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 | */ | |
21 | #include <stdio.h> | |
22 | #include <stdlib.h> | |
23 | #include <unistd.h> | |
24 | #include <fcntl.h> | |
25 | #include <string.h> | |
26 | #include <errno.h> | |
27 | #include <signal.h> | |
28 | #include <time.h> | |
29 | #include <math.h> | |
30 | #include <assert.h> | |
31 | #include <dirent.h> | |
32 | #include <libgen.h> | |
33 | #include <sys/types.h> | |
34 | #include <sys/stat.h> | |
35 | #include <sys/wait.h> | |
36 | #include <sys/ipc.h> | |
37 | #include <sys/shm.h> | |
38 | #include <sys/ioctl.h> | |
39 | #include <sys/mman.h> | |
40 | ||
41 | #include "fio.h" | |
42 | #include "os.h" | |
43 | ||
44 | #define MASK (4095) | |
45 | ||
46 | #define ALIGN(buf) (char *) (((unsigned long) (buf) + MASK) & ~(MASK)) | |
47 | ||
48 | int groupid = 0; | |
49 | int thread_number = 0; | |
50 | static char run_str[MAX_JOBS + 1]; | |
51 | int shm_id = 0; | |
52 | static LIST_HEAD(disk_list); | |
53 | static struct itimerval itimer; | |
5289b847 | 54 | static struct timeval genesis; |
ebac4655 JA |
55 | |
56 | static void update_io_ticks(void); | |
57 | static void disk_util_timer_arm(void); | |
58 | static void print_thread_status(void); | |
59 | ||
c04f7ec3 JA |
60 | extern unsigned long long mlock_size; |
61 | ||
ebac4655 JA |
62 | /* |
63 | * thread life cycle | |
64 | */ | |
65 | enum { | |
66 | TD_NOT_CREATED = 0, | |
67 | TD_CREATED, | |
68 | TD_RUNNING, | |
69 | TD_VERIFYING, | |
70 | TD_EXITED, | |
71 | TD_REAPED, | |
72 | }; | |
73 | ||
3d60d1ed | 74 | #define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync)) |
ebac4655 JA |
75 | |
76 | static sem_t startup_sem; | |
77 | ||
78 | #define TERMINATE_ALL (-1) | |
79 | ||
80 | static void terminate_threads(int group_id) | |
81 | { | |
82 | int i; | |
83 | ||
84 | for (i = 0; i < thread_number; i++) { | |
85 | struct thread_data *td = &threads[i]; | |
86 | ||
87 | if (group_id == TERMINATE_ALL || groupid == td->groupid) { | |
88 | td->terminate = 1; | |
89 | td->start_delay = 0; | |
90 | } | |
91 | } | |
92 | } | |
93 | ||
94 | static void sig_handler(int sig) | |
95 | { | |
96 | switch (sig) { | |
97 | case SIGALRM: | |
98 | update_io_ticks(); | |
99 | disk_util_timer_arm(); | |
100 | print_thread_status(); | |
101 | break; | |
102 | default: | |
103 | printf("\nfio: terminating on signal\n"); | |
104 | fflush(stdout); | |
105 | terminate_threads(TERMINATE_ALL); | |
106 | break; | |
107 | } | |
108 | } | |
109 | ||
110 | static unsigned long utime_since(struct timeval *s, struct timeval *e) | |
111 | { | |
112 | double sec, usec; | |
113 | ||
114 | sec = e->tv_sec - s->tv_sec; | |
115 | usec = e->tv_usec - s->tv_usec; | |
116 | if (sec > 0 && usec < 0) { | |
117 | sec--; | |
118 | usec += 1000000; | |
119 | } | |
120 | ||
121 | sec *= (double) 1000000; | |
122 | ||
123 | return sec + usec; | |
124 | } | |
125 | ||
126 | static unsigned long utime_since_now(struct timeval *s) | |
127 | { | |
128 | struct timeval t; | |
129 | ||
130 | gettimeofday(&t, NULL); | |
131 | return utime_since(s, &t); | |
132 | } | |
133 | ||
134 | static unsigned long mtime_since(struct timeval *s, struct timeval *e) | |
135 | { | |
136 | double sec, usec; | |
137 | ||
138 | sec = e->tv_sec - s->tv_sec; | |
139 | usec = e->tv_usec - s->tv_usec; | |
140 | if (sec > 0 && usec < 0) { | |
141 | sec--; | |
142 | usec += 1000000; | |
143 | } | |
144 | ||
145 | sec *= (double) 1000; | |
146 | usec /= (double) 1000; | |
147 | ||
148 | return sec + usec; | |
149 | } | |
150 | ||
151 | static unsigned long mtime_since_now(struct timeval *s) | |
152 | { | |
153 | struct timeval t; | |
154 | ||
155 | gettimeofday(&t, NULL); | |
156 | return mtime_since(s, &t); | |
157 | } | |
158 | ||
159 | static inline unsigned long msec_now(struct timeval *s) | |
160 | { | |
161 | return s->tv_sec * 1000 + s->tv_usec / 1000; | |
162 | } | |
163 | ||
5289b847 JA |
164 | static unsigned long time_since_now(struct timeval *s) |
165 | { | |
166 | return mtime_since_now(s) / 1000; | |
167 | } | |
168 | ||
ebac4655 JA |
169 | static int random_map_free(struct thread_data *td, unsigned long long block) |
170 | { | |
171 | unsigned int idx = RAND_MAP_IDX(td, block); | |
172 | unsigned int bit = RAND_MAP_BIT(td, block); | |
173 | ||
174 | return (td->file_map[idx] & (1UL << bit)) == 0; | |
175 | } | |
176 | ||
177 | static int get_next_free_block(struct thread_data *td, unsigned long long *b) | |
178 | { | |
179 | int i; | |
180 | ||
181 | *b = 0; | |
182 | i = 0; | |
183 | while ((*b) * td->min_bs < td->io_size) { | |
184 | if (td->file_map[i] != -1UL) { | |
185 | *b += ffz(td->file_map[i]); | |
186 | return 0; | |
187 | } | |
188 | ||
189 | *b += BLOCKS_PER_MAP; | |
190 | i++; | |
191 | } | |
192 | ||
193 | return 1; | |
194 | } | |
195 | ||
196 | static void mark_random_map(struct thread_data *td, struct io_u *io_u) | |
197 | { | |
198 | unsigned long block = io_u->offset / td->min_bs; | |
199 | unsigned int blocks = 0; | |
200 | ||
201 | while (blocks < (io_u->buflen / td->min_bs)) { | |
202 | unsigned int idx, bit; | |
203 | ||
204 | if (!random_map_free(td, block)) | |
205 | break; | |
206 | ||
207 | idx = RAND_MAP_IDX(td, block); | |
208 | bit = RAND_MAP_BIT(td, block); | |
209 | ||
210 | assert(idx < td->num_maps); | |
211 | ||
212 | td->file_map[idx] |= (1UL << bit); | |
213 | block++; | |
214 | blocks++; | |
215 | } | |
216 | ||
217 | if ((blocks * td->min_bs) < io_u->buflen) | |
218 | io_u->buflen = blocks * td->min_bs; | |
219 | } | |
220 | ||
ebac4655 JA |
221 | static inline void add_stat_sample(struct io_stat *is, unsigned long val) |
222 | { | |
223 | if (val > is->max_val) | |
224 | is->max_val = val; | |
225 | if (val < is->min_val) | |
226 | is->min_val = val; | |
227 | ||
228 | is->val += val; | |
229 | is->val_sq += val * val; | |
230 | is->samples++; | |
231 | } | |
232 | ||
233 | static void add_log_sample(struct thread_data *td, struct io_log *iolog, | |
234 | unsigned long val, int ddir) | |
235 | { | |
236 | if (iolog->nr_samples == iolog->max_samples) { | |
237 | int new_size = sizeof(struct io_sample) * iolog->max_samples*2; | |
238 | ||
239 | iolog->log = realloc(iolog->log, new_size); | |
240 | iolog->max_samples <<= 1; | |
241 | } | |
242 | ||
243 | iolog->log[iolog->nr_samples].val = val; | |
244 | iolog->log[iolog->nr_samples].time = mtime_since_now(&td->epoch); | |
245 | iolog->log[iolog->nr_samples].ddir = ddir; | |
246 | iolog->nr_samples++; | |
247 | } | |
248 | ||
249 | static void add_clat_sample(struct thread_data *td, int ddir,unsigned long msec) | |
250 | { | |
251 | add_stat_sample(&td->clat_stat[ddir], msec); | |
252 | ||
253 | if (td->clat_log) | |
254 | add_log_sample(td, td->clat_log, msec, ddir); | |
255 | } | |
256 | ||
257 | static void add_slat_sample(struct thread_data *td, int ddir,unsigned long msec) | |
258 | { | |
259 | add_stat_sample(&td->slat_stat[ddir], msec); | |
260 | ||
261 | if (td->slat_log) | |
262 | add_log_sample(td, td->slat_log, msec, ddir); | |
263 | } | |
264 | ||
265 | static void add_bw_sample(struct thread_data *td, int ddir) | |
266 | { | |
267 | unsigned long spent = mtime_since_now(&td->stat_sample_time[ddir]); | |
268 | unsigned long rate; | |
269 | ||
270 | if (spent < td->bw_avg_time) | |
271 | return; | |
272 | ||
273 | rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) / spent; | |
274 | add_stat_sample(&td->bw_stat[ddir], rate); | |
275 | ||
276 | if (td->bw_log) | |
277 | add_log_sample(td, td->bw_log, rate, ddir); | |
278 | ||
279 | gettimeofday(&td->stat_sample_time[ddir], NULL); | |
280 | td->stat_io_bytes[ddir] = td->this_io_bytes[ddir]; | |
281 | } | |
282 | ||
20dc95c4 JA |
283 | static int get_next_offset(struct thread_data *td, unsigned long long *offset) |
284 | { | |
285 | unsigned long long b, rb; | |
286 | long r; | |
287 | ||
288 | if (!td->sequential) { | |
289 | unsigned long max_blocks = td->io_size / td->min_bs; | |
290 | int loops = 50; | |
291 | ||
292 | do { | |
293 | lrand48_r(&td->random_state, &r); | |
294 | b = ((max_blocks - 1) * r / (RAND_MAX+1.0)); | |
295 | rb = b + (td->file_offset / td->min_bs); | |
296 | loops--; | |
297 | } while (!random_map_free(td, rb) && loops); | |
298 | ||
299 | if (!loops) { | |
300 | if (get_next_free_block(td, &b)) | |
301 | return 1; | |
302 | } | |
303 | } else | |
304 | b = td->last_pos / td->min_bs; | |
305 | ||
306 | *offset = (b * td->min_bs) + td->file_offset; | |
838a3cd3 | 307 | if (*offset > td->real_file_size) |
20dc95c4 JA |
308 | return 1; |
309 | ||
310 | return 0; | |
311 | } | |
312 | ||
313 | static unsigned int get_next_buflen(struct thread_data *td) | |
314 | { | |
315 | unsigned int buflen; | |
316 | long r; | |
317 | ||
318 | if (td->min_bs == td->max_bs) | |
319 | buflen = td->min_bs; | |
320 | else { | |
321 | lrand48_r(&td->bsrange_state, &r); | |
322 | buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0)); | |
323 | buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1); | |
324 | } | |
325 | ||
326 | if (buflen > td->io_size - td->this_io_bytes[td->ddir]) | |
327 | buflen = td->io_size - td->this_io_bytes[td->ddir]; | |
328 | ||
329 | return buflen; | |
330 | } | |
331 | ||
ebac4655 JA |
332 | /* |
333 | * busy looping version for the last few usec | |
334 | */ | |
335 | static void __usec_sleep(unsigned int usec) | |
336 | { | |
337 | struct timeval start; | |
338 | ||
339 | gettimeofday(&start, NULL); | |
340 | while (utime_since_now(&start) < usec) | |
341 | nop; | |
342 | } | |
343 | ||
344 | static void usec_sleep(struct thread_data *td, unsigned long usec) | |
345 | { | |
346 | struct timespec req, rem; | |
347 | ||
348 | req.tv_sec = usec / 1000000; | |
349 | req.tv_nsec = usec * 1000 - req.tv_sec * 1000000; | |
350 | ||
351 | do { | |
352 | if (usec < 5000) { | |
353 | __usec_sleep(usec); | |
354 | break; | |
355 | } | |
356 | ||
357 | rem.tv_sec = rem.tv_nsec = 0; | |
358 | if (nanosleep(&req, &rem) < 0) | |
359 | break; | |
360 | ||
361 | if ((rem.tv_sec + rem.tv_nsec) == 0) | |
362 | break; | |
363 | ||
364 | req.tv_nsec = rem.tv_nsec; | |
365 | req.tv_sec = rem.tv_sec; | |
366 | ||
367 | usec = rem.tv_sec * 1000000 + rem.tv_nsec / 1000; | |
368 | } while (!td->terminate); | |
369 | } | |
370 | ||
371 | static void rate_throttle(struct thread_data *td, unsigned long time_spent, | |
372 | unsigned int bytes) | |
373 | { | |
374 | unsigned long usec_cycle; | |
375 | ||
376 | if (!td->rate) | |
377 | return; | |
378 | ||
379 | usec_cycle = td->rate_usec_cycle * (bytes / td->min_bs); | |
380 | ||
381 | if (time_spent < usec_cycle) { | |
382 | unsigned long s = usec_cycle - time_spent; | |
383 | ||
384 | td->rate_pending_usleep += s; | |
385 | if (td->rate_pending_usleep >= 100000) { | |
386 | usec_sleep(td, td->rate_pending_usleep); | |
387 | td->rate_pending_usleep = 0; | |
388 | } | |
389 | } else { | |
390 | long overtime = time_spent - usec_cycle; | |
391 | ||
392 | td->rate_pending_usleep -= overtime; | |
393 | } | |
394 | } | |
395 | ||
396 | static int check_min_rate(struct thread_data *td, struct timeval *now) | |
397 | { | |
398 | unsigned long spent; | |
399 | unsigned long rate; | |
400 | int ddir = td->ddir; | |
401 | ||
402 | /* | |
403 | * allow a 2 second settle period in the beginning | |
404 | */ | |
405 | if (mtime_since(&td->start, now) < 2000) | |
406 | return 0; | |
407 | ||
408 | /* | |
409 | * if rate blocks is set, sample is running | |
410 | */ | |
411 | if (td->rate_bytes) { | |
412 | spent = mtime_since(&td->lastrate, now); | |
413 | if (spent < td->ratecycle) | |
414 | return 0; | |
415 | ||
416 | rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent; | |
417 | if (rate < td->ratemin) { | |
418 | printf("Client%d: min rate %d not met, got %ldKiB/sec\n", td->thread_number, td->ratemin, rate); | |
419 | if (rate_quit) | |
420 | terminate_threads(td->groupid); | |
421 | return 1; | |
422 | } | |
423 | } | |
424 | ||
425 | td->rate_bytes = td->this_io_bytes[ddir]; | |
426 | memcpy(&td->lastrate, now, sizeof(*now)); | |
427 | return 0; | |
428 | } | |
429 | ||
430 | static inline int runtime_exceeded(struct thread_data *td, struct timeval *t) | |
431 | { | |
432 | if (!td->timeout) | |
433 | return 0; | |
434 | if (mtime_since(&td->epoch, t) >= td->timeout * 1000) | |
435 | return 1; | |
436 | ||
437 | return 0; | |
438 | } | |
439 | ||
440 | static void fill_random_bytes(struct thread_data *td, | |
441 | unsigned char *p, unsigned int len) | |
442 | { | |
443 | unsigned int todo; | |
444 | double r; | |
445 | ||
446 | while (len) { | |
447 | drand48_r(&td->verify_state, &r); | |
448 | ||
449 | /* | |
450 | * lrand48_r seems to be broken and only fill the bottom | |
451 | * 32-bits, even on 64-bit archs with 64-bit longs | |
452 | */ | |
453 | todo = sizeof(r); | |
454 | if (todo > len) | |
455 | todo = len; | |
456 | ||
457 | memcpy(p, &r, todo); | |
458 | ||
459 | len -= todo; | |
460 | p += todo; | |
461 | } | |
462 | } | |
463 | ||
464 | static void hexdump(void *buffer, int len) | |
465 | { | |
466 | unsigned char *p = buffer; | |
467 | int i; | |
468 | ||
469 | for (i = 0; i < len; i++) | |
470 | printf("%02x", p[i]); | |
471 | printf("\n"); | |
472 | } | |
473 | ||
474 | static int verify_io_u_crc32(struct verify_header *hdr, struct io_u *io_u) | |
475 | { | |
476 | unsigned char *p = (unsigned char *) io_u->buf; | |
477 | unsigned long c; | |
478 | int ret; | |
479 | ||
480 | p += sizeof(*hdr); | |
481 | c = crc32(p, hdr->len - sizeof(*hdr)); | |
482 | ret = c != hdr->crc32; | |
483 | ||
484 | if (ret) { | |
485 | fprintf(stderr, "crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen); | |
486 | fprintf(stderr, "crc32: wanted %lx, got %lx\n", hdr->crc32, c); | |
487 | } | |
488 | ||
489 | return ret; | |
490 | } | |
491 | ||
492 | static int verify_io_u_md5(struct verify_header *hdr, struct io_u *io_u) | |
493 | { | |
494 | unsigned char *p = (unsigned char *) io_u->buf; | |
495 | struct md5_ctx md5_ctx; | |
496 | int ret; | |
497 | ||
498 | memset(&md5_ctx, 0, sizeof(md5_ctx)); | |
499 | p += sizeof(*hdr); | |
500 | md5_update(&md5_ctx, p, hdr->len - sizeof(*hdr)); | |
501 | ||
502 | ret = memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash)); | |
503 | if (ret) { | |
504 | fprintf(stderr, "md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen); | |
505 | hexdump(hdr->md5_digest, sizeof(hdr->md5_digest)); | |
506 | hexdump(md5_ctx.hash, sizeof(md5_ctx.hash)); | |
507 | } | |
508 | ||
509 | return ret; | |
510 | } | |
511 | ||
512 | static int verify_io_u(struct io_u *io_u) | |
513 | { | |
514 | struct verify_header *hdr = (struct verify_header *) io_u->buf; | |
515 | int ret; | |
516 | ||
517 | if (hdr->fio_magic != FIO_HDR_MAGIC) | |
518 | return 1; | |
519 | ||
520 | if (hdr->verify_type == VERIFY_MD5) | |
521 | ret = verify_io_u_md5(hdr, io_u); | |
522 | else if (hdr->verify_type == VERIFY_CRC32) | |
523 | ret = verify_io_u_crc32(hdr, io_u); | |
524 | else { | |
525 | fprintf(stderr, "Bad verify type %d\n", hdr->verify_type); | |
526 | ret = 1; | |
527 | } | |
528 | ||
529 | return ret; | |
530 | } | |
531 | ||
532 | static void fill_crc32(struct verify_header *hdr, void *p, unsigned int len) | |
533 | { | |
534 | hdr->crc32 = crc32(p, len); | |
535 | } | |
536 | ||
537 | static void fill_md5(struct verify_header *hdr, void *p, unsigned int len) | |
538 | { | |
539 | struct md5_ctx md5_ctx; | |
540 | ||
541 | memset(&md5_ctx, 0, sizeof(md5_ctx)); | |
542 | md5_update(&md5_ctx, p, len); | |
543 | memcpy(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash)); | |
544 | } | |
545 | ||
3d60d1ed JA |
546 | unsigned int hweight32(unsigned int w) |
547 | { | |
548 | unsigned int res = w - ((w >> 1) & 0x55555555); | |
549 | ||
550 | res = (res & 0x33333333) + ((res >> 2) & 0x33333333); | |
551 | res = (res + (res >> 4)) & 0x0F0F0F0F; | |
552 | res = res + (res >> 8); | |
553 | ||
554 | return (res + (res >> 16)) & 0x000000FF; | |
555 | } | |
556 | ||
557 | unsigned long hweight64(unsigned long long w) | |
558 | { | |
559 | #if __WORDSIZE == 32 | |
560 | return hweight32((unsigned int)(w >> 32)) + hweight32((unsigned int)w); | |
561 | #elif __WORDSIZE == 64 | |
562 | unsigned long long v = w - ((w >> 1) & 0x5555555555555555ul); | |
563 | ||
564 | v = (v & 0x3333333333333333ul) + ((v >> 2) & 0x3333333333333333ul); | |
565 | v = (v + (v >> 4)) & 0x0F0F0F0F0F0F0F0Ful; | |
566 | v = v + (v >> 8); | |
567 | v = v + (v >> 16); | |
568 | ||
569 | return (v + (v >> 32)) & 0x00000000000000FFul; | |
570 | #else | |
571 | #error __WORDSIZE not defined | |
572 | #endif | |
573 | } | |
574 | ||
575 | static int get_rw_ddir(struct thread_data *td) | |
576 | { | |
577 | /* | |
578 | * perhaps cheasy, but use the hamming weight of the position | |
579 | * as a randomizer for data direction. | |
580 | */ | |
581 | if (td_rw(td)) | |
582 | return hweight64(td->last_pos) & 1; | |
583 | else if (td_read(td)) | |
584 | return DDIR_READ; | |
585 | else | |
586 | return DDIR_WRITE; | |
587 | } | |
588 | ||
ebac4655 JA |
589 | /* |
590 | * fill body of io_u->buf with random data and add a header with the | |
591 | * (eg) sha1sum of that data. | |
592 | */ | |
593 | static void populate_io_u(struct thread_data *td, struct io_u *io_u) | |
594 | { | |
595 | unsigned char *p = (unsigned char *) io_u->buf; | |
596 | struct verify_header hdr; | |
597 | ||
598 | hdr.fio_magic = FIO_HDR_MAGIC; | |
599 | hdr.len = io_u->buflen; | |
600 | p += sizeof(hdr); | |
601 | fill_random_bytes(td, p, io_u->buflen - sizeof(hdr)); | |
602 | ||
603 | if (td->verify == VERIFY_MD5) { | |
604 | fill_md5(&hdr, p, io_u->buflen - sizeof(hdr)); | |
605 | hdr.verify_type = VERIFY_MD5; | |
606 | } else { | |
607 | fill_crc32(&hdr, p, io_u->buflen - sizeof(hdr)); | |
608 | hdr.verify_type = VERIFY_CRC32; | |
609 | } | |
610 | ||
611 | memcpy(io_u->buf, &hdr, sizeof(hdr)); | |
612 | } | |
613 | ||
aea47d44 | 614 | static int td_io_prep(struct thread_data *td, struct io_u *io_u) |
20dc95c4 | 615 | { |
20dc95c4 JA |
616 | if (td->io_prep && td->io_prep(td, io_u)) |
617 | return 1; | |
618 | ||
619 | return 0; | |
620 | } | |
621 | ||
b1ff3403 | 622 | void put_io_u(struct thread_data *td, struct io_u *io_u) |
ebac4655 JA |
623 | { |
624 | list_del(&io_u->list); | |
625 | list_add(&io_u->list, &td->io_u_freelist); | |
626 | td->cur_depth--; | |
627 | } | |
628 | ||
aea47d44 JA |
629 | static int fill_io_u(struct thread_data *td, struct io_u *io_u) |
630 | { | |
631 | /* | |
632 | * If using an iolog, grab next piece if any available. | |
633 | */ | |
634 | if (td->iolog) { | |
635 | struct io_piece *ipo; | |
636 | ||
637 | if (list_empty(&td->io_log_list)) | |
638 | return 1; | |
639 | ||
640 | ipo = list_entry(td->io_log_list.next, struct io_piece, list); | |
641 | list_del(&ipo->list); | |
642 | io_u->offset = ipo->offset; | |
643 | io_u->buflen = ipo->len; | |
644 | io_u->ddir = ipo->ddir; | |
645 | free(ipo); | |
646 | return 0; | |
647 | } | |
648 | ||
649 | /* | |
650 | * No log, let the seq/rand engine retrieve the next position. | |
651 | */ | |
652 | if (!get_next_offset(td, &io_u->offset)) { | |
653 | io_u->buflen = get_next_buflen(td); | |
654 | ||
655 | if (io_u->buflen) { | |
656 | io_u->ddir = get_rw_ddir(td); | |
657 | return 0; | |
658 | } | |
659 | } | |
660 | ||
661 | return 1; | |
662 | } | |
663 | ||
ebac4655 JA |
664 | #define queue_full(td) (list_empty(&(td)->io_u_freelist)) |
665 | ||
b1ff3403 | 666 | struct io_u *__get_io_u(struct thread_data *td) |
ebac4655 JA |
667 | { |
668 | struct io_u *io_u; | |
669 | ||
670 | if (queue_full(td)) | |
671 | return NULL; | |
672 | ||
673 | io_u = list_entry(td->io_u_freelist.next, struct io_u, list); | |
674 | io_u->error = 0; | |
675 | io_u->resid = 0; | |
676 | list_del(&io_u->list); | |
677 | list_add(&io_u->list, &td->io_u_busylist); | |
678 | td->cur_depth++; | |
679 | return io_u; | |
680 | } | |
681 | ||
ebac4655 JA |
682 | static struct io_u *get_io_u(struct thread_data *td) |
683 | { | |
684 | struct io_u *io_u; | |
685 | ||
686 | io_u = __get_io_u(td); | |
687 | if (!io_u) | |
688 | return NULL; | |
689 | ||
20dc95c4 JA |
690 | if (td->zone_bytes >= td->zone_size) { |
691 | td->zone_bytes = 0; | |
692 | td->last_pos += td->zone_skip; | |
693 | } | |
694 | ||
aea47d44 | 695 | if (fill_io_u(td, io_u)) { |
ebac4655 JA |
696 | put_io_u(td, io_u); |
697 | return NULL; | |
698 | } | |
699 | ||
838a3cd3 JA |
700 | if (io_u->buflen + io_u->offset > td->real_file_size) |
701 | io_u->buflen = td->real_file_size - io_u->offset; | |
ebac4655 JA |
702 | |
703 | if (!io_u->buflen) { | |
704 | put_io_u(td, io_u); | |
705 | return NULL; | |
706 | } | |
707 | ||
aea47d44 | 708 | if (!td->iolog && !td->sequential) |
ebac4655 JA |
709 | mark_random_map(td, io_u); |
710 | ||
20dc95c4 | 711 | td->last_pos += io_u->buflen; |
ebac4655 JA |
712 | |
713 | if (td->verify != VERIFY_NONE) | |
714 | populate_io_u(td, io_u); | |
715 | ||
aea47d44 | 716 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
717 | put_io_u(td, io_u); |
718 | return NULL; | |
719 | } | |
720 | ||
721 | gettimeofday(&io_u->start_time, NULL); | |
722 | return io_u; | |
723 | } | |
724 | ||
725 | static inline void td_set_runstate(struct thread_data *td, int runstate) | |
726 | { | |
727 | td->old_runstate = td->runstate; | |
728 | td->runstate = runstate; | |
729 | } | |
730 | ||
aea47d44 | 731 | static int get_next_verify(struct thread_data *td, struct io_u *io_u) |
ebac4655 JA |
732 | { |
733 | struct io_piece *ipo; | |
734 | ||
735 | if (list_empty(&td->io_hist_list)) | |
736 | return 1; | |
737 | ||
738 | ipo = list_entry(td->io_hist_list.next, struct io_piece, list); | |
739 | list_del(&ipo->list); | |
740 | ||
aea47d44 JA |
741 | io_u->offset = ipo->offset; |
742 | io_u->buflen = ipo->len; | |
743 | io_u->ddir = DDIR_READ; | |
ebac4655 JA |
744 | free(ipo); |
745 | return 0; | |
746 | } | |
747 | ||
748 | static void prune_io_piece_log(struct thread_data *td) | |
749 | { | |
750 | struct io_piece *ipo; | |
751 | ||
752 | while (!list_empty(&td->io_hist_list)) { | |
753 | ipo = list_entry(td->io_hist_list.next, struct io_piece, list); | |
754 | ||
755 | list_del(&ipo->list); | |
756 | free(ipo); | |
757 | } | |
758 | } | |
759 | ||
760 | /* | |
761 | * log a succesful write, so we can unwind the log for verify | |
762 | */ | |
763 | static void log_io_piece(struct thread_data *td, struct io_u *io_u) | |
764 | { | |
765 | struct io_piece *ipo = malloc(sizeof(struct io_piece)); | |
766 | struct list_head *entry; | |
767 | ||
768 | INIT_LIST_HEAD(&ipo->list); | |
769 | ipo->offset = io_u->offset; | |
770 | ipo->len = io_u->buflen; | |
771 | ||
772 | /* | |
773 | * for random io where the writes extend the file, it will typically | |
774 | * be laid out with the block scattered as written. it's faster to | |
775 | * read them in in that order again, so don't sort | |
776 | */ | |
777 | if (td->sequential || !td->overwrite) { | |
778 | list_add_tail(&ipo->list, &td->io_hist_list); | |
779 | return; | |
780 | } | |
781 | ||
782 | /* | |
783 | * for random io, sort the list so verify will run faster | |
784 | */ | |
785 | entry = &td->io_hist_list; | |
786 | while ((entry = entry->prev) != &td->io_hist_list) { | |
787 | struct io_piece *__ipo = list_entry(entry, struct io_piece, list); | |
788 | ||
789 | if (__ipo->offset < ipo->offset) | |
790 | break; | |
791 | } | |
792 | ||
793 | list_add(&ipo->list, entry); | |
794 | } | |
795 | ||
aea47d44 JA |
796 | static int init_iolog(struct thread_data *td) |
797 | { | |
798 | unsigned long long offset; | |
799 | unsigned int bytes; | |
800 | char *str, *p; | |
801 | FILE *f; | |
802 | int rw, i, reads, writes; | |
803 | ||
804 | if (!td->iolog) | |
805 | return 0; | |
806 | ||
807 | f = fopen(td->iolog_file, "r"); | |
808 | if (!f) { | |
809 | perror("fopen iolog"); | |
810 | return 1; | |
811 | } | |
812 | ||
813 | str = malloc(4096); | |
814 | reads = writes = i = 0; | |
815 | while ((p = fgets(str, 4096, f)) != NULL) { | |
816 | struct io_piece *ipo; | |
817 | ||
818 | if (sscanf(p, "%d,%llu,%u", &rw, &offset, &bytes) != 3) { | |
819 | fprintf(stderr, "bad iolog: %s\n", p); | |
820 | continue; | |
821 | } | |
822 | if (rw == DDIR_READ) | |
823 | reads++; | |
824 | else if (rw == DDIR_WRITE) | |
825 | writes++; | |
826 | else { | |
827 | fprintf(stderr, "bad ddir: %d\n", rw); | |
828 | continue; | |
829 | } | |
830 | ||
831 | ipo = malloc(sizeof(*ipo)); | |
832 | INIT_LIST_HEAD(&ipo->list); | |
833 | ipo->offset = offset; | |
834 | ipo->len = bytes; | |
835 | if (bytes > td->max_bs) | |
836 | td->max_bs = bytes; | |
837 | ipo->ddir = rw; | |
838 | list_add_tail(&ipo->list, &td->io_log_list); | |
839 | i++; | |
840 | } | |
841 | ||
842 | free(str); | |
843 | fclose(f); | |
844 | ||
845 | if (!i) | |
846 | return 1; | |
847 | ||
848 | if (reads && !writes) | |
849 | td->ddir = DDIR_READ; | |
850 | else if (!reads && writes) | |
851 | td->ddir = DDIR_READ; | |
852 | else | |
853 | td->iomix = 1; | |
854 | ||
855 | return 0; | |
856 | } | |
857 | ||
ebac4655 JA |
858 | static int sync_td(struct thread_data *td) |
859 | { | |
860 | if (td->io_sync) | |
861 | return td->io_sync(td); | |
862 | ||
863 | return 0; | |
864 | } | |
865 | ||
866 | static int io_u_getevents(struct thread_data *td, int min, int max, | |
867 | struct timespec *t) | |
868 | { | |
869 | return td->io_getevents(td, min, max, t); | |
870 | } | |
871 | ||
872 | static int io_u_queue(struct thread_data *td, struct io_u *io_u) | |
873 | { | |
874 | gettimeofday(&io_u->issue_time, NULL); | |
875 | ||
876 | return td->io_queue(td, io_u); | |
877 | } | |
878 | ||
879 | #define iocb_time(iocb) ((unsigned long) (iocb)->data) | |
880 | ||
881 | static void io_completed(struct thread_data *td, struct io_u *io_u, | |
882 | struct io_completion_data *icd) | |
883 | { | |
884 | struct timeval e; | |
885 | unsigned long msec; | |
886 | ||
887 | gettimeofday(&e, NULL); | |
888 | ||
889 | if (!io_u->error) { | |
20dc95c4 | 890 | unsigned int bytes = io_u->buflen - io_u->resid; |
3d60d1ed | 891 | const int idx = io_u->ddir; |
ebac4655 JA |
892 | |
893 | td->io_blocks[idx]++; | |
20dc95c4 JA |
894 | td->io_bytes[idx] += bytes; |
895 | td->zone_bytes += bytes; | |
896 | td->this_io_bytes[idx] += bytes; | |
ebac4655 JA |
897 | |
898 | msec = mtime_since(&io_u->issue_time, &e); | |
899 | ||
3d60d1ed JA |
900 | add_clat_sample(td, idx, msec); |
901 | add_bw_sample(td, idx); | |
ebac4655 | 902 | |
3d60d1ed | 903 | if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE) |
ebac4655 JA |
904 | log_io_piece(td, io_u); |
905 | ||
20dc95c4 | 906 | icd->bytes_done[idx] += bytes; |
ebac4655 JA |
907 | } else |
908 | icd->error = io_u->error; | |
909 | } | |
910 | ||
911 | static void ios_completed(struct thread_data *td,struct io_completion_data *icd) | |
912 | { | |
913 | struct io_u *io_u; | |
914 | int i; | |
915 | ||
916 | icd->error = 0; | |
917 | icd->bytes_done[0] = icd->bytes_done[1] = 0; | |
918 | ||
919 | for (i = 0; i < icd->nr; i++) { | |
920 | io_u = td->io_event(td, i); | |
921 | ||
922 | io_completed(td, io_u, icd); | |
923 | put_io_u(td, io_u); | |
924 | } | |
925 | } | |
926 | ||
927 | static void cleanup_pending_aio(struct thread_data *td) | |
928 | { | |
929 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
930 | struct list_head *entry, *n; | |
931 | struct io_completion_data icd; | |
932 | struct io_u *io_u; | |
933 | int r; | |
934 | ||
935 | /* | |
936 | * get immediately available events, if any | |
937 | */ | |
938 | r = io_u_getevents(td, 0, td->cur_depth, &ts); | |
939 | if (r > 0) { | |
940 | icd.nr = r; | |
941 | ios_completed(td, &icd); | |
942 | } | |
943 | ||
944 | /* | |
945 | * now cancel remaining active events | |
946 | */ | |
947 | if (td->io_cancel) { | |
948 | list_for_each_safe(entry, n, &td->io_u_busylist) { | |
949 | io_u = list_entry(entry, struct io_u, list); | |
950 | ||
951 | r = td->io_cancel(td, io_u); | |
952 | if (!r) | |
953 | put_io_u(td, io_u); | |
954 | } | |
955 | } | |
956 | ||
957 | if (td->cur_depth) { | |
958 | r = io_u_getevents(td, td->cur_depth, td->cur_depth, NULL); | |
959 | if (r > 0) { | |
960 | icd.nr = r; | |
961 | ios_completed(td, &icd); | |
962 | } | |
963 | } | |
964 | } | |
965 | ||
966 | static int do_io_u_verify(struct thread_data *td, struct io_u **io_u) | |
967 | { | |
968 | struct io_u *v_io_u = *io_u; | |
969 | int ret = 0; | |
970 | ||
971 | if (v_io_u) { | |
972 | ret = verify_io_u(v_io_u); | |
973 | put_io_u(td, v_io_u); | |
974 | *io_u = NULL; | |
975 | } | |
976 | ||
977 | return ret; | |
978 | } | |
979 | ||
980 | static void do_verify(struct thread_data *td) | |
981 | { | |
982 | struct timeval t; | |
983 | struct io_u *io_u, *v_io_u = NULL; | |
984 | struct io_completion_data icd; | |
985 | int ret; | |
986 | ||
987 | td_set_runstate(td, TD_VERIFYING); | |
988 | ||
989 | do { | |
990 | if (td->terminate) | |
991 | break; | |
992 | ||
993 | gettimeofday(&t, NULL); | |
994 | if (runtime_exceeded(td, &t)) | |
995 | break; | |
996 | ||
997 | io_u = __get_io_u(td); | |
998 | if (!io_u) | |
999 | break; | |
1000 | ||
aea47d44 | 1001 | if (get_next_verify(td, io_u)) { |
ebac4655 JA |
1002 | put_io_u(td, io_u); |
1003 | break; | |
1004 | } | |
1005 | ||
aea47d44 | 1006 | if (td_io_prep(td, io_u)) { |
ebac4655 JA |
1007 | put_io_u(td, io_u); |
1008 | break; | |
1009 | } | |
1010 | ||
1011 | ret = io_u_queue(td, io_u); | |
1012 | if (ret) { | |
1013 | put_io_u(td, io_u); | |
1014 | td_verror(td, ret); | |
1015 | break; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * we have one pending to verify, do that while | |
1020 | * we are doing io on the next one | |
1021 | */ | |
1022 | if (do_io_u_verify(td, &v_io_u)) | |
1023 | break; | |
1024 | ||
1025 | ret = io_u_getevents(td, 1, 1, NULL); | |
1026 | if (ret != 1) { | |
1027 | if (ret < 0) | |
1028 | td_verror(td, ret); | |
1029 | break; | |
1030 | } | |
1031 | ||
1032 | v_io_u = td->io_event(td, 0); | |
1033 | icd.nr = 1; | |
1034 | icd.error = 0; | |
1035 | io_completed(td, v_io_u, &icd); | |
1036 | ||
1037 | if (icd.error) { | |
1038 | td_verror(td, icd.error); | |
1039 | put_io_u(td, v_io_u); | |
1040 | v_io_u = NULL; | |
1041 | break; | |
1042 | } | |
1043 | ||
ebac4655 JA |
1044 | /* |
1045 | * if we can't submit more io, we need to verify now | |
1046 | */ | |
1047 | if (queue_full(td) && do_io_u_verify(td, &v_io_u)) | |
1048 | break; | |
1049 | ||
1050 | } while (1); | |
1051 | ||
1052 | do_io_u_verify(td, &v_io_u); | |
1053 | ||
1054 | if (td->cur_depth) | |
1055 | cleanup_pending_aio(td); | |
1056 | ||
1057 | td_set_runstate(td, TD_RUNNING); | |
1058 | } | |
1059 | ||
1060 | static void do_io(struct thread_data *td) | |
1061 | { | |
1062 | struct io_completion_data icd; | |
1063 | struct timeval s, e; | |
1064 | unsigned long usec; | |
1065 | ||
1066 | while (td->this_io_bytes[td->ddir] < td->io_size) { | |
1067 | struct timespec ts = { .tv_sec = 0, .tv_nsec = 0}; | |
1068 | struct timespec *timeout; | |
1069 | int ret, min_evts = 0; | |
1070 | struct io_u *io_u; | |
1071 | ||
1072 | if (td->terminate) | |
1073 | break; | |
1074 | ||
1075 | io_u = get_io_u(td); | |
1076 | if (!io_u) | |
1077 | break; | |
1078 | ||
1079 | memcpy(&s, &io_u->start_time, sizeof(s)); | |
1080 | ||
1081 | ret = io_u_queue(td, io_u); | |
1082 | if (ret) { | |
1083 | put_io_u(td, io_u); | |
1084 | td_verror(td, ret); | |
1085 | break; | |
1086 | } | |
1087 | ||
1088 | add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time)); | |
1089 | ||
1090 | if (td->cur_depth < td->iodepth) { | |
1091 | timeout = &ts; | |
1092 | min_evts = 0; | |
1093 | } else { | |
1094 | timeout = NULL; | |
1095 | min_evts = 1; | |
1096 | } | |
1097 | ||
1098 | ret = io_u_getevents(td, min_evts, td->cur_depth, timeout); | |
1099 | if (ret < 0) { | |
1100 | td_verror(td, ret); | |
1101 | break; | |
1102 | } else if (!ret) | |
1103 | continue; | |
1104 | ||
1105 | icd.nr = ret; | |
1106 | ios_completed(td, &icd); | |
1107 | if (icd.error) { | |
1108 | td_verror(td, icd.error); | |
1109 | break; | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * the rate is batched for now, it should work for batches | |
1114 | * of completions except the very first one which may look | |
1115 | * a little bursty | |
1116 | */ | |
1117 | gettimeofday(&e, NULL); | |
1118 | usec = utime_since(&s, &e); | |
1119 | ||
1120 | rate_throttle(td, usec, icd.bytes_done[td->ddir]); | |
1121 | ||
1122 | if (check_min_rate(td, &e)) { | |
1123 | td_verror(td, ENOMEM); | |
1124 | break; | |
1125 | } | |
1126 | ||
1127 | if (runtime_exceeded(td, &e)) | |
1128 | break; | |
1129 | ||
1130 | if (td->thinktime) | |
1131 | usec_sleep(td, td->thinktime); | |
1132 | ||
1133 | if (should_fsync(td) && td->fsync_blocks && | |
1134 | (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0) | |
1135 | sync_td(td); | |
1136 | } | |
1137 | ||
1138 | if (td->cur_depth) | |
1139 | cleanup_pending_aio(td); | |
1140 | ||
fc1a4713 | 1141 | if (should_fsync(td) && td->end_fsync) |
ebac4655 JA |
1142 | sync_td(td); |
1143 | } | |
1144 | ||
1145 | static void cleanup_io(struct thread_data *td) | |
1146 | { | |
1147 | if (td->io_cleanup) | |
1148 | td->io_cleanup(td); | |
1149 | } | |
1150 | ||
1151 | static int init_io(struct thread_data *td) | |
1152 | { | |
1153 | if (td->io_engine == FIO_SYNCIO) | |
1154 | return fio_syncio_init(td); | |
1155 | else if (td->io_engine == FIO_MMAPIO) | |
1156 | return fio_mmapio_init(td); | |
1157 | else if (td->io_engine == FIO_LIBAIO) | |
1158 | return fio_libaio_init(td); | |
1159 | else if (td->io_engine == FIO_POSIXAIO) | |
1160 | return fio_posixaio_init(td); | |
1161 | else if (td->io_engine == FIO_SGIO) | |
1162 | return fio_sgio_init(td); | |
8756e4d4 JA |
1163 | else if (td->io_engine == FIO_SPLICEIO) |
1164 | return fio_spliceio_init(td); | |
ebac4655 JA |
1165 | else { |
1166 | fprintf(stderr, "bad io_engine %d\n", td->io_engine); | |
1167 | return 1; | |
1168 | } | |
1169 | } | |
1170 | ||
1171 | static void cleanup_io_u(struct thread_data *td) | |
1172 | { | |
1173 | struct list_head *entry, *n; | |
1174 | struct io_u *io_u; | |
1175 | ||
1176 | list_for_each_safe(entry, n, &td->io_u_freelist) { | |
1177 | io_u = list_entry(entry, struct io_u, list); | |
1178 | ||
1179 | list_del(&io_u->list); | |
1180 | free(io_u); | |
1181 | } | |
1182 | ||
1183 | if (td->mem_type == MEM_MALLOC) | |
1184 | free(td->orig_buffer); | |
1185 | else if (td->mem_type == MEM_SHM) { | |
1186 | struct shmid_ds sbuf; | |
1187 | ||
1188 | shmdt(td->orig_buffer); | |
1189 | shmctl(td->shm_id, IPC_RMID, &sbuf); | |
1190 | } else if (td->mem_type == MEM_MMAP) | |
1191 | munmap(td->orig_buffer, td->orig_buffer_size); | |
1192 | else | |
1193 | fprintf(stderr, "Bad memory type %d\n", td->mem_type); | |
1194 | ||
1195 | td->orig_buffer = NULL; | |
1196 | } | |
1197 | ||
1198 | static int init_io_u(struct thread_data *td) | |
1199 | { | |
1200 | struct io_u *io_u; | |
1201 | int i, max_units; | |
1202 | char *p; | |
1203 | ||
1204 | if (td->io_engine & FIO_SYNCIO) | |
1205 | max_units = 1; | |
1206 | else | |
1207 | max_units = td->iodepth; | |
1208 | ||
1209 | td->orig_buffer_size = td->max_bs * max_units + MASK; | |
1210 | ||
1211 | if (td->mem_type == MEM_MALLOC) | |
1212 | td->orig_buffer = malloc(td->orig_buffer_size); | |
1213 | else if (td->mem_type == MEM_SHM) { | |
1214 | td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600); | |
1215 | if (td->shm_id < 0) { | |
1216 | td_verror(td, errno); | |
1217 | perror("shmget"); | |
1218 | return 1; | |
1219 | } | |
1220 | ||
1221 | td->orig_buffer = shmat(td->shm_id, NULL, 0); | |
1222 | if (td->orig_buffer == (void *) -1) { | |
1223 | td_verror(td, errno); | |
1224 | perror("shmat"); | |
1225 | td->orig_buffer = NULL; | |
1226 | return 1; | |
1227 | } | |
1228 | } else if (td->mem_type == MEM_MMAP) { | |
1229 | td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
1230 | if (td->orig_buffer == MAP_FAILED) { | |
1231 | td_verror(td, errno); | |
1232 | perror("mmap"); | |
1233 | td->orig_buffer = NULL; | |
1234 | return 1; | |
1235 | } | |
1236 | } | |
1237 | ||
ebac4655 JA |
1238 | p = ALIGN(td->orig_buffer); |
1239 | for (i = 0; i < max_units; i++) { | |
1240 | io_u = malloc(sizeof(*io_u)); | |
1241 | memset(io_u, 0, sizeof(*io_u)); | |
1242 | INIT_LIST_HEAD(&io_u->list); | |
1243 | ||
1244 | io_u->buf = p + td->max_bs * i; | |
b1ff3403 | 1245 | io_u->index = i; |
ebac4655 JA |
1246 | list_add(&io_u->list, &td->io_u_freelist); |
1247 | } | |
1248 | ||
1249 | return 0; | |
1250 | } | |
1251 | ||
1252 | static int create_file(struct thread_data *td, unsigned long long size, | |
1253 | int extend) | |
1254 | { | |
1255 | unsigned long long left; | |
1256 | unsigned int bs; | |
1257 | int r, oflags; | |
1258 | char *b; | |
1259 | ||
1260 | /* | |
1261 | * unless specifically asked for overwrite, let normal io extend it | |
1262 | */ | |
1263 | if (td_write(td) && !td->overwrite) | |
1264 | return 0; | |
1265 | ||
1266 | if (!size) { | |
1267 | fprintf(stderr, "Need size for create\n"); | |
1268 | td_verror(td, EINVAL); | |
1269 | return 1; | |
1270 | } | |
1271 | ||
1272 | if (!extend) { | |
1273 | oflags = O_CREAT | O_TRUNC; | |
1274 | printf("Client%d: Laying out IO file (%LuMiB)\n", td->thread_number, size >> 20); | |
1275 | } else { | |
1276 | oflags = O_APPEND; | |
1277 | printf("Client%d: Extending IO file (%Lu -> %LuMiB)\n", td->thread_number, (td->file_size - size) >> 20, td->file_size >> 20); | |
1278 | } | |
1279 | ||
1280 | td->fd = open(td->file_name, O_WRONLY | oflags, 0644); | |
1281 | if (td->fd < 0) { | |
1282 | td_verror(td, errno); | |
1283 | return 1; | |
1284 | } | |
1285 | ||
1286 | if (!extend && ftruncate(td->fd, td->file_size) == -1) { | |
1287 | td_verror(td, errno); | |
1288 | return 1; | |
1289 | } | |
1290 | ||
1291 | td->io_size = td->file_size; | |
1292 | b = malloc(td->max_bs); | |
1293 | memset(b, 0, td->max_bs); | |
1294 | ||
1295 | left = size; | |
1296 | while (left && !td->terminate) { | |
1297 | bs = td->max_bs; | |
1298 | if (bs > left) | |
1299 | bs = left; | |
1300 | ||
1301 | r = write(td->fd, b, bs); | |
1302 | ||
1303 | if (r == (int) bs) { | |
1304 | left -= bs; | |
1305 | continue; | |
1306 | } else { | |
1307 | if (r < 0) | |
1308 | td_verror(td, errno); | |
1309 | else | |
1310 | td_verror(td, EIO); | |
1311 | ||
1312 | break; | |
1313 | } | |
1314 | } | |
1315 | ||
1316 | if (td->terminate) | |
1317 | unlink(td->file_name); | |
1318 | else if (td->create_fsync) | |
1319 | fsync(td->fd); | |
1320 | ||
1321 | close(td->fd); | |
1322 | td->fd = -1; | |
1323 | free(b); | |
1324 | return 0; | |
1325 | } | |
1326 | ||
1327 | static int file_size(struct thread_data *td) | |
1328 | { | |
1329 | struct stat st; | |
1330 | ||
1331 | if (fstat(td->fd, &st) == -1) { | |
1332 | td_verror(td, errno); | |
1333 | return 1; | |
1334 | } | |
1335 | ||
838a3cd3 JA |
1336 | td->real_file_size = st.st_size; |
1337 | ||
1338 | if (!td->file_size || td->file_size > td->real_file_size) | |
1339 | td->file_size = td->real_file_size; | |
ebac4655 | 1340 | |
e8e387c1 | 1341 | td->file_size -= td->file_offset; |
ebac4655 JA |
1342 | return 0; |
1343 | } | |
1344 | ||
1345 | static int bdev_size(struct thread_data *td) | |
1346 | { | |
9104f874 | 1347 | unsigned long long bytes; |
ebac4655 JA |
1348 | int r; |
1349 | ||
1350 | r = blockdev_size(td->fd, &bytes); | |
1351 | if (r) { | |
1352 | td_verror(td, r); | |
1353 | return 1; | |
1354 | } | |
1355 | ||
838a3cd3 JA |
1356 | td->real_file_size = bytes; |
1357 | ||
ebac4655 JA |
1358 | /* |
1359 | * no extend possibilities, so limit size to device size if too large | |
1360 | */ | |
838a3cd3 JA |
1361 | if (!td->file_size || td->file_size > td->real_file_size) |
1362 | td->file_size = td->real_file_size; | |
ebac4655 | 1363 | |
e8e387c1 | 1364 | td->file_size -= td->file_offset; |
ebac4655 JA |
1365 | return 0; |
1366 | } | |
1367 | ||
1368 | static int get_file_size(struct thread_data *td) | |
1369 | { | |
0af7b542 | 1370 | int ret = 0; |
ebac4655 JA |
1371 | |
1372 | if (td->filetype == FIO_TYPE_FILE) | |
1373 | ret = file_size(td); | |
0af7b542 | 1374 | else if (td->filetype == FIO_TYPE_BD) |
ebac4655 | 1375 | ret = bdev_size(td); |
0af7b542 JA |
1376 | else |
1377 | td->real_file_size = -1; | |
ebac4655 JA |
1378 | |
1379 | if (ret) | |
1380 | return ret; | |
1381 | ||
e8e387c1 JA |
1382 | if (td->file_offset > td->real_file_size) { |
1383 | fprintf(stderr, "Client%d: offset extends end (%Lu > %Lu)\n", td->thread_number, td->file_offset, td->real_file_size); | |
ebac4655 JA |
1384 | return 1; |
1385 | } | |
1386 | ||
838a3cd3 | 1387 | td->io_size = td->file_size; |
ebac4655 JA |
1388 | if (td->io_size == 0) { |
1389 | fprintf(stderr, "Client%d: no io blocks\n", td->thread_number); | |
1390 | td_verror(td, EINVAL); | |
1391 | return 1; | |
1392 | } | |
1393 | ||
20dc95c4 JA |
1394 | if (!td->zone_size) |
1395 | td->zone_size = td->io_size; | |
1396 | ||
ebac4655 JA |
1397 | td->total_io_size = td->io_size * td->loops; |
1398 | return 0; | |
1399 | } | |
1400 | ||
1401 | static int setup_file_mmap(struct thread_data *td) | |
1402 | { | |
1403 | int flags; | |
1404 | ||
3d60d1ed JA |
1405 | if (td_rw(td)) |
1406 | flags = PROT_READ | PROT_WRITE; | |
1407 | else if (td_write(td)) { | |
ebac4655 JA |
1408 | flags = PROT_WRITE; |
1409 | ||
1410 | if (td->verify != VERIFY_NONE) | |
1411 | flags |= PROT_READ; | |
3d60d1ed JA |
1412 | } else |
1413 | flags = PROT_READ; | |
ebac4655 JA |
1414 | |
1415 | td->mmap = mmap(NULL, td->file_size, flags, MAP_SHARED, td->fd, td->file_offset); | |
1416 | if (td->mmap == MAP_FAILED) { | |
1417 | td->mmap = NULL; | |
1418 | td_verror(td, errno); | |
1419 | return 1; | |
1420 | } | |
1421 | ||
1422 | if (td->invalidate_cache) { | |
1423 | if (madvise(td->mmap, td->file_size, MADV_DONTNEED) < 0) { | |
1424 | td_verror(td, errno); | |
1425 | return 1; | |
1426 | } | |
1427 | } | |
1428 | ||
1429 | if (td->sequential) { | |
1430 | if (madvise(td->mmap, td->file_size, MADV_SEQUENTIAL) < 0) { | |
1431 | td_verror(td, errno); | |
1432 | return 1; | |
1433 | } | |
1434 | } else { | |
1435 | if (madvise(td->mmap, td->file_size, MADV_RANDOM) < 0) { | |
1436 | td_verror(td, errno); | |
1437 | return 1; | |
1438 | } | |
1439 | } | |
1440 | ||
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | static int setup_file_plain(struct thread_data *td) | |
1445 | { | |
1446 | if (td->invalidate_cache) { | |
1447 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_DONTNEED) < 0) { | |
1448 | td_verror(td, errno); | |
1449 | return 1; | |
1450 | } | |
1451 | } | |
1452 | ||
1453 | if (td->sequential) { | |
1454 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_SEQUENTIAL) < 0) { | |
1455 | td_verror(td, errno); | |
1456 | return 1; | |
1457 | } | |
1458 | } else { | |
1459 | if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_RANDOM) < 0) { | |
1460 | td_verror(td, errno); | |
1461 | return 1; | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | return 0; | |
1466 | } | |
1467 | ||
1468 | static int setup_file(struct thread_data *td) | |
1469 | { | |
1470 | struct stat st; | |
1471 | int flags = 0; | |
1472 | ||
1473 | if (stat(td->file_name, &st) == -1) { | |
1474 | if (errno != ENOENT) { | |
1475 | td_verror(td, errno); | |
1476 | return 1; | |
1477 | } | |
1478 | if (!td->create_file) { | |
1479 | td_verror(td, ENOENT); | |
1480 | return 1; | |
1481 | } | |
1482 | if (create_file(td, td->file_size, 0)) | |
1483 | return 1; | |
1484 | } else if (td->filetype == FIO_TYPE_FILE) { | |
6a0106a0 | 1485 | if (st.st_size < (off_t) td->file_size) { |
ebac4655 JA |
1486 | if (create_file(td, td->file_size - st.st_size, 1)) |
1487 | return 1; | |
1488 | } | |
1489 | } | |
1490 | ||
1491 | if (td->odirect) | |
1492 | flags |= O_DIRECT; | |
1493 | ||
3d60d1ed | 1494 | if (td_write(td) || td_rw(td)) { |
ebac4655 JA |
1495 | if (td->filetype == FIO_TYPE_FILE) { |
1496 | if (!td->overwrite) | |
1497 | flags |= O_TRUNC; | |
1498 | ||
1499 | flags |= O_CREAT; | |
1500 | } | |
1501 | if (td->sync_io) | |
1502 | flags |= O_SYNC; | |
1503 | ||
1504 | flags |= O_RDWR; | |
1505 | ||
1506 | td->fd = open(td->file_name, flags, 0600); | |
3d60d1ed JA |
1507 | } else { |
1508 | if (td->filetype == FIO_TYPE_CHAR) | |
1509 | flags |= O_RDWR; | |
1510 | else | |
1511 | flags |= O_RDONLY; | |
1512 | ||
1513 | td->fd = open(td->file_name, flags); | |
ebac4655 JA |
1514 | } |
1515 | ||
1516 | if (td->fd == -1) { | |
1517 | td_verror(td, errno); | |
1518 | return 1; | |
1519 | } | |
1520 | ||
1521 | if (get_file_size(td)) | |
1522 | return 1; | |
1523 | ||
1524 | if (td->io_engine != FIO_MMAPIO) | |
1525 | return setup_file_plain(td); | |
1526 | else | |
1527 | return setup_file_mmap(td); | |
1528 | } | |
1529 | ||
1530 | static int check_dev_match(dev_t dev, char *path) | |
1531 | { | |
1532 | unsigned int major, minor; | |
1533 | char line[256], *p; | |
1534 | FILE *f; | |
1535 | ||
1536 | f = fopen(path, "r"); | |
1537 | if (!f) { | |
1538 | perror("open path"); | |
1539 | return 1; | |
1540 | } | |
1541 | ||
1542 | p = fgets(line, sizeof(line), f); | |
1543 | if (!p) { | |
1544 | fclose(f); | |
1545 | return 1; | |
1546 | } | |
1547 | ||
1548 | if (sscanf(p, "%u:%u", &major, &minor) != 2) { | |
1549 | fclose(f); | |
1550 | return 1; | |
1551 | } | |
1552 | ||
1553 | if (((major << 8) | minor) == dev) { | |
1554 | fclose(f); | |
1555 | return 0; | |
1556 | } | |
1557 | ||
1558 | fclose(f); | |
1559 | return 1; | |
1560 | } | |
1561 | ||
1562 | static int find_block_dir(dev_t dev, char *path) | |
1563 | { | |
1564 | struct dirent *dir; | |
1565 | struct stat st; | |
1566 | int found = 0; | |
1567 | DIR *D; | |
1568 | ||
1569 | D = opendir(path); | |
1570 | if (!D) | |
1571 | return 0; | |
1572 | ||
1573 | while ((dir = readdir(D)) != NULL) { | |
1574 | char full_path[256]; | |
1575 | ||
1576 | if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, "..")) | |
1577 | continue; | |
1578 | if (!strcmp(dir->d_name, "device")) | |
1579 | continue; | |
1580 | ||
1581 | sprintf(full_path, "%s/%s", path, dir->d_name); | |
1582 | ||
1583 | if (!strcmp(dir->d_name, "dev")) { | |
1584 | if (!check_dev_match(dev, full_path)) { | |
1585 | found = 1; | |
1586 | break; | |
1587 | } | |
1588 | } | |
1589 | ||
1590 | if (stat(full_path, &st) == -1) { | |
1591 | perror("stat"); | |
1592 | break; | |
1593 | } | |
1594 | ||
1595 | if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode)) | |
1596 | continue; | |
1597 | ||
1598 | found = find_block_dir(dev, full_path); | |
1599 | if (found) { | |
1600 | strcpy(path, full_path); | |
1601 | break; | |
1602 | } | |
1603 | } | |
1604 | ||
1605 | closedir(D); | |
1606 | return found; | |
1607 | } | |
1608 | ||
1609 | static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus) | |
1610 | { | |
1611 | unsigned in_flight; | |
1612 | char line[256]; | |
1613 | FILE *f; | |
1614 | char *p; | |
1615 | ||
1616 | f = fopen(du->path, "r"); | |
1617 | if (!f) | |
1618 | return 1; | |
1619 | ||
1620 | p = fgets(line, sizeof(line), f); | |
1621 | if (!p) { | |
1622 | fclose(f); | |
1623 | return 1; | |
1624 | } | |
1625 | ||
1626 | if (sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], &dus->merges[0], &dus->sectors[0], &dus->ticks[0], &dus->ios[1], &dus->merges[1], &dus->sectors[1], &dus->ticks[1], &in_flight, &dus->io_ticks, &dus->time_in_queue) != 11) { | |
1627 | fclose(f); | |
1628 | return 1; | |
1629 | } | |
1630 | ||
1631 | fclose(f); | |
1632 | return 0; | |
1633 | } | |
1634 | ||
1635 | static void update_io_tick_disk(struct disk_util *du) | |
1636 | { | |
1637 | struct disk_util_stat __dus, *dus, *ldus; | |
1638 | struct timeval t; | |
1639 | ||
1640 | if (get_io_ticks(du, &__dus)) | |
1641 | return; | |
1642 | ||
1643 | dus = &du->dus; | |
1644 | ldus = &du->last_dus; | |
1645 | ||
1646 | dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]); | |
1647 | dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]); | |
1648 | dus->ios[0] += (__dus.ios[0] - ldus->ios[0]); | |
1649 | dus->ios[1] += (__dus.ios[1] - ldus->ios[1]); | |
1650 | dus->merges[0] += (__dus.merges[0] - ldus->merges[0]); | |
1651 | dus->merges[1] += (__dus.merges[1] - ldus->merges[1]); | |
1652 | dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]); | |
1653 | dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]); | |
1654 | dus->io_ticks += (__dus.io_ticks - ldus->io_ticks); | |
1655 | dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue); | |
1656 | ||
1657 | gettimeofday(&t, NULL); | |
1658 | du->msec += mtime_since(&du->time, &t); | |
1659 | memcpy(&du->time, &t, sizeof(t)); | |
1660 | memcpy(ldus, &__dus, sizeof(__dus)); | |
1661 | } | |
1662 | ||
1663 | static void update_io_ticks(void) | |
1664 | { | |
1665 | struct list_head *entry; | |
1666 | struct disk_util *du; | |
1667 | ||
1668 | list_for_each(entry, &disk_list) { | |
1669 | du = list_entry(entry, struct disk_util, list); | |
1670 | update_io_tick_disk(du); | |
1671 | } | |
1672 | } | |
1673 | ||
1674 | static int disk_util_exists(dev_t dev) | |
1675 | { | |
1676 | struct list_head *entry; | |
1677 | struct disk_util *du; | |
1678 | ||
1679 | list_for_each(entry, &disk_list) { | |
1680 | du = list_entry(entry, struct disk_util, list); | |
1681 | ||
1682 | if (du->dev == dev) | |
1683 | return 1; | |
1684 | } | |
1685 | ||
1686 | return 0; | |
1687 | } | |
1688 | ||
1689 | static void disk_util_add(dev_t dev, char *path) | |
1690 | { | |
1691 | struct disk_util *du = malloc(sizeof(*du)); | |
1692 | ||
1693 | memset(du, 0, sizeof(*du)); | |
1694 | INIT_LIST_HEAD(&du->list); | |
1695 | sprintf(du->path, "%s/stat", path); | |
1696 | du->name = strdup(basename(path)); | |
1697 | du->dev = dev; | |
1698 | ||
1699 | gettimeofday(&du->time, NULL); | |
1700 | get_io_ticks(du, &du->last_dus); | |
1701 | ||
1702 | list_add_tail(&du->list, &disk_list); | |
1703 | } | |
1704 | ||
1705 | static void init_disk_util(struct thread_data *td) | |
1706 | { | |
1707 | struct stat st; | |
1708 | char foo[256], tmp[256]; | |
1709 | dev_t dev; | |
1710 | char *p; | |
1711 | ||
1712 | if (!td->do_disk_util) | |
1713 | return; | |
1714 | ||
1715 | if (!stat(td->file_name, &st)) { | |
1716 | if (S_ISBLK(st.st_mode)) | |
1717 | dev = st.st_rdev; | |
1718 | else | |
1719 | dev = st.st_dev; | |
1720 | } else { | |
1721 | /* | |
1722 | * must be a file, open "." in that path | |
1723 | */ | |
1724 | strcpy(foo, td->file_name); | |
1725 | p = dirname(foo); | |
1726 | if (stat(p, &st)) { | |
1727 | perror("disk util stat"); | |
1728 | return; | |
1729 | } | |
1730 | ||
1731 | dev = st.st_dev; | |
1732 | } | |
1733 | ||
1734 | if (disk_util_exists(dev)) | |
1735 | return; | |
1736 | ||
1737 | sprintf(foo, "/sys/block"); | |
1738 | if (!find_block_dir(dev, foo)) | |
1739 | return; | |
1740 | ||
1741 | /* | |
64cb3e10 | 1742 | * for md/dm, there's no queue dir. we already have the right place |
ebac4655 | 1743 | */ |
64cb3e10 | 1744 | sprintf(tmp, "%s/stat", foo); |
ebac4655 | 1745 | if (stat(tmp, &st)) { |
64cb3e10 JA |
1746 | /* |
1747 | * if this is inside a partition dir, jump back to parent | |
1748 | */ | |
1749 | sprintf(tmp, "%s/queue", foo); | |
ebac4655 | 1750 | if (stat(tmp, &st)) { |
64cb3e10 JA |
1751 | p = dirname(foo); |
1752 | sprintf(tmp, "%s/queue", p); | |
1753 | if (stat(tmp, &st)) { | |
1754 | fprintf(stderr, "unknown sysfs layout\n"); | |
1755 | return; | |
1756 | } | |
1757 | sprintf(foo, "%s", p); | |
ebac4655 | 1758 | } |
ebac4655 JA |
1759 | } |
1760 | ||
1761 | disk_util_add(dev, foo); | |
1762 | } | |
1763 | ||
1764 | static void disk_util_timer_arm(void) | |
1765 | { | |
1766 | itimer.it_value.tv_sec = 0; | |
1767 | itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; | |
1768 | setitimer(ITIMER_REAL, &itimer, NULL); | |
1769 | } | |
1770 | ||
1771 | static void clear_io_state(struct thread_data *td) | |
1772 | { | |
1773 | if (td->io_engine == FIO_SYNCIO) | |
1774 | lseek(td->fd, SEEK_SET, 0); | |
1775 | ||
20dc95c4 | 1776 | td->last_pos = 0; |
ebac4655 JA |
1777 | td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0; |
1778 | td->this_io_bytes[0] = td->this_io_bytes[1] = 0; | |
20dc95c4 | 1779 | td->zone_bytes = 0; |
ebac4655 JA |
1780 | |
1781 | if (td->file_map) | |
1782 | memset(td->file_map, 0, td->num_maps * sizeof(long)); | |
1783 | } | |
1784 | ||
1785 | static void update_rusage_stat(struct thread_data *td) | |
1786 | { | |
1787 | if (!(td->runtime[0] + td->runtime[1])) | |
1788 | return; | |
1789 | ||
1790 | getrusage(RUSAGE_SELF, &td->ru_end); | |
1791 | ||
1792 | td->usr_time += mtime_since(&td->ru_start.ru_utime, &td->ru_end.ru_utime); | |
1793 | td->sys_time += mtime_since(&td->ru_start.ru_stime, &td->ru_end.ru_stime); | |
1794 | td->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw); | |
1795 | ||
1796 | ||
1797 | memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end)); | |
1798 | } | |
1799 | ||
1800 | static void *thread_main(void *data) | |
1801 | { | |
1802 | struct thread_data *td = data; | |
1803 | int ret = 1; | |
1804 | ||
1805 | if (!td->use_thread) | |
1806 | setsid(); | |
1807 | ||
1808 | td->pid = getpid(); | |
1809 | ||
aea47d44 JA |
1810 | INIT_LIST_HEAD(&td->io_u_freelist); |
1811 | INIT_LIST_HEAD(&td->io_u_busylist); | |
1812 | INIT_LIST_HEAD(&td->io_hist_list); | |
1813 | INIT_LIST_HEAD(&td->io_log_list); | |
1814 | ||
ebac4655 JA |
1815 | if (init_io_u(td)) |
1816 | goto err; | |
1817 | ||
1818 | if (fio_setaffinity(td) == -1) { | |
1819 | td_verror(td, errno); | |
1820 | goto err; | |
1821 | } | |
1822 | ||
1823 | if (init_io(td)) | |
1824 | goto err; | |
1825 | ||
aea47d44 JA |
1826 | if (init_iolog(td)) |
1827 | goto err; | |
1828 | ||
ebac4655 JA |
1829 | if (td->ioprio) { |
1830 | if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) { | |
1831 | td_verror(td, errno); | |
1832 | goto err; | |
1833 | } | |
1834 | } | |
1835 | ||
1836 | sem_post(&startup_sem); | |
1837 | sem_wait(&td->mutex); | |
1838 | ||
1839 | if (!td->create_serialize && setup_file(td)) | |
1840 | goto err; | |
1841 | ||
1842 | if (init_random_state(td)) | |
1843 | goto err; | |
1844 | ||
1845 | gettimeofday(&td->epoch, NULL); | |
1846 | ||
1847 | while (td->loops--) { | |
1848 | getrusage(RUSAGE_SELF, &td->ru_start); | |
1849 | gettimeofday(&td->start, NULL); | |
1850 | memcpy(&td->stat_sample_time, &td->start, sizeof(td->start)); | |
1851 | ||
1852 | if (td->ratemin) | |
1853 | memcpy(&td->lastrate, &td->stat_sample_time, sizeof(td->lastrate)); | |
1854 | ||
1855 | clear_io_state(td); | |
1856 | prune_io_piece_log(td); | |
1857 | ||
1858 | do_io(td); | |
1859 | ||
1860 | td->runtime[td->ddir] += mtime_since_now(&td->start); | |
aea47d44 | 1861 | if (td_rw(td) && td->io_bytes[td->ddir ^ 1]) |
3d60d1ed JA |
1862 | td->runtime[td->ddir ^ 1] = td->runtime[td->ddir]; |
1863 | ||
ebac4655 JA |
1864 | update_rusage_stat(td); |
1865 | ||
1866 | if (td->error || td->terminate) | |
1867 | break; | |
1868 | ||
1869 | if (td->verify == VERIFY_NONE) | |
1870 | continue; | |
1871 | ||
1872 | clear_io_state(td); | |
1873 | gettimeofday(&td->start, NULL); | |
1874 | ||
1875 | do_verify(td); | |
1876 | ||
1877 | td->runtime[DDIR_READ] += mtime_since_now(&td->start); | |
1878 | ||
1879 | if (td->error || td->terminate) | |
1880 | break; | |
1881 | } | |
1882 | ||
1883 | ret = 0; | |
1884 | ||
1885 | if (td->bw_log) | |
1886 | finish_log(td, td->bw_log, "bw"); | |
1887 | if (td->slat_log) | |
1888 | finish_log(td, td->slat_log, "slat"); | |
1889 | if (td->clat_log) | |
1890 | finish_log(td, td->clat_log, "clat"); | |
1891 | ||
1892 | if (exitall_on_terminate) | |
1893 | terminate_threads(td->groupid); | |
1894 | ||
1895 | err: | |
1896 | if (td->fd != -1) { | |
1897 | close(td->fd); | |
1898 | td->fd = -1; | |
1899 | } | |
1900 | if (td->mmap) | |
1901 | munmap(td->mmap, td->file_size); | |
1902 | cleanup_io(td); | |
1903 | cleanup_io_u(td); | |
1904 | if (ret) { | |
1905 | sem_post(&startup_sem); | |
1906 | sem_wait(&td->mutex); | |
1907 | } | |
1908 | td_set_runstate(td, TD_EXITED); | |
1909 | return NULL; | |
1910 | ||
1911 | } | |
1912 | ||
1913 | static void *fork_main(int shmid, int offset) | |
1914 | { | |
1915 | struct thread_data *td; | |
1916 | void *data; | |
1917 | ||
1918 | data = shmat(shmid, NULL, 0); | |
1919 | if (data == (void *) -1) { | |
1920 | perror("shmat"); | |
1921 | return NULL; | |
1922 | } | |
1923 | ||
1924 | td = data + offset * sizeof(struct thread_data); | |
1925 | thread_main(td); | |
1926 | shmdt(data); | |
1927 | return NULL; | |
1928 | } | |
1929 | ||
1930 | static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max, | |
1931 | double *mean, double *dev) | |
1932 | { | |
1933 | double n; | |
1934 | ||
1935 | if (is->samples == 0) | |
1936 | return 0; | |
1937 | ||
1938 | *min = is->min_val; | |
1939 | *max = is->max_val; | |
1940 | ||
1941 | n = (double) is->samples; | |
1942 | *mean = (double) is->val / n; | |
1943 | *dev = sqrt(((double) is->val_sq - (*mean * *mean) / n) / (n - 1)); | |
1944 | if (!(*min + *max) && !(*mean + *dev)) | |
1945 | return 0; | |
1946 | ||
1947 | return 1; | |
1948 | } | |
1949 | ||
1950 | static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs, | |
1951 | int ddir) | |
1952 | { | |
1953 | char *ddir_str[] = { "read ", "write" }; | |
1954 | unsigned long min, max, bw; | |
1955 | double mean, dev; | |
1956 | ||
1957 | if (!td->runtime[ddir]) | |
1958 | return; | |
1959 | ||
1960 | bw = td->io_bytes[ddir] / td->runtime[ddir]; | |
9104f874 | 1961 | printf(" %s: io=%6lluMiB, bw=%6luKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]); |
ebac4655 JA |
1962 | |
1963 | if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev)) | |
1964 | printf(" slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev); | |
1965 | ||
1966 | if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev)) | |
1967 | printf(" clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev); | |
1968 | ||
1969 | if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) { | |
1970 | double p_of_agg; | |
1971 | ||
1972 | p_of_agg = mean * 100 / (double) rs->agg[ddir]; | |
1973 | printf(" bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, dev=%5.02f\n", min, max, p_of_agg, mean, dev); | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | static void show_thread_status(struct thread_data *td, | |
1978 | struct group_run_stats *rs) | |
1979 | { | |
1980 | double usr_cpu, sys_cpu; | |
1981 | ||
1982 | if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error) | |
1983 | return; | |
1984 | ||
1985 | printf("Client%d (groupid=%d): err=%2d:\n", td->thread_number, td->groupid, td->error); | |
1986 | ||
1987 | show_ddir_status(td, rs, td->ddir); | |
aea47d44 JA |
1988 | if (td->io_bytes[td->ddir ^ 1]) |
1989 | show_ddir_status(td, rs, td->ddir ^ 1); | |
ebac4655 JA |
1990 | |
1991 | if (td->runtime[0] + td->runtime[1]) { | |
1992 | double runt = td->runtime[0] + td->runtime[1]; | |
1993 | ||
1994 | usr_cpu = (double) td->usr_time * 100 / runt; | |
1995 | sys_cpu = (double) td->sys_time * 100 / runt; | |
1996 | } else { | |
1997 | usr_cpu = 0; | |
1998 | sys_cpu = 0; | |
1999 | } | |
2000 | ||
2001 | printf(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx); | |
2002 | } | |
2003 | ||
2004 | static void check_str_update(struct thread_data *td) | |
2005 | { | |
2006 | char c = run_str[td->thread_number - 1]; | |
2007 | ||
2008 | if (td->runstate == td->old_runstate) | |
2009 | return; | |
2010 | ||
2011 | switch (td->runstate) { | |
2012 | case TD_REAPED: | |
2013 | c = '_'; | |
2014 | break; | |
2015 | case TD_EXITED: | |
2016 | c = 'E'; | |
2017 | break; | |
2018 | case TD_RUNNING: | |
3d60d1ed JA |
2019 | if (td_rw(td)) { |
2020 | if (td->sequential) | |
2021 | c = 'M'; | |
2022 | else | |
2023 | c = 'm'; | |
2024 | } else if (td_read(td)) { | |
ebac4655 JA |
2025 | if (td->sequential) |
2026 | c = 'R'; | |
2027 | else | |
2028 | c = 'r'; | |
2029 | } else { | |
2030 | if (td->sequential) | |
2031 | c = 'W'; | |
2032 | else | |
2033 | c = 'w'; | |
2034 | } | |
2035 | break; | |
2036 | case TD_VERIFYING: | |
2037 | c = 'V'; | |
2038 | break; | |
2039 | case TD_CREATED: | |
2040 | c = 'C'; | |
2041 | break; | |
2042 | case TD_NOT_CREATED: | |
2043 | c = 'P'; | |
2044 | break; | |
2045 | default: | |
2046 | printf("state %d\n", td->runstate); | |
2047 | } | |
2048 | ||
2049 | run_str[td->thread_number - 1] = c; | |
2050 | td->old_runstate = td->runstate; | |
2051 | } | |
2052 | ||
5289b847 JA |
2053 | static void eta_to_str(char *str, int eta_sec) |
2054 | { | |
2055 | unsigned int d, h, m, s; | |
2056 | static int always_d, always_h; | |
2057 | ||
2058 | d = h = m = s = 0; | |
2059 | ||
2060 | s = eta_sec % 60; | |
2061 | eta_sec /= 60; | |
2062 | m = eta_sec % 60; | |
2063 | eta_sec /= 60; | |
2064 | h = eta_sec % 24; | |
2065 | eta_sec /= 24; | |
2066 | d = eta_sec; | |
2067 | ||
2068 | if (d || always_d) { | |
2069 | always_d = 1; | |
2070 | str += sprintf(str, "%02dd:", d); | |
2071 | } | |
2072 | if (h || always_h) { | |
2073 | always_h = 1; | |
2074 | str += sprintf(str, "%02dh:", h); | |
2075 | } | |
2076 | ||
2077 | str += sprintf(str, "%02dm:", m); | |
2078 | str += sprintf(str, "%02ds", s); | |
2079 | } | |
2080 | ||
6a0106a0 JA |
2081 | static int thread_eta(struct thread_data *td, unsigned long elapsed) |
2082 | { | |
2083 | unsigned long long bytes_total, bytes_done; | |
2084 | unsigned int eta_sec = 0; | |
2085 | ||
2086 | bytes_total = td->total_io_size; | |
3d60d1ed JA |
2087 | |
2088 | /* | |
2089 | * if writing, bytes_total will be twice the size. If mixing, | |
2090 | * assume a 50/50 split and thus bytes_total will be 50% larger. | |
2091 | */ | |
2092 | if (td->verify) { | |
2093 | if (td_rw(td)) | |
2094 | bytes_total = bytes_total * 3 / 2; | |
2095 | else | |
2096 | bytes_total <<= 1; | |
2097 | } | |
6a0106a0 JA |
2098 | if (td->zone_size && td->zone_skip) |
2099 | bytes_total /= (td->zone_skip / td->zone_size); | |
2100 | ||
2101 | if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) { | |
2102 | double perc; | |
8b611c34 | 2103 | |
6a0106a0 JA |
2104 | bytes_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE]; |
2105 | perc = (double) bytes_done / (double) bytes_total; | |
2106 | if (perc > 1.0) | |
2107 | perc = 1.0; | |
2108 | ||
2109 | eta_sec = (elapsed * (1.0 / perc)) - elapsed; | |
2110 | ||
8b611c34 | 2111 | if (td->timeout && eta_sec > (td->timeout - elapsed)) |
6a0106a0 JA |
2112 | eta_sec = td->timeout - elapsed; |
2113 | } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED) { | |
2114 | int t_eta = 0, r_eta = 0; | |
2115 | ||
2116 | /* | |
2117 | * We can only guess - assume it'll run the full timeout | |
2118 | * if given, otherwise assume it'll run at the specified rate. | |
2119 | */ | |
2120 | if (td->timeout) | |
2121 | t_eta = td->timeout + td->start_delay - elapsed; | |
2122 | if (td->rate) { | |
2123 | r_eta = (bytes_total / 1024) / td->rate; | |
2124 | r_eta += td->start_delay - elapsed; | |
2125 | } | |
2126 | ||
2127 | if (r_eta && t_eta) | |
2128 | eta_sec = min(r_eta, t_eta); | |
2129 | else if (r_eta) | |
2130 | eta_sec = r_eta; | |
2131 | else if (t_eta) | |
2132 | eta_sec = t_eta; | |
2133 | else | |
2134 | eta_sec = INT_MAX; | |
2135 | } else { | |
2136 | /* | |
2137 | * thread is already done | |
2138 | */ | |
2139 | eta_sec = 0; | |
2140 | } | |
2141 | ||
2142 | return eta_sec; | |
2143 | } | |
2144 | ||
ebac4655 JA |
2145 | static void print_thread_status(void) |
2146 | { | |
6a0106a0 JA |
2147 | unsigned long elapsed = time_since_now(&genesis); |
2148 | int i, nr_running, t_rate, m_rate, *eta_secs, eta_sec; | |
5289b847 | 2149 | char eta_str[32]; |
6a0106a0 JA |
2150 | double perc = 0.0; |
2151 | ||
2152 | eta_secs = malloc(thread_number * sizeof(int)); | |
2153 | memset(eta_secs, 0, thread_number * sizeof(int)); | |
ebac4655 | 2154 | |
ebac4655 JA |
2155 | nr_running = t_rate = m_rate = 0; |
2156 | for (i = 0; i < thread_number; i++) { | |
2157 | struct thread_data *td = &threads[i]; | |
2158 | ||
2159 | if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING){ | |
2160 | nr_running++; | |
2161 | t_rate += td->rate; | |
2162 | m_rate += td->ratemin; | |
2163 | } | |
2164 | ||
6a0106a0 JA |
2165 | if (elapsed >= 3) |
2166 | eta_secs[i] = thread_eta(td, elapsed); | |
8b611c34 JA |
2167 | else |
2168 | eta_secs[i] = INT_MAX; | |
ebac4655 JA |
2169 | |
2170 | check_str_update(td); | |
2171 | } | |
2172 | ||
6a0106a0 JA |
2173 | if (exitall_on_terminate) |
2174 | eta_sec = INT_MAX; | |
2175 | else | |
2176 | eta_sec = 0; | |
5289b847 | 2177 | |
6a0106a0 JA |
2178 | for (i = 0; i < thread_number; i++) { |
2179 | if (exitall_on_terminate) { | |
2180 | if (eta_secs[i] < eta_sec) | |
2181 | eta_sec = eta_secs[i]; | |
2182 | } else { | |
2183 | if (eta_secs[i] > eta_sec) | |
2184 | eta_sec = eta_secs[i]; | |
5289b847 | 2185 | } |
6a0106a0 | 2186 | } |
5289b847 | 2187 | |
8b611c34 | 2188 | if (eta_sec != INT_MAX && elapsed) { |
6a0106a0 JA |
2189 | perc = (double) elapsed / (double) (elapsed + eta_sec); |
2190 | eta_to_str(eta_str, eta_sec); | |
ebac4655 JA |
2191 | } |
2192 | ||
5289b847 | 2193 | printf("Threads now running (%d)", nr_running); |
ebac4655 JA |
2194 | if (m_rate || t_rate) |
2195 | printf(", commitrate %d/%dKiB/sec", t_rate, m_rate); | |
8b611c34 | 2196 | if (eta_sec != INT_MAX) { |
6a0106a0 JA |
2197 | perc *= 100.0; |
2198 | printf(": [%s] [%3.2f%% done] [eta %s]", run_str, perc,eta_str); | |
2199 | } | |
5289b847 | 2200 | printf("\r"); |
ebac4655 | 2201 | fflush(stdout); |
6a0106a0 | 2202 | free(eta_secs); |
ebac4655 JA |
2203 | } |
2204 | ||
2205 | static void reap_threads(int *nr_running, int *t_rate, int *m_rate) | |
2206 | { | |
2207 | int i; | |
2208 | ||
2209 | /* | |
2210 | * reap exited threads (TD_EXITED -> TD_REAPED) | |
2211 | */ | |
2212 | for (i = 0; i < thread_number; i++) { | |
2213 | struct thread_data *td = &threads[i]; | |
2214 | ||
2215 | if (td->runstate != TD_EXITED) | |
2216 | continue; | |
2217 | ||
2218 | td_set_runstate(td, TD_REAPED); | |
2219 | ||
2220 | if (td->use_thread) { | |
2221 | long ret; | |
2222 | ||
2223 | if (pthread_join(td->thread, (void *) &ret)) | |
2224 | perror("thread_join"); | |
2225 | } else | |
2226 | waitpid(td->pid, NULL, 0); | |
2227 | ||
2228 | (*nr_running)--; | |
2229 | (*m_rate) -= td->ratemin; | |
2230 | (*t_rate) -= td->rate; | |
2231 | } | |
2232 | } | |
2233 | ||
fcb6ade2 JA |
2234 | static void fio_unpin_memory(void *pinned) |
2235 | { | |
2236 | if (pinned) { | |
2237 | if (munlock(pinned, mlock_size) < 0) | |
2238 | perror("munlock"); | |
2239 | munmap(pinned, mlock_size); | |
2240 | } | |
2241 | } | |
2242 | ||
2243 | static void *fio_pin_memory(void) | |
2244 | { | |
2245 | long pagesize, pages; | |
2246 | void *ptr; | |
2247 | ||
2248 | if (!mlock_size) | |
2249 | return NULL; | |
2250 | ||
2251 | /* | |
2252 | * Don't allow mlock of more than real_mem-128MB | |
2253 | */ | |
2254 | pagesize = sysconf(_SC_PAGESIZE); | |
2255 | pages = sysconf(_SC_PHYS_PAGES); | |
2256 | if (pages != -1 && pagesize != -1) { | |
2257 | unsigned long long real_mem = pages * pagesize; | |
2258 | ||
2259 | if ((mlock_size + 128 * 1024 * 1024) > real_mem) { | |
2260 | mlock_size = real_mem - 128 * 1024 * 1024; | |
2261 | printf("fio: limiting mlocked memory to %lluMiB\n", | |
2262 | mlock_size >> 20); | |
2263 | } | |
2264 | } | |
2265 | ||
2266 | ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0); | |
2267 | if (!ptr) { | |
2268 | perror("malloc locked mem"); | |
2269 | return NULL; | |
2270 | } | |
2271 | if (mlock(ptr, mlock_size) < 0) { | |
2272 | munmap(ptr, mlock_size); | |
2273 | perror("mlock"); | |
2274 | return NULL; | |
2275 | } | |
2276 | ||
2277 | return ptr; | |
2278 | } | |
2279 | ||
ebac4655 JA |
2280 | static void run_threads(void) |
2281 | { | |
ebac4655 JA |
2282 | struct thread_data *td; |
2283 | unsigned long spent; | |
2284 | int i, todo, nr_running, m_rate, t_rate, nr_started; | |
fcb6ade2 JA |
2285 | void *mlocked_mem; |
2286 | ||
2287 | mlocked_mem = fio_pin_memory(); | |
ebac4655 JA |
2288 | |
2289 | printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : ""); | |
fcb6ade2 | 2290 | fflush(stdout); |
c04f7ec3 | 2291 | |
4efa970e JA |
2292 | signal(SIGINT, sig_handler); |
2293 | signal(SIGALRM, sig_handler); | |
2294 | ||
ebac4655 JA |
2295 | todo = thread_number; |
2296 | nr_running = 0; | |
2297 | nr_started = 0; | |
2298 | m_rate = t_rate = 0; | |
2299 | ||
2300 | for (i = 0; i < thread_number; i++) { | |
2301 | td = &threads[i]; | |
2302 | ||
2303 | run_str[td->thread_number - 1] = 'P'; | |
2304 | ||
2305 | init_disk_util(td); | |
2306 | ||
2307 | if (!td->create_serialize) | |
2308 | continue; | |
2309 | ||
2310 | /* | |
2311 | * do file setup here so it happens sequentially, | |
2312 | * we don't want X number of threads getting their | |
2313 | * client data interspersed on disk | |
2314 | */ | |
2315 | if (setup_file(td)) { | |
2316 | td_set_runstate(td, TD_REAPED); | |
2317 | todo--; | |
2318 | } | |
2319 | } | |
2320 | ||
2321 | gettimeofday(&genesis, NULL); | |
2322 | ||
2323 | while (todo) { | |
2324 | /* | |
2325 | * create threads (TD_NOT_CREATED -> TD_CREATED) | |
2326 | */ | |
2327 | for (i = 0; i < thread_number; i++) { | |
2328 | td = &threads[i]; | |
2329 | ||
2330 | if (td->runstate != TD_NOT_CREATED) | |
2331 | continue; | |
2332 | ||
2333 | /* | |
2334 | * never got a chance to start, killed by other | |
2335 | * thread for some reason | |
2336 | */ | |
2337 | if (td->terminate) { | |
2338 | todo--; | |
2339 | continue; | |
2340 | } | |
2341 | ||
2342 | if (td->start_delay) { | |
2343 | spent = mtime_since_now(&genesis); | |
2344 | ||
2345 | if (td->start_delay * 1000 > spent) | |
2346 | continue; | |
2347 | } | |
2348 | ||
2349 | if (td->stonewall && (nr_started || nr_running)) | |
2350 | break; | |
2351 | ||
2352 | td_set_runstate(td, TD_CREATED); | |
2353 | sem_init(&startup_sem, 0, 1); | |
2354 | todo--; | |
2355 | nr_started++; | |
2356 | ||
2357 | if (td->use_thread) { | |
2358 | if (pthread_create(&td->thread, NULL, thread_main, td)) { | |
2359 | perror("thread_create"); | |
2360 | nr_started--; | |
2361 | } | |
2362 | } else { | |
2363 | if (fork()) | |
2364 | sem_wait(&startup_sem); | |
2365 | else { | |
2366 | fork_main(shm_id, i); | |
2367 | exit(0); | |
2368 | } | |
2369 | } | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * start created threads (TD_CREATED -> TD_RUNNING) | |
2374 | */ | |
2375 | for (i = 0; i < thread_number; i++) { | |
2376 | td = &threads[i]; | |
2377 | ||
2378 | if (td->runstate != TD_CREATED) | |
2379 | continue; | |
2380 | ||
2381 | td_set_runstate(td, TD_RUNNING); | |
2382 | nr_running++; | |
2383 | nr_started--; | |
2384 | m_rate += td->ratemin; | |
2385 | t_rate += td->rate; | |
2386 | sem_post(&td->mutex); | |
2387 | } | |
2388 | ||
2389 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2390 | ||
2391 | if (todo) | |
2392 | usleep(100000); | |
2393 | } | |
2394 | ||
2395 | while (nr_running) { | |
2396 | reap_threads(&nr_running, &t_rate, &m_rate); | |
2397 | usleep(10000); | |
2398 | } | |
2399 | ||
2400 | update_io_ticks(); | |
fcb6ade2 | 2401 | fio_unpin_memory(mlocked_mem); |
ebac4655 JA |
2402 | } |
2403 | ||
2404 | static void show_group_stats(struct group_run_stats *rs, int id) | |
2405 | { | |
2406 | printf("\nRun status group %d (all jobs):\n", id); | |
2407 | ||
2408 | if (rs->max_run[DDIR_READ]) | |
9104f874 | 2409 | printf(" READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_mb[0], rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]); |
ebac4655 | 2410 | if (rs->max_run[DDIR_WRITE]) |
9104f874 | 2411 | printf(" WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_mb[1], rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]); |
ebac4655 JA |
2412 | } |
2413 | ||
2414 | static void show_disk_util(void) | |
2415 | { | |
2416 | struct disk_util_stat *dus; | |
2417 | struct list_head *entry; | |
2418 | struct disk_util *du; | |
2419 | double util; | |
2420 | ||
2421 | printf("\nDisk stats (read/write):\n"); | |
2422 | ||
2423 | list_for_each(entry, &disk_list) { | |
2424 | du = list_entry(entry, struct disk_util, list); | |
2425 | dus = &du->dus; | |
2426 | ||
2427 | util = (double) 100 * du->dus.io_ticks / (double) du->msec; | |
2428 | if (util > 100.0) | |
2429 | util = 100.0; | |
2430 | ||
2431 | printf(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util); | |
2432 | } | |
2433 | } | |
2434 | ||
2435 | static void show_run_stats(void) | |
2436 | { | |
2437 | struct group_run_stats *runstats, *rs; | |
2438 | struct thread_data *td; | |
2439 | int i; | |
2440 | ||
2441 | runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1)); | |
2442 | ||
2443 | for (i = 0; i < groupid + 1; i++) { | |
2444 | rs = &runstats[i]; | |
2445 | ||
2446 | memset(rs, 0, sizeof(*rs)); | |
2447 | rs->min_bw[0] = rs->min_run[0] = ~0UL; | |
2448 | rs->min_bw[1] = rs->min_run[1] = ~0UL; | |
2449 | } | |
2450 | ||
2451 | for (i = 0; i < thread_number; i++) { | |
2452 | unsigned long rbw, wbw; | |
2453 | ||
2454 | td = &threads[i]; | |
2455 | ||
2456 | if (td->error) { | |
2457 | printf("Client%d: %s\n", td->thread_number, td->verror); | |
2458 | continue; | |
2459 | } | |
2460 | ||
2461 | rs = &runstats[td->groupid]; | |
2462 | ||
2463 | if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0]) | |
2464 | rs->min_run[0] = td->runtime[0]; | |
2465 | if (td->runtime[0] > rs->max_run[0]) | |
2466 | rs->max_run[0] = td->runtime[0]; | |
2467 | if (td->runtime[1] < rs->min_run[1] || !rs->min_run[1]) | |
2468 | rs->min_run[1] = td->runtime[1]; | |
2469 | if (td->runtime[1] > rs->max_run[1]) | |
2470 | rs->max_run[1] = td->runtime[1]; | |
2471 | ||
2472 | rbw = wbw = 0; | |
2473 | if (td->runtime[0]) | |
2474 | rbw = td->io_bytes[0] / td->runtime[0]; | |
2475 | if (td->runtime[1]) | |
2476 | wbw = td->io_bytes[1] / td->runtime[1]; | |
2477 | ||
2478 | if (rbw < rs->min_bw[0]) | |
2479 | rs->min_bw[0] = rbw; | |
2480 | if (wbw < rs->min_bw[1]) | |
2481 | rs->min_bw[1] = wbw; | |
2482 | if (rbw > rs->max_bw[0]) | |
2483 | rs->max_bw[0] = rbw; | |
2484 | if (wbw > rs->max_bw[1]) | |
2485 | rs->max_bw[1] = wbw; | |
2486 | ||
2487 | rs->io_mb[0] += td->io_bytes[0] >> 20; | |
2488 | rs->io_mb[1] += td->io_bytes[1] >> 20; | |
2489 | } | |
2490 | ||
2491 | for (i = 0; i < groupid + 1; i++) { | |
2492 | rs = &runstats[i]; | |
2493 | ||
2494 | if (rs->max_run[0]) | |
2495 | rs->agg[0] = (rs->io_mb[0]*1024*1000) / rs->max_run[0]; | |
2496 | if (rs->max_run[1]) | |
2497 | rs->agg[1] = (rs->io_mb[1]*1024*1000) / rs->max_run[1]; | |
2498 | } | |
2499 | ||
2500 | /* | |
2501 | * don't overwrite last signal output | |
2502 | */ | |
2503 | printf("\n"); | |
2504 | ||
2505 | for (i = 0; i < thread_number; i++) { | |
2506 | td = &threads[i]; | |
2507 | rs = &runstats[td->groupid]; | |
2508 | ||
2509 | show_thread_status(td, rs); | |
2510 | } | |
2511 | ||
2512 | for (i = 0; i < groupid + 1; i++) | |
2513 | show_group_stats(&runstats[i], i); | |
2514 | ||
2515 | show_disk_util(); | |
2516 | } | |
2517 | ||
2518 | int main(int argc, char *argv[]) | |
2519 | { | |
2520 | if (parse_options(argc, argv)) | |
2521 | return 1; | |
2522 | ||
2523 | if (!thread_number) { | |
2524 | printf("Nothing to do\n"); | |
2525 | return 1; | |
2526 | } | |
2527 | ||
2528 | disk_util_timer_arm(); | |
2529 | ||
2530 | run_threads(); | |
2531 | show_run_stats(); | |
2532 | ||
2533 | return 0; | |
2534 | } |