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; |
be4ecfdf | 76 | static pthread_t gtod_thread; |
7985701b | 77 | static pthread_t disk_util_thread; |
dae5341c | 78 | static struct flist_head *cgroup_list; |
6adb38a1 | 79 | static char *cgroup_mnt; |
ebac4655 | 80 | |
d3cc4ebf JA |
81 | unsigned long arch_flags = 0; |
82 | ||
bb3884d8 JA |
83 | struct io_log *agg_io_log[2]; |
84 | ||
75154845 | 85 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 | 86 | |
cca84643 JA |
87 | static const char *fio_os_strings[os_nr] = { |
88 | "Invalid", | |
89 | "Linux", | |
90 | "AIX", | |
91 | "FreeBSD", | |
92 | "HP-UX", | |
93 | "OSX", | |
94 | "NetBSD", | |
95 | "Solaris", | |
96 | "Windows" | |
97 | }; | |
98 | ||
99 | static const char *fio_arch_strings[arch_nr] = { | |
100 | "Invalid", | |
101 | "x86-64", | |
102 | "x86", | |
103 | "ppc", | |
104 | "ia64", | |
105 | "s390", | |
106 | "alpha", | |
107 | "sparc", | |
108 | "sparc64", | |
109 | "arm", | |
110 | "sh", | |
111 | "hppa", | |
112 | "generic" | |
113 | }; | |
114 | ||
115 | const char *fio_get_os_string(int nr) | |
116 | { | |
117 | if (nr < os_nr) | |
118 | return fio_os_strings[nr]; | |
119 | ||
120 | return NULL; | |
121 | } | |
122 | ||
123 | const char *fio_get_arch_string(int nr) | |
124 | { | |
125 | if (nr < arch_nr) | |
126 | return fio_arch_strings[nr]; | |
127 | ||
128 | return NULL; | |
129 | } | |
130 | ||
b29ee5b3 | 131 | void td_set_runstate(struct thread_data *td, int runstate) |
6ce15a32 | 132 | { |
d8fab1b6 JA |
133 | if (td->runstate == runstate) |
134 | return; | |
135 | ||
5921e80c JA |
136 | dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid, |
137 | td->runstate, runstate); | |
6ce15a32 JA |
138 | td->runstate = runstate; |
139 | } | |
140 | ||
cc0df00a | 141 | void fio_terminate_threads(int group_id) |
ebac4655 | 142 | { |
34572e28 | 143 | struct thread_data *td; |
ebac4655 JA |
144 | int i; |
145 | ||
6f88bcb0 JA |
146 | dprint(FD_PROCESS, "terminate group_id=%d\n", group_id); |
147 | ||
34572e28 | 148 | for_each_td(td, i) { |
ebac4655 | 149 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { |
6f88bcb0 | 150 | dprint(FD_PROCESS, "setting terminate on %s/%d\n", |
5921e80c | 151 | td->o.name, (int) td->pid); |
ad830ed7 JA |
152 | td->terminate = 1; |
153 | td->o.start_delay = 0; | |
154 | ||
7a93ab19 JA |
155 | /* |
156 | * if the thread is running, just let it exit | |
157 | */ | |
6aa27816 MC |
158 | if (!td->pid) |
159 | continue; | |
160 | else if (td->runstate < TD_RAMP) | |
03e20d68 | 161 | kill(td->pid, SIGTERM); |
f63f7f3b JA |
162 | else { |
163 | struct ioengine_ops *ops = td->io_ops; | |
164 | ||
03e20d68 BC |
165 | if (ops && (ops->flags & FIO_SIGTERM)) |
166 | kill(td->pid, SIGTERM); | |
f63f7f3b | 167 | } |
ebac4655 JA |
168 | } |
169 | } | |
170 | } | |
171 | ||
697a606c JA |
172 | static void sig_int(int sig) |
173 | { | |
174 | if (threads) { | |
7b821684 JA |
175 | if (is_backend) |
176 | fio_server_got_signal(sig); | |
177 | else { | |
178 | log_info("\nfio: terminating on signal %d\n", sig); | |
179 | fflush(stdout); | |
180 | exit_value = 128; | |
181 | } | |
182 | ||
cc0df00a | 183 | fio_terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
184 | } |
185 | } | |
186 | ||
7985701b | 187 | static void *disk_thread_main(void *data) |
a5b01f1b | 188 | { |
7985701b JA |
189 | fio_mutex_up(startup_mutex); |
190 | ||
191 | while (threads) { | |
192 | usleep(DISK_UTIL_MSEC * 1000); | |
9a793c2f JA |
193 | if (!threads) |
194 | break; | |
7985701b | 195 | update_io_ticks(); |
cf451d1e | 196 | |
af9c9fb3 | 197 | if (!is_backend) |
cf451d1e | 198 | print_thread_status(); |
7985701b JA |
199 | } |
200 | ||
201 | return NULL; | |
a5b01f1b JA |
202 | } |
203 | ||
7985701b | 204 | static int create_disk_util_thread(void) |
a5b01f1b | 205 | { |
7985701b | 206 | int ret; |
a5b01f1b | 207 | |
69204d6e | 208 | ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL); |
7985701b JA |
209 | if (ret) { |
210 | log_err("Can't create disk util thread: %s\n", strerror(ret)); | |
211 | return 1; | |
212 | } | |
a5b01f1b | 213 | |
7985701b JA |
214 | ret = pthread_detach(disk_util_thread); |
215 | if (ret) { | |
216 | log_err("Can't detatch disk util thread: %s\n", strerror(ret)); | |
217 | return 1; | |
218 | } | |
03e20d68 | 219 | |
7985701b JA |
220 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
221 | fio_mutex_down(startup_mutex); | |
222 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); | |
223 | return 0; | |
a5b01f1b JA |
224 | } |
225 | ||
697a606c JA |
226 | static void set_sig_handlers(void) |
227 | { | |
228 | struct sigaction act; | |
229 | ||
697a606c JA |
230 | memset(&act, 0, sizeof(act)); |
231 | act.sa_handler = sig_int; | |
232 | act.sa_flags = SA_RESTART; | |
233 | sigaction(SIGINT, &act, NULL); | |
de79c915 | 234 | |
5f8f9726 | 235 | memset(&act, 0, sizeof(act)); |
b20b6be4 | 236 | act.sa_handler = sig_int; |
5f8f9726 | 237 | act.sa_flags = SA_RESTART; |
03e20d68 | 238 | sigaction(SIGTERM, &act, NULL); |
cc0df00a JA |
239 | |
240 | if (is_backend) { | |
241 | memset(&act, 0, sizeof(act)); | |
242 | act.sa_handler = sig_int; | |
243 | act.sa_flags = SA_RESTART; | |
244 | sigaction(SIGPIPE, &act, NULL); | |
245 | } | |
697a606c JA |
246 | } |
247 | ||
906c8d75 JA |
248 | /* |
249 | * Check if we are above the minimum rate given. | |
250 | */ | |
581e7141 | 251 | static int __check_min_rate(struct thread_data *td, struct timeval *now, |
bd93e34b | 252 | enum fio_ddir ddir) |
ebac4655 | 253 | { |
0904200b | 254 | unsigned long long bytes = 0; |
4e991c23 | 255 | unsigned long iops = 0; |
ebac4655 JA |
256 | unsigned long spent; |
257 | unsigned long rate; | |
581e7141 JA |
258 | unsigned int ratemin = 0; |
259 | unsigned int rate_iops = 0; | |
260 | unsigned int rate_iops_min = 0; | |
ebac4655 | 261 | |
ff58fced JA |
262 | assert(ddir_rw(ddir)); |
263 | ||
d982d873 JA |
264 | if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir]) |
265 | return 0; | |
266 | ||
ebac4655 JA |
267 | /* |
268 | * allow a 2 second settle period in the beginning | |
269 | */ | |
270 | if (mtime_since(&td->start, now) < 2000) | |
271 | return 0; | |
272 | ||
c8eeb9df | 273 | iops += td->this_io_blocks[ddir]; |
581e7141 JA |
274 | bytes += td->this_io_bytes[ddir]; |
275 | ratemin += td->o.ratemin[ddir]; | |
276 | rate_iops += td->o.rate_iops[ddir]; | |
277 | rate_iops_min += td->o.rate_iops_min[ddir]; | |
0904200b | 278 | |
ebac4655 JA |
279 | /* |
280 | * if rate blocks is set, sample is running | |
281 | */ | |
581e7141 JA |
282 | if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) { |
283 | spent = mtime_since(&td->lastrate[ddir], now); | |
2dc1bbeb | 284 | if (spent < td->o.ratecycle) |
ebac4655 JA |
285 | return 0; |
286 | ||
581e7141 | 287 | if (td->o.rate[ddir]) { |
4e991c23 JA |
288 | /* |
289 | * check bandwidth specified rate | |
290 | */ | |
581e7141 | 291 | if (bytes < td->rate_bytes[ddir]) { |
5ec10eaa | 292 | log_err("%s: min rate %u not met\n", td->o.name, |
581e7141 | 293 | ratemin); |
4e991c23 JA |
294 | return 1; |
295 | } else { | |
581e7141 JA |
296 | rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent; |
297 | if (rate < ratemin || | |
298 | bytes < td->rate_bytes[ddir]) { | |
5ec10eaa | 299 | log_err("%s: min rate %u not met, got" |
b22989b9 | 300 | " %luKB/sec\n", td->o.name, |
581e7141 | 301 | ratemin, rate); |
4e991c23 JA |
302 | return 1; |
303 | } | |
304 | } | |
413dd459 | 305 | } else { |
4e991c23 JA |
306 | /* |
307 | * checks iops specified rate | |
308 | */ | |
581e7141 | 309 | if (iops < rate_iops) { |
5ec10eaa | 310 | log_err("%s: min iops rate %u not met\n", |
581e7141 | 311 | td->o.name, rate_iops); |
413dd459 | 312 | return 1; |
4e991c23 | 313 | } else { |
581e7141 JA |
314 | rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent; |
315 | if (rate < rate_iops_min || | |
316 | iops < td->rate_blocks[ddir]) { | |
5ec10eaa JA |
317 | log_err("%s: min iops rate %u not met," |
318 | " got %lu\n", td->o.name, | |
581e7141 | 319 | rate_iops_min, rate); |
4e991c23 | 320 | } |
413dd459 | 321 | } |
ebac4655 JA |
322 | } |
323 | } | |
324 | ||
581e7141 JA |
325 | td->rate_bytes[ddir] = bytes; |
326 | td->rate_blocks[ddir] = iops; | |
327 | memcpy(&td->lastrate[ddir], now, sizeof(*now)); | |
ebac4655 JA |
328 | return 0; |
329 | } | |
330 | ||
581e7141 JA |
331 | static int check_min_rate(struct thread_data *td, struct timeval *now, |
332 | unsigned long *bytes_done) | |
333 | { | |
334 | int ret = 0; | |
335 | ||
336 | if (bytes_done[0]) | |
337 | ret |= __check_min_rate(td, now, 0); | |
338 | if (bytes_done[1]) | |
339 | ret |= __check_min_rate(td, now, 1); | |
340 | ||
341 | return ret; | |
342 | } | |
343 | ||
ebac4655 JA |
344 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) |
345 | { | |
2dc1bbeb | 346 | if (!td->o.timeout) |
ebac4655 | 347 | return 0; |
2dc1bbeb | 348 | if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000) |
ebac4655 JA |
349 | return 1; |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
906c8d75 JA |
354 | /* |
355 | * When job exits, we can cancel the in-flight IO if we are using async | |
356 | * io. Attempt to do so. | |
357 | */ | |
ebac4655 JA |
358 | static void cleanup_pending_aio(struct thread_data *td) |
359 | { | |
01743ee1 | 360 | struct flist_head *entry, *n; |
ebac4655 JA |
361 | struct io_u *io_u; |
362 | int r; | |
363 | ||
364 | /* | |
365 | * get immediately available events, if any | |
366 | */ | |
581e7141 | 367 | r = io_u_queued_complete(td, 0, NULL); |
b2fdda43 JA |
368 | if (r < 0) |
369 | return; | |
ebac4655 JA |
370 | |
371 | /* | |
372 | * now cancel remaining active events | |
373 | */ | |
2866c82d | 374 | if (td->io_ops->cancel) { |
01743ee1 JA |
375 | flist_for_each_safe(entry, n, &td->io_u_busylist) { |
376 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 377 | |
0c6e7517 JA |
378 | /* |
379 | * if the io_u isn't in flight, then that generally | |
380 | * means someone leaked an io_u. complain but fix | |
381 | * it up, so we don't stall here. | |
382 | */ | |
383 | if ((io_u->flags & IO_U_F_FLIGHT) == 0) { | |
384 | log_err("fio: non-busy IO on busy list\n"); | |
ebac4655 | 385 | put_io_u(td, io_u); |
0c6e7517 JA |
386 | } else { |
387 | r = td->io_ops->cancel(td, io_u); | |
388 | if (!r) | |
389 | put_io_u(td, io_u); | |
390 | } | |
ebac4655 JA |
391 | } |
392 | } | |
393 | ||
97601024 | 394 | if (td->cur_depth) |
581e7141 | 395 | r = io_u_queued_complete(td, td->cur_depth, NULL); |
ebac4655 JA |
396 | } |
397 | ||
858a3d47 JA |
398 | /* |
399 | * Helper to handle the final sync of a file. Works just like the normal | |
400 | * io path, just does everything sync. | |
401 | */ | |
402 | static int fio_io_sync(struct thread_data *td, struct fio_file *f) | |
403 | { | |
404 | struct io_u *io_u = __get_io_u(td); | |
858a3d47 JA |
405 | int ret; |
406 | ||
407 | if (!io_u) | |
408 | return 1; | |
409 | ||
410 | io_u->ddir = DDIR_SYNC; | |
411 | io_u->file = f; | |
412 | ||
413 | if (td_io_prep(td, io_u)) { | |
414 | put_io_u(td, io_u); | |
415 | return 1; | |
416 | } | |
417 | ||
755200a3 | 418 | requeue: |
858a3d47 | 419 | ret = td_io_queue(td, io_u); |
36167d82 | 420 | if (ret < 0) { |
e1161c32 | 421 | td_verror(td, io_u->error, "td_io_queue"); |
858a3d47 | 422 | put_io_u(td, io_u); |
858a3d47 | 423 | return 1; |
36167d82 | 424 | } else if (ret == FIO_Q_QUEUED) { |
581e7141 | 425 | if (io_u_queued_complete(td, 1, NULL) < 0) |
36167d82 | 426 | return 1; |
36167d82 JA |
427 | } else if (ret == FIO_Q_COMPLETED) { |
428 | if (io_u->error) { | |
e1161c32 | 429 | td_verror(td, io_u->error, "td_io_queue"); |
36167d82 JA |
430 | return 1; |
431 | } | |
858a3d47 | 432 | |
581e7141 | 433 | if (io_u_sync_complete(td, io_u, NULL) < 0) |
b2fdda43 | 434 | return 1; |
755200a3 JA |
435 | } else if (ret == FIO_Q_BUSY) { |
436 | if (td_io_commit(td)) | |
437 | return 1; | |
438 | goto requeue; | |
858a3d47 JA |
439 | } |
440 | ||
441 | return 0; | |
442 | } | |
443 | ||
687c6a23 JA |
444 | static inline void __update_tv_cache(struct thread_data *td) |
445 | { | |
446 | fio_gettime(&td->tv_cache, NULL); | |
447 | } | |
448 | ||
993bf48b JA |
449 | static inline void update_tv_cache(struct thread_data *td) |
450 | { | |
451 | if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask) | |
687c6a23 | 452 | __update_tv_cache(td); |
993bf48b JA |
453 | } |
454 | ||
06842027 SL |
455 | static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir, |
456 | int *retptr) | |
f2bba182 RR |
457 | { |
458 | int ret = *retptr; | |
459 | ||
460 | if (ret < 0 || td->error) { | |
461 | int err; | |
462 | ||
f2bba182 RR |
463 | if (ret < 0) |
464 | err = -ret; | |
465 | else | |
466 | err = td->error; | |
467 | ||
06842027 SL |
468 | if (!(td->o.continue_on_error & td_error_type(ddir, err))) |
469 | return 1; | |
470 | ||
f2bba182 RR |
471 | if (td_non_fatal_error(err)) { |
472 | /* | |
473 | * Continue with the I/Os in case of | |
474 | * a non fatal error. | |
475 | */ | |
8a4f8a24 | 476 | update_error_count(td, err); |
f2bba182 RR |
477 | td_clear_error(td); |
478 | *retptr = 0; | |
479 | return 0; | |
8a4f8a24 RR |
480 | } else if (td->o.fill_device && err == ENOSPC) { |
481 | /* | |
482 | * We expect to hit this error if | |
483 | * fill_device option is set. | |
484 | */ | |
485 | td_clear_error(td); | |
486 | td->terminate = 1; | |
487 | return 1; | |
f2bba182 RR |
488 | } else { |
489 | /* | |
490 | * Stop the I/O in case of a fatal | |
491 | * error. | |
492 | */ | |
8a4f8a24 | 493 | update_error_count(td, err); |
f2bba182 RR |
494 | return 1; |
495 | } | |
496 | } | |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
906c8d75 | 501 | /* |
34403fb1 | 502 | * The main verify engine. Runs over the writes we previously submitted, |
906c8d75 JA |
503 | * reads the blocks back in, and checks the crc/md5 of the data. |
504 | */ | |
1e97cce9 | 505 | static void do_verify(struct thread_data *td) |
ebac4655 | 506 | { |
53cdc686 | 507 | struct fio_file *f; |
36167d82 | 508 | struct io_u *io_u; |
af52b345 JA |
509 | int ret, min_events; |
510 | unsigned int i; | |
e5b401d4 | 511 | |
008618a8 JA |
512 | dprint(FD_VERIFY, "starting loop\n"); |
513 | ||
e5b401d4 JA |
514 | /* |
515 | * sync io first and invalidate cache, to make sure we really | |
516 | * read from disk. | |
517 | */ | |
518 | for_each_file(td, f, i) { | |
d6aed795 | 519 | if (!fio_file_open(f)) |
3d7b485f | 520 | continue; |
b2fdda43 JA |
521 | if (fio_io_sync(td, f)) |
522 | break; | |
523 | if (file_invalidate_cache(td, f)) | |
524 | break; | |
e5b401d4 | 525 | } |
ebac4655 | 526 | |
b2fdda43 JA |
527 | if (td->error) |
528 | return; | |
529 | ||
ebac4655 JA |
530 | td_set_runstate(td, TD_VERIFYING); |
531 | ||
36167d82 JA |
532 | io_u = NULL; |
533 | while (!td->terminate) { | |
4950421a | 534 | int ret2, full; |
5451792e | 535 | |
993bf48b JA |
536 | update_tv_cache(td); |
537 | ||
538 | if (runtime_exceeded(td, &td->tv_cache)) { | |
687c6a23 JA |
539 | __update_tv_cache(td); |
540 | if (runtime_exceeded(td, &td->tv_cache)) { | |
541 | td->terminate = 1; | |
542 | break; | |
543 | } | |
069c2918 | 544 | } |
02bcaa8c | 545 | |
dcb4830d RR |
546 | io_u = __get_io_u(td); |
547 | if (!io_u) | |
548 | break; | |
549 | ||
069c2918 JA |
550 | if (get_next_verify(td, io_u)) { |
551 | put_io_u(td, io_u); | |
ebac4655 | 552 | break; |
069c2918 | 553 | } |
ebac4655 | 554 | |
069c2918 JA |
555 | if (td_io_prep(td, io_u)) { |
556 | put_io_u(td, io_u); | |
53cdc686 | 557 | break; |
069c2918 | 558 | } |
d7762cf8 | 559 | |
e8462bd8 JA |
560 | if (td->o.verify_async) |
561 | io_u->end_io = verify_io_u_async; | |
562 | else | |
563 | io_u->end_io = verify_io_u; | |
53cdc686 | 564 | |
11786802 | 565 | ret = td_io_queue(td, io_u); |
36167d82 JA |
566 | switch (ret) { |
567 | case FIO_Q_COMPLETED: | |
f2bba182 | 568 | if (io_u->error) { |
22819ec2 | 569 | ret = -io_u->error; |
f2bba182 RR |
570 | clear_io_u(td, io_u); |
571 | } else if (io_u->resid) { | |
36167d82 | 572 | int bytes = io_u->xfer_buflen - io_u->resid; |
ebac4655 | 573 | |
9e9d164e JA |
574 | /* |
575 | * zero read, fail | |
576 | */ | |
577 | if (!bytes) { | |
41d9f0fd | 578 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
579 | put_io_u(td, io_u); |
580 | break; | |
581 | } | |
8400d9b2 | 582 | |
36167d82 JA |
583 | io_u->xfer_buflen = io_u->resid; |
584 | io_u->xfer_buf += bytes; | |
8400d9b2 JA |
585 | io_u->offset += bytes; |
586 | ||
ff58fced JA |
587 | if (ddir_rw(io_u->ddir)) |
588 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 589 | |
2b13e716 | 590 | f = io_u->file; |
d460eb31 | 591 | if (io_u->offset == f->real_file_size) |
8400d9b2 JA |
592 | goto sync_done; |
593 | ||
11786802 JA |
594 | requeue_io_u(td, &io_u); |
595 | } else { | |
8400d9b2 | 596 | sync_done: |
581e7141 | 597 | ret = io_u_sync_complete(td, io_u, NULL); |
11786802 JA |
598 | if (ret < 0) |
599 | break; | |
36167d82 | 600 | } |
36167d82 JA |
601 | continue; |
602 | case FIO_Q_QUEUED: | |
603 | break; | |
755200a3 JA |
604 | case FIO_Q_BUSY: |
605 | requeue_io_u(td, &io_u); | |
5451792e JA |
606 | ret2 = td_io_commit(td); |
607 | if (ret2 < 0) | |
608 | ret = ret2; | |
755200a3 | 609 | break; |
36167d82 JA |
610 | default: |
611 | assert(ret < 0); | |
e1161c32 | 612 | td_verror(td, -ret, "td_io_queue"); |
ebac4655 JA |
613 | break; |
614 | } | |
615 | ||
06842027 | 616 | if (break_on_this_error(td, io_u->ddir, &ret)) |
3af6ef39 JA |
617 | break; |
618 | ||
ebac4655 | 619 | /* |
3af6ef39 | 620 | * if we can queue more, do so. but check if there are |
ccc02050 JA |
621 | * completed io_u's first. Note that we can get BUSY even |
622 | * without IO queued, if the system is resource starved. | |
ebac4655 | 623 | */ |
ccc02050 | 624 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 625 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
626 | min_events = min(td->o.iodepth_batch_complete, |
627 | td->cur_depth); | |
1f84f1d9 | 628 | if (full && !min_events && td->o.iodepth_batch_complete != 0) |
838bc709 | 629 | min_events = 1; |
e916b390 | 630 | |
4950421a JA |
631 | do { |
632 | /* | |
633 | * Reap required number of io units, if any, | |
634 | * and do the verification on them through | |
635 | * the callback handler | |
636 | */ | |
581e7141 | 637 | if (io_u_queued_complete(td, min_events, NULL) < 0) { |
4950421a JA |
638 | ret = -1; |
639 | break; | |
640 | } | |
641 | } while (full && (td->cur_depth > td->o.iodepth_low)); | |
642 | } | |
643 | if (ret < 0) | |
ebac4655 | 644 | break; |
36167d82 | 645 | } |
ebac4655 | 646 | |
c01c0395 JA |
647 | if (!td->error) { |
648 | min_events = td->cur_depth; | |
649 | ||
650 | if (min_events) | |
581e7141 | 651 | ret = io_u_queued_complete(td, min_events, NULL); |
c01c0395 | 652 | } else |
ebac4655 JA |
653 | cleanup_pending_aio(td); |
654 | ||
655 | td_set_runstate(td, TD_RUNNING); | |
008618a8 JA |
656 | |
657 | dprint(FD_VERIFY, "exiting loop\n"); | |
ebac4655 JA |
658 | } |
659 | ||
32cd46a0 | 660 | /* |
906c8d75 | 661 | * Main IO worker function. It retrieves io_u's to process and queues |
32cd46a0 JA |
662 | * and reaps them, checking for rate and errors along the way. |
663 | */ | |
ebac4655 JA |
664 | static void do_io(struct thread_data *td) |
665 | { | |
af52b345 JA |
666 | unsigned int i; |
667 | int ret = 0; | |
ebac4655 | 668 | |
b29ee5b3 JA |
669 | if (in_ramp_time(td)) |
670 | td_set_runstate(td, TD_RAMP); | |
671 | else | |
672 | td_set_runstate(td, TD_RUNNING); | |
5853e5a8 | 673 | |
457bf399 | 674 | while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) || |
a917a8b3 | 675 | (!flist_empty(&td->trim_list)) || |
457bf399 | 676 | ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) { |
97601024 | 677 | struct timeval comp_time; |
581e7141 | 678 | unsigned long bytes_done[2] = { 0, 0 }; |
84585003 | 679 | int min_evts = 0; |
ebac4655 | 680 | struct io_u *io_u; |
4950421a | 681 | int ret2, full; |
06842027 | 682 | enum fio_ddir ddir; |
ebac4655 JA |
683 | |
684 | if (td->terminate) | |
685 | break; | |
686 | ||
993bf48b | 687 | update_tv_cache(td); |
97601024 | 688 | |
993bf48b | 689 | if (runtime_exceeded(td, &td->tv_cache)) { |
687c6a23 JA |
690 | __update_tv_cache(td); |
691 | if (runtime_exceeded(td, &td->tv_cache)) { | |
692 | td->terminate = 1; | |
693 | break; | |
694 | } | |
97601024 | 695 | } |
36167d82 | 696 | |
dcb4830d RR |
697 | io_u = get_io_u(td); |
698 | if (!io_u) | |
699 | break; | |
700 | ||
06842027 SL |
701 | ddir = io_u->ddir; |
702 | ||
b67740d3 JA |
703 | /* |
704 | * Add verification end_io handler, if asked to verify | |
705 | * a previously written file. | |
706 | */ | |
715d2b3e JA |
707 | if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ && |
708 | !td_rw(td)) { | |
e8462bd8 JA |
709 | if (td->o.verify_async) |
710 | io_u->end_io = verify_io_u_async; | |
711 | else | |
712 | io_u->end_io = verify_io_u; | |
d8fab1b6 | 713 | td_set_runstate(td, TD_VERIFYING); |
b29ee5b3 JA |
714 | } else if (in_ramp_time(td)) |
715 | td_set_runstate(td, TD_RAMP); | |
716 | else | |
d8fab1b6 | 717 | td_set_runstate(td, TD_RUNNING); |
b67740d3 | 718 | |
11786802 | 719 | ret = td_io_queue(td, io_u); |
36167d82 JA |
720 | switch (ret) { |
721 | case FIO_Q_COMPLETED: | |
f2bba182 | 722 | if (io_u->error) { |
5451792e | 723 | ret = -io_u->error; |
f2bba182 RR |
724 | clear_io_u(td, io_u); |
725 | } else if (io_u->resid) { | |
36167d82 | 726 | int bytes = io_u->xfer_buflen - io_u->resid; |
d460eb31 | 727 | struct fio_file *f = io_u->file; |
36167d82 | 728 | |
9e9d164e JA |
729 | /* |
730 | * zero read, fail | |
731 | */ | |
732 | if (!bytes) { | |
41d9f0fd | 733 | td_verror(td, EIO, "full resid"); |
9e9d164e JA |
734 | put_io_u(td, io_u); |
735 | break; | |
736 | } | |
737 | ||
cec6b55d | 738 | io_u->xfer_buflen = io_u->resid; |
36167d82 | 739 | io_u->xfer_buf += bytes; |
5a7c5680 JA |
740 | io_u->offset += bytes; |
741 | ||
ff58fced JA |
742 | if (ddir_rw(io_u->ddir)) |
743 | td->ts.short_io_u[io_u->ddir]++; | |
30061b97 | 744 | |
d460eb31 | 745 | if (io_u->offset == f->real_file_size) |
5a7c5680 JA |
746 | goto sync_done; |
747 | ||
11786802 JA |
748 | requeue_io_u(td, &io_u); |
749 | } else { | |
5a7c5680 | 750 | sync_done: |
581e7141 JA |
751 | if (__should_check_rate(td, 0) || |
752 | __should_check_rate(td, 1)) | |
993bf48b JA |
753 | fio_gettime(&comp_time, NULL); |
754 | ||
581e7141 JA |
755 | ret = io_u_sync_complete(td, io_u, bytes_done); |
756 | if (ret < 0) | |
757 | break; | |
cec6b55d | 758 | } |
36167d82 JA |
759 | break; |
760 | case FIO_Q_QUEUED: | |
7e77dd02 JA |
761 | /* |
762 | * if the engine doesn't have a commit hook, | |
763 | * the io_u is really queued. if it does have such | |
764 | * a hook, it has to call io_u_queued() itself. | |
765 | */ | |
766 | if (td->io_ops->commit == NULL) | |
767 | io_u_queued(td, io_u); | |
36167d82 | 768 | break; |
755200a3 JA |
769 | case FIO_Q_BUSY: |
770 | requeue_io_u(td, &io_u); | |
5451792e JA |
771 | ret2 = td_io_commit(td); |
772 | if (ret2 < 0) | |
773 | ret = ret2; | |
755200a3 | 774 | break; |
36167d82 JA |
775 | default: |
776 | assert(ret < 0); | |
777 | put_io_u(td, io_u); | |
778 | break; | |
ebac4655 JA |
779 | } |
780 | ||
06842027 | 781 | if (break_on_this_error(td, ddir, &ret)) |
36167d82 JA |
782 | break; |
783 | ||
97601024 | 784 | /* |
ccc02050 JA |
785 | * See if we need to complete some commands. Note that we |
786 | * can get BUSY even without IO queued, if the system is | |
787 | * resource starved. | |
97601024 | 788 | */ |
ccc02050 | 789 | full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth); |
4950421a | 790 | if (full || !td->o.iodepth_batch_complete) { |
6a96276c RR |
791 | min_evts = min(td->o.iodepth_batch_complete, |
792 | td->cur_depth); | |
1f84f1d9 | 793 | if (full && !min_evts && td->o.iodepth_batch_complete != 0) |
36167d82 | 794 | min_evts = 1; |
4950421a | 795 | |
581e7141 JA |
796 | if (__should_check_rate(td, 0) || |
797 | __should_check_rate(td, 1)) | |
993bf48b | 798 | fio_gettime(&comp_time, NULL); |
4950421a JA |
799 | |
800 | do { | |
581e7141 JA |
801 | ret = io_u_queued_complete(td, min_evts, bytes_done); |
802 | if (ret < 0) | |
4950421a JA |
803 | break; |
804 | ||
4950421a | 805 | } while (full && (td->cur_depth > td->o.iodepth_low)); |
ebac4655 JA |
806 | } |
807 | ||
4950421a JA |
808 | if (ret < 0) |
809 | break; | |
581e7141 | 810 | if (!(bytes_done[0] + bytes_done[1])) |
97601024 JA |
811 | continue; |
812 | ||
581e7141 JA |
813 | if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) { |
814 | if (check_min_rate(td, &comp_time, bytes_done)) { | |
721938ae | 815 | if (exitall_on_terminate) |
cc0df00a | 816 | fio_terminate_threads(td->groupid); |
721938ae JA |
817 | td_verror(td, EIO, "check_min_rate"); |
818 | break; | |
819 | } | |
ebac4655 JA |
820 | } |
821 | ||
2dc1bbeb | 822 | if (td->o.thinktime) { |
9c1f7434 JA |
823 | unsigned long long b; |
824 | ||
825 | b = td->io_blocks[0] + td->io_blocks[1]; | |
2dc1bbeb | 826 | if (!(b % td->o.thinktime_blocks)) { |
48097d5c JA |
827 | int left; |
828 | ||
2dc1bbeb | 829 | if (td->o.thinktime_spin) |
6dd6f2cd | 830 | usec_spin(td->o.thinktime_spin); |
48097d5c | 831 | |
2dc1bbeb | 832 | left = td->o.thinktime - td->o.thinktime_spin; |
48097d5c JA |
833 | if (left) |
834 | usec_sleep(td, left); | |
835 | } | |
9c1f7434 | 836 | } |
ebac4655 JA |
837 | } |
838 | ||
d708ac5e | 839 | if (td->trim_entries) |
18320cf4 | 840 | log_err("fio: %d trim entries leaked?\n", td->trim_entries); |
a917a8b3 | 841 | |
aa31f1f1 SL |
842 | if (td->o.fill_device && td->error == ENOSPC) { |
843 | td->error = 0; | |
844 | td->terminate = 1; | |
845 | } | |
4d2413c6 | 846 | if (!td->error) { |
3d7c391d JA |
847 | struct fio_file *f; |
848 | ||
c01c0395 | 849 | i = td->cur_depth; |
d06093e7 | 850 | if (i) { |
581e7141 | 851 | ret = io_u_queued_complete(td, i, NULL); |
d06093e7 SL |
852 | if (td->o.fill_device && td->error == ENOSPC) |
853 | td->error = 0; | |
854 | } | |
ebac4655 | 855 | |
2dc1bbeb | 856 | if (should_fsync(td) && td->o.end_fsync) { |
84585003 | 857 | td_set_runstate(td, TD_FSYNCING); |
3d7b485f JA |
858 | |
859 | for_each_file(td, f, i) { | |
d6aed795 | 860 | if (!fio_file_open(f)) |
3d7b485f | 861 | continue; |
858a3d47 | 862 | fio_io_sync(td, f); |
3d7b485f | 863 | } |
84585003 | 864 | } |
c01c0395 JA |
865 | } else |
866 | cleanup_pending_aio(td); | |
2ba1c290 JA |
867 | |
868 | /* | |
869 | * stop job if we failed doing any IO | |
870 | */ | |
871 | if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0) | |
872 | td->done = 1; | |
ebac4655 JA |
873 | } |
874 | ||
ebac4655 JA |
875 | static void cleanup_io_u(struct thread_data *td) |
876 | { | |
01743ee1 | 877 | struct flist_head *entry, *n; |
ebac4655 JA |
878 | struct io_u *io_u; |
879 | ||
01743ee1 JA |
880 | flist_for_each_safe(entry, n, &td->io_u_freelist) { |
881 | io_u = flist_entry(entry, struct io_u, list); | |
ebac4655 | 882 | |
01743ee1 | 883 | flist_del(&io_u->list); |
91aea6ec | 884 | fio_memfree(io_u, sizeof(*io_u)); |
ebac4655 JA |
885 | } |
886 | ||
2f9ade3c | 887 | free_io_mem(td); |
ebac4655 JA |
888 | } |
889 | ||
890 | static int init_io_u(struct thread_data *td) | |
891 | { | |
892 | struct io_u *io_u; | |
a00735e6 | 893 | unsigned int max_bs; |
eb7ccf38 | 894 | int cl_align, i, max_units; |
ebac4655 JA |
895 | char *p; |
896 | ||
1e3c09af | 897 | max_units = td->o.iodepth; |
2dc1bbeb | 898 | max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]); |
5ec10eaa JA |
899 | td->orig_buffer_size = (unsigned long long) max_bs |
900 | * (unsigned long long) max_units; | |
901 | ||
902 | if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) { | |
903 | unsigned long bs; | |
74b025b0 | 904 | |
5ec10eaa JA |
905 | bs = td->orig_buffer_size + td->o.hugepage_size - 1; |
906 | td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1); | |
907 | } | |
ebac4655 | 908 | |
c3990645 JA |
909 | if (td->orig_buffer_size != (size_t) td->orig_buffer_size) { |
910 | log_err("fio: IO memory too large. Reduce max_bs or iodepth\n"); | |
911 | return 1; | |
912 | } | |
913 | ||
2f9ade3c JA |
914 | if (allocate_io_mem(td)) |
915 | return 1; | |
ebac4655 | 916 | |
1c04bf78 JA |
917 | if (td->o.odirect || td->o.mem_align || |
918 | (td->io_ops->flags & FIO_RAWIO)) | |
d529ee19 | 919 | p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align; |
b3f378e9 JA |
920 | else |
921 | p = td->orig_buffer; | |
922 | ||
eb7ccf38 JA |
923 | cl_align = os_cache_line_size(); |
924 | ||
ebac4655 | 925 | for (i = 0; i < max_units; i++) { |
eb7ccf38 JA |
926 | void *ptr; |
927 | ||
db1cd90b JA |
928 | if (td->terminate) |
929 | return 1; | |
eb7ccf38 | 930 | |
91aea6ec JA |
931 | ptr = fio_memalign(cl_align, sizeof(*io_u)); |
932 | if (!ptr) { | |
933 | log_err("fio: unable to allocate aligned memory\n"); | |
eb7ccf38 JA |
934 | break; |
935 | } | |
936 | ||
937 | io_u = ptr; | |
ebac4655 | 938 | memset(io_u, 0, sizeof(*io_u)); |
01743ee1 | 939 | INIT_FLIST_HEAD(&io_u->list); |
d529ee19 | 940 | dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i); |
ebac4655 | 941 | |
b4c5e1ac | 942 | if (!(td->io_ops->flags & FIO_NOIO)) { |
cf00f975 | 943 | io_u->buf = p; |
d529ee19 | 944 | dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf); |
e9459e5a | 945 | |
3545a109 | 946 | if (td_write(td)) |
5973cafb | 947 | io_u_fill_buffer(td, io_u, max_bs); |
3545a109 | 948 | if (td_write(td) && td->o.verify_pattern_bytes) { |
cbe8d756 RR |
949 | /* |
950 | * Fill the buffer with the pattern if we are | |
951 | * going to be doing writes. | |
952 | */ | |
7d9fb455 | 953 | fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0); |
cbe8d756 | 954 | } |
b4c5e1ac | 955 | } |
6b9cea23 | 956 | |
b1ff3403 | 957 | io_u->index = i; |
0c6e7517 | 958 | io_u->flags = IO_U_F_FREE; |
01743ee1 | 959 | flist_add(&io_u->list, &td->io_u_freelist); |
cf00f975 | 960 | p += max_bs; |
ebac4655 JA |
961 | } |
962 | ||
963 | return 0; | |
964 | } | |
965 | ||
da86774e JA |
966 | static int switch_ioscheduler(struct thread_data *td) |
967 | { | |
968 | char tmp[256], tmp2[128]; | |
969 | FILE *f; | |
970 | int ret; | |
971 | ||
ba0fbe10 | 972 | if (td->io_ops->flags & FIO_DISKLESSIO) |
f48b467c JA |
973 | return 0; |
974 | ||
da86774e JA |
975 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); |
976 | ||
977 | f = fopen(tmp, "r+"); | |
978 | if (!f) { | |
cbf5c1d7 | 979 | if (errno == ENOENT) { |
5ec10eaa JA |
980 | log_err("fio: os or kernel doesn't support IO scheduler" |
981 | " switching\n"); | |
cbf5c1d7 JA |
982 | return 0; |
983 | } | |
984 | td_verror(td, errno, "fopen iosched"); | |
da86774e JA |
985 | return 1; |
986 | } | |
987 | ||
988 | /* | |
989 | * Set io scheduler. | |
990 | */ | |
2dc1bbeb | 991 | ret = fwrite(td->o.ioscheduler, strlen(td->o.ioscheduler), 1, f); |
da86774e | 992 | if (ferror(f) || ret != 1) { |
e1161c32 | 993 | td_verror(td, errno, "fwrite"); |
da86774e JA |
994 | fclose(f); |
995 | return 1; | |
996 | } | |
997 | ||
998 | rewind(f); | |
999 | ||
1000 | /* | |
1001 | * Read back and check that the selected scheduler is now the default. | |
1002 | */ | |
1003 | ret = fread(tmp, 1, sizeof(tmp), f); | |
1004 | if (ferror(f) || ret < 0) { | |
e1161c32 | 1005 | td_verror(td, errno, "fread"); |
da86774e JA |
1006 | fclose(f); |
1007 | return 1; | |
1008 | } | |
1009 | ||
2dc1bbeb | 1010 | sprintf(tmp2, "[%s]", td->o.ioscheduler); |
da86774e | 1011 | if (!strstr(tmp, tmp2)) { |
2dc1bbeb | 1012 | log_err("fio: io scheduler %s not found\n", td->o.ioscheduler); |
e1161c32 | 1013 | td_verror(td, EINVAL, "iosched_switch"); |
da86774e JA |
1014 | fclose(f); |
1015 | return 1; | |
1016 | } | |
1017 | ||
1018 | fclose(f); | |
1019 | return 0; | |
1020 | } | |
1021 | ||
492158cf JA |
1022 | static int keep_running(struct thread_data *td) |
1023 | { | |
1024 | unsigned long long io_done; | |
1025 | ||
20e354ef JA |
1026 | if (td->done) |
1027 | return 0; | |
492158cf JA |
1028 | if (td->o.time_based) |
1029 | return 1; | |
1030 | if (td->o.loops) { | |
1031 | td->o.loops--; | |
1032 | return 1; | |
1033 | } | |
1034 | ||
5ec10eaa JA |
1035 | io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] |
1036 | + td->io_skip_bytes; | |
492158cf JA |
1037 | if (io_done < td->o.size) |
1038 | return 1; | |
1039 | ||
1040 | return 0; | |
1041 | } | |
1042 | ||
b29ee5b3 | 1043 | static void reset_io_counters(struct thread_data *td) |
ebac4655 | 1044 | { |
f0505a14 | 1045 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
ebac4655 | 1046 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; |
c8eeb9df JA |
1047 | td->stat_io_blocks[0] = td->stat_io_blocks[1] = 0; |
1048 | td->this_io_blocks[0] = td->this_io_blocks[1] = 0; | |
20dc95c4 | 1049 | td->zone_bytes = 0; |
581e7141 JA |
1050 | td->rate_bytes[0] = td->rate_bytes[1] = 0; |
1051 | td->rate_blocks[0] = td->rate_blocks[1] = 0; | |
ebac4655 | 1052 | |
c1324df1 JA |
1053 | td->last_was_sync = 0; |
1054 | ||
2ba1c290 JA |
1055 | /* |
1056 | * reset file done count if we are to start over | |
1057 | */ | |
1058 | if (td->o.time_based || td->o.loops) | |
48f5abd3 | 1059 | td->nr_done_files = 0; |
b29ee5b3 JA |
1060 | } |
1061 | ||
1062 | void reset_all_stats(struct thread_data *td) | |
1063 | { | |
1064 | struct timeval tv; | |
1065 | int i; | |
1066 | ||
1067 | reset_io_counters(td); | |
1068 | ||
1069 | for (i = 0; i < 2; i++) { | |
1070 | td->io_bytes[i] = 0; | |
1071 | td->io_blocks[i] = 0; | |
1072 | td->io_issues[i] = 0; | |
1073 | td->ts.total_io_u[i] = 0; | |
1074 | } | |
9b836561 | 1075 | |
b29ee5b3 | 1076 | fio_gettime(&tv, NULL); |
cda99fa0 SSP |
1077 | td->ts.runtime[0] = 0; |
1078 | td->ts.runtime[1] = 0; | |
b29ee5b3 JA |
1079 | memcpy(&td->epoch, &tv, sizeof(tv)); |
1080 | memcpy(&td->start, &tv, sizeof(tv)); | |
1081 | } | |
1082 | ||
df9c26b1 | 1083 | static void clear_io_state(struct thread_data *td) |
b29ee5b3 JA |
1084 | { |
1085 | struct fio_file *f; | |
1086 | unsigned int i; | |
b29ee5b3 JA |
1087 | |
1088 | reset_io_counters(td); | |
281f9b63 | 1089 | |
24ffd2c2 | 1090 | close_files(td); |
df9c26b1 | 1091 | for_each_file(td, f, i) |
d6aed795 | 1092 | fio_file_clear_done(f); |
f0153948 JA |
1093 | |
1094 | /* | |
1095 | * Set the same seed to get repeatable runs | |
1096 | */ | |
1097 | td_fill_rand_seeds(td); | |
ebac4655 JA |
1098 | } |
1099 | ||
398e8c4d JA |
1100 | static int exec_string(const char *string) |
1101 | { | |
1102 | int ret, newlen = strlen(string) + 1 + 8; | |
1103 | char *str; | |
1104 | ||
1105 | str = malloc(newlen); | |
1106 | sprintf(str, "sh -c %s", string); | |
1107 | ||
1108 | ret = system(str); | |
1109 | if (ret == -1) | |
1110 | log_err("fio: exec of cmd <%s> failed\n", str); | |
1111 | ||
1112 | free(str); | |
1113 | return ret; | |
1114 | } | |
1115 | ||
906c8d75 JA |
1116 | /* |
1117 | * Entry point for the thread based jobs. The process based jobs end up | |
1118 | * here as well, after a little setup. | |
1119 | */ | |
ebac4655 JA |
1120 | static void *thread_main(void *data) |
1121 | { | |
cda99fa0 | 1122 | unsigned long long elapsed; |
ebac4655 | 1123 | struct thread_data *td = data; |
e8462bd8 | 1124 | pthread_condattr_t attr; |
a978ba68 | 1125 | int clear_state; |
ebac4655 | 1126 | |
47f767c1 | 1127 | if (!td->o.use_thread) { |
ebac4655 | 1128 | setsid(); |
47f767c1 JA |
1129 | td->pid = getpid(); |
1130 | } else | |
1131 | td->pid = gettid(); | |
ebac4655 | 1132 | |
5921e80c | 1133 | dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid); |
ee56ad50 | 1134 | |
01743ee1 JA |
1135 | INIT_FLIST_HEAD(&td->io_u_freelist); |
1136 | INIT_FLIST_HEAD(&td->io_u_busylist); | |
1137 | INIT_FLIST_HEAD(&td->io_u_requeues); | |
1138 | INIT_FLIST_HEAD(&td->io_log_list); | |
1139 | INIT_FLIST_HEAD(&td->io_hist_list); | |
e8462bd8 | 1140 | INIT_FLIST_HEAD(&td->verify_list); |
0d29de83 | 1141 | INIT_FLIST_HEAD(&td->trim_list); |
e8462bd8 | 1142 | pthread_mutex_init(&td->io_u_lock, NULL); |
bb5d7d0b | 1143 | td->io_hist_tree = RB_ROOT; |
aea47d44 | 1144 | |
e8462bd8 JA |
1145 | pthread_condattr_init(&attr); |
1146 | pthread_cond_init(&td->verify_cond, &attr); | |
1147 | pthread_cond_init(&td->free_cond, &attr); | |
1148 | ||
db1cd90b | 1149 | td_set_runstate(td, TD_INITIALIZED); |
29adda3c | 1150 | dprint(FD_MUTEX, "up startup_mutex\n"); |
cdd18ad8 | 1151 | fio_mutex_up(startup_mutex); |
29adda3c | 1152 | dprint(FD_MUTEX, "wait on td->mutex\n"); |
cdd18ad8 | 1153 | fio_mutex_down(td->mutex); |
29adda3c | 1154 | dprint(FD_MUTEX, "done waiting on td->mutex\n"); |
db1cd90b JA |
1155 | |
1156 | /* | |
cdd18ad8 | 1157 | * the ->mutex mutex is now no longer used, close it to avoid |
db1cd90b JA |
1158 | * eating a file descriptor |
1159 | */ | |
cdd18ad8 | 1160 | fio_mutex_remove(td->mutex); |
db1cd90b | 1161 | |
dfc6e751 SH |
1162 | /* |
1163 | * A new gid requires privilege, so we need to do this before setting | |
1164 | * the uid. | |
1165 | */ | |
e0b0d892 JA |
1166 | if (td->o.gid != -1U && setgid(td->o.gid)) { |
1167 | td_verror(td, errno, "setgid"); | |
1168 | goto err; | |
1169 | } | |
dfc6e751 SH |
1170 | if (td->o.uid != -1U && setuid(td->o.uid)) { |
1171 | td_verror(td, errno, "setuid"); | |
1172 | goto err; | |
1173 | } | |
e0b0d892 | 1174 | |
be18a94f JA |
1175 | /* |
1176 | * If we have a gettimeofday() thread, make sure we exclude that | |
1177 | * thread from this job | |
1178 | */ | |
1179 | if (td->o.gtod_cpu) | |
1180 | fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu); | |
1181 | ||
1182 | /* | |
1183 | * Set affinity first, in case it has an impact on the memory | |
1184 | * allocations. | |
1185 | */ | |
1186 | if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) { | |
1187 | td_verror(td, errno, "cpu_set_affinity"); | |
1188 | goto err; | |
1189 | } | |
1190 | ||
d84f8d49 JA |
1191 | /* |
1192 | * May alter parameters that init_io_u() will use, so we need to | |
1193 | * do this first. | |
1194 | */ | |
1195 | if (init_iolog(td)) | |
1196 | goto err; | |
1197 | ||
ebac4655 | 1198 | if (init_io_u(td)) |
db1cd90b | 1199 | goto err; |
ebac4655 | 1200 | |
e8462bd8 JA |
1201 | if (td->o.verify_async && verify_async_init(td)) |
1202 | goto err; | |
1203 | ||
ac684785 | 1204 | if (td->ioprio_set) { |
ebac4655 | 1205 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { |
e1161c32 | 1206 | td_verror(td, errno, "ioprio_set"); |
db1cd90b | 1207 | goto err; |
ebac4655 JA |
1208 | } |
1209 | } | |
1210 | ||
6adb38a1 | 1211 | if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt)) |
a696fa2a JA |
1212 | goto err; |
1213 | ||
2dc1bbeb | 1214 | if (nice(td->o.nice) == -1) { |
e1161c32 | 1215 | td_verror(td, errno, "nice"); |
db1cd90b | 1216 | goto err; |
b6f4d880 JA |
1217 | } |
1218 | ||
2dc1bbeb | 1219 | if (td->o.ioscheduler && switch_ioscheduler(td)) |
db1cd90b | 1220 | goto err; |
37f56873 | 1221 | |
2dc1bbeb | 1222 | if (!td->o.create_serialize && setup_files(td)) |
ebac4655 JA |
1223 | goto err; |
1224 | ||
7d6c5283 JA |
1225 | if (td_io_init(td)) |
1226 | goto err; | |
1227 | ||
68727076 JA |
1228 | if (init_random_map(td)) |
1229 | goto err; | |
1230 | ||
2dc1bbeb | 1231 | if (td->o.exec_prerun) { |
398e8c4d | 1232 | if (exec_string(td->o.exec_prerun)) |
69cfd7e0 JA |
1233 | goto err; |
1234 | } | |
4e0ba8af | 1235 | |
afad68f7 ZY |
1236 | if (td->o.pre_read) { |
1237 | if (pre_read_files(td) < 0) | |
1238 | goto err; | |
1239 | } | |
1240 | ||
69008999 | 1241 | fio_gettime(&td->epoch, NULL); |
c8aaba19 | 1242 | getrusage(RUSAGE_SELF, &td->ru_start); |
69008999 | 1243 | |
a978ba68 | 1244 | clear_state = 0; |
492158cf | 1245 | while (keep_running(td)) { |
02bcaa8c | 1246 | fio_gettime(&td->start, NULL); |
c8eeb9df JA |
1247 | memcpy(&td->bw_sample_time, &td->start, sizeof(td->start)); |
1248 | memcpy(&td->iops_sample_time, &td->start, sizeof(td->start)); | |
993bf48b | 1249 | memcpy(&td->tv_cache, &td->start, sizeof(td->start)); |
ebac4655 | 1250 | |
cf2dd8d6 | 1251 | if (td->o.ratemin[0] || td->o.ratemin[1]) |
c8eeb9df | 1252 | memcpy(&td->lastrate, &td->bw_sample_time, |
5ec10eaa | 1253 | sizeof(td->lastrate)); |
ebac4655 | 1254 | |
df9c26b1 JA |
1255 | if (clear_state) |
1256 | clear_io_state(td); | |
a978ba68 | 1257 | |
ebac4655 JA |
1258 | prune_io_piece_log(td); |
1259 | ||
ba0fbe10 | 1260 | do_io(td); |
ebac4655 | 1261 | |
a978ba68 JA |
1262 | clear_state = 1; |
1263 | ||
38d77cae | 1264 | if (td_read(td) && td->io_bytes[DDIR_READ]) { |
0a4957c6 | 1265 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1266 | td->ts.runtime[DDIR_READ] += elapsed; |
38d77cae JA |
1267 | } |
1268 | if (td_write(td) && td->io_bytes[DDIR_WRITE]) { | |
0a4957c6 | 1269 | elapsed = utime_since_now(&td->start); |
cda99fa0 | 1270 | td->ts.runtime[DDIR_WRITE] += elapsed; |
38d77cae | 1271 | } |
5ec10eaa | 1272 | |
ebac4655 JA |
1273 | if (td->error || td->terminate) |
1274 | break; | |
1275 | ||
e84c73a8 SL |
1276 | if (!td->o.do_verify || |
1277 | td->o.verify == VERIFY_NONE || | |
b67740d3 | 1278 | (td->io_ops->flags & FIO_UNIDIR)) |
ebac4655 JA |
1279 | continue; |
1280 | ||
df9c26b1 | 1281 | clear_io_state(td); |
a978ba68 | 1282 | |
02bcaa8c | 1283 | fio_gettime(&td->start, NULL); |
ebac4655 JA |
1284 | |
1285 | do_verify(td); | |
1286 | ||
cda99fa0 | 1287 | td->ts.runtime[DDIR_READ] += utime_since_now(&td->start); |
ebac4655 JA |
1288 | |
1289 | if (td->error || td->terminate) | |
1290 | break; | |
1291 | } | |
1292 | ||
36dff966 | 1293 | update_rusage_stat(td); |
cda99fa0 SSP |
1294 | td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000; |
1295 | td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000; | |
756867bd JA |
1296 | td->ts.total_run_time = mtime_since_now(&td->epoch); |
1297 | td->ts.io_bytes[0] = td->io_bytes[0]; | |
1298 | td->ts.io_bytes[1] = td->io_bytes[1]; | |
1299 | ||
d2bbb8d9 | 1300 | fio_mutex_down(writeout_mutex); |
7b9f733a | 1301 | if (td->bw_log) { |
f484470b | 1302 | if (td->o.bw_log_file) { |
7b9f733a | 1303 | finish_log_named(td, td->bw_log, |
f484470b JA |
1304 | td->o.bw_log_file, "bw"); |
1305 | } else | |
7b9f733a | 1306 | finish_log(td, td->bw_log, "bw"); |
e3cedca7 | 1307 | } |
7b9f733a | 1308 | if (td->lat_log) { |
02af0988 | 1309 | if (td->o.lat_log_file) { |
7b9f733a | 1310 | finish_log_named(td, td->lat_log, |
02af0988 JA |
1311 | td->o.lat_log_file, "lat"); |
1312 | } else | |
7b9f733a | 1313 | finish_log(td, td->lat_log, "lat"); |
02af0988 | 1314 | } |
7b9f733a | 1315 | if (td->slat_log) { |
f484470b | 1316 | if (td->o.lat_log_file) { |
7b9f733a | 1317 | finish_log_named(td, td->slat_log, |
1ca95738 | 1318 | td->o.lat_log_file, "slat"); |
f484470b | 1319 | } else |
7b9f733a | 1320 | finish_log(td, td->slat_log, "slat"); |
e3cedca7 | 1321 | } |
7b9f733a | 1322 | if (td->clat_log) { |
f484470b | 1323 | if (td->o.lat_log_file) { |
7b9f733a | 1324 | finish_log_named(td, td->clat_log, |
f484470b JA |
1325 | td->o.lat_log_file, "clat"); |
1326 | } else | |
7b9f733a | 1327 | finish_log(td, td->clat_log, "clat"); |
e3cedca7 | 1328 | } |
c8eeb9df JA |
1329 | if (td->iops_log) { |
1330 | if (td->o.iops_log_file) { | |
1331 | finish_log_named(td, td->iops_log, | |
1332 | td->o.iops_log_file, "iops"); | |
1333 | } else | |
1334 | finish_log(td, td->iops_log, "iops"); | |
1335 | } | |
1336 | ||
d2bbb8d9 | 1337 | fio_mutex_up(writeout_mutex); |
398e8c4d JA |
1338 | if (td->o.exec_postrun) |
1339 | exec_string(td->o.exec_postrun); | |
ebac4655 JA |
1340 | |
1341 | if (exitall_on_terminate) | |
cc0df00a | 1342 | fio_terminate_threads(td->groupid); |
ebac4655 JA |
1343 | |
1344 | err: | |
5bf13a5a | 1345 | if (td->error) |
4ceb30d4 | 1346 | log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error, |
5ec10eaa | 1347 | td->verror); |
b4707ed7 JA |
1348 | |
1349 | if (td->o.verify_async) | |
1350 | verify_async_exit(td); | |
1351 | ||
24ffd2c2 | 1352 | close_and_free_files(td); |
2866c82d | 1353 | close_ioengine(td); |
ebac4655 | 1354 | cleanup_io_u(td); |
6adb38a1 | 1355 | cgroup_shutdown(td, &cgroup_mnt); |
f29b25a3 | 1356 | |
d2ce18b5 JA |
1357 | if (td->o.cpumask_set) { |
1358 | int ret = fio_cpuset_exit(&td->o.cpumask); | |
1359 | ||
1360 | td_verror(td, ret, "fio_cpuset_exit"); | |
1361 | } | |
4e78e405 | 1362 | |
f29b25a3 JA |
1363 | /* |
1364 | * do this very late, it will log file closing as well | |
1365 | */ | |
1366 | if (td->o.write_iolog_file) | |
1367 | write_iolog_close(td); | |
1368 | ||
ebac4655 | 1369 | td_set_runstate(td, TD_EXITED); |
43d76807 | 1370 | return (void *) (unsigned long) td->error; |
ebac4655 JA |
1371 | } |
1372 | ||
906c8d75 JA |
1373 | /* |
1374 | * We cannot pass the td data into a forked process, so attach the td and | |
1375 | * pass it to the thread worker. | |
1376 | */ | |
a6418147 | 1377 | static int fork_main(int shmid, int offset) |
ebac4655 JA |
1378 | { |
1379 | struct thread_data *td; | |
a6418147 | 1380 | void *data, *ret; |
ebac4655 | 1381 | |
65258961 | 1382 | #ifndef __hpux |
ebac4655 JA |
1383 | data = shmat(shmid, NULL, 0); |
1384 | if (data == (void *) -1) { | |
a6418147 JA |
1385 | int __err = errno; |
1386 | ||
ebac4655 | 1387 | perror("shmat"); |
a6418147 | 1388 | return __err; |
ebac4655 | 1389 | } |
65258961 JA |
1390 | #else |
1391 | /* | |
1392 | * HP-UX inherits shm mappings? | |
1393 | */ | |
1394 | data = threads; | |
1395 | #endif | |
ebac4655 JA |
1396 | |
1397 | td = data + offset * sizeof(struct thread_data); | |
a6418147 | 1398 | ret = thread_main(td); |
ebac4655 | 1399 | shmdt(data); |
43d76807 | 1400 | return (int) (unsigned long) ret; |
ebac4655 JA |
1401 | } |
1402 | ||
906c8d75 JA |
1403 | /* |
1404 | * Run over the job map and reap the threads that have exited, if any. | |
1405 | */ | |
11e950bd JA |
1406 | static void reap_threads(unsigned int *nr_running, unsigned int *t_rate, |
1407 | unsigned int *m_rate) | |
ebac4655 | 1408 | { |
34572e28 | 1409 | struct thread_data *td; |
11e950bd JA |
1410 | unsigned int cputhreads, realthreads, pending; |
1411 | int i, status, ret; | |
ebac4655 JA |
1412 | |
1413 | /* | |
1414 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1415 | */ | |
1f809d15 | 1416 | realthreads = pending = cputhreads = 0; |
34572e28 | 1417 | for_each_td(td, i) { |
3707f45b | 1418 | int flags = 0; |
a2f77c9f | 1419 | |
84585003 JA |
1420 | /* |
1421 | * ->io_ops is NULL for a thread that has closed its | |
1422 | * io engine | |
1423 | */ | |
ba0fbe10 | 1424 | if (td->io_ops && !strcmp(td->io_ops->name, "cpuio")) |
b990b5c0 | 1425 | cputhreads++; |
1f809d15 JA |
1426 | else |
1427 | realthreads++; | |
b990b5c0 | 1428 | |
a1dedc1f AC |
1429 | if (!td->pid) { |
1430 | pending++; | |
1431 | continue; | |
1432 | } | |
1433 | if (td->runstate == TD_REAPED) | |
a2f77c9f | 1434 | continue; |
2dc1bbeb | 1435 | if (td->o.use_thread) { |
3707f45b JA |
1436 | if (td->runstate == TD_EXITED) { |
1437 | td_set_runstate(td, TD_REAPED); | |
1438 | goto reaped; | |
1439 | } | |
1440 | continue; | |
1441 | } | |
a2f77c9f JA |
1442 | |
1443 | flags = WNOHANG; | |
1444 | if (td->runstate == TD_EXITED) | |
1445 | flags = 0; | |
1446 | ||
1447 | /* | |
1448 | * check if someone quit or got killed in an unusual way | |
1449 | */ | |
1450 | ret = waitpid(td->pid, &status, flags); | |
3707f45b | 1451 | if (ret < 0) { |
a2f77c9f | 1452 | if (errno == ECHILD) { |
5921e80c JA |
1453 | log_err("fio: pid=%d disappeared %d\n", |
1454 | (int) td->pid, td->runstate); | |
a2f77c9f JA |
1455 | td_set_runstate(td, TD_REAPED); |
1456 | goto reaped; | |
1457 | } | |
1458 | perror("waitpid"); | |
1459 | } else if (ret == td->pid) { | |
1460 | if (WIFSIGNALED(status)) { | |
fab6aa71 JA |
1461 | int sig = WTERMSIG(status); |
1462 | ||
03e20d68 | 1463 | if (sig != SIGTERM) |
5ec10eaa | 1464 | log_err("fio: pid=%d, got signal=%d\n", |
5921e80c | 1465 | (int) td->pid, sig); |
fab6aa71 JA |
1466 | td_set_runstate(td, TD_REAPED); |
1467 | goto reaped; | |
1468 | } | |
a2f77c9f JA |
1469 | if (WIFEXITED(status)) { |
1470 | if (WEXITSTATUS(status) && !td->error) | |
1471 | td->error = WEXITSTATUS(status); | |
a2f77c9f | 1472 | |
a2f77c9f JA |
1473 | td_set_runstate(td, TD_REAPED); |
1474 | goto reaped; | |
a6418147 JA |
1475 | } |
1476 | } | |
ebac4655 | 1477 | |
a2f77c9f JA |
1478 | /* |
1479 | * thread is not dead, continue | |
1480 | */ | |
e48676ba | 1481 | pending++; |
a2f77c9f | 1482 | continue; |
fab6aa71 | 1483 | reaped: |
ebac4655 | 1484 | (*nr_running)--; |
581e7141 JA |
1485 | (*m_rate) -= (td->o.ratemin[0] + td->o.ratemin[1]); |
1486 | (*t_rate) -= (td->o.rate[0] + td->o.rate[1]); | |
6f88bcb0 JA |
1487 | if (!td->pid) |
1488 | pending--; | |
a2f77c9f JA |
1489 | |
1490 | if (td->error) | |
1491 | exit_value++; | |
d3eeeabc JA |
1492 | |
1493 | done_secs += mtime_since_now(&td->epoch) / 1000; | |
ebac4655 | 1494 | } |
b990b5c0 | 1495 | |
1f809d15 | 1496 | if (*nr_running == cputhreads && !pending && realthreads) |
cc0df00a | 1497 | fio_terminate_threads(TERMINATE_ALL); |
ebac4655 JA |
1498 | } |
1499 | ||
be4ecfdf JA |
1500 | static void *gtod_thread_main(void *data) |
1501 | { | |
1502 | fio_mutex_up(startup_mutex); | |
1503 | ||
1504 | /* | |
1505 | * As long as we have jobs around, update the clock. It would be nice | |
1506 | * to have some way of NOT hammering that CPU with gettimeofday(), | |
1507 | * but I'm not sure what to use outside of a simple CPU nop to relax | |
1508 | * it - we don't want to lose precision. | |
1509 | */ | |
1510 | while (threads) { | |
1511 | fio_gtod_update(); | |
1512 | nop; | |
1513 | } | |
1514 | ||
1515 | return NULL; | |
1516 | } | |
1517 | ||
1518 | static int fio_start_gtod_thread(void) | |
1519 | { | |
304a47c7 | 1520 | pthread_attr_t attr; |
f42c153d JA |
1521 | int ret; |
1522 | ||
304a47c7 VA |
1523 | pthread_attr_init(&attr); |
1524 | pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN); | |
1525 | ret = pthread_create(>od_thread, &attr, gtod_thread_main, NULL); | |
1526 | pthread_attr_destroy(&attr); | |
f42c153d JA |
1527 | if (ret) { |
1528 | log_err("Can't create gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1529 | return 1; |
1530 | } | |
f42c153d JA |
1531 | |
1532 | ret = pthread_detach(gtod_thread); | |
1533 | if (ret) { | |
1534 | log_err("Can't detatch gtod thread: %s\n", strerror(ret)); | |
be4ecfdf JA |
1535 | return 1; |
1536 | } | |
1537 | ||
29adda3c | 1538 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
be4ecfdf | 1539 | fio_mutex_down(startup_mutex); |
29adda3c | 1540 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
be4ecfdf JA |
1541 | return 0; |
1542 | } | |
1543 | ||
906c8d75 JA |
1544 | /* |
1545 | * Main function for kicking off and reaping jobs, as needed. | |
1546 | */ | |
ebac4655 JA |
1547 | static void run_threads(void) |
1548 | { | |
ebac4655 JA |
1549 | struct thread_data *td; |
1550 | unsigned long spent; | |
11e950bd | 1551 | unsigned int i, todo, nr_running, m_rate, t_rate, nr_started; |
fcb6ade2 | 1552 | |
2f9ade3c JA |
1553 | if (fio_pin_memory()) |
1554 | return; | |
ebac4655 | 1555 | |
be4ecfdf JA |
1556 | if (fio_gtod_offload && fio_start_gtod_thread()) |
1557 | return; | |
1558 | ||
cc0df00a JA |
1559 | set_sig_handlers(); |
1560 | ||
c6ae0a5b | 1561 | if (!terse_output) { |
4ceb30d4 | 1562 | log_info("Starting "); |
9cedf167 | 1563 | if (nr_thread) |
4ceb30d4 | 1564 | log_info("%d thread%s", nr_thread, |
5ec10eaa | 1565 | nr_thread > 1 ? "s" : ""); |
9cedf167 JA |
1566 | if (nr_process) { |
1567 | if (nr_thread) | |
cc604381 | 1568 | log_info(" and "); |
4ceb30d4 | 1569 | log_info("%d process%s", nr_process, |
5ec10eaa | 1570 | nr_process > 1 ? "es" : ""); |
9cedf167 | 1571 | } |
4ceb30d4 | 1572 | log_info("\n"); |
c6ae0a5b JA |
1573 | fflush(stdout); |
1574 | } | |
c04f7ec3 | 1575 | |
ebac4655 JA |
1576 | todo = thread_number; |
1577 | nr_running = 0; | |
1578 | nr_started = 0; | |
1579 | m_rate = t_rate = 0; | |
1580 | ||
34572e28 | 1581 | for_each_td(td, i) { |
263e529f | 1582 | print_status_init(td->thread_number - 1); |
ebac4655 | 1583 | |
067430bb | 1584 | if (!td->o.create_serialize) |
ebac4655 JA |
1585 | continue; |
1586 | ||
1587 | /* | |
1588 | * do file setup here so it happens sequentially, | |
1589 | * we don't want X number of threads getting their | |
1590 | * client data interspersed on disk | |
1591 | */ | |
53cdc686 | 1592 | if (setup_files(td)) { |
5bf13a5a JA |
1593 | exit_value++; |
1594 | if (td->error) | |
5921e80c JA |
1595 | log_err("fio: pid=%d, err=%d/%s\n", |
1596 | (int) td->pid, td->error, td->verror); | |
ebac4655 JA |
1597 | td_set_runstate(td, TD_REAPED); |
1598 | todo--; | |
c69e8875 JA |
1599 | } else { |
1600 | struct fio_file *f; | |
2b13e716 | 1601 | unsigned int j; |
c69e8875 JA |
1602 | |
1603 | /* | |
1604 | * for sharing to work, each job must always open | |
1605 | * its own files. so close them, if we opened them | |
1606 | * for creation | |
1607 | */ | |
2b13e716 | 1608 | for_each_file(td, f, j) { |
22a57ba8 JA |
1609 | if (fio_file_open(f)) |
1610 | td_io_close_file(td, f); | |
22a57ba8 | 1611 | } |
5ec10eaa | 1612 | } |
ebac4655 JA |
1613 | } |
1614 | ||
a2f77c9f JA |
1615 | set_genesis_time(); |
1616 | ||
ebac4655 | 1617 | while (todo) { |
fca70358 | 1618 | struct thread_data *map[REAL_MAX_JOBS]; |
75154845 JA |
1619 | struct timeval this_start; |
1620 | int this_jobs = 0, left; | |
1621 | ||
ebac4655 JA |
1622 | /* |
1623 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1624 | */ | |
34572e28 | 1625 | for_each_td(td, i) { |
ebac4655 JA |
1626 | if (td->runstate != TD_NOT_CREATED) |
1627 | continue; | |
1628 | ||
1629 | /* | |
1630 | * never got a chance to start, killed by other | |
1631 | * thread for some reason | |
1632 | */ | |
1633 | if (td->terminate) { | |
1634 | todo--; | |
1635 | continue; | |
1636 | } | |
1637 | ||
2dc1bbeb | 1638 | if (td->o.start_delay) { |
263e529f | 1639 | spent = mtime_since_genesis(); |
ebac4655 | 1640 | |
2dc1bbeb | 1641 | if (td->o.start_delay * 1000 > spent) |
ebac4655 JA |
1642 | continue; |
1643 | } | |
1644 | ||
6f88bcb0 JA |
1645 | if (td->o.stonewall && (nr_started || nr_running)) { |
1646 | dprint(FD_PROCESS, "%s: stonewall wait\n", | |
1647 | td->o.name); | |
ebac4655 | 1648 | break; |
6f88bcb0 | 1649 | } |
ebac4655 | 1650 | |
067430bb JA |
1651 | init_disk_util(td); |
1652 | ||
75154845 JA |
1653 | /* |
1654 | * Set state to created. Thread will transition | |
1655 | * to TD_INITIALIZED when it's done setting up. | |
1656 | */ | |
ebac4655 | 1657 | td_set_runstate(td, TD_CREATED); |
75154845 | 1658 | map[this_jobs++] = td; |
ebac4655 JA |
1659 | nr_started++; |
1660 | ||
2dc1bbeb | 1661 | if (td->o.use_thread) { |
f42c153d JA |
1662 | int ret; |
1663 | ||
ee56ad50 | 1664 | dprint(FD_PROCESS, "will pthread_create\n"); |
f42c153d JA |
1665 | ret = pthread_create(&td->thread, NULL, |
1666 | thread_main, td); | |
1667 | if (ret) { | |
1668 | log_err("pthread_create: %s\n", | |
1669 | strerror(ret)); | |
ebac4655 | 1670 | nr_started--; |
c8bb6faf | 1671 | break; |
ebac4655 | 1672 | } |
f42c153d JA |
1673 | ret = pthread_detach(td->thread); |
1674 | if (ret) | |
1675 | log_err("pthread_detach: %s", | |
1676 | strerror(ret)); | |
ebac4655 | 1677 | } else { |
5e1d306e | 1678 | pid_t pid; |
ee56ad50 | 1679 | dprint(FD_PROCESS, "will fork\n"); |
5e1d306e JA |
1680 | pid = fork(); |
1681 | if (!pid) { | |
a6418147 JA |
1682 | int ret = fork_main(shm_id, i); |
1683 | ||
5e1d306e JA |
1684 | _exit(ret); |
1685 | } else if (i == fio_debug_jobno) | |
1686 | *fio_debug_jobp = pid; | |
ebac4655 | 1687 | } |
29adda3c | 1688 | dprint(FD_MUTEX, "wait on startup_mutex\n"); |
656b1393 JA |
1689 | if (fio_mutex_down_timeout(startup_mutex, 10)) { |
1690 | log_err("fio: job startup hung? exiting.\n"); | |
cc0df00a | 1691 | fio_terminate_threads(TERMINATE_ALL); |
656b1393 JA |
1692 | fio_abort = 1; |
1693 | nr_started--; | |
1694 | break; | |
1695 | } | |
29adda3c | 1696 | dprint(FD_MUTEX, "done waiting on startup_mutex\n"); |
ebac4655 JA |
1697 | } |
1698 | ||
1699 | /* | |
75154845 JA |
1700 | * Wait for the started threads to transition to |
1701 | * TD_INITIALIZED. | |
ebac4655 | 1702 | */ |
02bcaa8c | 1703 | fio_gettime(&this_start, NULL); |
75154845 | 1704 | left = this_jobs; |
6ce15a32 | 1705 | while (left && !fio_abort) { |
75154845 JA |
1706 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) |
1707 | break; | |
1708 | ||
1709 | usleep(100000); | |
1710 | ||
1711 | for (i = 0; i < this_jobs; i++) { | |
1712 | td = map[i]; | |
1713 | if (!td) | |
1714 | continue; | |
b6f4d880 | 1715 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1716 | map[i] = NULL; |
1717 | left--; | |
b6f4d880 JA |
1718 | } else if (td->runstate >= TD_EXITED) { |
1719 | map[i] = NULL; | |
1720 | left--; | |
1721 | todo--; | |
1722 | nr_running++; /* work-around... */ | |
75154845 JA |
1723 | } |
1724 | } | |
1725 | } | |
1726 | ||
1727 | if (left) { | |
3b70d7e5 | 1728 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1729 | for (i = 0; i < this_jobs; i++) { |
1730 | td = map[i]; | |
1731 | if (!td) | |
1732 | continue; | |
1733 | kill(td->pid, SIGTERM); | |
1734 | } | |
1735 | break; | |
1736 | } | |
1737 | ||
1738 | /* | |
b6f4d880 | 1739 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1740 | */ |
34572e28 | 1741 | for_each_td(td, i) { |
75154845 | 1742 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1743 | continue; |
1744 | ||
b29ee5b3 JA |
1745 | if (in_ramp_time(td)) |
1746 | td_set_runstate(td, TD_RAMP); | |
1747 | else | |
1748 | td_set_runstate(td, TD_RUNNING); | |
ebac4655 JA |
1749 | nr_running++; |
1750 | nr_started--; | |
581e7141 JA |
1751 | m_rate += td->o.ratemin[0] + td->o.ratemin[1]; |
1752 | t_rate += td->o.rate[0] + td->o.rate[1]; | |
75154845 | 1753 | todo--; |
cdd18ad8 | 1754 | fio_mutex_up(td->mutex); |
ebac4655 JA |
1755 | } |
1756 | ||
1757 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1758 | ||
602d6ff9 JA |
1759 | if (todo) { |
1760 | if (is_backend) | |
1761 | fio_server_idle_loop(); | |
1762 | else | |
1763 | usleep(100000); | |
1764 | } | |
ebac4655 JA |
1765 | } |
1766 | ||
1767 | while (nr_running) { | |
1768 | reap_threads(&nr_running, &t_rate, &m_rate); | |
cc0df00a JA |
1769 | |
1770 | if (is_backend) | |
1771 | fio_server_idle_loop(); | |
70e0c316 JA |
1772 | else |
1773 | usleep(10000); | |
ebac4655 JA |
1774 | } |
1775 | ||
1776 | update_io_ticks(); | |
2f9ade3c | 1777 | fio_unpin_memory(); |
ebac4655 JA |
1778 | } |
1779 | ||
50d16976 | 1780 | int exec_run(void) |
ebac4655 | 1781 | { |
7e356b2d JA |
1782 | struct thread_data *td; |
1783 | int i; | |
1784 | ||
a37f69b7 | 1785 | if (nr_clients) |
37db14fe | 1786 | return fio_handle_clients(); |
7e356b2d JA |
1787 | if (exec_profile) { |
1788 | if (load_profile(exec_profile)) | |
1789 | return 1; | |
1790 | free(exec_profile); | |
1791 | exec_profile = NULL; | |
1792 | } | |
74929ac2 JA |
1793 | if (!thread_number) |
1794 | return 0; | |
1795 | ||
bb3884d8 JA |
1796 | if (write_bw_log) { |
1797 | setup_log(&agg_io_log[DDIR_READ]); | |
1798 | setup_log(&agg_io_log[DDIR_WRITE]); | |
1799 | } | |
1800 | ||
cdd18ad8 | 1801 | startup_mutex = fio_mutex_init(0); |
03e20d68 BC |
1802 | if (startup_mutex == NULL) |
1803 | return 1; | |
d2bbb8d9 | 1804 | writeout_mutex = fio_mutex_init(1); |
03e20d68 BC |
1805 | if (writeout_mutex == NULL) |
1806 | return 1; | |
07739b57 | 1807 | |
a2f77c9f | 1808 | set_genesis_time(); |
7985701b | 1809 | create_disk_util_thread(); |
ebac4655 | 1810 | |
dae5341c JA |
1811 | cgroup_list = smalloc(sizeof(*cgroup_list)); |
1812 | INIT_FLIST_HEAD(cgroup_list); | |
1813 | ||
ebac4655 | 1814 | run_threads(); |
6ce15a32 | 1815 | |
bb3884d8 | 1816 | if (!fio_abort) { |
6ce15a32 | 1817 | show_run_stats(); |
bb3884d8 | 1818 | if (write_bw_log) { |
5ec10eaa JA |
1819 | __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log"); |
1820 | __finish_log(agg_io_log[DDIR_WRITE], | |
1821 | "agg-write_bw.log"); | |
bb3884d8 JA |
1822 | } |
1823 | } | |
ebac4655 | 1824 | |
7e356b2d JA |
1825 | for_each_td(td, i) |
1826 | fio_options_free(td); | |
1827 | ||
dae5341c JA |
1828 | cgroup_kill(cgroup_list); |
1829 | sfree(cgroup_list); | |
61eb313e | 1830 | sfree(cgroup_mnt); |
39f22027 | 1831 | |
cdd18ad8 | 1832 | fio_mutex_remove(startup_mutex); |
d2bbb8d9 | 1833 | fio_mutex_remove(writeout_mutex); |
437c9b71 | 1834 | return exit_value; |
ebac4655 | 1835 | } |
50d16976 JA |
1836 | |
1837 | void reset_fio_state(void) | |
1838 | { | |
1839 | groupid = 0; | |
1840 | thread_number = 0; | |
1841 | nr_process = 0; | |
1842 | nr_thread = 0; | |
1843 | done_secs = 0; | |
1844 | } | |
1845 | ||
0ad5edcc JA |
1846 | static int endian_check(void) |
1847 | { | |
1848 | union { | |
1849 | uint8_t c[8]; | |
1850 | uint64_t v; | |
1851 | } u; | |
1852 | int le = 0, be = 0; | |
1853 | ||
1854 | u.v = 0x12; | |
1855 | if (u.c[7] == 0x12) | |
1856 | be = 1; | |
1857 | else if (u.c[0] == 0x12) | |
1858 | le = 1; | |
1859 | ||
1860 | #if defined(FIO_LITTLE_ENDIAN) | |
1861 | if (be) | |
1862 | return 1; | |
1863 | #elif defined(FIO_BIG_ENDIAN) | |
1864 | if (le) | |
1865 | return 1; | |
1866 | #else | |
1867 | return 1; | |
1868 | #endif | |
1869 | ||
1870 | if (!le && !be) | |
1871 | return 1; | |
1872 | ||
1873 | return 0; | |
1874 | } | |
1875 | ||
50d16976 JA |
1876 | int main(int argc, char *argv[], char *envp[]) |
1877 | { | |
1878 | long ps; | |
1879 | ||
0ad5edcc JA |
1880 | if (endian_check()) { |
1881 | log_err("fio: endianness settings appear wrong.\n"); | |
1882 | log_err("fio: please report this to fio@vger.kernel.org\n"); | |
1883 | return 1; | |
1884 | } | |
1885 | ||
50d16976 JA |
1886 | arch_init(envp); |
1887 | ||
1888 | sinit(); | |
1889 | ||
1890 | /* | |
1891 | * We need locale for number printing, if it isn't set then just | |
1892 | * go with the US format. | |
1893 | */ | |
1894 | if (!getenv("LC_NUMERIC")) | |
1895 | setlocale(LC_NUMERIC, "en_US"); | |
1896 | ||
1897 | ps = sysconf(_SC_PAGESIZE); | |
1898 | if (ps < 0) { | |
1899 | log_err("Failed to get page size\n"); | |
1900 | return 1; | |
1901 | } | |
1902 | ||
1903 | page_size = ps; | |
1904 | page_mask = ps - 1; | |
1905 | ||
1906 | fio_keywords_init(); | |
1907 | ||
1908 | if (parse_options(argc, argv)) | |
1909 | return 1; | |
1910 | ||
1911 | return exec_run(); | |
1912 | } |