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; |
88c6ed80 | 46 | static int temp_stall_ts; |
ebac4655 | 47 | |
ebac4655 JA |
48 | static void print_thread_status(void); |
49 | ||
c04f7ec3 JA |
50 | extern unsigned long long mlock_size; |
51 | ||
ebac4655 | 52 | /* |
79809113 JA |
53 | * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it |
54 | * will never back again. It may cycle between running/verififying/fsyncing. | |
55 | * Once the thread reaches TD_EXITED, it is just waiting for the core to | |
56 | * reap it. | |
ebac4655 JA |
57 | */ |
58 | enum { | |
59 | TD_NOT_CREATED = 0, | |
60 | TD_CREATED, | |
75154845 | 61 | TD_INITIALIZED, |
ebac4655 JA |
62 | TD_RUNNING, |
63 | TD_VERIFYING, | |
5853e5a8 | 64 | TD_FSYNCING, |
ebac4655 JA |
65 | TD_EXITED, |
66 | TD_REAPED, | |
67 | }; | |
68 | ||
3d60d1ed | 69 | #define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync)) |
ebac4655 | 70 | |
bbfd6b00 | 71 | static volatile int startup_sem; |
ebac4655 JA |
72 | |
73 | #define TERMINATE_ALL (-1) | |
75154845 | 74 | #define JOB_START_TIMEOUT (5 * 1000) |
ebac4655 JA |
75 | |
76 | static void terminate_threads(int group_id) | |
77 | { | |
78 | int i; | |
79 | ||
80 | for (i = 0; i < thread_number; i++) { | |
81 | struct thread_data *td = &threads[i]; | |
82 | ||
83 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { | |
84 | td->terminate = 1; | |
85 | td->start_delay = 0; | |
86 | } | |
87 | } | |
88 | } | |
89 | ||
90 | static void sig_handler(int sig) | |
91 | { | |
92 | switch (sig) { | |
93 | case SIGALRM: | |
94 | update_io_ticks(); | |
95 | disk_util_timer_arm(); | |
96 | print_thread_status(); | |
97 | break; | |
98 | default: | |
99 | printf("\nfio: terminating on signal\n"); | |
100 | fflush(stdout); | |
101 | terminate_threads(TERMINATE_ALL); | |
102 | break; | |
103 | } | |
104 | } | |
105 | ||
ebac4655 JA |
106 | static int random_map_free(struct thread_data *td, unsigned long long block) |
107 | { | |
108 | unsigned int idx = RAND_MAP_IDX(td, block); | |
109 | unsigned int bit = RAND_MAP_BIT(td, block); | |
110 | ||
111 | return (td->file_map[idx] & (1UL << bit)) == 0; | |
112 | } | |
113 | ||
114 | static int get_next_free_block(struct thread_data *td, unsigned long long *b) | |
115 | { | |
116 | int i; | |
117 | ||
118 | *b = 0; | |
119 | i = 0; | |
120 | while ((*b) * td->min_bs < td->io_size) { | |
121 | if (td->file_map[i] != -1UL) { | |
122 | *b += ffz(td->file_map[i]); | |
123 | return 0; | |
124 | } | |
125 | ||
126 | *b += BLOCKS_PER_MAP; | |
127 | i++; | |
128 | } | |
129 | ||
130 | return 1; | |
131 | } | |
132 | ||
133 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) | |
134 | { | |
200bc855 | 135 | unsigned long long block = io_u->offset / (unsigned long long) td->min_bs; |
ebac4655 JA |
136 | unsigned int blocks = 0; |
137 | ||
138 | while (blocks < (io_u->buflen / td->min_bs)) { | |
139 | unsigned int idx, bit; | |
140 | ||
141 | if (!random_map_free(td, block)) | |
142 | break; | |
143 | ||
144 | idx = RAND_MAP_IDX(td, block); | |
145 | bit = RAND_MAP_BIT(td, block); | |
146 | ||
147 | assert(idx < td->num_maps); | |
148 | ||
149 | td->file_map[idx] |= (1UL << bit); | |
150 | block++; | |
151 | blocks++; | |
152 | } | |
153 | ||
154 | if ((blocks * td->min_bs) < io_u->buflen) | |
155 | io_u->buflen = blocks * td->min_bs; | |
156 | } | |
157 | ||
20dc95c4 JA |
158 | static int get_next_offset(struct thread_data *td, unsigned long long *offset) |
159 | { | |
160 | unsigned long long b, rb; | |
161 | long r; | |
162 | ||
163 | if (!td->sequential) { | |
085227ad | 164 | unsigned long long max_blocks = td->io_size / td->min_bs; |
20dc95c4 JA |
165 | int loops = 50; |
166 | ||
167 | do { | |
6dfd46b9 | 168 | r = os_random_long(&td->random_state); |
085227ad | 169 | b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0)); |
20dc95c4 JA |
170 | rb = b + (td->file_offset / td->min_bs); |
171 | loops--; | |
172 | } while (!random_map_free(td, rb) && loops); | |
173 | ||
174 | if (!loops) { | |
175 | if (get_next_free_block(td, &b)) | |
176 | return 1; | |
177 | } | |
178 | } else | |
179 | b = td->last_pos / td->min_bs; | |
180 | ||
181 | *offset = (b * td->min_bs) + td->file_offset; | |
838a3cd3 | 182 | if (*offset > td->real_file_size) |
20dc95c4 JA |
183 | return 1; |
184 | ||
185 | return 0; | |
186 | } | |
187 | ||
188 | static unsigned int get_next_buflen(struct thread_data *td) | |
189 | { | |
190 | unsigned int buflen; | |
191 | long r; | |
192 | ||
193 | if (td->min_bs == td->max_bs) | |
194 | buflen = td->min_bs; | |
195 | else { | |
6dfd46b9 | 196 | r = os_random_long(&td->bsrange_state); |
20dc95c4 JA |
197 | buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0)); |
198 | buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1); | |
199 | } | |
200 | ||
201 | if (buflen > td->io_size - td->this_io_bytes[td->ddir]) | |
202 | buflen = td->io_size - td->this_io_bytes[td->ddir]; | |
203 | ||
204 | return buflen; | |
205 | } | |
206 | ||
ebac4655 JA |
207 | static int check_min_rate(struct thread_data *td, struct timeval *now) |
208 | { | |
209 | unsigned long spent; | |
210 | unsigned long rate; | |
211 | int ddir = td->ddir; | |
212 | ||
213 | /* | |
214 | * allow a 2 second settle period in the beginning | |
215 | */ | |
216 | if (mtime_since(&td->start, now) < 2000) | |
217 | return 0; | |
218 | ||
219 | /* | |
220 | * if rate blocks is set, sample is running | |
221 | */ | |
222 | if (td->rate_bytes) { | |
223 | spent = mtime_since(&td->lastrate, now); | |
224 | if (spent < td->ratecycle) | |
225 | return 0; | |
226 | ||
227 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
228 | if (rate < td->ratemin) { | |
eb8bbf48 | 229 | fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate); |
ebac4655 JA |
230 | if (rate_quit) |
231 | terminate_threads(td->groupid); | |
232 | return 1; | |
233 | } | |
234 | } | |
235 | ||
236 | td->rate_bytes = td->this_io_bytes[ddir]; | |
237 | memcpy(&td->lastrate, now, sizeof(*now)); | |
238 | return 0; | |
239 | } | |
240 | ||
241 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
242 | { | |
243 | if (!td->timeout) | |
244 | return 0; | |
245 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
246 | return 1; | |
247 | ||
248 | return 0; | |
249 | } | |
250 | ||
251 | static void fill_random_bytes(struct thread_data *td, | |
252 | unsigned char *p, unsigned int len) | |
253 | { | |
254 | unsigned int todo; | |
255 | double r; | |
256 | ||
257 | while (len) { | |
6dfd46b9 | 258 | r = os_random_double(&td->verify_state); |
ebac4655 JA |
259 | |
260 | /* | |
261 | * lrand48_r seems to be broken and only fill the bottom | |
262 | * 32-bits, even on 64-bit archs with 64-bit longs | |
263 | */ | |
264 | todo = sizeof(r); | |
265 | if (todo > len) | |
266 | todo = len; | |
267 | ||
268 | memcpy(p, &r, todo); | |
269 | ||
270 | len -= todo; | |
271 | p += todo; | |
272 | } | |
273 | } | |
274 | ||
275 | static void hexdump(void *buffer, int len) | |
276 | { | |
277 | unsigned char *p = buffer; | |
278 | int i; | |
279 | ||
280 | for (i = 0; i < len; i++) | |
eb8bbf48 JA |
281 | fprintf(f_out, "%02x", p[i]); |
282 | fprintf(f_out, "\n"); | |
ebac4655 JA |
283 | } |
284 | ||
285 | static int verify_io_u_crc32(struct verify_header *hdr, struct io_u *io_u) | |
286 | { | |
287 | unsigned char *p = (unsigned char *) io_u->buf; | |
288 | unsigned long c; | |
ebac4655 JA |
289 | |
290 | p += sizeof(*hdr); | |
291 | c = crc32(p, hdr->len - sizeof(*hdr)); | |
ebac4655 | 292 | |
22f78b32 | 293 | if (c != hdr->crc32) { |
3b70d7e5 JA |
294 | log_err("crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen); |
295 | log_err("crc32: wanted %lx, got %lx\n", hdr->crc32, c); | |
22f78b32 | 296 | return 1; |
ebac4655 JA |
297 | } |
298 | ||
22f78b32 | 299 | return 0; |
ebac4655 JA |
300 | } |
301 | ||
302 | static int verify_io_u_md5(struct verify_header *hdr, struct io_u *io_u) | |
303 | { | |
304 | unsigned char *p = (unsigned char *) io_u->buf; | |
305 | struct md5_ctx md5_ctx; | |
ebac4655 JA |
306 | |
307 | memset(&md5_ctx, 0, sizeof(md5_ctx)); | |
308 | p += sizeof(*hdr); | |
309 | md5_update(&md5_ctx, p, hdr->len - sizeof(*hdr)); | |
310 | ||
22f78b32 | 311 | if (memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash))) { |
3b70d7e5 | 312 | log_err("md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen); |
ebac4655 JA |
313 | hexdump(hdr->md5_digest, sizeof(hdr->md5_digest)); |
314 | hexdump(md5_ctx.hash, sizeof(md5_ctx.hash)); | |
22f78b32 | 315 | return 1; |
ebac4655 JA |
316 | } |
317 | ||
22f78b32 | 318 | return 0; |
ebac4655 JA |
319 | } |
320 | ||
321 | static int verify_io_u(struct io_u *io_u) | |
322 | { | |
323 | struct verify_header *hdr = (struct verify_header *) io_u->buf; | |
324 | int ret; | |
325 | ||
326 | if (hdr->fio_magic != FIO_HDR_MAGIC) | |
327 | return 1; | |
328 | ||
329 | if (hdr->verify_type == VERIFY_MD5) | |
330 | ret = verify_io_u_md5(hdr, io_u); | |
331 | else if (hdr->verify_type == VERIFY_CRC32) | |
332 | ret = verify_io_u_crc32(hdr, io_u); | |
333 | else { | |
3b70d7e5 | 334 | log_err("Bad verify type %d\n", hdr->verify_type); |
ebac4655 JA |
335 | ret = 1; |
336 | } | |
337 | ||
338 | return ret; | |
339 | } | |
340 | ||
341 | static void fill_crc32(struct verify_header *hdr, void *p, unsigned int len) | |
342 | { | |
343 | hdr->crc32 = crc32(p, len); | |
344 | } | |
345 | ||
346 | static void fill_md5(struct verify_header *hdr, void *p, unsigned int len) | |
347 | { | |
348 | struct md5_ctx md5_ctx; | |
349 | ||
350 | memset(&md5_ctx, 0, sizeof(md5_ctx)); | |
351 | md5_update(&md5_ctx, p, len); | |
352 | memcpy(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash)); | |
353 | } | |
354 | ||
a6ccc7be | 355 | static int get_rw_ddir(struct thread_data *td) |
3d60d1ed | 356 | { |
a6ccc7be JA |
357 | if (td_rw(td)) { |
358 | struct timeval now; | |
359 | unsigned long elapsed; | |
3d60d1ed | 360 | |
a6ccc7be JA |
361 | gettimeofday(&now, NULL); |
362 | elapsed = mtime_since_now(&td->rwmix_switch); | |
3d60d1ed | 363 | |
a6ccc7be JA |
364 | /* |
365 | * Check if it's time to seed a new data direction. | |
366 | */ | |
367 | if (elapsed >= td->rwmixcycle) { | |
c1ee2ca4 | 368 | int v; |
a6ccc7be | 369 | long r; |
3d60d1ed | 370 | |
c1ee2ca4 JA |
371 | r = os_random_long(&td->rwmix_state); |
372 | v = 1 + (int) (100.0 * (r / (RAND_MAX + 1.0))); | |
a6ccc7be JA |
373 | if (v < td->rwmixread) |
374 | td->rwmix_ddir = DDIR_READ; | |
375 | else | |
376 | td->rwmix_ddir = DDIR_WRITE; | |
377 | memcpy(&td->rwmix_switch, &now, sizeof(now)); | |
378 | } | |
379 | return td->rwmix_ddir; | |
380 | } else if (td_read(td)) | |
3d60d1ed JA |
381 | return DDIR_READ; |
382 | else | |
383 | return DDIR_WRITE; | |
384 | } | |
385 | ||
ebac4655 JA |
386 | /* |
387 | * fill body of io_u->buf with random data and add a header with the | |
22f78b32 | 388 | * crc32 or md5 sum of that data. |
ebac4655 JA |
389 | */ |
390 | static void populate_io_u(struct thread_data *td, struct io_u *io_u) | |
391 | { | |
392 | unsigned char *p = (unsigned char *) io_u->buf; | |
393 | struct verify_header hdr; | |
394 | ||
395 | hdr.fio_magic = FIO_HDR_MAGIC; | |
396 | hdr.len = io_u->buflen; | |
397 | p += sizeof(hdr); | |
398 | fill_random_bytes(td, p, io_u->buflen - sizeof(hdr)); | |
399 | ||
400 | if (td->verify == VERIFY_MD5) { | |
401 | fill_md5(&hdr, p, io_u->buflen - sizeof(hdr)); | |
402 | hdr.verify_type = VERIFY_MD5; | |
403 | } else { | |
404 | fill_crc32(&hdr, p, io_u->buflen - sizeof(hdr)); | |
405 | hdr.verify_type = VERIFY_CRC32; | |
406 | } | |
407 | ||
408 | memcpy(io_u->buf, &hdr, sizeof(hdr)); | |
409 | } | |
410 | ||
aea47d44 | 411 | static int td_io_prep(struct thread_data *td, struct io_u *io_u) |
20dc95c4 | 412 | { |
20dc95c4 JA |
413 | if (td->io_prep && td->io_prep(td, io_u)) |
414 | return 1; | |
415 | ||
416 | return 0; | |
417 | } | |
418 | ||
b1ff3403 | 419 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
ebac4655 JA |
420 | { |
421 | list_del(&io_u->list); | |
422 | list_add(&io_u->list, &td->io_u_freelist); | |
423 | td->cur_depth--; | |
424 | } | |
425 | ||
843a7413 JA |
426 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
427 | { | |
428 | /* | |
429 | * If using an iolog, grab next piece if any available. | |
430 | */ | |
431 | if (td->read_iolog) | |
432 | return read_iolog_get(td, io_u); | |
433 | ||
aea47d44 JA |
434 | /* |
435 | * No log, let the seq/rand engine retrieve the next position. | |
436 | */ | |
437 | if (!get_next_offset(td, &io_u->offset)) { | |
438 | io_u->buflen = get_next_buflen(td); | |
439 | ||
440 | if (io_u->buflen) { | |
441 | io_u->ddir = get_rw_ddir(td); | |
843a7413 JA |
442 | |
443 | /* | |
444 | * If using a write iolog, store this entry. | |
445 | */ | |
446 | if (td->write_iolog) | |
447 | write_iolog_put(td, io_u); | |
448 | ||
aea47d44 JA |
449 | return 0; |
450 | } | |
451 | } | |
452 | ||
453 | return 1; | |
454 | } | |
455 | ||
22f78b32 | 456 | #define queue_full(td) list_empty(&(td)->io_u_freelist) |
ebac4655 | 457 | |
b1ff3403 | 458 | struct io_u *__get_io_u(struct thread_data *td) |
ebac4655 | 459 | { |
22f78b32 | 460 | struct io_u *io_u = NULL; |
ebac4655 | 461 | |
22f78b32 JA |
462 | if (!queue_full(td)) { |
463 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); | |
464 | ||
465 | io_u->error = 0; | |
466 | io_u->resid = 0; | |
467 | list_del(&io_u->list); | |
468 | list_add(&io_u->list, &td->io_u_busylist); | |
469 | td->cur_depth++; | |
470 | } | |
ebac4655 | 471 | |
ebac4655 JA |
472 | return io_u; |
473 | } | |
474 | ||
ebac4655 JA |
475 | static struct io_u *get_io_u(struct thread_data *td) |
476 | { | |
477 | struct io_u *io_u; | |
478 | ||
479 | io_u = __get_io_u(td); | |
480 | if (!io_u) | |
481 | return NULL; | |
482 | ||
20dc95c4 JA |
483 | if (td->zone_bytes >= td->zone_size) { |
484 | td->zone_bytes = 0; | |
485 | td->last_pos += td->zone_skip; | |
486 | } | |
487 | ||
aea47d44 | 488 | if (fill_io_u(td, io_u)) { |
ebac4655 JA |
489 | put_io_u(td, io_u); |
490 | return NULL; | |
491 | } | |
492 | ||
838a3cd3 JA |
493 | if (io_u->buflen + io_u->offset > td->real_file_size) |
494 | io_u->buflen = td->real_file_size - io_u->offset; | |
ebac4655 JA |
495 | |
496 | if (!io_u->buflen) { | |
497 | put_io_u(td, io_u); | |
498 | return NULL; | |
499 | } | |
500 | ||
843a7413 | 501 | if (!td->read_iolog && !td->sequential) |
ebac4655 JA |
502 | mark_random_map(td, io_u); |
503 | ||
20dc95c4 | 504 | td->last_pos += io_u->buflen; |
ebac4655 JA |
505 | |
506 | if (td->verify != VERIFY_NONE) | |
507 | populate_io_u(td, io_u); | |
508 | ||
aea47d44 | 509 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
510 | put_io_u(td, io_u); |
511 | return NULL; | |
512 | } | |
513 | ||
514 | gettimeofday(&io_u->start_time, NULL); | |
515 | return io_u; | |
516 | } | |
517 | ||
518 | static inline void td_set_runstate(struct thread_data *td, int runstate) | |
519 | { | |
ebac4655 JA |
520 | td->runstate = runstate; |
521 | } | |
522 | ||
aea47d44 | 523 | static int get_next_verify(struct thread_data *td, struct io_u *io_u) |
ebac4655 JA |
524 | { |
525 | struct io_piece *ipo; | |
526 | ||
22f78b32 JA |
527 | if (!list_empty(&td->io_hist_list)) { |
528 | ipo = list_entry(td->io_hist_list.next, struct io_piece, list); | |
ebac4655 | 529 | |
22f78b32 | 530 | list_del(&ipo->list); |
ebac4655 | 531 | |
22f78b32 JA |
532 | io_u->offset = ipo->offset; |
533 | io_u->buflen = ipo->len; | |
534 | io_u->ddir = DDIR_READ; | |
535 | free(ipo); | |
536 | return 0; | |
537 | } | |
538 | ||
539 | return 1; | |
ebac4655 JA |
540 | } |
541 | ||
ebac4655 JA |
542 | static int sync_td(struct thread_data *td) |
543 | { | |
544 | if (td->io_sync) | |
545 | return td->io_sync(td); | |
546 | ||
547 | return 0; | |
548 | } | |
549 | ||
550 | static int io_u_getevents(struct thread_data *td, int min, int max, | |
551 | struct timespec *t) | |
552 | { | |
553 | return td->io_getevents(td, min, max, t); | |
554 | } | |
555 | ||
556 | static int io_u_queue(struct thread_data *td, struct io_u *io_u) | |
557 | { | |
558 | gettimeofday(&io_u->issue_time, NULL); | |
559 | ||
560 | return td->io_queue(td, io_u); | |
561 | } | |
562 | ||
563 | #define iocb_time(iocb) ((unsigned long) (iocb)->data) | |
564 | ||
565 | static void io_completed(struct thread_data *td, struct io_u *io_u, | |
566 | struct io_completion_data *icd) | |
567 | { | |
568 | struct timeval e; | |
569 | unsigned long msec; | |
570 | ||
571 | gettimeofday(&e, NULL); | |
572 | ||
573 | if (!io_u->error) { | |
20dc95c4 | 574 | unsigned int bytes = io_u->buflen - io_u->resid; |
3d60d1ed | 575 | const int idx = io_u->ddir; |
ebac4655 JA |
576 | |
577 | td->io_blocks[idx]++; | |
20dc95c4 JA |
578 | td->io_bytes[idx] += bytes; |
579 | td->zone_bytes += bytes; | |
580 | td->this_io_bytes[idx] += bytes; | |
ebac4655 JA |
581 | |
582 | msec = mtime_since(&io_u->issue_time, &e); | |
583 | ||
3d60d1ed JA |
584 | add_clat_sample(td, idx, msec); |
585 | add_bw_sample(td, idx); | |
ebac4655 | 586 | |
3d60d1ed | 587 | if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) |
ebac4655 JA |
588 | log_io_piece(td, io_u); |
589 | ||
20dc95c4 | 590 | icd->bytes_done[idx] += bytes; |
ebac4655 JA |
591 | } else |
592 | icd->error = io_u->error; | |
593 | } | |
594 | ||
595 | static void ios_completed(struct thread_data *td,struct io_completion_data *icd) | |
596 | { | |
597 | struct io_u *io_u; | |
598 | int i; | |
599 | ||
600 | icd->error = 0; | |
601 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
602 | ||
603 | for (i = 0; i < icd->nr; i++) { | |
604 | io_u = td->io_event(td, i); | |
605 | ||
606 | io_completed(td, io_u, icd); | |
607 | put_io_u(td, io_u); | |
608 | } | |
609 | } | |
610 | ||
611 | static void cleanup_pending_aio(struct thread_data *td) | |
612 | { | |
613 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
614 | struct list_head *entry, *n; | |
615 | struct io_completion_data icd; | |
616 | struct io_u *io_u; | |
617 | int r; | |
618 | ||
619 | /* | |
620 | * get immediately available events, if any | |
621 | */ | |
622 | r = io_u_getevents(td, 0, td->cur_depth, &ts); | |
623 | if (r > 0) { | |
624 | icd.nr = r; | |
625 | ios_completed(td, &icd); | |
626 | } | |
627 | ||
628 | /* | |
629 | * now cancel remaining active events | |
630 | */ | |
631 | if (td->io_cancel) { | |
632 | list_for_each_safe(entry, n, &td->io_u_busylist) { | |
633 | io_u = list_entry(entry, struct io_u, list); | |
634 | ||
635 | r = td->io_cancel(td, io_u); | |
636 | if (!r) | |
637 | put_io_u(td, io_u); | |
638 | } | |
639 | } | |
640 | ||
641 | if (td->cur_depth) { | |
642 | r = io_u_getevents(td, td->cur_depth, td->cur_depth, NULL); | |
643 | if (r > 0) { | |
644 | icd.nr = r; | |
645 | ios_completed(td, &icd); | |
646 | } | |
647 | } | |
648 | } | |
649 | ||
650 | static int do_io_u_verify(struct thread_data *td, struct io_u **io_u) | |
651 | { | |
652 | struct io_u *v_io_u = *io_u; | |
653 | int ret = 0; | |
654 | ||
655 | if (v_io_u) { | |
656 | ret = verify_io_u(v_io_u); | |
657 | put_io_u(td, v_io_u); | |
658 | *io_u = NULL; | |
659 | } | |
660 | ||
661 | return ret; | |
662 | } | |
663 | ||
664 | static void do_verify(struct thread_data *td) | |
665 | { | |
666 | struct timeval t; | |
667 | struct io_u *io_u, *v_io_u = NULL; | |
668 | struct io_completion_data icd; | |
669 | int ret; | |
670 | ||
671 | td_set_runstate(td, TD_VERIFYING); | |
672 | ||
673 | do { | |
674 | if (td->terminate) | |
675 | break; | |
676 | ||
677 | gettimeofday(&t, NULL); | |
678 | if (runtime_exceeded(td, &t)) | |
679 | break; | |
680 | ||
681 | io_u = __get_io_u(td); | |
682 | if (!io_u) | |
683 | break; | |
684 | ||
aea47d44 | 685 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
686 | put_io_u(td, io_u); |
687 | break; | |
688 | } | |
689 | ||
aea47d44 | 690 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
691 | put_io_u(td, io_u); |
692 | break; | |
693 | } | |
694 | ||
695 | ret = io_u_queue(td, io_u); | |
696 | if (ret) { | |
697 | put_io_u(td, io_u); | |
698 | td_verror(td, ret); | |
699 | break; | |
700 | } | |
701 | ||
702 | /* | |
703 | * we have one pending to verify, do that while | |
704 | * we are doing io on the next one | |
705 | */ | |
706 | if (do_io_u_verify(td, &v_io_u)) | |
707 | break; | |
708 | ||
709 | ret = io_u_getevents(td, 1, 1, NULL); | |
710 | if (ret != 1) { | |
711 | if (ret < 0) | |
712 | td_verror(td, ret); | |
713 | break; | |
714 | } | |
715 | ||
716 | v_io_u = td->io_event(td, 0); | |
717 | icd.nr = 1; | |
718 | icd.error = 0; | |
719 | io_completed(td, v_io_u, &icd); | |
720 | ||
721 | if (icd.error) { | |
722 | td_verror(td, icd.error); | |
723 | put_io_u(td, v_io_u); | |
724 | v_io_u = NULL; | |
725 | break; | |
726 | } | |
727 | ||
ebac4655 JA |
728 | /* |
729 | * if we can't submit more io, we need to verify now | |
730 | */ | |
731 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
732 | break; | |
733 | ||
734 | } while (1); | |
735 | ||
736 | do_io_u_verify(td, &v_io_u); | |
737 | ||
738 | if (td->cur_depth) | |
739 | cleanup_pending_aio(td); | |
740 | ||
741 | td_set_runstate(td, TD_RUNNING); | |
742 | } | |
743 | ||
32cd46a0 JA |
744 | /* |
745 | * Main IO worker functions. It retrieves io_u's to process and queues | |
746 | * and reaps them, checking for rate and errors along the way. | |
747 | */ | |
ebac4655 JA |
748 | static void do_io(struct thread_data *td) |
749 | { | |
750 | struct io_completion_data icd; | |
751 | struct timeval s, e; | |
752 | unsigned long usec; | |
753 | ||
5853e5a8 JA |
754 | td_set_runstate(td, TD_RUNNING); |
755 | ||
ebac4655 JA |
756 | while (td->this_io_bytes[td->ddir] < td->io_size) { |
757 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
758 | struct timespec *timeout; | |
759 | int ret, min_evts = 0; | |
760 | struct io_u *io_u; | |
761 | ||
762 | if (td->terminate) | |
763 | break; | |
764 | ||
765 | io_u = get_io_u(td); | |
766 | if (!io_u) | |
767 | break; | |
768 | ||
769 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
770 | ||
771 | ret = io_u_queue(td, io_u); | |
772 | if (ret) { | |
773 | put_io_u(td, io_u); | |
774 | td_verror(td, ret); | |
775 | break; | |
776 | } | |
777 | ||
778 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
779 | ||
780 | if (td->cur_depth < td->iodepth) { | |
781 | timeout = &ts; | |
782 | min_evts = 0; | |
783 | } else { | |
784 | timeout = NULL; | |
785 | min_evts = 1; | |
786 | } | |
787 | ||
788 | ret = io_u_getevents(td, min_evts, td->cur_depth, timeout); | |
789 | if (ret < 0) { | |
790 | td_verror(td, ret); | |
791 | break; | |
792 | } else if (!ret) | |
793 | continue; | |
794 | ||
795 | icd.nr = ret; | |
796 | ios_completed(td, &icd); | |
797 | if (icd.error) { | |
798 | td_verror(td, icd.error); | |
799 | break; | |
800 | } | |
801 | ||
802 | /* | |
803 | * the rate is batched for now, it should work for batches | |
804 | * of completions except the very first one which may look | |
805 | * a little bursty | |
806 | */ | |
807 | gettimeofday(&e, NULL); | |
808 | usec = utime_since(&s, &e); | |
809 | ||
810 | rate_throttle(td, usec, icd.bytes_done[td->ddir]); | |
811 | ||
812 | if (check_min_rate(td, &e)) { | |
813 | td_verror(td, ENOMEM); | |
814 | break; | |
815 | } | |
816 | ||
817 | if (runtime_exceeded(td, &e)) | |
818 | break; | |
819 | ||
820 | if (td->thinktime) | |
821 | usec_sleep(td, td->thinktime); | |
822 | ||
823 | if (should_fsync(td) && td->fsync_blocks && | |
824 | (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0) | |
825 | sync_td(td); | |
826 | } | |
827 | ||
828 | if (td->cur_depth) | |
829 | cleanup_pending_aio(td); | |
830 | ||
5853e5a8 JA |
831 | if (should_fsync(td) && td->end_fsync) { |
832 | td_set_runstate(td, TD_FSYNCING); | |
ebac4655 | 833 | sync_td(td); |
5853e5a8 | 834 | } |
ebac4655 JA |
835 | } |
836 | ||
837 | static void cleanup_io(struct thread_data *td) | |
838 | { | |
839 | if (td->io_cleanup) | |
840 | td->io_cleanup(td); | |
841 | } | |
842 | ||
843 | static int init_io(struct thread_data *td) | |
844 | { | |
845 | if (td->io_engine == FIO_SYNCIO) | |
846 | return fio_syncio_init(td); | |
847 | else if (td->io_engine == FIO_MMAPIO) | |
848 | return fio_mmapio_init(td); | |
849 | else if (td->io_engine == FIO_LIBAIO) | |
850 | return fio_libaio_init(td); | |
851 | else if (td->io_engine == FIO_POSIXAIO) | |
852 | return fio_posixaio_init(td); | |
853 | else if (td->io_engine == FIO_SGIO) | |
854 | return fio_sgio_init(td); | |
8756e4d4 JA |
855 | else if (td->io_engine == FIO_SPLICEIO) |
856 | return fio_spliceio_init(td); | |
ebac4655 | 857 | else { |
3b70d7e5 | 858 | log_err("bad io_engine %d\n", td->io_engine); |
ebac4655 JA |
859 | return 1; |
860 | } | |
861 | } | |
862 | ||
863 | static void cleanup_io_u(struct thread_data *td) | |
864 | { | |
865 | struct list_head *entry, *n; | |
866 | struct io_u *io_u; | |
867 | ||
868 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
869 | io_u = list_entry(entry, struct io_u, list); | |
870 | ||
871 | list_del(&io_u->list); | |
872 | free(io_u); | |
873 | } | |
874 | ||
875 | if (td->mem_type == MEM_MALLOC) | |
876 | free(td->orig_buffer); | |
877 | else if (td->mem_type == MEM_SHM) { | |
878 | struct shmid_ds sbuf; | |
879 | ||
880 | shmdt(td->orig_buffer); | |
881 | shmctl(td->shm_id, IPC_RMID, &sbuf); | |
882 | } else if (td->mem_type == MEM_MMAP) | |
883 | munmap(td->orig_buffer, td->orig_buffer_size); | |
884 | else | |
3b70d7e5 | 885 | log_err("Bad memory type %d\n", td->mem_type); |
ebac4655 JA |
886 | |
887 | td->orig_buffer = NULL; | |
888 | } | |
889 | ||
890 | static int init_io_u(struct thread_data *td) | |
891 | { | |
892 | struct io_u *io_u; | |
893 | int i, max_units; | |
894 | char *p; | |
895 | ||
896 | if (td->io_engine & FIO_SYNCIO) | |
897 | max_units = 1; | |
898 | else | |
899 | max_units = td->iodepth; | |
900 | ||
901 | td->orig_buffer_size = td->max_bs * max_units + MASK; | |
902 | ||
903 | if (td->mem_type == MEM_MALLOC) | |
904 | td->orig_buffer = malloc(td->orig_buffer_size); | |
905 | else if (td->mem_type == MEM_SHM) { | |
906 | td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600); | |
907 | if (td->shm_id < 0) { | |
908 | td_verror(td, errno); | |
909 | perror("shmget"); | |
910 | return 1; | |
911 | } | |
912 | ||
913 | td->orig_buffer = shmat(td->shm_id, NULL, 0); | |
914 | if (td->orig_buffer == (void *) -1) { | |
915 | td_verror(td, errno); | |
916 | perror("shmat"); | |
917 | td->orig_buffer = NULL; | |
918 | return 1; | |
919 | } | |
920 | } else if (td->mem_type == MEM_MMAP) { | |
921 | td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
922 | if (td->orig_buffer == MAP_FAILED) { | |
923 | td_verror(td, errno); | |
924 | perror("mmap"); | |
925 | td->orig_buffer = NULL; | |
926 | return 1; | |
927 | } | |
928 | } | |
929 | ||
ebac4655 JA |
930 | p = ALIGN(td->orig_buffer); |
931 | for (i = 0; i < max_units; i++) { | |
932 | io_u = malloc(sizeof(*io_u)); | |
933 | memset(io_u, 0, sizeof(*io_u)); | |
934 | INIT_LIST_HEAD(&io_u->list); | |
935 | ||
936 | io_u->buf = p + td->max_bs * i; | |
b1ff3403 | 937 | io_u->index = i; |
ebac4655 JA |
938 | list_add(&io_u->list, &td->io_u_freelist); |
939 | } | |
940 | ||
941 | return 0; | |
942 | } | |
943 | ||
944 | static int create_file(struct thread_data *td, unsigned long long size, | |
945 | int extend) | |
946 | { | |
947 | unsigned long long left; | |
948 | unsigned int bs; | |
949 | int r, oflags; | |
950 | char *b; | |
951 | ||
952 | /* | |
953 | * unless specifically asked for overwrite, let normal io extend it | |
954 | */ | |
c20d25b4 JA |
955 | if (td_write(td) && !td->overwrite) { |
956 | td->real_file_size = size; | |
ebac4655 | 957 | return 0; |
c20d25b4 | 958 | } |
ebac4655 JA |
959 | |
960 | if (!size) { | |
3b70d7e5 | 961 | log_err("Need size for create\n"); |
ebac4655 JA |
962 | td_verror(td, EINVAL); |
963 | return 1; | |
964 | } | |
965 | ||
88c6ed80 JA |
966 | temp_stall_ts = 1; |
967 | ||
ebac4655 JA |
968 | if (!extend) { |
969 | oflags = O_CREAT | O_TRUNC; | |
eb8bbf48 | 970 | fprintf(f_out, "%s: Laying out IO file (%LuMiB)\n", td->name, size >> 20); |
ebac4655 JA |
971 | } else { |
972 | oflags = O_APPEND; | |
eb8bbf48 | 973 | fprintf(f_out, "%s: Extending IO file (%Lu -> %LuMiB)\n", td->name, (td->file_size - size) >> 20, td->file_size >> 20); |
ebac4655 JA |
974 | } |
975 | ||
976 | td->fd = open(td->file_name, O_WRONLY | oflags, 0644); | |
977 | if (td->fd < 0) { | |
978 | td_verror(td, errno); | |
c20d25b4 | 979 | goto done_noclose; |
ebac4655 JA |
980 | } |
981 | ||
982 | if (!extend && ftruncate(td->fd, td->file_size) == -1) { | |
983 | td_verror(td, errno); | |
c20d25b4 | 984 | goto done; |
ebac4655 JA |
985 | } |
986 | ||
987 | td->io_size = td->file_size; | |
988 | b = malloc(td->max_bs); | |
989 | memset(b, 0, td->max_bs); | |
990 | ||
991 | left = size; | |
992 | while (left && !td->terminate) { | |
993 | bs = td->max_bs; | |
994 | if (bs > left) | |
995 | bs = left; | |
996 | ||
997 | r = write(td->fd, b, bs); | |
998 | ||
999 | if (r == (int) bs) { | |
1000 | left -= bs; | |
1001 | continue; | |
1002 | } else { | |
1003 | if (r < 0) | |
1004 | td_verror(td, errno); | |
1005 | else | |
1006 | td_verror(td, EIO); | |
1007 | ||
1008 | break; | |
1009 | } | |
1010 | } | |
1011 | ||
1012 | if (td->terminate) | |
1013 | unlink(td->file_name); | |
1014 | else if (td->create_fsync) | |
1015 | fsync(td->fd); | |
1016 | ||
c20d25b4 JA |
1017 | free(b); |
1018 | done: | |
ebac4655 JA |
1019 | close(td->fd); |
1020 | td->fd = -1; | |
c20d25b4 JA |
1021 | done_noclose: |
1022 | temp_stall_ts = 0; | |
ebac4655 JA |
1023 | return 0; |
1024 | } | |
1025 | ||
1026 | static int file_size(struct thread_data *td) | |
1027 | { | |
1028 | struct stat st; | |
1029 | ||
c20d25b4 JA |
1030 | if (td->overwrite) { |
1031 | if (fstat(td->fd, &st) == -1) { | |
1032 | td_verror(td, errno); | |
1033 | return 1; | |
1034 | } | |
ebac4655 | 1035 | |
c20d25b4 | 1036 | td->real_file_size = st.st_size; |
838a3cd3 | 1037 | |
c20d25b4 JA |
1038 | if (!td->file_size || td->file_size > td->real_file_size) |
1039 | td->file_size = td->real_file_size; | |
1040 | } | |
ebac4655 | 1041 | |
e8e387c1 | 1042 | td->file_size -= td->file_offset; |
ebac4655 JA |
1043 | return 0; |
1044 | } | |
1045 | ||
1046 | static int bdev_size(struct thread_data *td) | |
1047 | { | |
9104f874 | 1048 | unsigned long long bytes; |
ebac4655 JA |
1049 | int r; |
1050 | ||
1051 | r = blockdev_size(td->fd, &bytes); | |
1052 | if (r) { | |
1053 | td_verror(td, r); | |
1054 | return 1; | |
1055 | } | |
1056 | ||
838a3cd3 JA |
1057 | td->real_file_size = bytes; |
1058 | ||
ebac4655 JA |
1059 | /* |
1060 | * no extend possibilities, so limit size to device size if too large | |
1061 | */ | |
838a3cd3 JA |
1062 | if (!td->file_size || td->file_size > td->real_file_size) |
1063 | td->file_size = td->real_file_size; | |
ebac4655 | 1064 | |
e8e387c1 | 1065 | td->file_size -= td->file_offset; |
ebac4655 JA |
1066 | return 0; |
1067 | } | |
1068 | ||
1069 | static int get_file_size(struct thread_data *td) | |
1070 | { | |
0af7b542 | 1071 | int ret = 0; |
ebac4655 JA |
1072 | |
1073 | if (td->filetype == FIO_TYPE_FILE) | |
1074 | ret = file_size(td); | |
0af7b542 | 1075 | else if (td->filetype == FIO_TYPE_BD) |
ebac4655 | 1076 | ret = bdev_size(td); |
0af7b542 JA |
1077 | else |
1078 | td->real_file_size = -1; | |
ebac4655 JA |
1079 | |
1080 | if (ret) | |
1081 | return ret; | |
1082 | ||
e8e387c1 | 1083 | if (td->file_offset > td->real_file_size) { |
3b70d7e5 | 1084 | log_err("%s: offset extends end (%Lu > %Lu)\n", td->name, td->file_offset, td->real_file_size); |
ebac4655 JA |
1085 | return 1; |
1086 | } | |
1087 | ||
838a3cd3 | 1088 | td->io_size = td->file_size; |
ebac4655 | 1089 | if (td->io_size == 0) { |
3b70d7e5 | 1090 | log_err("%s: no io blocks\n", td->name); |
ebac4655 JA |
1091 | td_verror(td, EINVAL); |
1092 | return 1; | |
1093 | } | |
1094 | ||
20dc95c4 JA |
1095 | if (!td->zone_size) |
1096 | td->zone_size = td->io_size; | |
1097 | ||
ebac4655 JA |
1098 | td->total_io_size = td->io_size * td->loops; |
1099 | return 0; | |
1100 | } | |
1101 | ||
1102 | static int setup_file_mmap(struct thread_data *td) | |
1103 | { | |
1104 | int flags; | |
1105 | ||
3d60d1ed JA |
1106 | if (td_rw(td)) |
1107 | flags = PROT_READ | PROT_WRITE; | |
1108 | else if (td_write(td)) { | |
ebac4655 JA |
1109 | flags = PROT_WRITE; |
1110 | ||
1111 | if (td->verify != VERIFY_NONE) | |
1112 | flags |= PROT_READ; | |
3d60d1ed JA |
1113 | } else |
1114 | flags = PROT_READ; | |
ebac4655 JA |
1115 | |
1116 | td->mmap = mmap(NULL, td->file_size, flags, MAP_SHARED, td->fd, td->file_offset); | |
1117 | if (td->mmap == MAP_FAILED) { | |
1118 | td->mmap = NULL; | |
1119 | td_verror(td, errno); | |
1120 | return 1; | |
1121 | } | |
1122 | ||
1123 | if (td->invalidate_cache) { | |
1124 | if (madvise(td->mmap, td->file_size, MADV_DONTNEED) < 0) { | |
1125 | td_verror(td, errno); | |
1126 | return 1; | |
1127 | } | |
1128 | } | |
1129 | ||
1130 | if (td->sequential) { | |
1131 | if (madvise(td->mmap, td->file_size, MADV_SEQUENTIAL) < 0) { | |
1132 | td_verror(td, errno); | |
1133 | return 1; | |
1134 | } | |
1135 | } else { | |
1136 | if (madvise(td->mmap, td->file_size, MADV_RANDOM) < 0) { | |
1137 | td_verror(td, errno); | |
1138 | return 1; | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | return 0; | |
1143 | } | |
1144 | ||
1145 | static int setup_file_plain(struct thread_data *td) | |
1146 | { | |
1147 | if (td->invalidate_cache) { | |
1148 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_DONTNEED) < 0) { | |
1149 | td_verror(td, errno); | |
1150 | return 1; | |
1151 | } | |
1152 | } | |
1153 | ||
1154 | if (td->sequential) { | |
1155 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_SEQUENTIAL) < 0) { | |
1156 | td_verror(td, errno); | |
1157 | return 1; | |
1158 | } | |
1159 | } else { | |
1160 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_RANDOM) < 0) { | |
1161 | td_verror(td, errno); | |
1162 | return 1; | |
1163 | } | |
1164 | } | |
1165 | ||
1166 | return 0; | |
1167 | } | |
1168 | ||
1169 | static int setup_file(struct thread_data *td) | |
1170 | { | |
1171 | struct stat st; | |
1172 | int flags = 0; | |
1173 | ||
1174 | if (stat(td->file_name, &st) == -1) { | |
1175 | if (errno != ENOENT) { | |
1176 | td_verror(td, errno); | |
1177 | return 1; | |
1178 | } | |
1179 | if (!td->create_file) { | |
1180 | td_verror(td, ENOENT); | |
1181 | return 1; | |
1182 | } | |
1183 | if (create_file(td, td->file_size, 0)) | |
1184 | return 1; | |
1185 | } else if (td->filetype == FIO_TYPE_FILE) { | |
6a0106a0 | 1186 | if (st.st_size < (off_t) td->file_size) { |
ebac4655 JA |
1187 | if (create_file(td, td->file_size - st.st_size, 1)) |
1188 | return 1; | |
1189 | } | |
1190 | } | |
1191 | ||
1192 | if (td->odirect) | |
2c0ecd28 | 1193 | flags |= OS_O_DIRECT; |
ebac4655 | 1194 | |
3d60d1ed | 1195 | if (td_write(td) || td_rw(td)) { |
ebac4655 JA |
1196 | if (td->filetype == FIO_TYPE_FILE) { |
1197 | if (!td->overwrite) | |
1198 | flags |= O_TRUNC; | |
1199 | ||
1200 | flags |= O_CREAT; | |
1201 | } | |
1202 | if (td->sync_io) | |
1203 | flags |= O_SYNC; | |
1204 | ||
1205 | flags |= O_RDWR; | |
1206 | ||
1207 | td->fd = open(td->file_name, flags, 0600); | |
3d60d1ed JA |
1208 | } else { |
1209 | if (td->filetype == FIO_TYPE_CHAR) | |
1210 | flags |= O_RDWR; | |
1211 | else | |
1212 | flags |= O_RDONLY; | |
1213 | ||
1214 | td->fd = open(td->file_name, flags); | |
ebac4655 JA |
1215 | } |
1216 | ||
1217 | if (td->fd == -1) { | |
1218 | td_verror(td, errno); | |
1219 | return 1; | |
1220 | } | |
1221 | ||
1222 | if (get_file_size(td)) | |
1223 | return 1; | |
1224 | ||
1225 | if (td->io_engine != FIO_MMAPIO) | |
1226 | return setup_file_plain(td); | |
1227 | else | |
1228 | return setup_file_mmap(td); | |
1229 | } | |
1230 | ||
da86774e JA |
1231 | static int switch_ioscheduler(struct thread_data *td) |
1232 | { | |
1233 | char tmp[256], tmp2[128]; | |
1234 | FILE *f; | |
1235 | int ret; | |
1236 | ||
1237 | sprintf(tmp, "%s/queue/scheduler", td->sysfs_root); | |
1238 | ||
1239 | f = fopen(tmp, "r+"); | |
1240 | if (!f) { | |
1241 | td_verror(td, errno); | |
1242 | return 1; | |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * Set io scheduler. | |
1247 | */ | |
1248 | ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f); | |
1249 | if (ferror(f) || ret != 1) { | |
1250 | td_verror(td, errno); | |
1251 | fclose(f); | |
1252 | return 1; | |
1253 | } | |
1254 | ||
1255 | rewind(f); | |
1256 | ||
1257 | /* | |
1258 | * Read back and check that the selected scheduler is now the default. | |
1259 | */ | |
1260 | ret = fread(tmp, 1, sizeof(tmp), f); | |
1261 | if (ferror(f) || ret < 0) { | |
1262 | td_verror(td, errno); | |
1263 | fclose(f); | |
1264 | return 1; | |
1265 | } | |
1266 | ||
1267 | sprintf(tmp2, "[%s]", td->ioscheduler); | |
1268 | if (!strstr(tmp, tmp2)) { | |
3b70d7e5 | 1269 | log_err("fio: io scheduler %s not found\n", td->ioscheduler); |
da86774e JA |
1270 | td_verror(td, EINVAL); |
1271 | fclose(f); | |
1272 | return 1; | |
1273 | } | |
1274 | ||
1275 | fclose(f); | |
1276 | return 0; | |
1277 | } | |
1278 | ||
ebac4655 JA |
1279 | static void clear_io_state(struct thread_data *td) |
1280 | { | |
1281 | if (td->io_engine == FIO_SYNCIO) | |
1282 | lseek(td->fd, SEEK_SET, 0); | |
1283 | ||
20dc95c4 | 1284 | td->last_pos = 0; |
ebac4655 JA |
1285 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
1286 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 1287 | td->zone_bytes = 0; |
ebac4655 JA |
1288 | |
1289 | if (td->file_map) | |
1290 | memset(td->file_map, 0, td->num_maps * sizeof(long)); | |
1291 | } | |
1292 | ||
ebac4655 JA |
1293 | static void *thread_main(void *data) |
1294 | { | |
1295 | struct thread_data *td = data; | |
ebac4655 JA |
1296 | |
1297 | if (!td->use_thread) | |
1298 | setsid(); | |
1299 | ||
1300 | td->pid = getpid(); | |
1301 | ||
aea47d44 JA |
1302 | INIT_LIST_HEAD(&td->io_u_freelist); |
1303 | INIT_LIST_HEAD(&td->io_u_busylist); | |
1304 | INIT_LIST_HEAD(&td->io_hist_list); | |
1305 | INIT_LIST_HEAD(&td->io_log_list); | |
1306 | ||
ebac4655 JA |
1307 | if (init_io_u(td)) |
1308 | goto err; | |
1309 | ||
1310 | if (fio_setaffinity(td) == -1) { | |
1311 | td_verror(td, errno); | |
1312 | goto err; | |
1313 | } | |
1314 | ||
1315 | if (init_io(td)) | |
1316 | goto err; | |
1317 | ||
aea47d44 JA |
1318 | if (init_iolog(td)) |
1319 | goto err; | |
1320 | ||
ebac4655 JA |
1321 | if (td->ioprio) { |
1322 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
1323 | td_verror(td, errno); | |
1324 | goto err; | |
1325 | } | |
1326 | } | |
1327 | ||
b6f4d880 JA |
1328 | if (nice(td->nice) < 0) { |
1329 | td_verror(td, errno); | |
1330 | goto err; | |
1331 | } | |
1332 | ||
75154845 JA |
1333 | if (init_random_state(td)) |
1334 | goto err; | |
1335 | ||
da86774e JA |
1336 | if (td->ioscheduler && switch_ioscheduler(td)) |
1337 | goto err; | |
1338 | ||
75154845 | 1339 | td_set_runstate(td, TD_INITIALIZED); |
bbfd6b00 JA |
1340 | fio_sem_up(&startup_sem); |
1341 | fio_sem_down(&td->mutex); | |
ebac4655 JA |
1342 | |
1343 | if (!td->create_serialize && setup_file(td)) | |
1344 | goto err; | |
1345 | ||
ebac4655 JA |
1346 | gettimeofday(&td->epoch, NULL); |
1347 | ||
4e0ba8af JA |
1348 | if (td->exec_prerun) |
1349 | system(td->exec_prerun); | |
1350 | ||
ebac4655 JA |
1351 | while (td->loops--) { |
1352 | getrusage(RUSAGE_SELF, &td->ru_start); | |
1353 | gettimeofday(&td->start, NULL); | |
1354 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); | |
1355 | ||
1356 | if (td->ratemin) | |
1357 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
1358 | ||
1359 | clear_io_state(td); | |
1360 | prune_io_piece_log(td); | |
1361 | ||
1362 | do_io(td); | |
1363 | ||
1364 | td->runtime[td->ddir] += mtime_since_now(&td->start); | |
aea47d44 | 1365 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
3d60d1ed JA |
1366 | td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; |
1367 | ||
ebac4655 JA |
1368 | update_rusage_stat(td); |
1369 | ||
1370 | if (td->error || td->terminate) | |
1371 | break; | |
1372 | ||
1373 | if (td->verify == VERIFY_NONE) | |
1374 | continue; | |
1375 | ||
1376 | clear_io_state(td); | |
1377 | gettimeofday(&td->start, NULL); | |
1378 | ||
1379 | do_verify(td); | |
1380 | ||
1381 | td->runtime[DDIR_READ] += mtime_since_now(&td->start); | |
1382 | ||
1383 | if (td->error || td->terminate) | |
1384 | break; | |
1385 | } | |
1386 | ||
ebac4655 JA |
1387 | if (td->bw_log) |
1388 | finish_log(td, td->bw_log, "bw"); | |
1389 | if (td->slat_log) | |
1390 | finish_log(td, td->slat_log, "slat"); | |
1391 | if (td->clat_log) | |
1392 | finish_log(td, td->clat_log, "clat"); | |
843a7413 JA |
1393 | if (td->write_iolog) |
1394 | write_iolog_close(td); | |
4e0ba8af JA |
1395 | if (td->exec_postrun) |
1396 | system(td->exec_postrun); | |
ebac4655 JA |
1397 | |
1398 | if (exitall_on_terminate) | |
1399 | terminate_threads(td->groupid); | |
1400 | ||
1401 | err: | |
1402 | if (td->fd != -1) { | |
1403 | close(td->fd); | |
1404 | td->fd = -1; | |
1405 | } | |
1406 | if (td->mmap) | |
1407 | munmap(td->mmap, td->file_size); | |
1408 | cleanup_io(td); | |
1409 | cleanup_io_u(td); | |
ebac4655 JA |
1410 | td_set_runstate(td, TD_EXITED); |
1411 | return NULL; | |
1412 | ||
1413 | } | |
1414 | ||
1415 | static void *fork_main(int shmid, int offset) | |
1416 | { | |
1417 | struct thread_data *td; | |
1418 | void *data; | |
1419 | ||
1420 | data = shmat(shmid, NULL, 0); | |
1421 | if (data == (void *) -1) { | |
1422 | perror("shmat"); | |
1423 | return NULL; | |
1424 | } | |
1425 | ||
1426 | td = data + offset * sizeof(struct thread_data); | |
1427 | thread_main(td); | |
1428 | shmdt(data); | |
1429 | return NULL; | |
1430 | } | |
1431 | ||
ebac4655 JA |
1432 | static void check_str_update(struct thread_data *td) |
1433 | { | |
1434 | char c = run_str[td->thread_number - 1]; | |
1435 | ||
ebac4655 JA |
1436 | switch (td->runstate) { |
1437 | case TD_REAPED: | |
1438 | c = '_'; | |
1439 | break; | |
1440 | case TD_EXITED: | |
1441 | c = 'E'; | |
1442 | break; | |
1443 | case TD_RUNNING: | |
3d60d1ed JA |
1444 | if (td_rw(td)) { |
1445 | if (td->sequential) | |
1446 | c = 'M'; | |
1447 | else | |
1448 | c = 'm'; | |
1449 | } else if (td_read(td)) { | |
ebac4655 JA |
1450 | if (td->sequential) |
1451 | c = 'R'; | |
1452 | else | |
1453 | c = 'r'; | |
1454 | } else { | |
1455 | if (td->sequential) | |
1456 | c = 'W'; | |
1457 | else | |
1458 | c = 'w'; | |
1459 | } | |
1460 | break; | |
1461 | case TD_VERIFYING: | |
1462 | c = 'V'; | |
1463 | break; | |
5853e5a8 JA |
1464 | case TD_FSYNCING: |
1465 | c = 'F'; | |
1466 | break; | |
ebac4655 JA |
1467 | case TD_CREATED: |
1468 | c = 'C'; | |
1469 | break; | |
75154845 JA |
1470 | case TD_INITIALIZED: |
1471 | c = 'I'; | |
1472 | break; | |
ebac4655 JA |
1473 | case TD_NOT_CREATED: |
1474 | c = 'P'; | |
1475 | break; | |
1476 | default: | |
3b70d7e5 | 1477 | log_err("state %d\n", td->runstate); |
ebac4655 JA |
1478 | } |
1479 | ||
1480 | run_str[td->thread_number - 1] = c; | |
ebac4655 JA |
1481 | } |
1482 | ||
5289b847 JA |
1483 | static void eta_to_str(char *str, int eta_sec) |
1484 | { | |
1485 | unsigned int d, h, m, s; | |
1486 | static int always_d, always_h; | |
1487 | ||
1488 | d = h = m = s = 0; | |
1489 | ||
1490 | s = eta_sec % 60; | |
1491 | eta_sec /= 60; | |
1492 | m = eta_sec % 60; | |
1493 | eta_sec /= 60; | |
1494 | h = eta_sec % 24; | |
1495 | eta_sec /= 24; | |
1496 | d = eta_sec; | |
1497 | ||
1498 | if (d || always_d) { | |
1499 | always_d = 1; | |
1500 | str += sprintf(str, "%02dd:", d); | |
1501 | } | |
1502 | if (h || always_h) { | |
1503 | always_h = 1; | |
1504 | str += sprintf(str, "%02dh:", h); | |
1505 | } | |
1506 | ||
1507 | str += sprintf(str, "%02dm:", m); | |
1508 | str += sprintf(str, "%02ds", s); | |
1509 | } | |
1510 | ||
6a0106a0 JA |
1511 | static int thread_eta(struct thread_data *td, unsigned long elapsed) |
1512 | { | |
1513 | unsigned long long bytes_total, bytes_done; | |
1514 | unsigned int eta_sec = 0; | |
1515 | ||
1516 | bytes_total = td->total_io_size; | |
3d60d1ed JA |
1517 | |
1518 | /* | |
1519 | * if writing, bytes_total will be twice the size. If mixing, | |
1520 | * assume a 50/50 split and thus bytes_total will be 50% larger. | |
1521 | */ | |
1522 | if (td->verify) { | |
1523 | if (td_rw(td)) | |
1524 | bytes_total = bytes_total * 3 / 2; | |
1525 | else | |
1526 | bytes_total <<= 1; | |
1527 | } | |
6a0106a0 JA |
1528 | if (td->zone_size && td->zone_skip) |
1529 | bytes_total /= (td->zone_skip / td->zone_size); | |
1530 | ||
1531 | if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) { | |
1532 | double perc; | |
8b611c34 | 1533 | |
6a0106a0 JA |
1534 | bytes_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE]; |
1535 | perc = (double) bytes_done / (double) bytes_total; | |
1536 | if (perc > 1.0) | |
1537 | perc = 1.0; | |
1538 | ||
1539 | eta_sec = (elapsed * (1.0 / perc)) - elapsed; | |
1540 | ||
8b611c34 | 1541 | if (td->timeout && eta_sec > (td->timeout - elapsed)) |
6a0106a0 | 1542 | eta_sec = td->timeout - elapsed; |
75154845 JA |
1543 | } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED |
1544 | || td->runstate == TD_INITIALIZED) { | |
6a0106a0 JA |
1545 | int t_eta = 0, r_eta = 0; |
1546 | ||
1547 | /* | |
1548 | * We can only guess - assume it'll run the full timeout | |
1549 | * if given, otherwise assume it'll run at the specified rate. | |
1550 | */ | |
1551 | if (td->timeout) | |
1552 | t_eta = td->timeout + td->start_delay - elapsed; | |
1553 | if (td->rate) { | |
1554 | r_eta = (bytes_total / 1024) / td->rate; | |
1555 | r_eta += td->start_delay - elapsed; | |
1556 | } | |
1557 | ||
1558 | if (r_eta && t_eta) | |
1559 | eta_sec = min(r_eta, t_eta); | |
1560 | else if (r_eta) | |
1561 | eta_sec = r_eta; | |
1562 | else if (t_eta) | |
1563 | eta_sec = t_eta; | |
1564 | else | |
972cfd25 | 1565 | eta_sec = 0; |
6a0106a0 JA |
1566 | } else { |
1567 | /* | |
5853e5a8 | 1568 | * thread is already done or waiting for fsync |
6a0106a0 JA |
1569 | */ |
1570 | eta_sec = 0; | |
1571 | } | |
1572 | ||
1573 | return eta_sec; | |
1574 | } | |
1575 | ||
ebac4655 JA |
1576 | static void print_thread_status(void) |
1577 | { | |
6a0106a0 | 1578 | unsigned long elapsed = time_since_now(&genesis); |
71a751ce | 1579 | int i, nr_running, nr_pending, t_rate, m_rate, *eta_secs, eta_sec; |
5289b847 | 1580 | char eta_str[32]; |
6a0106a0 JA |
1581 | double perc = 0.0; |
1582 | ||
c6ae0a5b | 1583 | if (temp_stall_ts || terse_output) |
88c6ed80 JA |
1584 | return; |
1585 | ||
6a0106a0 JA |
1586 | eta_secs = malloc(thread_number * sizeof(int)); |
1587 | memset(eta_secs, 0, thread_number * sizeof(int)); | |
ebac4655 | 1588 | |
71a751ce | 1589 | nr_pending = nr_running = t_rate = m_rate = 0; |
ebac4655 JA |
1590 | for (i = 0; i < thread_number; i++) { |
1591 | struct thread_data *td = &threads[i]; | |
1592 | ||
5853e5a8 JA |
1593 | if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING|| |
1594 | td->runstate == TD_FSYNCING) { | |
ebac4655 JA |
1595 | nr_running++; |
1596 | t_rate += td->rate; | |
1597 | m_rate += td->ratemin; | |
71a751ce JA |
1598 | } else if (td->runstate < TD_RUNNING) |
1599 | nr_pending++; | |
ebac4655 | 1600 | |
6a0106a0 JA |
1601 | if (elapsed >= 3) |
1602 | eta_secs[i] = thread_eta(td, elapsed); | |
8b611c34 JA |
1603 | else |
1604 | eta_secs[i] = INT_MAX; | |
ebac4655 JA |
1605 | |
1606 | check_str_update(td); | |
1607 | } | |
1608 | ||
6a0106a0 JA |
1609 | if (exitall_on_terminate) |
1610 | eta_sec = INT_MAX; | |
1611 | else | |
1612 | eta_sec = 0; | |
5289b847 | 1613 | |
6a0106a0 JA |
1614 | for (i = 0; i < thread_number; i++) { |
1615 | if (exitall_on_terminate) { | |
1616 | if (eta_secs[i] < eta_sec) | |
1617 | eta_sec = eta_secs[i]; | |
1618 | } else { | |
1619 | if (eta_secs[i] > eta_sec) | |
1620 | eta_sec = eta_secs[i]; | |
5289b847 | 1621 | } |
6a0106a0 | 1622 | } |
5289b847 | 1623 | |
8b611c34 | 1624 | if (eta_sec != INT_MAX && elapsed) { |
6a0106a0 JA |
1625 | perc = (double) elapsed / (double) (elapsed + eta_sec); |
1626 | eta_to_str(eta_str, eta_sec); | |
ebac4655 JA |
1627 | } |
1628 | ||
71a751ce | 1629 | if (!nr_running && !nr_pending) |
f44c1d46 JA |
1630 | return; |
1631 | ||
eb8bbf48 | 1632 | printf("Threads running: %d", nr_running); |
ebac4655 JA |
1633 | if (m_rate || t_rate) |
1634 | printf(", commitrate %d/%dKiB/sec", t_rate, m_rate); | |
8b611c34 | 1635 | if (eta_sec != INT_MAX) { |
6a0106a0 JA |
1636 | perc *= 100.0; |
1637 | printf(": [%s] [%3.2f%% done] [eta %s]", run_str, perc,eta_str); | |
1638 | } | |
5289b847 | 1639 | printf("\r"); |
ebac4655 | 1640 | fflush(stdout); |
6a0106a0 | 1641 | free(eta_secs); |
ebac4655 JA |
1642 | } |
1643 | ||
1644 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) | |
1645 | { | |
1646 | int i; | |
1647 | ||
1648 | /* | |
1649 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
1650 | */ | |
1651 | for (i = 0; i < thread_number; i++) { | |
1652 | struct thread_data *td = &threads[i]; | |
1653 | ||
1654 | if (td->runstate != TD_EXITED) | |
1655 | continue; | |
1656 | ||
1657 | td_set_runstate(td, TD_REAPED); | |
1658 | ||
1659 | if (td->use_thread) { | |
1660 | long ret; | |
1661 | ||
1662 | if (pthread_join(td->thread, (void *) &ret)) | |
1663 | perror("thread_join"); | |
1664 | } else | |
1665 | waitpid(td->pid, NULL, 0); | |
1666 | ||
1667 | (*nr_running)--; | |
1668 | (*m_rate) -= td->ratemin; | |
1669 | (*t_rate) -= td->rate; | |
1670 | } | |
1671 | } | |
1672 | ||
fcb6ade2 JA |
1673 | static void fio_unpin_memory(void *pinned) |
1674 | { | |
1675 | if (pinned) { | |
1676 | if (munlock(pinned, mlock_size) < 0) | |
1677 | perror("munlock"); | |
1678 | munmap(pinned, mlock_size); | |
1679 | } | |
1680 | } | |
1681 | ||
1682 | static void *fio_pin_memory(void) | |
1683 | { | |
32cd46a0 | 1684 | unsigned long long phys_mem; |
fcb6ade2 JA |
1685 | void *ptr; |
1686 | ||
1687 | if (!mlock_size) | |
1688 | return NULL; | |
1689 | ||
1690 | /* | |
1691 | * Don't allow mlock of more than real_mem-128MB | |
1692 | */ | |
32cd46a0 JA |
1693 | phys_mem = os_phys_mem(); |
1694 | if (phys_mem) { | |
1695 | if ((mlock_size + 128 * 1024 * 1024) > phys_mem) { | |
1696 | mlock_size = phys_mem - 128 * 1024 * 1024; | |
eb8bbf48 | 1697 | fprintf(f_out, "fio: limiting mlocked memory to %lluMiB\n", mlock_size >> 20); |
fcb6ade2 JA |
1698 | } |
1699 | } | |
1700 | ||
1701 | ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
1702 | if (!ptr) { | |
1703 | perror("malloc locked mem"); | |
1704 | return NULL; | |
1705 | } | |
1706 | if (mlock(ptr, mlock_size) < 0) { | |
1707 | munmap(ptr, mlock_size); | |
1708 | perror("mlock"); | |
1709 | return NULL; | |
1710 | } | |
1711 | ||
1712 | return ptr; | |
1713 | } | |
1714 | ||
ebac4655 JA |
1715 | static void run_threads(void) |
1716 | { | |
ebac4655 JA |
1717 | struct thread_data *td; |
1718 | unsigned long spent; | |
1719 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 JA |
1720 | void *mlocked_mem; |
1721 | ||
1722 | mlocked_mem = fio_pin_memory(); | |
ebac4655 | 1723 | |
c6ae0a5b JA |
1724 | if (!terse_output) { |
1725 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
1726 | fflush(stdout); | |
1727 | } | |
c04f7ec3 | 1728 | |
4efa970e JA |
1729 | signal(SIGINT, sig_handler); |
1730 | signal(SIGALRM, sig_handler); | |
1731 | ||
ebac4655 JA |
1732 | todo = thread_number; |
1733 | nr_running = 0; | |
1734 | nr_started = 0; | |
1735 | m_rate = t_rate = 0; | |
1736 | ||
1737 | for (i = 0; i < thread_number; i++) { | |
1738 | td = &threads[i]; | |
1739 | ||
1740 | run_str[td->thread_number - 1] = 'P'; | |
1741 | ||
1742 | init_disk_util(td); | |
1743 | ||
1744 | if (!td->create_serialize) | |
1745 | continue; | |
1746 | ||
1747 | /* | |
1748 | * do file setup here so it happens sequentially, | |
1749 | * we don't want X number of threads getting their | |
1750 | * client data interspersed on disk | |
1751 | */ | |
1752 | if (setup_file(td)) { | |
1753 | td_set_runstate(td, TD_REAPED); | |
1754 | todo--; | |
1755 | } | |
1756 | } | |
1757 | ||
1758 | gettimeofday(&genesis, NULL); | |
1759 | ||
1760 | while (todo) { | |
75154845 JA |
1761 | struct thread_data *map[MAX_JOBS]; |
1762 | struct timeval this_start; | |
1763 | int this_jobs = 0, left; | |
1764 | ||
ebac4655 JA |
1765 | /* |
1766 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
1767 | */ | |
1768 | for (i = 0; i < thread_number; i++) { | |
1769 | td = &threads[i]; | |
1770 | ||
1771 | if (td->runstate != TD_NOT_CREATED) | |
1772 | continue; | |
1773 | ||
1774 | /* | |
1775 | * never got a chance to start, killed by other | |
1776 | * thread for some reason | |
1777 | */ | |
1778 | if (td->terminate) { | |
1779 | todo--; | |
1780 | continue; | |
1781 | } | |
1782 | ||
1783 | if (td->start_delay) { | |
1784 | spent = mtime_since_now(&genesis); | |
1785 | ||
1786 | if (td->start_delay * 1000 > spent) | |
1787 | continue; | |
1788 | } | |
1789 | ||
1790 | if (td->stonewall && (nr_started || nr_running)) | |
1791 | break; | |
1792 | ||
75154845 JA |
1793 | /* |
1794 | * Set state to created. Thread will transition | |
1795 | * to TD_INITIALIZED when it's done setting up. | |
1796 | */ | |
ebac4655 | 1797 | td_set_runstate(td, TD_CREATED); |
75154845 | 1798 | map[this_jobs++] = td; |
bbfd6b00 | 1799 | fio_sem_init(&startup_sem, 1); |
ebac4655 JA |
1800 | nr_started++; |
1801 | ||
1802 | if (td->use_thread) { | |
1803 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
1804 | perror("thread_create"); | |
1805 | nr_started--; | |
1806 | } | |
1807 | } else { | |
1808 | if (fork()) | |
bbfd6b00 | 1809 | fio_sem_down(&startup_sem); |
ebac4655 JA |
1810 | else { |
1811 | fork_main(shm_id, i); | |
1812 | exit(0); | |
1813 | } | |
1814 | } | |
1815 | } | |
1816 | ||
1817 | /* | |
75154845 JA |
1818 | * Wait for the started threads to transition to |
1819 | * TD_INITIALIZED. | |
ebac4655 | 1820 | */ |
75154845 JA |
1821 | gettimeofday(&this_start, NULL); |
1822 | left = this_jobs; | |
1823 | while (left) { | |
1824 | if (mtime_since_now(&this_start) > JOB_START_TIMEOUT) | |
1825 | break; | |
1826 | ||
1827 | usleep(100000); | |
1828 | ||
1829 | for (i = 0; i < this_jobs; i++) { | |
1830 | td = map[i]; | |
1831 | if (!td) | |
1832 | continue; | |
b6f4d880 | 1833 | if (td->runstate == TD_INITIALIZED) { |
75154845 JA |
1834 | map[i] = NULL; |
1835 | left--; | |
b6f4d880 JA |
1836 | } else if (td->runstate >= TD_EXITED) { |
1837 | map[i] = NULL; | |
1838 | left--; | |
1839 | todo--; | |
1840 | nr_running++; /* work-around... */ | |
75154845 JA |
1841 | } |
1842 | } | |
1843 | } | |
1844 | ||
1845 | if (left) { | |
3b70d7e5 | 1846 | log_err("fio: %d jobs failed to start\n", left); |
75154845 JA |
1847 | for (i = 0; i < this_jobs; i++) { |
1848 | td = map[i]; | |
1849 | if (!td) | |
1850 | continue; | |
1851 | kill(td->pid, SIGTERM); | |
1852 | } | |
1853 | break; | |
1854 | } | |
1855 | ||
1856 | /* | |
b6f4d880 | 1857 | * start created threads (TD_INITIALIZED -> TD_RUNNING). |
75154845 | 1858 | */ |
ebac4655 JA |
1859 | for (i = 0; i < thread_number; i++) { |
1860 | td = &threads[i]; | |
1861 | ||
75154845 | 1862 | if (td->runstate != TD_INITIALIZED) |
ebac4655 JA |
1863 | continue; |
1864 | ||
1865 | td_set_runstate(td, TD_RUNNING); | |
1866 | nr_running++; | |
1867 | nr_started--; | |
1868 | m_rate += td->ratemin; | |
1869 | t_rate += td->rate; | |
75154845 | 1870 | todo--; |
bbfd6b00 | 1871 | fio_sem_up(&td->mutex); |
ebac4655 JA |
1872 | } |
1873 | ||
1874 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1875 | ||
1876 | if (todo) | |
1877 | usleep(100000); | |
1878 | } | |
1879 | ||
1880 | while (nr_running) { | |
1881 | reap_threads(&nr_running, &t_rate, &m_rate); | |
1882 | usleep(10000); | |
1883 | } | |
1884 | ||
1885 | update_io_ticks(); | |
fcb6ade2 | 1886 | fio_unpin_memory(mlocked_mem); |
ebac4655 JA |
1887 | } |
1888 | ||
ebac4655 JA |
1889 | int main(int argc, char *argv[]) |
1890 | { | |
1891 | if (parse_options(argc, argv)) | |
1892 | return 1; | |
1893 | ||
1894 | if (!thread_number) { | |
3b70d7e5 | 1895 | log_err("Nothing to do\n"); |
ebac4655 JA |
1896 | return 1; |
1897 | } | |
1898 | ||
1899 | disk_util_timer_arm(); | |
1900 | ||
1901 | run_threads(); | |
1902 | show_run_stats(); | |
1903 | ||
1904 | return 0; | |
1905 | } |