• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * block queue tracing parse application
3  *
4  * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
5  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
6  *
7  *  This program is free software; you can redistribute it and/or modify
8  *  it under the terms of the GNU General Public License as published by
9  *  the Free Software Foundation; either version 2 of the License, or
10  *  (at your option) any later version.
11  *
12  *  This program is distributed in the hope that it will be useful,
13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *  GNU General Public License for more details.
16  *
17  *  You should have received a copy of the GNU General Public License
18  *  along with this program; if not, write to the Free Software
19  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
20  *
21  */
22 #include <sys/types.h>
23 #include <sys/stat.h>
24 #include <unistd.h>
25 #include <stdio.h>
26 #include <fcntl.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <getopt.h>
30 #include <errno.h>
31 #include <signal.h>
32 #include <locale.h>
33 #include <libgen.h>
34 
35 #include "blktrace.h"
36 #include "rbtree.h"
37 #include "jhash.h"
38 
39 static char blkparse_version[] = "1.0.1";
40 
41 struct skip_info {
42 	unsigned long start, end;
43 	struct skip_info *prev, *next;
44 };
45 
46 struct per_dev_info {
47 	dev_t dev;
48 	char *name;
49 
50 	int backwards;
51 	unsigned long long events;
52 	unsigned long long first_reported_time;
53 	unsigned long long last_reported_time;
54 	unsigned long long last_read_time;
55 	struct io_stats io_stats;
56 	unsigned long skips;
57 	unsigned long long seq_skips;
58 	unsigned int max_depth[2];
59 	unsigned int cur_depth[2];
60 
61 	struct rb_root rb_track;
62 
63 	int nfiles;
64 	int ncpus;
65 
66 	unsigned long *cpu_map;
67 	unsigned int cpu_map_max;
68 
69 	struct per_cpu_info *cpus;
70 };
71 
72 /*
73  * some duplicated effort here, we can unify this hash and the ppi hash later
74  */
75 struct process_pid_map {
76 	pid_t pid;
77 	char comm[16];
78 	struct process_pid_map *hash_next, *list_next;
79 };
80 
81 #define PPM_HASH_SHIFT	(8)
82 #define PPM_HASH_SIZE	(1 << PPM_HASH_SHIFT)
83 #define PPM_HASH_MASK	(PPM_HASH_SIZE - 1)
84 static struct process_pid_map *ppm_hash_table[PPM_HASH_SIZE];
85 
86 struct per_process_info {
87 	struct process_pid_map *ppm;
88 	struct io_stats io_stats;
89 	struct per_process_info *hash_next, *list_next;
90 	int more_than_one;
91 
92 	/*
93 	 * individual io stats
94 	 */
95 	unsigned long long longest_allocation_wait[2];
96 	unsigned long long longest_dispatch_wait[2];
97 	unsigned long long longest_completion_wait[2];
98 };
99 
100 #define PPI_HASH_SHIFT	(8)
101 #define PPI_HASH_SIZE	(1 << PPI_HASH_SHIFT)
102 #define PPI_HASH_MASK	(PPI_HASH_SIZE - 1)
103 static struct per_process_info *ppi_hash_table[PPI_HASH_SIZE];
104 static struct per_process_info *ppi_list;
105 static int ppi_list_entries;
106 
107 static struct option l_opts[] = {
108  	{
109 		.name = "act-mask",
110 		.has_arg = required_argument,
111 		.flag = NULL,
112 		.val = 'a'
113 	},
114 	{
115 		.name = "set-mask",
116 		.has_arg = required_argument,
117 		.flag = NULL,
118 		.val = 'A'
119 	},
120 	{
121 		.name = "batch",
122 		.has_arg = required_argument,
123 		.flag = NULL,
124 		.val = 'b'
125 	},
126 	{
127 		.name = "input-directory",
128 		.has_arg = required_argument,
129 		.flag = NULL,
130 		.val = 'D'
131 	},
132 	{
133 		.name = "dump-binary",
134 		.has_arg = required_argument,
135 		.flag = NULL,
136 		.val = 'd'
137 	},
138 	{
139 		.name = "format",
140 		.has_arg = required_argument,
141 		.flag = NULL,
142 		.val = 'f'
143 	},
144 	{
145 		.name = "format-spec",
146 		.has_arg = required_argument,
147 		.flag = NULL,
148 		.val = 'F'
149 	},
150 	{
151 		.name = "hash-by-name",
152 		.has_arg = no_argument,
153 		.flag = NULL,
154 		.val = 'h'
155 	},
156 	{
157 		.name = "input",
158 		.has_arg = required_argument,
159 		.flag = NULL,
160 		.val = 'i'
161 	},
162 	{
163 		.name = "no-msgs",
164 		.has_arg = no_argument,
165 		.flag = NULL,
166 		.val = 'M'
167 	},
168 	{
169 		.name = "output",
170 		.has_arg = required_argument,
171 		.flag = NULL,
172 		.val = 'o'
173 	},
174 	{
175 		.name = "no-text-output",
176 		.has_arg = no_argument,
177 		.flag = NULL,
178 		.val = 'O'
179 	},
180 	{
181 		.name = "quiet",
182 		.has_arg = no_argument,
183 		.flag = NULL,
184 		.val = 'q'
185 	},
186 	{
187 		.name = "per-program-stats",
188 		.has_arg = no_argument,
189 		.flag = NULL,
190 		.val = 's'
191 	},
192 	{
193 		.name = "track-ios",
194 		.has_arg = no_argument,
195 		.flag = NULL,
196 		.val = 't'
197 	},
198 	{
199 		.name = "stopwatch",
200 		.has_arg = required_argument,
201 		.flag = NULL,
202 		.val = 'w'
203 	},
204 	{
205 		.name = "verbose",
206 		.has_arg = no_argument,
207 		.flag = NULL,
208 		.val = 'v'
209 	},
210 	{
211 		.name = "version",
212 		.has_arg = no_argument,
213 		.flag = NULL,
214 		.val = 'V'
215 	},
216 	{
217 		.name = NULL,
218 	}
219 };
220 
221 /*
222  * for sorting the displayed output
223  */
224 struct trace {
225 	struct blk_io_trace *bit;
226 	struct rb_node rb_node;
227 	struct trace *next;
228 	unsigned long read_sequence;
229 };
230 
231 static struct rb_root rb_sort_root;
232 static unsigned long rb_sort_entries;
233 
234 static struct trace *trace_list;
235 
236 /*
237  * allocation cache
238  */
239 static struct blk_io_trace *bit_alloc_list;
240 static struct trace *t_alloc_list;
241 
242 /*
243  * for tracking individual ios
244  */
245 struct io_track {
246 	struct rb_node rb_node;
247 
248 	struct process_pid_map *ppm;
249 	__u64 sector;
250 	unsigned long long allocation_time;
251 	unsigned long long queue_time;
252 	unsigned long long dispatch_time;
253 	unsigned long long completion_time;
254 };
255 
256 static int ndevices;
257 static struct per_dev_info *devices;
258 static char *get_dev_name(struct per_dev_info *, char *, int);
259 static int trace_rb_insert_last(struct per_dev_info *, struct trace *);
260 
261 FILE *ofp = NULL;
262 static char *output_name;
263 static char *input_dir;
264 
265 static unsigned long long genesis_time;
266 static unsigned long long last_allowed_time;
267 static unsigned long long stopwatch_start;	/* start from zero by default */
268 static unsigned long long stopwatch_end = -1ULL;	/* "infinity" */
269 static unsigned long read_sequence;
270 
271 static int per_process_stats;
272 static int per_device_and_cpu_stats = 1;
273 static int track_ios;
274 static int ppi_hash_by_pid = 1;
275 static int verbose;
276 static unsigned int act_mask = -1U;
277 static int stats_printed;
278 static int bin_output_msgs = 1;
279 int data_is_native = -1;
280 
281 static FILE *dump_fp;
282 static char *dump_binary;
283 
284 static unsigned int t_alloc_cache;
285 static unsigned int bit_alloc_cache;
286 
287 #define RB_BATCH_DEFAULT	(512)
288 static unsigned int rb_batch = RB_BATCH_DEFAULT;
289 
290 static int pipeline;
291 static char *pipename;
292 
293 static int text_output = 1;
294 
295 #define is_done()	(*(volatile int *)(&done))
296 static volatile int done;
297 
298 struct timespec		abs_start_time;
299 static unsigned long long start_timestamp;
300 
301 static int have_drv_data = 0;
302 
303 #define JHASH_RANDOM	(0x3af5f2ee)
304 
305 #define CPUS_PER_LONG	(8 * sizeof(unsigned long))
306 #define CPU_IDX(cpu)	((cpu) / CPUS_PER_LONG)
307 #define CPU_BIT(cpu)	((cpu) & (CPUS_PER_LONG - 1))
308 
output_binary(void * buf,int len)309 static void output_binary(void *buf, int len)
310 {
311 	if (dump_binary) {
312 		size_t n = fwrite(buf, len, 1, dump_fp);
313 		if (n != 1) {
314 			perror(dump_binary);
315 			fclose(dump_fp);
316 			dump_binary = NULL;
317 		}
318 	}
319 }
320 
resize_cpu_info(struct per_dev_info * pdi,int cpu)321 static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
322 {
323 	struct per_cpu_info *cpus = pdi->cpus;
324 	int ncpus = pdi->ncpus;
325 	int new_count = cpu + 1;
326 	int new_space, size;
327 	char *new_start;
328 
329 	size = new_count * sizeof(struct per_cpu_info);
330 	cpus = realloc(cpus, size);
331 	if (!cpus) {
332 		char name[20];
333 		fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
334 			get_dev_name(pdi, name, sizeof(name)), size);
335 		exit(1);
336 	}
337 
338 	new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
339 	new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
340 	memset(new_start, 0, new_space);
341 
342 	pdi->ncpus = new_count;
343 	pdi->cpus = cpus;
344 
345 	for (new_count = 0; new_count < pdi->ncpus; new_count++) {
346 		struct per_cpu_info *pci = &pdi->cpus[new_count];
347 
348 		if (!pci->fd) {
349 			pci->fd = -1;
350 			memset(&pci->rb_last, 0, sizeof(pci->rb_last));
351 			pci->rb_last_entries = 0;
352 			pci->last_sequence = -1;
353 		}
354 	}
355 }
356 
get_cpu_info(struct per_dev_info * pdi,int cpu)357 static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
358 {
359 	struct per_cpu_info *pci;
360 
361 	if (cpu >= pdi->ncpus)
362 		resize_cpu_info(pdi, cpu);
363 
364 	pci = &pdi->cpus[cpu];
365 	pci->cpu = cpu;
366 	return pci;
367 }
368 
369 
resize_devices(char * name)370 static int resize_devices(char *name)
371 {
372 	int size = (ndevices + 1) * sizeof(struct per_dev_info);
373 
374 	devices = realloc(devices, size);
375 	if (!devices) {
376 		fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
377 		return 1;
378 	}
379 	memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
380 	devices[ndevices].name = name;
381 	ndevices++;
382 	return 0;
383 }
384 
get_dev_info(dev_t dev)385 static struct per_dev_info *get_dev_info(dev_t dev)
386 {
387 	struct per_dev_info *pdi;
388 	int i;
389 
390 	for (i = 0; i < ndevices; i++) {
391 		if (!devices[i].dev)
392 			devices[i].dev = dev;
393 		if (devices[i].dev == dev)
394 			return &devices[i];
395 	}
396 
397 	if (resize_devices(NULL))
398 		return NULL;
399 
400 	pdi = &devices[ndevices - 1];
401 	pdi->dev = dev;
402 	pdi->first_reported_time = 0;
403 	pdi->last_read_time = 0;
404 
405 	return pdi;
406 }
407 
insert_skip(struct per_cpu_info * pci,unsigned long start,unsigned long end)408 static void insert_skip(struct per_cpu_info *pci, unsigned long start,
409 			unsigned long end)
410 {
411 	struct skip_info *sip;
412 
413 	for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) {
414 		if (end == (sip->start - 1)) {
415 			sip->start = start;
416 			return;
417 		} else if (start == (sip->end + 1)) {
418 			sip->end = end;
419 			return;
420 		}
421 	}
422 
423 	sip = malloc(sizeof(struct skip_info));
424 	sip->start = start;
425 	sip->end = end;
426 	sip->prev = sip->next = NULL;
427 	if (pci->skips_tail == NULL)
428 		pci->skips_head = pci->skips_tail = sip;
429 	else {
430 		sip->prev = pci->skips_tail;
431 		pci->skips_tail->next = sip;
432 		pci->skips_tail = sip;
433 	}
434 }
435 
remove_sip(struct per_cpu_info * pci,struct skip_info * sip)436 static void remove_sip(struct per_cpu_info *pci, struct skip_info *sip)
437 {
438 	if (sip->prev == NULL) {
439 		if (sip->next == NULL)
440 			pci->skips_head = pci->skips_tail = NULL;
441 		else {
442 			pci->skips_head = sip->next;
443 			sip->next->prev = NULL;
444 		}
445 	} else if (sip->next == NULL) {
446 		pci->skips_tail = sip->prev;
447 		sip->prev->next = NULL;
448 	} else {
449 		sip->prev->next = sip->next;
450 		sip->next->prev = sip->prev;
451 	}
452 
453 	sip->prev = sip->next = NULL;
454 	free(sip);
455 }
456 
457 #define IN_SKIP(sip,seq) (((sip)->start <= (seq)) && ((seq) <= sip->end))
check_current_skips(struct per_cpu_info * pci,unsigned long seq)458 static int check_current_skips(struct per_cpu_info *pci, unsigned long seq)
459 {
460 	struct skip_info *sip;
461 
462 	for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) {
463 		if (IN_SKIP(sip, seq)) {
464 			if (sip->start == seq) {
465 				if (sip->end == seq)
466 					remove_sip(pci, sip);
467 				else
468 					sip->start += 1;
469 			} else if (sip->end == seq)
470 				sip->end -= 1;
471 			else {
472 				sip->end = seq - 1;
473 				insert_skip(pci, seq + 1, sip->end);
474 			}
475 			return 1;
476 		}
477 	}
478 
479 	return 0;
480 }
481 
collect_pdi_skips(struct per_dev_info * pdi)482 static void collect_pdi_skips(struct per_dev_info *pdi)
483 {
484 	struct skip_info *sip;
485 	int cpu;
486 
487 	pdi->skips = 0;
488 	pdi->seq_skips = 0;
489 
490 	for (cpu = 0; cpu < pdi->ncpus; cpu++) {
491 		struct per_cpu_info *pci = &pdi->cpus[cpu];
492 
493 		for (sip = pci->skips_head; sip != NULL; sip = sip->next) {
494 			pdi->skips++;
495 			pdi->seq_skips += (sip->end - sip->start + 1);
496 			if (verbose)
497 				fprintf(stderr,"(%d,%d): skipping %lu -> %lu\n",
498 					MAJOR(pdi->dev), MINOR(pdi->dev),
499 					sip->start, sip->end);
500 		}
501 	}
502 }
503 
cpu_mark_online(struct per_dev_info * pdi,unsigned int cpu)504 static void cpu_mark_online(struct per_dev_info *pdi, unsigned int cpu)
505 {
506 	if (cpu >= pdi->cpu_map_max || !pdi->cpu_map) {
507 		int new_max = (cpu + CPUS_PER_LONG) & ~(CPUS_PER_LONG - 1);
508 		unsigned long *map = malloc(new_max / sizeof(long));
509 
510 		memset(map, 0, new_max / sizeof(long));
511 
512 		if (pdi->cpu_map) {
513 			memcpy(map, pdi->cpu_map, pdi->cpu_map_max / sizeof(long));
514 			free(pdi->cpu_map);
515 		}
516 
517 		pdi->cpu_map = map;
518 		pdi->cpu_map_max = new_max;
519 	}
520 
521 	pdi->cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu));
522 }
523 
cpu_mark_offline(struct per_dev_info * pdi,int cpu)524 static inline void cpu_mark_offline(struct per_dev_info *pdi, int cpu)
525 {
526 	pdi->cpu_map[CPU_IDX(cpu)] &= ~(1UL << CPU_BIT(cpu));
527 }
528 
cpu_is_online(struct per_dev_info * pdi,int cpu)529 static inline int cpu_is_online(struct per_dev_info *pdi, int cpu)
530 {
531 	return (pdi->cpu_map[CPU_IDX(cpu)] & (1UL << CPU_BIT(cpu))) != 0;
532 }
533 
ppm_hash_pid(pid_t pid)534 static inline int ppm_hash_pid(pid_t pid)
535 {
536 	return jhash_1word(pid, JHASH_RANDOM) & PPM_HASH_MASK;
537 }
538 
find_ppm(pid_t pid)539 static struct process_pid_map *find_ppm(pid_t pid)
540 {
541 	const int hash_idx = ppm_hash_pid(pid);
542 	struct process_pid_map *ppm;
543 
544 	ppm = ppm_hash_table[hash_idx];
545 	while (ppm) {
546 		if (ppm->pid == pid)
547 			return ppm;
548 
549 		ppm = ppm->hash_next;
550 	}
551 
552 	return NULL;
553 }
554 
add_ppm_hash(pid_t pid,const char * name)555 static struct process_pid_map *add_ppm_hash(pid_t pid, const char *name)
556 {
557 	const int hash_idx = ppm_hash_pid(pid);
558 	struct process_pid_map *ppm;
559 
560 	ppm = find_ppm(pid);
561 	if (!ppm) {
562 		ppm = malloc(sizeof(*ppm));
563 		memset(ppm, 0, sizeof(*ppm));
564 		ppm->pid = pid;
565 		strcpy(ppm->comm, name);
566 		ppm->hash_next = ppm_hash_table[hash_idx];
567 		ppm_hash_table[hash_idx] = ppm;
568 	}
569 
570 	return ppm;
571 }
572 
handle_notify(struct blk_io_trace * bit)573 static void handle_notify(struct blk_io_trace *bit)
574 {
575 	void	*payload = (caddr_t) bit + sizeof(*bit);
576 	__u32	two32[2];
577 
578 	switch (bit->action) {
579 	case BLK_TN_PROCESS:
580 		add_ppm_hash(bit->pid, payload);
581 		break;
582 
583 	case BLK_TN_TIMESTAMP:
584 		if (bit->pdu_len != sizeof(two32))
585 			return;
586 		memcpy(two32, payload, sizeof(two32));
587 		if (!data_is_native) {
588 			two32[0] = be32_to_cpu(two32[0]);
589 			two32[1] = be32_to_cpu(two32[1]);
590 		}
591 		start_timestamp = bit->time;
592 		abs_start_time.tv_sec  = two32[0];
593 		abs_start_time.tv_nsec = two32[1];
594 		if (abs_start_time.tv_nsec < 0) {
595 			abs_start_time.tv_sec--;
596 			abs_start_time.tv_nsec += 1000000000;
597 		}
598 
599 		break;
600 
601 	case BLK_TN_MESSAGE:
602 		if (bit->pdu_len > 0) {
603 			char msg[bit->pdu_len+1];
604 
605 			memcpy(msg, (char *)payload, bit->pdu_len);
606 			msg[bit->pdu_len] = '\0';
607 
608 			fprintf(ofp,
609 				"%3d,%-3d %2d %8s %5d.%09lu %5u %2s %3s %s\n",
610 				MAJOR(bit->device), MINOR(bit->device),
611 				bit->cpu, "0", (int) SECONDS(bit->time),
612 				(unsigned long) NANO_SECONDS(bit->time),
613 				0, "m", "N", msg);
614 		}
615 		break;
616 
617 	default:
618 		/* Ignore unknown notify events */
619 		;
620 	}
621 }
622 
find_process_name(pid_t pid)623 char *find_process_name(pid_t pid)
624 {
625 	struct process_pid_map *ppm = find_ppm(pid);
626 
627 	if (ppm)
628 		return ppm->comm;
629 
630 	return NULL;
631 }
632 
ppi_hash_pid(pid_t pid)633 static inline int ppi_hash_pid(pid_t pid)
634 {
635 	return jhash_1word(pid, JHASH_RANDOM) & PPI_HASH_MASK;
636 }
637 
ppi_hash_name(const char * name)638 static inline int ppi_hash_name(const char *name)
639 {
640 	return jhash(name, 16, JHASH_RANDOM) & PPI_HASH_MASK;
641 }
642 
ppi_hash(struct per_process_info * ppi)643 static inline int ppi_hash(struct per_process_info *ppi)
644 {
645 	struct process_pid_map *ppm = ppi->ppm;
646 
647 	if (ppi_hash_by_pid)
648 		return ppi_hash_pid(ppm->pid);
649 
650 	return ppi_hash_name(ppm->comm);
651 }
652 
add_ppi_to_hash(struct per_process_info * ppi)653 static inline void add_ppi_to_hash(struct per_process_info *ppi)
654 {
655 	const int hash_idx = ppi_hash(ppi);
656 
657 	ppi->hash_next = ppi_hash_table[hash_idx];
658 	ppi_hash_table[hash_idx] = ppi;
659 }
660 
add_ppi_to_list(struct per_process_info * ppi)661 static inline void add_ppi_to_list(struct per_process_info *ppi)
662 {
663 	ppi->list_next = ppi_list;
664 	ppi_list = ppi;
665 	ppi_list_entries++;
666 }
667 
find_ppi_by_name(char * name)668 static struct per_process_info *find_ppi_by_name(char *name)
669 {
670 	const int hash_idx = ppi_hash_name(name);
671 	struct per_process_info *ppi;
672 
673 	ppi = ppi_hash_table[hash_idx];
674 	while (ppi) {
675 		struct process_pid_map *ppm = ppi->ppm;
676 
677 		if (!strcmp(ppm->comm, name))
678 			return ppi;
679 
680 		ppi = ppi->hash_next;
681 	}
682 
683 	return NULL;
684 }
685 
find_ppi_by_pid(pid_t pid)686 static struct per_process_info *find_ppi_by_pid(pid_t pid)
687 {
688 	const int hash_idx = ppi_hash_pid(pid);
689 	struct per_process_info *ppi;
690 
691 	ppi = ppi_hash_table[hash_idx];
692 	while (ppi) {
693 		struct process_pid_map *ppm = ppi->ppm;
694 
695 		if (ppm->pid == pid)
696 			return ppi;
697 
698 		ppi = ppi->hash_next;
699 	}
700 
701 	return NULL;
702 }
703 
find_ppi(pid_t pid)704 static struct per_process_info *find_ppi(pid_t pid)
705 {
706 	struct per_process_info *ppi;
707 	char *name;
708 
709 	if (ppi_hash_by_pid)
710 		return find_ppi_by_pid(pid);
711 
712 	name = find_process_name(pid);
713 	if (!name)
714 		return NULL;
715 
716 	ppi = find_ppi_by_name(name);
717 	if (ppi && ppi->ppm->pid != pid)
718 		ppi->more_than_one = 1;
719 
720 	return ppi;
721 }
722 
723 /*
724  * struct trace and blktrace allocation cache, we do potentially
725  * millions of mallocs for these structures while only using at most
726  * a few thousand at the time
727  */
t_free(struct trace * t)728 static inline void t_free(struct trace *t)
729 {
730 	if (t_alloc_cache < 1024) {
731 		t->next = t_alloc_list;
732 		t_alloc_list = t;
733 		t_alloc_cache++;
734 	} else
735 		free(t);
736 }
737 
t_alloc(void)738 static inline struct trace *t_alloc(void)
739 {
740 	struct trace *t = t_alloc_list;
741 
742 	if (t) {
743 		t_alloc_list = t->next;
744 		t_alloc_cache--;
745 		return t;
746 	}
747 
748 	return malloc(sizeof(*t));
749 }
750 
bit_free(struct blk_io_trace * bit)751 static inline void bit_free(struct blk_io_trace *bit)
752 {
753 	if (bit_alloc_cache < 1024 && !bit->pdu_len) {
754 		/*
755 		 * abuse a 64-bit field for a next pointer for the free item
756 		 */
757 		bit->time = (__u64) (unsigned long) bit_alloc_list;
758 		bit_alloc_list = (struct blk_io_trace *) bit;
759 		bit_alloc_cache++;
760 	} else
761 		free(bit);
762 }
763 
bit_alloc(void)764 static inline struct blk_io_trace *bit_alloc(void)
765 {
766 	struct blk_io_trace *bit = bit_alloc_list;
767 
768 	if (bit) {
769 		bit_alloc_list = (struct blk_io_trace *) (unsigned long) \
770 				 bit->time;
771 		bit_alloc_cache--;
772 		return bit;
773 	}
774 
775 	return malloc(sizeof(*bit));
776 }
777 
__put_trace_last(struct per_dev_info * pdi,struct trace * t)778 static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t)
779 {
780 	struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu);
781 
782 	rb_erase(&t->rb_node, &pci->rb_last);
783 	pci->rb_last_entries--;
784 
785 	bit_free(t->bit);
786 	t_free(t);
787 }
788 
put_trace(struct per_dev_info * pdi,struct trace * t)789 static void put_trace(struct per_dev_info *pdi, struct trace *t)
790 {
791 	rb_erase(&t->rb_node, &rb_sort_root);
792 	rb_sort_entries--;
793 
794 	trace_rb_insert_last(pdi, t);
795 }
796 
trace_rb_insert(struct trace * t,struct rb_root * root)797 static inline int trace_rb_insert(struct trace *t, struct rb_root *root)
798 {
799 	struct rb_node **p = &root->rb_node;
800 	struct rb_node *parent = NULL;
801 	struct trace *__t;
802 
803 	while (*p) {
804 		parent = *p;
805 
806 		__t = rb_entry(parent, struct trace, rb_node);
807 
808 		if (t->bit->time < __t->bit->time)
809 			p = &(*p)->rb_left;
810 		else if (t->bit->time > __t->bit->time)
811 			p = &(*p)->rb_right;
812 		else if (t->bit->device < __t->bit->device)
813 			p = &(*p)->rb_left;
814 		else if (t->bit->device > __t->bit->device)
815 			p = &(*p)->rb_right;
816 		else if (t->bit->sequence < __t->bit->sequence)
817 			p = &(*p)->rb_left;
818 		else	/* >= sequence */
819 			p = &(*p)->rb_right;
820 	}
821 
822 	rb_link_node(&t->rb_node, parent, p);
823 	rb_insert_color(&t->rb_node, root);
824 	return 0;
825 }
826 
trace_rb_insert_sort(struct trace * t)827 static inline int trace_rb_insert_sort(struct trace *t)
828 {
829 	if (!trace_rb_insert(t, &rb_sort_root)) {
830 		rb_sort_entries++;
831 		return 0;
832 	}
833 
834 	return 1;
835 }
836 
trace_rb_insert_last(struct per_dev_info * pdi,struct trace * t)837 static int trace_rb_insert_last(struct per_dev_info *pdi, struct trace *t)
838 {
839 	struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu);
840 
841 	if (trace_rb_insert(t, &pci->rb_last))
842 		return 1;
843 
844 	pci->rb_last_entries++;
845 
846 	if (pci->rb_last_entries > rb_batch * pdi->nfiles) {
847 		struct rb_node *n = rb_first(&pci->rb_last);
848 
849 		t = rb_entry(n, struct trace, rb_node);
850 		__put_trace_last(pdi, t);
851 	}
852 
853 	return 0;
854 }
855 
trace_rb_find(dev_t device,unsigned long sequence,struct rb_root * root,int order)856 static struct trace *trace_rb_find(dev_t device, unsigned long sequence,
857 				   struct rb_root *root, int order)
858 {
859 	struct rb_node *n = root->rb_node;
860 	struct rb_node *prev = NULL;
861 	struct trace *__t;
862 
863 	while (n) {
864 		__t = rb_entry(n, struct trace, rb_node);
865 		prev = n;
866 
867 		if (device < __t->bit->device)
868 			n = n->rb_left;
869 		else if (device > __t->bit->device)
870 			n = n->rb_right;
871 		else if (sequence < __t->bit->sequence)
872 			n = n->rb_left;
873 		else if (sequence > __t->bit->sequence)
874 			n = n->rb_right;
875 		else
876 			return __t;
877 	}
878 
879 	/*
880 	 * hack - the list may not be sequence ordered because some
881 	 * events don't have sequence and time matched. so we end up
882 	 * being a little off in the rb lookup here, because we don't
883 	 * know the time we are looking for. compensate by browsing
884 	 * a little ahead from the last entry to find the match
885 	 */
886 	if (order && prev) {
887 		int max = 5;
888 
889 		while (((n = rb_next(prev)) != NULL) && max--) {
890 			__t = rb_entry(n, struct trace, rb_node);
891 
892 			if (__t->bit->device == device &&
893 			    __t->bit->sequence == sequence)
894 				return __t;
895 
896 			prev = n;
897 		}
898 	}
899 
900 	return NULL;
901 }
902 
trace_rb_find_last(struct per_dev_info * pdi,struct per_cpu_info * pci,unsigned long seq)903 static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi,
904 					       struct per_cpu_info *pci,
905 					       unsigned long seq)
906 {
907 	return trace_rb_find(pdi->dev, seq, &pci->rb_last, 0);
908 }
909 
track_rb_insert(struct per_dev_info * pdi,struct io_track * iot)910 static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot)
911 {
912 	struct rb_node **p = &pdi->rb_track.rb_node;
913 	struct rb_node *parent = NULL;
914 	struct io_track *__iot;
915 
916 	while (*p) {
917 		parent = *p;
918 		__iot = rb_entry(parent, struct io_track, rb_node);
919 
920 		if (iot->sector < __iot->sector)
921 			p = &(*p)->rb_left;
922 		else if (iot->sector > __iot->sector)
923 			p = &(*p)->rb_right;
924 		else {
925 			fprintf(stderr,
926 				"sector alias (%Lu) on device %d,%d!\n",
927 				(unsigned long long) iot->sector,
928 				MAJOR(pdi->dev), MINOR(pdi->dev));
929 			return 1;
930 		}
931 	}
932 
933 	rb_link_node(&iot->rb_node, parent, p);
934 	rb_insert_color(&iot->rb_node, &pdi->rb_track);
935 	return 0;
936 }
937 
__find_track(struct per_dev_info * pdi,__u64 sector)938 static struct io_track *__find_track(struct per_dev_info *pdi, __u64 sector)
939 {
940 	struct rb_node *n = pdi->rb_track.rb_node;
941 	struct io_track *__iot;
942 
943 	while (n) {
944 		__iot = rb_entry(n, struct io_track, rb_node);
945 
946 		if (sector < __iot->sector)
947 			n = n->rb_left;
948 		else if (sector > __iot->sector)
949 			n = n->rb_right;
950 		else
951 			return __iot;
952 	}
953 
954 	return NULL;
955 }
956 
find_track(struct per_dev_info * pdi,pid_t pid,__u64 sector)957 static struct io_track *find_track(struct per_dev_info *pdi, pid_t pid,
958 				   __u64 sector)
959 {
960 	struct io_track *iot;
961 
962 	iot = __find_track(pdi, sector);
963 	if (!iot) {
964 		iot = malloc(sizeof(*iot));
965 		iot->ppm = find_ppm(pid);
966 		if (!iot->ppm)
967 			iot->ppm = add_ppm_hash(pid, "unknown");
968 		iot->sector = sector;
969 		track_rb_insert(pdi, iot);
970 	}
971 
972 	return iot;
973 }
974 
log_track_frontmerge(struct per_dev_info * pdi,struct blk_io_trace * t)975 static void log_track_frontmerge(struct per_dev_info *pdi,
976 				 struct blk_io_trace *t)
977 {
978 	struct io_track *iot;
979 
980 	if (!track_ios)
981 		return;
982 
983 	iot = __find_track(pdi, t->sector + t_sec(t));
984 	if (!iot) {
985 		if (verbose)
986 			fprintf(stderr, "merge not found for (%d,%d): %llu\n",
987 				MAJOR(pdi->dev), MINOR(pdi->dev),
988 				(unsigned long long) t->sector + t_sec(t));
989 		return;
990 	}
991 
992 	rb_erase(&iot->rb_node, &pdi->rb_track);
993 	iot->sector -= t_sec(t);
994 	track_rb_insert(pdi, iot);
995 }
996 
log_track_getrq(struct per_dev_info * pdi,struct blk_io_trace * t)997 static void log_track_getrq(struct per_dev_info *pdi, struct blk_io_trace *t)
998 {
999 	struct io_track *iot;
1000 
1001 	if (!track_ios)
1002 		return;
1003 
1004 	iot = find_track(pdi, t->pid, t->sector);
1005 	iot->allocation_time = t->time;
1006 }
1007 
is_remapper(struct per_dev_info * pdi)1008 static inline int is_remapper(struct per_dev_info *pdi)
1009 {
1010 	int major = MAJOR(pdi->dev);
1011 
1012 	return (major == 253 || major == 9);
1013 }
1014 
1015 /*
1016  * for md/dm setups, the interesting cycle is Q -> C. So track queueing
1017  * time here, as dispatch time
1018  */
log_track_queue(struct per_dev_info * pdi,struct blk_io_trace * t)1019 static void log_track_queue(struct per_dev_info *pdi, struct blk_io_trace *t)
1020 {
1021 	struct io_track *iot;
1022 
1023 	if (!track_ios)
1024 		return;
1025 	if (!is_remapper(pdi))
1026 		return;
1027 
1028 	iot = find_track(pdi, t->pid, t->sector);
1029 	iot->dispatch_time = t->time;
1030 }
1031 
1032 /*
1033  * return time between rq allocation and insertion
1034  */
log_track_insert(struct per_dev_info * pdi,struct blk_io_trace * t)1035 static unsigned long long log_track_insert(struct per_dev_info *pdi,
1036 					   struct blk_io_trace *t)
1037 {
1038 	unsigned long long elapsed;
1039 	struct io_track *iot;
1040 
1041 	if (!track_ios)
1042 		return -1;
1043 
1044 	iot = find_track(pdi, t->pid, t->sector);
1045 	iot->queue_time = t->time;
1046 
1047 	if (!iot->allocation_time)
1048 		return -1;
1049 
1050 	elapsed = iot->queue_time - iot->allocation_time;
1051 
1052 	if (per_process_stats) {
1053 		struct per_process_info *ppi = find_ppi(iot->ppm->pid);
1054 		int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1055 
1056 		if (ppi && elapsed > ppi->longest_allocation_wait[w])
1057 			ppi->longest_allocation_wait[w] = elapsed;
1058 	}
1059 
1060 	return elapsed;
1061 }
1062 
1063 /*
1064  * return time between queue and issue
1065  */
log_track_issue(struct per_dev_info * pdi,struct blk_io_trace * t)1066 static unsigned long long log_track_issue(struct per_dev_info *pdi,
1067 					  struct blk_io_trace *t)
1068 {
1069 	unsigned long long elapsed;
1070 	struct io_track *iot;
1071 
1072 	if (!track_ios)
1073 		return -1;
1074 	if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
1075 		return -1;
1076 
1077 	iot = __find_track(pdi, t->sector);
1078 	if (!iot) {
1079 		if (verbose)
1080 			fprintf(stderr, "issue not found for (%d,%d): %llu\n",
1081 				MAJOR(pdi->dev), MINOR(pdi->dev),
1082 				(unsigned long long) t->sector);
1083 		return -1;
1084 	}
1085 
1086 	iot->dispatch_time = t->time;
1087 	elapsed = iot->dispatch_time - iot->queue_time;
1088 
1089 	if (per_process_stats) {
1090 		struct per_process_info *ppi = find_ppi(iot->ppm->pid);
1091 		int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1092 
1093 		if (ppi && elapsed > ppi->longest_dispatch_wait[w])
1094 			ppi->longest_dispatch_wait[w] = elapsed;
1095 	}
1096 
1097 	return elapsed;
1098 }
1099 
1100 /*
1101  * return time between dispatch and complete
1102  */
log_track_complete(struct per_dev_info * pdi,struct blk_io_trace * t)1103 static unsigned long long log_track_complete(struct per_dev_info *pdi,
1104 					     struct blk_io_trace *t)
1105 {
1106 	unsigned long long elapsed;
1107 	struct io_track *iot;
1108 
1109 	if (!track_ios)
1110 		return -1;
1111 
1112 	iot = __find_track(pdi, t->sector);
1113 	if (!iot) {
1114 		if (verbose)
1115 			fprintf(stderr,"complete not found for (%d,%d): %llu\n",
1116 				MAJOR(pdi->dev), MINOR(pdi->dev),
1117 				(unsigned long long) t->sector);
1118 		return -1;
1119 	}
1120 
1121 	iot->completion_time = t->time;
1122 	elapsed = iot->completion_time - iot->dispatch_time;
1123 
1124 	if (per_process_stats) {
1125 		struct per_process_info *ppi = find_ppi(iot->ppm->pid);
1126 		int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1127 
1128 		if (ppi && elapsed > ppi->longest_completion_wait[w])
1129 			ppi->longest_completion_wait[w] = elapsed;
1130 	}
1131 
1132 	/*
1133 	 * kill the trace, we don't need it after completion
1134 	 */
1135 	rb_erase(&iot->rb_node, &pdi->rb_track);
1136 	free(iot);
1137 
1138 	return elapsed;
1139 }
1140 
1141 
find_process_io_stats(pid_t pid)1142 static struct io_stats *find_process_io_stats(pid_t pid)
1143 {
1144 	struct per_process_info *ppi = find_ppi(pid);
1145 
1146 	if (!ppi) {
1147 		ppi = malloc(sizeof(*ppi));
1148 		memset(ppi, 0, sizeof(*ppi));
1149 		ppi->ppm = find_ppm(pid);
1150 		if (!ppi->ppm)
1151 			ppi->ppm = add_ppm_hash(pid, "unknown");
1152 		add_ppi_to_hash(ppi);
1153 		add_ppi_to_list(ppi);
1154 	}
1155 
1156 	return &ppi->io_stats;
1157 }
1158 
get_dev_name(struct per_dev_info * pdi,char * buffer,int size)1159 static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size)
1160 {
1161 	if (pdi->name)
1162 		snprintf(buffer, size, "%s", pdi->name);
1163 	else
1164 		snprintf(buffer, size, "%d,%d",MAJOR(pdi->dev),MINOR(pdi->dev));
1165 	return buffer;
1166 }
1167 
check_time(struct per_dev_info * pdi,struct blk_io_trace * bit)1168 static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit)
1169 {
1170 	unsigned long long this = bit->time;
1171 	unsigned long long last = pdi->last_reported_time;
1172 
1173 	pdi->backwards = (this < last) ? 'B' : ' ';
1174 	pdi->last_reported_time = this;
1175 }
1176 
__account_m(struct io_stats * ios,struct blk_io_trace * t,int rw)1177 static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t,
1178 			       int rw)
1179 {
1180 	if (rw) {
1181 		ios->mwrites++;
1182 		ios->mwrite_kb += t_kb(t);
1183 	} else {
1184 		ios->mreads++;
1185 		ios->mread_kb += t_kb(t);
1186 	}
1187 }
1188 
account_m(struct blk_io_trace * t,struct per_cpu_info * pci,int rw)1189 static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci,
1190 			     int rw)
1191 {
1192 	__account_m(&pci->io_stats, t, rw);
1193 
1194 	if (per_process_stats) {
1195 		struct io_stats *ios = find_process_io_stats(t->pid);
1196 
1197 		__account_m(ios, t, rw);
1198 	}
1199 }
1200 
__account_pc_queue(struct io_stats * ios,struct blk_io_trace * t,int rw)1201 static inline void __account_pc_queue(struct io_stats *ios,
1202 				      struct blk_io_trace *t, int rw)
1203 {
1204 	if (rw) {
1205 		ios->qwrites_pc++;
1206 		ios->qwrite_kb_pc += t_kb(t);
1207 	} else {
1208 		ios->qreads_pc++;
1209 		ios->qread_kb += t_kb(t);
1210 	}
1211 }
1212 
account_pc_queue(struct blk_io_trace * t,struct per_cpu_info * pci,int rw)1213 static inline void account_pc_queue(struct blk_io_trace *t,
1214 				    struct per_cpu_info *pci, int rw)
1215 {
1216 	__account_pc_queue(&pci->io_stats, t, rw);
1217 
1218 	if (per_process_stats) {
1219 		struct io_stats *ios = find_process_io_stats(t->pid);
1220 
1221 		__account_pc_queue(ios, t, rw);
1222 	}
1223 }
1224 
__account_pc_issue(struct io_stats * ios,int rw,unsigned int bytes)1225 static inline void __account_pc_issue(struct io_stats *ios, int rw,
1226 				      unsigned int bytes)
1227 {
1228 	if (rw) {
1229 		ios->iwrites_pc++;
1230 		ios->iwrite_kb_pc += bytes >> 10;
1231 	} else {
1232 		ios->ireads_pc++;
1233 		ios->iread_kb_pc += bytes >> 10;
1234 	}
1235 }
1236 
account_pc_issue(struct blk_io_trace * t,struct per_cpu_info * pci,int rw)1237 static inline void account_pc_issue(struct blk_io_trace *t,
1238 				    struct per_cpu_info *pci, int rw)
1239 {
1240 	__account_pc_issue(&pci->io_stats, rw, t->bytes);
1241 
1242 	if (per_process_stats) {
1243 		struct io_stats *ios = find_process_io_stats(t->pid);
1244 
1245 		__account_pc_issue(ios, rw, t->bytes);
1246 	}
1247 }
1248 
__account_pc_requeue(struct io_stats * ios,struct blk_io_trace * t,int rw)1249 static inline void __account_pc_requeue(struct io_stats *ios,
1250 					struct blk_io_trace *t, int rw)
1251 {
1252 	if (rw) {
1253 		ios->wrqueue_pc++;
1254 		ios->iwrite_kb_pc -= t_kb(t);
1255 	} else {
1256 		ios->rrqueue_pc++;
1257 		ios->iread_kb_pc -= t_kb(t);
1258 	}
1259 }
1260 
account_pc_requeue(struct blk_io_trace * t,struct per_cpu_info * pci,int rw)1261 static inline void account_pc_requeue(struct blk_io_trace *t,
1262 				      struct per_cpu_info *pci, int rw)
1263 {
1264 	__account_pc_requeue(&pci->io_stats, t, rw);
1265 
1266 	if (per_process_stats) {
1267 		struct io_stats *ios = find_process_io_stats(t->pid);
1268 
1269 		__account_pc_requeue(ios, t, rw);
1270 	}
1271 }
1272 
__account_pc_c(struct io_stats * ios,int rw)1273 static inline void __account_pc_c(struct io_stats *ios, int rw)
1274 {
1275 	if (rw)
1276 		ios->cwrites_pc++;
1277 	else
1278 		ios->creads_pc++;
1279 }
1280 
account_pc_c(struct blk_io_trace * t,struct per_cpu_info * pci,int rw)1281 static inline void account_pc_c(struct blk_io_trace *t,
1282 				struct per_cpu_info *pci, int rw)
1283 {
1284 	__account_pc_c(&pci->io_stats, rw);
1285 
1286 	if (per_process_stats) {
1287 		struct io_stats *ios = find_process_io_stats(t->pid);
1288 
1289 		__account_pc_c(ios, rw);
1290 	}
1291 }
1292 
__account_queue(struct io_stats * ios,struct blk_io_trace * t,int rw)1293 static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t,
1294 				   int rw)
1295 {
1296 	if (rw) {
1297 		ios->qwrites++;
1298 		ios->qwrite_kb += t_kb(t);
1299 	} else {
1300 		ios->qreads++;
1301 		ios->qread_kb += t_kb(t);
1302 	}
1303 }
1304 
account_queue(struct blk_io_trace * t,struct per_cpu_info * pci,int rw)1305 static inline void account_queue(struct blk_io_trace *t,
1306 				 struct per_cpu_info *pci, int rw)
1307 {
1308 	__account_queue(&pci->io_stats, t, rw);
1309 
1310 	if (per_process_stats) {
1311 		struct io_stats *ios = find_process_io_stats(t->pid);
1312 
1313 		__account_queue(ios, t, rw);
1314 	}
1315 }
1316 
__account_c(struct io_stats * ios,int rw,int bytes)1317 static inline void __account_c(struct io_stats *ios, int rw, int bytes)
1318 {
1319 	if (rw) {
1320 		ios->cwrites++;
1321 		ios->cwrite_kb += bytes >> 10;
1322 	} else {
1323 		ios->creads++;
1324 		ios->cread_kb += bytes >> 10;
1325 	}
1326 }
1327 
account_c(struct blk_io_trace * t,struct per_cpu_info * pci,int rw,int bytes)1328 static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci,
1329 			     int rw, int bytes)
1330 {
1331 	__account_c(&pci->io_stats, rw, bytes);
1332 
1333 	if (per_process_stats) {
1334 		struct io_stats *ios = find_process_io_stats(t->pid);
1335 
1336 		__account_c(ios, rw, bytes);
1337 	}
1338 }
1339 
__account_issue(struct io_stats * ios,int rw,unsigned int bytes)1340 static inline void __account_issue(struct io_stats *ios, int rw,
1341 				   unsigned int bytes)
1342 {
1343 	if (rw) {
1344 		ios->iwrites++;
1345 		ios->iwrite_kb += bytes >> 10;
1346 	} else {
1347 		ios->ireads++;
1348 		ios->iread_kb += bytes >> 10;
1349 	}
1350 }
1351 
account_issue(struct blk_io_trace * t,struct per_cpu_info * pci,int rw)1352 static inline void account_issue(struct blk_io_trace *t,
1353 				 struct per_cpu_info *pci, int rw)
1354 {
1355 	__account_issue(&pci->io_stats, rw, t->bytes);
1356 
1357 	if (per_process_stats) {
1358 		struct io_stats *ios = find_process_io_stats(t->pid);
1359 
1360 		__account_issue(ios, rw, t->bytes);
1361 	}
1362 }
1363 
__account_unplug(struct io_stats * ios,int timer)1364 static inline void __account_unplug(struct io_stats *ios, int timer)
1365 {
1366 	if (timer)
1367 		ios->timer_unplugs++;
1368 	else
1369 		ios->io_unplugs++;
1370 }
1371 
account_unplug(struct blk_io_trace * t,struct per_cpu_info * pci,int timer)1372 static inline void account_unplug(struct blk_io_trace *t,
1373 				  struct per_cpu_info *pci, int timer)
1374 {
1375 	__account_unplug(&pci->io_stats, timer);
1376 
1377 	if (per_process_stats) {
1378 		struct io_stats *ios = find_process_io_stats(t->pid);
1379 
1380 		__account_unplug(ios, timer);
1381 	}
1382 }
1383 
__account_requeue(struct io_stats * ios,struct blk_io_trace * t,int rw)1384 static inline void __account_requeue(struct io_stats *ios,
1385 				     struct blk_io_trace *t, int rw)
1386 {
1387 	if (rw) {
1388 		ios->wrqueue++;
1389 		ios->iwrite_kb -= t_kb(t);
1390 	} else {
1391 		ios->rrqueue++;
1392 		ios->iread_kb -= t_kb(t);
1393 	}
1394 }
1395 
account_requeue(struct blk_io_trace * t,struct per_cpu_info * pci,int rw)1396 static inline void account_requeue(struct blk_io_trace *t,
1397 				   struct per_cpu_info *pci, int rw)
1398 {
1399 	__account_requeue(&pci->io_stats, t, rw);
1400 
1401 	if (per_process_stats) {
1402 		struct io_stats *ios = find_process_io_stats(t->pid);
1403 
1404 		__account_requeue(ios, t, rw);
1405 	}
1406 }
1407 
log_complete(struct per_dev_info * pdi,struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1408 static void log_complete(struct per_dev_info *pdi, struct per_cpu_info *pci,
1409 			 struct blk_io_trace *t, char *act)
1410 {
1411 	process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL);
1412 }
1413 
log_insert(struct per_dev_info * pdi,struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1414 static void log_insert(struct per_dev_info *pdi, struct per_cpu_info *pci,
1415 		       struct blk_io_trace *t, char *act)
1416 {
1417 	process_fmt(act, pci, t, log_track_insert(pdi, t), 0, NULL);
1418 }
1419 
log_queue(struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1420 static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
1421 		      char *act)
1422 {
1423 	process_fmt(act, pci, t, -1, 0, NULL);
1424 }
1425 
log_issue(struct per_dev_info * pdi,struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1426 static void log_issue(struct per_dev_info *pdi, struct per_cpu_info *pci,
1427 		      struct blk_io_trace *t, char *act)
1428 {
1429 	process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL);
1430 }
1431 
log_merge(struct per_dev_info * pdi,struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1432 static void log_merge(struct per_dev_info *pdi, struct per_cpu_info *pci,
1433 		      struct blk_io_trace *t, char *act)
1434 {
1435 	if (act[0] == 'F')
1436 		log_track_frontmerge(pdi, t);
1437 
1438 	process_fmt(act, pci, t, -1ULL, 0, NULL);
1439 }
1440 
log_action(struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1441 static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t,
1442 			char *act)
1443 {
1444 	process_fmt(act, pci, t, -1ULL, 0, NULL);
1445 }
1446 
log_generic(struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1447 static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
1448 			char *act)
1449 {
1450 	process_fmt(act, pci, t, -1ULL, 0, NULL);
1451 }
1452 
log_unplug(struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1453 static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
1454 		      char *act)
1455 {
1456 	process_fmt(act, pci, t, -1ULL, 0, NULL);
1457 }
1458 
log_split(struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1459 static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t,
1460 		      char *act)
1461 {
1462 	process_fmt(act, pci, t, -1ULL, 0, NULL);
1463 }
1464 
log_pc(struct per_cpu_info * pci,struct blk_io_trace * t,char * act)1465 static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act)
1466 {
1467 	unsigned char *buf = (unsigned char *) t + sizeof(*t);
1468 
1469 	process_fmt(act, pci, t, -1ULL, t->pdu_len, buf);
1470 }
1471 
dump_trace_pc(struct blk_io_trace * t,struct per_dev_info * pdi,struct per_cpu_info * pci)1472 static void dump_trace_pc(struct blk_io_trace *t, struct per_dev_info *pdi,
1473 			  struct per_cpu_info *pci)
1474 {
1475 	int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1476 	int act = t->action & 0xffff;
1477 
1478 	switch (act) {
1479 		case __BLK_TA_QUEUE:
1480 			log_generic(pci, t, "Q");
1481 			account_pc_queue(t, pci, w);
1482 			break;
1483 		case __BLK_TA_GETRQ:
1484 			log_generic(pci, t, "G");
1485 			break;
1486 		case __BLK_TA_SLEEPRQ:
1487 			log_generic(pci, t, "S");
1488 			break;
1489 		case __BLK_TA_REQUEUE:
1490 			/*
1491 			 * can happen if we miss traces, don't let it go
1492 			 * below zero
1493 			 */
1494 			if (pdi->cur_depth[w])
1495 				pdi->cur_depth[w]--;
1496 			account_pc_requeue(t, pci, w);
1497 			log_generic(pci, t, "R");
1498 			break;
1499 		case __BLK_TA_ISSUE:
1500 			account_pc_issue(t, pci, w);
1501 			pdi->cur_depth[w]++;
1502 			if (pdi->cur_depth[w] > pdi->max_depth[w])
1503 				pdi->max_depth[w] = pdi->cur_depth[w];
1504 			log_pc(pci, t, "D");
1505 			break;
1506 		case __BLK_TA_COMPLETE:
1507 			if (pdi->cur_depth[w])
1508 				pdi->cur_depth[w]--;
1509 			log_pc(pci, t, "C");
1510 			account_pc_c(t, pci, w);
1511 			break;
1512 		case __BLK_TA_INSERT:
1513 			log_pc(pci, t, "I");
1514 			break;
1515 		default:
1516 			fprintf(stderr, "Bad pc action %x\n", act);
1517 			break;
1518 	}
1519 }
1520 
dump_trace_fs(struct blk_io_trace * t,struct per_dev_info * pdi,struct per_cpu_info * pci)1521 static void dump_trace_fs(struct blk_io_trace *t, struct per_dev_info *pdi,
1522 			  struct per_cpu_info *pci)
1523 {
1524 	int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1525 	int act = t->action & 0xffff;
1526 
1527 	switch (act) {
1528 		case __BLK_TA_QUEUE:
1529 			log_track_queue(pdi, t);
1530 			account_queue(t, pci, w);
1531 			log_queue(pci, t, "Q");
1532 			break;
1533 		case __BLK_TA_INSERT:
1534 			log_insert(pdi, pci, t, "I");
1535 			break;
1536 		case __BLK_TA_BACKMERGE:
1537 			account_m(t, pci, w);
1538 			log_merge(pdi, pci, t, "M");
1539 			break;
1540 		case __BLK_TA_FRONTMERGE:
1541 			account_m(t, pci, w);
1542 			log_merge(pdi, pci, t, "F");
1543 			break;
1544 		case __BLK_TA_GETRQ:
1545 			log_track_getrq(pdi, t);
1546 			log_generic(pci, t, "G");
1547 			break;
1548 		case __BLK_TA_SLEEPRQ:
1549 			log_generic(pci, t, "S");
1550 			break;
1551 		case __BLK_TA_REQUEUE:
1552 			/*
1553 			 * can happen if we miss traces, don't let it go
1554 			 * below zero
1555 			 */
1556 			if (pdi->cur_depth[w])
1557 				pdi->cur_depth[w]--;
1558 			account_requeue(t, pci, w);
1559 			log_queue(pci, t, "R");
1560 			break;
1561 		case __BLK_TA_ISSUE:
1562 			account_issue(t, pci, w);
1563 			pdi->cur_depth[w]++;
1564 			if (pdi->cur_depth[w] > pdi->max_depth[w])
1565 				pdi->max_depth[w] = pdi->cur_depth[w];
1566 			log_issue(pdi, pci, t, "D");
1567 			break;
1568 		case __BLK_TA_COMPLETE:
1569 			if (pdi->cur_depth[w])
1570 				pdi->cur_depth[w]--;
1571 			account_c(t, pci, w, t->bytes);
1572 			log_complete(pdi, pci, t, "C");
1573 			break;
1574 		case __BLK_TA_PLUG:
1575 			log_action(pci, t, "P");
1576 			break;
1577 		case __BLK_TA_UNPLUG_IO:
1578 			account_unplug(t, pci, 0);
1579 			log_unplug(pci, t, "U");
1580 			break;
1581 		case __BLK_TA_UNPLUG_TIMER:
1582 			account_unplug(t, pci, 1);
1583 			log_unplug(pci, t, "UT");
1584 			break;
1585 		case __BLK_TA_SPLIT:
1586 			log_split(pci, t, "X");
1587 			break;
1588 		case __BLK_TA_BOUNCE:
1589 			log_generic(pci, t, "B");
1590 			break;
1591 		case __BLK_TA_REMAP:
1592 			log_generic(pci, t, "A");
1593 			break;
1594 		case __BLK_TA_DRV_DATA:
1595 			have_drv_data = 1;
1596 			/* dump to binary file only */
1597 			break;
1598 		default:
1599 			fprintf(stderr, "Bad fs action %x\n", t->action);
1600 			break;
1601 	}
1602 }
1603 
dump_trace(struct blk_io_trace * t,struct per_cpu_info * pci,struct per_dev_info * pdi)1604 static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
1605 		       struct per_dev_info *pdi)
1606 {
1607 	if (text_output) {
1608 		if (t->action == BLK_TN_MESSAGE)
1609 			handle_notify(t);
1610 		else if (t->action & BLK_TC_ACT(BLK_TC_PC))
1611 			dump_trace_pc(t, pdi, pci);
1612 		else
1613 			dump_trace_fs(t, pdi, pci);
1614 	}
1615 
1616 	if (!pdi->events)
1617 		pdi->first_reported_time = t->time;
1618 
1619 	pdi->events++;
1620 
1621 	if (bin_output_msgs ||
1622 			    !(t->action & BLK_TC_ACT(BLK_TC_NOTIFY) &&
1623 			      t->action == BLK_TN_MESSAGE))
1624 		output_binary(t, sizeof(*t) + t->pdu_len);
1625 }
1626 
1627 /*
1628  * print in a proper way, not too small and not too big. if more than
1629  * 1000,000K, turn into M and so on
1630  */
size_cnv(char * dst,unsigned long long num,int in_kb)1631 static char *size_cnv(char *dst, unsigned long long num, int in_kb)
1632 {
1633 	char suff[] = { '\0', 'K', 'M', 'G', 'P' };
1634 	unsigned int i = 0;
1635 
1636 	if (in_kb)
1637 		i++;
1638 
1639 	while (num > 1000 * 1000ULL && (i < sizeof(suff) - 1)) {
1640 		i++;
1641 		num /= 1000;
1642 	}
1643 
1644 	sprintf(dst, "%'8Lu%c", num, suff[i]);
1645 	return dst;
1646 }
1647 
dump_io_stats(struct per_dev_info * pdi,struct io_stats * ios,char * msg)1648 static void dump_io_stats(struct per_dev_info *pdi, struct io_stats *ios,
1649 			  char *msg)
1650 {
1651 	static char x[256], y[256];
1652 
1653 	fprintf(ofp, "%s\n", msg);
1654 
1655 	fprintf(ofp, " Reads Queued:    %s, %siB\t", size_cnv(x, ios->qreads, 0), size_cnv(y, ios->qread_kb, 1));
1656 	fprintf(ofp, " Writes Queued:    %s, %siB\n", size_cnv(x, ios->qwrites, 0), size_cnv(y, ios->qwrite_kb, 1));
1657 	fprintf(ofp, " Read Dispatches: %s, %siB\t", size_cnv(x, ios->ireads, 0), size_cnv(y, ios->iread_kb, 1));
1658 	fprintf(ofp, " Write Dispatches: %s, %siB\n", size_cnv(x, ios->iwrites, 0), size_cnv(y, ios->iwrite_kb, 1));
1659 	fprintf(ofp, " Reads Requeued:  %s\t\t", size_cnv(x, ios->rrqueue, 0));
1660 	fprintf(ofp, " Writes Requeued:  %s\n", size_cnv(x, ios->wrqueue, 0));
1661 	fprintf(ofp, " Reads Completed: %s, %siB\t", size_cnv(x, ios->creads, 0), size_cnv(y, ios->cread_kb, 1));
1662 	fprintf(ofp, " Writes Completed: %s, %siB\n", size_cnv(x, ios->cwrites, 0), size_cnv(y, ios->cwrite_kb, 1));
1663 	fprintf(ofp, " Read Merges:     %s, %siB\t", size_cnv(x, ios->mreads, 0), size_cnv(y, ios->mread_kb, 1));
1664 	fprintf(ofp, " Write Merges:     %s, %siB\n", size_cnv(x, ios->mwrites, 0), size_cnv(y, ios->mwrite_kb, 1));
1665 	if (pdi) {
1666 		fprintf(ofp, " Read depth:      %'8u%8c\t", pdi->max_depth[0], ' ');
1667 		fprintf(ofp, " Write depth:      %'8u\n", pdi->max_depth[1]);
1668 	}
1669 	if (ios->qreads_pc || ios->qwrites_pc || ios->ireads_pc || ios->iwrites_pc ||
1670 	    ios->rrqueue_pc || ios->wrqueue_pc || ios->creads_pc || ios->cwrites_pc) {
1671 		fprintf(ofp, " PC Reads Queued: %s, %siB\t", size_cnv(x, ios->qreads_pc, 0), size_cnv(y, ios->qread_kb_pc, 1));
1672 		fprintf(ofp, " PC Writes Queued: %s, %siB\n", size_cnv(x, ios->qwrites_pc, 0), size_cnv(y, ios->qwrite_kb_pc, 1));
1673 		fprintf(ofp, " PC Read Disp.:   %s, %siB\t", size_cnv(x, ios->ireads_pc, 0), size_cnv(y, ios->iread_kb_pc, 1));
1674 		fprintf(ofp, " PC Write Disp.:   %s, %siB\n", size_cnv(x, ios->iwrites_pc, 0), size_cnv(y, ios->iwrite_kb_pc, 1));
1675 		fprintf(ofp, " PC Reads Req.:   %s\t\t", size_cnv(x, ios->rrqueue_pc, 0));
1676 		fprintf(ofp, " PC Writes Req.:   %s\n", size_cnv(x, ios->wrqueue_pc, 0));
1677 		fprintf(ofp, " PC Reads Compl.: %s\t\t", size_cnv(x, ios->creads_pc, 0));
1678 		fprintf(ofp, " PC Writes Compl.: %s\n", size_cnv(x, ios->cwrites, 0));
1679 	}
1680 	fprintf(ofp, " IO unplugs:      %'8lu%8c\t", ios->io_unplugs, ' ');
1681 	fprintf(ofp, " Timer unplugs:    %'8lu\n", ios->timer_unplugs);
1682 }
1683 
dump_wait_stats(struct per_process_info * ppi)1684 static void dump_wait_stats(struct per_process_info *ppi)
1685 {
1686 	unsigned long rawait = ppi->longest_allocation_wait[0] / 1000;
1687 	unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000;
1688 	unsigned long rcwait = ppi->longest_completion_wait[0] / 1000;
1689 	unsigned long wawait = ppi->longest_allocation_wait[1] / 1000;
1690 	unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000;
1691 	unsigned long wcwait = ppi->longest_completion_wait[1] / 1000;
1692 
1693 	fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' ');
1694 	fprintf(ofp, " Allocation wait:  %'8lu\n", wawait);
1695 	fprintf(ofp, " Dispatch wait:   %'8lu%8c\t", rdwait, ' ');
1696 	fprintf(ofp, " Dispatch wait:    %'8lu\n", wdwait);
1697 	fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' ');
1698 	fprintf(ofp, " Completion wait:  %'8lu\n", wcwait);
1699 }
1700 
ppi_name_compare(const void * p1,const void * p2)1701 static int ppi_name_compare(const void *p1, const void *p2)
1702 {
1703 	struct per_process_info *ppi1 = *((struct per_process_info **) p1);
1704 	struct per_process_info *ppi2 = *((struct per_process_info **) p2);
1705 	int res;
1706 
1707 	res = strverscmp(ppi1->ppm->comm, ppi2->ppm->comm);
1708 	if (!res)
1709 		res = ppi1->ppm->pid > ppi2->ppm->pid;
1710 
1711 	return res;
1712 }
1713 
sort_process_list(void)1714 static void sort_process_list(void)
1715 {
1716 	struct per_process_info **ppis;
1717 	struct per_process_info *ppi;
1718 	int i = 0;
1719 
1720 	ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *));
1721 
1722 	ppi = ppi_list;
1723 	while (ppi) {
1724 		ppis[i++] = ppi;
1725 		ppi = ppi->list_next;
1726 	}
1727 
1728 	qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare);
1729 
1730 	i = ppi_list_entries - 1;
1731 	ppi_list = NULL;
1732 	while (i >= 0) {
1733 		ppi = ppis[i];
1734 
1735 		ppi->list_next = ppi_list;
1736 		ppi_list = ppi;
1737 		i--;
1738 	}
1739 
1740 	free(ppis);
1741 }
1742 
show_process_stats(void)1743 static void show_process_stats(void)
1744 {
1745 	struct per_process_info *ppi;
1746 
1747 	sort_process_list();
1748 
1749 	ppi = ppi_list;
1750 	while (ppi) {
1751 		struct process_pid_map *ppm = ppi->ppm;
1752 		char name[64];
1753 
1754 		if (ppi->more_than_one)
1755 			sprintf(name, "%s (%u, ...)", ppm->comm, ppm->pid);
1756 		else
1757 			sprintf(name, "%s (%u)", ppm->comm, ppm->pid);
1758 
1759 		dump_io_stats(NULL, &ppi->io_stats, name);
1760 		dump_wait_stats(ppi);
1761 		ppi = ppi->list_next;
1762 	}
1763 
1764 	fprintf(ofp, "\n");
1765 }
1766 
show_device_and_cpu_stats(void)1767 static void show_device_and_cpu_stats(void)
1768 {
1769 	struct per_dev_info *pdi;
1770 	struct per_cpu_info *pci;
1771 	struct io_stats total, *ios;
1772 	unsigned long long rrate, wrate, msec;
1773 	int i, j, pci_events;
1774 	char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
1775 	char name[32];
1776 	double ratio;
1777 
1778 	for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1779 
1780 		memset(&total, 0, sizeof(total));
1781 		pci_events = 0;
1782 
1783 		if (i > 0)
1784 			fprintf(ofp, "\n");
1785 
1786 		for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) {
1787 			if (!pci->nelems)
1788 				continue;
1789 
1790 			ios = &pci->io_stats;
1791 			total.qreads += ios->qreads;
1792 			total.qwrites += ios->qwrites;
1793 			total.creads += ios->creads;
1794 			total.cwrites += ios->cwrites;
1795 			total.mreads += ios->mreads;
1796 			total.mwrites += ios->mwrites;
1797 			total.ireads += ios->ireads;
1798 			total.iwrites += ios->iwrites;
1799 			total.rrqueue += ios->rrqueue;
1800 			total.wrqueue += ios->wrqueue;
1801 			total.qread_kb += ios->qread_kb;
1802 			total.qwrite_kb += ios->qwrite_kb;
1803 			total.cread_kb += ios->cread_kb;
1804 			total.cwrite_kb += ios->cwrite_kb;
1805 			total.iread_kb += ios->iread_kb;
1806 			total.iwrite_kb += ios->iwrite_kb;
1807 			total.mread_kb += ios->mread_kb;
1808 			total.mwrite_kb += ios->mwrite_kb;
1809 
1810 			total.qreads_pc += ios->qreads_pc;
1811 			total.qwrites_pc += ios->qwrites_pc;
1812 			total.creads_pc += ios->creads_pc;
1813 			total.cwrites_pc += ios->cwrites_pc;
1814 			total.ireads_pc += ios->ireads_pc;
1815 			total.iwrites_pc += ios->iwrites_pc;
1816 			total.rrqueue_pc += ios->rrqueue_pc;
1817 			total.wrqueue_pc += ios->wrqueue_pc;
1818 			total.qread_kb_pc += ios->qread_kb_pc;
1819 			total.qwrite_kb_pc += ios->qwrite_kb_pc;
1820 			total.iread_kb_pc += ios->iread_kb_pc;
1821 			total.iwrite_kb_pc += ios->iwrite_kb_pc;
1822 
1823 			total.timer_unplugs += ios->timer_unplugs;
1824 			total.io_unplugs += ios->io_unplugs;
1825 
1826 			snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
1827 				 j, get_dev_name(pdi, name, sizeof(name)));
1828 			dump_io_stats(pdi, ios, line);
1829 			pci_events++;
1830 		}
1831 
1832 		if (pci_events > 1) {
1833 			fprintf(ofp, "\n");
1834 			snprintf(line, sizeof(line) - 1, "Total (%s):",
1835 				 get_dev_name(pdi, name, sizeof(name)));
1836 			dump_io_stats(NULL, &total, line);
1837 		}
1838 
1839 		wrate = rrate = 0;
1840 		msec = (pdi->last_reported_time - pdi->first_reported_time) / 1000000;
1841 		if (msec) {
1842 			rrate = 1000 * total.cread_kb / msec;
1843 			wrate = 1000 * total.cwrite_kb / msec;
1844 		}
1845 
1846 		fprintf(ofp, "\nThroughput (R/W): %'LuKiB/s / %'LuKiB/s\n",
1847 			rrate, wrate);
1848 		fprintf(ofp, "Events (%s): %'Lu entries\n",
1849 			get_dev_name(pdi, line, sizeof(line)), pdi->events);
1850 
1851 		collect_pdi_skips(pdi);
1852 		if (!pdi->skips && !pdi->events)
1853 			ratio = 0.0;
1854 		else
1855 			ratio = 100.0 * ((double)pdi->seq_skips /
1856 					(double)(pdi->events + pdi->seq_skips));
1857 		fprintf(ofp, "Skips: %'lu forward (%'llu - %5.1lf%%)\n",
1858 			pdi->skips, pdi->seq_skips, ratio);
1859 	}
1860 }
1861 
find_genesis(void)1862 static void find_genesis(void)
1863 {
1864 	struct trace *t = trace_list;
1865 
1866 	genesis_time = -1ULL;
1867 	while (t != NULL) {
1868 		if (t->bit->time < genesis_time)
1869 			genesis_time = t->bit->time;
1870 
1871 		t = t->next;
1872 	}
1873 
1874 	/* The time stamp record will usually be the first
1875 	 * record in the trace, but not always.
1876 	 */
1877 	if (start_timestamp
1878 	 && start_timestamp != genesis_time) {
1879 		long delta = genesis_time - start_timestamp;
1880 
1881 		abs_start_time.tv_sec  += SECONDS(delta);
1882 		abs_start_time.tv_nsec += NANO_SECONDS(delta);
1883 		if (abs_start_time.tv_nsec < 0) {
1884 			abs_start_time.tv_nsec += 1000000000;
1885 			abs_start_time.tv_sec -= 1;
1886 		} else
1887 		if (abs_start_time.tv_nsec > 1000000000) {
1888 			abs_start_time.tv_nsec -= 1000000000;
1889 			abs_start_time.tv_sec += 1;
1890 		}
1891 	}
1892 }
1893 
check_stopwatch(struct blk_io_trace * bit)1894 static inline int check_stopwatch(struct blk_io_trace *bit)
1895 {
1896 	if (bit->time < stopwatch_end &&
1897 	    bit->time >= stopwatch_start)
1898 		return 0;
1899 
1900 	return 1;
1901 }
1902 
1903 /*
1904  * return youngest entry read
1905  */
sort_entries(unsigned long long * youngest)1906 static int sort_entries(unsigned long long *youngest)
1907 {
1908 	struct per_dev_info *pdi = NULL;
1909 	struct per_cpu_info *pci = NULL;
1910 	struct trace *t;
1911 
1912 	if (!genesis_time)
1913 		find_genesis();
1914 
1915 	*youngest = 0;
1916 	while ((t = trace_list) != NULL) {
1917 		struct blk_io_trace *bit = t->bit;
1918 
1919 		trace_list = t->next;
1920 
1921 		bit->time -= genesis_time;
1922 
1923 		if (bit->time < *youngest || !*youngest)
1924 			*youngest = bit->time;
1925 
1926 		if (!pdi || pdi->dev != bit->device) {
1927 			pdi = get_dev_info(bit->device);
1928 			pci = NULL;
1929 		}
1930 
1931 		if (!pci || pci->cpu != bit->cpu)
1932 			pci = get_cpu_info(pdi, bit->cpu);
1933 
1934 		if (bit->sequence < pci->smallest_seq_read)
1935 			pci->smallest_seq_read = bit->sequence;
1936 
1937 		if (check_stopwatch(bit)) {
1938 			bit_free(bit);
1939 			t_free(t);
1940 			continue;
1941 		}
1942 
1943 		if (trace_rb_insert_sort(t))
1944 			return -1;
1945 	}
1946 
1947 	return 0;
1948 }
1949 
1950 /*
1951  * to continue, we must have traces from all online cpus in the tree
1952  */
check_cpu_map(struct per_dev_info * pdi)1953 static int check_cpu_map(struct per_dev_info *pdi)
1954 {
1955 	unsigned long *cpu_map;
1956 	struct rb_node *n;
1957 	struct trace *__t;
1958 	unsigned int i;
1959 	int ret, cpu;
1960 
1961 	/*
1962 	 * create a map of the cpus we have traces for
1963 	 */
1964 	cpu_map = malloc(pdi->cpu_map_max / sizeof(long));
1965 	n = rb_first(&rb_sort_root);
1966 	while (n) {
1967 		__t = rb_entry(n, struct trace, rb_node);
1968 		cpu = __t->bit->cpu;
1969 
1970 		cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu));
1971 		n = rb_next(n);
1972 	}
1973 
1974 	/*
1975 	 * we can't continue if pdi->cpu_map has entries set that we don't
1976 	 * have in the sort rbtree. the opposite is not a problem, though
1977 	 */
1978 	ret = 0;
1979 	for (i = 0; i < pdi->cpu_map_max / CPUS_PER_LONG; i++) {
1980 		if (pdi->cpu_map[i] & ~(cpu_map[i])) {
1981 			ret = 1;
1982 			break;
1983 		}
1984 	}
1985 
1986 	free(cpu_map);
1987 	return ret;
1988 }
1989 
check_sequence(struct per_dev_info * pdi,struct trace * t,int force)1990 static int check_sequence(struct per_dev_info *pdi, struct trace *t, int force)
1991 {
1992 	struct blk_io_trace *bit = t->bit;
1993 	unsigned long expected_sequence;
1994 	struct per_cpu_info *pci;
1995 	struct trace *__t;
1996 
1997 	pci = get_cpu_info(pdi, bit->cpu);
1998 	expected_sequence = pci->last_sequence + 1;
1999 
2000 	if (!expected_sequence) {
2001 		/*
2002 		 * 1 should be the first entry, just allow it
2003 		 */
2004 		if (bit->sequence == 1)
2005 			return 0;
2006 		if (bit->sequence == pci->smallest_seq_read)
2007 			return 0;
2008 
2009 		return check_cpu_map(pdi);
2010 	}
2011 
2012 	if (bit->sequence == expected_sequence)
2013 		return 0;
2014 
2015 	/*
2016 	 * we may not have seen that sequence yet. if we are not doing
2017 	 * the final run, break and wait for more entries.
2018 	 */
2019 	if (expected_sequence < pci->smallest_seq_read) {
2020 		__t = trace_rb_find_last(pdi, pci, expected_sequence);
2021 		if (!__t)
2022 			goto skip;
2023 
2024 		__put_trace_last(pdi, __t);
2025 		return 0;
2026 	} else if (!force) {
2027 		return 1;
2028 	} else {
2029 skip:
2030 		if (check_current_skips(pci, bit->sequence))
2031 			return 0;
2032 
2033 		if (expected_sequence < bit->sequence)
2034 			insert_skip(pci, expected_sequence, bit->sequence - 1);
2035 		return 0;
2036 	}
2037 }
2038 
show_entries_rb(int force)2039 static void show_entries_rb(int force)
2040 {
2041 	struct per_dev_info *pdi = NULL;
2042 	struct per_cpu_info *pci = NULL;
2043 	struct blk_io_trace *bit;
2044 	struct rb_node *n;
2045 	struct trace *t;
2046 
2047 	while ((n = rb_first(&rb_sort_root)) != NULL) {
2048 		if (is_done() && !force && !pipeline)
2049 			break;
2050 
2051 		t = rb_entry(n, struct trace, rb_node);
2052 		bit = t->bit;
2053 
2054 		if (read_sequence - t->read_sequence < 1 && !force)
2055 			break;
2056 
2057 		if (!pdi || pdi->dev != bit->device) {
2058 			pdi = get_dev_info(bit->device);
2059 			pci = NULL;
2060 		}
2061 
2062 		if (!pdi) {
2063 			fprintf(stderr, "Unknown device ID? (%d,%d)\n",
2064 				MAJOR(bit->device), MINOR(bit->device));
2065 			break;
2066 		}
2067 
2068 		if (check_sequence(pdi, t, force))
2069 			break;
2070 
2071 		if (!force && bit->time > last_allowed_time)
2072 			break;
2073 
2074 		check_time(pdi, bit);
2075 
2076 		if (!pci || pci->cpu != bit->cpu)
2077 			pci = get_cpu_info(pdi, bit->cpu);
2078 
2079 		pci->last_sequence = bit->sequence;
2080 
2081 		pci->nelems++;
2082 
2083 		if (bit->action & (act_mask << BLK_TC_SHIFT))
2084 			dump_trace(bit, pci, pdi);
2085 
2086 		put_trace(pdi, t);
2087 	}
2088 }
2089 
read_data(int fd,void * buffer,int bytes,int block,int * fdblock)2090 static int read_data(int fd, void *buffer, int bytes, int block, int *fdblock)
2091 {
2092 	int ret, bytes_left, fl;
2093 	void *p;
2094 
2095 	if (block != *fdblock) {
2096 		fl = fcntl(fd, F_GETFL);
2097 
2098 		if (!block) {
2099 			*fdblock = 0;
2100 			fcntl(fd, F_SETFL, fl | O_NONBLOCK);
2101 		} else {
2102 			*fdblock = 1;
2103 			fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
2104 		}
2105 	}
2106 
2107 	bytes_left = bytes;
2108 	p = buffer;
2109 	while (bytes_left > 0) {
2110 		ret = read(fd, p, bytes_left);
2111 		if (!ret)
2112 			return 1;
2113 		else if (ret < 0) {
2114 			if (errno != EAGAIN) {
2115 				perror("read");
2116 				return -1;
2117 			}
2118 
2119 			/*
2120 			 * never do partial reads. we can return if we
2121 			 * didn't read anything and we should not block,
2122 			 * otherwise wait for data
2123 			 */
2124 			if ((bytes_left == bytes) && !block)
2125 				return 1;
2126 
2127 			usleep(10);
2128 			continue;
2129 		} else {
2130 			p += ret;
2131 			bytes_left -= ret;
2132 		}
2133 	}
2134 
2135 	return 0;
2136 }
2137 
get_pdulen(struct blk_io_trace * bit)2138 static inline __u16 get_pdulen(struct blk_io_trace *bit)
2139 {
2140 	if (data_is_native)
2141 		return bit->pdu_len;
2142 
2143 	return __bswap_16(bit->pdu_len);
2144 }
2145 
get_magic(struct blk_io_trace * bit)2146 static inline __u32 get_magic(struct blk_io_trace *bit)
2147 {
2148 	if (data_is_native)
2149 		return bit->magic;
2150 
2151 	return __bswap_32(bit->magic);
2152 }
2153 
read_events(int fd,int always_block,int * fdblock)2154 static int read_events(int fd, int always_block, int *fdblock)
2155 {
2156 	struct per_dev_info *pdi = NULL;
2157 	unsigned int events = 0;
2158 
2159 	while (!is_done() && events < rb_batch) {
2160 		struct blk_io_trace *bit;
2161 		struct trace *t;
2162 		int pdu_len, should_block, ret;
2163 		__u32 magic;
2164 
2165 		bit = bit_alloc();
2166 
2167 		should_block = !events || always_block;
2168 
2169 		ret = read_data(fd, bit, sizeof(*bit), should_block, fdblock);
2170 		if (ret) {
2171 			bit_free(bit);
2172 			if (!events && ret < 0)
2173 				events = ret;
2174 			break;
2175 		}
2176 
2177 		/*
2178 		 * look at first trace to check whether we need to convert
2179 		 * data in the future
2180 		 */
2181 		if (data_is_native == -1 && check_data_endianness(bit->magic))
2182 			break;
2183 
2184 		magic = get_magic(bit);
2185 		if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
2186 			fprintf(stderr, "Bad magic %x\n", magic);
2187 			break;
2188 		}
2189 
2190 		pdu_len = get_pdulen(bit);
2191 		if (pdu_len) {
2192 			void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
2193 
2194 			if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1, fdblock)) {
2195 				bit_free(ptr);
2196 				break;
2197 			}
2198 
2199 			bit = ptr;
2200 		}
2201 
2202 		trace_to_cpu(bit);
2203 
2204 		if (verify_trace(bit)) {
2205 			bit_free(bit);
2206 			continue;
2207 		}
2208 
2209 		/*
2210 		 * not a real trace, so grab and handle it here
2211 		 */
2212 		if (bit->action & BLK_TC_ACT(BLK_TC_NOTIFY) && bit->action != BLK_TN_MESSAGE) {
2213 			handle_notify(bit);
2214 			output_binary(bit, sizeof(*bit) + bit->pdu_len);
2215 			continue;
2216 		}
2217 
2218 		t = t_alloc();
2219 		memset(t, 0, sizeof(*t));
2220 		t->bit = bit;
2221 		t->read_sequence = read_sequence;
2222 
2223 		t->next = trace_list;
2224 		trace_list = t;
2225 
2226 		if (!pdi || pdi->dev != bit->device)
2227 			pdi = get_dev_info(bit->device);
2228 
2229 		if (bit->time > pdi->last_read_time)
2230 			pdi->last_read_time = bit->time;
2231 
2232 		events++;
2233 	}
2234 
2235 	return events;
2236 }
2237 
2238 /*
2239  * Managing input streams
2240  */
2241 
2242 struct ms_stream {
2243 	struct ms_stream *next;
2244 	struct trace *first, *last;
2245 	struct per_dev_info *pdi;
2246 	unsigned int cpu;
2247 };
2248 
2249 #define MS_HASH(d, c) ((MAJOR(d) & 0xff) ^ (MINOR(d) & 0xff) ^ (cpu & 0xff))
2250 
2251 struct ms_stream *ms_head;
2252 struct ms_stream *ms_hash[256];
2253 
2254 static void ms_sort(struct ms_stream *msp);
2255 static int ms_prime(struct ms_stream *msp);
2256 
ms_peek(struct ms_stream * msp)2257 static inline struct trace *ms_peek(struct ms_stream *msp)
2258 {
2259 	return (msp == NULL) ? NULL : msp->first;
2260 }
2261 
ms_peek_time(struct ms_stream * msp)2262 static inline __u64 ms_peek_time(struct ms_stream *msp)
2263 {
2264 	return ms_peek(msp)->bit->time;
2265 }
2266 
ms_resort(struct ms_stream * msp)2267 static inline void ms_resort(struct ms_stream *msp)
2268 {
2269 	if (msp->next && ms_peek_time(msp) > ms_peek_time(msp->next)) {
2270 		ms_head = msp->next;
2271 		msp->next = NULL;
2272 		ms_sort(msp);
2273 	}
2274 }
2275 
ms_deq(struct ms_stream * msp)2276 static inline void ms_deq(struct ms_stream *msp)
2277 {
2278 	msp->first = msp->first->next;
2279 	if (!msp->first) {
2280 		msp->last = NULL;
2281 		if (!ms_prime(msp)) {
2282 			ms_head = msp->next;
2283 			msp->next = NULL;
2284 			return;
2285 		}
2286 	}
2287 
2288 	ms_resort(msp);
2289 }
2290 
ms_sort(struct ms_stream * msp)2291 static void ms_sort(struct ms_stream *msp)
2292 {
2293 	__u64 msp_t = ms_peek_time(msp);
2294 	struct ms_stream *this_msp = ms_head;
2295 
2296 	if (this_msp == NULL)
2297 		ms_head = msp;
2298 	else if (msp_t < ms_peek_time(this_msp)) {
2299 		msp->next = this_msp;
2300 		ms_head = msp;
2301 	}
2302 	else {
2303 		while (this_msp->next && ms_peek_time(this_msp->next) < msp_t)
2304 			this_msp = this_msp->next;
2305 
2306 		msp->next = this_msp->next;
2307 		this_msp->next = msp;
2308 	}
2309 }
2310 
ms_prime(struct ms_stream * msp)2311 static int ms_prime(struct ms_stream *msp)
2312 {
2313 	__u32 magic;
2314 	unsigned int i;
2315 	struct trace *t;
2316 	struct per_dev_info *pdi = msp->pdi;
2317 	struct per_cpu_info *pci = get_cpu_info(pdi, msp->cpu);
2318 	struct blk_io_trace *bit = NULL;
2319 	int ret, pdu_len, ndone = 0;
2320 
2321 	for (i = 0; !is_done() && pci->fd >= 0 && i < rb_batch; i++) {
2322 		bit = bit_alloc();
2323 		ret = read_data(pci->fd, bit, sizeof(*bit), 1, &pci->fdblock);
2324 		if (ret)
2325 			goto err;
2326 
2327 		if (data_is_native == -1 && check_data_endianness(bit->magic))
2328 			goto err;
2329 
2330 		magic = get_magic(bit);
2331 		if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
2332 			fprintf(stderr, "Bad magic %x\n", magic);
2333 			goto err;
2334 
2335 		}
2336 
2337 		pdu_len = get_pdulen(bit);
2338 		if (pdu_len) {
2339 			void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
2340 			ret = read_data(pci->fd, ptr + sizeof(*bit), pdu_len,
2341 							     1, &pci->fdblock);
2342 			if (ret) {
2343 				free(ptr);
2344 				bit = NULL;
2345 				goto err;
2346 			}
2347 
2348 			bit = ptr;
2349 		}
2350 
2351 		trace_to_cpu(bit);
2352 		if (verify_trace(bit))
2353 			goto err;
2354 
2355 		if (bit->action & BLK_TC_ACT(BLK_TC_NOTIFY) && bit->action != BLK_TN_MESSAGE) {
2356 			handle_notify(bit);
2357 			output_binary(bit, sizeof(*bit) + bit->pdu_len);
2358 			bit_free(bit);
2359 
2360 			i -= 1;
2361 			continue;
2362 		}
2363 
2364 		if (bit->time > pdi->last_read_time)
2365 			pdi->last_read_time = bit->time;
2366 
2367 		t = t_alloc();
2368 		memset(t, 0, sizeof(*t));
2369 		t->bit = bit;
2370 
2371 		if (msp->first == NULL)
2372 			msp->first = msp->last = t;
2373 		else {
2374 			msp->last->next = t;
2375 			msp->last = t;
2376 		}
2377 
2378 		ndone++;
2379 	}
2380 
2381 	return ndone;
2382 
2383 err:
2384 	if (bit) bit_free(bit);
2385 
2386 	cpu_mark_offline(pdi, pci->cpu);
2387 	close(pci->fd);
2388 	pci->fd = -1;
2389 
2390 	return ndone;
2391 }
2392 
ms_alloc(struct per_dev_info * pdi,int cpu)2393 static struct ms_stream *ms_alloc(struct per_dev_info *pdi, int cpu)
2394 {
2395 	struct ms_stream *msp = malloc(sizeof(*msp));
2396 
2397 	msp->next = NULL;
2398 	msp->first = msp->last = NULL;
2399 	msp->pdi = pdi;
2400 	msp->cpu = cpu;
2401 
2402 	if (ms_prime(msp))
2403 		ms_sort(msp);
2404 
2405 	return msp;
2406 }
2407 
setup_file(struct per_dev_info * pdi,int cpu)2408 static int setup_file(struct per_dev_info *pdi, int cpu)
2409 {
2410 	int len = 0;
2411 	struct stat st;
2412 	char *p, *dname;
2413 	struct per_cpu_info *pci = get_cpu_info(pdi, cpu);
2414 
2415 	pci->cpu = cpu;
2416 	pci->fdblock = -1;
2417 
2418 	p = strdup(pdi->name);
2419 	dname = dirname(p);
2420 	if (strcmp(dname, ".")) {
2421 		input_dir = dname;
2422 		p = strdup(pdi->name);
2423 		strcpy(pdi->name, basename(p));
2424 	}
2425 	free(p);
2426 
2427 	if (input_dir)
2428 		len = sprintf(pci->fname, "%s/", input_dir);
2429 
2430 	snprintf(pci->fname + len, sizeof(pci->fname)-1-len,
2431 		 "%s.blktrace.%d", pdi->name, pci->cpu);
2432 	if (stat(pci->fname, &st) < 0)
2433 		return 0;
2434 	if (!st.st_size)
2435 		return 1;
2436 
2437 	pci->fd = open(pci->fname, O_RDONLY);
2438 	if (pci->fd < 0) {
2439 		perror(pci->fname);
2440 		return 0;
2441 	}
2442 
2443 	printf("Input file %s added\n", pci->fname);
2444 	cpu_mark_online(pdi, pci->cpu);
2445 
2446 	pdi->nfiles++;
2447 	ms_alloc(pdi, pci->cpu);
2448 
2449 	return 1;
2450 }
2451 
handle(struct ms_stream * msp)2452 static int handle(struct ms_stream *msp)
2453 {
2454 	struct trace *t;
2455 	struct per_dev_info *pdi;
2456 	struct per_cpu_info *pci;
2457 	struct blk_io_trace *bit;
2458 
2459 	t = ms_peek(msp);
2460 
2461 	bit = t->bit;
2462 	pdi = msp->pdi;
2463 	pci = get_cpu_info(pdi, msp->cpu);
2464 	pci->nelems++;
2465 	bit->time -= genesis_time;
2466 
2467 	if (t->bit->time > stopwatch_end)
2468 		return 0;
2469 
2470 	pdi->last_reported_time = bit->time;
2471 	if ((bit->action & (act_mask << BLK_TC_SHIFT))&&
2472 	    t->bit->time >= stopwatch_start)
2473 		dump_trace(bit, pci, pdi);
2474 
2475 	ms_deq(msp);
2476 
2477 	if (text_output)
2478 		trace_rb_insert_last(pdi, t);
2479 	else {
2480 		bit_free(t->bit);
2481 		t_free(t);
2482 	}
2483 
2484 	return 1;
2485 }
2486 
2487 /*
2488  * Check if we need to sanitize the name. We allow 'foo', or if foo.blktrace.X
2489  * is given, then strip back down to 'foo' to avoid missing files.
2490  */
name_fixup(char * name)2491 static int name_fixup(char *name)
2492 {
2493 	char *b;
2494 
2495 	if (!name)
2496 		return 1;
2497 
2498 	b = strstr(name, ".blktrace.");
2499 	if (b)
2500 		*b = '\0';
2501 
2502 	return 0;
2503 }
2504 
do_file(void)2505 static int do_file(void)
2506 {
2507 	int i, cpu, ret;
2508 	struct per_dev_info *pdi;
2509 
2510 	/*
2511 	 * first prepare all files for reading
2512 	 */
2513 	for (i = 0; i < ndevices; i++) {
2514 		pdi = &devices[i];
2515 		ret = name_fixup(pdi->name);
2516 		if (ret)
2517 			return ret;
2518 
2519 		for (cpu = 0; setup_file(pdi, cpu); cpu++)
2520 			;
2521 	}
2522 
2523 	/*
2524 	 * Get the initial time stamp
2525 	 */
2526 	if (ms_head)
2527 		genesis_time = ms_peek_time(ms_head);
2528 
2529 	/*
2530 	 * Keep processing traces while any are left
2531 	 */
2532 	while (!is_done() && ms_head && handle(ms_head))
2533 		;
2534 
2535 	return 0;
2536 }
2537 
do_pipe(int fd)2538 static void do_pipe(int fd)
2539 {
2540 	unsigned long long youngest;
2541 	int events, fdblock;
2542 
2543 	last_allowed_time = -1ULL;
2544 	fdblock = -1;
2545 	while ((events = read_events(fd, 0, &fdblock)) > 0) {
2546 		read_sequence++;
2547 
2548 #if 0
2549 		smallest_seq_read = -1U;
2550 #endif
2551 
2552 		if (sort_entries(&youngest))
2553 			break;
2554 
2555 		if (youngest > stopwatch_end)
2556 			break;
2557 
2558 		show_entries_rb(0);
2559 	}
2560 
2561 	if (rb_sort_entries)
2562 		show_entries_rb(1);
2563 }
2564 
do_fifo(void)2565 static int do_fifo(void)
2566 {
2567 	int fd;
2568 
2569 	if (!strcmp(pipename, "-"))
2570 		fd = dup(STDIN_FILENO);
2571 	else
2572 		fd = open(pipename, O_RDONLY);
2573 
2574 	if (fd == -1) {
2575 		perror("dup stdin");
2576 		return -1;
2577 	}
2578 
2579 	do_pipe(fd);
2580 	close(fd);
2581 	return 0;
2582 }
2583 
show_stats(void)2584 static void show_stats(void)
2585 {
2586 	if (!ofp)
2587 		return;
2588 	if (stats_printed)
2589 		return;
2590 
2591 	stats_printed = 1;
2592 
2593 	if (per_process_stats)
2594 		show_process_stats();
2595 
2596 	if (per_device_and_cpu_stats)
2597 		show_device_and_cpu_stats();
2598 
2599 	fflush(ofp);
2600 }
2601 
handle_sigint(int sig)2602 static void handle_sigint(__attribute__((__unused__)) int sig)
2603 {
2604 	done = 1;
2605 }
2606 
2607 /*
2608  * Extract start and duration times from a string, allowing
2609  * us to specify a time interval of interest within a trace.
2610  * Format: "duration" (start is zero) or "start:duration".
2611  */
find_stopwatch_interval(char * string)2612 static int find_stopwatch_interval(char *string)
2613 {
2614 	double value;
2615 	char *sp;
2616 
2617 	value = strtod(string, &sp);
2618 	if (sp == string) {
2619 		fprintf(stderr,"Invalid stopwatch timer: %s\n", string);
2620 		return 1;
2621 	}
2622 	if (*sp == ':') {
2623 		stopwatch_start = DOUBLE_TO_NANO_ULL(value);
2624 		string = sp + 1;
2625 		value = strtod(string, &sp);
2626 		if (sp == string || *sp != '\0') {
2627 			fprintf(stderr,"Invalid stopwatch duration time: %s\n",
2628 				string);
2629 			return 1;
2630 		}
2631 	} else if (*sp != '\0') {
2632 		fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
2633 		return 1;
2634 	}
2635 	stopwatch_end = DOUBLE_TO_NANO_ULL(value);
2636 	if (stopwatch_end <= stopwatch_start) {
2637 		fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n",
2638 			stopwatch_start, stopwatch_end);
2639 		return 1;
2640 	}
2641 
2642 	return 0;
2643 }
2644 
is_pipe(const char * str)2645 static int is_pipe(const char *str)
2646 {
2647 	struct stat st;
2648 
2649 	if (!strcmp(str, "-"))
2650 		return 1;
2651 	if (!stat(str, &st) && S_ISFIFO(st.st_mode))
2652 		return 1;
2653 
2654 	return 0;
2655 }
2656 
2657 #define S_OPTS  "a:A:b:D:d:f:F:hi:o:Oqstw:vVM"
2658 static char usage_str[] =    "\n\n" \
2659 	"-i <file>           | --input=<file>\n" \
2660 	"[ -a <action field> | --act-mask=<action field> ]\n" \
2661 	"[ -A <action mask>  | --set-mask=<action mask> ]\n" \
2662 	"[ -b <traces>       | --batch=<traces> ]\n" \
2663 	"[ -d <file>         | --dump-binary=<file> ]\n" \
2664 	"[ -D <dir>          | --input-directory=<dir> ]\n" \
2665 	"[ -f <format>       | --format=<format> ]\n" \
2666 	"[ -F <spec>         | --format-spec=<spec> ]\n" \
2667 	"[ -h                | --hash-by-name ]\n" \
2668 	"[ -o <file>         | --output=<file> ]\n" \
2669 	"[ -O                | --no-text-output ]\n" \
2670 	"[ -q                | --quiet ]\n" \
2671 	"[ -s                | --per-program-stats ]\n" \
2672 	"[ -t                | --track-ios ]\n" \
2673 	"[ -w <time>         | --stopwatch=<time> ]\n" \
2674 	"[ -M                | --no-msgs\n" \
2675 	"[ -v                | --verbose ]\n" \
2676 	"[ -V                | --version ]\n\n" \
2677 	"\t-b stdin read batching\n" \
2678 	"\t-d Output file. If specified, binary data is written to file\n" \
2679 	"\t-D Directory to prepend to input file names\n" \
2680 	"\t-f Output format. Customize the output format. The format field\n" \
2681 	"\t   identifies can be found in the documentation\n" \
2682 	"\t-F Format specification. Can be found in the documentation\n" \
2683 	"\t-h Hash processes by name, not pid\n" \
2684 	"\t-i Input file containing trace data, or '-' for stdin\n" \
2685 	"\t-o Output file. If not given, output is stdout\n" \
2686 	"\t-O Do NOT output text data\n" \
2687 	"\t-q Quiet. Don't display any stats at the end of the trace\n" \
2688 	"\t-s Show per-program io statistics\n" \
2689 	"\t-t Track individual ios. Will tell you the time a request took\n" \
2690 	"\t   to get queued, to get dispatched, and to get completed\n" \
2691 	"\t-w Only parse data between the given time interval in seconds.\n" \
2692 	"\t   If 'start' isn't given, blkparse defaults the start time to 0\n" \
2693 	"\t-M Do not output messages to binary file\n" \
2694 	"\t-v More verbose for marginal errors\n" \
2695 	"\t-V Print program version info\n\n";
2696 
usage(char * prog)2697 static void usage(char *prog)
2698 {
2699 	fprintf(stderr, "Usage: %s %s %s", prog, blkparse_version, usage_str);
2700 }
2701 
main(int argc,char * argv[])2702 int main(int argc, char *argv[])
2703 {
2704 	int i, c, ret, mode;
2705 	int act_mask_tmp = 0;
2706 	char *ofp_buffer = NULL;
2707 	char *bin_ofp_buffer = NULL;
2708 
2709 	while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
2710 		switch (c) {
2711 		case 'a':
2712 			i = find_mask_map(optarg);
2713 			if (i < 0) {
2714 				fprintf(stderr,"Invalid action mask %s\n",
2715 					optarg);
2716 				return 1;
2717 			}
2718 			act_mask_tmp |= i;
2719 			break;
2720 
2721 		case 'A':
2722 			if ((sscanf(optarg, "%x", &i) != 1) ||
2723 							!valid_act_opt(i)) {
2724 				fprintf(stderr,
2725 					"Invalid set action mask %s/0x%x\n",
2726 					optarg, i);
2727 				return 1;
2728 			}
2729 			act_mask_tmp = i;
2730 			break;
2731 		case 'i':
2732 			if (is_pipe(optarg) && !pipeline) {
2733 				pipeline = 1;
2734 				pipename = strdup(optarg);
2735 			} else if (resize_devices(optarg) != 0)
2736 				return 1;
2737 			break;
2738 		case 'D':
2739 			input_dir = optarg;
2740 			break;
2741 		case 'o':
2742 			output_name = optarg;
2743 			break;
2744 		case 'O':
2745 			text_output = 0;
2746 			break;
2747 		case 'b':
2748 			rb_batch = atoi(optarg);
2749 			if (rb_batch <= 0)
2750 				rb_batch = RB_BATCH_DEFAULT;
2751 			break;
2752 		case 's':
2753 			per_process_stats = 1;
2754 			break;
2755 		case 't':
2756 			track_ios = 1;
2757 			break;
2758 		case 'q':
2759 			per_device_and_cpu_stats = 0;
2760 			break;
2761 		case 'w':
2762 			if (find_stopwatch_interval(optarg) != 0)
2763 				return 1;
2764 			break;
2765 		case 'f':
2766 			set_all_format_specs(optarg);
2767 			break;
2768 		case 'F':
2769 			if (add_format_spec(optarg) != 0)
2770 				return 1;
2771 			break;
2772 		case 'h':
2773 			ppi_hash_by_pid = 0;
2774 			break;
2775 		case 'v':
2776 			verbose++;
2777 			break;
2778 		case 'V':
2779 			printf("%s version %s\n", argv[0], blkparse_version);
2780 			return 0;
2781 		case 'd':
2782 			dump_binary = optarg;
2783 			break;
2784 		case 'M':
2785 			bin_output_msgs = 0;
2786 			break;
2787 		default:
2788 			usage(argv[0]);
2789 			return 1;
2790 		}
2791 	}
2792 
2793 	while (optind < argc) {
2794 		if (is_pipe(argv[optind]) && !pipeline) {
2795 			pipeline = 1;
2796 			pipename = strdup(argv[optind]);
2797 		} else if (resize_devices(argv[optind]) != 0)
2798 			return 1;
2799 		optind++;
2800 	}
2801 
2802 	if (!pipeline && !ndevices) {
2803 		usage(argv[0]);
2804 		return 1;
2805 	}
2806 
2807 	if (act_mask_tmp != 0)
2808 		act_mask = act_mask_tmp;
2809 
2810 	memset(&rb_sort_root, 0, sizeof(rb_sort_root));
2811 
2812 	signal(SIGINT, handle_sigint);
2813 	signal(SIGHUP, handle_sigint);
2814 	signal(SIGTERM, handle_sigint);
2815 
2816 	setlocale(LC_NUMERIC, "en_US");
2817 
2818 	if (text_output) {
2819 		if (!output_name) {
2820 			ofp = fdopen(STDOUT_FILENO, "w");
2821 			mode = _IOLBF;
2822 		} else {
2823 			char ofname[128];
2824 
2825 			snprintf(ofname, sizeof(ofname) - 1, "%s", output_name);
2826 			ofp = fopen(ofname, "w");
2827 			mode = _IOFBF;
2828 		}
2829 
2830 		if (!ofp) {
2831 			perror("fopen");
2832 			return 1;
2833 		}
2834 
2835 		ofp_buffer = malloc(4096);
2836 		if (setvbuf(ofp, ofp_buffer, mode, 4096)) {
2837 			perror("setvbuf");
2838 			return 1;
2839 		}
2840 	}
2841 
2842 	if (dump_binary) {
2843 		dump_fp = fopen(dump_binary, "w");
2844 		if (!dump_fp) {
2845 			perror(dump_binary);
2846 			dump_binary = NULL;
2847 			return 1;
2848 		}
2849 		bin_ofp_buffer = malloc(128 * 1024);
2850 		if (setvbuf(dump_fp, bin_ofp_buffer, _IOFBF, 128 * 1024)) {
2851 			perror("setvbuf binary");
2852 			return 1;
2853 		}
2854 	}
2855 
2856 	if (pipeline)
2857 		ret = do_fifo();
2858 	else
2859 		ret = do_file();
2860 
2861 	if (!ret)
2862 		show_stats();
2863 
2864 	if (have_drv_data && !dump_binary)
2865 		printf("\ndiscarded traces containing low-level device driver "
2866 		       "specific data (only available in binary output)\n");
2867 
2868 	if (ofp_buffer) {
2869 		fflush(ofp);
2870 		free(ofp_buffer);
2871 	}
2872 	if (bin_ofp_buffer) {
2873 		fflush(dump_fp);
2874 		free(bin_ofp_buffer);
2875 	}
2876 	return ret;
2877 }
2878