Commit | Line | Data |
---|---|---|
ebac4655 JA |
1 | /* |
2 | * fio - the flexible io tester | |
3 | * | |
4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program; if not, write to the Free Software | |
18 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
19 | * | |
20 | */ | |
ebac4655 JA |
21 | #include <unistd.h> |
22 | #include <fcntl.h> | |
23 | #include <string.h> | |
ebac4655 JA |
24 | #include <signal.h> |
25 | #include <time.h> | |
ebac4655 | 26 | #include <assert.h> |
ebac4655 JA |
27 | #include <sys/stat.h> |
28 | #include <sys/wait.h> | |
29 | #include <sys/ipc.h> | |
30 | #include <sys/shm.h> | |
31 | #include <sys/ioctl.h> | |
32 | #include <sys/mman.h> | |
33 | ||
34 | #include "fio.h" | |
35 | #include "os.h" | |
36 | ||
37 | #define MASK (4095) | |
38 | ||
39 | #define ALIGN(buf) (char *) (((unsigned long) (buf) + MASK) & ~(MASK)) | |
40 | ||
41 | int groupid = 0; | |
42 | int thread_number = 0; | |
43 | static char run_str[MAX_JOBS + 1]; | |
44 | int shm_id = 0; | |
5289b847 | 45 | static struct timeval genesis; |
ebac4655 | 46 | |
ebac4655 JA |
47 | static void print_thread_status(void); |
48 | ||
c04f7ec3 JA |
49 | extern unsigned long long mlock_size; |
50 | ||
ebac4655 JA |
51 | /* |
52 | * thread life cycle | |
53 | */ | |
54 | enum { | |
55 | TD_NOT_CREATED = 0, | |
56 | TD_CREATED, | |
75154845 | 57 | TD_INITIALIZED, |
ebac4655 JA |
58 | TD_RUNNING, |
59 | TD_VERIFYING, | |
5853e5a8 | 60 | TD_FSYNCING, |
ebac4655 JA |
61 | TD_EXITED, |
62 | TD_REAPED, | |
63 | }; | |
64 | ||
3d60d1ed | 65 | #define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync)) |
ebac4655 | 66 | |
bbfd6b00 | 67 | static volatile int startup_sem; |
ebac4655 JA |
68 | |
69 | #define TERMINATE_ALL (-1) | |
75154845 | 70 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 JA |
71 | |
72 | static void terminate_threads(int group_id) | |
73 | { | |
74 | int i; | |
75 | ||
76 | for (i = 0; i < thread_number; i++) { | |
77 | struct thread_data *td = &threads[i]; | |
78 | ||
79 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { | |
80 | td->terminate = 1; | |
81 | td->start_delay = 0; | |
82 | } | |
83 | } | |
84 | } | |
85 | ||
86 | static void sig_handler(int sig) | |
87 | { | |
88 | switch (sig) { | |
89 | case SIGALRM: | |
90 | update_io_ticks(); | |
91 | disk_util_timer_arm(); | |
92 | print_thread_status(); | |
93 | break; | |
94 | default: | |
95 | printf("\nfio: terminating on signal\n"); | |
96 | fflush(stdout); | |
97 | terminate_threads(TERMINATE_ALL); | |
98 | break; | |
99 | } | |
100 | } | |
101 | ||
ebac4655 JA |
102 | static int random_map_free(struct thread_data *td, unsigned long long block) |
103 | { | |
104 | unsigned int idx = RAND_MAP_IDX(td, block); | |
105 | unsigned int bit = RAND_MAP_BIT(td, block); | |
106 | ||
107 | return (td->file_map[idx] & (1UL << bit)) == 0; | |
108 | } | |
109 | ||
110 | static int get_next_free_block(struct thread_data *td, unsigned long long *b) | |
111 | { | |
112 | int i; | |
113 | ||
114 | *b = 0; | |
115 | i = 0; | |
116 | while ((*b) * td->min_bs < td->io_size) { | |
117 | if (td->file_map[i] != -1UL) { | |
118 | *b += ffz(td->file_map[i]); | |
119 | return 0; | |
120 | } | |
121 | ||
122 | *b += BLOCKS_PER_MAP; | |
123 | i++; | |
124 | } | |
125 | ||
126 | return 1; | |
127 | } | |
128 | ||
129 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) | |
130 | { | |
200bc855 | 131 | unsigned long long block = io_u->offset / (unsigned long long) td->min_bs; |
ebac4655 JA |
132 | unsigned int blocks = 0; |
133 | ||
134 | while (blocks < (io_u->buflen / td->min_bs)) { | |
135 | unsigned int idx, bit; | |
136 | ||
137 | if (!random_map_free(td, block)) | |
138 | break; | |
139 | ||
140 | idx = RAND_MAP_IDX(td, block); | |
141 | bit = RAND_MAP_BIT(td, block); | |
142 | ||
143 | assert(idx < td->num_maps); | |
144 | ||
145 | td->file_map[idx] |= (1UL << bit); | |
146 | block++; | |
147 | blocks++; | |
148 | } | |
149 | ||
150 | if ((blocks * td->min_bs) < io_u->buflen) | |
151 | io_u->buflen = blocks * td->min_bs; | |
152 | } | |
153 | ||
20dc95c4 JA |
154 | static int get_next_offset(struct thread_data *td, unsigned long long *offset) |
155 | { | |
156 | unsigned long long b, rb; | |
157 | long r; | |
158 | ||
159 | if (!td->sequential) { | |
085227ad | 160 | unsigned long long max_blocks = td->io_size / td->min_bs; |
20dc95c4 JA |
161 | int loops = 50; |
162 | ||
163 | do { | |
6dfd46b9 | 164 | r = os_random_long(&td->random_state); |
085227ad | 165 | b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0)); |
20dc95c4 JA |
166 | rb = b + (td->file_offset / td->min_bs); |
167 | loops--; | |
168 | } while (!random_map_free(td, rb) && loops); | |
169 | ||
170 | if (!loops) { | |
171 | if (get_next_free_block(td, &b)) | |
172 | return 1; | |
173 | } | |
174 | } else | |
175 | b = td->last_pos / td->min_bs; | |
176 | ||
177 | *offset = (b * td->min_bs) + td->file_offset; | |
838a3cd3 | 178 | if (*offset > td->real_file_size) |
20dc95c4 JA |
179 | return 1; |
180 | ||
181 | return 0; | |
182 | } | |
183 | ||
184 | static unsigned int get_next_buflen(struct thread_data *td) | |
185 | { | |
186 | unsigned int buflen; | |
187 | long r; | |
188 | ||
189 | if (td->min_bs == td->max_bs) | |
190 | buflen = td->min_bs; | |
191 | else { | |
6dfd46b9 | 192 | r = os_random_long(&td->bsrange_state); |
20dc95c4 JA |
193 | buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0)); |
194 | buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1); | |
195 | } | |
196 | ||
197 | if (buflen > td->io_size - td->this_io_bytes[td->ddir]) | |
198 | buflen = td->io_size - td->this_io_bytes[td->ddir]; | |
199 | ||
200 | return buflen; | |
201 | } | |
202 | ||
ebac4655 JA |
203 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
204 | { | |
205 | unsigned long spent; | |
206 | unsigned long rate; | |
207 | int ddir = td->ddir; | |
208 | ||
209 | /* | |
210 | * allow a 2 second settle period in the beginning | |
211 | */ | |
212 | if (mtime_since(&td->start, now) < 2000) | |
213 | return 0; | |
214 | ||
215 | /* | |
216 | * if rate blocks is set, sample is running | |
217 | */ | |
218 | if (td->rate_bytes) { | |
219 | spent = mtime_since(&td->lastrate, now); | |
220 | if (spent < td->ratecycle) | |
221 | return 0; | |
222 | ||
223 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
224 | if (rate < td->ratemin) { | |
eb8bbf48 | 225 | fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
226 | if (rate_quit) |
227 | terminate_threads(td->groupid); | |
228 | return 1; | |
229 | } | |
230 | } | |
231 | ||
232 | td->rate_bytes = td->this_io_bytes[ddir]; | |
233 | memcpy(&td->lastrate, now, sizeof(*now)); | |
234 | return 0; | |
235 | } | |
236 | ||
237 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
238 | { | |
239 | if (!td->timeout) | |
240 | return 0; | |
241 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
242 | return 1; | |
243 | ||
244 | return 0; | |
245 | } | |
246 | ||
247 | static void fill_random_bytes(struct thread_data *td, | |
248 | unsigned char *p, unsigned int len) | |
249 | { | |
250 | unsigned int todo; | |
251 | double r; | |
252 | ||
253 | while (len) { | |
6dfd46b9 | 254 | r = os_random_double(&td->verify_state); |
ebac4655 JA |
255 | |
256 | /* | |
257 | * lrand48_r seems to be broken and only fill the bottom | |
258 | * 32-bits, even on 64-bit archs with 64-bit longs | |
259 | */ | |
260 | todo = sizeof(r); | |
261 | if (todo > len) | |
262 | todo = len; | |
263 | ||
264 | memcpy(p, &r, todo); | |
265 | ||
266 | len -= todo; | |
267 | p += todo; | |
268 | } | |
269 | } | |
270 | ||
271 | static void hexdump(void *buffer, int len) | |
272 | { | |
273 | unsigned char *p = buffer; | |
274 | int i; | |
275 | ||
276 | for (i = 0; i < len; i++) | |
eb8bbf48 JA |
277 | fprintf(f_out, "%02x", p[i]); |
278 | fprintf(f_out, "\n"); | |
ebac4655 JA |
279 | } |
280 | ||
281 | static int verify_io_u_crc32(struct verify_header *hdr, struct io_u *io_u) | |
282 | { | |
283 | unsigned char *p = (unsigned char *) io_u->buf; | |
284 | unsigned long c; | |
ebac4655 JA |
285 | |
286 | p += sizeof(*hdr); | |
287 | c = crc32(p, hdr->len - sizeof(*hdr)); | |
ebac4655 | 288 | |
22f78b32 | 289 | if (c != hdr->crc32) { |
eb8bbf48 JA |
290 | fprintf(f_err, "crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen); |
291 | fprintf(f_err, "crc32: wanted %lx, got %lx\n", hdr->crc32, c); | |
22f78b32 | 292 | return 1; |
ebac4655 JA |
293 | } |
294 | ||
22f78b32 | 295 | return 0; |
ebac4655 JA |
296 | } |
297 | ||
298 | static int verify_io_u_md5(struct verify_header *hdr, struct io_u *io_u) | |
299 | { | |
300 | unsigned char *p = (unsigned char *) io_u->buf; | |
301 | struct md5_ctx md5_ctx; | |
ebac4655 JA |
302 | |
303 | memset(&md5_ctx, 0, sizeof(md5_ctx)); | |
304 | p += sizeof(*hdr); | |
305 | md5_update(&md5_ctx, p, hdr->len - sizeof(*hdr)); | |
306 | ||
22f78b32 | 307 | if (memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash))) { |
eb8bbf48 | 308 | fprintf(f_err, "md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen); |
ebac4655 JA |
309 | hexdump(hdr->md5_digest, sizeof(hdr->md5_digest)); |
310 | hexdump(md5_ctx.hash, sizeof(md5_ctx.hash)); | |
22f78b32 | 311 | return 1; |
ebac4655 JA |
312 | } |
313 | ||
22f78b32 | 314 | return 0; |
ebac4655 JA |
315 | } |
316 | ||
317 | static int verify_io_u(struct io_u *io_u) | |
318 | { | |
319 | struct verify_header *hdr = (struct verify_header *) io_u->buf; | |
320 | int ret; | |
321 | ||
322 | if (hdr->fio_magic != FIO_HDR_MAGIC) | |
323 | return 1; | |
324 | ||
325 | if (hdr->verify_type == VERIFY_MD5) | |
326 | ret = verify_io_u_md5(hdr, io_u); | |
327 | else if (hdr->verify_type == VERIFY_CRC32) | |
328 | ret = verify_io_u_crc32(hdr, io_u); | |
329 | else { | |
eb8bbf48 | 330 | fprintf(f_err, "Bad verify type %d\n", hdr->verify_type); |
ebac4655 JA |
331 | ret = 1; |
332 | } | |
333 | ||
334 | return ret; | |
335 | } | |
336 | ||
337 | static void fill_crc32(struct verify_header *hdr, void *p, unsigned int len) | |
338 | { | |
339 | hdr->crc32 = crc32(p, len); | |
340 | } | |
341 | ||
342 | static void fill_md5(struct verify_header *hdr, void *p, unsigned int len) | |
343 | { | |
344 | struct md5_ctx md5_ctx; | |
345 | ||
346 | memset(&md5_ctx, 0, sizeof(md5_ctx)); | |
347 | md5_update(&md5_ctx, p, len); | |
348 | memcpy(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash)); | |
349 | } | |
350 | ||
a6ccc7be | 351 | static int get_rw_ddir(struct thread_data *td) |
3d60d1ed | 352 | { |
a6ccc7be JA |
353 | if (td_rw(td)) { |
354 | struct timeval now; | |
355 | unsigned long elapsed; | |
3d60d1ed | 356 | |
a6ccc7be JA |
357 | gettimeofday(&now, NULL); |
358 | elapsed = mtime_since_now(&td->rwmix_switch); | |
3d60d1ed | 359 | |
a6ccc7be JA |
360 | /* |
361 | * Check if it's time to seed a new data direction. | |
362 | */ | |
363 | if (elapsed >= td->rwmixcycle) { | |
c1ee2ca4 | 364 | int v; |
a6ccc7be | 365 | long r; |
3d60d1ed | 366 | |
c1ee2ca4 JA |
367 | r = os_random_long(&td->rwmix_state); |
368 | v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); | |
a6ccc7be JA |
369 | if (v < td->rwmixread) |
370 | td->rwmix_ddir = DDIR_READ; | |
371 | else | |
372 | td->rwmix_ddir = DDIR_WRITE; | |
373 | memcpy(&td->rwmix_switch, &now, sizeof(now)); | |
374 | } | |
375 | return td->rwmix_ddir; | |
376 | } else if (td_read(td)) | |
3d60d1ed JA |
377 | return DDIR_READ; |
378 | else | |
379 | return DDIR_WRITE; | |
380 | } | |
381 | ||
ebac4655 JA |
382 | /* |
383 | * fill body of io_u->buf with random data and add a header with the | |
22f78b32 | 384 | * crc32 or md5 sum of that data. |
ebac4655 JA |
385 | */ |
386 | static void populate_io_u(struct thread_data *td, struct io_u *io_u) | |
387 | { | |
388 | unsigned char *p = (unsigned char *) io_u->buf; | |
389 | struct verify_header hdr; | |
390 | ||
391 | hdr.fio_magic = FIO_HDR_MAGIC; | |
392 | hdr.len = io_u->buflen; | |
393 | p += sizeof(hdr); | |
394 | fill_random_bytes(td, p, io_u->buflen - sizeof(hdr)); | |
395 | ||
396 | if (td->verify == VERIFY_MD5) { | |
397 | fill_md5(&hdr, p, io_u->buflen - sizeof(hdr)); | |
398 | hdr.verify_type = VERIFY_MD5; | |
399 | } else { | |
400 | fill_crc32(&hdr, p, io_u->buflen - sizeof(hdr)); | |
401 | hdr.verify_type = VERIFY_CRC32; | |
402 | } | |
403 | ||
404 | memcpy(io_u->buf, &hdr, sizeof(hdr)); | |
405 | } | |
406 | ||
aea47d44 | 407 | static int td_io_prep(struct thread_data *td, struct io_u *io_u) |
20dc95c4 | 408 | { |
20dc95c4 JA |
409 | if (td->io_prep && td->io_prep(td, io_u)) |
410 | return 1; | |
411 | ||
412 | return 0; | |
413 | } | |
414 | ||
b1ff3403 | 415 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
ebac4655 JA |
416 | { |
417 | list_del(&io_u->list); | |
418 | list_add(&io_u->list, &td->io_u_freelist); | |
419 | td->cur_depth--; | |
420 | } | |
421 | ||
843a7413 JA |
422 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
423 | { | |
424 | /* | |
425 | * If using an iolog, grab next piece if any available. | |
426 | */ | |
427 | if (td->read_iolog) | |
428 | return read_iolog_get(td, io_u); | |
429 | ||
aea47d44 JA |
430 | /* |
431 | * No log, let the seq/rand engine retrieve the next position. | |
432 | */ | |
433 | if (!get_next_offset(td, &io_u->offset)) { | |
434 | io_u->buflen = get_next_buflen(td); | |
435 | ||
436 | if (io_u->buflen) { | |
437 | io_u->ddir = get_rw_ddir(td); | |
843a7413 JA |
438 | |
439 | /* | |
440 | * If using a write iolog, store this entry. | |
441 | */ | |
442 | if (td->write_iolog) | |
443 | write_iolog_put(td, io_u); | |
444 | ||
aea47d44 JA |
445 | return 0; |
446 | } | |
447 | } | |
448 | ||
449 | return 1; | |
450 | } | |
451 | ||
22f78b32 | 452 | #define queue_full(td) list_empty(&(td)->io_u_freelist) |
ebac4655 | 453 | |
b1ff3403 | 454 | struct io_u *__get_io_u(struct thread_data *td) |
ebac4655 | 455 | { |
22f78b32 | 456 | struct io_u *io_u = NULL; |
ebac4655 | 457 | |
22f78b32 JA |
458 | if (!queue_full(td)) { |
459 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); | |
460 | ||
461 | io_u->error = 0; | |
462 | io_u->resid = 0; | |
463 | list_del(&io_u->list); | |
464 | list_add(&io_u->list, &td->io_u_busylist); | |
465 | td->cur_depth++; | |
466 | } | |
ebac4655 | 467 | |
ebac4655 JA |
468 | return io_u; |
469 | } | |
470 | ||
ebac4655 JA |
471 | static struct io_u *get_io_u(struct thread_data *td) |
472 | { | |
473 | struct io_u *io_u; | |
474 | ||
475 | io_u = __get_io_u(td); | |
476 | if (!io_u) | |
477 | return NULL; | |
478 | ||
20dc95c4 JA |
479 | if (td->zone_bytes >= td->zone_size) { |
480 | td->zone_bytes = 0; | |
481 | td->last_pos += td->zone_skip; | |
482 | } | |
483 | ||
aea47d44 | 484 | if (fill_io_u(td, io_u)) { |
ebac4655 JA |
485 | put_io_u(td, io_u); |
486 | return NULL; | |
487 | } | |
488 | ||
838a3cd3 JA |
489 | if (io_u->buflen + io_u->offset > td->real_file_size) |
490 | io_u->buflen = td->real_file_size - io_u->offset; | |
ebac4655 JA |
491 | |
492 | if (!io_u->buflen) { | |
493 | put_io_u(td, io_u); | |
494 | return NULL; | |
495 | } | |
496 | ||
843a7413 | 497 | if (!td->read_iolog && !td->sequential) |
ebac4655 JA |
498 | mark_random_map(td, io_u); |
499 | ||
20dc95c4 | 500 | td->last_pos += io_u->buflen; |
ebac4655 JA |
501 | |
502 | if (td->verify != VERIFY_NONE) | |
503 | populate_io_u(td, io_u); | |
504 | ||
aea47d44 | 505 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
506 | put_io_u(td, io_u); |
507 | return NULL; | |
508 | } | |
509 | ||
510 | gettimeofday(&io_u->start_time, NULL); | |
511 | return io_u; | |
512 | } | |
513 | ||
514 | static inline void td_set_runstate(struct thread_data *td, int runstate) | |
515 | { | |
ebac4655 JA |
516 | td->runstate = runstate; |
517 | } | |
518 | ||
aea47d44 | 519 | static int get_next_verify(struct thread_data *td, struct io_u *io_u) |
ebac4655 JA |
520 | { |
521 | struct io_piece *ipo; | |
522 | ||
22f78b32 JA |
523 | if (!list_empty(&td->io_hist_list)) { |
524 | ipo = list_entry(td->io_hist_list.next, struct io_piece, list); | |
ebac4655 | 525 | |
22f78b32 | 526 | list_del(&ipo->list); |
ebac4655 | 527 | |
22f78b32 JA |
528 | io_u->offset = ipo->offset; |
529 | io_u->buflen = ipo->len; | |
530 | io_u->ddir = DDIR_READ; | |
531 | free(ipo); | |
532 | return 0; | |
533 | } | |
534 | ||
535 | return 1; | |
ebac4655 JA |
536 | } |
537 | ||
ebac4655 JA |
538 | static int sync_td(struct thread_data *td) |
539 | { | |
540 | if (td->io_sync) | |
541 | return td->io_sync(td); | |
542 | ||
543 | return 0; | |
544 | } | |
545 | ||
546 | static int io_u_getevents(struct thread_data *td, int min, int max, | |
547 | struct timespec *t) | |
548 | { | |
549 | return td->io_getevents(td, min, max, t); | |
550 | } | |
551 | ||
552 | static int io_u_queue(struct thread_data *td, struct io_u *io_u) | |
553 | { | |
554 | gettimeofday(&io_u->issue_time, NULL); | |
555 | ||
556 | return td->io_queue(td, io_u); | |
557 | } | |
558 | ||
559 | #define iocb_time(iocb) ((unsigned long) (iocb)->data) | |
560 | ||
561 | static void io_completed(struct thread_data *td, struct io_u *io_u, | |
562 | struct io_completion_data *icd) | |
563 | { | |
564 | struct timeval e; | |
565 | unsigned long msec; | |
566 | ||
567 | gettimeofday(&e, NULL); | |
568 | ||
569 | if (!io_u->error) { | |
20dc95c4 | 570 | unsigned int bytes = io_u->buflen - io_u->resid; |
3d60d1ed | 571 | const int idx = io_u->ddir; |
ebac4655 JA |
572 | |
573 | td->io_blocks[idx]++; | |
20dc95c4 JA |
574 | td->io_bytes[idx] += bytes; |
575 | td->zone_bytes += bytes; | |
576 | td->this_io_bytes[idx] += bytes; | |
ebac4655 JA |
577 | |
578 | msec = mtime_since(&io_u->issue_time, &e); | |
579 | ||
3d60d1ed JA |
580 | add_clat_sample(td, idx, msec); |
581 | add_bw_sample(td, idx); | |
ebac4655 | 582 | |
3d60d1ed | 583 | if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) |
ebac4655 JA |
584 | log_io_piece(td, io_u); |
585 | ||
20dc95c4 | 586 | icd->bytes_done[idx] += bytes; |
ebac4655 JA |
587 | } else |
588 | icd->error = io_u->error; | |
589 | } | |
590 | ||
591 | static void ios_completed(struct thread_data *td,struct io_completion_data *icd) | |
592 | { | |
593 | struct io_u *io_u; | |
594 | int i; | |
595 | ||
596 | icd->error = 0; | |
597 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
598 | ||
599 | for (i = 0; i < icd->nr; i++) { | |
600 | io_u = td->io_event(td, i); | |
601 | ||
602 | io_completed(td, io_u, icd); | |
603 | put_io_u(td, io_u); | |
604 | } | |
605 | } | |
606 | ||
607 | static void cleanup_pending_aio(struct thread_data *td) | |
608 | { | |
609 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
610 | struct list_head *entry, *n; | |
611 | struct io_completion_data icd; | |
612 | struct io_u *io_u; | |
613 | int r; | |
614 | ||
615 | /* | |
616 | * get immediately available events, if any | |
617 | */ | |
618 | r = io_u_getevents(td, 0, td->cur_depth, &ts); | |
619 | if (r > 0) { | |
620 | icd.nr = r; | |
621 | ios_completed(td, &icd); | |
622 | } | |
623 | ||
624 | /* | |
625 | * now cancel remaining active events | |
626 | */ | |
627 | if (td->io_cancel) { | |
628 | list_for_each_safe(entry, n, &td->io_u_busylist) { | |
629 | io_u = list_entry(entry, struct io_u, list); | |
630 | ||
631 | r = td->io_cancel(td, io_u); | |
632 | if (!r) | |
633 | put_io_u(td, io_u); | |
634 | } | |
635 | } | |
636 | ||
637 | if (td->cur_depth) { | |
638 | r = io_u_getevents(td, td->cur_depth, td->cur_depth, NULL); | |
639 | if (r > 0) { | |
640 | icd.nr = r; | |
641 | ios_completed(td, &icd); | |
642 | } | |
643 | } | |
644 | } | |
645 | ||
646 | static int do_io_u_verify(struct thread_data *td, struct io_u **io_u) | |
647 | { | |
648 | struct io_u *v_io_u = *io_u; | |
649 | int ret = 0; | |
650 | ||
651 | if (v_io_u) { | |
652 | ret = verify_io_u(v_io_u); | |
653 | put_io_u(td, v_io_u); | |
654 | *io_u = NULL; | |
655 | } | |
656 | ||
657 | return ret; | |
658 | } | |
659 | ||
660 | static void do_verify(struct thread_data *td) | |
661 | { | |
662 | struct timeval t; | |
663 | struct io_u *io_u, *v_io_u = NULL; | |
664 | struct io_completion_data icd; | |
665 | int ret; | |
666 | ||
667 | td_set_runstate(td, TD_VERIFYING); | |
668 | ||
669 | do { | |
670 | if (td->terminate) | |
671 | break; | |
672 | ||
673 | gettimeofday(&t, NULL); | |
674 | if (runtime_exceeded(td, &t)) | |
675 | break; | |
676 | ||
677 | io_u = __get_io_u(td); | |
678 | if (!io_u) | |
679 | break; | |
680 | ||
aea47d44 | 681 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
682 | put_io_u(td, io_u); |
683 | break; | |
684 | } | |
685 | ||
aea47d44 | 686 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
687 | put_io_u(td, io_u); |
688 | break; | |
689 | } | |
690 | ||
691 | ret = io_u_queue(td, io_u); | |
692 | if (ret) { | |
693 | put_io_u(td, io_u); | |
694 | td_verror(td, ret); | |
695 | break; | |
696 | } | |
697 | ||
698 | /* | |
699 | * we have one pending to verify, do that while | |
700 | * we are doing io on the next one | |
701 | */ | |
702 | if (do_io_u_verify(td, &v_io_u)) | |
703 | break; | |
704 | ||
705 | ret = io_u_getevents(td, 1, 1, NULL); | |
706 | if (ret != 1) { | |
707 | if (ret < 0) | |
708 | td_verror(td, ret); | |
709 | break; | |
710 | } | |
711 | ||
712 | v_io_u = td->io_event(td, 0); | |
713 | icd.nr = 1; | |
714 | icd.error = 0; | |
715 | io_completed(td, v_io_u, &icd); | |
716 | ||
717 | if (icd.error) { | |
718 | td_verror(td, icd.error); | |
719 | put_io_u(td, v_io_u); | |
720 | v_io_u = NULL; | |
721 | break; | |
722 | } | |
723 | ||
ebac4655 JA |
724 | /* |
725 | * if we can't submit more io, we need to verify now | |
726 | */ | |
727 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
728 | break; | |
729 | ||
730 | } while (1); | |
731 | ||
732 | do_io_u_verify(td, &v_io_u); | |
733 | ||
734 | if (td->cur_depth) | |
735 | cleanup_pending_aio(td); | |
736 | ||
737 | td_set_runstate(td, TD_RUNNING); | |
738 | } | |
739 | ||
32cd46a0 JA |
740 | /* |
741 | * Main IO worker functions. It retrieves io_u's to process and queues | |
742 | * and reaps them, checking for rate and errors along the way. | |
743 | */ | |
ebac4655 JA |
744 | static void do_io(struct thread_data *td) |
745 | { | |
746 | struct io_completion_data icd; | |
747 | struct timeval s, e; | |
748 | unsigned long usec; | |
749 | ||
5853e5a8 JA |
750 | td_set_runstate(td, TD_RUNNING); |
751 | ||
ebac4655 JA |
752 | while (td->this_io_bytes[td->ddir] < td->io_size) { |
753 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
754 | struct timespec *timeout; | |
755 | int ret, min_evts = 0; | |
756 | struct io_u *io_u; | |
757 | ||
758 | if (td->terminate) | |
759 | break; | |
760 | ||
761 | io_u = get_io_u(td); | |
762 | if (!io_u) | |
763 | break; | |
764 | ||
765 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
766 | ||
767 | ret = io_u_queue(td, io_u); | |
768 | if (ret) { | |
769 | put_io_u(td, io_u); | |
770 | td_verror(td, ret); | |
771 | break; | |
772 | } | |
773 | ||
774 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
775 | ||
776 | if (td->cur_depth < td->iodepth) { | |
777 | timeout = &ts; | |
778 | min_evts = 0; | |
779 | } else { | |
780 | timeout = NULL; | |
781 | min_evts = 1; | |
782 | } | |
783 | ||
784 | ret = io_u_getevents(td, min_evts, td->cur_depth, timeout); | |
785 | if (ret < 0) { | |
786 | td_verror(td, ret); | |
787 | break; | |
788 | } else if (!ret) | |
789 | continue; | |
790 | ||
791 | icd.nr = ret; | |
792 | ios_completed(td, &icd); | |
793 | if (icd.error) { | |
794 | td_verror(td, icd.error); | |
795 | break; | |
796 | } | |
797 | ||
798 | /* | |
799 | * the rate is batched for now, it should work for batches | |
800 | * of completions except the very first one which may look | |
801 | * a little bursty | |
802 | */ | |
803 | gettimeofday(&e, NULL); | |
804 | usec = utime_since(&s, &e); | |
805 | ||
806 | rate_throttle(td, usec, icd.bytes_done[td->ddir]); | |
807 | ||
808 | if (check_min_rate(td, &e)) { | |
809 | td_verror(td, ENOMEM); | |
810 | break; | |
811 | } | |
812 | ||
813 | if (runtime_exceeded(td, &e)) | |
814 | break; | |
815 | ||
816 | if (td->thinktime) | |
817 | usec_sleep(td, td->thinktime); | |
818 | ||
819 | if (should_fsync(td) && td->fsync_blocks && | |
820 | (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0) | |
821 | sync_td(td); | |
822 | } | |
823 | ||
824 | if (td->cur_depth) | |
825 | cleanup_pending_aio(td); | |
826 | ||
5853e5a8 JA |
827 | if (should_fsync(td) && td->end_fsync) { |
828 | td_set_runstate(td, TD_FSYNCING); | |
ebac4655 | 829 | sync_td(td); |
5853e5a8 | 830 | } |
ebac4655 JA |
831 | } |
832 | ||
833 | static void cleanup_io(struct thread_data *td) | |
834 | { | |
835 | if (td->io_cleanup) | |
836 | td->io_cleanup(td); | |
837 | } | |
838 | ||
839 | static int init_io(struct thread_data *td) | |
840 | { | |
841 | if (td->io_engine == FIO_SYNCIO) | |
842 | return fio_syncio_init(td); | |
843 | else if (td->io_engine == FIO_MMAPIO) | |
844 | return fio_mmapio_init(td); | |
845 | else if (td->io_engine == FIO_LIBAIO) | |
846 | return fio_libaio_init(td); | |
847 | else if (td->io_engine == FIO_POSIXAIO) | |
848 | return fio_posixaio_init(td); | |
849 | else if (td->io_engine == FIO_SGIO) | |
850 | return fio_sgio_init(td); | |
8756e4d4 JA |
851 | else if (td->io_engine == FIO_SPLICEIO) |
852 | return fio_spliceio_init(td); | |
ebac4655 | 853 | else { |
eb8bbf48 | 854 | fprintf(f_err, "bad io_engine %d\n", td->io_engine); |
ebac4655 JA |
855 | return 1; |
856 | } | |
857 | } | |
858 | ||
859 | static void cleanup_io_u(struct thread_data *td) | |
860 | { | |
861 | struct list_head *entry, *n; | |
862 | struct io_u *io_u; | |
863 | ||
864 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
865 | io_u = list_entry(entry, struct io_u, list); | |
866 | ||
867 | list_del(&io_u->list); | |
868 | free(io_u); | |
869 | } | |
870 | ||
871 | if (td->mem_type == MEM_MALLOC) | |
872 | free(td->orig_buffer); | |
873 | else if (td->mem_type == MEM_SHM) { | |
874 | struct shmid_ds sbuf; | |
875 | ||
876 | shmdt(td->orig_buffer); | |
877 | shmctl(td->shm_id, IPC_RMID, &sbuf); | |
878 | } else if (td->mem_type == MEM_MMAP) | |
879 | munmap(td->orig_buffer, td->orig_buffer_size); | |
880 | else | |
eb8bbf48 | 881 | fprintf(f_err, "Bad memory type %d\n", td->mem_type); |
ebac4655 JA |
882 | |
883 | td->orig_buffer = NULL; | |
884 | } | |
885 | ||
886 | static int init_io_u(struct thread_data *td) | |
887 | { | |
888 | struct io_u *io_u; | |
889 | int i, max_units; | |
890 | char *p; | |
891 | ||
892 | if (td->io_engine & FIO_SYNCIO) | |
893 | max_units = 1; | |
894 | else | |
895 | max_units = td->iodepth; | |
896 | ||
897 | td->orig_buffer_size = td->max_bs * max_units + MASK; | |
898 | ||
899 | if (td->mem_type == MEM_MALLOC) | |
900 | td->orig_buffer = malloc(td->orig_buffer_size); | |
901 | else if (td->mem_type == MEM_SHM) { | |
902 | td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600); | |
903 | if (td->shm_id < 0) { | |
904 | td_verror(td, errno); | |
905 | perror("shmget"); | |
906 | return 1; | |
907 | } | |
908 | ||
909 | td->orig_buffer = shmat(td->shm_id, NULL, 0); | |
910 | if (td->orig_buffer == (void *) -1) { | |
911 | td_verror(td, errno); | |
912 | perror("shmat"); | |
913 | td->orig_buffer = NULL; | |
914 | return 1; | |
915 | } | |
916 | } else if (td->mem_type == MEM_MMAP) { | |
917 | td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
918 | if (td->orig_buffer == MAP_FAILED) { | |
919 | td_verror(td, errno); | |
920 | perror("mmap"); | |
921 | td->orig_buffer = NULL; | |
922 | return 1; | |
923 | } | |
924 | } | |
925 | ||
ebac4655 JA |
926 | p = ALIGN(td->orig_buffer); |
927 | for (i = 0; i < max_units; i++) { | |
928 | io_u = malloc(sizeof(*io_u)); | |
929 | memset(io_u, 0, sizeof(*io_u)); | |
930 | INIT_LIST_HEAD(&io_u->list); | |
931 | ||
932 | io_u->buf = p + td->max_bs * i; | |
b1ff3403 | 933 | io_u->index = i; |
ebac4655 JA |
934 | list_add(&io_u->list, &td->io_u_freelist); |
935 | } | |
936 | ||
937 | return 0; | |
938 | } | |
939 | ||
940 | static int create_file(struct thread_data *td, unsigned long long size, | |
941 | int extend) | |
942 | { | |
943 | unsigned long long left; | |
944 | unsigned int bs; | |
945 | int r, oflags; | |
946 | char *b; | |
947 | ||
948 | /* | |
949 | * unless specifically asked for overwrite, let normal io extend it | |
950 | */ | |
951 | if (td_write(td) && !td->overwrite) | |
952 | return 0; | |
953 | ||
954 | if (!size) { | |
eb8bbf48 | 955 | fprintf(f_err, "Need size for create\n"); |
ebac4655 JA |
956 | td_verror(td, EINVAL); |
957 | return 1; | |
958 | } | |
959 | ||
960 | if (!extend) { | |
961 | oflags = O_CREAT | O_TRUNC; | |
eb8bbf48 | 962 | fprintf(f_out, "%s: Laying out IO file (%LuMiB)\n", td->name, size >> 20); |
ebac4655 JA |
963 | } else { |
964 | oflags = O_APPEND; | |
eb8bbf48 | 965 | fprintf(f_out, "%s: Extending IO file (%Lu -> %LuMiB)\n", td->name, (td->file_size - size) >> 20, td->file_size >> 20); |
ebac4655 JA |
966 | } |
967 | ||
968 | td->fd = open(td->file_name, O_WRONLY | oflags, 0644); | |
969 | if (td->fd < 0) { | |
970 | td_verror(td, errno); | |
971 | return 1; | |
972 | } | |
973 | ||
974 | if (!extend && ftruncate(td->fd, td->file_size) == -1) { | |
975 | td_verror(td, errno); | |
976 | return 1; | |
977 | } | |
978 | ||
979 | td->io_size = td->file_size; | |
980 | b = malloc(td->max_bs); | |
981 | memset(b, 0, td->max_bs); | |
982 | ||
983 | left = size; | |
984 | while (left && !td->terminate) { | |
985 | bs = td->max_bs; | |
986 | if (bs > left) | |
987 | bs = left; | |
988 | ||
989 | r = write(td->fd, b, bs); | |
990 | ||
991 | if (r == (int) bs) { | |
992 | left -= bs; | |
993 | continue; | |
994 | } else { | |
995 | if (r < 0) | |
996 | td_verror(td, errno); | |
997 | else | |
998 | td_verror(td, EIO); | |
999 | ||
1000 | break; | |
1001 | } | |
1002 | } | |
1003 | ||
1004 | if (td->terminate) | |
1005 | unlink(td->file_name); | |
1006 | else if (td->create_fsync) | |
1007 | fsync(td->fd); | |
1008 | ||
1009 | close(td->fd); | |
1010 | td->fd = -1; | |
1011 | free(b); | |
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | static int file_size(struct thread_data *td) | |
1016 | { | |
1017 | struct stat st; | |
1018 | ||
1019 | if (fstat(td->fd, &st) == -1) { | |
1020 | td_verror(td, errno); | |
1021 | return 1; | |
1022 | } | |
1023 | ||
838a3cd3 JA |
1024 | td->real_file_size = st.st_size; |
1025 | ||
1026 | if (!td->file_size || td->file_size > td->real_file_size) | |
1027 | td->file_size = td->real_file_size; | |
ebac4655 | 1028 | |
e8e387c1 | 1029 | td->file_size -= td->file_offset; |
ebac4655 JA |
1030 | return 0; |
1031 | } | |
1032 | ||
1033 | static int bdev_size(struct thread_data *td) | |
1034 | { | |
9104f874 | 1035 | unsigned long long bytes; |
ebac4655 JA |
1036 | int r; |
1037 | ||
1038 | r = blockdev_size(td->fd, &bytes); | |
1039 | if (r) { | |
1040 | td_verror(td, r); | |
1041 | return 1; | |
1042 | } | |
1043 | ||
838a3cd3 JA |
1044 | td->real_file_size = bytes; |
1045 | ||
ebac4655 JA |
1046 | /* |
1047 | * no extend possibilities, so limit size to device size if too large | |
1048 | */ | |
838a3cd3 JA |
1049 | if (!td->file_size || td->file_size > td->real_file_size) |
1050 | td->file_size = td->real_file_size; | |
ebac4655 | 1051 | |
e8e387c1 | 1052 | td->file_size -= td->file_offset; |
ebac4655 JA |
1053 | return 0; |
1054 | } | |
1055 | ||
1056 | static int get_file_size(struct thread_data *td) | |
1057 | { | |
0af7b542 | 1058 | int ret = 0; |
ebac4655 JA |
1059 | |
1060 | if (td->filetype == FIO_TYPE_FILE) | |
1061 | ret = file_size(td); | |
0af7b542 | 1062 | else if (td->filetype == FIO_TYPE_BD) |
ebac4655 | 1063 | ret = bdev_size(td); |
0af7b542 JA |
1064 | else |
1065 | td->real_file_size = -1; | |
ebac4655 JA |
1066 | |
1067 | if (ret) | |
1068 | return ret; | |
1069 | ||
e8e387c1 | 1070 | if (td->file_offset > td->real_file_size) { |
eb8bbf48 | 1071 | fprintf(f_err, "%s: offset extends end (%Lu > %Lu)\n", td->name, td->file_offset, td->real_file_size); |
ebac4655 JA |
1072 | return 1; |
1073 | } | |
1074 | ||
838a3cd3 | 1075 | td->io_size = td->file_size; |
ebac4655 | 1076 | if (td->io_size == 0) { |
eb8bbf48 | 1077 | fprintf(f_err, "%s: no io blocks\n", td->name); |
ebac4655 JA |
1078 | td_verror(td, EINVAL); |
1079 | return 1; | |
1080 | } | |
1081 | ||
20dc95c4 JA |
1082 | if (!td->zone_size) |
1083 | td->zone_size = td->io_size; | |
1084 | ||
ebac4655 JA |
1085 | td->total_io_size = td->io_size * td->loops; |
1086 | return 0; | |
1087 | } | |
1088 | ||
1089 | static int setup_file_mmap(struct thread_data *td) | |
1090 | { | |
1091 | int flags; | |
1092 | ||
3d60d1ed JA |
1093 | if (td_rw(td)) |
1094 | flags = PROT_READ | PROT_WRITE; | |
1095 | else if (td_write(td)) { | |
ebac4655 JA |
1096 | flags = PROT_WRITE; |
1097 | ||
1098 | if (td->verify != VERIFY_NONE) | |
1099 | flags |= PROT_READ; | |
3d60d1ed JA |
1100 | } else |
1101 | flags = PROT_READ; | |
ebac4655 JA |
1102 | |
1103 | td->mmap = mmap(NULL, td->file_size, flags, MAP_SHARED, td->fd, td->file_offset); | |
1104 | if (td->mmap == MAP_FAILED) { | |
1105 | td->mmap = NULL; | |
1106 | td_verror(td, errno); | |
1107 | return 1; | |
1108 | } | |
1109 | ||
1110 | if (td->invalidate_cache) { | |
1111 | if (madvise(td->mmap, td->file_size, MADV_DONTNEED) < 0) { | |
1112 | td_verror(td, errno); | |
1113 | return 1; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | if (td->sequential) { | |
1118 | if (madvise(td->mmap, td->file_size, MADV_SEQUENTIAL) < 0) { | |
1119 | td_verror(td, errno); | |
1120 | return 1; | |
1121 | } | |
1122 | } else { | |
1123 | if (madvise(td->mmap, td->file_size, MADV_RANDOM) < 0) { | |
1124 | td_verror(td, errno); | |
1125 | return 1; | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | return 0; | |
1130 | } | |
1131 | ||
1132 | static int setup_file_plain(struct thread_data *td) | |
1133 | { | |
1134 | if (td->invalidate_cache) { | |
1135 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_DONTNEED) < 0) { | |
1136 | td_verror(td, errno); | |
1137 | return 1; | |
1138 | } | |
1139 | } | |
1140 | ||
1141 | if (td->sequential) { | |
1142 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_SEQUENTIAL) < 0) { | |
1143 | td_verror(td, errno); | |
1144 | return 1; | |
1145 | } | |
1146 | } else { | |
1147 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_RANDOM) < 0) { | |
1148 | td_verror(td, errno); | |
1149 | return 1; | |
1150 | } | |
1151 | } | |
1152 | ||
1153 | return 0; | |
1154 | } | |
1155 | ||
1156 | static int setup_file(struct thread_data *td) | |
1157 | { | |
1158 | struct stat st; | |
1159 | int flags = 0; | |
1160 | ||
1161 | if (stat(td->file_name, &st) == -1) { | |
1162 | if (errno != ENOENT) { | |
1163 | td_verror(td, errno); | |
1164 | return 1; | |
1165 | } | |
1166 | if (!td->create_file) { | |
1167 | td_verror(td, ENOENT); | |
1168 | return 1; | |
1169 | } | |
1170 | if (create_file(td, td->file_size, 0)) | |
1171 | return 1; | |
1172 | } else if (td->filetype == FIO_TYPE_FILE) { | |
6a0106a0 | 1173 | if (st.st_size < (off_t) td->file_size) { |
ebac4655 JA |
1174 | if (create_file(td, td->file_size - st.st_size, 1)) |
1175 | return 1; | |
1176 | } | |
1177 | } | |
1178 | ||
1179 | if (td->odirect) | |
2c0ecd28 | 1180 | flags |= OS_O_DIRECT; |
ebac4655 | 1181 | |
3d60d1ed | 1182 | if (td_write(td) || td_rw(td)) { |
ebac4655 JA |
1183 | if (td->filetype == FIO_TYPE_FILE) { |
1184 | if (!td->overwrite) | |
1185 | flags |= O_TRUNC; | |
1186 | ||
1187 | flags |= O_CREAT; | |
1188 | } | |
1189 | if (td->sync_io) | |
1190 | flags |= O_SYNC; | |
1191 | ||
1192 | flags |= O_RDWR; | |
1193 | ||
1194 | td->fd = open(td->file_name, flags, 0600); | |
3d60d1ed JA |
1195 | } else { |
1196 | if (td->filetype == FIO_TYPE_CHAR) | |
1197 | flags |= O_RDWR; | |
1198 | else | |
1199 | flags |= O_RDONLY; | |
1200 | ||
1201 | td->fd = open(td->file_name, flags); | |
ebac4655 JA |
1202 | } |
1203 | ||
1204 | if (td->fd == -1) { | |
1205 | td_verror(td, errno); | |
1206 | return 1; | |
1207 | } | |
1208 | ||
1209 | if (get_file_size(td)) | |
1210 | return 1; | |
1211 | ||
1212 | if (td->io_engine != FIO_MMAPIO) | |
1213 | return setup_file_plain(td); | |
1214 | else | |
1215 | return setup_file_mmap(td); | |
1216 | } | |
1217 | ||
da86774e JA |
1218 | static int switch_ioscheduler(struct thread_data *td) |
1219 | { | |
1220 | char tmp[256], tmp2[128]; | |
1221 | FILE *f; | |
1222 | int ret; | |
1223 | ||
1224 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
1225 | ||
1226 | f = fopen(tmp, "r+"); | |
1227 | if (!f) { | |
1228 | td_verror(td, errno); | |
1229 | return 1; | |
1230 | } | |
1231 | ||
1232 | /* | |
1233 | * Set io scheduler. | |
1234 | */ | |
1235 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
1236 | if (ferror(f) || ret != 1) { | |
1237 | td_verror(td, errno); | |
1238 | fclose(f); | |
1239 | return 1; | |
1240 | } | |
1241 | ||
1242 | rewind(f); | |
1243 | ||
1244 | /* | |
1245 | * Read back and check that the selected scheduler is now the default. | |
1246 | */ | |
1247 | ret = fread(tmp, 1, sizeof(tmp), f); | |
1248 | if (ferror(f) || ret < 0) { | |
1249 | td_verror(td, errno); | |
1250 | fclose(f); | |
1251 | return 1; | |
1252 | } | |
1253 | ||
1254 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
1255 | if (!strstr(tmp, tmp2)) { | |
eb8bbf48 | 1256 | fprintf(f_err, "fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
1257 | td_verror(td, EINVAL); |
1258 | fclose(f); | |
1259 | return 1; | |
1260 | } | |
1261 | ||
1262 | fclose(f); | |
1263 | return 0; | |
1264 | } | |
1265 | ||
ebac4655 JA |
1266 | static void clear_io_state(struct thread_data *td) |
1267 | { | |
1268 | if (td->io_engine == FIO_SYNCIO) | |
1269 | lseek(td->fd, SEEK_SET, 0); | |
1270 | ||
20dc95c4 | 1271 | td->last_pos = 0; |
ebac4655 JA |
1272 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
1273 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 1274 | td->zone_bytes = 0; |
ebac4655 JA |
1275 | |
1276 | if (td->file_map) | |
1277 | memset(td->file_map, 0, td->num_maps * sizeof(long)); | |
1278 | } | |
1279 | ||
ebac4655 JA |
1280 | static void *thread_main(void *data) |
1281 | { | |
1282 | struct thread_data *td = data; | |
ebac4655 JA |
1283 | |
1284 | if (!td->use_thread) | |
1285 | setsid(); | |
1286 | ||
1287 | td->pid = getpid(); | |
1288 | ||
aea47d44 JA |
1289 | INIT_LIST_HEAD(&td->io_u_freelist); |
1290 | INIT_LIST_HEAD(&td->io_u_busylist); | |
1291 | INIT_LIST_HEAD(&td->io_hist_list); | |
1292 | INIT_LIST_HEAD(&td->io_log_list); | |
1293 | ||
ebac4655 JA |
1294 | if (init_io_u(td)) |
1295 | goto err; | |
1296 | ||
1297 | if (fio_setaffinity(td) == -1) { | |
1298 | td_verror(td, errno); | |
1299 | goto err; | |
1300 | } | |
1301 | ||
1302 | if (init_io(td)) | |
1303 | goto err; | |
1304 | ||
aea47d44 JA |
1305 | if (init_iolog(td)) |
1306 | goto err; | |
1307 | ||
ebac4655 JA |
1308 | if (td->ioprio) { |
1309 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
1310 | td_verror(td, errno); | |
1311 | goto err; | |
1312 | } | |
1313 | } | |
1314 | ||
b6f4d880 JA |
1315 | if (nice(td->nice) < 0) { |
1316 | td_verror(td, errno); | |
1317 | goto err; | |
1318 | } | |
1319 | ||
75154845 JA |
1320 | if (init_random_state(td)) |
1321 | goto err; | |
1322 | ||
da86774e JA |
1323 | if (td->ioscheduler && switch_ioscheduler(td)) |
1324 | goto err; | |
1325 | ||
75154845 | 1326 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
1327 | fio_sem_up(&startup_sem); |
1328 | fio_sem_down(&td->mutex); | |
ebac4655 JA |
1329 | |
1330 | if (!td->create_serialize && setup_file(td)) | |
1331 | goto err; | |
1332 | ||
ebac4655 JA |
1333 | gettimeofday(&td->epoch, NULL); |
1334 | ||
4e0ba8af JA |
1335 | if (td->exec_prerun) |
1336 | system(td->exec_prerun); | |
1337 | ||
ebac4655 JA |
1338 | while (td->loops--) { |
1339 | getrusage(RUSAGE_SELF, &td->ru_start); | |
1340 | gettimeofday(&td->start, NULL); | |
1341 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); | |
1342 | ||
1343 | if (td->ratemin) | |
1344 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
1345 | ||
1346 | clear_io_state(td); | |
1347 | prune_io_piece_log(td); | |
1348 | ||
1349 | do_io(td); | |
1350 | ||
1351 | td->runtime[td->ddir] += mtime_since_now(&td->start); | |
aea47d44 | 1352 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
3d60d1ed JA |
1353 | td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; |
1354 | ||
ebac4655 JA |
1355 | update_rusage_stat(td); |
1356 | ||
1357 | if (td->error || td->terminate) | |
1358 | break; | |
1359 | ||
1360 | if (td->verify == VERIFY_NONE) | |
1361 | continue; | |
1362 | ||
1363 | clear_io_state(td); | |
1364 | gettimeofday(&td->start, NULL); | |
1365 | ||
1366 | do_verify(td); | |
1367 | ||
1368 | td->runtime[DDIR_READ] += mtime_since_now(&td->start); | |
1369 | ||
1370 | if (td->error || td->terminate) | |
1371 | break; | |
1372 | } | |
1373 | ||
ebac4655 JA |
1374 | if (td->bw_log) |
1375 | finish_log(td, td->bw_log, "bw"); | |
1376 | if (td->slat_log) | |
1377 | finish_log(td, td->slat_log, "slat"); | |
1378 | if (td->clat_log) | |
1379 | finish_log(td, td->clat_log, "clat"); | |
843a7413 JA |
1380 | if (td->write_iolog) |
1381 | write_iolog_close(td); | |
4e0ba8af JA |
1382 | if (td->exec_postrun) |
1383 | system(td->exec_postrun); | |
ebac4655 JA |
1384 | |
1385 | if (exitall_on_terminate) | |
1386 | terminate_threads(td->groupid); | |
1387 | ||
1388 | err: | |
1389 | if (td->fd != -1) { | |
1390 | close(td->fd); | |
1391 | td->fd = -1; | |
1392 | } | |
1393 | if (td->mmap) | |
1394 | munmap(td->mmap, td->file_size); | |
1395 | cleanup_io(td); | |
1396 | cleanup_io_u(td); | |
ebac4655 JA |
1397 | td_set_runstate(td, TD_EXITED); |
1398 | return NULL; | |
1399 | ||
1400 | } | |
1401 | ||
1402 | static void *fork_main(int shmid, int offset) | |
1403 | { | |
1404 | struct thread_data *td; | |
1405 | void *data; | |
1406 | ||
1407 | data = shmat(shmid, NULL, 0); | |
1408 | if (data == (void *) -1) { | |
1409 | perror("shmat"); | |
1410 | return NULL; | |
1411 | } | |
1412 | ||
1413 | td = data + offset * sizeof(struct thread_data); | |
1414 | thread_main(td); | |
1415 | shmdt(data); | |
1416 | return NULL; | |
1417 | } | |
1418 | ||
ebac4655 JA |
1419 | static void check_str_update(struct thread_data *td) |
1420 | { | |
1421 | char c = run_str[td->thread_number - 1]; | |
1422 | ||
ebac4655 JA |
1423 | switch (td->runstate) { |
1424 | case TD_REAPED: | |
1425 | c = '_'; | |
1426 | break; | |
1427 | case TD_EXITED: | |
1428 | c = 'E'; | |
1429 | break; | |
1430 | case TD_RUNNING: | |
3d60d1ed JA |
1431 | if (td_rw(td)) { |
1432 | if (td->sequential) | |
1433 | c = 'M'; | |
1434 | else | |
1435 | c = 'm'; | |
1436 | } else if (td_read(td)) { | |
ebac4655 JA |
1437 | if (td->sequential) |
1438 | c = 'R'; | |
1439 | else | |
1440 | c = 'r'; | |
1441 | } else { | |
1442 | if (td->sequential) | |
1443 | c = 'W'; | |
1444 | else | |
1445 | c = 'w'; | |
1446 | } | |
1447 | break; | |
1448 | case TD_VERIFYING: | |
1449 | c = 'V'; | |
1450 | break; | |
5853e5a8 JA |
1451 | case TD_FSYNCING: |
1452 | c = 'F'; | |
1453 | break; | |
ebac4655 JA |
1454 | case TD_CREATED: |
1455 | c = 'C'; | |
1456 | break; | |
75154845 JA |
1457 | case TD_INITIALIZED: |
1458 | c = 'I'; | |
1459 | break; | |
ebac4655 JA |
1460 | case TD_NOT_CREATED: |
1461 | c = 'P'; | |
1462 | break; | |
1463 | default: | |
eb8bbf48 | 1464 | fprintf(f_err, "state %d\n", td->runstate); |
ebac4655 JA |
1465 | } |
1466 | ||
1467 | run_str[td->thread_number - 1] = c; | |
ebac4655 JA |
1468 | } |
1469 | ||
5289b847 JA |
1470 | static void eta_to_str(char *str, int eta_sec) |
1471 | { | |
1472 | unsigned int d, h, m, s; | |
1473 | static int always_d, always_h; | |
1474 | ||
1475 | d = h = m = s = 0; | |
1476 | ||
1477 | s = eta_sec % 60; | |
1478 | eta_sec /= 60; | |
1479 | m = eta_sec % 60; | |
1480 | eta_sec /= 60; | |
1481 | h = eta_sec % 24; | |
1482 | eta_sec /= 24; | |
1483 | d = eta_sec; | |
1484 | ||
1485 | if (d || always_d) { | |
1486 | always_d = 1; | |
1487 | str += sprintf(str, "%02dd:", d); | |
1488 | } | |
1489 | if (h || always_h) { | |
1490 | always_h = 1; | |
1491 | str += sprintf(str, "%02dh:", h); | |
1492 | } | |
1493 | ||
1494 | str += sprintf(str, "%02dm:", m); | |
1495 | str += sprintf(str, "%02ds", s); | |
1496 | } | |
1497 | ||
6a0106a0 JA |
1498 | static int thread_eta(struct thread_data *td, unsigned long elapsed) |
1499 | { | |
1500 | unsigned long long bytes_total, bytes_done; | |
1501 | unsigned int eta_sec = 0; | |
1502 | ||
1503 | bytes_total = td->total_io_size; | |
3d60d1ed JA |
1504 | |
1505 | /* | |
1506 | * if writing, bytes_total will be twice the size. If mixing, | |
1507 | * assume a 50/50 split and thus bytes_total will be 50% larger. | |
1508 | */ | |
1509 | if (td->verify) { | |
1510 | if (td_rw(td)) | |
1511 | bytes_total = bytes_total * 3 / 2; | |
1512 | else | |
1513 | bytes_total <<= 1; | |
1514 | } | |
6a0106a0 JA |
1515 | if (td->zone_size && td->zone_skip) |
1516 | bytes_total /= (td->zone_skip / td->zone_size); | |
1517 | ||
1518 | if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) { | |
1519 | double perc; | |
8b611c34 | 1520 | |
6a0106a0 JA |
1521 | bytes_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE]; |
1522 | perc = (double) bytes_done / (double) bytes_total; | |
1523 | if (perc > 1.0) | |
1524 | perc = 1.0; | |
1525 | ||
1526 | eta_sec = (elapsed * (1.0 / perc)) - elapsed; | |
1527 | ||
8b611c34 | 1528 | if (td->timeout && eta_sec > (td->timeout - elapsed)) |
6a0106a0 | 1529 | eta_sec = td->timeout - elapsed; |
75154845 JA |
1530 | } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED |
1531 | || td->runstate == TD_INITIALIZED) { | |
6a0106a0 JA |
1532 | int t_eta = 0, r_eta = 0; |
1533 | ||
1534 | /* | |
1535 | * We can only guess - assume it'll run the full timeout | |
1536 | * if given, otherwise assume it'll run at the specified rate. | |
1537 | */ | |
1538 | if (td->timeout) | |
1539 | t_eta = td->timeout + td->start_delay - elapsed; | |
1540 | if (td->rate) { | |
1541 | r_eta = (bytes_total / 1024) / td->rate; | |
1542 | r_eta += td->start_delay - elapsed; | |
1543 | } | |
1544 | ||
1545 | if (r_eta && t_eta) | |
1546 | eta_sec = min(r_eta, t_eta); | |
1547 | else if (r_eta) | |
1548 | eta_sec = r_eta; | |
1549 | else if (t_eta) | |
1550 | eta_sec = t_eta; | |
1551 | else | |
1552 | eta_sec = INT_MAX; | |
1553 | } else { | |
1554 | /* | |
5853e5a8 | 1555 | * thread is already done or waiting for fsync |
6a0106a0 JA |
1556 | */ |
1557 | eta_sec = 0; | |
1558 | } | |
1559 | ||
1560 | return eta_sec; | |
1561 | } | |
1562 | ||
ebac4655 JA |
1563 | static void print_thread_status(void) |
1564 | { | |
6a0106a0 | 1565 | unsigned long elapsed = time_since_now(&genesis); |
71a751ce | 1566 | int i, nr_running, nr_pending, t_rate, m_rate, *eta_secs, eta_sec; |
5289b847 | 1567 | char eta_str[32]; |
6a0106a0 JA |
1568 | double perc = 0.0; |
1569 | ||
1570 | eta_secs = malloc(thread_number * sizeof(int)); | |
1571 | memset(eta_secs, 0, thread_number * sizeof(int)); | |
ebac4655 | 1572 | |
71a751ce | 1573 | nr_pending = nr_running = t_rate = m_rate = 0; |
ebac4655 JA |
1574 | for (i = 0; i < thread_number; i++) { |
1575 | struct thread_data *td = &threads[i]; | |
1576 | ||
5853e5a8 JA |
1577 | if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING|| |
1578 | td->runstate == TD_FSYNCING) { | |
ebac4655 JA |
1579 | nr_running++; |
1580 | t_rate += td->rate; | |
1581 | m_rate += td->ratemin; | |
71a751ce JA |
1582 | } else if (td->runstate < TD_RUNNING) |
1583 | nr_pending++; | |
ebac4655 | 1584 | |
6a0106a0 JA |
1585 | if (elapsed >= 3) |
1586 | eta_secs[i] = thread_eta(td, elapsed); | |
8b611c34 JA |
1587 | else |
1588 | eta_secs[i] = INT_MAX; | |
ebac4655 JA |
1589 | |
1590 | check_str_update(td); | |
1591 | } | |
1592 | ||
6a0106a0 JA |
1593 | if (exitall_on_terminate) |
1594 | eta_sec = INT_MAX; | |
1595 | else | |
1596 | eta_sec = 0; | |
5289b847 | 1597 | |
6a0106a0 JA |
1598 | for (i = 0; i < thread_number; i++) { |
1599 | if (exitall_on_terminate) { | |
1600 | if (eta_secs[i] < eta_sec) | |
1601 | eta_sec = eta_secs[i]; | |
1602 | } else { | |
1603 | if (eta_secs[i] > eta_sec) | |
1604 | eta_sec = eta_secs[i]; | |
5289b847 | 1605 | } |
6a0106a0 | 1606 | } |
5289b847 | 1607 | |
8b611c34 | 1608 | if (eta_sec != INT_MAX && elapsed) { |
6a0106a0 JA |
1609 | perc = (double) elapsed / (double) (elapsed + eta_sec); |
1610 | eta_to_str(eta_str, eta_sec); | |
ebac4655 JA |
1611 | } |
1612 | ||
71a751ce | 1613 | if (!nr_running && !nr_pending) |
f44c1d46 JA |
1614 | return; |
1615 | ||
eb8bbf48 | 1616 | printf("Threads running: %d", nr_running); |
ebac4655 JA |
1617 | if (m_rate || t_rate) |
1618 | printf(", commitrate %d/%dKiB/sec", t_rate, m_rate); | |
8b611c34 | 1619 | if (eta_sec != INT_MAX) { |
6a0106a0 JA |
1620 | perc *= 100.0; |
1621 | printf(": [%s] [%3.2f%% done] [eta %s]", run_str, perc,eta_str); | |
1622 | } | |
5289b847 | 1623 | printf("\r"); |
ebac4655 | 1624 | fflush(stdout); |
6a0106a0 | 1625 | free(eta_secs); |
ebac4655 JA |
1626 | } |
1627 | ||
1628 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) | |
1629 | { | |
1630 | int i; | |
1631 | ||
1632 | /* | |
1633 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1634 | */ | |
1635 | for (i = 0; i < thread_number; i++) { | |
1636 | struct thread_data *td = &threads[i]; | |
1637 | ||
1638 | if (td->runstate != TD_EXITED) | |
1639 | continue; | |
1640 | ||
1641 | td_set_runstate(td, TD_REAPED); | |
1642 | ||
1643 | if (td->use_thread) { | |
1644 | long ret; | |
1645 | ||
1646 | if (pthread_join(td->thread, (void *) &ret)) | |
1647 | perror("thread_join"); | |
1648 | } else | |
1649 | waitpid(td->pid, NULL, 0); | |
1650 | ||
1651 | (*nr_running)--; | |
1652 | (*m_rate) -= td->ratemin; | |
1653 | (*t_rate) -= td->rate; | |
1654 | } | |
1655 | } | |
1656 | ||
fcb6ade2 JA |
1657 | static void fio_unpin_memory(void *pinned) |
1658 | { | |
1659 | if (pinned) { | |
1660 | if (munlock(pinned, mlock_size) < 0) | |
1661 | perror("munlock"); | |
1662 | munmap(pinned, mlock_size); | |
1663 | } | |
1664 | } | |
1665 | ||
1666 | static void *fio_pin_memory(void) | |
1667 | { | |
32cd46a0 | 1668 | unsigned long long phys_mem; |
fcb6ade2 JA |
1669 | void *ptr; |
1670 | ||
1671 | if (!mlock_size) | |
1672 | return NULL; | |
1673 | ||
1674 | /* | |
1675 | * Don't allow mlock of more than real_mem-128MB | |
1676 | */ | |
32cd46a0 JA |
1677 | phys_mem = os_phys_mem(); |
1678 | if (phys_mem) { | |
1679 | if ((mlock_size + 128 * 1024 * 1024) > phys_mem) { | |
1680 | mlock_size = phys_mem - 128 * 1024 * 1024; | |
eb8bbf48 | 1681 | fprintf(f_out, "fio: limiting mlocked memory to %lluMiB\n", mlock_size >> 20); |
fcb6ade2 JA |
1682 | } |
1683 | } | |
1684 | ||
1685 | ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
1686 | if (!ptr) { | |
1687 | perror("malloc locked mem"); | |
1688 | return NULL; | |
1689 | } | |
1690 | if (mlock(ptr, mlock_size) < 0) { | |
1691 | munmap(ptr, mlock_size); | |
1692 | perror("mlock"); | |
1693 | return NULL; | |
1694 | } | |
1695 | ||
1696 | return ptr; | |
1697 | } | |
1698 | ||
ebac4655 JA |
1699 | static void run_threads(void) |
1700 | { | |
ebac4655 JA |
1701 | struct thread_data *td; |
1702 | unsigned long spent; | |
1703 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 JA |
1704 | void *mlocked_mem; |
1705 | ||
1706 | mlocked_mem = fio_pin_memory(); | |
ebac4655 JA |
1707 | |
1708 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
fcb6ade2 | 1709 | fflush(stdout); |
c04f7ec3 | 1710 | |
4efa970e JA |
1711 | signal(SIGINT, sig_handler); |
1712 | signal(SIGALRM, sig_handler); | |
1713 | ||
ebac4655 JA |
1714 | todo = thread_number; |
1715 | nr_running = 0; | |
1716 | nr_started = 0; | |
1717 | m_rate = t_rate = 0; | |
1718 | ||
1719 | for (i = 0; i < thread_number; i++) { | |
1720 | td = &threads[i]; | |
1721 | ||
1722 | run_str[td->thread_number - 1] = 'P'; | |
1723 | ||
1724 | init_disk_util(td); | |
1725 | ||
1726 | if (!td->create_serialize) | |
1727 | continue; | |
1728 | ||
1729 | /* | |
1730 | * do file setup here so it happens sequentially, | |
1731 | * we don't want X number of threads getting their | |
1732 | * client data interspersed on disk | |
1733 | */ | |
1734 | if (setup_file(td)) { | |
1735 | td_set_runstate(td, TD_REAPED); | |
1736 | todo--; | |
1737 | } | |
1738 | } | |
1739 | ||
1740 | gettimeofday(&genesis, NULL); | |
1741 | ||
1742 | while (todo) { | |
75154845 JA |
1743 | struct thread_data *map[MAX_JOBS]; |
1744 | struct timeval this_start; | |
1745 | int this_jobs = 0, left; | |
1746 | ||
ebac4655 JA |
1747 | /* |
1748 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1749 | */ | |
1750 | for (i = 0; i < thread_number; i++) { | |
1751 | td = &threads[i]; | |
1752 | ||
1753 | if (td->runstate != TD_NOT_CREATED) | |
1754 | continue; | |
1755 | ||
1756 | /* | |
1757 | * never got a chance to start, killed by other | |
1758 | * thread for some reason | |
1759 | */ | |
1760 | if (td->terminate) { | |
1761 | todo--; | |
1762 | continue; | |
1763 | } | |
1764 | ||
1765 | if (td->start_delay) { | |
1766 | spent = mtime_since_now(&genesis); | |
1767 | ||
1768 | if (td->start_delay * 1000 > spent) | |
1769 | continue; | |
1770 | } | |
1771 | ||
1772 | if (td->stonewall && (nr_started || nr_running)) | |
1773 | break; | |
1774 | ||
75154845 JA |
1775 | /* |
1776 | * Set state to created. Thread will transition | |
1777 | * to TD_INITIALIZED when it's done setting up. | |
1778 | */ | |
ebac4655 | 1779 | td_set_runstate(td, TD_CREATED); |
75154845 | 1780 | map[this_jobs++] = td; |
bbfd6b00 | 1781 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
1782 | nr_started++; |
1783 | ||
1784 | if (td->use_thread) { | |
1785 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
1786 | perror("thread_create"); | |
1787 | nr_started--; | |
1788 | } | |
1789 | } else { | |
1790 | if (fork()) | |
bbfd6b00 | 1791 | fio_sem_down(&startup_sem); |
ebac4655 JA |
1792 | else { |
1793 | fork_main(shm_id, i); | |
1794 | exit(0); | |
1795 | } | |
1796 | } | |
1797 | } | |
1798 | ||
1799 | /* | |
75154845 JA |
1800 | * Wait for the started threads to transition to |
1801 | * TD_INITIALIZED. | |
ebac4655 | 1802 | */ |
75154845 JA |
1803 | printf("fio: Waiting for threads to initialize...\n"); |
1804 | gettimeofday(&this_start, NULL); | |
1805 | left = this_jobs; | |
1806 | while (left) { | |
1807 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
1808 | break; | |
1809 | ||
1810 | usleep(100000); | |
1811 | ||
1812 | for (i = 0; i < this_jobs; i++) { | |
1813 | td = map[i]; | |
1814 | if (!td) | |
1815 | continue; | |
b6f4d880 | 1816 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1817 | map[i] = NULL; |
1818 | left--; | |
b6f4d880 JA |
1819 | } else if (td->runstate >= TD_EXITED) { |
1820 | map[i] = NULL; | |
1821 | left--; | |
1822 | todo--; | |
1823 | nr_running++; /* work-around... */ | |
75154845 JA |
1824 | } |
1825 | } | |
1826 | } | |
1827 | ||
1828 | if (left) { | |
eb8bbf48 | 1829 | fprintf(f_err, "fio: %d jobs failed to start\n", left); |
75154845 JA |
1830 | for (i = 0; i < this_jobs; i++) { |
1831 | td = map[i]; | |
1832 | if (!td) | |
1833 | continue; | |
1834 | kill(td->pid, SIGTERM); | |
1835 | } | |
1836 | break; | |
1837 | } | |
1838 | ||
1839 | /* | |
b6f4d880 | 1840 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 JA |
1841 | */ |
1842 | printf("fio: Go for launch\n"); | |
ebac4655 JA |
1843 | for (i = 0; i < thread_number; i++) { |
1844 | td = &threads[i]; | |
1845 | ||
75154845 | 1846 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1847 | continue; |
1848 | ||
1849 | td_set_runstate(td, TD_RUNNING); | |
1850 | nr_running++; | |
1851 | nr_started--; | |
1852 | m_rate += td->ratemin; | |
1853 | t_rate += td->rate; | |
75154845 | 1854 | todo--; |
bbfd6b00 | 1855 | fio_sem_up(&td->mutex); |
ebac4655 JA |
1856 | } |
1857 | ||
1858 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1859 | ||
1860 | if (todo) | |
1861 | usleep(100000); | |
1862 | } | |
1863 | ||
1864 | while (nr_running) { | |
1865 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1866 | usleep(10000); | |
1867 | } | |
1868 | ||
1869 | update_io_ticks(); | |
fcb6ade2 | 1870 | fio_unpin_memory(mlocked_mem); |
ebac4655 JA |
1871 | } |
1872 | ||
ebac4655 JA |
1873 | int main(int argc, char *argv[]) |
1874 | { | |
1875 | if (parse_options(argc, argv)) | |
1876 | return 1; | |
1877 | ||
1878 | if (!thread_number) { | |
eb8bbf48 | 1879 | fprintf(f_err, "Nothing to do\n"); |
ebac4655 JA |
1880 | return 1; |
1881 | } | |
1882 | ||
1883 | disk_util_timer_arm(); | |
1884 | ||
1885 | run_threads(); | |
1886 | show_run_stats(); | |
1887 | ||
1888 | return 0; | |
1889 | } |