Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
aae22ca7 | 5 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
ebac4655 | 6 | * |
8e9fe637 JA |
7 | * The license below covers all files distributed with fio unless otherwise |
8 | * noted in the file itself. | |
9 | * | |
ebac4655 | 10 | * This program is free software; you can redistribute it and/or modify |
8e9fe637 JA |
11 | * it under the terms of the GNU General Public License version 2 as |
12 | * published by the Free Software Foundation. | |
ebac4655 JA |
13 | * |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
21 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
22 | * | |
23 | */ | |
ebac4655 JA |
24 | #include <unistd.h> |
25 | #include <fcntl.h> | |
26 | #include <string.h> | |
ebac4655 JA |
27 | #include <signal.h> |
28 | #include <time.h> | |
dbe1125e | 29 | #include <locale.h> |
36167d82 | 30 | #include <assert.h> |
ebac4655 JA |
31 | #include <sys/stat.h> |
32 | #include <sys/wait.h> | |
33 | #include <sys/ipc.h> | |
34 | #include <sys/shm.h> | |
ebac4655 JA |
35 | #include <sys/mman.h> |
36 | ||
37 | #include "fio.h" | |
38 | #include "os.h" | |
39 | ||
29d610e1 JA |
40 | static unsigned long page_mask; |
41 | #define ALIGN(buf) \ | |
42 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) | |
ebac4655 JA |
43 | |
44 | int groupid = 0; | |
45 | int thread_number = 0; | |
ebac4655 | 46 | int shm_id = 0; |
53cdc686 | 47 | int temp_stall_ts; |
ebac4655 | 48 | |
bbfd6b00 | 49 | static volatile int startup_sem; |
6ce15a32 | 50 | static volatile int fio_abort; |
437c9b71 | 51 | static int exit_value; |
ebac4655 | 52 | |
bb3884d8 JA |
53 | struct io_log *agg_io_log[2]; |
54 | ||
ebac4655 | 55 | #define TERMINATE_ALL (-1) |
75154845 | 56 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 57 | |
6ce15a32 JA |
58 | static inline void td_set_runstate(struct thread_data *td, int runstate) |
59 | { | |
60 | td->runstate = runstate; | |
61 | } | |
62 | ||
63 | static void terminate_threads(int group_id, int forced_kill) | |
ebac4655 | 64 | { |
34572e28 | 65 | struct thread_data *td; |
ebac4655 JA |
66 | int i; |
67 | ||
34572e28 | 68 | for_each_td(td, i) { |
ebac4655 JA |
69 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
70 | td->terminate = 1; | |
71 | td->start_delay = 0; | |
6ce15a32 JA |
72 | if (forced_kill) |
73 | td_set_runstate(td, TD_EXITED); | |
ebac4655 JA |
74 | } |
75 | } | |
76 | } | |
77 | ||
78 | static void sig_handler(int sig) | |
79 | { | |
80 | switch (sig) { | |
81 | case SIGALRM: | |
82 | update_io_ticks(); | |
83 | disk_util_timer_arm(); | |
84 | print_thread_status(); | |
85 | break; | |
86 | default: | |
6ce15a32 | 87 | printf("\nfio: terminating on signal %d\n", sig); |
ebac4655 | 88 | fflush(stdout); |
6ce15a32 | 89 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
90 | break; |
91 | } | |
92 | } | |
93 | ||
906c8d75 JA |
94 | /* |
95 | * Check if we are above the minimum rate given. | |
96 | */ | |
ebac4655 JA |
97 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
98 | { | |
99 | unsigned long spent; | |
100 | unsigned long rate; | |
101 | int ddir = td->ddir; | |
102 | ||
103 | /* | |
104 | * allow a 2 second settle period in the beginning | |
105 | */ | |
106 | if (mtime_since(&td->start, now) < 2000) | |
107 | return 0; | |
108 | ||
109 | /* | |
110 | * if rate blocks is set, sample is running | |
111 | */ | |
112 | if (td->rate_bytes) { | |
113 | spent = mtime_since(&td->lastrate, now); | |
114 | if (spent < td->ratecycle) | |
115 | return 0; | |
116 | ||
117 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
118 | if (rate < td->ratemin) { | |
1e97cce9 | 119 | fprintf(f_out, "%s: min rate %u not met, got %luKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
120 | return 1; |
121 | } | |
122 | } | |
123 | ||
124 | td->rate_bytes = td->this_io_bytes[ddir]; | |
125 | memcpy(&td->lastrate, now, sizeof(*now)); | |
126 | return 0; | |
127 | } | |
128 | ||
129 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
130 | { | |
131 | if (!td->timeout) | |
132 | return 0; | |
133 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
134 | return 1; | |
135 | ||
136 | return 0; | |
137 | } | |
138 | ||
906c8d75 JA |
139 | /* |
140 | * When job exits, we can cancel the in-flight IO if we are using async | |
141 | * io. Attempt to do so. | |
142 | */ | |
ebac4655 JA |
143 | static void cleanup_pending_aio(struct thread_data *td) |
144 | { | |
ebac4655 | 145 | struct list_head *entry, *n; |
ebac4655 JA |
146 | struct io_u *io_u; |
147 | int r; | |
148 | ||
149 | /* | |
150 | * get immediately available events, if any | |
151 | */ | |
97601024 | 152 | io_u_queued_complete(td, 0, NULL); |
ebac4655 JA |
153 | |
154 | /* | |
155 | * now cancel remaining active events | |
156 | */ | |
2866c82d | 157 | if (td->io_ops->cancel) { |
ebac4655 JA |
158 | list_for_each_safe(entry, n, &td->io_u_busylist) { |
159 | io_u = list_entry(entry, struct io_u, list); | |
160 | ||
2866c82d | 161 | r = td->io_ops->cancel(td, io_u); |
ebac4655 JA |
162 | if (!r) |
163 | put_io_u(td, io_u); | |
164 | } | |
165 | } | |
166 | ||
97601024 JA |
167 | if (td->cur_depth) |
168 | io_u_queued_complete(td, td->cur_depth, NULL); | |
ebac4655 JA |
169 | } |
170 | ||
858a3d47 JA |
171 | /* |
172 | * Helper to handle the final sync of a file. Works just like the normal | |
173 | * io path, just does everything sync. | |
174 | */ | |
175 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
176 | { | |
177 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
178 | int ret; |
179 | ||
180 | if (!io_u) | |
181 | return 1; | |
182 | ||
183 | io_u->ddir = DDIR_SYNC; | |
184 | io_u->file = f; | |
185 | ||
186 | if (td_io_prep(td, io_u)) { | |
187 | put_io_u(td, io_u); | |
188 | return 1; | |
189 | } | |
190 | ||
755200a3 | 191 | requeue: |
858a3d47 | 192 | ret = td_io_queue(td, io_u); |
36167d82 | 193 | if (ret < 0) { |
353a7e0e | 194 | td_verror(td, io_u->error); |
858a3d47 | 195 | put_io_u(td, io_u); |
858a3d47 | 196 | return 1; |
36167d82 | 197 | } else if (ret == FIO_Q_QUEUED) { |
97601024 | 198 | if (io_u_queued_complete(td, 1, NULL)) |
36167d82 | 199 | return 1; |
36167d82 JA |
200 | } else if (ret == FIO_Q_COMPLETED) { |
201 | if (io_u->error) { | |
202 | td_verror(td, io_u->error); | |
203 | return 1; | |
204 | } | |
858a3d47 | 205 | |
97601024 | 206 | io_u_sync_complete(td, io_u, NULL); |
755200a3 JA |
207 | } else if (ret == FIO_Q_BUSY) { |
208 | if (td_io_commit(td)) | |
209 | return 1; | |
210 | goto requeue; | |
858a3d47 JA |
211 | } |
212 | ||
213 | return 0; | |
214 | } | |
215 | ||
906c8d75 JA |
216 | /* |
217 | * The main verify engine. Runs over the writes we previusly submitted, | |
218 | * reads the blocks back in, and checks the crc/md5 of the data. | |
219 | */ | |
1e97cce9 | 220 | static void do_verify(struct thread_data *td) |
ebac4655 | 221 | { |
53cdc686 | 222 | struct fio_file *f; |
36167d82 | 223 | struct io_u *io_u; |
3af6ef39 | 224 | int ret, i, min_events; |
e5b401d4 JA |
225 | |
226 | /* | |
227 | * sync io first and invalidate cache, to make sure we really | |
228 | * read from disk. | |
229 | */ | |
230 | for_each_file(td, f, i) { | |
858a3d47 | 231 | fio_io_sync(td, f); |
e5b401d4 JA |
232 | file_invalidate_cache(td, f); |
233 | } | |
ebac4655 JA |
234 | |
235 | td_set_runstate(td, TD_VERIFYING); | |
236 | ||
36167d82 JA |
237 | io_u = NULL; |
238 | while (!td->terminate) { | |
ebac4655 JA |
239 | io_u = __get_io_u(td); |
240 | if (!io_u) | |
241 | break; | |
242 | ||
36167d82 | 243 | if (runtime_exceeded(td, &io_u->start_time)) |
02bcaa8c | 244 | break; |
02bcaa8c | 245 | |
36167d82 | 246 | if (get_next_verify(td, io_u)) |
ebac4655 | 247 | break; |
ebac4655 | 248 | |
36167d82 | 249 | if (td_io_prep(td, io_u)) |
53cdc686 JA |
250 | break; |
251 | ||
36167d82 JA |
252 | requeue: |
253 | ret = td_io_queue(td, io_u); | |
53cdc686 | 254 | |
36167d82 JA |
255 | switch (ret) { |
256 | case FIO_Q_COMPLETED: | |
257 | if (io_u->error) | |
22819ec2 | 258 | ret = -io_u->error; |
36167d82 JA |
259 | if (io_u->xfer_buflen != io_u->resid && io_u->resid) { |
260 | int bytes = io_u->xfer_buflen - io_u->resid; | |
ebac4655 | 261 | |
36167d82 JA |
262 | io_u->xfer_buflen = io_u->resid; |
263 | io_u->xfer_buf += bytes; | |
264 | goto requeue; | |
265 | } | |
99784632 JA |
266 | ret = io_u_sync_complete(td, io_u, verify_io_u); |
267 | if (ret) | |
36167d82 | 268 | break; |
36167d82 JA |
269 | continue; |
270 | case FIO_Q_QUEUED: | |
271 | break; | |
755200a3 JA |
272 | case FIO_Q_BUSY: |
273 | requeue_io_u(td, &io_u); | |
274 | ret = td_io_commit(td); | |
275 | break; | |
36167d82 JA |
276 | default: |
277 | assert(ret < 0); | |
22819ec2 | 278 | td_verror(td, -ret); |
ebac4655 JA |
279 | break; |
280 | } | |
281 | ||
99784632 | 282 | if (ret < 0 || td->error) |
3af6ef39 JA |
283 | break; |
284 | ||
ebac4655 | 285 | /* |
3af6ef39 JA |
286 | * if we can queue more, do so. but check if there are |
287 | * completed io_u's first. | |
ebac4655 | 288 | */ |
97601024 | 289 | min_events = 0; |
755200a3 | 290 | if (queue_full(td) || ret == FIO_Q_BUSY) |
3af6ef39 | 291 | min_events = 1; |
3af6ef39 JA |
292 | |
293 | /* | |
294 | * Reap required number of io units, if any, and do the | |
295 | * verification on them through the callback handler | |
296 | */ | |
97601024 | 297 | if (io_u_queued_complete(td, min_events, verify_io_u)) |
ebac4655 | 298 | break; |
36167d82 | 299 | } |
ebac4655 | 300 | |
36167d82 JA |
301 | if (io_u) |
302 | put_io_u(td, io_u); | |
ebac4655 JA |
303 | |
304 | if (td->cur_depth) | |
305 | cleanup_pending_aio(td); | |
306 | ||
307 | td_set_runstate(td, TD_RUNNING); | |
308 | } | |
309 | ||
b990b5c0 JA |
310 | /* |
311 | * Not really an io thread, all it does is burn CPU cycles in the specified | |
312 | * manner. | |
313 | */ | |
314 | static void do_cpuio(struct thread_data *td) | |
315 | { | |
316 | struct timeval e; | |
317 | int split = 100 / td->cpuload; | |
318 | int i = 0; | |
319 | ||
320 | while (!td->terminate) { | |
02bcaa8c | 321 | fio_gettime(&e, NULL); |
b990b5c0 JA |
322 | |
323 | if (runtime_exceeded(td, &e)) | |
324 | break; | |
325 | ||
326 | if (!(i % split)) | |
327 | __usec_sleep(10000); | |
328 | else | |
329 | usec_sleep(td, 10000); | |
330 | ||
331 | i++; | |
332 | } | |
333 | } | |
334 | ||
32cd46a0 | 335 | /* |
906c8d75 | 336 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
337 | * and reaps them, checking for rate and errors along the way. |
338 | */ | |
ebac4655 JA |
339 | static void do_io(struct thread_data *td) |
340 | { | |
02bcaa8c | 341 | struct timeval s; |
ebac4655 | 342 | unsigned long usec; |
84585003 | 343 | int i, ret = 0; |
ebac4655 | 344 | |
5853e5a8 JA |
345 | td_set_runstate(td, TD_RUNNING); |
346 | ||
3951414d | 347 | while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) { |
97601024 JA |
348 | struct timeval comp_time; |
349 | long bytes_done = 0; | |
84585003 | 350 | int min_evts = 0; |
ebac4655 JA |
351 | struct io_u *io_u; |
352 | ||
353 | if (td->terminate) | |
354 | break; | |
355 | ||
3d7c391d | 356 | io_u = get_io_u(td); |
ebac4655 JA |
357 | if (!io_u) |
358 | break; | |
359 | ||
360 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
97601024 JA |
361 | |
362 | if (runtime_exceeded(td, &s)) { | |
363 | put_io_u(td, io_u); | |
364 | break; | |
365 | } | |
cec6b55d | 366 | requeue: |
45bee283 | 367 | ret = td_io_queue(td, io_u); |
36167d82 JA |
368 | |
369 | switch (ret) { | |
370 | case FIO_Q_COMPLETED: | |
371 | if (io_u->error) { | |
372 | ret = io_u->error; | |
373 | break; | |
374 | } | |
375 | if (io_u->xfer_buflen != io_u->resid && io_u->resid) { | |
376 | int bytes = io_u->xfer_buflen - io_u->resid; | |
377 | ||
cec6b55d | 378 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 379 | io_u->xfer_buf += bytes; |
cec6b55d | 380 | goto requeue; |
cec6b55d | 381 | } |
97601024 JA |
382 | fio_gettime(&comp_time, NULL); |
383 | bytes_done = io_u_sync_complete(td, io_u, NULL); | |
99784632 JA |
384 | if (bytes_done < 0) |
385 | ret = bytes_done; | |
36167d82 JA |
386 | break; |
387 | case FIO_Q_QUEUED: | |
388 | break; | |
755200a3 JA |
389 | case FIO_Q_BUSY: |
390 | requeue_io_u(td, &io_u); | |
391 | ret = td_io_commit(td); | |
392 | break; | |
36167d82 JA |
393 | default: |
394 | assert(ret < 0); | |
395 | put_io_u(td, io_u); | |
396 | break; | |
ebac4655 JA |
397 | } |
398 | ||
99784632 | 399 | if (ret < 0 || td->error) |
36167d82 JA |
400 | break; |
401 | ||
755200a3 JA |
402 | if (io_u) |
403 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
ebac4655 | 404 | |
97601024 JA |
405 | /* |
406 | * See if we need to complete some commands | |
407 | */ | |
755200a3 | 408 | if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) { |
97601024 | 409 | min_evts = 0; |
755200a3 | 410 | if (queue_full(td) || ret == FIO_Q_BUSY) |
36167d82 | 411 | min_evts = 1; |
ebac4655 | 412 | |
97601024 JA |
413 | fio_gettime(&comp_time, NULL); |
414 | bytes_done = io_u_queued_complete(td, min_evts, NULL); | |
415 | if (bytes_done < 0) | |
36167d82 | 416 | break; |
ebac4655 JA |
417 | } |
418 | ||
97601024 JA |
419 | if (!bytes_done) |
420 | continue; | |
421 | ||
ebac4655 JA |
422 | /* |
423 | * the rate is batched for now, it should work for batches | |
424 | * of completions except the very first one which may look | |
425 | * a little bursty | |
426 | */ | |
97601024 | 427 | usec = utime_since(&s, &comp_time); |
ebac4655 | 428 | |
97601024 | 429 | rate_throttle(td, usec, bytes_done, td->ddir); |
ebac4655 | 430 | |
97601024 | 431 | if (check_min_rate(td, &comp_time)) { |
98aa62d8 | 432 | if (exitall_on_terminate) |
6ce15a32 | 433 | terminate_threads(td->groupid, 0); |
fd841467 | 434 | td_verror(td, ENODATA); |
ebac4655 JA |
435 | break; |
436 | } | |
437 | ||
9c1f7434 JA |
438 | if (td->thinktime) { |
439 | unsigned long long b; | |
440 | ||
441 | b = td->io_blocks[0] + td->io_blocks[1]; | |
48097d5c JA |
442 | if (!(b % td->thinktime_blocks)) { |
443 | int left; | |
444 | ||
445 | if (td->thinktime_spin) | |
446 | __usec_sleep(td->thinktime_spin); | |
447 | ||
448 | left = td->thinktime - td->thinktime_spin; | |
449 | if (left) | |
450 | usec_sleep(td, left); | |
451 | } | |
9c1f7434 | 452 | } |
ebac4655 JA |
453 | } |
454 | ||
4d2413c6 | 455 | if (!td->error) { |
3d7c391d JA |
456 | struct fio_file *f; |
457 | ||
84585003 JA |
458 | if (td->cur_depth) |
459 | cleanup_pending_aio(td); | |
ebac4655 | 460 | |
84585003 JA |
461 | if (should_fsync(td) && td->end_fsync) { |
462 | td_set_runstate(td, TD_FSYNCING); | |
463 | for_each_file(td, f, i) | |
858a3d47 | 464 | fio_io_sync(td, f); |
84585003 | 465 | } |
5853e5a8 | 466 | } |
ebac4655 JA |
467 | } |
468 | ||
ebac4655 JA |
469 | static void cleanup_io_u(struct thread_data *td) |
470 | { | |
471 | struct list_head *entry, *n; | |
472 | struct io_u *io_u; | |
473 | ||
474 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
475 | io_u = list_entry(entry, struct io_u, list); | |
476 | ||
477 | list_del(&io_u->list); | |
478 | free(io_u); | |
479 | } | |
480 | ||
2f9ade3c | 481 | free_io_mem(td); |
ebac4655 JA |
482 | } |
483 | ||
6b9cea23 JA |
484 | /* |
485 | * "randomly" fill the buffer contents | |
486 | */ | |
66eeb296 | 487 | static void fill_rand_buf(struct io_u *io_u, int max_bs) |
6b9cea23 | 488 | { |
66eeb296 | 489 | int *ptr = io_u->buf; |
6b9cea23 JA |
490 | |
491 | while ((void *) ptr - io_u->buf < max_bs) { | |
492 | *ptr = rand() * 0x9e370001; | |
493 | ptr++; | |
494 | } | |
495 | } | |
496 | ||
ebac4655 JA |
497 | static int init_io_u(struct thread_data *td) |
498 | { | |
499 | struct io_u *io_u; | |
a00735e6 | 500 | unsigned int max_bs; |
ebac4655 JA |
501 | int i, max_units; |
502 | char *p; | |
503 | ||
2866c82d | 504 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
505 | return 0; |
506 | ||
2866c82d | 507 | if (td->io_ops->flags & FIO_SYNCIO) |
ebac4655 JA |
508 | max_units = 1; |
509 | else | |
510 | max_units = td->iodepth; | |
511 | ||
a00735e6 | 512 | max_bs = max(td->max_bs[DDIR_READ], td->max_bs[DDIR_WRITE]); |
74b025b0 JA |
513 | td->orig_buffer_size = max_bs * max_units; |
514 | ||
d0bdaf49 | 515 | if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE) |
56bb17f2 | 516 | td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1); |
74b025b0 | 517 | else |
29d610e1 | 518 | td->orig_buffer_size += page_mask; |
ebac4655 | 519 | |
2f9ade3c JA |
520 | if (allocate_io_mem(td)) |
521 | return 1; | |
ebac4655 | 522 | |
ebac4655 JA |
523 | p = ALIGN(td->orig_buffer); |
524 | for (i = 0; i < max_units; i++) { | |
525 | io_u = malloc(sizeof(*io_u)); | |
526 | memset(io_u, 0, sizeof(*io_u)); | |
527 | INIT_LIST_HEAD(&io_u->list); | |
528 | ||
a00735e6 | 529 | io_u->buf = p + max_bs * i; |
6b9cea23 | 530 | if (td_write(td) || td_rw(td)) |
a00735e6 | 531 | fill_rand_buf(io_u, max_bs); |
6b9cea23 | 532 | |
b1ff3403 | 533 | io_u->index = i; |
ebac4655 JA |
534 | list_add(&io_u->list, &td->io_u_freelist); |
535 | } | |
536 | ||
537 | return 0; | |
538 | } | |
539 | ||
da86774e JA |
540 | static int switch_ioscheduler(struct thread_data *td) |
541 | { | |
542 | char tmp[256], tmp2[128]; | |
543 | FILE *f; | |
544 | int ret; | |
545 | ||
f48b467c JA |
546 | if (td->io_ops->flags & FIO_CPUIO) |
547 | return 0; | |
548 | ||
da86774e JA |
549 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
550 | ||
551 | f = fopen(tmp, "r+"); | |
552 | if (!f) { | |
553 | td_verror(td, errno); | |
554 | return 1; | |
555 | } | |
556 | ||
557 | /* | |
558 | * Set io scheduler. | |
559 | */ | |
560 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
561 | if (ferror(f) || ret != 1) { | |
562 | td_verror(td, errno); | |
563 | fclose(f); | |
564 | return 1; | |
565 | } | |
566 | ||
567 | rewind(f); | |
568 | ||
569 | /* | |
570 | * Read back and check that the selected scheduler is now the default. | |
571 | */ | |
572 | ret = fread(tmp, 1, sizeof(tmp), f); | |
573 | if (ferror(f) || ret < 0) { | |
574 | td_verror(td, errno); | |
575 | fclose(f); | |
576 | return 1; | |
577 | } | |
578 | ||
579 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
580 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 581 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
582 | td_verror(td, EINVAL); |
583 | fclose(f); | |
584 | return 1; | |
585 | } | |
586 | ||
587 | fclose(f); | |
588 | return 0; | |
589 | } | |
590 | ||
ebac4655 JA |
591 | static void clear_io_state(struct thread_data *td) |
592 | { | |
53cdc686 JA |
593 | struct fio_file *f; |
594 | int i; | |
ebac4655 | 595 | |
ebac4655 JA |
596 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
597 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 598 | td->zone_bytes = 0; |
ebac4655 | 599 | |
c1324df1 JA |
600 | td->last_was_sync = 0; |
601 | ||
53cdc686 | 602 | for_each_file(td, f, i) { |
c1324df1 JA |
603 | f->last_completed_pos = 0; |
604 | ||
53cdc686 JA |
605 | f->last_pos = 0; |
606 | if (td->io_ops->flags & FIO_SYNCIO) | |
607 | lseek(f->fd, SEEK_SET, 0); | |
608 | ||
609 | if (f->file_map) | |
610 | memset(f->file_map, 0, f->num_maps * sizeof(long)); | |
611 | } | |
ebac4655 JA |
612 | } |
613 | ||
906c8d75 JA |
614 | /* |
615 | * Entry point for the thread based jobs. The process based jobs end up | |
616 | * here as well, after a little setup. | |
617 | */ | |
ebac4655 JA |
618 | static void *thread_main(void *data) |
619 | { | |
69008999 | 620 | unsigned long long runtime[2]; |
ebac4655 | 621 | struct thread_data *td = data; |
ebac4655 JA |
622 | |
623 | if (!td->use_thread) | |
624 | setsid(); | |
625 | ||
626 | td->pid = getpid(); | |
627 | ||
aea47d44 JA |
628 | INIT_LIST_HEAD(&td->io_u_freelist); |
629 | INIT_LIST_HEAD(&td->io_u_busylist); | |
755200a3 | 630 | INIT_LIST_HEAD(&td->io_u_requeues); |
aea47d44 JA |
631 | INIT_LIST_HEAD(&td->io_hist_list); |
632 | INIT_LIST_HEAD(&td->io_log_list); | |
633 | ||
ebac4655 JA |
634 | if (init_io_u(td)) |
635 | goto err; | |
636 | ||
637 | if (fio_setaffinity(td) == -1) { | |
638 | td_verror(td, errno); | |
639 | goto err; | |
640 | } | |
641 | ||
aea47d44 JA |
642 | if (init_iolog(td)) |
643 | goto err; | |
644 | ||
ebac4655 JA |
645 | if (td->ioprio) { |
646 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
647 | td_verror(td, errno); | |
648 | goto err; | |
649 | } | |
650 | } | |
651 | ||
1056eaad | 652 | if (nice(td->nice) == -1) { |
b6f4d880 JA |
653 | td_verror(td, errno); |
654 | goto err; | |
655 | } | |
656 | ||
75154845 JA |
657 | if (init_random_state(td)) |
658 | goto err; | |
659 | ||
56f9498d JA |
660 | if (td->ioscheduler && switch_ioscheduler(td)) |
661 | goto err; | |
da86774e | 662 | |
75154845 | 663 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
664 | fio_sem_up(&startup_sem); |
665 | fio_sem_down(&td->mutex); | |
ebac4655 | 666 | |
53cdc686 | 667 | if (!td->create_serialize && setup_files(td)) |
ebac4655 | 668 | goto err; |
21972cde JA |
669 | if (open_files(td)) |
670 | goto err; | |
ebac4655 | 671 | |
7d6c5283 JA |
672 | /* |
673 | * Do this late, as some IO engines would like to have the | |
674 | * files setup prior to initializing structures. | |
675 | */ | |
676 | if (td_io_init(td)) | |
677 | goto err; | |
678 | ||
69cfd7e0 JA |
679 | if (td->exec_prerun) { |
680 | if (system(td->exec_prerun) < 0) | |
681 | goto err; | |
682 | } | |
4e0ba8af | 683 | |
69008999 | 684 | fio_gettime(&td->epoch, NULL); |
36dff966 | 685 | getrusage(RUSAGE_SELF, &td->ru_start); |
69008999 JA |
686 | |
687 | runtime[0] = runtime[1] = 0; | |
ebac4655 | 688 | while (td->loops--) { |
02bcaa8c | 689 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
690 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); |
691 | ||
692 | if (td->ratemin) | |
693 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
694 | ||
695 | clear_io_state(td); | |
696 | prune_io_piece_log(td); | |
697 | ||
2866c82d | 698 | if (td->io_ops->flags & FIO_CPUIO) |
b990b5c0 JA |
699 | do_cpuio(td); |
700 | else | |
701 | do_io(td); | |
ebac4655 | 702 | |
69008999 | 703 | runtime[td->ddir] += utime_since_now(&td->start); |
aea47d44 | 704 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
69008999 | 705 | runtime[td->ddir ^ 1] = runtime[td->ddir]; |
3d60d1ed | 706 | |
ebac4655 JA |
707 | if (td->error || td->terminate) |
708 | break; | |
709 | ||
710 | if (td->verify == VERIFY_NONE) | |
711 | continue; | |
712 | ||
713 | clear_io_state(td); | |
02bcaa8c | 714 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
715 | |
716 | do_verify(td); | |
717 | ||
69008999 | 718 | runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
719 | |
720 | if (td->error || td->terminate) | |
721 | break; | |
722 | } | |
723 | ||
36dff966 | 724 | update_rusage_stat(td); |
69008999 JA |
725 | fio_gettime(&td->end_time, NULL); |
726 | td->runtime[0] = runtime[0] / 1000; | |
727 | td->runtime[1] = runtime[1] / 1000; | |
728 | ||
ebac4655 JA |
729 | if (td->bw_log) |
730 | finish_log(td, td->bw_log, "bw"); | |
731 | if (td->slat_log) | |
732 | finish_log(td, td->slat_log, "slat"); | |
733 | if (td->clat_log) | |
734 | finish_log(td, td->clat_log, "clat"); | |
076efc7c | 735 | if (td->write_iolog_file) |
843a7413 | 736 | write_iolog_close(td); |
69cfd7e0 JA |
737 | if (td->exec_postrun) { |
738 | if (system(td->exec_postrun) < 0) | |
739 | log_err("fio: postrun %s failed\n", td->exec_postrun); | |
740 | } | |
ebac4655 JA |
741 | |
742 | if (exitall_on_terminate) | |
6ce15a32 | 743 | terminate_threads(td->groupid, 0); |
ebac4655 JA |
744 | |
745 | err: | |
5bf13a5a JA |
746 | if (td->error) |
747 | printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
53cdc686 | 748 | close_files(td); |
2866c82d | 749 | close_ioengine(td); |
ebac4655 | 750 | cleanup_io_u(td); |
ebac4655 | 751 | td_set_runstate(td, TD_EXITED); |
a6418147 | 752 | return (void *) td->error; |
ebac4655 JA |
753 | } |
754 | ||
906c8d75 JA |
755 | /* |
756 | * We cannot pass the td data into a forked process, so attach the td and | |
757 | * pass it to the thread worker. | |
758 | */ | |
a6418147 | 759 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
760 | { |
761 | struct thread_data *td; | |
a6418147 | 762 | void *data, *ret; |
ebac4655 JA |
763 | |
764 | data = shmat(shmid, NULL, 0); | |
765 | if (data == (void *) -1) { | |
a6418147 JA |
766 | int __err = errno; |
767 | ||
ebac4655 | 768 | perror("shmat"); |
a6418147 | 769 | return __err; |
ebac4655 JA |
770 | } |
771 | ||
772 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 773 | ret = thread_main(td); |
ebac4655 | 774 | shmdt(data); |
a6418147 | 775 | return (int) ret; |
ebac4655 JA |
776 | } |
777 | ||
906c8d75 JA |
778 | /* |
779 | * Run over the job map and reap the threads that have exited, if any. | |
780 | */ | |
ebac4655 JA |
781 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
782 | { | |
34572e28 | 783 | struct thread_data *td; |
fab6aa71 | 784 | int i, cputhreads, pending, status, ret; |
ebac4655 JA |
785 | |
786 | /* | |
787 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
788 | */ | |
4d2413c6 | 789 | pending = cputhreads = 0; |
34572e28 | 790 | for_each_td(td, i) { |
84585003 JA |
791 | /* |
792 | * ->io_ops is NULL for a thread that has closed its | |
793 | * io engine | |
794 | */ | |
795 | if (td->io_ops && td->io_ops->flags & FIO_CPUIO) | |
b990b5c0 JA |
796 | cputhreads++; |
797 | ||
fab6aa71 JA |
798 | if (td->runstate < TD_EXITED) { |
799 | /* | |
800 | * check if someone quit or got killed in an unusual way | |
801 | */ | |
802 | ret = waitpid(td->pid, &status, WNOHANG); | |
803 | if (ret < 0) | |
804 | perror("waitpid"); | |
805 | else if ((ret == td->pid) && WIFSIGNALED(status)) { | |
806 | int sig = WTERMSIG(status); | |
807 | ||
808 | log_err("fio: pid=%d, got signal=%d\n", td->pid, sig); | |
809 | td_set_runstate(td, TD_REAPED); | |
810 | goto reaped; | |
811 | } | |
812 | } | |
813 | ||
4d2413c6 JA |
814 | if (td->runstate != TD_EXITED) { |
815 | if (td->runstate < TD_RUNNING) | |
816 | pending++; | |
817 | ||
ebac4655 | 818 | continue; |
4d2413c6 | 819 | } |
ebac4655 | 820 | |
437c9b71 JA |
821 | if (td->error) |
822 | exit_value++; | |
823 | ||
ebac4655 JA |
824 | td_set_runstate(td, TD_REAPED); |
825 | ||
826 | if (td->use_thread) { | |
827 | long ret; | |
828 | ||
829 | if (pthread_join(td->thread, (void *) &ret)) | |
830 | perror("thread_join"); | |
a6418147 JA |
831 | } else { |
832 | int status; | |
833 | ||
fab6aa71 JA |
834 | ret = waitpid(td->pid, &status, 0); |
835 | if (ret < 0) | |
836 | perror("waitpid"); | |
73170f19 | 837 | else if (WIFEXITED(status) && WEXITSTATUS(status)) { |
a6418147 JA |
838 | if (!exit_value) |
839 | exit_value++; | |
840 | } | |
841 | } | |
ebac4655 | 842 | |
fab6aa71 | 843 | reaped: |
ebac4655 JA |
844 | (*nr_running)--; |
845 | (*m_rate) -= td->ratemin; | |
846 | (*t_rate) -= td->rate; | |
847 | } | |
b990b5c0 | 848 | |
4d2413c6 | 849 | if (*nr_running == cputhreads && !pending) |
6ce15a32 | 850 | terminate_threads(TERMINATE_ALL, 0); |
ebac4655 JA |
851 | } |
852 | ||
906c8d75 JA |
853 | /* |
854 | * Main function for kicking off and reaping jobs, as needed. | |
855 | */ | |
ebac4655 JA |
856 | static void run_threads(void) |
857 | { | |
ebac4655 JA |
858 | struct thread_data *td; |
859 | unsigned long spent; | |
860 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 861 | |
2f9ade3c JA |
862 | if (fio_pin_memory()) |
863 | return; | |
ebac4655 | 864 | |
c6ae0a5b JA |
865 | if (!terse_output) { |
866 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
867 | fflush(stdout); | |
868 | } | |
c04f7ec3 | 869 | |
4efa970e JA |
870 | signal(SIGINT, sig_handler); |
871 | signal(SIGALRM, sig_handler); | |
872 | ||
ebac4655 JA |
873 | todo = thread_number; |
874 | nr_running = 0; | |
875 | nr_started = 0; | |
876 | m_rate = t_rate = 0; | |
877 | ||
34572e28 | 878 | for_each_td(td, i) { |
263e529f | 879 | print_status_init(td->thread_number - 1); |
ebac4655 | 880 | |
380cf265 JA |
881 | if (!td->create_serialize) { |
882 | init_disk_util(td); | |
ebac4655 | 883 | continue; |
380cf265 | 884 | } |
ebac4655 JA |
885 | |
886 | /* | |
887 | * do file setup here so it happens sequentially, | |
888 | * we don't want X number of threads getting their | |
889 | * client data interspersed on disk | |
890 | */ | |
53cdc686 | 891 | if (setup_files(td)) { |
5bf13a5a JA |
892 | exit_value++; |
893 | if (td->error) | |
894 | log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror); | |
ebac4655 JA |
895 | td_set_runstate(td, TD_REAPED); |
896 | todo--; | |
897 | } | |
380cf265 JA |
898 | |
899 | init_disk_util(td); | |
ebac4655 JA |
900 | } |
901 | ||
ebac4655 | 902 | while (todo) { |
75154845 JA |
903 | struct thread_data *map[MAX_JOBS]; |
904 | struct timeval this_start; | |
905 | int this_jobs = 0, left; | |
906 | ||
ebac4655 JA |
907 | /* |
908 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
909 | */ | |
34572e28 | 910 | for_each_td(td, i) { |
ebac4655 JA |
911 | if (td->runstate != TD_NOT_CREATED) |
912 | continue; | |
913 | ||
914 | /* | |
915 | * never got a chance to start, killed by other | |
916 | * thread for some reason | |
917 | */ | |
918 | if (td->terminate) { | |
919 | todo--; | |
920 | continue; | |
921 | } | |
922 | ||
923 | if (td->start_delay) { | |
263e529f | 924 | spent = mtime_since_genesis(); |
ebac4655 JA |
925 | |
926 | if (td->start_delay * 1000 > spent) | |
927 | continue; | |
928 | } | |
929 | ||
930 | if (td->stonewall && (nr_started || nr_running)) | |
931 | break; | |
932 | ||
75154845 JA |
933 | /* |
934 | * Set state to created. Thread will transition | |
935 | * to TD_INITIALIZED when it's done setting up. | |
936 | */ | |
ebac4655 | 937 | td_set_runstate(td, TD_CREATED); |
75154845 | 938 | map[this_jobs++] = td; |
bbfd6b00 | 939 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
940 | nr_started++; |
941 | ||
942 | if (td->use_thread) { | |
943 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
944 | perror("thread_create"); | |
945 | nr_started--; | |
946 | } | |
947 | } else { | |
948 | if (fork()) | |
bbfd6b00 | 949 | fio_sem_down(&startup_sem); |
ebac4655 | 950 | else { |
a6418147 JA |
951 | int ret = fork_main(shm_id, i); |
952 | ||
953 | exit(ret); | |
ebac4655 JA |
954 | } |
955 | } | |
956 | } | |
957 | ||
958 | /* | |
75154845 JA |
959 | * Wait for the started threads to transition to |
960 | * TD_INITIALIZED. | |
ebac4655 | 961 | */ |
02bcaa8c | 962 | fio_gettime(&this_start, NULL); |
75154845 | 963 | left = this_jobs; |
6ce15a32 | 964 | while (left && !fio_abort) { |
75154845 JA |
965 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
966 | break; | |
967 | ||
968 | usleep(100000); | |
969 | ||
970 | for (i = 0; i < this_jobs; i++) { | |
971 | td = map[i]; | |
972 | if (!td) | |
973 | continue; | |
b6f4d880 | 974 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
975 | map[i] = NULL; |
976 | left--; | |
b6f4d880 JA |
977 | } else if (td->runstate >= TD_EXITED) { |
978 | map[i] = NULL; | |
979 | left--; | |
980 | todo--; | |
981 | nr_running++; /* work-around... */ | |
75154845 JA |
982 | } |
983 | } | |
984 | } | |
985 | ||
986 | if (left) { | |
3b70d7e5 | 987 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
988 | for (i = 0; i < this_jobs; i++) { |
989 | td = map[i]; | |
990 | if (!td) | |
991 | continue; | |
992 | kill(td->pid, SIGTERM); | |
993 | } | |
994 | break; | |
995 | } | |
996 | ||
997 | /* | |
b6f4d880 | 998 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 999 | */ |
34572e28 | 1000 | for_each_td(td, i) { |
75154845 | 1001 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1002 | continue; |
1003 | ||
1004 | td_set_runstate(td, TD_RUNNING); | |
1005 | nr_running++; | |
1006 | nr_started--; | |
1007 | m_rate += td->ratemin; | |
1008 | t_rate += td->rate; | |
75154845 | 1009 | todo--; |
bbfd6b00 | 1010 | fio_sem_up(&td->mutex); |
ebac4655 JA |
1011 | } |
1012 | ||
1013 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1014 | ||
1015 | if (todo) | |
1016 | usleep(100000); | |
1017 | } | |
1018 | ||
1019 | while (nr_running) { | |
1020 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1021 | usleep(10000); | |
1022 | } | |
1023 | ||
1024 | update_io_ticks(); | |
2f9ade3c | 1025 | fio_unpin_memory(); |
ebac4655 JA |
1026 | } |
1027 | ||
ebac4655 JA |
1028 | int main(int argc, char *argv[]) |
1029 | { | |
29d610e1 JA |
1030 | long ps; |
1031 | ||
dbe1125e JA |
1032 | /* |
1033 | * We need locale for number printing, if it isn't set then just | |
1034 | * go with the US format. | |
1035 | */ | |
1036 | if (!getenv("LC_NUMERIC")) | |
1037 | setlocale(LC_NUMERIC, "en_US"); | |
1038 | ||
ebac4655 JA |
1039 | if (parse_options(argc, argv)) |
1040 | return 1; | |
1041 | ||
1042 | if (!thread_number) { | |
3b70d7e5 | 1043 | log_err("Nothing to do\n"); |
ebac4655 JA |
1044 | return 1; |
1045 | } | |
1046 | ||
29d610e1 JA |
1047 | ps = sysconf(_SC_PAGESIZE); |
1048 | if (ps < 0) { | |
1049 | log_err("Failed to get page size\n"); | |
1050 | return 1; | |
1051 | } | |
1052 | ||
1053 | page_mask = ps - 1; | |
1054 | ||
bb3884d8 JA |
1055 | if (write_bw_log) { |
1056 | setup_log(&agg_io_log[DDIR_READ]); | |
1057 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1058 | } | |
1059 | ||
ebac4655 JA |
1060 | disk_util_timer_arm(); |
1061 | ||
1062 | run_threads(); | |
6ce15a32 | 1063 | |
bb3884d8 | 1064 | if (!fio_abort) { |
6ce15a32 | 1065 | show_run_stats(); |
bb3884d8 JA |
1066 | if (write_bw_log) { |
1067 | __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log"); | |
1068 | __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log"); | |
1069 | } | |
1070 | } | |
ebac4655 | 1071 | |
437c9b71 | 1072 | return exit_value; |
ebac4655 | 1073 | } |