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; | |
11e950bd JA |
58 | unsigned int thread_number = 0; |
59 | unsigned int nr_process = 0; | |
60 | unsigned 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 | |
d09a64a0 JA |
65 | /* |
66 | * Just expose an empty list, if the OS does not support disk util stats | |
67 | */ | |
68 | #ifndef FIO_HAVE_DISK_UTIL | |
69 | FLIST_HEAD(disk_list); | |
70 | #endif | |
71 | ||
cdd18ad8 | 72 | static struct fio_mutex *startup_mutex; |
d2bbb8d9 | 73 | static struct fio_mutex *writeout_mutex; |
6ce15a32 | 74 | static volatile int fio_abort; |
437c9b71 | 75 | static int exit_value; |
7985701b | 76 | static pthread_t disk_util_thread; |
dae5341c | 77 | static struct flist_head *cgroup_list; |
6adb38a1 | 78 | static char *cgroup_mnt; |
ebac4655 | 79 | |
d3cc4ebf JA |
80 | unsigned long arch_flags = 0; |
81 | ||
bb3884d8 JA |
82 | struct io_log *agg_io_log[2]; |
83 | ||
75154845 | 84 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 85 | |
cca84643 JA |
86 | static const char *fio_os_strings[os_nr] = { |
87 | "Invalid", | |
88 | "Linux", | |
89 | "AIX", | |
90 | "FreeBSD", | |
91 | "HP-UX", | |
92 | "OSX", | |
93 | "NetBSD", | |
94 | "Solaris", | |
95 | "Windows" | |
96 | }; | |
97 | ||
98 | static const char *fio_arch_strings[arch_nr] = { | |
99 | "Invalid", | |
100 | "x86-64", | |
101 | "x86", | |
102 | "ppc", | |
103 | "ia64", | |
104 | "s390", | |
105 | "alpha", | |
106 | "sparc", | |
107 | "sparc64", | |
108 | "arm", | |
109 | "sh", | |
110 | "hppa", | |
111 | "generic" | |
112 | }; | |
113 | ||
114 | const char *fio_get_os_string(int nr) | |
115 | { | |
116 | if (nr < os_nr) | |
117 | return fio_os_strings[nr]; | |
118 | ||
119 | return NULL; | |
120 | } | |
121 | ||
122 | const char *fio_get_arch_string(int nr) | |
123 | { | |
124 | if (nr < arch_nr) | |
125 | return fio_arch_strings[nr]; | |
126 | ||
127 | return NULL; | |
128 | } | |
129 | ||
b29ee5b3 | 130 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 131 | { |
d8fab1b6 JA |
132 | if (td->runstate == runstate) |
133 | return; | |
134 | ||
5921e80c JA |
135 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
136 | td->runstate, runstate); | |
6ce15a32 JA |
137 | td->runstate = runstate; |
138 | } | |
139 | ||
cc0df00a | 140 | void fio_terminate_threads(int group_id) |
ebac4655 | 141 | { |
34572e28 | 142 | struct thread_data *td; |
ebac4655 JA |
143 | int i; |
144 | ||
6f88bcb0 JA |
145 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
146 | ||
34572e28 | 147 | for_each_td(td, i) { |
ebac4655 | 148 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 149 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 150 | td->o.name, (int) td->pid); |
ad830ed7 JA |
151 | td->terminate = 1; |
152 | td->o.start_delay = 0; | |
153 | ||
7a93ab19 JA |
154 | /* |
155 | * if the thread is running, just let it exit | |
156 | */ | |
6aa27816 MC |
157 | if (!td->pid) |
158 | continue; | |
159 | else if (td->runstate < TD_RAMP) | |
03e20d68 | 160 | kill(td->pid, SIGTERM); |
f63f7f3b JA |
161 | else { |
162 | struct ioengine_ops *ops = td->io_ops; | |
163 | ||
03e20d68 BC |
164 | if (ops && (ops->flags & FIO_SIGTERM)) |
165 | kill(td->pid, SIGTERM); | |
f63f7f3b | 166 | } |
ebac4655 JA |
167 | } |
168 | } | |
169 | } | |
170 | ||
697a606c JA |
171 | static void sig_int(int sig) |
172 | { | |
173 | if (threads) { | |
7b821684 JA |
174 | if (is_backend) |
175 | fio_server_got_signal(sig); | |
176 | else { | |
177 | log_info("\nfio: terminating on signal %d\n", sig); | |
178 | fflush(stdout); | |
179 | exit_value = 128; | |
180 | } | |
181 | ||
cc0df00a | 182 | fio_terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
183 | } |
184 | } | |
185 | ||
7985701b | 186 | static void *disk_thread_main(void *data) |
a5b01f1b | 187 | { |
7985701b JA |
188 | fio_mutex_up(startup_mutex); |
189 | ||
190 | while (threads) { | |
191 | usleep(DISK_UTIL_MSEC * 1000); | |
9a793c2f JA |
192 | if (!threads) |
193 | break; | |
7985701b | 194 | update_io_ticks(); |
cf451d1e | 195 | |
af9c9fb3 | 196 | if (!is_backend) |
cf451d1e | 197 | print_thread_status(); |
7985701b JA |
198 | } |
199 | ||
200 | return NULL; | |
a5b01f1b JA |
201 | } |
202 | ||
7985701b | 203 | static int create_disk_util_thread(void) |
a5b01f1b | 204 | { |
7985701b | 205 | int ret; |
a5b01f1b | 206 | |
69204d6e | 207 | ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); |
7985701b JA |
208 | if (ret) { |
209 | log_err("Can't create disk util thread: %s\n", strerror(ret)); | |
210 | return 1; | |
211 | } | |
a5b01f1b | 212 | |
7985701b JA |
213 | ret = pthread_detach(disk_util_thread); |
214 | if (ret) { | |
215 | log_err("Can't detatch disk util thread: %s\n", strerror(ret)); | |
216 | return 1; | |
217 | } | |
03e20d68 | 218 | |
7985701b JA |
219 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
220 | fio_mutex_down(startup_mutex); | |
221 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
222 | return 0; | |
a5b01f1b JA |
223 | } |
224 | ||
697a606c JA |
225 | static void set_sig_handlers(void) |
226 | { | |
227 | struct sigaction act; | |
228 | ||
697a606c JA |
229 | memset(&act, 0, sizeof(act)); |
230 | act.sa_handler = sig_int; | |
231 | act.sa_flags = SA_RESTART; | |
232 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 233 | |
5f8f9726 | 234 | memset(&act, 0, sizeof(act)); |
b20b6be4 | 235 | act.sa_handler = sig_int; |
5f8f9726 | 236 | act.sa_flags = SA_RESTART; |
03e20d68 | 237 | sigaction(SIGTERM, &act, NULL); |
cc0df00a JA |
238 | |
239 | if (is_backend) { | |
240 | memset(&act, 0, sizeof(act)); | |
241 | act.sa_handler = sig_int; | |
242 | act.sa_flags = SA_RESTART; | |
243 | sigaction(SIGPIPE, &act, NULL); | |
244 | } | |
697a606c JA |
245 | } |
246 | ||
906c8d75 JA |
247 | /* |
248 | * Check if we are above the minimum rate given. | |
249 | */ | |
581e7141 | 250 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
bd93e34b | 251 | enum fio_ddir ddir) |
ebac4655 | 252 | { |
0904200b | 253 | unsigned long long bytes = 0; |
4e991c23 | 254 | unsigned long iops = 0; |
ebac4655 JA |
255 | unsigned long spent; |
256 | unsigned long rate; | |
581e7141 JA |
257 | unsigned int ratemin = 0; |
258 | unsigned int rate_iops = 0; | |
259 | unsigned int rate_iops_min = 0; | |
ebac4655 | 260 | |
ff58fced JA |
261 | assert(ddir_rw(ddir)); |
262 | ||
d982d873 JA |
263 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
264 | return 0; | |
265 | ||
ebac4655 JA |
266 | /* |
267 | * allow a 2 second settle period in the beginning | |
268 | */ | |
269 | if (mtime_since(&td->start, now) < 2000) | |
270 | return 0; | |
271 | ||
c8eeb9df | 272 | iops += td->this_io_blocks[ddir]; |
581e7141 JA |
273 | bytes += td->this_io_bytes[ddir]; |
274 | ratemin += td->o.ratemin[ddir]; | |
275 | rate_iops += td->o.rate_iops[ddir]; | |
276 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 277 | |
ebac4655 JA |
278 | /* |
279 | * if rate blocks is set, sample is running | |
280 | */ | |
581e7141 JA |
281 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
282 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 283 | if (spent < td->o.ratecycle) |
ebac4655 JA |
284 | return 0; |
285 | ||
581e7141 | 286 | if (td->o.rate[ddir]) { |
4e991c23 JA |
287 | /* |
288 | * check bandwidth specified rate | |
289 | */ | |
581e7141 | 290 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 291 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 292 | ratemin); |
4e991c23 JA |
293 | return 1; |
294 | } else { | |
581e7141 JA |
295 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
296 | if (rate < ratemin || | |
297 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 298 | log_err("%s: min rate %u not met, got" |
b22989b9 | 299 | " %luKB/sec\n", td->o.name, |
581e7141 | 300 | ratemin, rate); |
4e991c23 JA |
301 | return 1; |
302 | } | |
303 | } | |
413dd459 | 304 | } else { |
4e991c23 JA |
305 | /* |
306 | * checks iops specified rate | |
307 | */ | |
581e7141 | 308 | if (iops < rate_iops) { |
5ec10eaa | 309 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 310 | td->o.name, rate_iops); |
413dd459 | 311 | return 1; |
4e991c23 | 312 | } else { |
581e7141 JA |
313 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
314 | if (rate < rate_iops_min || | |
315 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
316 | log_err("%s: min iops rate %u not met," |
317 | " got %lu\n", td->o.name, | |
581e7141 | 318 | rate_iops_min, rate); |
4e991c23 | 319 | } |
413dd459 | 320 | } |
ebac4655 JA |
321 | } |
322 | } | |
323 | ||
581e7141 JA |
324 | td->rate_bytes[ddir] = bytes; |
325 | td->rate_blocks[ddir] = iops; | |
326 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
327 | return 0; |
328 | } | |
329 | ||
581e7141 JA |
330 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
331 | unsigned long *bytes_done) | |
332 | { | |
333 | int ret = 0; | |
334 | ||
335 | if (bytes_done[0]) | |
336 | ret |= __check_min_rate(td, now, 0); | |
337 | if (bytes_done[1]) | |
338 | ret |= __check_min_rate(td, now, 1); | |
339 | ||
340 | return ret; | |
341 | } | |
342 | ||
ebac4655 JA |
343 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
344 | { | |
e4413f05 SN |
345 | if (in_ramp_time(td)) |
346 | return 0; | |
2dc1bbeb | 347 | if (!td->o.timeout) |
ebac4655 | 348 | return 0; |
2dc1bbeb | 349 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
350 | return 1; |
351 | ||
352 | return 0; | |
353 | } | |
354 | ||
906c8d75 JA |
355 | /* |
356 | * When job exits, we can cancel the in-flight IO if we are using async | |
357 | * io. Attempt to do so. | |
358 | */ | |
ebac4655 JA |
359 | static void cleanup_pending_aio(struct thread_data *td) |
360 | { | |
01743ee1 | 361 | struct flist_head *entry, *n; |
ebac4655 JA |
362 | struct io_u *io_u; |
363 | int r; | |
364 | ||
365 | /* | |
366 | * get immediately available events, if any | |
367 | */ | |
581e7141 | 368 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
369 | if (r < 0) |
370 | return; | |
ebac4655 JA |
371 | |
372 | /* | |
373 | * now cancel remaining active events | |
374 | */ | |
2866c82d | 375 | if (td->io_ops->cancel) { |
01743ee1 JA |
376 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
377 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 378 | |
0c6e7517 JA |
379 | /* |
380 | * if the io_u isn't in flight, then that generally | |
381 | * means someone leaked an io_u. complain but fix | |
382 | * it up, so we don't stall here. | |
383 | */ | |
384 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
385 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 386 | put_io_u(td, io_u); |
0c6e7517 JA |
387 | } else { |
388 | r = td->io_ops->cancel(td, io_u); | |
389 | if (!r) | |
390 | put_io_u(td, io_u); | |
391 | } | |
ebac4655 JA |
392 | } |
393 | } | |
394 | ||
97601024 | 395 | if (td->cur_depth) |
581e7141 | 396 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
397 | } |
398 | ||
858a3d47 JA |
399 | /* |
400 | * Helper to handle the final sync of a file. Works just like the normal | |
401 | * io path, just does everything sync. | |
402 | */ | |
403 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
404 | { | |
405 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
406 | int ret; |
407 | ||
408 | if (!io_u) | |
409 | return 1; | |
410 | ||
411 | io_u->ddir = DDIR_SYNC; | |
412 | io_u->file = f; | |
413 | ||
414 | if (td_io_prep(td, io_u)) { | |
415 | put_io_u(td, io_u); | |
416 | return 1; | |
417 | } | |
418 | ||
755200a3 | 419 | requeue: |
858a3d47 | 420 | ret = td_io_queue(td, io_u); |
36167d82 | 421 | if (ret < 0) { |
e1161c32 | 422 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 423 | put_io_u(td, io_u); |
858a3d47 | 424 | return 1; |
36167d82 | 425 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 426 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 427 | return 1; |
36167d82 JA |
428 | } else if (ret == FIO_Q_COMPLETED) { |
429 | if (io_u->error) { | |
e1161c32 | 430 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
431 | return 1; |
432 | } | |
858a3d47 | 433 | |
581e7141 | 434 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 435 | return 1; |
755200a3 JA |
436 | } else if (ret == FIO_Q_BUSY) { |
437 | if (td_io_commit(td)) | |
438 | return 1; | |
439 | goto requeue; | |
858a3d47 JA |
440 | } |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
687c6a23 JA |
445 | static inline void __update_tv_cache(struct thread_data *td) |
446 | { | |
447 | fio_gettime(&td->tv_cache, NULL); | |
448 | } | |
449 | ||
993bf48b JA |
450 | static inline void update_tv_cache(struct thread_data *td) |
451 | { | |
452 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
687c6a23 | 453 | __update_tv_cache(td); |
993bf48b JA |
454 | } |
455 | ||
06842027 SL |
456 | static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, |
457 | int *retptr) | |
f2bba182 RR |
458 | { |
459 | int ret = *retptr; | |
460 | ||
461 | if (ret < 0 || td->error) { | |
462 | int err; | |
463 | ||
f2bba182 RR |
464 | if (ret < 0) |
465 | err = -ret; | |
466 | else | |
467 | err = td->error; | |
468 | ||
06842027 SL |
469 | if (!(td->o.continue_on_error & td_error_type(ddir, err))) |
470 | return 1; | |
471 | ||
f2bba182 RR |
472 | if (td_non_fatal_error(err)) { |
473 | /* | |
474 | * Continue with the I/Os in case of | |
475 | * a non fatal error. | |
476 | */ | |
8a4f8a24 | 477 | update_error_count(td, err); |
f2bba182 RR |
478 | td_clear_error(td); |
479 | *retptr = 0; | |
480 | return 0; | |
8a4f8a24 RR |
481 | } else if (td->o.fill_device && err == ENOSPC) { |
482 | /* | |
483 | * We expect to hit this error if | |
484 | * fill_device option is set. | |
485 | */ | |
486 | td_clear_error(td); | |
487 | td->terminate = 1; | |
488 | return 1; | |
f2bba182 RR |
489 | } else { |
490 | /* | |
491 | * Stop the I/O in case of a fatal | |
492 | * error. | |
493 | */ | |
8a4f8a24 | 494 | update_error_count(td, err); |
f2bba182 RR |
495 | return 1; |
496 | } | |
497 | } | |
498 | ||
499 | return 0; | |
500 | } | |
501 | ||
906c8d75 | 502 | /* |
34403fb1 | 503 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
504 | * reads the blocks back in, and checks the crc/md5 of the data. |
505 | */ | |
1e97cce9 | 506 | static void do_verify(struct thread_data *td) |
ebac4655 | 507 | { |
53cdc686 | 508 | struct fio_file *f; |
36167d82 | 509 | struct io_u *io_u; |
af52b345 JA |
510 | int ret, min_events; |
511 | unsigned int i; | |
e5b401d4 | 512 | |
008618a8 JA |
513 | dprint(FD_VERIFY, "starting loop\n"); |
514 | ||
e5b401d4 JA |
515 | /* |
516 | * sync io first and invalidate cache, to make sure we really | |
517 | * read from disk. | |
518 | */ | |
519 | for_each_file(td, f, i) { | |
d6aed795 | 520 | if (!fio_file_open(f)) |
3d7b485f | 521 | continue; |
b2fdda43 JA |
522 | if (fio_io_sync(td, f)) |
523 | break; | |
524 | if (file_invalidate_cache(td, f)) | |
525 | break; | |
e5b401d4 | 526 | } |
ebac4655 | 527 | |
b2fdda43 JA |
528 | if (td->error) |
529 | return; | |
530 | ||
ebac4655 JA |
531 | td_set_runstate(td, TD_VERIFYING); |
532 | ||
36167d82 JA |
533 | io_u = NULL; |
534 | while (!td->terminate) { | |
4950421a | 535 | int ret2, full; |
5451792e | 536 | |
993bf48b JA |
537 | update_tv_cache(td); |
538 | ||
539 | if (runtime_exceeded(td, &td->tv_cache)) { | |
687c6a23 JA |
540 | __update_tv_cache(td); |
541 | if (runtime_exceeded(td, &td->tv_cache)) { | |
542 | td->terminate = 1; | |
543 | break; | |
544 | } | |
069c2918 | 545 | } |
02bcaa8c | 546 | |
dcb4830d RR |
547 | io_u = __get_io_u(td); |
548 | if (!io_u) | |
549 | break; | |
550 | ||
069c2918 JA |
551 | if (get_next_verify(td, io_u)) { |
552 | put_io_u(td, io_u); | |
ebac4655 | 553 | break; |
069c2918 | 554 | } |
ebac4655 | 555 | |
069c2918 JA |
556 | if (td_io_prep(td, io_u)) { |
557 | put_io_u(td, io_u); | |
53cdc686 | 558 | break; |
069c2918 | 559 | } |
d7762cf8 | 560 | |
e8462bd8 JA |
561 | if (td->o.verify_async) |
562 | io_u->end_io = verify_io_u_async; | |
563 | else | |
564 | io_u->end_io = verify_io_u; | |
53cdc686 | 565 | |
11786802 | 566 | ret = td_io_queue(td, io_u); |
36167d82 JA |
567 | switch (ret) { |
568 | case FIO_Q_COMPLETED: | |
f2bba182 | 569 | if (io_u->error) { |
22819ec2 | 570 | ret = -io_u->error; |
f2bba182 RR |
571 | clear_io_u(td, io_u); |
572 | } else if (io_u->resid) { | |
36167d82 | 573 | int bytes = io_u->xfer_buflen - io_u->resid; |
ebac4655 | 574 | |
9e9d164e JA |
575 | /* |
576 | * zero read, fail | |
577 | */ | |
578 | if (!bytes) { | |
41d9f0fd | 579 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
580 | put_io_u(td, io_u); |
581 | break; | |
582 | } | |
8400d9b2 | 583 | |
36167d82 JA |
584 | io_u->xfer_buflen = io_u->resid; |
585 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
586 | io_u->offset += bytes; |
587 | ||
ff58fced JA |
588 | if (ddir_rw(io_u->ddir)) |
589 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 590 | |
2b13e716 | 591 | f = io_u->file; |
d460eb31 | 592 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
593 | goto sync_done; |
594 | ||
11786802 JA |
595 | requeue_io_u(td, &io_u); |
596 | } else { | |
8400d9b2 | 597 | sync_done: |
581e7141 | 598 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
599 | if (ret < 0) |
600 | break; | |
36167d82 | 601 | } |
36167d82 JA |
602 | continue; |
603 | case FIO_Q_QUEUED: | |
604 | break; | |
755200a3 JA |
605 | case FIO_Q_BUSY: |
606 | requeue_io_u(td, &io_u); | |
5451792e JA |
607 | ret2 = td_io_commit(td); |
608 | if (ret2 < 0) | |
609 | ret = ret2; | |
755200a3 | 610 | break; |
36167d82 JA |
611 | default: |
612 | assert(ret < 0); | |
e1161c32 | 613 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
614 | break; |
615 | } | |
616 | ||
06842027 | 617 | if (break_on_this_error(td, io_u->ddir, &ret)) |
3af6ef39 JA |
618 | break; |
619 | ||
ebac4655 | 620 | /* |
3af6ef39 | 621 | * if we can queue more, do so. but check if there are |
ccc02050 JA |
622 | * completed io_u's first. Note that we can get BUSY even |
623 | * without IO queued, if the system is resource starved. | |
ebac4655 | 624 | */ |
ccc02050 | 625 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 626 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
627 | min_events = min(td->o.iodepth_batch_complete, |
628 | td->cur_depth); | |
1f84f1d9 | 629 | if (full && !min_events && td->o.iodepth_batch_complete != 0) |
838bc709 | 630 | min_events = 1; |
e916b390 | 631 | |
4950421a JA |
632 | do { |
633 | /* | |
634 | * Reap required number of io units, if any, | |
635 | * and do the verification on them through | |
636 | * the callback handler | |
637 | */ | |
581e7141 | 638 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
639 | ret = -1; |
640 | break; | |
641 | } | |
642 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
643 | } | |
644 | if (ret < 0) | |
ebac4655 | 645 | break; |
36167d82 | 646 | } |
ebac4655 | 647 | |
c01c0395 JA |
648 | if (!td->error) { |
649 | min_events = td->cur_depth; | |
650 | ||
651 | if (min_events) | |
581e7141 | 652 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 653 | } else |
ebac4655 JA |
654 | cleanup_pending_aio(td); |
655 | ||
656 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
657 | |
658 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
659 | } |
660 | ||
32cd46a0 | 661 | /* |
906c8d75 | 662 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
663 | * and reaps them, checking for rate and errors along the way. |
664 | */ | |
ebac4655 JA |
665 | static void do_io(struct thread_data *td) |
666 | { | |
af52b345 JA |
667 | unsigned int i; |
668 | int ret = 0; | |
ebac4655 | 669 | |
b29ee5b3 JA |
670 | if (in_ramp_time(td)) |
671 | td_set_runstate(td, TD_RAMP); | |
672 | else | |
673 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 674 | |
457bf399 | 675 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
a917a8b3 | 676 | (!flist_empty(&td->trim_list)) || |
457bf399 | 677 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { |
97601024 | 678 | struct timeval comp_time; |
581e7141 | 679 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 680 | int min_evts = 0; |
ebac4655 | 681 | struct io_u *io_u; |
4950421a | 682 | int ret2, full; |
06842027 | 683 | enum fio_ddir ddir; |
ebac4655 JA |
684 | |
685 | if (td->terminate) | |
686 | break; | |
687 | ||
993bf48b | 688 | update_tv_cache(td); |
97601024 | 689 | |
993bf48b | 690 | if (runtime_exceeded(td, &td->tv_cache)) { |
687c6a23 JA |
691 | __update_tv_cache(td); |
692 | if (runtime_exceeded(td, &td->tv_cache)) { | |
693 | td->terminate = 1; | |
694 | break; | |
695 | } | |
97601024 | 696 | } |
36167d82 | 697 | |
dcb4830d RR |
698 | io_u = get_io_u(td); |
699 | if (!io_u) | |
700 | break; | |
701 | ||
06842027 SL |
702 | ddir = io_u->ddir; |
703 | ||
b67740d3 JA |
704 | /* |
705 | * Add verification end_io handler, if asked to verify | |
706 | * a previously written file. | |
707 | */ | |
715d2b3e JA |
708 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
709 | !td_rw(td)) { | |
e8462bd8 JA |
710 | if (td->o.verify_async) |
711 | io_u->end_io = verify_io_u_async; | |
712 | else | |
713 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 714 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
715 | } else if (in_ramp_time(td)) |
716 | td_set_runstate(td, TD_RAMP); | |
717 | else | |
d8fab1b6 | 718 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 719 | |
11786802 | 720 | ret = td_io_queue(td, io_u); |
36167d82 JA |
721 | switch (ret) { |
722 | case FIO_Q_COMPLETED: | |
f2bba182 | 723 | if (io_u->error) { |
5451792e | 724 | ret = -io_u->error; |
f2bba182 RR |
725 | clear_io_u(td, io_u); |
726 | } else if (io_u->resid) { | |
36167d82 | 727 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 728 | struct fio_file *f = io_u->file; |
36167d82 | 729 | |
9e9d164e JA |
730 | /* |
731 | * zero read, fail | |
732 | */ | |
733 | if (!bytes) { | |
41d9f0fd | 734 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
735 | put_io_u(td, io_u); |
736 | break; | |
737 | } | |
738 | ||
cec6b55d | 739 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 740 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
741 | io_u->offset += bytes; |
742 | ||
ff58fced JA |
743 | if (ddir_rw(io_u->ddir)) |
744 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 745 | |
d460eb31 | 746 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
747 | goto sync_done; |
748 | ||
11786802 JA |
749 | requeue_io_u(td, &io_u); |
750 | } else { | |
5a7c5680 | 751 | sync_done: |
581e7141 JA |
752 | if (__should_check_rate(td, 0) || |
753 | __should_check_rate(td, 1)) | |
993bf48b JA |
754 | fio_gettime(&comp_time, NULL); |
755 | ||
581e7141 JA |
756 | ret = io_u_sync_complete(td, io_u, bytes_done); |
757 | if (ret < 0) | |
758 | break; | |
cec6b55d | 759 | } |
36167d82 JA |
760 | break; |
761 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
762 | /* |
763 | * if the engine doesn't have a commit hook, | |
764 | * the io_u is really queued. if it does have such | |
765 | * a hook, it has to call io_u_queued() itself. | |
766 | */ | |
767 | if (td->io_ops->commit == NULL) | |
768 | io_u_queued(td, io_u); | |
36167d82 | 769 | break; |
755200a3 JA |
770 | case FIO_Q_BUSY: |
771 | requeue_io_u(td, &io_u); | |
5451792e JA |
772 | ret2 = td_io_commit(td); |
773 | if (ret2 < 0) | |
774 | ret = ret2; | |
755200a3 | 775 | break; |
36167d82 JA |
776 | default: |
777 | assert(ret < 0); | |
778 | put_io_u(td, io_u); | |
779 | break; | |
ebac4655 JA |
780 | } |
781 | ||
06842027 | 782 | if (break_on_this_error(td, ddir, &ret)) |
36167d82 JA |
783 | break; |
784 | ||
97601024 | 785 | /* |
ccc02050 JA |
786 | * See if we need to complete some commands. Note that we |
787 | * can get BUSY even without IO queued, if the system is | |
788 | * resource starved. | |
97601024 | 789 | */ |
ccc02050 | 790 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 791 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
792 | min_evts = min(td->o.iodepth_batch_complete, |
793 | td->cur_depth); | |
1f84f1d9 | 794 | if (full && !min_evts && td->o.iodepth_batch_complete != 0) |
36167d82 | 795 | min_evts = 1; |
4950421a | 796 | |
581e7141 JA |
797 | if (__should_check_rate(td, 0) || |
798 | __should_check_rate(td, 1)) | |
993bf48b | 799 | fio_gettime(&comp_time, NULL); |
4950421a JA |
800 | |
801 | do { | |
581e7141 JA |
802 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
803 | if (ret < 0) | |
4950421a JA |
804 | break; |
805 | ||
4950421a | 806 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
807 | } |
808 | ||
4950421a JA |
809 | if (ret < 0) |
810 | break; | |
581e7141 | 811 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
812 | continue; |
813 | ||
581e7141 JA |
814 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
815 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae | 816 | if (exitall_on_terminate) |
cc0df00a | 817 | fio_terminate_threads(td->groupid); |
721938ae JA |
818 | td_verror(td, EIO, "check_min_rate"); |
819 | break; | |
820 | } | |
ebac4655 JA |
821 | } |
822 | ||
2dc1bbeb | 823 | if (td->o.thinktime) { |
9c1f7434 JA |
824 | unsigned long long b; |
825 | ||
826 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 827 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
828 | int left; |
829 | ||
2dc1bbeb | 830 | if (td->o.thinktime_spin) |
6dd6f2cd | 831 | usec_spin(td->o.thinktime_spin); |
48097d5c | 832 | |
2dc1bbeb | 833 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
834 | if (left) |
835 | usec_sleep(td, left); | |
836 | } | |
9c1f7434 | 837 | } |
ebac4655 JA |
838 | } |
839 | ||
d708ac5e | 840 | if (td->trim_entries) |
18320cf4 | 841 | log_err("fio: %d trim entries leaked?\n", td->trim_entries); |
a917a8b3 | 842 | |
aa31f1f1 SL |
843 | if (td->o.fill_device && td->error == ENOSPC) { |
844 | td->error = 0; | |
845 | td->terminate = 1; | |
846 | } | |
4d2413c6 | 847 | if (!td->error) { |
3d7c391d JA |
848 | struct fio_file *f; |
849 | ||
c01c0395 | 850 | i = td->cur_depth; |
d06093e7 | 851 | if (i) { |
581e7141 | 852 | ret = io_u_queued_complete(td, i, NULL); |
d06093e7 SL |
853 | if (td->o.fill_device && td->error == ENOSPC) |
854 | td->error = 0; | |
855 | } | |
ebac4655 | 856 | |
2dc1bbeb | 857 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 858 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
859 | |
860 | for_each_file(td, f, i) { | |
d6aed795 | 861 | if (!fio_file_open(f)) |
3d7b485f | 862 | continue; |
858a3d47 | 863 | fio_io_sync(td, f); |
3d7b485f | 864 | } |
84585003 | 865 | } |
c01c0395 JA |
866 | } else |
867 | cleanup_pending_aio(td); | |
2ba1c290 JA |
868 | |
869 | /* | |
870 | * stop job if we failed doing any IO | |
871 | */ | |
872 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
873 | td->done = 1; | |
ebac4655 JA |
874 | } |
875 | ||
ebac4655 JA |
876 | static void cleanup_io_u(struct thread_data *td) |
877 | { | |
01743ee1 | 878 | struct flist_head *entry, *n; |
ebac4655 JA |
879 | struct io_u *io_u; |
880 | ||
01743ee1 JA |
881 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
882 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 883 | |
01743ee1 | 884 | flist_del(&io_u->list); |
91aea6ec | 885 | fio_memfree(io_u, sizeof(*io_u)); |
ebac4655 JA |
886 | } |
887 | ||
2f9ade3c | 888 | free_io_mem(td); |
ebac4655 JA |
889 | } |
890 | ||
891 | static int init_io_u(struct thread_data *td) | |
892 | { | |
893 | struct io_u *io_u; | |
a00735e6 | 894 | unsigned int max_bs; |
eb7ccf38 | 895 | int cl_align, i, max_units; |
ebac4655 JA |
896 | char *p; |
897 | ||
1e3c09af | 898 | max_units = td->o.iodepth; |
2dc1bbeb | 899 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
900 | td->orig_buffer_size = (unsigned long long) max_bs |
901 | * (unsigned long long) max_units; | |
902 | ||
903 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
904 | unsigned long bs; | |
74b025b0 | 905 | |
5ec10eaa JA |
906 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
907 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
908 | } | |
ebac4655 | 909 | |
c3990645 JA |
910 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
911 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
912 | return 1; | |
913 | } | |
914 | ||
2f9ade3c JA |
915 | if (allocate_io_mem(td)) |
916 | return 1; | |
ebac4655 | 917 | |
1c04bf78 JA |
918 | if (td->o.odirect || td->o.mem_align || |
919 | (td->io_ops->flags & FIO_RAWIO)) | |
d529ee19 | 920 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
921 | else |
922 | p = td->orig_buffer; | |
923 | ||
eb7ccf38 JA |
924 | cl_align = os_cache_line_size(); |
925 | ||
ebac4655 | 926 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
927 | void *ptr; |
928 | ||
db1cd90b JA |
929 | if (td->terminate) |
930 | return 1; | |
eb7ccf38 | 931 | |
91aea6ec JA |
932 | ptr = fio_memalign(cl_align, sizeof(*io_u)); |
933 | if (!ptr) { | |
934 | log_err("fio: unable to allocate aligned memory\n"); | |
eb7ccf38 JA |
935 | break; |
936 | } | |
937 | ||
938 | io_u = ptr; | |
ebac4655 | 939 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 940 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 941 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 942 | |
b4c5e1ac | 943 | if (!(td->io_ops->flags & FIO_NOIO)) { |
cf00f975 | 944 | io_u->buf = p; |
d529ee19 | 945 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 946 | |
3545a109 | 947 | if (td_write(td)) |
5973cafb | 948 | io_u_fill_buffer(td, io_u, max_bs); |
3545a109 | 949 | if (td_write(td) && td->o.verify_pattern_bytes) { |
cbe8d756 RR |
950 | /* |
951 | * Fill the buffer with the pattern if we are | |
952 | * going to be doing writes. | |
953 | */ | |
7d9fb455 | 954 | fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
cbe8d756 | 955 | } |
b4c5e1ac | 956 | } |
6b9cea23 | 957 | |
b1ff3403 | 958 | io_u->index = i; |
0c6e7517 | 959 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 960 | flist_add(&io_u->list, &td->io_u_freelist); |
cf00f975 | 961 | p += max_bs; |
ebac4655 JA |
962 | } |
963 | ||
964 | return 0; | |
965 | } | |
966 | ||
da86774e JA |
967 | static int switch_ioscheduler(struct thread_data *td) |
968 | { | |
969 | char tmp[256], tmp2[128]; | |
970 | FILE *f; | |
971 | int ret; | |
972 | ||
ba0fbe10 | 973 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
974 | return 0; |
975 | ||
da86774e JA |
976 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
977 | ||
978 | f = fopen(tmp, "r+"); | |
979 | if (!f) { | |
cbf5c1d7 | 980 | if (errno == ENOENT) { |
5ec10eaa JA |
981 | log_err("fio: os or kernel doesn't support IO scheduler" |
982 | " switching\n"); | |
cbf5c1d7 JA |
983 | return 0; |
984 | } | |
985 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
986 | return 1; |
987 | } | |
988 | ||
989 | /* | |
990 | * Set io scheduler. | |
991 | */ | |
2dc1bbeb | 992 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 993 | if (ferror(f) || ret != 1) { |
e1161c32 | 994 | td_verror(td, errno, "fwrite"); |
da86774e JA |
995 | fclose(f); |
996 | return 1; | |
997 | } | |
998 | ||
999 | rewind(f); | |
1000 | ||
1001 | /* | |
1002 | * Read back and check that the selected scheduler is now the default. | |
1003 | */ | |
1004 | ret = fread(tmp, 1, sizeof(tmp), f); | |
1005 | if (ferror(f) || ret < 0) { | |
e1161c32 | 1006 | td_verror(td, errno, "fread"); |
da86774e JA |
1007 | fclose(f); |
1008 | return 1; | |
1009 | } | |
1010 | ||
2dc1bbeb | 1011 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 1012 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 1013 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 1014 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
1015 | fclose(f); |
1016 | return 1; | |
1017 | } | |
1018 | ||
1019 | fclose(f); | |
1020 | return 0; | |
1021 | } | |
1022 | ||
492158cf JA |
1023 | static int keep_running(struct thread_data *td) |
1024 | { | |
1025 | unsigned long long io_done; | |
1026 | ||
20e354ef JA |
1027 | if (td->done) |
1028 | return 0; | |
492158cf JA |
1029 | if (td->o.time_based) |
1030 | return 1; | |
1031 | if (td->o.loops) { | |
1032 | td->o.loops--; | |
1033 | return 1; | |
1034 | } | |
1035 | ||
5ec10eaa JA |
1036 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
1037 | + td->io_skip_bytes; | |
492158cf JA |
1038 | if (io_done < td->o.size) |
1039 | return 1; | |
1040 | ||
1041 | return 0; | |
1042 | } | |
1043 | ||
b29ee5b3 | 1044 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 1045 | { |
f0505a14 | 1046 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
ebac4655 | 1047 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
c8eeb9df JA |
1048 | td->stat_io_blocks[0] = td->stat_io_blocks[1] = 0; |
1049 | td->this_io_blocks[0] = td->this_io_blocks[1] = 0; | |
20dc95c4 | 1050 | td->zone_bytes = 0; |
581e7141 JA |
1051 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
1052 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 1053 | |
c1324df1 JA |
1054 | td->last_was_sync = 0; |
1055 | ||
2ba1c290 JA |
1056 | /* |
1057 | * reset file done count if we are to start over | |
1058 | */ | |
1059 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 1060 | td->nr_done_files = 0; |
b29ee5b3 JA |
1061 | } |
1062 | ||
1063 | void reset_all_stats(struct thread_data *td) | |
1064 | { | |
1065 | struct timeval tv; | |
1066 | int i; | |
1067 | ||
1068 | reset_io_counters(td); | |
1069 | ||
1070 | for (i = 0; i < 2; i++) { | |
1071 | td->io_bytes[i] = 0; | |
1072 | td->io_blocks[i] = 0; | |
1073 | td->io_issues[i] = 0; | |
1074 | td->ts.total_io_u[i] = 0; | |
1075 | } | |
9b836561 | 1076 | |
b29ee5b3 | 1077 | fio_gettime(&tv, NULL); |
cda99fa0 SSP |
1078 | td->ts.runtime[0] = 0; |
1079 | td->ts.runtime[1] = 0; | |
b29ee5b3 JA |
1080 | memcpy(&td->epoch, &tv, sizeof(tv)); |
1081 | memcpy(&td->start, &tv, sizeof(tv)); | |
1082 | } | |
1083 | ||
df9c26b1 | 1084 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
1085 | { |
1086 | struct fio_file *f; | |
1087 | unsigned int i; | |
b29ee5b3 JA |
1088 | |
1089 | reset_io_counters(td); | |
281f9b63 | 1090 | |
24ffd2c2 | 1091 | close_files(td); |
df9c26b1 | 1092 | for_each_file(td, f, i) |
d6aed795 | 1093 | fio_file_clear_done(f); |
f0153948 JA |
1094 | |
1095 | /* | |
1096 | * Set the same seed to get repeatable runs | |
1097 | */ | |
1098 | td_fill_rand_seeds(td); | |
ebac4655 JA |
1099 | } |
1100 | ||
398e8c4d JA |
1101 | static int exec_string(const char *string) |
1102 | { | |
1103 | int ret, newlen = strlen(string) + 1 + 8; | |
1104 | char *str; | |
1105 | ||
1106 | str = malloc(newlen); | |
1107 | sprintf(str, "sh -c %s", string); | |
1108 | ||
1109 | ret = system(str); | |
1110 | if (ret == -1) | |
1111 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1112 | ||
1113 | free(str); | |
1114 | return ret; | |
1115 | } | |
1116 | ||
906c8d75 JA |
1117 | /* |
1118 | * Entry point for the thread based jobs. The process based jobs end up | |
1119 | * here as well, after a little setup. | |
1120 | */ | |
ebac4655 JA |
1121 | static void *thread_main(void *data) |
1122 | { | |
cda99fa0 | 1123 | unsigned long long elapsed; |
ebac4655 | 1124 | struct thread_data *td = data; |
e8462bd8 | 1125 | pthread_condattr_t attr; |
a978ba68 | 1126 | int clear_state; |
ebac4655 | 1127 | |
47f767c1 | 1128 | if (!td->o.use_thread) { |
ebac4655 | 1129 | setsid(); |
47f767c1 JA |
1130 | td->pid = getpid(); |
1131 | } else | |
1132 | td->pid = gettid(); | |
ebac4655 | 1133 | |
5921e80c | 1134 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1135 | |
01743ee1 JA |
1136 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1137 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1138 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1139 | INIT_FLIST_HEAD(&td->io_log_list); | |
1140 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 | 1141 | INIT_FLIST_HEAD(&td->verify_list); |
0d29de83 | 1142 | INIT_FLIST_HEAD(&td->trim_list); |
e8462bd8 | 1143 | pthread_mutex_init(&td->io_u_lock, NULL); |
bb5d7d0b | 1144 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1145 | |
e8462bd8 JA |
1146 | pthread_condattr_init(&attr); |
1147 | pthread_cond_init(&td->verify_cond, &attr); | |
1148 | pthread_cond_init(&td->free_cond, &attr); | |
1149 | ||
db1cd90b | 1150 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1151 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1152 | fio_mutex_up(startup_mutex); |
29adda3c | 1153 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1154 | fio_mutex_down(td->mutex); |
29adda3c | 1155 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1156 | |
1157 | /* | |
cdd18ad8 | 1158 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1159 | * eating a file descriptor |
1160 | */ | |
cdd18ad8 | 1161 | fio_mutex_remove(td->mutex); |
db1cd90b | 1162 | |
dfc6e751 SH |
1163 | /* |
1164 | * A new gid requires privilege, so we need to do this before setting | |
1165 | * the uid. | |
1166 | */ | |
e0b0d892 JA |
1167 | if (td->o.gid != -1U && setgid(td->o.gid)) { |
1168 | td_verror(td, errno, "setgid"); | |
1169 | goto err; | |
1170 | } | |
dfc6e751 SH |
1171 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1172 | td_verror(td, errno, "setuid"); | |
1173 | goto err; | |
1174 | } | |
e0b0d892 | 1175 | |
be18a94f JA |
1176 | /* |
1177 | * If we have a gettimeofday() thread, make sure we exclude that | |
1178 | * thread from this job | |
1179 | */ | |
1180 | if (td->o.gtod_cpu) | |
1181 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); | |
1182 | ||
1183 | /* | |
1184 | * Set affinity first, in case it has an impact on the memory | |
1185 | * allocations. | |
1186 | */ | |
1187 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
1188 | td_verror(td, errno, "cpu_set_affinity"); | |
1189 | goto err; | |
1190 | } | |
1191 | ||
d84f8d49 JA |
1192 | /* |
1193 | * May alter parameters that init_io_u() will use, so we need to | |
1194 | * do this first. | |
1195 | */ | |
1196 | if (init_iolog(td)) | |
1197 | goto err; | |
1198 | ||
ebac4655 | 1199 | if (init_io_u(td)) |
db1cd90b | 1200 | goto err; |
ebac4655 | 1201 | |
e8462bd8 JA |
1202 | if (td->o.verify_async && verify_async_init(td)) |
1203 | goto err; | |
1204 | ||
ac684785 | 1205 | if (td->ioprio_set) { |
ebac4655 | 1206 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1207 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1208 | goto err; |
ebac4655 JA |
1209 | } |
1210 | } | |
1211 | ||
6adb38a1 | 1212 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1213 | goto err; |
1214 | ||
2dc1bbeb | 1215 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1216 | td_verror(td, errno, "nice"); |
db1cd90b | 1217 | goto err; |
b6f4d880 JA |
1218 | } |
1219 | ||
2dc1bbeb | 1220 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1221 | goto err; |
37f56873 | 1222 | |
2dc1bbeb | 1223 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1224 | goto err; |
1225 | ||
7d6c5283 JA |
1226 | if (td_io_init(td)) |
1227 | goto err; | |
1228 | ||
68727076 JA |
1229 | if (init_random_map(td)) |
1230 | goto err; | |
1231 | ||
2dc1bbeb | 1232 | if (td->o.exec_prerun) { |
398e8c4d | 1233 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1234 | goto err; |
1235 | } | |
4e0ba8af | 1236 | |
afad68f7 ZY |
1237 | if (td->o.pre_read) { |
1238 | if (pre_read_files(td) < 0) | |
1239 | goto err; | |
1240 | } | |
1241 | ||
69008999 | 1242 | fio_gettime(&td->epoch, NULL); |
c8aaba19 | 1243 | getrusage(RUSAGE_SELF, &td->ru_start); |
69008999 | 1244 | |
a978ba68 | 1245 | clear_state = 0; |
492158cf | 1246 | while (keep_running(td)) { |
02bcaa8c | 1247 | fio_gettime(&td->start, NULL); |
c8eeb9df JA |
1248 | memcpy(&td->bw_sample_time, &td->start, sizeof(td->start)); |
1249 | memcpy(&td->iops_sample_time, &td->start, sizeof(td->start)); | |
993bf48b | 1250 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1251 | |
8a322057 JC |
1252 | if (td->o.ratemin[0] || td->o.ratemin[1]) { |
1253 | memcpy(&td->lastrate[0], &td->bw_sample_time, | |
1254 | sizeof(td->bw_sample_time)); | |
1255 | memcpy(&td->lastrate[1], &td->bw_sample_time, | |
1256 | sizeof(td->bw_sample_time)); | |
1257 | } | |
ebac4655 | 1258 | |
df9c26b1 JA |
1259 | if (clear_state) |
1260 | clear_io_state(td); | |
a978ba68 | 1261 | |
ebac4655 JA |
1262 | prune_io_piece_log(td); |
1263 | ||
ba0fbe10 | 1264 | do_io(td); |
ebac4655 | 1265 | |
a978ba68 JA |
1266 | clear_state = 1; |
1267 | ||
38d77cae | 1268 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1269 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1270 | td->ts.runtime[DDIR_READ] += elapsed; |
38d77cae JA |
1271 | } |
1272 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1273 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1274 | td->ts.runtime[DDIR_WRITE] += elapsed; |
38d77cae | 1275 | } |
5ec10eaa | 1276 | |
ebac4655 JA |
1277 | if (td->error || td->terminate) |
1278 | break; | |
1279 | ||
e84c73a8 SL |
1280 | if (!td->o.do_verify || |
1281 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1282 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1283 | continue; |
1284 | ||
df9c26b1 | 1285 | clear_io_state(td); |
a978ba68 | 1286 | |
02bcaa8c | 1287 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1288 | |
1289 | do_verify(td); | |
1290 | ||
cda99fa0 | 1291 | td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1292 | |
1293 | if (td->error || td->terminate) | |
1294 | break; | |
1295 | } | |
1296 | ||
36dff966 | 1297 | update_rusage_stat(td); |
cda99fa0 SSP |
1298 | td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000; |
1299 | td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000; | |
756867bd JA |
1300 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1301 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1302 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1303 | ||
d2bbb8d9 | 1304 | fio_mutex_down(writeout_mutex); |
7b9f733a | 1305 | if (td->bw_log) { |
f484470b | 1306 | if (td->o.bw_log_file) { |
7b9f733a | 1307 | finish_log_named(td, td->bw_log, |
f484470b JA |
1308 | td->o.bw_log_file, "bw"); |
1309 | } else | |
7b9f733a | 1310 | finish_log(td, td->bw_log, "bw"); |
e3cedca7 | 1311 | } |
7b9f733a | 1312 | if (td->lat_log) { |
02af0988 | 1313 | if (td->o.lat_log_file) { |
7b9f733a | 1314 | finish_log_named(td, td->lat_log, |
02af0988 JA |
1315 | td->o.lat_log_file, "lat"); |
1316 | } else | |
7b9f733a | 1317 | finish_log(td, td->lat_log, "lat"); |
02af0988 | 1318 | } |
7b9f733a | 1319 | if (td->slat_log) { |
f484470b | 1320 | if (td->o.lat_log_file) { |
7b9f733a | 1321 | finish_log_named(td, td->slat_log, |
1ca95738 | 1322 | td->o.lat_log_file, "slat"); |
f484470b | 1323 | } else |
7b9f733a | 1324 | finish_log(td, td->slat_log, "slat"); |
e3cedca7 | 1325 | } |
7b9f733a | 1326 | if (td->clat_log) { |
f484470b | 1327 | if (td->o.lat_log_file) { |
7b9f733a | 1328 | finish_log_named(td, td->clat_log, |
f484470b JA |
1329 | td->o.lat_log_file, "clat"); |
1330 | } else | |
7b9f733a | 1331 | finish_log(td, td->clat_log, "clat"); |
e3cedca7 | 1332 | } |
c8eeb9df JA |
1333 | if (td->iops_log) { |
1334 | if (td->o.iops_log_file) { | |
1335 | finish_log_named(td, td->iops_log, | |
1336 | td->o.iops_log_file, "iops"); | |
1337 | } else | |
1338 | finish_log(td, td->iops_log, "iops"); | |
1339 | } | |
1340 | ||
d2bbb8d9 | 1341 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1342 | if (td->o.exec_postrun) |
1343 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1344 | |
1345 | if (exitall_on_terminate) | |
cc0df00a | 1346 | fio_terminate_threads(td->groupid); |
ebac4655 JA |
1347 | |
1348 | err: | |
5bf13a5a | 1349 | if (td->error) |
4ceb30d4 | 1350 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1351 | td->verror); |
b4707ed7 JA |
1352 | |
1353 | if (td->o.verify_async) | |
1354 | verify_async_exit(td); | |
1355 | ||
24ffd2c2 | 1356 | close_and_free_files(td); |
2866c82d | 1357 | close_ioengine(td); |
ebac4655 | 1358 | cleanup_io_u(td); |
6adb38a1 | 1359 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1360 | |
d2ce18b5 JA |
1361 | if (td->o.cpumask_set) { |
1362 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1363 | ||
1364 | td_verror(td, ret, "fio_cpuset_exit"); | |
1365 | } | |
4e78e405 | 1366 | |
f29b25a3 JA |
1367 | /* |
1368 | * do this very late, it will log file closing as well | |
1369 | */ | |
1370 | if (td->o.write_iolog_file) | |
1371 | write_iolog_close(td); | |
1372 | ||
ebac4655 | 1373 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1374 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1375 | } |
1376 | ||
906c8d75 JA |
1377 | /* |
1378 | * We cannot pass the td data into a forked process, so attach the td and | |
1379 | * pass it to the thread worker. | |
1380 | */ | |
a6418147 | 1381 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1382 | { |
1383 | struct thread_data *td; | |
a6418147 | 1384 | void *data, *ret; |
ebac4655 | 1385 | |
65258961 | 1386 | #ifndef __hpux |
ebac4655 JA |
1387 | data = shmat(shmid, NULL, 0); |
1388 | if (data == (void *) -1) { | |
a6418147 JA |
1389 | int __err = errno; |
1390 | ||
ebac4655 | 1391 | perror("shmat"); |
a6418147 | 1392 | return __err; |
ebac4655 | 1393 | } |
65258961 JA |
1394 | #else |
1395 | /* | |
1396 | * HP-UX inherits shm mappings? | |
1397 | */ | |
1398 | data = threads; | |
1399 | #endif | |
ebac4655 JA |
1400 | |
1401 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1402 | ret = thread_main(td); |
ebac4655 | 1403 | shmdt(data); |
43d76807 | 1404 | return (int) (unsigned long) ret; |
ebac4655 JA |
1405 | } |
1406 | ||
906c8d75 JA |
1407 | /* |
1408 | * Run over the job map and reap the threads that have exited, if any. | |
1409 | */ | |
11e950bd JA |
1410 | static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, |
1411 | unsigned int *m_rate) | |
ebac4655 | 1412 | { |
34572e28 | 1413 | struct thread_data *td; |
11e950bd JA |
1414 | unsigned int cputhreads, realthreads, pending; |
1415 | int i, status, ret; | |
ebac4655 JA |
1416 | |
1417 | /* | |
1418 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1419 | */ | |
1f809d15 | 1420 | realthreads = pending = cputhreads = 0; |
34572e28 | 1421 | for_each_td(td, i) { |
3707f45b | 1422 | int flags = 0; |
a2f77c9f | 1423 | |
84585003 JA |
1424 | /* |
1425 | * ->io_ops is NULL for a thread that has closed its | |
1426 | * io engine | |
1427 | */ | |
ba0fbe10 | 1428 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1429 | cputhreads++; |
1f809d15 JA |
1430 | else |
1431 | realthreads++; | |
b990b5c0 | 1432 | |
a1dedc1f AC |
1433 | if (!td->pid) { |
1434 | pending++; | |
1435 | continue; | |
1436 | } | |
1437 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1438 | continue; |
2dc1bbeb | 1439 | if (td->o.use_thread) { |
3707f45b JA |
1440 | if (td->runstate == TD_EXITED) { |
1441 | td_set_runstate(td, TD_REAPED); | |
1442 | goto reaped; | |
1443 | } | |
1444 | continue; | |
1445 | } | |
a2f77c9f JA |
1446 | |
1447 | flags = WNOHANG; | |
1448 | if (td->runstate == TD_EXITED) | |
1449 | flags = 0; | |
1450 | ||
1451 | /* | |
1452 | * check if someone quit or got killed in an unusual way | |
1453 | */ | |
1454 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1455 | if (ret < 0) { |
a2f77c9f | 1456 | if (errno == ECHILD) { |
5921e80c JA |
1457 | log_err("fio: pid=%d disappeared %d\n", |
1458 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1459 | td_set_runstate(td, TD_REAPED); |
1460 | goto reaped; | |
1461 | } | |
1462 | perror("waitpid"); | |
1463 | } else if (ret == td->pid) { | |
1464 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1465 | int sig = WTERMSIG(status); |
1466 | ||
03e20d68 | 1467 | if (sig != SIGTERM) |
5ec10eaa | 1468 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1469 | (int) td->pid, sig); |
fab6aa71 JA |
1470 | td_set_runstate(td, TD_REAPED); |
1471 | goto reaped; | |
1472 | } | |
a2f77c9f JA |
1473 | if (WIFEXITED(status)) { |
1474 | if (WEXITSTATUS(status) && !td->error) | |
1475 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1476 | |
a2f77c9f JA |
1477 | td_set_runstate(td, TD_REAPED); |
1478 | goto reaped; | |
a6418147 JA |
1479 | } |
1480 | } | |
ebac4655 | 1481 | |
a2f77c9f JA |
1482 | /* |
1483 | * thread is not dead, continue | |
1484 | */ | |
e48676ba | 1485 | pending++; |
a2f77c9f | 1486 | continue; |
fab6aa71 | 1487 | reaped: |
ebac4655 | 1488 | (*nr_running)--; |
581e7141 JA |
1489 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1490 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1491 | if (!td->pid) |
1492 | pending--; | |
a2f77c9f JA |
1493 | |
1494 | if (td->error) | |
1495 | exit_value++; | |
d3eeeabc JA |
1496 | |
1497 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1498 | } |
b990b5c0 | 1499 | |
1f809d15 | 1500 | if (*nr_running == cputhreads && !pending && realthreads) |
cc0df00a | 1501 | fio_terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1502 | } |
1503 | ||
906c8d75 JA |
1504 | /* |
1505 | * Main function for kicking off and reaping jobs, as needed. | |
1506 | */ | |
ebac4655 JA |
1507 | static void run_threads(void) |
1508 | { | |
ebac4655 JA |
1509 | struct thread_data *td; |
1510 | unsigned long spent; | |
11e950bd | 1511 | unsigned int i, todo, nr_running, m_rate, t_rate, nr_started; |
fcb6ade2 | 1512 | |
2f9ade3c JA |
1513 | if (fio_pin_memory()) |
1514 | return; | |
ebac4655 | 1515 | |
be4ecfdf JA |
1516 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1517 | return; | |
1518 | ||
cc0df00a JA |
1519 | set_sig_handlers(); |
1520 | ||
c6ae0a5b | 1521 | if (!terse_output) { |
4ceb30d4 | 1522 | log_info("Starting "); |
9cedf167 | 1523 | if (nr_thread) |
4ceb30d4 | 1524 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1525 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1526 | if (nr_process) { |
1527 | if (nr_thread) | |
cc604381 | 1528 | log_info(" and "); |
4ceb30d4 | 1529 | log_info("%d process%s", nr_process, |
5ec10eaa | 1530 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1531 | } |
4ceb30d4 | 1532 | log_info("\n"); |
c6ae0a5b JA |
1533 | fflush(stdout); |
1534 | } | |
c04f7ec3 | 1535 | |
ebac4655 JA |
1536 | todo = thread_number; |
1537 | nr_running = 0; | |
1538 | nr_started = 0; | |
1539 | m_rate = t_rate = 0; | |
1540 | ||
34572e28 | 1541 | for_each_td(td, i) { |
263e529f | 1542 | print_status_init(td->thread_number - 1); |
ebac4655 | 1543 | |
067430bb | 1544 | if (!td->o.create_serialize) |
ebac4655 JA |
1545 | continue; |
1546 | ||
1547 | /* | |
1548 | * do file setup here so it happens sequentially, | |
1549 | * we don't want X number of threads getting their | |
1550 | * client data interspersed on disk | |
1551 | */ | |
53cdc686 | 1552 | if (setup_files(td)) { |
5bf13a5a JA |
1553 | exit_value++; |
1554 | if (td->error) | |
5921e80c JA |
1555 | log_err("fio: pid=%d, err=%d/%s\n", |
1556 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1557 | td_set_runstate(td, TD_REAPED); |
1558 | todo--; | |
c69e8875 JA |
1559 | } else { |
1560 | struct fio_file *f; | |
2b13e716 | 1561 | unsigned int j; |
c69e8875 JA |
1562 | |
1563 | /* | |
1564 | * for sharing to work, each job must always open | |
1565 | * its own files. so close them, if we opened them | |
1566 | * for creation | |
1567 | */ | |
2b13e716 | 1568 | for_each_file(td, f, j) { |
22a57ba8 JA |
1569 | if (fio_file_open(f)) |
1570 | td_io_close_file(td, f); | |
22a57ba8 | 1571 | } |
5ec10eaa | 1572 | } |
ebac4655 JA |
1573 | } |
1574 | ||
a2f77c9f JA |
1575 | set_genesis_time(); |
1576 | ||
ebac4655 | 1577 | while (todo) { |
fca70358 | 1578 | struct thread_data *map[REAL_MAX_JOBS]; |
75154845 JA |
1579 | struct timeval this_start; |
1580 | int this_jobs = 0, left; | |
1581 | ||
ebac4655 JA |
1582 | /* |
1583 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1584 | */ | |
34572e28 | 1585 | for_each_td(td, i) { |
ebac4655 JA |
1586 | if (td->runstate != TD_NOT_CREATED) |
1587 | continue; | |
1588 | ||
1589 | /* | |
1590 | * never got a chance to start, killed by other | |
1591 | * thread for some reason | |
1592 | */ | |
1593 | if (td->terminate) { | |
1594 | todo--; | |
1595 | continue; | |
1596 | } | |
1597 | ||
2dc1bbeb | 1598 | if (td->o.start_delay) { |
263e529f | 1599 | spent = mtime_since_genesis(); |
ebac4655 | 1600 | |
2dc1bbeb | 1601 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1602 | continue; |
1603 | } | |
1604 | ||
6f88bcb0 JA |
1605 | if (td->o.stonewall && (nr_started || nr_running)) { |
1606 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1607 | td->o.name); | |
ebac4655 | 1608 | break; |
6f88bcb0 | 1609 | } |
ebac4655 | 1610 | |
067430bb JA |
1611 | init_disk_util(td); |
1612 | ||
75154845 JA |
1613 | /* |
1614 | * Set state to created. Thread will transition | |
1615 | * to TD_INITIALIZED when it's done setting up. | |
1616 | */ | |
ebac4655 | 1617 | td_set_runstate(td, TD_CREATED); |
75154845 | 1618 | map[this_jobs++] = td; |
ebac4655 JA |
1619 | nr_started++; |
1620 | ||
2dc1bbeb | 1621 | if (td->o.use_thread) { |
f42c153d JA |
1622 | int ret; |
1623 | ||
ee56ad50 | 1624 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1625 | ret = pthread_create(&td->thread, NULL, |
1626 | thread_main, td); | |
1627 | if (ret) { | |
1628 | log_err("pthread_create: %s\n", | |
1629 | strerror(ret)); | |
ebac4655 | 1630 | nr_started--; |
c8bb6faf | 1631 | break; |
ebac4655 | 1632 | } |
f42c153d JA |
1633 | ret = pthread_detach(td->thread); |
1634 | if (ret) | |
1635 | log_err("pthread_detach: %s", | |
1636 | strerror(ret)); | |
ebac4655 | 1637 | } else { |
5e1d306e | 1638 | pid_t pid; |
ee56ad50 | 1639 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1640 | pid = fork(); |
1641 | if (!pid) { | |
a6418147 JA |
1642 | int ret = fork_main(shm_id, i); |
1643 | ||
5e1d306e JA |
1644 | _exit(ret); |
1645 | } else if (i == fio_debug_jobno) | |
1646 | *fio_debug_jobp = pid; | |
ebac4655 | 1647 | } |
29adda3c | 1648 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1649 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1650 | log_err("fio: job startup hung? exiting.\n"); | |
cc0df00a | 1651 | fio_terminate_threads(TERMINATE_ALL); |
656b1393 JA |
1652 | fio_abort = 1; |
1653 | nr_started--; | |
1654 | break; | |
1655 | } | |
29adda3c | 1656 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1657 | } |
1658 | ||
1659 | /* | |
75154845 JA |
1660 | * Wait for the started threads to transition to |
1661 | * TD_INITIALIZED. | |
ebac4655 | 1662 | */ |
02bcaa8c | 1663 | fio_gettime(&this_start, NULL); |
75154845 | 1664 | left = this_jobs; |
6ce15a32 | 1665 | while (left && !fio_abort) { |
75154845 JA |
1666 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1667 | break; | |
1668 | ||
1669 | usleep(100000); | |
1670 | ||
1671 | for (i = 0; i < this_jobs; i++) { | |
1672 | td = map[i]; | |
1673 | if (!td) | |
1674 | continue; | |
b6f4d880 | 1675 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1676 | map[i] = NULL; |
1677 | left--; | |
b6f4d880 JA |
1678 | } else if (td->runstate >= TD_EXITED) { |
1679 | map[i] = NULL; | |
1680 | left--; | |
1681 | todo--; | |
1682 | nr_running++; /* work-around... */ | |
75154845 JA |
1683 | } |
1684 | } | |
1685 | } | |
1686 | ||
1687 | if (left) { | |
3b70d7e5 | 1688 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1689 | for (i = 0; i < this_jobs; i++) { |
1690 | td = map[i]; | |
1691 | if (!td) | |
1692 | continue; | |
1693 | kill(td->pid, SIGTERM); | |
1694 | } | |
1695 | break; | |
1696 | } | |
1697 | ||
1698 | /* | |
b6f4d880 | 1699 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1700 | */ |
34572e28 | 1701 | for_each_td(td, i) { |
75154845 | 1702 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1703 | continue; |
1704 | ||
b29ee5b3 JA |
1705 | if (in_ramp_time(td)) |
1706 | td_set_runstate(td, TD_RAMP); | |
1707 | else | |
1708 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1709 | nr_running++; |
1710 | nr_started--; | |
581e7141 JA |
1711 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1712 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1713 | todo--; |
cdd18ad8 | 1714 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1715 | } |
1716 | ||
1717 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1718 | ||
602d6ff9 JA |
1719 | if (todo) { |
1720 | if (is_backend) | |
1721 | fio_server_idle_loop(); | |
1722 | else | |
1723 | usleep(100000); | |
1724 | } | |
ebac4655 JA |
1725 | } |
1726 | ||
1727 | while (nr_running) { | |
1728 | reap_threads(&nr_running, &t_rate, &m_rate); | |
cc0df00a JA |
1729 | |
1730 | if (is_backend) | |
1731 | fio_server_idle_loop(); | |
70e0c316 JA |
1732 | else |
1733 | usleep(10000); | |
ebac4655 JA |
1734 | } |
1735 | ||
1736 | update_io_ticks(); | |
2f9ade3c | 1737 | fio_unpin_memory(); |
ebac4655 JA |
1738 | } |
1739 | ||
50d16976 | 1740 | int exec_run(void) |
ebac4655 | 1741 | { |
7e356b2d JA |
1742 | struct thread_data *td; |
1743 | int i; | |
1744 | ||
a37f69b7 | 1745 | if (nr_clients) |
37db14fe | 1746 | return fio_handle_clients(); |
7e356b2d JA |
1747 | if (exec_profile) { |
1748 | if (load_profile(exec_profile)) | |
1749 | return 1; | |
1750 | free(exec_profile); | |
1751 | exec_profile = NULL; | |
1752 | } | |
74929ac2 JA |
1753 | if (!thread_number) |
1754 | return 0; | |
1755 | ||
bb3884d8 | 1756 | if (write_bw_log) { |
b8bc8cba JA |
1757 | setup_log(&agg_io_log[DDIR_READ], 0); |
1758 | setup_log(&agg_io_log[DDIR_WRITE], 0); | |
bb3884d8 JA |
1759 | } |
1760 | ||
cdd18ad8 | 1761 | startup_mutex = fio_mutex_init(0); |
03e20d68 BC |
1762 | if (startup_mutex == NULL) |
1763 | return 1; | |
d2bbb8d9 | 1764 | writeout_mutex = fio_mutex_init(1); |
03e20d68 BC |
1765 | if (writeout_mutex == NULL) |
1766 | return 1; | |
07739b57 | 1767 | |
a2f77c9f | 1768 | set_genesis_time(); |
7985701b | 1769 | create_disk_util_thread(); |
ebac4655 | 1770 | |
dae5341c JA |
1771 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1772 | INIT_FLIST_HEAD(cgroup_list); | |
1773 | ||
ebac4655 | 1774 | run_threads(); |
6ce15a32 | 1775 | |
bb3884d8 | 1776 | if (!fio_abort) { |
6ce15a32 | 1777 | show_run_stats(); |
bb3884d8 | 1778 | if (write_bw_log) { |
5ec10eaa JA |
1779 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1780 | __finish_log(agg_io_log[DDIR_WRITE], | |
1781 | "agg-write_bw.log"); | |
bb3884d8 JA |
1782 | } |
1783 | } | |
ebac4655 | 1784 | |
7e356b2d JA |
1785 | for_each_td(td, i) |
1786 | fio_options_free(td); | |
1787 | ||
dae5341c JA |
1788 | cgroup_kill(cgroup_list); |
1789 | sfree(cgroup_list); | |
61eb313e | 1790 | sfree(cgroup_mnt); |
39f22027 | 1791 | |
cdd18ad8 | 1792 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1793 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1794 | return exit_value; |
ebac4655 | 1795 | } |
50d16976 JA |
1796 | |
1797 | void reset_fio_state(void) | |
1798 | { | |
1799 | groupid = 0; | |
1800 | thread_number = 0; | |
1801 | nr_process = 0; | |
1802 | nr_thread = 0; | |
1803 | done_secs = 0; | |
1804 | } | |
1805 | ||
0ad5edcc JA |
1806 | static int endian_check(void) |
1807 | { | |
1808 | union { | |
1809 | uint8_t c[8]; | |
1810 | uint64_t v; | |
1811 | } u; | |
1812 | int le = 0, be = 0; | |
1813 | ||
1814 | u.v = 0x12; | |
1815 | if (u.c[7] == 0x12) | |
1816 | be = 1; | |
1817 | else if (u.c[0] == 0x12) | |
1818 | le = 1; | |
1819 | ||
1820 | #if defined(FIO_LITTLE_ENDIAN) | |
1821 | if (be) | |
1822 | return 1; | |
1823 | #elif defined(FIO_BIG_ENDIAN) | |
1824 | if (le) | |
1825 | return 1; | |
1826 | #else | |
1827 | return 1; | |
1828 | #endif | |
1829 | ||
1830 | if (!le && !be) | |
1831 | return 1; | |
1832 | ||
1833 | return 0; | |
1834 | } | |
1835 | ||
50d16976 JA |
1836 | int main(int argc, char *argv[], char *envp[]) |
1837 | { | |
1838 | long ps; | |
1839 | ||
0ad5edcc JA |
1840 | if (endian_check()) { |
1841 | log_err("fio: endianness settings appear wrong.\n"); | |
1842 | log_err("fio: please report this to fio@vger.kernel.org\n"); | |
1843 | return 1; | |
1844 | } | |
1845 | ||
50d16976 JA |
1846 | arch_init(envp); |
1847 | ||
1848 | sinit(); | |
1849 | ||
1850 | /* | |
1851 | * We need locale for number printing, if it isn't set then just | |
1852 | * go with the US format. | |
1853 | */ | |
1854 | if (!getenv("LC_NUMERIC")) | |
1855 | setlocale(LC_NUMERIC, "en_US"); | |
1856 | ||
1857 | ps = sysconf(_SC_PAGESIZE); | |
1858 | if (ps < 0) { | |
1859 | log_err("Failed to get page size\n"); | |
1860 | return 1; | |
1861 | } | |
1862 | ||
1863 | page_size = ps; | |
1864 | page_mask = ps - 1; | |
1865 | ||
1866 | fio_keywords_init(); | |
1867 | ||
1868 | if (parse_options(argc, argv)) | |
1869 | return 1; | |
1870 | ||
1871 | return exec_run(); | |
1872 | } |