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