stdout); | |
873 | } | |
c04f7ec3 | 874 | |
4efa970e JA |
875 | signal(SIGINT, sig_handler); |
876 | signal(SIGALRM, sig_handler); | |
877 | ||
ebac4655 JA |
878 | todo = thread_number; |
879 | nr_running = 0; | |
880 | nr_started = 0; | |
881 | m_rate = t_rate = 0; | |
882 | ||
34572e28 | 883 | for_each_td(td, i) { |
263e529f | 884 | print_status_init(td->thread_number - 1); |
ebac4655 | 885 | |
380cf265 JA |
886 | if (!td->create_serialize) { |
887 | init_disk_util(td); | |
ebac4655 | 888 | continue; |
380cf265 | 889 | } |
ebac4655 JA |
890 | |
891 | /* | |
892 | * do file setup here so it happens sequentially, | |
893 | * we don't want X number of threads getting their | |
894 | * client data interspersed on disk | |
895 | */ | |
53cdc686 | 896 | if (setup_files(td)) { |
5bf13a5a JA |
897 | exit_value++; |
898 | if (td->error) | |
899 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
ebac4655 JA |
900 | td_set_runstate(td, TD_REAPED); |
901 | todo--; | |
902 | } | |
380cf265 JA |
903 | |
904 | init_disk_util(td); | |
ebac4655 JA |
905 | } |
906 | ||
ebac4655 | 907 | while (todo) { |
75154845 JA |
908 | struct thread_data *map[MAX_JOBS]; |
909 | struct timeval this_start; | |
910 | int this_jobs = 0, left; | |
911 | ||
ebac4655 JA |
912 | /* |
913 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
914 | */ | |
34572e28 | 915 | for_each_td(td, i) { |
ebac4655 JA |
916 | if (td->runstate != TD_NOT_CREATED) |
917 | continue; | |
918 | ||
919 | /* | |
920 | * never got a chance to start, killed by other | |
921 | * thread for some reason | |
922 | */ | |
923 | if (td->terminate) { | |
924 | todo--; | |
925 | continue; | |
926 | } | |
927 | ||
928 | if (td->start_delay) { | |
263e529f | 929 | spent = mtime_since_genesis(); |
ebac4655 JA |
930 | |
931 | if (td->start_delay * 1000 > spent) | |
932 | continue; | |
933 | } | |
934 | ||
935 | if (td->stonewall && (nr_started || nr_running)) | |
936 | break; | |
937 | ||
75154845 JA |
938 | /* |
939 | * Set state to created. Thread will transition | |
940 | * to TD_INITIALIZED when it's done setting up. | |
941 | */ | |
ebac4655 | 942 | td_set_runstate(td, TD_CREATED); |
75154845 | 943 | map[this_jobs++] = td; |
bbfd6b00 | 944 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
945 | nr_started++; |
946 | ||
947 | if (td->use_thread) { | |
948 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
949 | perror("thread_create"); | |
950 | nr_started--; | |
951 | } | |
952 | } else { | |
953 | if (fork()) | |
bbfd6b00 | 954 | fio_sem_down(&startup_sem); |
ebac4655 | 955 | else { |
a6418147 JA |
956 | int ret = fork_main(shm_id, i); |
957 | ||
958 | exit(ret); | |
ebac4655 JA |
959 | } |
960 | } | |
961 | } | |
962 | ||
963 | /* | |
75154845 JA |
964 | * Wait for the started threads to transition to |
965 | * TD_INITIALIZED. | |
ebac4655 | 966 | */ |
02bcaa8c | 967 | fio_gettime(&this_start, NULL); |
75154845 | 968 | left = this_jobs; |
6ce15a32 | 969 | while (left && !fio_abort) { |
75154845 JA |
970 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
971 | break; | |
972 | ||
973 | usleep(100000); | |
974 | ||
975 | for (i = 0; i < this_jobs; i++) { | |
976 | td = map[i]; | |
977 | if (!td) | |
978 | continue; | |
b6f4d880 | 979 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
980 | map[i] = NULL; |
981 | left--; | |
b6f4d880 JA |
982 | } else if (td->runstate >= TD_EXITED) { |
983 | map[i] = NULL; | |
984 | left--; | |
985 | todo--; | |
986 | nr_running++; /* work-around... */ | |
75154845 JA |
987 | } |
988 | } | |
989 | } | |
990 | ||
991 | if (left) { | |
3b70d7e5 | 992 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
993 | for (i = 0; i < this_jobs; i++) { |
994 | td = map[i]; | |
995 | if (!td) | |
996 | continue; | |
997 | kill(td->pid, SIGTERM); | |
998 | } | |
999 | break; | |
1000 | } | |
1001 | ||
1002 | /* | |
b6f4d880 | 1003 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1004 | */ |
34572e28 | 1005 | for_each_td(td, i) { |
75154845 | 1006 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1007 | continue; |
1008 | ||
1009 | td_set_runstate(td, TD_RUNNING); | |
1010 | nr_running++; | |
1011 | nr_started--; | |
1012 | m_rate += td->ratemin; | |
1013 | t_rate += td->rate; | |
75154845 | 1014 | todo--; |
bbfd6b00 | 1015 | fio_sem_up(&td->mutex); |
ebac4655 JA |
1016 | } |
1017 | ||
1018 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1019 | ||
1020 | if (todo) | |
1021 | usleep(100000); | |
1022 | } | |
1023 | ||
1024 | while (nr_running) { | |
1025 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1026 | usleep(10000); | |
1027 | } | |
1028 | ||
1029 | update_io_ticks(); | |
2f9ade3c | 1030 | fio_unpin_memory(); |
ebac4655 JA |
1031 | } |
1032 | ||
ebac4655 JA |
1033 | int main(int argc, char *argv[]) |
1034 | { | |
29d610e1 JA |
1035 | long ps; |
1036 | ||
dbe1125e JA |
1037 | /* |
1038 | * We need locale for number printing, if it isn't set then just | |
1039 | * go with the US format. | |
1040 | */ | |
1041 | if (!getenv("LC_NUMERIC")) | |
1042 | setlocale(LC_NUMERIC, "en_US"); | |
1043 | ||
ebac4655 JA |
1044 | if (parse_options(argc, argv)) |
1045 | return 1; | |
1046 | ||
1047 | if (!thread_number) { | |
3b70d7e5 | 1048 | log_err("Nothing to do\n"); |
ebac4655 JA |
1049 | return 1; |
1050 | } | |
1051 | ||
29d610e1 JA |
1052 | ps = sysconf(_SC_PAGESIZE); |
1053 | if (ps < 0) { | |
1054 | log_err("Failed to get page size\n"); | |
1055 | return 1; | |
1056 | } | |
1057 | ||
1058 | page_mask = ps - 1; | |
1059 | ||
bb3884d8 JA |
1060 | if (write_bw_log) { |
1061 | setup_log(&agg_io_log[DDIR_READ]); | |
1062 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1063 | } | |
1064 | ||
ebac4655 JA |
1065 | disk_util_timer_arm(); |
1066 | ||
1067 | run_threads(); | |
6ce15a32 | 1068 | |
bb3884d8 | 1069 | if (!fio_abort) { |
6ce15a32 | 1070 | show_run_stats(); |
bb3884d8 JA |
1071 | if (write_bw_log) { |
1072 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1073 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1074 | } | |
1075 | } | |
ebac4655 | 1076 | |
437c9b71 | 1077 | return exit_value; |
ebac4655 | 1078 | } |