Commit | Line | Data |
---|---|---|
d0ca268b JA |
1 | /* |
2 | * block queue tracing application | |
3 | * | |
d956a2cd JA |
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 | * | |
d0ca268b JA |
20 | */ |
21 | #include <pthread.h> | |
22 | #include <sys/types.h> | |
23 | #include <sys/stat.h> | |
24 | #include <unistd.h> | |
25 | #include <locale.h> | |
26 | #include <signal.h> | |
27 | #include <fcntl.h> | |
28 | #include <string.h> | |
29 | #include <sys/ioctl.h> | |
b9d4294e | 30 | #include <sys/param.h> |
d0ca268b JA |
31 | #include <stdio.h> |
32 | #include <stdlib.h> | |
33 | #include <sched.h> | |
d39c04ca AB |
34 | #include <ctype.h> |
35 | #include <getopt.h> | |
d0ca268b JA |
36 | |
37 | #include "blktrace.h" | |
38 | ||
39 | #define BUF_SIZE (128 *1024) | |
40 | #define BUF_NR (4) | |
41 | ||
d39c04ca AB |
42 | #define DECLARE_MASK_MAP(mask) { BLK_TC_##mask, #mask, "BLK_TC_"#mask } |
43 | #define COMPARE_MASK_MAP(mmp, str) \ | |
75da3c6a NS |
44 | (!strcasecmp((mmp)->short_form, (str)) || \ |
45 | !strcasecmp((mmp)->long_form, (str))) | |
d39c04ca AB |
46 | |
47 | #define VALID_SET(x) ((1 <= (x)) && ((x) < (1 << BLK_TC_SHIFT))) | |
48 | ||
49 | struct mask_map { | |
50 | int mask; | |
51 | char *short_form; | |
52 | char *long_form; | |
53 | }; | |
54 | ||
55 | struct mask_map mask_maps[] = { | |
5c86134e JA |
56 | DECLARE_MASK_MAP(READ), |
57 | DECLARE_MASK_MAP(WRITE), | |
58 | DECLARE_MASK_MAP(BARRIER), | |
59 | DECLARE_MASK_MAP(SYNC), | |
60 | DECLARE_MASK_MAP(QUEUE), | |
61 | DECLARE_MASK_MAP(REQUEUE), | |
62 | DECLARE_MASK_MAP(ISSUE), | |
63 | DECLARE_MASK_MAP(COMPLETE), | |
64 | DECLARE_MASK_MAP(FS), | |
65 | DECLARE_MASK_MAP(PC), | |
d39c04ca AB |
66 | }; |
67 | ||
d5396421 JA |
68 | #define S_OPTS "d:a:A:r:o:" |
69 | static struct option l_opts[] = { | |
5c86134e | 70 | { |
d39c04ca AB |
71 | .name = "dev", |
72 | .has_arg = 1, | |
73 | .flag = NULL, | |
74 | .val = 'd' | |
75 | }, | |
5c86134e | 76 | { |
d39c04ca AB |
77 | .name = "act-mask", |
78 | .has_arg = 1, | |
79 | .flag = NULL, | |
80 | .val = 'a' | |
81 | }, | |
5c86134e | 82 | { |
d39c04ca AB |
83 | .name = "set-mask", |
84 | .has_arg = 1, | |
85 | .flag = NULL, | |
86 | .val = 'A' | |
87 | }, | |
5c86134e | 88 | { |
5270dddd JA |
89 | .name = "relay", |
90 | .has_arg = 1, | |
91 | .flag = NULL, | |
92 | .val = 'r' | |
93 | }, | |
d5396421 JA |
94 | { |
95 | .name = "output", | |
96 | .has_arg = 1, | |
97 | .flag = NULL, | |
98 | .val = 'o' | |
99 | }, | |
d39c04ca AB |
100 | { |
101 | .name = NULL, | |
102 | .has_arg = 0, | |
103 | .flag = NULL, | |
104 | .val = 0 | |
105 | } | |
106 | }; | |
107 | ||
d0ca268b JA |
108 | struct thread_information { |
109 | int cpu; | |
110 | pthread_t thread; | |
b9d4294e JA |
111 | |
112 | int fd; | |
113 | char fn[MAXPATHLEN + 64]; | |
114 | ||
d5396421 JA |
115 | pthread_mutex_t *fd_lock; |
116 | int ofd; | |
117 | ||
d0ca268b JA |
118 | unsigned long events_processed; |
119 | }; | |
120 | ||
5270dddd | 121 | static char *relay_path; |
d0ca268b JA |
122 | |
123 | #define is_done() (*(volatile int *)(&done)) | |
124 | static volatile int done; | |
125 | ||
126 | static int devfd, ncpus; | |
127 | static struct thread_information *thread_information; | |
128 | static char *buts_name_p; | |
d39c04ca | 129 | static char *dev; |
d5396421 | 130 | static char *output_name; |
5c86134e | 131 | static int act_mask = ~0U; |
830fd65c | 132 | static int trace_started; |
d39c04ca | 133 | |
d5396421 JA |
134 | static pthread_mutex_t stdout_mutex = PTHREAD_MUTEX_INITIALIZER; |
135 | ||
d39c04ca AB |
136 | int find_mask_map(char *string) |
137 | { | |
5c86134e JA |
138 | int i; |
139 | ||
140 | for (i = 0; i < sizeof(mask_maps)/sizeof(mask_maps[0]); i++) | |
75da3c6a | 141 | if (COMPARE_MASK_MAP(&mask_maps[i], string)) |
5c86134e | 142 | return mask_maps[i].mask; |
d39c04ca | 143 | |
d39c04ca AB |
144 | return -1; |
145 | } | |
d0ca268b | 146 | |
3aabcd89 | 147 | static int start_trace(char *dev) |
d0ca268b JA |
148 | { |
149 | struct blk_user_trace_setup buts; | |
150 | ||
151 | devfd = open(dev, O_RDONLY); | |
152 | if (devfd < 0) { | |
153 | perror(dev); | |
154 | return 1; | |
155 | } | |
156 | ||
157 | memset(&buts, sizeof(buts), 0); | |
158 | buts.buf_size = BUF_SIZE; | |
159 | buts.buf_nr = BUF_NR; | |
d39c04ca | 160 | buts.act_mask = act_mask; |
d0ca268b | 161 | |
d0ca268b JA |
162 | if (ioctl(devfd, BLKSTARTTRACE, &buts) < 0) { |
163 | perror("BLKSTARTTRACE"); | |
164 | return 1; | |
165 | } | |
166 | ||
830fd65c | 167 | trace_started = 1; |
d0ca268b JA |
168 | buts_name_p = strdup(buts.name); |
169 | return 0; | |
170 | } | |
171 | ||
3aabcd89 | 172 | static void stop_trace(void) |
d0ca268b | 173 | { |
707b0914 JA |
174 | if (trace_started) { |
175 | if (ioctl(devfd, BLKSTOPTRACE) < 0) | |
176 | perror("BLKSTOPTRACE"); | |
d0ca268b | 177 | |
707b0914 JA |
178 | close(devfd); |
179 | trace_started = 0; | |
180 | } | |
d0ca268b JA |
181 | } |
182 | ||
d5396421 | 183 | static void extract_data(struct thread_information *tip, char *ofn, int nb) |
d0ca268b JA |
184 | { |
185 | int ret, bytes_left; | |
87b72777 | 186 | unsigned char *buf, *p; |
d0ca268b | 187 | |
87b72777 | 188 | buf = malloc(nb); |
d0ca268b JA |
189 | p = buf; |
190 | bytes_left = nb; | |
191 | while (bytes_left > 0) { | |
b9d4294e | 192 | ret = read(tip->fd, p, bytes_left); |
3aabcd89 JA |
193 | if (!ret) |
194 | usleep(1000); | |
195 | else if (ret < 0) { | |
b9d4294e | 196 | perror(tip->fn); |
d0ca268b | 197 | fprintf(stderr, "Thread %d extract_data %s failed\n", |
b9d4294e | 198 | tip->cpu, tip->fn); |
87b72777 | 199 | free(buf); |
d0ca268b | 200 | exit(1); |
3aabcd89 | 201 | } else { |
d0ca268b JA |
202 | p += ret; |
203 | bytes_left -= ret; | |
204 | } | |
205 | } | |
206 | ||
d5396421 | 207 | ret = write(tip->ofd, buf, nb); |
d0ca268b JA |
208 | if (ret != nb) { |
209 | perror(ofn); | |
b9d4294e | 210 | fprintf(stderr,"Thread %d extract_data %s failed\n", tip->cpu, ofn); |
87b72777 | 211 | free(buf); |
d0ca268b JA |
212 | exit(1); |
213 | } | |
87b72777 JA |
214 | |
215 | free(buf); | |
d0ca268b JA |
216 | } |
217 | ||
d5396421 JA |
218 | static inline void tip_fd_unlock(struct thread_information *tip) |
219 | { | |
220 | if (tip->fd_lock) | |
221 | pthread_mutex_unlock(tip->fd_lock); | |
222 | } | |
223 | ||
224 | static inline void tip_fd_lock(struct thread_information *tip) | |
225 | { | |
226 | if (tip->fd_lock) | |
227 | pthread_mutex_lock(tip->fd_lock); | |
228 | } | |
229 | ||
3aabcd89 | 230 | static void *extract(void *arg) |
d0ca268b JA |
231 | { |
232 | struct thread_information *tip = arg; | |
d5396421 JA |
233 | int ret, pdu_len; |
234 | char dp[64]; | |
d0ca268b JA |
235 | struct blk_io_trace t; |
236 | pid_t pid = getpid(); | |
237 | cpu_set_t cpu_mask; | |
238 | ||
239 | CPU_ZERO(&cpu_mask); | |
b9d4294e | 240 | CPU_SET((tip->cpu), &cpu_mask); |
d0ca268b JA |
241 | |
242 | if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) { | |
243 | perror("sched_setaffinity"); | |
244 | exit(1); | |
245 | } | |
246 | ||
b9d4294e JA |
247 | snprintf(tip->fn, sizeof(tip->fn), |
248 | "%s/block/%s/trace%d", relay_path, buts_name_p, tip->cpu); | |
249 | tip->fd = open(tip->fn, O_RDONLY); | |
250 | if (tip->fd < 0) { | |
251 | perror(tip->fn); | |
5c86134e JA |
252 | fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu, |
253 | tip->fn); | |
d0ca268b JA |
254 | exit(1); |
255 | } | |
256 | ||
257 | while (!is_done()) { | |
b9d4294e | 258 | ret = read(tip->fd, &t, sizeof(t)); |
d0ca268b JA |
259 | if (ret != sizeof(t)) { |
260 | if (ret < 0) { | |
b9d4294e | 261 | perror(tip->fn); |
d0ca268b | 262 | fprintf(stderr,"Thread %d failed read of %s\n", |
b9d4294e | 263 | tip->cpu, tip->fn); |
d0ca268b JA |
264 | exit(1); |
265 | } else if (ret > 0) { | |
8fc0abbc | 266 | fprintf(stderr,"Thread %d misread %s %d,%d\n", |
b9d4294e | 267 | tip->cpu, tip->fn, ret, (int)sizeof(t)); |
d0ca268b JA |
268 | exit(1); |
269 | } else { | |
270 | usleep(10000); | |
271 | continue; | |
272 | } | |
273 | } | |
274 | ||
275 | if (verify_trace(&t)) | |
276 | exit(1); | |
277 | ||
18ada3d4 JA |
278 | pdu_len = t.pdu_len; |
279 | ||
6fe4709e JA |
280 | trace_to_be(&t); |
281 | ||
d5396421 JA |
282 | tip_fd_lock(tip); |
283 | ||
284 | ret = write(tip->ofd, &t, sizeof(t)); | |
d0ca268b | 285 | if (ret < 0) { |
d5396421 JA |
286 | fprintf(stderr,"Thread %d failed write\n", tip->cpu); |
287 | tip_fd_unlock(tip); | |
d0ca268b JA |
288 | exit(1); |
289 | } | |
290 | ||
18ada3d4 | 291 | if (pdu_len) |
d5396421 JA |
292 | extract_data(tip, dp, pdu_len); |
293 | ||
294 | tip_fd_unlock(tip); | |
87b72777 | 295 | |
d0ca268b JA |
296 | tip->events_processed++; |
297 | } | |
298 | ||
299 | return NULL; | |
300 | } | |
301 | ||
3aabcd89 | 302 | static int start_threads(void) |
d0ca268b JA |
303 | { |
304 | struct thread_information *tip; | |
d5396421 | 305 | char op[64]; |
d0ca268b JA |
306 | int i; |
307 | ||
308 | ncpus = sysconf(_SC_NPROCESSORS_ONLN); | |
309 | if (ncpus < 0) { | |
310 | fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n"); | |
311 | return 1; | |
312 | } | |
d0ca268b JA |
313 | |
314 | thread_information = malloc(ncpus * sizeof(struct thread_information)); | |
315 | for (i = 0, tip = thread_information; i < ncpus; i++, tip++) { | |
d5396421 | 316 | tip->fd_lock = NULL; |
d0ca268b JA |
317 | tip->cpu = i; |
318 | tip->events_processed = 0; | |
319 | ||
d5396421 JA |
320 | if (!strcmp(output_name, "-")) { |
321 | tip->ofd = dup(1); | |
322 | tip->fd_lock = &stdout_mutex; | |
323 | } else { | |
324 | sprintf(op, "%s_out.%d", output_name, tip->cpu); | |
325 | tip->ofd = open(op, O_CREAT|O_TRUNC|O_WRONLY, 0644); | |
326 | } | |
327 | ||
328 | if (tip->ofd < 0) { | |
329 | perror(op); | |
330 | return 1; | |
331 | } | |
332 | ||
d0ca268b JA |
333 | if (pthread_create(&tip->thread, NULL, extract, tip)) { |
334 | perror( "pthread_create"); | |
335 | return 0; | |
336 | } | |
337 | } | |
338 | ||
339 | return ncpus; | |
340 | } | |
341 | ||
3aabcd89 JA |
342 | static void stop_threads(void) |
343 | { | |
344 | struct thread_information *tip = thread_information; | |
345 | int i; | |
346 | ||
347 | for (i = 0; i < ncpus; i++, tip++) { | |
348 | int ret; | |
349 | ||
350 | if (pthread_join(tip->thread, (void *) &ret)) | |
351 | perror("thread_join"); | |
d5396421 | 352 | |
b9d4294e | 353 | close(tip->fd); |
d5396421 | 354 | close(tip->ofd); |
3aabcd89 JA |
355 | } |
356 | } | |
357 | ||
d0ca268b JA |
358 | void show_stats(void) |
359 | { | |
360 | int i; | |
361 | struct thread_information *tip; | |
362 | unsigned long events_processed = 0; | |
363 | ||
d5396421 JA |
364 | if (!strcmp(output_name, "-")) |
365 | return; | |
366 | ||
d0ca268b JA |
367 | for (i = 0, tip = thread_information; i < ncpus; i++, tip++) { |
368 | printf("CPU%3d: %20ld events\n", | |
369 | tip->cpu, tip->events_processed); | |
370 | events_processed += tip->events_processed; | |
371 | } | |
372 | ||
373 | printf("Total: %20ld events\n", events_processed); | |
374 | } | |
375 | ||
376 | void handle_sigint(int sig) | |
377 | { | |
d0ca268b JA |
378 | done = 1; |
379 | } | |
380 | ||
381 | int main(int argc, char *argv[]) | |
382 | { | |
5270dddd | 383 | static char default_relay_path[] = "/relay"; |
d0ca268b | 384 | struct stat st; |
d39c04ca AB |
385 | int i, c; |
386 | int act_mask_tmp = 0; | |
387 | ||
388 | while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) { | |
389 | switch (c) { | |
390 | case 'a': | |
391 | i = find_mask_map(optarg); | |
392 | if (i < 0) { | |
393 | fprintf(stderr,"Invalid action mask %s\n", | |
394 | optarg); | |
395 | return 4; | |
396 | } | |
397 | act_mask_tmp |= i; | |
398 | break; | |
399 | ||
400 | case 'A': | |
401 | if ((sscanf(optarg, "%x", &i) != 1) || !VALID_SET(i)) { | |
402 | fprintf(stderr, | |
403 | "Invalid set action mask %s/0x%x\n", | |
404 | optarg, i); | |
405 | return 4; | |
406 | } | |
407 | act_mask_tmp = i; | |
408 | break; | |
d0ca268b | 409 | |
d39c04ca AB |
410 | case 'd': |
411 | dev = strdup(optarg); | |
412 | break; | |
413 | ||
5270dddd JA |
414 | case 'r': |
415 | relay_path = optarg; | |
416 | break; | |
417 | ||
d5396421 JA |
418 | case 'o': |
419 | output_name = strdup(optarg); | |
420 | break; | |
421 | ||
d39c04ca AB |
422 | default: |
423 | fprintf(stderr,"Usage: %s -d <dev> " | |
424 | "[-a <trace> [-a <trace>]]\n", argv[0]); | |
425 | return 4; | |
426 | } | |
427 | } | |
428 | ||
429 | if ((dev == NULL) || (optind < argc)) { | |
430 | fprintf(stderr,"Usage: %s -d <dev> " | |
431 | "[-a <trace> [-a <trace>]]\n", argv[0]); | |
432 | return 4; | |
433 | } | |
434 | ||
5270dddd JA |
435 | if (!relay_path) |
436 | relay_path = default_relay_path; | |
437 | ||
d5396421 | 438 | if (act_mask_tmp != 0) |
d39c04ca | 439 | act_mask = act_mask_tmp; |
d0ca268b JA |
440 | |
441 | if (stat(relay_path, &st) < 0) { | |
d39c04ca | 442 | fprintf(stderr,"%s does not appear to be mounted\n", |
d0ca268b JA |
443 | relay_path); |
444 | return 2; | |
445 | } | |
446 | ||
d39c04ca | 447 | if (start_trace(dev)) { |
5270dddd | 448 | close(devfd); |
d39c04ca | 449 | fprintf(stderr, "Failed to start trace on %s\n", dev); |
d0ca268b JA |
450 | return 3; |
451 | } | |
452 | ||
453 | setlocale(LC_NUMERIC, "en_US"); | |
454 | ||
d5396421 JA |
455 | if (!output_name) |
456 | output_name = strdup(buts_name_p); | |
457 | ||
d0ca268b JA |
458 | i = start_threads(); |
459 | if (!i) { | |
460 | fprintf(stderr, "Failed to start worker threads\n"); | |
461 | stop_trace(); | |
462 | return 4; | |
463 | } | |
464 | ||
d0ca268b JA |
465 | signal(SIGINT, handle_sigint); |
466 | signal(SIGHUP, handle_sigint); | |
467 | signal(SIGTERM, handle_sigint); | |
468 | ||
707b0914 | 469 | atexit(stop_trace); |
830fd65c | 470 | |
d0ca268b JA |
471 | while (!is_done()) |
472 | sleep(1); | |
473 | ||
b9d4294e | 474 | stop_threads(); |
3a9995b9 | 475 | stop_trace(); |
d0ca268b JA |
476 | show_stats(); |
477 | ||
478 | return 0; | |
479 | } | |
480 |