| 1 | /* |
| 2 | * block queue tracing application |
| 3 | * |
| 4 | * Copyright (C) 2005 Jens Axboe <axboe@suse.de> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 19 | * |
| 20 | */ |
| 21 | #include <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> |
| 30 | #include <sys/param.h> |
| 31 | #include <stdio.h> |
| 32 | #include <stdlib.h> |
| 33 | #include <sched.h> |
| 34 | #include <ctype.h> |
| 35 | #include <getopt.h> |
| 36 | |
| 37 | #include "blktrace.h" |
| 38 | |
| 39 | #define BUF_SIZE (128 *1024) |
| 40 | #define BUF_NR (4) |
| 41 | |
| 42 | #define DECLARE_MASK_MAP(mask) { BLK_TC_##mask, #mask, "BLK_TC_"#mask } |
| 43 | #define COMPARE_MASK_MAP(mmp, str) \ |
| 44 | (!strcasecmp((mmp)->short_form, (str)) || \ |
| 45 | !strcasecmp((mmp)->long_form, (str))) |
| 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[] = { |
| 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), |
| 66 | }; |
| 67 | |
| 68 | #define S_OPTS "d:a:A:r:o:" |
| 69 | static struct option l_opts[] = { |
| 70 | { |
| 71 | .name = "dev", |
| 72 | .has_arg = 1, |
| 73 | .flag = NULL, |
| 74 | .val = 'd' |
| 75 | }, |
| 76 | { |
| 77 | .name = "act-mask", |
| 78 | .has_arg = 1, |
| 79 | .flag = NULL, |
| 80 | .val = 'a' |
| 81 | }, |
| 82 | { |
| 83 | .name = "set-mask", |
| 84 | .has_arg = 1, |
| 85 | .flag = NULL, |
| 86 | .val = 'A' |
| 87 | }, |
| 88 | { |
| 89 | .name = "relay", |
| 90 | .has_arg = 1, |
| 91 | .flag = NULL, |
| 92 | .val = 'r' |
| 93 | }, |
| 94 | { |
| 95 | .name = "output", |
| 96 | .has_arg = 1, |
| 97 | .flag = NULL, |
| 98 | .val = 'o' |
| 99 | }, |
| 100 | { |
| 101 | .name = NULL, |
| 102 | .has_arg = 0, |
| 103 | .flag = NULL, |
| 104 | .val = 0 |
| 105 | } |
| 106 | }; |
| 107 | |
| 108 | struct thread_information { |
| 109 | int cpu; |
| 110 | pthread_t thread; |
| 111 | |
| 112 | int fd; |
| 113 | char fn[MAXPATHLEN + 64]; |
| 114 | |
| 115 | pthread_mutex_t *fd_lock; |
| 116 | int ofd; |
| 117 | |
| 118 | unsigned long events_processed; |
| 119 | }; |
| 120 | |
| 121 | static char *relay_path; |
| 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; |
| 129 | static char *dev; |
| 130 | static char *output_name; |
| 131 | static int act_mask = ~0U; |
| 132 | static int trace_started; |
| 133 | |
| 134 | static pthread_mutex_t stdout_mutex = PTHREAD_MUTEX_INITIALIZER; |
| 135 | |
| 136 | int find_mask_map(char *string) |
| 137 | { |
| 138 | int i; |
| 139 | |
| 140 | for (i = 0; i < sizeof(mask_maps)/sizeof(mask_maps[0]); i++) |
| 141 | if (COMPARE_MASK_MAP(&mask_maps[i], string)) |
| 142 | return mask_maps[i].mask; |
| 143 | |
| 144 | return -1; |
| 145 | } |
| 146 | |
| 147 | static int start_trace(char *dev) |
| 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; |
| 160 | buts.act_mask = act_mask; |
| 161 | |
| 162 | if (ioctl(devfd, BLKSTARTTRACE, &buts) < 0) { |
| 163 | perror("BLKSTARTTRACE"); |
| 164 | return 1; |
| 165 | } |
| 166 | |
| 167 | trace_started = 1; |
| 168 | buts_name_p = strdup(buts.name); |
| 169 | return 0; |
| 170 | } |
| 171 | |
| 172 | static void stop_trace(void) |
| 173 | { |
| 174 | if (trace_started) { |
| 175 | if (ioctl(devfd, BLKSTOPTRACE) < 0) |
| 176 | perror("BLKSTOPTRACE"); |
| 177 | |
| 178 | close(devfd); |
| 179 | trace_started = 0; |
| 180 | } |
| 181 | } |
| 182 | |
| 183 | static void extract_data(struct thread_information *tip, char *ofn, int nb) |
| 184 | { |
| 185 | int ret, bytes_left; |
| 186 | unsigned char *buf, *p; |
| 187 | |
| 188 | buf = malloc(nb); |
| 189 | p = buf; |
| 190 | bytes_left = nb; |
| 191 | while (bytes_left > 0) { |
| 192 | ret = read(tip->fd, p, bytes_left); |
| 193 | if (!ret) |
| 194 | usleep(1000); |
| 195 | else if (ret < 0) { |
| 196 | perror(tip->fn); |
| 197 | fprintf(stderr, "Thread %d extract_data %s failed\n", |
| 198 | tip->cpu, tip->fn); |
| 199 | free(buf); |
| 200 | exit(1); |
| 201 | } else { |
| 202 | p += ret; |
| 203 | bytes_left -= ret; |
| 204 | } |
| 205 | } |
| 206 | |
| 207 | ret = write(tip->ofd, buf, nb); |
| 208 | if (ret != nb) { |
| 209 | perror(ofn); |
| 210 | fprintf(stderr,"Thread %d extract_data %s failed\n", tip->cpu, ofn); |
| 211 | free(buf); |
| 212 | exit(1); |
| 213 | } |
| 214 | |
| 215 | free(buf); |
| 216 | } |
| 217 | |
| 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 | |
| 230 | static void *extract(void *arg) |
| 231 | { |
| 232 | struct thread_information *tip = arg; |
| 233 | int ret, pdu_len; |
| 234 | char dp[64]; |
| 235 | struct blk_io_trace t; |
| 236 | pid_t pid = getpid(); |
| 237 | cpu_set_t cpu_mask; |
| 238 | |
| 239 | CPU_ZERO(&cpu_mask); |
| 240 | CPU_SET((tip->cpu), &cpu_mask); |
| 241 | |
| 242 | if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) { |
| 243 | perror("sched_setaffinity"); |
| 244 | exit(1); |
| 245 | } |
| 246 | |
| 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); |
| 252 | fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu, |
| 253 | tip->fn); |
| 254 | exit(1); |
| 255 | } |
| 256 | |
| 257 | while (!is_done()) { |
| 258 | ret = read(tip->fd, &t, sizeof(t)); |
| 259 | if (ret != sizeof(t)) { |
| 260 | if (ret < 0) { |
| 261 | perror(tip->fn); |
| 262 | fprintf(stderr,"Thread %d failed read of %s\n", |
| 263 | tip->cpu, tip->fn); |
| 264 | exit(1); |
| 265 | } else if (ret > 0) { |
| 266 | fprintf(stderr,"Thread %d misread %s %d,%d\n", |
| 267 | tip->cpu, tip->fn, ret, (int)sizeof(t)); |
| 268 | exit(1); |
| 269 | } else { |
| 270 | usleep(10000); |
| 271 | continue; |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | if (verify_trace(&t)) |
| 276 | exit(1); |
| 277 | |
| 278 | pdu_len = t.pdu_len; |
| 279 | |
| 280 | trace_to_be(&t); |
| 281 | |
| 282 | tip_fd_lock(tip); |
| 283 | |
| 284 | ret = write(tip->ofd, &t, sizeof(t)); |
| 285 | if (ret < 0) { |
| 286 | fprintf(stderr,"Thread %d failed write\n", tip->cpu); |
| 287 | tip_fd_unlock(tip); |
| 288 | exit(1); |
| 289 | } |
| 290 | |
| 291 | if (pdu_len) |
| 292 | extract_data(tip, dp, pdu_len); |
| 293 | |
| 294 | tip_fd_unlock(tip); |
| 295 | |
| 296 | tip->events_processed++; |
| 297 | } |
| 298 | |
| 299 | return NULL; |
| 300 | } |
| 301 | |
| 302 | static int start_threads(void) |
| 303 | { |
| 304 | struct thread_information *tip; |
| 305 | char op[64]; |
| 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 | } |
| 313 | |
| 314 | thread_information = malloc(ncpus * sizeof(struct thread_information)); |
| 315 | for (i = 0, tip = thread_information; i < ncpus; i++, tip++) { |
| 316 | tip->fd_lock = NULL; |
| 317 | tip->cpu = i; |
| 318 | tip->events_processed = 0; |
| 319 | |
| 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 | |
| 333 | if (pthread_create(&tip->thread, NULL, extract, tip)) { |
| 334 | perror( "pthread_create"); |
| 335 | return 0; |
| 336 | } |
| 337 | } |
| 338 | |
| 339 | return ncpus; |
| 340 | } |
| 341 | |
| 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"); |
| 352 | |
| 353 | close(tip->fd); |
| 354 | close(tip->ofd); |
| 355 | } |
| 356 | } |
| 357 | |
| 358 | void show_stats(void) |
| 359 | { |
| 360 | int i; |
| 361 | struct thread_information *tip; |
| 362 | unsigned long events_processed = 0; |
| 363 | |
| 364 | if (!strcmp(output_name, "-")) |
| 365 | return; |
| 366 | |
| 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 | { |
| 378 | done = 1; |
| 379 | } |
| 380 | |
| 381 | int main(int argc, char *argv[]) |
| 382 | { |
| 383 | static char default_relay_path[] = "/relay"; |
| 384 | struct stat st; |
| 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; |
| 409 | |
| 410 | case 'd': |
| 411 | dev = strdup(optarg); |
| 412 | break; |
| 413 | |
| 414 | case 'r': |
| 415 | relay_path = optarg; |
| 416 | break; |
| 417 | |
| 418 | case 'o': |
| 419 | output_name = strdup(optarg); |
| 420 | break; |
| 421 | |
| 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 | |
| 435 | if (!relay_path) |
| 436 | relay_path = default_relay_path; |
| 437 | |
| 438 | if (act_mask_tmp != 0) |
| 439 | act_mask = act_mask_tmp; |
| 440 | |
| 441 | if (stat(relay_path, &st) < 0) { |
| 442 | fprintf(stderr,"%s does not appear to be mounted\n", |
| 443 | relay_path); |
| 444 | return 2; |
| 445 | } |
| 446 | |
| 447 | if (start_trace(dev)) { |
| 448 | close(devfd); |
| 449 | fprintf(stderr, "Failed to start trace on %s\n", dev); |
| 450 | return 3; |
| 451 | } |
| 452 | |
| 453 | setlocale(LC_NUMERIC, "en_US"); |
| 454 | |
| 455 | if (!output_name) |
| 456 | output_name = strdup(buts_name_p); |
| 457 | |
| 458 | i = start_threads(); |
| 459 | if (!i) { |
| 460 | fprintf(stderr, "Failed to start worker threads\n"); |
| 461 | stop_trace(); |
| 462 | return 4; |
| 463 | } |
| 464 | |
| 465 | signal(SIGINT, handle_sigint); |
| 466 | signal(SIGHUP, handle_sigint); |
| 467 | signal(SIGTERM, handle_sigint); |
| 468 | |
| 469 | atexit(stop_trace); |
| 470 | |
| 471 | while (!is_done()) |
| 472 | sleep(1); |
| 473 | |
| 474 | stop_threads(); |
| 475 | stop_trace(); |
| 476 | show_stats(); |
| 477 | |
| 478 | return 0; |
| 479 | } |
| 480 | |