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