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> | |
9b836561 | 27 | #include <limits.h> |
ebac4655 JA |
28 | #include <signal.h> |
29 | #include <time.h> | |
dbe1125e | 30 | #include <locale.h> |
36167d82 | 31 | #include <assert.h> |
a5b01f1b | 32 | #include <time.h> |
ebac4655 JA |
33 | #include <sys/stat.h> |
34 | #include <sys/wait.h> | |
35 | #include <sys/ipc.h> | |
36 | #include <sys/shm.h> | |
ebac4655 JA |
37 | #include <sys/mman.h> |
38 | ||
39 | #include "fio.h" | |
210f43b3 | 40 | #include "hash.h" |
2e5cdb11 | 41 | #include "smalloc.h" |
4f5af7b2 | 42 | #include "verify.h" |
a917a8b3 | 43 | #include "trim.h" |
7c9b1bce | 44 | #include "diskutil.h" |
a696fa2a | 45 | #include "cgroup.h" |
07b3232d | 46 | #include "profile.h" |
d54a144d | 47 | #include "lib/rand.h" |
91aea6ec | 48 | #include "memalign.h" |
009b1be4 | 49 | #include "server.h" |
ebac4655 | 50 | |
cfc99db7 JA |
51 | unsigned long page_mask; |
52 | unsigned long page_size; | |
d529ee19 JA |
53 | |
54 | #define PAGE_ALIGN(buf) \ | |
29d610e1 | 55 | (char *) (((unsigned long) (buf) + page_mask) & ~page_mask) |
ebac4655 JA |
56 | |
57 | int groupid = 0; | |
58 | int thread_number = 0; | |
9cedf167 JA |
59 | int nr_process = 0; |
60 | int nr_thread = 0; | |
ebac4655 | 61 | int shm_id = 0; |
53cdc686 | 62 | int temp_stall_ts; |
d3eeeabc | 63 | unsigned long done_secs = 0; |
ebac4655 | 64 | |
cdd18ad8 | 65 | static struct fio_mutex *startup_mutex; |
d2bbb8d9 | 66 | static struct fio_mutex *writeout_mutex; |
6ce15a32 | 67 | static volatile int fio_abort; |
437c9b71 | 68 | static int exit_value; |
be4ecfdf | 69 | static pthread_t gtod_thread; |
7985701b | 70 | static pthread_t disk_util_thread; |
dae5341c | 71 | static struct flist_head *cgroup_list; |
6adb38a1 | 72 | static char *cgroup_mnt; |
ebac4655 | 73 | |
d3cc4ebf JA |
74 | unsigned long arch_flags = 0; |
75 | ||
bb3884d8 JA |
76 | struct io_log *agg_io_log[2]; |
77 | ||
ebac4655 | 78 | #define TERMINATE_ALL (-1) |
75154845 | 79 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 80 | |
b29ee5b3 | 81 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 82 | { |
d8fab1b6 JA |
83 | if (td->runstate == runstate) |
84 | return; | |
85 | ||
5921e80c JA |
86 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
87 | td->runstate, runstate); | |
6ce15a32 JA |
88 | td->runstate = runstate; |
89 | } | |
90 | ||
390c40e2 | 91 | static void terminate_threads(int group_id) |
ebac4655 | 92 | { |
34572e28 | 93 | struct thread_data *td; |
ebac4655 JA |
94 | int i; |
95 | ||
6f88bcb0 JA |
96 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
97 | ||
34572e28 | 98 | for_each_td(td, i) { |
ebac4655 | 99 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 100 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 101 | td->o.name, (int) td->pid); |
ad830ed7 JA |
102 | td->terminate = 1; |
103 | td->o.start_delay = 0; | |
104 | ||
7a93ab19 JA |
105 | /* |
106 | * if the thread is running, just let it exit | |
107 | */ | |
6aa27816 MC |
108 | if (!td->pid) |
109 | continue; | |
110 | else if (td->runstate < TD_RAMP) | |
03e20d68 | 111 | kill(td->pid, SIGTERM); |
f63f7f3b JA |
112 | else { |
113 | struct ioengine_ops *ops = td->io_ops; | |
114 | ||
03e20d68 BC |
115 | if (ops && (ops->flags & FIO_SIGTERM)) |
116 | kill(td->pid, SIGTERM); | |
f63f7f3b | 117 | } |
ebac4655 JA |
118 | } |
119 | } | |
120 | } | |
121 | ||
697a606c JA |
122 | static void sig_int(int sig) |
123 | { | |
124 | if (threads) { | |
4ceb30d4 | 125 | log_info("\nfio: terminating on signal %d\n", sig); |
009b1be4 | 126 | exit_backend = 1; |
5ec10eaa | 127 | fflush(stdout); |
deaa1af2 | 128 | exit_value = 128; |
5ec10eaa | 129 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
130 | } |
131 | } | |
132 | ||
7985701b | 133 | static void *disk_thread_main(void *data) |
a5b01f1b | 134 | { |
7985701b JA |
135 | fio_mutex_up(startup_mutex); |
136 | ||
137 | while (threads) { | |
138 | usleep(DISK_UTIL_MSEC * 1000); | |
9a793c2f JA |
139 | if (!threads) |
140 | break; | |
7985701b | 141 | update_io_ticks(); |
37db14fe | 142 | print_thread_status(); |
7985701b JA |
143 | } |
144 | ||
145 | return NULL; | |
a5b01f1b JA |
146 | } |
147 | ||
7985701b | 148 | static int create_disk_util_thread(void) |
a5b01f1b | 149 | { |
7985701b | 150 | int ret; |
a5b01f1b | 151 | |
69204d6e | 152 | ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); |
7985701b JA |
153 | if (ret) { |
154 | log_err("Can't create disk util thread: %s\n", strerror(ret)); | |
155 | return 1; | |
156 | } | |
a5b01f1b | 157 | |
7985701b JA |
158 | ret = pthread_detach(disk_util_thread); |
159 | if (ret) { | |
160 | log_err("Can't detatch disk util thread: %s\n", strerror(ret)); | |
161 | return 1; | |
162 | } | |
03e20d68 | 163 | |
7985701b JA |
164 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
165 | fio_mutex_down(startup_mutex); | |
166 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
167 | return 0; | |
a5b01f1b JA |
168 | } |
169 | ||
697a606c JA |
170 | static void set_sig_handlers(void) |
171 | { | |
172 | struct sigaction act; | |
173 | ||
697a606c JA |
174 | memset(&act, 0, sizeof(act)); |
175 | act.sa_handler = sig_int; | |
176 | act.sa_flags = SA_RESTART; | |
177 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 178 | |
5f8f9726 | 179 | memset(&act, 0, sizeof(act)); |
b20b6be4 | 180 | act.sa_handler = sig_int; |
5f8f9726 | 181 | act.sa_flags = SA_RESTART; |
03e20d68 | 182 | sigaction(SIGTERM, &act, NULL); |
697a606c JA |
183 | } |
184 | ||
906c8d75 JA |
185 | /* |
186 | * Check if we are above the minimum rate given. | |
187 | */ | |
581e7141 | 188 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
bd93e34b | 189 | enum fio_ddir ddir) |
ebac4655 | 190 | { |
0904200b | 191 | unsigned long long bytes = 0; |
4e991c23 | 192 | unsigned long iops = 0; |
ebac4655 JA |
193 | unsigned long spent; |
194 | unsigned long rate; | |
581e7141 JA |
195 | unsigned int ratemin = 0; |
196 | unsigned int rate_iops = 0; | |
197 | unsigned int rate_iops_min = 0; | |
ebac4655 | 198 | |
ff58fced JA |
199 | assert(ddir_rw(ddir)); |
200 | ||
d982d873 JA |
201 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
202 | return 0; | |
203 | ||
ebac4655 JA |
204 | /* |
205 | * allow a 2 second settle period in the beginning | |
206 | */ | |
207 | if (mtime_since(&td->start, now) < 2000) | |
208 | return 0; | |
209 | ||
581e7141 JA |
210 | iops += td->io_blocks[ddir]; |
211 | bytes += td->this_io_bytes[ddir]; | |
212 | ratemin += td->o.ratemin[ddir]; | |
213 | rate_iops += td->o.rate_iops[ddir]; | |
214 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 215 | |
ebac4655 JA |
216 | /* |
217 | * if rate blocks is set, sample is running | |
218 | */ | |
581e7141 JA |
219 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
220 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 221 | if (spent < td->o.ratecycle) |
ebac4655 JA |
222 | return 0; |
223 | ||
581e7141 | 224 | if (td->o.rate[ddir]) { |
4e991c23 JA |
225 | /* |
226 | * check bandwidth specified rate | |
227 | */ | |
581e7141 | 228 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 229 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 230 | ratemin); |
4e991c23 JA |
231 | return 1; |
232 | } else { | |
581e7141 JA |
233 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
234 | if (rate < ratemin || | |
235 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 236 | log_err("%s: min rate %u not met, got" |
b22989b9 | 237 | " %luKB/sec\n", td->o.name, |
581e7141 | 238 | ratemin, rate); |
4e991c23 JA |
239 | return 1; |
240 | } | |
241 | } | |
413dd459 | 242 | } else { |
4e991c23 JA |
243 | /* |
244 | * checks iops specified rate | |
245 | */ | |
581e7141 | 246 | if (iops < rate_iops) { |
5ec10eaa | 247 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 248 | td->o.name, rate_iops); |
413dd459 | 249 | return 1; |
4e991c23 | 250 | } else { |
581e7141 JA |
251 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
252 | if (rate < rate_iops_min || | |
253 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
254 | log_err("%s: min iops rate %u not met," |
255 | " got %lu\n", td->o.name, | |
581e7141 | 256 | rate_iops_min, rate); |
4e991c23 | 257 | } |
413dd459 | 258 | } |
ebac4655 JA |
259 | } |
260 | } | |
261 | ||
581e7141 JA |
262 | td->rate_bytes[ddir] = bytes; |
263 | td->rate_blocks[ddir] = iops; | |
264 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
265 | return 0; |
266 | } | |
267 | ||
581e7141 JA |
268 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
269 | unsigned long *bytes_done) | |
270 | { | |
271 | int ret = 0; | |
272 | ||
273 | if (bytes_done[0]) | |
274 | ret |= __check_min_rate(td, now, 0); | |
275 | if (bytes_done[1]) | |
276 | ret |= __check_min_rate(td, now, 1); | |
277 | ||
278 | return ret; | |
279 | } | |
280 | ||
ebac4655 JA |
281 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
282 | { | |
2dc1bbeb | 283 | if (!td->o.timeout) |
ebac4655 | 284 | return 0; |
2dc1bbeb | 285 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
286 | return 1; |
287 | ||
288 | return 0; | |
289 | } | |
290 | ||
906c8d75 JA |
291 | /* |
292 | * When job exits, we can cancel the in-flight IO if we are using async | |
293 | * io. Attempt to do so. | |
294 | */ | |
ebac4655 JA |
295 | static void cleanup_pending_aio(struct thread_data *td) |
296 | { | |
01743ee1 | 297 | struct flist_head *entry, *n; |
ebac4655 JA |
298 | struct io_u *io_u; |
299 | int r; | |
300 | ||
301 | /* | |
302 | * get immediately available events, if any | |
303 | */ | |
581e7141 | 304 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
305 | if (r < 0) |
306 | return; | |
ebac4655 JA |
307 | |
308 | /* | |
309 | * now cancel remaining active events | |
310 | */ | |
2866c82d | 311 | if (td->io_ops->cancel) { |
01743ee1 JA |
312 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
313 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 314 | |
0c6e7517 JA |
315 | /* |
316 | * if the io_u isn't in flight, then that generally | |
317 | * means someone leaked an io_u. complain but fix | |
318 | * it up, so we don't stall here. | |
319 | */ | |
320 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
321 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 322 | put_io_u(td, io_u); |
0c6e7517 JA |
323 | } else { |
324 | r = td->io_ops->cancel(td, io_u); | |
325 | if (!r) | |
326 | put_io_u(td, io_u); | |
327 | } | |
ebac4655 JA |
328 | } |
329 | } | |
330 | ||
97601024 | 331 | if (td->cur_depth) |
581e7141 | 332 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
333 | } |
334 | ||
858a3d47 JA |
335 | /* |
336 | * Helper to handle the final sync of a file. Works just like the normal | |
337 | * io path, just does everything sync. | |
338 | */ | |
339 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
340 | { | |
341 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
342 | int ret; |
343 | ||
344 | if (!io_u) | |
345 | return 1; | |
346 | ||
347 | io_u->ddir = DDIR_SYNC; | |
348 | io_u->file = f; | |
349 | ||
350 | if (td_io_prep(td, io_u)) { | |
351 | put_io_u(td, io_u); | |
352 | return 1; | |
353 | } | |
354 | ||
755200a3 | 355 | requeue: |
858a3d47 | 356 | ret = td_io_queue(td, io_u); |
36167d82 | 357 | if (ret < 0) { |
e1161c32 | 358 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 359 | put_io_u(td, io_u); |
858a3d47 | 360 | return 1; |
36167d82 | 361 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 362 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 363 | return 1; |
36167d82 JA |
364 | } else if (ret == FIO_Q_COMPLETED) { |
365 | if (io_u->error) { | |
e1161c32 | 366 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
367 | return 1; |
368 | } | |
858a3d47 | 369 | |
581e7141 | 370 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 371 | return 1; |
755200a3 JA |
372 | } else if (ret == FIO_Q_BUSY) { |
373 | if (td_io_commit(td)) | |
374 | return 1; | |
375 | goto requeue; | |
858a3d47 JA |
376 | } |
377 | ||
378 | return 0; | |
379 | } | |
380 | ||
687c6a23 JA |
381 | static inline void __update_tv_cache(struct thread_data *td) |
382 | { | |
383 | fio_gettime(&td->tv_cache, NULL); | |
384 | } | |
385 | ||
993bf48b JA |
386 | static inline void update_tv_cache(struct thread_data *td) |
387 | { | |
388 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
687c6a23 | 389 | __update_tv_cache(td); |
993bf48b JA |
390 | } |
391 | ||
f2bba182 RR |
392 | static int break_on_this_error(struct thread_data *td, int *retptr) |
393 | { | |
394 | int ret = *retptr; | |
395 | ||
396 | if (ret < 0 || td->error) { | |
397 | int err; | |
398 | ||
1ec99eea JA |
399 | if (!td->o.continue_on_error) |
400 | return 1; | |
f2bba182 RR |
401 | |
402 | if (ret < 0) | |
403 | err = -ret; | |
404 | else | |
405 | err = td->error; | |
406 | ||
f2bba182 RR |
407 | if (td_non_fatal_error(err)) { |
408 | /* | |
409 | * Continue with the I/Os in case of | |
410 | * a non fatal error. | |
411 | */ | |
8a4f8a24 | 412 | update_error_count(td, err); |
f2bba182 RR |
413 | td_clear_error(td); |
414 | *retptr = 0; | |
415 | return 0; | |
8a4f8a24 RR |
416 | } else if (td->o.fill_device && err == ENOSPC) { |
417 | /* | |
418 | * We expect to hit this error if | |
419 | * fill_device option is set. | |
420 | */ | |
421 | td_clear_error(td); | |
422 | td->terminate = 1; | |
423 | return 1; | |
f2bba182 RR |
424 | } else { |
425 | /* | |
426 | * Stop the I/O in case of a fatal | |
427 | * error. | |
428 | */ | |
8a4f8a24 | 429 | update_error_count(td, err); |
f2bba182 RR |
430 | return 1; |
431 | } | |
432 | } | |
433 | ||
434 | return 0; | |
435 | } | |
436 | ||
906c8d75 | 437 | /* |
34403fb1 | 438 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
439 | * reads the blocks back in, and checks the crc/md5 of the data. |
440 | */ | |
1e97cce9 | 441 | static void do_verify(struct thread_data *td) |
ebac4655 | 442 | { |
53cdc686 | 443 | struct fio_file *f; |
36167d82 | 444 | struct io_u *io_u; |
af52b345 JA |
445 | int ret, min_events; |
446 | unsigned int i; | |
e5b401d4 | 447 | |
008618a8 JA |
448 | dprint(FD_VERIFY, "starting loop\n"); |
449 | ||
e5b401d4 JA |
450 | /* |
451 | * sync io first and invalidate cache, to make sure we really | |
452 | * read from disk. | |
453 | */ | |
454 | for_each_file(td, f, i) { | |
d6aed795 | 455 | if (!fio_file_open(f)) |
3d7b485f | 456 | continue; |
b2fdda43 JA |
457 | if (fio_io_sync(td, f)) |
458 | break; | |
459 | if (file_invalidate_cache(td, f)) | |
460 | break; | |
e5b401d4 | 461 | } |
ebac4655 | 462 | |
b2fdda43 JA |
463 | if (td->error) |
464 | return; | |
465 | ||
ebac4655 JA |
466 | td_set_runstate(td, TD_VERIFYING); |
467 | ||
36167d82 JA |
468 | io_u = NULL; |
469 | while (!td->terminate) { | |
4950421a | 470 | int ret2, full; |
5451792e | 471 | |
993bf48b JA |
472 | update_tv_cache(td); |
473 | ||
474 | if (runtime_exceeded(td, &td->tv_cache)) { | |
687c6a23 JA |
475 | __update_tv_cache(td); |
476 | if (runtime_exceeded(td, &td->tv_cache)) { | |
477 | td->terminate = 1; | |
478 | break; | |
479 | } | |
069c2918 | 480 | } |
02bcaa8c | 481 | |
dcb4830d RR |
482 | io_u = __get_io_u(td); |
483 | if (!io_u) | |
484 | break; | |
485 | ||
069c2918 JA |
486 | if (get_next_verify(td, io_u)) { |
487 | put_io_u(td, io_u); | |
ebac4655 | 488 | break; |
069c2918 | 489 | } |
ebac4655 | 490 | |
069c2918 JA |
491 | if (td_io_prep(td, io_u)) { |
492 | put_io_u(td, io_u); | |
53cdc686 | 493 | break; |
069c2918 | 494 | } |
d7762cf8 | 495 | |
e8462bd8 JA |
496 | if (td->o.verify_async) |
497 | io_u->end_io = verify_io_u_async; | |
498 | else | |
499 | io_u->end_io = verify_io_u; | |
53cdc686 | 500 | |
11786802 | 501 | ret = td_io_queue(td, io_u); |
36167d82 JA |
502 | switch (ret) { |
503 | case FIO_Q_COMPLETED: | |
f2bba182 | 504 | if (io_u->error) { |
22819ec2 | 505 | ret = -io_u->error; |
f2bba182 RR |
506 | clear_io_u(td, io_u); |
507 | } else if (io_u->resid) { | |
36167d82 | 508 | int bytes = io_u->xfer_buflen - io_u->resid; |
ebac4655 | 509 | |
9e9d164e JA |
510 | /* |
511 | * zero read, fail | |
512 | */ | |
513 | if (!bytes) { | |
41d9f0fd | 514 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
515 | put_io_u(td, io_u); |
516 | break; | |
517 | } | |
8400d9b2 | 518 | |
36167d82 JA |
519 | io_u->xfer_buflen = io_u->resid; |
520 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
521 | io_u->offset += bytes; |
522 | ||
ff58fced JA |
523 | if (ddir_rw(io_u->ddir)) |
524 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 525 | |
2b13e716 | 526 | f = io_u->file; |
d460eb31 | 527 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
528 | goto sync_done; |
529 | ||
11786802 JA |
530 | requeue_io_u(td, &io_u); |
531 | } else { | |
8400d9b2 | 532 | sync_done: |
581e7141 | 533 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
534 | if (ret < 0) |
535 | break; | |
36167d82 | 536 | } |
36167d82 JA |
537 | continue; |
538 | case FIO_Q_QUEUED: | |
539 | break; | |
755200a3 JA |
540 | case FIO_Q_BUSY: |
541 | requeue_io_u(td, &io_u); | |
5451792e JA |
542 | ret2 = td_io_commit(td); |
543 | if (ret2 < 0) | |
544 | ret = ret2; | |
755200a3 | 545 | break; |
36167d82 JA |
546 | default: |
547 | assert(ret < 0); | |
e1161c32 | 548 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
549 | break; |
550 | } | |
551 | ||
f2bba182 | 552 | if (break_on_this_error(td, &ret)) |
3af6ef39 JA |
553 | break; |
554 | ||
ebac4655 | 555 | /* |
3af6ef39 | 556 | * if we can queue more, do so. but check if there are |
ccc02050 JA |
557 | * completed io_u's first. Note that we can get BUSY even |
558 | * without IO queued, if the system is resource starved. | |
ebac4655 | 559 | */ |
ccc02050 | 560 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 561 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
562 | min_events = min(td->o.iodepth_batch_complete, |
563 | td->cur_depth); | |
1f84f1d9 | 564 | if (full && !min_events && td->o.iodepth_batch_complete != 0) |
838bc709 | 565 | min_events = 1; |
e916b390 | 566 | |
4950421a JA |
567 | do { |
568 | /* | |
569 | * Reap required number of io units, if any, | |
570 | * and do the verification on them through | |
571 | * the callback handler | |
572 | */ | |
581e7141 | 573 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
574 | ret = -1; |
575 | break; | |
576 | } | |
577 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
578 | } | |
579 | if (ret < 0) | |
ebac4655 | 580 | break; |
36167d82 | 581 | } |
ebac4655 | 582 | |
c01c0395 JA |
583 | if (!td->error) { |
584 | min_events = td->cur_depth; | |
585 | ||
586 | if (min_events) | |
581e7141 | 587 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 588 | } else |
ebac4655 JA |
589 | cleanup_pending_aio(td); |
590 | ||
591 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
592 | |
593 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
594 | } |
595 | ||
32cd46a0 | 596 | /* |
906c8d75 | 597 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
598 | * and reaps them, checking for rate and errors along the way. |
599 | */ | |
ebac4655 JA |
600 | static void do_io(struct thread_data *td) |
601 | { | |
af52b345 JA |
602 | unsigned int i; |
603 | int ret = 0; | |
ebac4655 | 604 | |
b29ee5b3 JA |
605 | if (in_ramp_time(td)) |
606 | td_set_runstate(td, TD_RAMP); | |
607 | else | |
608 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 609 | |
457bf399 | 610 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
a917a8b3 | 611 | (!flist_empty(&td->trim_list)) || |
457bf399 | 612 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { |
97601024 | 613 | struct timeval comp_time; |
581e7141 | 614 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 615 | int min_evts = 0; |
ebac4655 | 616 | struct io_u *io_u; |
4950421a | 617 | int ret2, full; |
ebac4655 JA |
618 | |
619 | if (td->terminate) | |
620 | break; | |
621 | ||
993bf48b | 622 | update_tv_cache(td); |
97601024 | 623 | |
993bf48b | 624 | if (runtime_exceeded(td, &td->tv_cache)) { |
687c6a23 JA |
625 | __update_tv_cache(td); |
626 | if (runtime_exceeded(td, &td->tv_cache)) { | |
627 | td->terminate = 1; | |
628 | break; | |
629 | } | |
97601024 | 630 | } |
36167d82 | 631 | |
dcb4830d RR |
632 | io_u = get_io_u(td); |
633 | if (!io_u) | |
634 | break; | |
635 | ||
b67740d3 JA |
636 | /* |
637 | * Add verification end_io handler, if asked to verify | |
638 | * a previously written file. | |
639 | */ | |
715d2b3e JA |
640 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
641 | !td_rw(td)) { | |
e8462bd8 JA |
642 | if (td->o.verify_async) |
643 | io_u->end_io = verify_io_u_async; | |
644 | else | |
645 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 646 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
647 | } else if (in_ramp_time(td)) |
648 | td_set_runstate(td, TD_RAMP); | |
649 | else | |
d8fab1b6 | 650 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 651 | |
11786802 | 652 | ret = td_io_queue(td, io_u); |
36167d82 JA |
653 | switch (ret) { |
654 | case FIO_Q_COMPLETED: | |
f2bba182 | 655 | if (io_u->error) { |
5451792e | 656 | ret = -io_u->error; |
f2bba182 RR |
657 | clear_io_u(td, io_u); |
658 | } else if (io_u->resid) { | |
36167d82 | 659 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 660 | struct fio_file *f = io_u->file; |
36167d82 | 661 | |
9e9d164e JA |
662 | /* |
663 | * zero read, fail | |
664 | */ | |
665 | if (!bytes) { | |
41d9f0fd | 666 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
667 | put_io_u(td, io_u); |
668 | break; | |
669 | } | |
670 | ||
cec6b55d | 671 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 672 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
673 | io_u->offset += bytes; |
674 | ||
ff58fced JA |
675 | if (ddir_rw(io_u->ddir)) |
676 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 677 | |
d460eb31 | 678 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
679 | goto sync_done; |
680 | ||
11786802 JA |
681 | requeue_io_u(td, &io_u); |
682 | } else { | |
5a7c5680 | 683 | sync_done: |
581e7141 JA |
684 | if (__should_check_rate(td, 0) || |
685 | __should_check_rate(td, 1)) | |
993bf48b JA |
686 | fio_gettime(&comp_time, NULL); |
687 | ||
581e7141 JA |
688 | ret = io_u_sync_complete(td, io_u, bytes_done); |
689 | if (ret < 0) | |
690 | break; | |
cec6b55d | 691 | } |
36167d82 JA |
692 | break; |
693 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
694 | /* |
695 | * if the engine doesn't have a commit hook, | |
696 | * the io_u is really queued. if it does have such | |
697 | * a hook, it has to call io_u_queued() itself. | |
698 | */ | |
699 | if (td->io_ops->commit == NULL) | |
700 | io_u_queued(td, io_u); | |
36167d82 | 701 | break; |
755200a3 JA |
702 | case FIO_Q_BUSY: |
703 | requeue_io_u(td, &io_u); | |
5451792e JA |
704 | ret2 = td_io_commit(td); |
705 | if (ret2 < 0) | |
706 | ret = ret2; | |
755200a3 | 707 | break; |
36167d82 JA |
708 | default: |
709 | assert(ret < 0); | |
710 | put_io_u(td, io_u); | |
711 | break; | |
ebac4655 JA |
712 | } |
713 | ||
f2bba182 | 714 | if (break_on_this_error(td, &ret)) |
36167d82 JA |
715 | break; |
716 | ||
97601024 | 717 | /* |
ccc02050 JA |
718 | * See if we need to complete some commands. Note that we |
719 | * can get BUSY even without IO queued, if the system is | |
720 | * resource starved. | |
97601024 | 721 | */ |
ccc02050 | 722 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 723 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
724 | min_evts = min(td->o.iodepth_batch_complete, |
725 | td->cur_depth); | |
1f84f1d9 | 726 | if (full && !min_evts && td->o.iodepth_batch_complete != 0) |
36167d82 | 727 | min_evts = 1; |
4950421a | 728 | |
581e7141 JA |
729 | if (__should_check_rate(td, 0) || |
730 | __should_check_rate(td, 1)) | |
993bf48b | 731 | fio_gettime(&comp_time, NULL); |
4950421a JA |
732 | |
733 | do { | |
581e7141 JA |
734 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
735 | if (ret < 0) | |
4950421a JA |
736 | break; |
737 | ||
4950421a | 738 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
739 | } |
740 | ||
4950421a JA |
741 | if (ret < 0) |
742 | break; | |
581e7141 | 743 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
744 | continue; |
745 | ||
581e7141 JA |
746 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
747 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae JA |
748 | if (exitall_on_terminate) |
749 | terminate_threads(td->groupid); | |
750 | td_verror(td, EIO, "check_min_rate"); | |
751 | break; | |
752 | } | |
ebac4655 JA |
753 | } |
754 | ||
2dc1bbeb | 755 | if (td->o.thinktime) { |
9c1f7434 JA |
756 | unsigned long long b; |
757 | ||
758 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 759 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
760 | int left; |
761 | ||
2dc1bbeb | 762 | if (td->o.thinktime_spin) |
6dd6f2cd | 763 | usec_spin(td->o.thinktime_spin); |
48097d5c | 764 | |
2dc1bbeb | 765 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
766 | if (left) |
767 | usec_sleep(td, left); | |
768 | } | |
9c1f7434 | 769 | } |
ebac4655 JA |
770 | } |
771 | ||
d708ac5e JA |
772 | if (td->trim_entries) |
773 | printf("trim entries %ld\n", td->trim_entries); | |
a917a8b3 | 774 | |
aa31f1f1 SL |
775 | if (td->o.fill_device && td->error == ENOSPC) { |
776 | td->error = 0; | |
777 | td->terminate = 1; | |
778 | } | |
4d2413c6 | 779 | if (!td->error) { |
3d7c391d JA |
780 | struct fio_file *f; |
781 | ||
c01c0395 | 782 | i = td->cur_depth; |
d06093e7 | 783 | if (i) { |
581e7141 | 784 | ret = io_u_queued_complete(td, i, NULL); |
d06093e7 SL |
785 | if (td->o.fill_device && td->error == ENOSPC) |
786 | td->error = 0; | |
787 | } | |
ebac4655 | 788 | |
2dc1bbeb | 789 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 790 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
791 | |
792 | for_each_file(td, f, i) { | |
d6aed795 | 793 | if (!fio_file_open(f)) |
3d7b485f | 794 | continue; |
858a3d47 | 795 | fio_io_sync(td, f); |
3d7b485f | 796 | } |
84585003 | 797 | } |
c01c0395 JA |
798 | } else |
799 | cleanup_pending_aio(td); | |
2ba1c290 JA |
800 | |
801 | /* | |
802 | * stop job if we failed doing any IO | |
803 | */ | |
804 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
805 | td->done = 1; | |
ebac4655 JA |
806 | } |
807 | ||
ebac4655 JA |
808 | static void cleanup_io_u(struct thread_data *td) |
809 | { | |
01743ee1 | 810 | struct flist_head *entry, *n; |
ebac4655 JA |
811 | struct io_u *io_u; |
812 | ||
01743ee1 JA |
813 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
814 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 815 | |
01743ee1 | 816 | flist_del(&io_u->list); |
91aea6ec | 817 | fio_memfree(io_u, sizeof(*io_u)); |
ebac4655 JA |
818 | } |
819 | ||
2f9ade3c | 820 | free_io_mem(td); |
ebac4655 JA |
821 | } |
822 | ||
823 | static int init_io_u(struct thread_data *td) | |
824 | { | |
825 | struct io_u *io_u; | |
a00735e6 | 826 | unsigned int max_bs; |
eb7ccf38 | 827 | int cl_align, i, max_units; |
ebac4655 JA |
828 | char *p; |
829 | ||
1e3c09af | 830 | max_units = td->o.iodepth; |
2dc1bbeb | 831 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
832 | td->orig_buffer_size = (unsigned long long) max_bs |
833 | * (unsigned long long) max_units; | |
834 | ||
835 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
836 | unsigned long bs; | |
74b025b0 | 837 | |
5ec10eaa JA |
838 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
839 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
840 | } | |
ebac4655 | 841 | |
c3990645 JA |
842 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
843 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
844 | return 1; | |
845 | } | |
846 | ||
2f9ade3c JA |
847 | if (allocate_io_mem(td)) |
848 | return 1; | |
ebac4655 | 849 | |
1c04bf78 JA |
850 | if (td->o.odirect || td->o.mem_align || |
851 | (td->io_ops->flags & FIO_RAWIO)) | |
d529ee19 | 852 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
853 | else |
854 | p = td->orig_buffer; | |
855 | ||
eb7ccf38 JA |
856 | cl_align = os_cache_line_size(); |
857 | ||
ebac4655 | 858 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
859 | void *ptr; |
860 | ||
db1cd90b JA |
861 | if (td->terminate) |
862 | return 1; | |
eb7ccf38 | 863 | |
91aea6ec JA |
864 | ptr = fio_memalign(cl_align, sizeof(*io_u)); |
865 | if (!ptr) { | |
866 | log_err("fio: unable to allocate aligned memory\n"); | |
eb7ccf38 JA |
867 | break; |
868 | } | |
869 | ||
870 | io_u = ptr; | |
ebac4655 | 871 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 872 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 873 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 874 | |
b4c5e1ac JA |
875 | if (!(td->io_ops->flags & FIO_NOIO)) { |
876 | io_u->buf = p + max_bs * i; | |
d529ee19 | 877 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 878 | |
3545a109 | 879 | if (td_write(td)) |
5973cafb | 880 | io_u_fill_buffer(td, io_u, max_bs); |
3545a109 | 881 | if (td_write(td) && td->o.verify_pattern_bytes) { |
cbe8d756 RR |
882 | /* |
883 | * Fill the buffer with the pattern if we are | |
884 | * going to be doing writes. | |
885 | */ | |
7d9fb455 | 886 | fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
cbe8d756 | 887 | } |
b4c5e1ac | 888 | } |
6b9cea23 | 889 | |
b1ff3403 | 890 | io_u->index = i; |
0c6e7517 | 891 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 892 | flist_add(&io_u->list, &td->io_u_freelist); |
ebac4655 JA |
893 | } |
894 | ||
895 | return 0; | |
896 | } | |
897 | ||
da86774e JA |
898 | static int switch_ioscheduler(struct thread_data *td) |
899 | { | |
900 | char tmp[256], tmp2[128]; | |
901 | FILE *f; | |
902 | int ret; | |
903 | ||
ba0fbe10 | 904 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
905 | return 0; |
906 | ||
da86774e JA |
907 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
908 | ||
909 | f = fopen(tmp, "r+"); | |
910 | if (!f) { | |
cbf5c1d7 | 911 | if (errno == ENOENT) { |
5ec10eaa JA |
912 | log_err("fio: os or kernel doesn't support IO scheduler" |
913 | " switching\n"); | |
cbf5c1d7 JA |
914 | return 0; |
915 | } | |
916 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
917 | return 1; |
918 | } | |
919 | ||
920 | /* | |
921 | * Set io scheduler. | |
922 | */ | |
2dc1bbeb | 923 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 924 | if (ferror(f) || ret != 1) { |
e1161c32 | 925 | td_verror(td, errno, "fwrite"); |
da86774e JA |
926 | fclose(f); |
927 | return 1; | |
928 | } | |
929 | ||
930 | rewind(f); | |
931 | ||
932 | /* | |
933 | * Read back and check that the selected scheduler is now the default. | |
934 | */ | |
935 | ret = fread(tmp, 1, sizeof(tmp), f); | |
936 | if (ferror(f) || ret < 0) { | |
e1161c32 | 937 | td_verror(td, errno, "fread"); |
da86774e JA |
938 | fclose(f); |
939 | return 1; | |
940 | } | |
941 | ||
2dc1bbeb | 942 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 943 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 944 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 945 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
946 | fclose(f); |
947 | return 1; | |
948 | } | |
949 | ||
950 | fclose(f); | |
951 | return 0; | |
952 | } | |
953 | ||
492158cf JA |
954 | static int keep_running(struct thread_data *td) |
955 | { | |
956 | unsigned long long io_done; | |
957 | ||
20e354ef JA |
958 | if (td->done) |
959 | return 0; | |
492158cf JA |
960 | if (td->o.time_based) |
961 | return 1; | |
962 | if (td->o.loops) { | |
963 | td->o.loops--; | |
964 | return 1; | |
965 | } | |
966 | ||
5ec10eaa JA |
967 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
968 | + td->io_skip_bytes; | |
492158cf JA |
969 | if (io_done < td->o.size) |
970 | return 1; | |
971 | ||
972 | return 0; | |
973 | } | |
974 | ||
b29ee5b3 | 975 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 976 | { |
f0505a14 | 977 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
ebac4655 | 978 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
20dc95c4 | 979 | td->zone_bytes = 0; |
581e7141 JA |
980 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
981 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 982 | |
c1324df1 JA |
983 | td->last_was_sync = 0; |
984 | ||
2ba1c290 JA |
985 | /* |
986 | * reset file done count if we are to start over | |
987 | */ | |
988 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 989 | td->nr_done_files = 0; |
b29ee5b3 JA |
990 | } |
991 | ||
992 | void reset_all_stats(struct thread_data *td) | |
993 | { | |
994 | struct timeval tv; | |
995 | int i; | |
996 | ||
997 | reset_io_counters(td); | |
998 | ||
999 | for (i = 0; i < 2; i++) { | |
1000 | td->io_bytes[i] = 0; | |
1001 | td->io_blocks[i] = 0; | |
1002 | td->io_issues[i] = 0; | |
1003 | td->ts.total_io_u[i] = 0; | |
1004 | } | |
9b836561 | 1005 | |
b29ee5b3 | 1006 | fio_gettime(&tv, NULL); |
cda99fa0 SSP |
1007 | td->ts.runtime[0] = 0; |
1008 | td->ts.runtime[1] = 0; | |
b29ee5b3 JA |
1009 | memcpy(&td->epoch, &tv, sizeof(tv)); |
1010 | memcpy(&td->start, &tv, sizeof(tv)); | |
1011 | } | |
1012 | ||
df9c26b1 | 1013 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
1014 | { |
1015 | struct fio_file *f; | |
1016 | unsigned int i; | |
b29ee5b3 JA |
1017 | |
1018 | reset_io_counters(td); | |
281f9b63 | 1019 | |
24ffd2c2 | 1020 | close_files(td); |
df9c26b1 | 1021 | for_each_file(td, f, i) |
d6aed795 | 1022 | fio_file_clear_done(f); |
f0153948 JA |
1023 | |
1024 | /* | |
1025 | * Set the same seed to get repeatable runs | |
1026 | */ | |
1027 | td_fill_rand_seeds(td); | |
ebac4655 JA |
1028 | } |
1029 | ||
398e8c4d JA |
1030 | static int exec_string(const char *string) |
1031 | { | |
1032 | int ret, newlen = strlen(string) + 1 + 8; | |
1033 | char *str; | |
1034 | ||
1035 | str = malloc(newlen); | |
1036 | sprintf(str, "sh -c %s", string); | |
1037 | ||
1038 | ret = system(str); | |
1039 | if (ret == -1) | |
1040 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1041 | ||
1042 | free(str); | |
1043 | return ret; | |
1044 | } | |
1045 | ||
906c8d75 JA |
1046 | /* |
1047 | * Entry point for the thread based jobs. The process based jobs end up | |
1048 | * here as well, after a little setup. | |
1049 | */ | |
ebac4655 JA |
1050 | static void *thread_main(void *data) |
1051 | { | |
cda99fa0 | 1052 | unsigned long long elapsed; |
ebac4655 | 1053 | struct thread_data *td = data; |
e8462bd8 | 1054 | pthread_condattr_t attr; |
a978ba68 | 1055 | int clear_state; |
ebac4655 | 1056 | |
47f767c1 | 1057 | if (!td->o.use_thread) { |
ebac4655 | 1058 | setsid(); |
47f767c1 JA |
1059 | td->pid = getpid(); |
1060 | } else | |
1061 | td->pid = gettid(); | |
ebac4655 | 1062 | |
5921e80c | 1063 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1064 | |
01743ee1 JA |
1065 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1066 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1067 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1068 | INIT_FLIST_HEAD(&td->io_log_list); | |
1069 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 | 1070 | INIT_FLIST_HEAD(&td->verify_list); |
0d29de83 | 1071 | INIT_FLIST_HEAD(&td->trim_list); |
e8462bd8 | 1072 | pthread_mutex_init(&td->io_u_lock, NULL); |
bb5d7d0b | 1073 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1074 | |
e8462bd8 JA |
1075 | pthread_condattr_init(&attr); |
1076 | pthread_cond_init(&td->verify_cond, &attr); | |
1077 | pthread_cond_init(&td->free_cond, &attr); | |
1078 | ||
db1cd90b | 1079 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1080 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1081 | fio_mutex_up(startup_mutex); |
29adda3c | 1082 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1083 | fio_mutex_down(td->mutex); |
29adda3c | 1084 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1085 | |
1086 | /* | |
cdd18ad8 | 1087 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1088 | * eating a file descriptor |
1089 | */ | |
cdd18ad8 | 1090 | fio_mutex_remove(td->mutex); |
db1cd90b | 1091 | |
dfc6e751 SH |
1092 | /* |
1093 | * A new gid requires privilege, so we need to do this before setting | |
1094 | * the uid. | |
1095 | */ | |
e0b0d892 JA |
1096 | if (td->o.gid != -1U && setgid(td->o.gid)) { |
1097 | td_verror(td, errno, "setgid"); | |
1098 | goto err; | |
1099 | } | |
dfc6e751 SH |
1100 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1101 | td_verror(td, errno, "setuid"); | |
1102 | goto err; | |
1103 | } | |
e0b0d892 | 1104 | |
be18a94f JA |
1105 | /* |
1106 | * If we have a gettimeofday() thread, make sure we exclude that | |
1107 | * thread from this job | |
1108 | */ | |
1109 | if (td->o.gtod_cpu) | |
1110 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); | |
1111 | ||
1112 | /* | |
1113 | * Set affinity first, in case it has an impact on the memory | |
1114 | * allocations. | |
1115 | */ | |
1116 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
1117 | td_verror(td, errno, "cpu_set_affinity"); | |
1118 | goto err; | |
1119 | } | |
1120 | ||
d84f8d49 JA |
1121 | /* |
1122 | * May alter parameters that init_io_u() will use, so we need to | |
1123 | * do this first. | |
1124 | */ | |
1125 | if (init_iolog(td)) | |
1126 | goto err; | |
1127 | ||
ebac4655 | 1128 | if (init_io_u(td)) |
db1cd90b | 1129 | goto err; |
ebac4655 | 1130 | |
e8462bd8 JA |
1131 | if (td->o.verify_async && verify_async_init(td)) |
1132 | goto err; | |
1133 | ||
ac684785 | 1134 | if (td->ioprio_set) { |
ebac4655 | 1135 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1136 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1137 | goto err; |
ebac4655 JA |
1138 | } |
1139 | } | |
1140 | ||
6adb38a1 | 1141 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1142 | goto err; |
1143 | ||
2dc1bbeb | 1144 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1145 | td_verror(td, errno, "nice"); |
db1cd90b | 1146 | goto err; |
b6f4d880 JA |
1147 | } |
1148 | ||
2dc1bbeb | 1149 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1150 | goto err; |
37f56873 | 1151 | |
2dc1bbeb | 1152 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1153 | goto err; |
1154 | ||
7d6c5283 JA |
1155 | if (td_io_init(td)) |
1156 | goto err; | |
1157 | ||
68727076 JA |
1158 | if (init_random_map(td)) |
1159 | goto err; | |
1160 | ||
2dc1bbeb | 1161 | if (td->o.exec_prerun) { |
398e8c4d | 1162 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1163 | goto err; |
1164 | } | |
4e0ba8af | 1165 | |
afad68f7 ZY |
1166 | if (td->o.pre_read) { |
1167 | if (pre_read_files(td) < 0) | |
1168 | goto err; | |
1169 | } | |
1170 | ||
69008999 | 1171 | fio_gettime(&td->epoch, NULL); |
c8aaba19 | 1172 | getrusage(RUSAGE_SELF, &td->ru_start); |
69008999 | 1173 | |
a978ba68 | 1174 | clear_state = 0; |
492158cf | 1175 | while (keep_running(td)) { |
02bcaa8c | 1176 | fio_gettime(&td->start, NULL); |
f0505a14 JA |
1177 | memcpy(&td->stat_sample_time[0], &td->start, sizeof(td->start)); |
1178 | memcpy(&td->stat_sample_time[1], &td->start, sizeof(td->start)); | |
993bf48b | 1179 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1180 | |
cf2dd8d6 | 1181 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
f0505a14 | 1182 | memcpy(&td->lastrate, &td->stat_sample_time, |
5ec10eaa | 1183 | sizeof(td->lastrate)); |
ebac4655 | 1184 | |
df9c26b1 JA |
1185 | if (clear_state) |
1186 | clear_io_state(td); | |
a978ba68 | 1187 | |
ebac4655 JA |
1188 | prune_io_piece_log(td); |
1189 | ||
ba0fbe10 | 1190 | do_io(td); |
ebac4655 | 1191 | |
a978ba68 JA |
1192 | clear_state = 1; |
1193 | ||
38d77cae | 1194 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1195 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1196 | td->ts.runtime[DDIR_READ] += elapsed; |
38d77cae JA |
1197 | } |
1198 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1199 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1200 | td->ts.runtime[DDIR_WRITE] += elapsed; |
38d77cae | 1201 | } |
5ec10eaa | 1202 | |
ebac4655 JA |
1203 | if (td->error || td->terminate) |
1204 | break; | |
1205 | ||
e84c73a8 SL |
1206 | if (!td->o.do_verify || |
1207 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1208 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1209 | continue; |
1210 | ||
df9c26b1 | 1211 | clear_io_state(td); |
a978ba68 | 1212 | |
02bcaa8c | 1213 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1214 | |
1215 | do_verify(td); | |
1216 | ||
cda99fa0 | 1217 | td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1218 | |
1219 | if (td->error || td->terminate) | |
1220 | break; | |
1221 | } | |
1222 | ||
36dff966 | 1223 | update_rusage_stat(td); |
cda99fa0 SSP |
1224 | td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000; |
1225 | td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000; | |
756867bd JA |
1226 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1227 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1228 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1229 | ||
d2bbb8d9 | 1230 | fio_mutex_down(writeout_mutex); |
7b9f733a | 1231 | if (td->bw_log) { |
f484470b | 1232 | if (td->o.bw_log_file) { |
7b9f733a | 1233 | finish_log_named(td, td->bw_log, |
f484470b JA |
1234 | td->o.bw_log_file, "bw"); |
1235 | } else | |
7b9f733a | 1236 | finish_log(td, td->bw_log, "bw"); |
e3cedca7 | 1237 | } |
7b9f733a | 1238 | if (td->lat_log) { |
02af0988 | 1239 | if (td->o.lat_log_file) { |
7b9f733a | 1240 | finish_log_named(td, td->lat_log, |
02af0988 JA |
1241 | td->o.lat_log_file, "lat"); |
1242 | } else | |
7b9f733a | 1243 | finish_log(td, td->lat_log, "lat"); |
02af0988 | 1244 | } |
7b9f733a | 1245 | if (td->slat_log) { |
f484470b | 1246 | if (td->o.lat_log_file) { |
7b9f733a | 1247 | finish_log_named(td, td->slat_log, |
1ca95738 | 1248 | td->o.lat_log_file, "slat"); |
f484470b | 1249 | } else |
7b9f733a | 1250 | finish_log(td, td->slat_log, "slat"); |
e3cedca7 | 1251 | } |
7b9f733a | 1252 | if (td->clat_log) { |
f484470b | 1253 | if (td->o.lat_log_file) { |
7b9f733a | 1254 | finish_log_named(td, td->clat_log, |
f484470b JA |
1255 | td->o.lat_log_file, "clat"); |
1256 | } else | |
7b9f733a | 1257 | finish_log(td, td->clat_log, "clat"); |
e3cedca7 | 1258 | } |
d2bbb8d9 | 1259 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1260 | if (td->o.exec_postrun) |
1261 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1262 | |
1263 | if (exitall_on_terminate) | |
390c40e2 | 1264 | terminate_threads(td->groupid); |
ebac4655 JA |
1265 | |
1266 | err: | |
5bf13a5a | 1267 | if (td->error) |
4ceb30d4 | 1268 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1269 | td->verror); |
b4707ed7 JA |
1270 | |
1271 | if (td->o.verify_async) | |
1272 | verify_async_exit(td); | |
1273 | ||
24ffd2c2 | 1274 | close_and_free_files(td); |
2866c82d | 1275 | close_ioengine(td); |
ebac4655 | 1276 | cleanup_io_u(td); |
6adb38a1 | 1277 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1278 | |
d2ce18b5 JA |
1279 | if (td->o.cpumask_set) { |
1280 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1281 | ||
1282 | td_verror(td, ret, "fio_cpuset_exit"); | |
1283 | } | |
4e78e405 | 1284 | |
f29b25a3 JA |
1285 | /* |
1286 | * do this very late, it will log file closing as well | |
1287 | */ | |
1288 | if (td->o.write_iolog_file) | |
1289 | write_iolog_close(td); | |
1290 | ||
d23bb327 | 1291 | options_mem_free(td); |
ebac4655 | 1292 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1293 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1294 | } |
1295 | ||
906c8d75 JA |
1296 | /* |
1297 | * We cannot pass the td data into a forked process, so attach the td and | |
1298 | * pass it to the thread worker. | |
1299 | */ | |
a6418147 | 1300 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1301 | { |
1302 | struct thread_data *td; | |
a6418147 | 1303 | void *data, *ret; |
ebac4655 | 1304 | |
65258961 | 1305 | #ifndef __hpux |
ebac4655 JA |
1306 | data = shmat(shmid, NULL, 0); |
1307 | if (data == (void *) -1) { | |
a6418147 JA |
1308 | int __err = errno; |
1309 | ||
ebac4655 | 1310 | perror("shmat"); |
a6418147 | 1311 | return __err; |
ebac4655 | 1312 | } |
65258961 JA |
1313 | #else |
1314 | /* | |
1315 | * HP-UX inherits shm mappings? | |
1316 | */ | |
1317 | data = threads; | |
1318 | #endif | |
ebac4655 JA |
1319 | |
1320 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1321 | ret = thread_main(td); |
ebac4655 | 1322 | shmdt(data); |
43d76807 | 1323 | return (int) (unsigned long) ret; |
ebac4655 JA |
1324 | } |
1325 | ||
906c8d75 JA |
1326 | /* |
1327 | * Run over the job map and reap the threads that have exited, if any. | |
1328 | */ | |
ebac4655 JA |
1329 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) |
1330 | { | |
34572e28 | 1331 | struct thread_data *td; |
1f809d15 | 1332 | int i, cputhreads, realthreads, pending, status, ret; |
ebac4655 JA |
1333 | |
1334 | /* | |
1335 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1336 | */ | |
1f809d15 | 1337 | realthreads = pending = cputhreads = 0; |
34572e28 | 1338 | for_each_td(td, i) { |
3707f45b | 1339 | int flags = 0; |
a2f77c9f | 1340 | |
84585003 JA |
1341 | /* |
1342 | * ->io_ops is NULL for a thread that has closed its | |
1343 | * io engine | |
1344 | */ | |
ba0fbe10 | 1345 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1346 | cputhreads++; |
1f809d15 JA |
1347 | else |
1348 | realthreads++; | |
b990b5c0 | 1349 | |
a1dedc1f AC |
1350 | if (!td->pid) { |
1351 | pending++; | |
1352 | continue; | |
1353 | } | |
1354 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1355 | continue; |
2dc1bbeb | 1356 | if (td->o.use_thread) { |
3707f45b JA |
1357 | if (td->runstate == TD_EXITED) { |
1358 | td_set_runstate(td, TD_REAPED); | |
1359 | goto reaped; | |
1360 | } | |
1361 | continue; | |
1362 | } | |
a2f77c9f JA |
1363 | |
1364 | flags = WNOHANG; | |
1365 | if (td->runstate == TD_EXITED) | |
1366 | flags = 0; | |
1367 | ||
1368 | /* | |
1369 | * check if someone quit or got killed in an unusual way | |
1370 | */ | |
1371 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1372 | if (ret < 0) { |
a2f77c9f | 1373 | if (errno == ECHILD) { |
5921e80c JA |
1374 | log_err("fio: pid=%d disappeared %d\n", |
1375 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1376 | td_set_runstate(td, TD_REAPED); |
1377 | goto reaped; | |
1378 | } | |
1379 | perror("waitpid"); | |
1380 | } else if (ret == td->pid) { | |
1381 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1382 | int sig = WTERMSIG(status); |
1383 | ||
03e20d68 | 1384 | if (sig != SIGTERM) |
5ec10eaa | 1385 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1386 | (int) td->pid, sig); |
fab6aa71 JA |
1387 | td_set_runstate(td, TD_REAPED); |
1388 | goto reaped; | |
1389 | } | |
a2f77c9f JA |
1390 | if (WIFEXITED(status)) { |
1391 | if (WEXITSTATUS(status) && !td->error) | |
1392 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1393 | |
a2f77c9f JA |
1394 | td_set_runstate(td, TD_REAPED); |
1395 | goto reaped; | |
a6418147 JA |
1396 | } |
1397 | } | |
ebac4655 | 1398 | |
a2f77c9f JA |
1399 | /* |
1400 | * thread is not dead, continue | |
1401 | */ | |
e48676ba | 1402 | pending++; |
a2f77c9f | 1403 | continue; |
fab6aa71 | 1404 | reaped: |
ebac4655 | 1405 | (*nr_running)--; |
581e7141 JA |
1406 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1407 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1408 | if (!td->pid) |
1409 | pending--; | |
a2f77c9f JA |
1410 | |
1411 | if (td->error) | |
1412 | exit_value++; | |
d3eeeabc JA |
1413 | |
1414 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1415 | } |
b990b5c0 | 1416 | |
1f809d15 | 1417 | if (*nr_running == cputhreads && !pending && realthreads) |
390c40e2 | 1418 | terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1419 | } |
1420 | ||
be4ecfdf JA |
1421 | static void *gtod_thread_main(void *data) |
1422 | { | |
1423 | fio_mutex_up(startup_mutex); | |
1424 | ||
1425 | /* | |
1426 | * As long as we have jobs around, update the clock. It would be nice | |
1427 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1428 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1429 | * it - we don't want to lose precision. | |
1430 | */ | |
1431 | while (threads) { | |
1432 | fio_gtod_update(); | |
1433 | nop; | |
1434 | } | |
1435 | ||
1436 | return NULL; | |
1437 | } | |
1438 | ||
1439 | static int fio_start_gtod_thread(void) | |
1440 | { | |
304a47c7 | 1441 | pthread_attr_t attr; |
f42c153d JA |
1442 | int ret; |
1443 | ||
304a47c7 VA |
1444 | pthread_attr_init(&attr); |
1445 | pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); | |
1446 | ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); | |
1447 | pthread_attr_destroy(&attr); | |
f42c153d JA |
1448 | if (ret) { |
1449 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1450 | return 1; |
1451 | } | |
f42c153d JA |
1452 | |
1453 | ret = pthread_detach(gtod_thread); | |
1454 | if (ret) { | |
1455 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1456 | return 1; |
1457 | } | |
1458 | ||
29adda3c | 1459 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1460 | fio_mutex_down(startup_mutex); |
29adda3c | 1461 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1462 | return 0; |
1463 | } | |
1464 | ||
906c8d75 JA |
1465 | /* |
1466 | * Main function for kicking off and reaping jobs, as needed. | |
1467 | */ | |
ebac4655 JA |
1468 | static void run_threads(void) |
1469 | { | |
ebac4655 JA |
1470 | struct thread_data *td; |
1471 | unsigned long spent; | |
1472 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 | 1473 | |
2f9ade3c JA |
1474 | if (fio_pin_memory()) |
1475 | return; | |
ebac4655 | 1476 | |
be4ecfdf JA |
1477 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1478 | return; | |
1479 | ||
c6ae0a5b | 1480 | if (!terse_output) { |
4ceb30d4 | 1481 | log_info("Starting "); |
9cedf167 | 1482 | if (nr_thread) |
4ceb30d4 | 1483 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1484 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1485 | if (nr_process) { |
1486 | if (nr_thread) | |
1487 | printf(" and "); | |
4ceb30d4 | 1488 | log_info("%d process%s", nr_process, |
5ec10eaa | 1489 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1490 | } |
4ceb30d4 | 1491 | log_info("\n"); |
c6ae0a5b JA |
1492 | fflush(stdout); |
1493 | } | |
c04f7ec3 | 1494 | |
697a606c | 1495 | set_sig_handlers(); |
4efa970e | 1496 | |
ebac4655 JA |
1497 | todo = thread_number; |
1498 | nr_running = 0; | |
1499 | nr_started = 0; | |
1500 | m_rate = t_rate = 0; | |
1501 | ||
34572e28 | 1502 | for_each_td(td, i) { |
263e529f | 1503 | print_status_init(td->thread_number - 1); |
ebac4655 | 1504 | |
067430bb | 1505 | if (!td->o.create_serialize) |
ebac4655 JA |
1506 | continue; |
1507 | ||
1508 | /* | |
1509 | * do file setup here so it happens sequentially, | |
1510 | * we don't want X number of threads getting their | |
1511 | * client data interspersed on disk | |
1512 | */ | |
53cdc686 | 1513 | if (setup_files(td)) { |
5bf13a5a JA |
1514 | exit_value++; |
1515 | if (td->error) | |
5921e80c JA |
1516 | log_err("fio: pid=%d, err=%d/%s\n", |
1517 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1518 | td_set_runstate(td, TD_REAPED); |
1519 | todo--; | |
c69e8875 JA |
1520 | } else { |
1521 | struct fio_file *f; | |
2b13e716 | 1522 | unsigned int j; |
c69e8875 JA |
1523 | |
1524 | /* | |
1525 | * for sharing to work, each job must always open | |
1526 | * its own files. so close them, if we opened them | |
1527 | * for creation | |
1528 | */ | |
2b13e716 | 1529 | for_each_file(td, f, j) { |
22a57ba8 JA |
1530 | if (fio_file_open(f)) |
1531 | td_io_close_file(td, f); | |
22a57ba8 | 1532 | } |
5ec10eaa | 1533 | } |
ebac4655 JA |
1534 | } |
1535 | ||
a2f77c9f JA |
1536 | set_genesis_time(); |
1537 | ||
ebac4655 | 1538 | while (todo) { |
fca70358 | 1539 | struct thread_data *map[REAL_MAX_JOBS]; |
75154845 JA |
1540 | struct timeval this_start; |
1541 | int this_jobs = 0, left; | |
1542 | ||
ebac4655 JA |
1543 | /* |
1544 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1545 | */ | |
34572e28 | 1546 | for_each_td(td, i) { |
ebac4655 JA |
1547 | if (td->runstate != TD_NOT_CREATED) |
1548 | continue; | |
1549 | ||
1550 | /* | |
1551 | * never got a chance to start, killed by other | |
1552 | * thread for some reason | |
1553 | */ | |
1554 | if (td->terminate) { | |
1555 | todo--; | |
1556 | continue; | |
1557 | } | |
1558 | ||
2dc1bbeb | 1559 | if (td->o.start_delay) { |
263e529f | 1560 | spent = mtime_since_genesis(); |
ebac4655 | 1561 | |
2dc1bbeb | 1562 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1563 | continue; |
1564 | } | |
1565 | ||
6f88bcb0 JA |
1566 | if (td->o.stonewall && (nr_started || nr_running)) { |
1567 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1568 | td->o.name); | |
ebac4655 | 1569 | break; |
6f88bcb0 | 1570 | } |
ebac4655 | 1571 | |
067430bb JA |
1572 | init_disk_util(td); |
1573 | ||
75154845 JA |
1574 | /* |
1575 | * Set state to created. Thread will transition | |
1576 | * to TD_INITIALIZED when it's done setting up. | |
1577 | */ | |
ebac4655 | 1578 | td_set_runstate(td, TD_CREATED); |
75154845 | 1579 | map[this_jobs++] = td; |
ebac4655 JA |
1580 | nr_started++; |
1581 | ||
2dc1bbeb | 1582 | if (td->o.use_thread) { |
f42c153d JA |
1583 | int ret; |
1584 | ||
ee56ad50 | 1585 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1586 | ret = pthread_create(&td->thread, NULL, |
1587 | thread_main, td); | |
1588 | if (ret) { | |
1589 | log_err("pthread_create: %s\n", | |
1590 | strerror(ret)); | |
ebac4655 | 1591 | nr_started--; |
c8bb6faf | 1592 | break; |
ebac4655 | 1593 | } |
f42c153d JA |
1594 | ret = pthread_detach(td->thread); |
1595 | if (ret) | |
1596 | log_err("pthread_detach: %s", | |
1597 | strerror(ret)); | |
ebac4655 | 1598 | } else { |
5e1d306e | 1599 | pid_t pid; |
ee56ad50 | 1600 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1601 | pid = fork(); |
1602 | if (!pid) { | |
a6418147 JA |
1603 | int ret = fork_main(shm_id, i); |
1604 | ||
5e1d306e JA |
1605 | _exit(ret); |
1606 | } else if (i == fio_debug_jobno) | |
1607 | *fio_debug_jobp = pid; | |
ebac4655 | 1608 | } |
29adda3c | 1609 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1610 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1611 | log_err("fio: job startup hung? exiting.\n"); | |
1612 | terminate_threads(TERMINATE_ALL); | |
1613 | fio_abort = 1; | |
1614 | nr_started--; | |
1615 | break; | |
1616 | } | |
29adda3c | 1617 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1618 | } |
1619 | ||
1620 | /* | |
75154845 JA |
1621 | * Wait for the started threads to transition to |
1622 | * TD_INITIALIZED. | |
ebac4655 | 1623 | */ |
02bcaa8c | 1624 | fio_gettime(&this_start, NULL); |
75154845 | 1625 | left = this_jobs; |
6ce15a32 | 1626 | while (left && !fio_abort) { |
75154845 JA |
1627 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1628 | break; | |
1629 | ||
1630 | usleep(100000); | |
1631 | ||
1632 | for (i = 0; i < this_jobs; i++) { | |
1633 | td = map[i]; | |
1634 | if (!td) | |
1635 | continue; | |
b6f4d880 | 1636 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1637 | map[i] = NULL; |
1638 | left--; | |
b6f4d880 JA |
1639 | } else if (td->runstate >= TD_EXITED) { |
1640 | map[i] = NULL; | |
1641 | left--; | |
1642 | todo--; | |
1643 | nr_running++; /* work-around... */ | |
75154845 JA |
1644 | } |
1645 | } | |
1646 | } | |
1647 | ||
1648 | if (left) { | |
3b70d7e5 | 1649 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1650 | for (i = 0; i < this_jobs; i++) { |
1651 | td = map[i]; | |
1652 | if (!td) | |
1653 | continue; | |
1654 | kill(td->pid, SIGTERM); | |
1655 | } | |
1656 | break; | |
1657 | } | |
1658 | ||
1659 | /* | |
b6f4d880 | 1660 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1661 | */ |
34572e28 | 1662 | for_each_td(td, i) { |
75154845 | 1663 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1664 | continue; |
1665 | ||
b29ee5b3 JA |
1666 | if (in_ramp_time(td)) |
1667 | td_set_runstate(td, TD_RAMP); | |
1668 | else | |
1669 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1670 | nr_running++; |
1671 | nr_started--; | |
581e7141 JA |
1672 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1673 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1674 | todo--; |
cdd18ad8 | 1675 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1676 | } |
1677 | ||
1678 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1679 | ||
1680 | if (todo) | |
1681 | usleep(100000); | |
1682 | } | |
1683 | ||
1684 | while (nr_running) { | |
1685 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1686 | usleep(10000); | |
1687 | } | |
1688 | ||
1689 | update_io_ticks(); | |
2f9ade3c | 1690 | fio_unpin_memory(); |
ebac4655 JA |
1691 | } |
1692 | ||
50d16976 | 1693 | int exec_run(void) |
ebac4655 | 1694 | { |
a37f69b7 | 1695 | if (nr_clients) |
37db14fe | 1696 | return fio_handle_clients(); |
07b3232d JA |
1697 | if (exec_profile && load_profile(exec_profile)) |
1698 | return 1; | |
1699 | ||
74929ac2 JA |
1700 | if (!thread_number) |
1701 | return 0; | |
1702 | ||
bb3884d8 JA |
1703 | if (write_bw_log) { |
1704 | setup_log(&agg_io_log[DDIR_READ]); | |
1705 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1706 | } | |
1707 | ||
cdd18ad8 | 1708 | startup_mutex = fio_mutex_init(0); |
03e20d68 BC |
1709 | if (startup_mutex == NULL) |
1710 | return 1; | |
d2bbb8d9 | 1711 | writeout_mutex = fio_mutex_init(1); |
03e20d68 BC |
1712 | if (writeout_mutex == NULL) |
1713 | return 1; | |
07739b57 | 1714 | |
a2f77c9f | 1715 | set_genesis_time(); |
7985701b | 1716 | create_disk_util_thread(); |
ebac4655 | 1717 | |
dae5341c JA |
1718 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1719 | INIT_FLIST_HEAD(cgroup_list); | |
1720 | ||
ebac4655 | 1721 | run_threads(); |
6ce15a32 | 1722 | |
bb3884d8 | 1723 | if (!fio_abort) { |
6ce15a32 | 1724 | show_run_stats(); |
bb3884d8 | 1725 | if (write_bw_log) { |
5ec10eaa JA |
1726 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1727 | __finish_log(agg_io_log[DDIR_WRITE], | |
1728 | "agg-write_bw.log"); | |
bb3884d8 JA |
1729 | } |
1730 | } | |
ebac4655 | 1731 | |
dae5341c JA |
1732 | cgroup_kill(cgroup_list); |
1733 | sfree(cgroup_list); | |
61eb313e | 1734 | sfree(cgroup_mnt); |
39f22027 | 1735 | |
cdd18ad8 | 1736 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1737 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1738 | return exit_value; |
ebac4655 | 1739 | } |
50d16976 JA |
1740 | |
1741 | void reset_fio_state(void) | |
1742 | { | |
1743 | groupid = 0; | |
1744 | thread_number = 0; | |
1745 | nr_process = 0; | |
1746 | nr_thread = 0; | |
1747 | done_secs = 0; | |
1748 | } | |
1749 | ||
1750 | int main(int argc, char *argv[], char *envp[]) | |
1751 | { | |
1752 | long ps; | |
1753 | ||
1754 | arch_init(envp); | |
1755 | ||
1756 | sinit(); | |
1757 | ||
1758 | /* | |
1759 | * We need locale for number printing, if it isn't set then just | |
1760 | * go with the US format. | |
1761 | */ | |
1762 | if (!getenv("LC_NUMERIC")) | |
1763 | setlocale(LC_NUMERIC, "en_US"); | |
1764 | ||
1765 | ps = sysconf(_SC_PAGESIZE); | |
1766 | if (ps < 0) { | |
1767 | log_err("Failed to get page size\n"); | |
1768 | return 1; | |
1769 | } | |
1770 | ||
1771 | page_size = ps; | |
1772 | page_mask = ps - 1; | |
1773 | ||
1774 | fio_keywords_init(); | |
1775 | ||
1776 | if (parse_options(argc, argv)) | |
1777 | return 1; | |
1778 | ||
1779 | return exec_run(); | |
1780 | } |