projects / blktrace.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
[PATCH] verify_blkparse: correct sscanf return
[blktrace.git] / blkparse.c
... / ...
CommitLineData
1/*
2 * block queue tracing parse application
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 <sys/types.h>
22#include <sys/stat.h>
23#include <unistd.h>
24#include <stdio.h>
25#include <fcntl.h>
26#include <stdlib.h>
27#include <string.h>
28#include <getopt.h>
29#include <errno.h>
30#include <signal.h>
31#include <locale.h>
32#include <limits.h>
33
34#include "blktrace.h"
35#include "rbtree.h"
36#include "jhash.h"
37
38static char blkparse_version[] = "0.90";
39
40struct per_dev_info {
41 dev_t dev;
42 char *name;
43
44 int backwards;
45 unsigned long long events;
46 unsigned long long last_reported_time;
47 unsigned long long last_read_time;
48 struct io_stats io_stats;
49 unsigned long last_sequence;
50 unsigned long skips;
51
52 struct rb_root rb_last;
53 unsigned long rb_last_entries;
54
55 struct rb_root rb_track;
56
57 int nfiles;
58 int ncpus;
59 struct per_cpu_info *cpus;
60};
61
62struct per_process_info {
63 char name[16];
64 __u32 pid;
65 struct io_stats io_stats;
66 struct per_process_info *hash_next, *list_next;
67 int more_than_one;
68
69 /*
70 * individual io stats
71 */
72 unsigned long long longest_allocation_wait[2];
73 unsigned long long longest_dispatch_wait[2];
74 unsigned long long longest_completion_wait[2];
75};
76
77#define PPI_HASH_SHIFT (8)
78#define PPI_HASH_SIZE (1 << PPI_HASH_SHIFT)
79#define PPI_HASH_MASK (PPI_HASH_SIZE - 1)
80static struct per_process_info *ppi_hash_table[PPI_HASH_SIZE];
81static struct per_process_info *ppi_list;
82static int ppi_list_entries;
83
84#define S_OPTS "a:A:i:o:b:stqw:f:F:vnmD:"
85static struct option l_opts[] = {
86 {
87 .name = "act-mask",
88 .has_arg = required_argument,
89 .flag = NULL,
90 .val = 'a'
91 },
92 {
93 .name = "set-mask",
94 .has_arg = required_argument,
95 .flag = NULL,
96 .val = 'A'
97 },
98 {
99 .name = "input",
100 .has_arg = required_argument,
101 .flag = NULL,
102 .val = 'i'
103 },
104 {
105 .name = "output",
106 .has_arg = required_argument,
107 .flag = NULL,
108 .val = 'o'
109 },
110 {
111 .name = "batch",
112 .has_arg = required_argument,
113 .flag = NULL,
114 .val = 'b'
115 },
116 {
117 .name = "per-program-stats",
118 .has_arg = no_argument,
119 .flag = NULL,
120 .val = 's'
121 },
122 {
123 .name = "track-ios",
124 .has_arg = no_argument,
125 .flag = NULL,
126 .val = 't'
127 },
128 {
129 .name = "quiet",
130 .has_arg = no_argument,
131 .flag = NULL,
132 .val = 'q'
133 },
134 {
135 .name = "stopwatch",
136 .has_arg = required_argument,
137 .flag = NULL,
138 .val = 'w'
139 },
140 {
141 .name = "format",
142 .has_arg = required_argument,
143 .flag = NULL,
144 .val = 'f'
145 },
146 {
147 .name = "format-spec",
148 .has_arg = required_argument,
149 .flag = NULL,
150 .val = 'F'
151 },
152 {
153 .name = "hash-by-name",
154 .has_arg = no_argument,
155 .flag = NULL,
156 .val = 'n'
157 },
158 {
159 .name = "missing",
160 .has_arg = no_argument,
161 .flag = NULL,
162 .val = 'm'
163 },
164 {
165 .name = "version",
166 .has_arg = no_argument,
167 .flag = NULL,
168 .val = 'v'
169 },
170 {
171 .name = "input-directory",
172 .has_arg = required_argument,
173 .flag = NULL,
174 .val = 'D'
175 },
176 {
177 .name = NULL,
178 }
179};
180
181/*
182 * for sorting the displayed output
183 */
184struct trace {
185 struct blk_io_trace *bit;
186 struct rb_node rb_node;
187 struct trace *next;
188};
189
190static struct rb_root rb_sort_root;
191static unsigned long rb_sort_entries;
192
193static struct trace *trace_list;
194
195/*
196 * allocation cache
197 */
198static struct blk_io_trace *bit_alloc_list;
199static struct trace *t_alloc_list;
200
201/*
202 * for tracking individual ios
203 */
204struct io_track {
205 struct rb_node rb_node;
206
207 __u64 sector;
208 __u32 pid;
209 char comm[16];
210 unsigned long long allocation_time;
211 unsigned long long queue_time;
212 unsigned long long dispatch_time;
213 unsigned long long completion_time;
214};
215
216static int ndevices;
217static struct per_dev_info *devices;
218static char *get_dev_name(struct per_dev_info *, char *, int);
219
220FILE *ofp = NULL;
221static char *output_name;
222static char *input_dir;
223
224static unsigned long long genesis_time;
225static unsigned long long last_allowed_time;
226static unsigned int smallest_seq_read;
227static unsigned long long stopwatch_start; /* start from zero by default */
228static unsigned long long stopwatch_end = ULONG_LONG_MAX; /* "infinity" */
229
230static int per_process_stats;
231static int track_ios;
232static int ppi_hash_by_pid = 1;
233static int print_missing;
234static unsigned int act_mask = -1U;
235
236static unsigned int t_alloc_cache;
237static unsigned int bit_alloc_cache;
238
239#define RB_BATCH_DEFAULT (512)
240static unsigned int rb_batch = RB_BATCH_DEFAULT;
241
242static int pipeline;
243
244#define is_done() (*(volatile int *)(&done))
245static volatile int done;
246
247#define JHASH_RANDOM (0x3af5f2ee)
248
249static inline int ppi_hash_pid(__u32 pid)
250{
251 return jhash_1word(pid, JHASH_RANDOM) & PPI_HASH_MASK;
252}
253
254static inline int ppi_hash_name(const char *name)
255{
256 return jhash(name, 16, JHASH_RANDOM) & PPI_HASH_MASK;
257}
258
259static inline int ppi_hash(struct per_process_info *ppi)
260{
261 if (ppi_hash_by_pid)
262 return ppi_hash_pid(ppi->pid);
263
264 return ppi_hash_name(ppi->name);
265}
266
267static inline void add_process_to_hash(struct per_process_info *ppi)
268{
269 const int hash_idx = ppi_hash(ppi);
270
271 ppi->hash_next = ppi_hash_table[hash_idx];
272 ppi_hash_table[hash_idx] = ppi;
273}
274
275static inline void add_process_to_list(struct per_process_info *ppi)
276{
277 ppi->list_next = ppi_list;
278 ppi_list = ppi;
279 ppi_list_entries++;
280}
281
282static struct per_process_info *find_process_by_name(char *name)
283{
284 const int hash_idx = ppi_hash_name(name);
285 struct per_process_info *ppi;
286
287 ppi = ppi_hash_table[hash_idx];
288 while (ppi) {
289 if (!strcmp(ppi->name, name))
290 return ppi;
291
292 ppi = ppi->hash_next;
293 }
294
295 return NULL;
296}
297
298static struct per_process_info *find_process_by_pid(__u32 pid)
299{
300 const int hash_idx = ppi_hash_pid(pid);
301 struct per_process_info *ppi;
302
303 ppi = ppi_hash_table[hash_idx];
304 while (ppi) {
305 if (ppi->pid == pid)
306 return ppi;
307
308 ppi = ppi->hash_next;
309 }
310
311 return NULL;
312}
313
314static struct per_process_info *find_process(__u32 pid, char *name)
315{
316 struct per_process_info *ppi;
317
318 if (ppi_hash_by_pid)
319 return find_process_by_pid(pid);
320
321 ppi = find_process_by_name(name);
322 if (ppi && ppi->pid != pid)
323 ppi->more_than_one = 1;
324
325 return ppi;
326}
327
328static inline int trace_rb_insert(struct trace *t, struct rb_root *root,
329 int check_time)
330{
331 struct rb_node **p = &root->rb_node;
332 struct rb_node *parent = NULL;
333 struct trace *__t;
334
335 while (*p) {
336 parent = *p;
337
338 __t = rb_entry(parent, struct trace, rb_node);
339
340 if (check_time) {
341 if (t->bit->time < __t->bit->time) {
342 p = &(*p)->rb_left;
343 continue;
344 } else if (t->bit->time > __t->bit->time) {
345 p = &(*p)->rb_right;
346 continue;
347 }
348 }
349 if (t->bit->device < __t->bit->device)
350 p = &(*p)->rb_left;
351 else if (t->bit->device > __t->bit->device)
352 p = &(*p)->rb_right;
353 else if (t->bit->sequence < __t->bit->sequence)
354 p = &(*p)->rb_left;
355 else /* >= sequence */
356 p = &(*p)->rb_right;
357 }
358
359 rb_link_node(&t->rb_node, parent, p);
360 rb_insert_color(&t->rb_node, root);
361 return 0;
362}
363
364static inline int trace_rb_insert_sort(struct trace *t)
365{
366 if (!trace_rb_insert(t, &rb_sort_root, 1)) {
367 rb_sort_entries++;
368 return 0;
369 }
370
371 return 1;
372}
373
374static inline int trace_rb_insert_last(struct per_dev_info *pdi,struct trace *t)
375{
376 if (!trace_rb_insert(t, &pdi->rb_last, 1)) {
377 pdi->rb_last_entries++;
378 return 0;
379 }
380
381 return 1;
382}
383
384static struct trace *trace_rb_find(dev_t device, unsigned long sequence,
385 struct rb_root *root, int order)
386{
387 struct rb_node *n = root->rb_node;
388 struct rb_node *prev = NULL;
389 struct trace *__t;
390
391 while (n) {
392 __t = rb_entry(n, struct trace, rb_node);
393 prev = n;
394
395 if (device < __t->bit->device)
396 n = n->rb_left;
397 else if (device > __t->bit->device)
398 n = n->rb_right;
399 else if (sequence < __t->bit->sequence)
400 n = n->rb_left;
401 else if (sequence > __t->bit->sequence)
402 n = n->rb_right;
403 else
404 return __t;
405 }
406
407 /*
408 * hack - the list may not be sequence ordered because some
409 * events don't have sequence and time matched. so we end up
410 * being a little off in the rb lookup here, because we don't
411 * know the time we are looking for. compensate by browsing
412 * a little ahead from the last entry to find the match
413 */
414 if (order && prev) {
415 int max = 5;
416
417 while (((n = rb_next(prev)) != NULL) && max--) {
418 __t = rb_entry(n, struct trace, rb_node);
419
420 if (__t->bit->device == device &&
421 __t->bit->sequence == sequence)
422 return __t;
423
424 prev = n;
425 }
426 }
427
428 return NULL;
429}
430
431static inline struct trace *trace_rb_find_sort(dev_t dev, unsigned long seq)
432{
433 return trace_rb_find(dev, seq, &rb_sort_root, 1);
434}
435
436static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi,
437 unsigned long seq)
438{
439 return trace_rb_find(pdi->dev, seq, &pdi->rb_last, 0);
440}
441
442static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot)
443{
444 struct rb_node **p = &pdi->rb_track.rb_node;
445 struct rb_node *parent = NULL;
446 struct io_track *__iot;
447
448 while (*p) {
449 parent = *p;
450 __iot = rb_entry(parent, struct io_track, rb_node);
451
452 if (iot->sector < __iot->sector)
453 p = &(*p)->rb_left;
454 else if (iot->sector > __iot->sector)
455 p = &(*p)->rb_right;
456 else {
457 fprintf(stderr,
458 "sector alias (%Lu) on device %d,%d!\n",
459 (unsigned long long) iot->sector,
460 MAJOR(pdi->dev), MINOR(pdi->dev));
461 return 1;
462 }
463 }
464
465 rb_link_node(&iot->rb_node, parent, p);
466 rb_insert_color(&iot->rb_node, &pdi->rb_track);
467 return 0;
468}
469
470static struct io_track *__find_track(struct per_dev_info *pdi, __u64 sector)
471{
472 struct rb_node *n = pdi->rb_track.rb_node;
473 struct io_track *__iot;
474
475 while (n) {
476 __iot = rb_entry(n, struct io_track, rb_node);
477
478 if (sector < __iot->sector)
479 n = n->rb_left;
480 else if (sector > __iot->sector)
481 n = n->rb_right;
482 else
483 return __iot;
484 }
485
486 return NULL;
487}
488
489static struct io_track *find_track(struct per_dev_info *pdi, __u32 pid,
490 char *comm, __u64 sector)
491{
492 struct io_track *iot;
493
494 iot = __find_track(pdi, sector);
495 if (!iot) {
496 iot = malloc(sizeof(*iot));
497 iot->pid = pid;
498 memcpy(iot->comm, comm, sizeof(iot->comm));
499 iot->sector = sector;
500 track_rb_insert(pdi, iot);
501 }
502
503 return iot;
504}
505
506static void log_track_frontmerge(struct per_dev_info *pdi,
507 struct blk_io_trace *t)
508{
509 struct io_track *iot;
510
511 if (!track_ios)
512 return;
513
514 iot = __find_track(pdi, t->sector + (t->bytes >> 9));
515 if (!iot) {
516 fprintf(stderr, "merge not found for (%d,%d): %llu\n",
517 MAJOR(pdi->dev), MINOR(pdi->dev),
518 (unsigned long long) t->sector + (t->bytes >> 9));
519 return;
520 }
521
522 rb_erase(&iot->rb_node, &pdi->rb_track);
523 iot->sector -= t->bytes >> 9;
524 track_rb_insert(pdi, iot);
525}
526
527static void log_track_getrq(struct per_dev_info *pdi, struct blk_io_trace *t)
528{
529 struct io_track *iot;
530
531 if (!track_ios)
532 return;
533
534 iot = find_track(pdi, t->pid, t->comm, t->sector);
535 iot->allocation_time = t->time;
536}
537
538/*
539 * return time between rq allocation and insertion
540 */
541static unsigned long long log_track_insert(struct per_dev_info *pdi,
542 struct blk_io_trace *t)
543{
544 unsigned long long elapsed;
545 struct io_track *iot;
546
547 if (!track_ios)
548 return -1;
549
550 iot = find_track(pdi, t->pid, t->comm, t->sector);
551 iot->queue_time = t->time;
552
553 if (!iot->allocation_time)
554 return -1;
555
556 elapsed = iot->queue_time - iot->allocation_time;
557
558 if (per_process_stats) {
559 struct per_process_info *ppi = find_process(iot->pid,iot->comm);
560 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
561
562 if (ppi && elapsed > ppi->longest_allocation_wait[w])
563 ppi->longest_allocation_wait[w] = elapsed;
564 }
565
566 return elapsed;
567}
568
569/*
570 * return time between queue and issue
571 */
572static unsigned long long log_track_issue(struct per_dev_info *pdi,
573 struct blk_io_trace *t)
574{
575 unsigned long long elapsed;
576 struct io_track *iot;
577
578 if (!track_ios)
579 return -1;
580 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
581 return -1;
582
583 iot = __find_track(pdi, t->sector);
584 if (!iot) {
585 fprintf(stderr, "issue not found for (%d,%d): %llu\n",
586 MAJOR(pdi->dev), MINOR(pdi->dev),
587 (unsigned long long) t->sector);
588 return -1;
589 }
590
591 iot->dispatch_time = t->time;
592 elapsed = iot->dispatch_time - iot->queue_time;
593
594 if (per_process_stats) {
595 struct per_process_info *ppi = find_process(iot->pid,iot->comm);
596 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
597
598 if (ppi && elapsed > ppi->longest_dispatch_wait[w])
599 ppi->longest_dispatch_wait[w] = elapsed;
600 }
601
602 return elapsed;
603}
604
605/*
606 * return time between dispatch and complete
607 */
608static unsigned long long log_track_complete(struct per_dev_info *pdi,
609 struct blk_io_trace *t)
610{
611 unsigned long long elapsed;
612 struct io_track *iot;
613
614 if (!track_ios)
615 return -1;
616 if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
617 return -1;
618
619 iot = __find_track(pdi, t->sector);
620 if (!iot) {
621 fprintf(stderr, "complete not found for (%d,%d): %llu\n",
622 MAJOR(pdi->dev), MINOR(pdi->dev),
623 (unsigned long long) t->sector);
624 return -1;
625 }
626
627 iot->completion_time = t->time;
628 elapsed = iot->completion_time - iot->dispatch_time;
629
630 if (per_process_stats) {
631 struct per_process_info *ppi = find_process(iot->pid,iot->comm);
632 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
633
634 if (ppi && elapsed > ppi->longest_completion_wait[w])
635 ppi->longest_completion_wait[w] = elapsed;
636 }
637
638 /*
639 * kill the trace, we don't need it after completion
640 */
641 rb_erase(&iot->rb_node, &pdi->rb_track);
642 free(iot);
643
644 return elapsed;
645}
646
647
648static struct io_stats *find_process_io_stats(__u32 pid, char *name)
649{
650 struct per_process_info *ppi = find_process(pid, name);
651
652 if (!ppi) {
653 ppi = malloc(sizeof(*ppi));
654 memset(ppi, 0, sizeof(*ppi));
655 memcpy(ppi->name, name, 16);
656 ppi->pid = pid;
657 add_process_to_hash(ppi);
658 add_process_to_list(ppi);
659 }
660
661 return &ppi->io_stats;
662}
663
664static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
665{
666 struct per_cpu_info *cpus = pdi->cpus;
667 int ncpus = pdi->ncpus;
668 int new_count = cpu + 1;
669 int new_space, size;
670 char *new_start;
671
672 size = new_count * sizeof(struct per_cpu_info);
673 cpus = realloc(cpus, size);
674 if (!cpus) {
675 char name[20];
676 fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
677 get_dev_name(pdi, name, sizeof(name)), size);
678 exit(1);
679 }
680
681 new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
682 new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
683 memset(new_start, 0, new_space);
684
685 pdi->ncpus = new_count;
686 pdi->cpus = cpus;
687}
688
689static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
690{
691 struct per_cpu_info *pci;
692
693 if (cpu >= pdi->ncpus)
694 resize_cpu_info(pdi, cpu);
695
696 pci = &pdi->cpus[cpu];
697 pci->cpu = cpu;
698 return pci;
699}
700
701
702static int resize_devices(char *name)
703{
704 int size = (ndevices + 1) * sizeof(struct per_dev_info);
705
706 devices = realloc(devices, size);
707 if (!devices) {
708 fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
709 return 1;
710 }
711 memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
712 devices[ndevices].name = name;
713 ndevices++;
714 return 0;
715}
716
717static struct per_dev_info *get_dev_info(dev_t dev)
718{
719 struct per_dev_info *pdi;
720 int i;
721
722 for (i = 0; i < ndevices; i++) {
723 if (!devices[i].dev)
724 devices[i].dev = dev;
725 if (devices[i].dev == dev)
726 return &devices[i];
727 }
728
729 if (resize_devices(NULL))
730 return NULL;
731
732 pdi = &devices[ndevices - 1];
733 pdi->dev = dev;
734 pdi->last_sequence = -1;
735 pdi->last_read_time = 0;
736 memset(&pdi->rb_last, 0, sizeof(pdi->rb_last));
737 pdi->rb_last_entries = 0;
738 return pdi;
739}
740
741static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size)
742{
743 if (pdi->name)
744 snprintf(buffer, size, "%s", pdi->name);
745 else
746 snprintf(buffer, size, "%d,%d",MAJOR(pdi->dev),MINOR(pdi->dev));
747 return buffer;
748}
749
750static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit)
751{
752 unsigned long long this = bit->time;
753 unsigned long long last = pdi->last_reported_time;
754
755 pdi->backwards = (this < last) ? 'B' : ' ';
756 pdi->last_reported_time = this;
757}
758
759static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t,
760 int rw)
761{
762 if (rw) {
763 ios->mwrites++;
764 ios->qwrite_kb += t->bytes >> 10;
765 } else {
766 ios->mreads++;
767 ios->qread_kb += t->bytes >> 10;
768 }
769}
770
771static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci,
772 int rw)
773{
774 __account_m(&pci->io_stats, t, rw);
775
776 if (per_process_stats) {
777 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
778
779 __account_m(ios, t, rw);
780 }
781}
782
783static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t,
784 int rw)
785{
786 if (rw) {
787 ios->qwrites++;
788 ios->qwrite_kb += t->bytes >> 10;
789 } else {
790 ios->qreads++;
791 ios->qread_kb += t->bytes >> 10;
792 }
793}
794
795static inline void account_queue(struct blk_io_trace *t,
796 struct per_cpu_info *pci, int rw)
797{
798 __account_queue(&pci->io_stats, t, rw);
799
800 if (per_process_stats) {
801 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
802
803 __account_queue(ios, t, rw);
804 }
805}
806
807static inline void __account_c(struct io_stats *ios, int rw, unsigned int bytes)
808{
809 if (rw) {
810 ios->cwrites++;
811 ios->cwrite_kb += bytes >> 10;
812 } else {
813 ios->creads++;
814 ios->cread_kb += bytes >> 10;
815 }
816}
817
818static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci,
819 int rw, int bytes)
820{
821 __account_c(&pci->io_stats, rw, bytes);
822
823 if (per_process_stats) {
824 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
825
826 __account_c(ios, rw, bytes);
827 }
828}
829
830static inline void __account_issue(struct io_stats *ios, int rw,
831 unsigned int bytes)
832{
833 if (rw) {
834 ios->iwrites++;
835 ios->iwrite_kb += bytes >> 10;
836 } else {
837 ios->ireads++;
838 ios->iread_kb += bytes >> 10;
839 }
840}
841
842static inline void account_issue(struct blk_io_trace *t,
843 struct per_cpu_info *pci, int rw)
844{
845 __account_issue(&pci->io_stats, rw, t->bytes);
846
847 if (per_process_stats) {
848 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
849
850 __account_issue(ios, rw, t->bytes);
851 }
852}
853
854static inline void __account_unplug(struct io_stats *ios, int timer)
855{
856 if (timer)
857 ios->timer_unplugs++;
858 else
859 ios->io_unplugs++;
860}
861
862static inline void account_unplug(struct blk_io_trace *t,
863 struct per_cpu_info *pci, int timer)
864{
865 __account_unplug(&pci->io_stats, timer);
866
867 if (per_process_stats) {
868 struct io_stats *ios = find_process_io_stats(t->pid, t->comm);
869
870 __account_unplug(ios, timer);
871 }
872}
873
874static void log_complete(struct per_dev_info *pdi, struct per_cpu_info *pci,
875 struct blk_io_trace *t, char *act)
876{
877 process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL);
878}
879
880static void log_insert(struct per_dev_info *pdi, struct per_cpu_info *pci,
881 struct blk_io_trace *t, char *act)
882{
883 process_fmt(act, pci, t, log_track_insert(pdi, t), 0, NULL);
884}
885
886static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
887 char *act)
888{
889 process_fmt(act, pci, t, -1, 0, NULL);
890}
891
892static void log_issue(struct per_dev_info *pdi, struct per_cpu_info *pci,
893 struct blk_io_trace *t, char *act)
894{
895 process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL);
896}
897
898static void log_merge(struct per_dev_info *pdi, struct per_cpu_info *pci,
899 struct blk_io_trace *t, char *act)
900{
901 if (act[0] == 'F')
902 log_track_frontmerge(pdi, t);
903
904 process_fmt(act, pci, t, -1ULL, 0, NULL);
905}
906
907static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t,
908 char *act)
909{
910 process_fmt(act, pci, t, -1ULL, 0, NULL);
911}
912
913static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
914 char *act)
915{
916 process_fmt(act, pci, t, -1ULL, 0, NULL);
917}
918
919static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
920 char *act)
921{
922 process_fmt(act, pci, t, -1ULL, 0, NULL);
923}
924
925static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t,
926 char *act)
927{
928 process_fmt(act, pci, t, -1ULL, 0, NULL);
929}
930
931static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act)
932{
933 unsigned char *buf = (unsigned char *) t + sizeof(*t);
934
935 process_fmt(act, pci, t, -1ULL, t->pdu_len, buf);
936}
937
938static void dump_trace_pc(struct blk_io_trace *t, struct per_cpu_info *pci)
939{
940 int act = t->action & 0xffff;
941
942 switch (act) {
943 case __BLK_TA_QUEUE:
944 log_generic(pci, t, "Q");
945 break;
946 case __BLK_TA_GETRQ:
947 log_generic(pci, t, "G");
948 break;
949 case __BLK_TA_SLEEPRQ:
950 log_generic(pci, t, "S");
951 break;
952 case __BLK_TA_REQUEUE:
953 log_generic(pci, t, "R");
954 break;
955 case __BLK_TA_ISSUE:
956 log_pc(pci, t, "D");
957 break;
958 case __BLK_TA_COMPLETE:
959 log_pc(pci, t, "C");
960 break;
961 case __BLK_TA_INSERT:
962 log_pc(pci, t, "I");
963 break;
964 default:
965 fprintf(stderr, "Bad pc action %x\n", act);
966 break;
967 }
968}
969
970static void dump_trace_fs(struct blk_io_trace *t, struct per_dev_info *pdi,
971 struct per_cpu_info *pci)
972{
973 int w = t->action & BLK_TC_ACT(BLK_TC_WRITE);
974 int act = t->action & 0xffff;
975
976 switch (act) {
977 case __BLK_TA_QUEUE:
978 account_queue(t, pci, w);
979 log_queue(pci, t, "Q");
980 break;
981 case __BLK_TA_INSERT:
982 log_insert(pdi, pci, t, "I");
983 break;
984 case __BLK_TA_BACKMERGE:
985 account_m(t, pci, w);
986 log_merge(pdi, pci, t, "M");
987 break;
988 case __BLK_TA_FRONTMERGE:
989 account_m(t, pci, w);
990 log_merge(pdi, pci, t, "F");
991 break;
992 case __BLK_TA_GETRQ:
993 log_track_getrq(pdi, t);
994 log_generic(pci, t, "G");
995 break;
996 case __BLK_TA_SLEEPRQ:
997 log_generic(pci, t, "S");
998 break;
999 case __BLK_TA_REQUEUE:
1000 account_c(t, pci, w, -t->bytes);
1001 log_queue(pci, t, "R");
1002 break;
1003 case __BLK_TA_ISSUE:
1004 account_issue(t, pci, w);
1005 log_issue(pdi, pci, t, "D");
1006 break;
1007 case __BLK_TA_COMPLETE:
1008 account_c(t, pci, w, t->bytes);
1009 log_complete(pdi, pci, t, "C");
1010 break;
1011 case __BLK_TA_PLUG:
1012 log_action(pci, t, "P");
1013 break;
1014 case __BLK_TA_UNPLUG_IO:
1015 account_unplug(t, pci, 0);
1016 log_unplug(pci, t, "U");
1017 break;
1018 case __BLK_TA_UNPLUG_TIMER:
1019 account_unplug(t, pci, 1);
1020 log_unplug(pci, t, "UT");
1021 break;
1022 case __BLK_TA_SPLIT:
1023 log_split(pci, t, "X");
1024 break;
1025 case __BLK_TA_BOUNCE:
1026 log_generic(pci, t, "B");
1027 break;
1028 default:
1029 fprintf(stderr, "Bad fs action %x\n", t->action);
1030 break;
1031 }
1032}
1033
1034static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
1035 struct per_dev_info *pdi)
1036{
1037 if (t->action & BLK_TC_ACT(BLK_TC_PC))
1038 dump_trace_pc(t, pci);
1039 else
1040 dump_trace_fs(t, pdi, pci);
1041
1042 pdi->events++;
1043}
1044
1045static void dump_io_stats(struct io_stats *ios, char *msg)
1046{
1047 fprintf(ofp, "%s\n", msg);
1048
1049 fprintf(ofp, " Reads Queued: %'8lu, %'8LuKiB\t", ios->qreads, ios->qread_kb);
1050 fprintf(ofp, " Writes Queued: %'8lu, %'8LuKiB\n", ios->qwrites,ios->qwrite_kb);
1051
1052 fprintf(ofp, " Read Dispatches: %'8lu, %'8LuKiB\t", ios->ireads, ios->iread_kb);
1053 fprintf(ofp, " Write Dispatches: %'8lu, %'8LuKiB\n", ios->iwrites,ios->iwrite_kb);
1054 fprintf(ofp, " Reads Completed: %'8lu, %'8LuKiB\t", ios->creads, ios->cread_kb);
1055 fprintf(ofp, " Writes Completed: %'8lu, %'8LuKiB\n", ios->cwrites,ios->cwrite_kb);
1056 fprintf(ofp, " Read Merges: %'8lu%8c\t", ios->mreads, ' ');
1057 fprintf(ofp, " Write Merges: %'8lu\n", ios->mwrites);
1058 fprintf(ofp, " IO unplugs: %'8lu%8c\t", ios->io_unplugs, ' ');
1059 fprintf(ofp, " Timer unplugs: %'8lu\n", ios->timer_unplugs);
1060}
1061
1062static void dump_wait_stats(struct per_process_info *ppi)
1063{
1064 unsigned long rawait = ppi->longest_allocation_wait[0] / 1000;
1065 unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000;
1066 unsigned long rcwait = ppi->longest_completion_wait[0] / 1000;
1067 unsigned long wawait = ppi->longest_allocation_wait[1] / 1000;
1068 unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000;
1069 unsigned long wcwait = ppi->longest_completion_wait[1] / 1000;
1070
1071 fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' ');
1072 fprintf(ofp, " Allocation wait: %'8lu\n", wawait);
1073 fprintf(ofp, " Dispatch wait: %'8lu%8c\t", rdwait, ' ');
1074 fprintf(ofp, " Dispatch wait: %'8lu\n", wdwait);
1075 fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' ');
1076 fprintf(ofp, " Completion wait: %'8lu\n", wcwait);
1077}
1078
1079static int ppi_name_compare(const void *p1, const void *p2)
1080{
1081 struct per_process_info *ppi1 = *((struct per_process_info **) p1);
1082 struct per_process_info *ppi2 = *((struct per_process_info **) p2);
1083 int res;
1084
1085 res = strverscmp(ppi1->name, ppi2->name);
1086 if (!res)
1087 res = ppi1->pid > ppi2->pid;
1088
1089 return res;
1090}
1091
1092static void sort_process_list(void)
1093{
1094 struct per_process_info **ppis;
1095 struct per_process_info *ppi;
1096 int i = 0;
1097
1098 ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *));
1099
1100 ppi = ppi_list;
1101 while (ppi) {
1102 ppis[i++] = ppi;
1103 ppi = ppi->list_next;
1104 }
1105
1106 qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare);
1107
1108 i = ppi_list_entries - 1;
1109 ppi_list = NULL;
1110 while (i >= 0) {
1111 ppi = ppis[i];
1112
1113 ppi->list_next = ppi_list;
1114 ppi_list = ppi;
1115 i--;
1116 }
1117
1118 free(ppis);
1119}
1120
1121static void show_process_stats(void)
1122{
1123 struct per_process_info *ppi;
1124
1125 sort_process_list();
1126
1127 ppi = ppi_list;
1128 while (ppi) {
1129 char name[64];
1130
1131 if (ppi->more_than_one)
1132 sprintf(name, "%s (%u, ...)", ppi->name, ppi->pid);
1133 else
1134 sprintf(name, "%s (%u)", ppi->name, ppi->pid);
1135
1136 dump_io_stats(&ppi->io_stats, name);
1137 dump_wait_stats(ppi);
1138 ppi = ppi->list_next;
1139 }
1140
1141 fprintf(ofp, "\n");
1142}
1143
1144static void show_device_and_cpu_stats(void)
1145{
1146 struct per_dev_info *pdi;
1147 struct per_cpu_info *pci;
1148 struct io_stats total, *ios;
1149 int i, j, pci_events;
1150 char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
1151 char name[32];
1152
1153 for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1154
1155 memset(&total, 0, sizeof(total));
1156 pci_events = 0;
1157
1158 if (i > 0)
1159 fprintf(ofp, "\n");
1160
1161 for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) {
1162 if (!pci->nelems)
1163 continue;
1164
1165 ios = &pci->io_stats;
1166 total.qreads += ios->qreads;
1167 total.qwrites += ios->qwrites;
1168 total.creads += ios->creads;
1169 total.cwrites += ios->cwrites;
1170 total.mreads += ios->mreads;
1171 total.mwrites += ios->mwrites;
1172 total.ireads += ios->ireads;
1173 total.iwrites += ios->iwrites;
1174 total.qread_kb += ios->qread_kb;
1175 total.qwrite_kb += ios->qwrite_kb;
1176 total.cread_kb += ios->cread_kb;
1177 total.cwrite_kb += ios->cwrite_kb;
1178 total.iread_kb += ios->iread_kb;
1179 total.iwrite_kb += ios->iwrite_kb;
1180 total.timer_unplugs += ios->timer_unplugs;
1181 total.io_unplugs += ios->io_unplugs;
1182
1183 snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
1184 j, get_dev_name(pdi, name, sizeof(name)));
1185 dump_io_stats(ios, line);
1186 pci_events++;
1187 }
1188
1189 if (pci_events > 1) {
1190 fprintf(ofp, "\n");
1191 snprintf(line, sizeof(line) - 1, "Total (%s):",
1192 get_dev_name(pdi, name, sizeof(name)));
1193 dump_io_stats(&total, line);
1194 }
1195
1196 fprintf(ofp, "\nEvents (%s): %'Lu entries, %'lu skips\n",
1197 get_dev_name(pdi, line, sizeof(line)), pdi->events,
1198 pdi->skips);
1199 }
1200}
1201
1202/*
1203 * struct trace and blktrace allocation cache, we do potentially
1204 * millions of mallocs for these structures while only using at most
1205 * a few thousand at the time
1206 */
1207static inline void t_free(struct trace *t)
1208{
1209 if (t_alloc_cache < 1024) {
1210 t->next = t_alloc_list;
1211 t_alloc_list = t;
1212 t_alloc_cache++;
1213 } else
1214 free(t);
1215}
1216
1217static inline struct trace *t_alloc(void)
1218{
1219 struct trace *t = t_alloc_list;
1220
1221 if (t) {
1222 t_alloc_list = t->next;
1223 t_alloc_cache--;
1224 return t;
1225 }
1226
1227 return malloc(sizeof(*t));
1228}
1229
1230static inline void bit_free(struct blk_io_trace *bit)
1231{
1232 if (bit_alloc_cache < 1024) {
1233 /*
1234 * abuse a 64-bit field for a next pointer for the free item
1235 */
1236 bit->time = (__u64) (unsigned long) bit_alloc_list;
1237 bit_alloc_list = (struct blk_io_trace *) bit;
1238 bit_alloc_cache++;
1239 } else
1240 free(bit);
1241}
1242
1243static inline struct blk_io_trace *bit_alloc(void)
1244{
1245 struct blk_io_trace *bit = bit_alloc_list;
1246
1247 if (bit) {
1248 bit_alloc_list = (struct blk_io_trace *) (unsigned long) \
1249 bit->time;
1250 bit_alloc_cache--;
1251 return bit;
1252 }
1253
1254 return malloc(sizeof(*bit));
1255}
1256
1257static void find_genesis(void)
1258{
1259 struct trace *t = trace_list;
1260
1261 genesis_time = -1ULL;
1262 while (t != NULL) {
1263 if (t->bit->time < genesis_time)
1264 genesis_time = t->bit->time;
1265
1266 t = t->next;
1267 }
1268}
1269
1270static inline int check_stopwatch(struct blk_io_trace *bit)
1271{
1272 if (bit->time < stopwatch_end &&
1273 bit->time >= stopwatch_start)
1274 return 0;
1275
1276 return 1;
1277}
1278
1279/*
1280 * return youngest entry read
1281 */
1282static int sort_entries(unsigned long long *youngest)
1283{
1284 struct trace *t;
1285
1286 if (!genesis_time)
1287 find_genesis();
1288
1289 *youngest = 0;
1290 while ((t = trace_list) != NULL) {
1291 struct blk_io_trace *bit = t->bit;
1292
1293 trace_list = t->next;
1294
1295 bit->time -= genesis_time;
1296
1297 if (bit->time < *youngest || !*youngest)
1298 *youngest = bit->time;
1299
1300 if (check_stopwatch(bit)) {
1301 bit_free(bit);
1302 t_free(t);
1303 continue;
1304 }
1305
1306 if (trace_rb_insert_sort(t))
1307 return -1;
1308
1309 if (bit->sequence < smallest_seq_read)
1310 smallest_seq_read = bit->sequence;
1311 }
1312
1313 return 0;
1314}
1315
1316static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t)
1317{
1318 rb_erase(&t->rb_node, &pdi->rb_last);
1319 pdi->rb_last_entries--;
1320
1321 bit_free(t->bit);
1322 t_free(t);
1323}
1324
1325static void put_trace(struct per_dev_info *pdi, struct trace *t)
1326{
1327 rb_erase(&t->rb_node, &rb_sort_root);
1328 rb_sort_entries--;
1329
1330 trace_rb_insert_last(pdi, t);
1331
1332 if (pdi->rb_last_entries > rb_batch * pdi->nfiles) {
1333 struct rb_node *n = rb_first(&pdi->rb_last);
1334
1335 t = rb_entry(n, struct trace, rb_node);
1336 __put_trace_last(pdi, t);
1337 }
1338}
1339
1340static int check_sequence(struct per_dev_info *pdi, struct blk_io_trace *bit)
1341{
1342 unsigned long expected_sequence = pdi->last_sequence + 1;
1343 struct trace *t;
1344
1345 /*
1346 * first entry, always ok
1347 */
1348 if (!expected_sequence)
1349 return 0;
1350
1351 if (bit->sequence == expected_sequence)
1352 return 0;
1353
1354 /*
1355 * we may not have seen that sequence yet. if we are not doing
1356 * the final run, break and wait for more entries.
1357 */
1358 if (expected_sequence < smallest_seq_read) {
1359 t = trace_rb_find_last(pdi, expected_sequence);
1360 if (!t)
1361 goto skip;
1362
1363 __put_trace_last(pdi, t);
1364 return 0;
1365 } else {
1366skip:
1367 if (print_missing) {
1368 fprintf(stderr, "(%d,%d): skipping %lu -> %u\n",
1369 MAJOR(pdi->dev), MINOR(pdi->dev),
1370 pdi->last_sequence, bit->sequence);
1371 }
1372 pdi->skips++;
1373 return 0;
1374 }
1375}
1376
1377static void show_entries_rb(int force)
1378{
1379 struct per_dev_info *pdi = NULL;
1380 struct per_cpu_info *pci = NULL;
1381 struct blk_io_trace *bit;
1382 struct rb_node *n;
1383 struct trace *t;
1384
1385 while ((n = rb_first(&rb_sort_root)) != NULL) {
1386 if (is_done() && !force && !pipeline)
1387 break;
1388
1389 t = rb_entry(n, struct trace, rb_node);
1390 bit = t->bit;
1391
1392 if (!pdi || pdi->dev != bit->device)
1393 pdi = get_dev_info(bit->device);
1394
1395 if (!pdi) {
1396 fprintf(stderr, "Unknown device ID? (%d,%d)\n",
1397 MAJOR(bit->device), MINOR(bit->device));
1398 break;
1399 }
1400
1401 if (!force) {
1402 if (check_sequence(pdi, bit))
1403 break;
1404
1405 if (bit->time > last_allowed_time)
1406 break;
1407 }
1408
1409 pdi->last_sequence = bit->sequence;
1410
1411 check_time(pdi, bit);
1412
1413 if (!pci || pci->cpu != bit->cpu)
1414 pci = get_cpu_info(pdi, bit->cpu);
1415
1416 if (bit->action & (act_mask << BLK_TC_SHIFT))
1417 dump_trace(bit, pci, pdi);
1418
1419 put_trace(pdi, t);
1420 }
1421}
1422
1423static int read_data(int fd, void *buffer, int bytes, int block)
1424{
1425 int ret, bytes_left, fl;
1426 void *p;
1427
1428 fl = fcntl(fd, F_GETFL);
1429
1430 if (!block)
1431 fcntl(fd, F_SETFL, fl | O_NONBLOCK);
1432 else
1433 fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
1434
1435 bytes_left = bytes;
1436 p = buffer;
1437 while (bytes_left > 0) {
1438 ret = read(fd, p, bytes_left);
1439 if (!ret)
1440 return 1;
1441 else if (ret < 0) {
1442 if (errno != EAGAIN)
1443 perror("read");
1444
1445 return -1;
1446 } else {
1447 p += ret;
1448 bytes_left -= ret;
1449 }
1450 }
1451
1452 return 0;
1453}
1454
1455static int read_events(int fd, int always_block)
1456{
1457 struct per_dev_info *pdi = NULL;
1458 unsigned int events = 0;
1459
1460 while (!is_done() && events < rb_batch) {
1461 struct blk_io_trace *bit;
1462 struct trace *t;
1463 int pdu_len;
1464 __u32 magic;
1465
1466 bit = bit_alloc();
1467
1468 if (read_data(fd, bit, sizeof(*bit), !events || always_block))
1469 break;
1470
1471 magic = be32_to_cpu(bit->magic);
1472 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1473 fprintf(stderr, "Bad magic %x\n", magic);
1474 break;
1475 }
1476
1477 pdu_len = be16_to_cpu(bit->pdu_len);
1478 if (pdu_len) {
1479 void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
1480
1481 if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1))
1482 break;
1483
1484 bit = ptr;
1485 }
1486
1487 trace_to_cpu(bit);
1488
1489 if (verify_trace(bit)) {
1490 bit_free(bit);
1491 continue;
1492 }
1493
1494 t = t_alloc();
1495 memset(t, 0, sizeof(*t));
1496 t->bit = bit;
1497
1498 t->next = trace_list;
1499 trace_list = t;
1500
1501 if (!pdi || pdi->dev != bit->device)
1502 pdi = get_dev_info(bit->device);
1503
1504 if (bit->time > pdi->last_read_time)
1505 pdi->last_read_time = bit->time;
1506
1507 events++;
1508 }
1509
1510 return events;
1511}
1512
1513static int do_file(void)
1514{
1515 struct per_cpu_info *pci;
1516 struct per_dev_info *pdi;
1517 int i, j, events, events_added;
1518
1519 /*
1520 * first prepare all files for reading
1521 */
1522 for (i = 0; i < ndevices; i++) {
1523 pdi = &devices[i];
1524 pdi->nfiles = 0;
1525 pdi->last_sequence = -1;
1526
1527 for (j = 0;; j++) {
1528 struct stat st;
1529 int len = 0;
1530
1531 pci = get_cpu_info(pdi, j);
1532 pci->cpu = j;
1533 pci->fd = -1;
1534
1535 if (input_dir)
1536 len = sprintf(pci->fname, "%s/", input_dir);
1537
1538 snprintf(pci->fname + len, sizeof(pci->fname)-1-len,
1539 "%s.blktrace.%d", pdi->name, pci->cpu);
1540 if (stat(pci->fname, &st) < 0)
1541 break;
1542 if (st.st_size) {
1543 pci->fd = open(pci->fname, O_RDONLY);
1544 if (pci->fd < 0) {
1545 perror(pci->fname);
1546 continue;
1547 }
1548 }
1549
1550 printf("Input file %s added\n", pci->fname);
1551 pdi->nfiles++;
1552 }
1553 }
1554
1555 /*
1556 * now loop over the files reading in the data
1557 */
1558 do {
1559 unsigned long long youngest;
1560
1561 events_added = 0;
1562 last_allowed_time = -1ULL;
1563 smallest_seq_read = -1U;
1564
1565 for (i = 0; i < ndevices; i++) {
1566 pdi = &devices[i];
1567
1568 for (j = 0; j < pdi->nfiles; j++) {
1569
1570 pci = get_cpu_info(pdi, j);
1571
1572 if (pci->fd == -1)
1573 continue;
1574
1575 events = read_events(pci->fd, 1);
1576 if (!events) {
1577 close(pci->fd);
1578 pci->fd = -1;
1579 continue;
1580 }
1581
1582 if (pdi->last_read_time < last_allowed_time)
1583 last_allowed_time = pdi->last_read_time;
1584
1585 events_added += events;
1586 }
1587 }
1588
1589 if (sort_entries(&youngest))
1590 break;
1591
1592 if (youngest > stopwatch_end)
1593 break;
1594
1595 show_entries_rb(0);
1596
1597 } while (events_added);
1598
1599 if (rb_sort_entries)
1600 show_entries_rb(1);
1601
1602 return 0;
1603}
1604
1605static int do_stdin(void)
1606{
1607 unsigned long long youngest;
1608 int fd, events, loops;
1609
1610 last_allowed_time = -1ULL;
1611 fd = dup(STDIN_FILENO);
1612 if (fd == -1) {
1613 perror("dup stdin");
1614 return -1;
1615 }
1616
1617 loops = 0;
1618 while ((events = read_events(fd, 0)) != 0) {
1619
1620 smallest_seq_read = -1U;
1621
1622 if (sort_entries(&youngest))
1623 break;
1624
1625 if (youngest > stopwatch_end)
1626 break;
1627
1628 if (loops++ & 1)
1629 show_entries_rb(0);
1630 }
1631
1632 if (rb_sort_entries)
1633 show_entries_rb(1);
1634
1635 close(fd);
1636 return 0;
1637}
1638
1639static void flush_output(void)
1640{
1641 fflush(ofp);
1642}
1643
1644static void handle_sigint(__attribute__((__unused__)) int sig)
1645{
1646 done = 1;
1647 flush_output();
1648}
1649
1650/*
1651 * Extract start and duration times from a string, allowing
1652 * us to specify a time interval of interest within a trace.
1653 * Format: "duration" (start is zero) or "start:duration".
1654 */
1655static int find_stopwatch_interval(char *string)
1656{
1657 double value;
1658 char *sp;
1659
1660 value = strtod(string, &sp);
1661 if (sp == string) {
1662 fprintf(stderr,"Invalid stopwatch timer: %s\n", string);
1663 return 1;
1664 }
1665 if (*sp == ':') {
1666 stopwatch_start = DOUBLE_TO_NANO_ULL(value);
1667 string = sp + 1;
1668 value = strtod(string, &sp);
1669 if (sp == string || *sp != '\0') {
1670 fprintf(stderr,"Invalid stopwatch duration time: %s\n",
1671 string);
1672 return 1;
1673 }
1674 } else if (*sp != '\0') {
1675 fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
1676 return 1;
1677 }
1678 stopwatch_end = DOUBLE_TO_NANO_ULL(value);
1679 if (stopwatch_end <= stopwatch_start) {
1680 fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n",
1681 stopwatch_start, stopwatch_end);
1682 return 1;
1683 }
1684
1685 return 0;
1686}
1687
1688static char usage_str[] = \
1689 "[ -i <input name> ] [-o <output name> [ -s ] [ -t ] [ -q ]\n" \
1690 "[ -w start:stop ] [ -f output format ] [ -F format spec ] [ -v] \n\n" \
1691 "\t-i Input file containing trace data, or '-' for stdin\n" \
1692 "\t-D Directory to prepend to input file names\n" \
1693 "\t-o Output file. If not given, output is stdout\n" \
1694 "\t-b stdin read batching\n" \
1695 "\t-s Show per-program io statistics\n" \
1696 "\t-n Hash processes by name, not pid\n" \
1697 "\t-t Track individual ios. Will tell you the time a request took\n" \
1698 "\t to get queued, to get dispatched, and to get completed\n" \
1699 "\t-q Quiet. Don't display any stats at the end of the trace\n" \
1700 "\t-w Only parse data between the given time interval in seconds.\n" \
1701 "\t If 'start' isn't given, blkparse defaults the start time to 0\n" \
1702 "\t -f Output format. Customize the output format. The format field\n" \
1703 "\t identifies can be found in the documentation\n" \
1704 "\t-F Format specification. Can be found in the documentation\n" \
1705 "\t-m Print missing entries\n" \
1706 "\t-v Print program version info\n\n";
1707
1708static void usage(char *prog)
1709{
1710 fprintf(stderr, "Usage: %s %s %s", prog, blkparse_version, usage_str);
1711}
1712
1713int main(int argc, char *argv[])
1714{
1715 char *ofp_buffer;
1716 int i, c, ret, mode;
1717 int per_device_and_cpu_stats = 1;
1718 int act_mask_tmp = 0;
1719
1720 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
1721 switch (c) {
1722 case 'a':
1723 i = find_mask_map(optarg);
1724 if (i < 0) {
1725 fprintf(stderr,"Invalid action mask %s\n",
1726 optarg);
1727 return 1;
1728 }
1729 act_mask_tmp |= i;
1730 break;
1731
1732 case 'A':
1733 if ((sscanf(optarg, "%x", &i) != 1) ||
1734 !valid_act_opt(i)) {
1735 fprintf(stderr,
1736 "Invalid set action mask %s/0x%x\n",
1737 optarg, i);
1738 return 1;
1739 }
1740 act_mask_tmp = i;
1741 break;
1742 case 'i':
1743 if (!strcmp(optarg, "-") && !pipeline)
1744 pipeline = 1;
1745 else if (resize_devices(optarg) != 0)
1746 return 1;
1747 break;
1748 case 'D':
1749 input_dir = optarg;
1750 break;
1751 case 'o':
1752 output_name = optarg;
1753 break;
1754 case 'b':
1755 rb_batch = atoi(optarg);
1756 if (rb_batch <= 0)
1757 rb_batch = RB_BATCH_DEFAULT;
1758 break;
1759 case 's':
1760 per_process_stats = 1;
1761 break;
1762 case 't':
1763 track_ios = 1;
1764 break;
1765 case 'q':
1766 per_device_and_cpu_stats = 0;
1767 break;
1768 case 'w':
1769 if (find_stopwatch_interval(optarg) != 0)
1770 return 1;
1771 break;
1772 case 'f':
1773 set_all_format_specs(optarg);
1774 break;
1775 case 'F':
1776 if (add_format_spec(optarg) != 0)
1777 return 1;
1778 break;
1779 case 'n':
1780 ppi_hash_by_pid = 0;
1781 break;
1782 case 'm':
1783 print_missing = 1;
1784 break;
1785 case 'v':
1786 printf("%s version %s\n", argv[0], blkparse_version);
1787 return 0;
1788 default:
1789 usage(argv[0]);
1790 return 1;
1791 }
1792 }
1793
1794 while (optind < argc) {
1795 if (!strcmp(argv[optind], "-") && !pipeline)
1796 pipeline = 1;
1797 else if (resize_devices(argv[optind]) != 0)
1798 return 1;
1799 optind++;
1800 }
1801
1802 if (!pipeline && !ndevices) {
1803 usage(argv[0]);
1804 return 1;
1805 }
1806
1807 if (act_mask_tmp != 0)
1808 act_mask = act_mask_tmp;
1809
1810 memset(&rb_sort_root, 0, sizeof(rb_sort_root));
1811
1812 signal(SIGINT, handle_sigint);
1813 signal(SIGHUP, handle_sigint);
1814 signal(SIGTERM, handle_sigint);
1815
1816 setlocale(LC_NUMERIC, "en_US");
1817
1818 if (!output_name) {
1819 ofp = fdopen(STDOUT_FILENO, "w");
1820 mode = _IOLBF;
1821 } else {
1822 char ofname[128];
1823
1824 snprintf(ofname, sizeof(ofname) - 1, "%s", output_name);
1825 ofp = fopen(ofname, "w");
1826 mode = _IOFBF;
1827 }
1828
1829 if (!ofp) {
1830 perror("fopen");
1831 return 1;
1832 }
1833
1834 ofp_buffer = malloc(4096);
1835 if (setvbuf(ofp, ofp_buffer, mode, 4096)) {
1836 perror("setvbuf");
1837 return 1;
1838 }
1839
1840 if (pipeline)
1841 ret = do_stdin();
1842 else
1843 ret = do_file();
1844
1845 if (per_process_stats)
1846 show_process_stats();
1847
1848 if (per_device_and_cpu_stats)
1849 show_device_and_cpu_stats();
1850
1851 flush_output();
1852 return ret;
1853}
Block queue io tracer