blktrace.c

   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 <sys/statfs.h>
  32 #include <sys/poll.h>
  33 #include <sys/mman.h>
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <sched.h>
  37 #include <ctype.h>
  38 #include <getopt.h>
  39 #include <errno.h>
  40
  41 #include "blktrace.h"
  42 #include "barrier.h"
  43
  44 static char blktrace_version[] = "0.99";
  45
  46 /*
  47  * You may want to increase this even more, if you are logging at a high
  48  * rate and see skipped/missed events
  49  */
  50 #define BUF_SIZE        (512 * 1024)
  51 #define BUF_NR          (4)
  52
  53 #define OFILE_BUF       (128 * 1024)
  54
  55 #define RELAYFS_TYPE    0xF0B4A981
  56
  57 #define S_OPTS  "d:a:A:r:o:kw:Vb:n:D:"
  58 static struct option l_opts[] = {
  59         {
  60                 .name = "dev",
  61                 .has_arg = required_argument,
  62                 .flag = NULL,
  63                 .val = 'd'
  64         },
  65         {
  66                 .name = "act-mask",
  67                 .has_arg = required_argument,
  68                 .flag = NULL,
  69                 .val = 'a'
  70         },
  71         {
  72                 .name = "set-mask",
  73                 .has_arg = required_argument,
  74                 .flag = NULL,
  75                 .val = 'A'
  76         },
  77         {
  78                 .name = "relay",
  79                 .has_arg = required_argument,
  80                 .flag = NULL,
  81                 .val = 'r'
  82         },
  83         {
  84                 .name = "output",
  85                 .has_arg = required_argument,
  86                 .flag = NULL,
  87                 .val = 'o'
  88         },
  89         {
  90                 .name = "kill",
  91                 .has_arg = no_argument,
  92                 .flag = NULL,
  93                 .val = 'k'
  94         },
  95         {
  96                 .name = "stopwatch",
  97                 .has_arg = required_argument,
  98                 .flag = NULL,
  99                 .val = 'w'
 100         },
 101         {
 102                 .name = "version",
 103                 .has_arg = no_argument,
 104                 .flag = NULL,
 105                 .val = 'V'
 106         },
 107         {
 108                 .name = "buffer-size",
 109                 .has_arg = required_argument,
 110                 .flag = NULL,
 111                 .val = 'b'
 112         },
 113         {
 114                 .name = "num-sub-buffers",
 115                 .has_arg = required_argument,
 116                 .flag = NULL,
 117                 .val = 'n'
 118         },
 119         {
 120                 .name = "output-dir",
 121                 .has_arg = required_argument,
 122                 .flag = NULL,
 123                 .val = 'D'
 124         },
 125         {
 126                 .name = NULL,
 127         }
 128 };
 129
 130 struct tip_subbuf {
 131         void *buf;
 132         unsigned int len;
 133         unsigned int max_len;
 134 };
 135
 136 #define FIFO_SIZE       (1024)  /* should be plenty big! */
 137 #define CL_SIZE         (128)   /* cache line, any bigger? */
 138
 139 struct tip_subbuf_fifo {
 140         int tail __attribute__((aligned(CL_SIZE)));
 141         int head __attribute__((aligned(CL_SIZE)));
 142         struct tip_subbuf *q[FIFO_SIZE];
 143 };
 144
 145 struct thread_information {
 146         int cpu;
 147         pthread_t thread;
 148
 149         int fd;
 150         void *fd_buf;
 151         char fn[MAXPATHLEN + 64];
 152
 153         FILE *ofile;
 154         char *ofile_buffer;
 155         int ofile_stdout;
 156
 157         unsigned long events_processed;
 158         unsigned long long data_read;
 159         struct device_information *device;
 160
 161         int exited;
 162
 163         /*
 164          * piped fifo buffers
 165          */
 166         struct tip_subbuf_fifo fifo;
 167         struct tip_subbuf *leftover_ts;
 168
 169         /*
 170          * mmap controlled output files
 171          */
 172         unsigned long long fs_size;
 173         unsigned long long fs_max_size;
 174         unsigned long fs_off;
 175         void *fs_buf;
 176         unsigned long fs_buf_len;
 177 };
 178
 179 struct device_information {
 180         int fd;
 181         char *path;
 182         char buts_name[32];
 183         volatile int trace_started;
 184         unsigned long drop_count;
 185         struct thread_information *threads;
 186 };
 187
 188 static int ncpus;
 189 static struct thread_information *thread_information;
 190 static int ndevs;
 191 static struct device_information *device_information;
 192
 193 /* command line option globals */
 194 static char *relay_path;
 195 static char *output_name;
 196 static char *output_dir;
 197 static int act_mask = ~0U;
 198 static int kill_running_trace;
 199 static unsigned long buf_size = BUF_SIZE;
 200 static unsigned long buf_nr = BUF_NR;
 201 static unsigned int page_size;
 202
 203 #define is_done()       (*(volatile int *)(&done))
 204 static volatile int done;
 205
 206 #define is_trace_stopped()      (*(volatile int *)(&trace_stopped))
 207 static volatile int trace_stopped;
 208
 209 #define is_stat_shown() (*(volatile int *)(&stat_shown))
 210 static volatile int stat_shown;
 211
 212 static void exit_trace(int status);
 213
 214 #define dip_tracing(dip)        (*(volatile int *)(&(dip)->trace_started))
 215 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
 216
 217 #define __for_each_dip(__d, __i, __e)   \
 218         for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
 219
 220 #define for_each_dip(__d, __i)  __for_each_dip(__d, __i, ndevs)
 221 #define for_each_tip(__d, __t, __j)     \
 222         for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
 223
 224 static int get_dropped_count(const char *buts_name)
 225 {
 226         int fd;
 227         char tmp[MAXPATHLEN + 64];
 228
 229         snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
 230                  relay_path, buts_name);
 231
 232         fd = open(tmp, O_RDONLY);
 233         if (fd < 0) {
 234                 /*
 235                  * this may be ok, if the kernel doesn't support dropped counts
 236                  */
 237                 if (errno == ENOENT)
 238                         return 0;
 239
 240                 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
 241                 return -1;
 242         }
 243
 244         if (read(fd, tmp, sizeof(tmp)) < 0) {
 245                 perror(tmp);
 246                 close(fd);
 247                 return -1;
 248         }
 249
 250         close(fd);
 251
 252         return atoi(tmp);
 253 }
 254
 255 static int start_trace(struct device_information *dip)
 256 {
 257         struct blk_user_trace_setup buts;
 258
 259         memset(&buts, 0, sizeof(buts));
 260         buts.buf_size = buf_size;
 261         buts.buf_nr = buf_nr;
 262         buts.act_mask = act_mask;
 263
 264         if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
 265                 perror("BLKTRACESETUP");
 266                 return 1;
 267         }
 268
 269         if (ioctl(dip->fd, BLKTRACESTART) < 0) {
 270                 perror("BLKTRACESTART");
 271                 return 1;
 272         }
 273
 274         memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
 275         dip_set_tracing(dip, 1);
 276         return 0;
 277 }
 278
 279 static void stop_trace(struct device_information *dip)
 280 {
 281         if (dip_tracing(dip) || kill_running_trace) {
 282                 dip_set_tracing(dip, 0);
 283
 284                 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
 285                         perror("BLKTRACESTOP");
 286                 if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
 287                         perror("BLKTRACETEARDOWN");
 288
 289                 close(dip->fd);
 290                 dip->fd = -1;
 291         }
 292 }
 293
 294 static void stop_all_traces(void)
 295 {
 296         struct device_information *dip;
 297         int i;
 298
 299         for_each_dip(dip, i) {
 300                 dip->drop_count = get_dropped_count(dip->buts_name);
 301                 stop_trace(dip);
 302         }
 303 }
 304
 305 static void wait_for_data(struct thread_information *tip)
 306 {
 307         struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
 308
 309         do {
 310                 poll(&pfd, 1, 100);
 311                 if (pfd.revents & POLLIN)
 312                         break;
 313                 if (tip->ofile_stdout)
 314                         break;
 315         } while (!is_done());
 316 }
 317
 318 static int read_data(struct thread_information *tip, void *buf, int len)
 319 {
 320         int ret = 0;
 321
 322         do {
 323                 wait_for_data(tip);
 324
 325                 ret = read(tip->fd, buf, len);
 326                 if (!ret)
 327                         continue;
 328                 else if (ret > 0)
 329                         return ret;
 330                 else {
 331                         if (errno != EAGAIN) {
 332                                 perror(tip->fn);
 333                                 fprintf(stderr,"Thread %d failed read of %s\n",
 334                                         tip->cpu, tip->fn);
 335                                 break;
 336                         }
 337                         continue;
 338                 }
 339         } while (!is_done());
 340
 341         return ret;
 342 }
 343
 344 static inline struct tip_subbuf *subbuf_fifo_dequeue(struct thread_information *tip)
 345 {
 346         const int head = tip->fifo.head;
 347         const int next = (head + 1) & (FIFO_SIZE - 1);
 348
 349         if (head != tip->fifo.tail) {
 350                 struct tip_subbuf *ts = tip->fifo.q[head];
 351
 352                 store_barrier();
 353                 tip->fifo.head = next;
 354                 return ts;
 355         }
 356
 357         return NULL;
 358 }
 359
 360 static inline int subbuf_fifo_queue(struct thread_information *tip,
 361                                     struct tip_subbuf *ts)
 362 {
 363         const int tail = tip->fifo.tail;
 364         const int next = (tail + 1) & (FIFO_SIZE - 1);
 365
 366         if (next != tip->fifo.head) {
 367                 tip->fifo.q[tail] = ts;
 368                 store_barrier();
 369                 tip->fifo.tail = next;
 370                 return 0;
 371         }
 372
 373         fprintf(stderr, "fifo too small!\n");
 374         return 1;
 375 }
 376
 377 /*
 378  * For file output, truncate and mmap the file appropriately
 379  */
 380 static int mmap_subbuf(struct thread_information *tip)
 381 {
 382         int ofd = fileno(tip->ofile);
 383         int ret;
 384
 385         /*
 386          * extend file, if we have to. use chunks of 16 subbuffers.
 387          */
 388         if (tip->fs_off + buf_size > tip->fs_buf_len) {
 389                 if (tip->fs_buf) {
 390                         munmap(tip->fs_buf, tip->fs_buf_len);
 391                         tip->fs_buf = NULL;
 392                 }
 393
 394                 tip->fs_off = tip->fs_size & (page_size - 1);
 395                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 396                 tip->fs_max_size += tip->fs_buf_len;
 397
 398                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 399                         perror("ftruncate");
 400                         return -1;
 401                 }
 402
 403                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 404                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 405                 if (tip->fs_buf == MAP_FAILED) {
 406                         perror("mmap");
 407                         return -1;
 408                 }
 409         }
 410
 411         ret = read_data(tip, tip->fs_buf + tip->fs_off, buf_size);
 412         if (ret >= 0) {
 413                 tip->data_read += ret;
 414                 tip->fs_size += ret;
 415                 tip->fs_off += ret;
 416                 return 0;
 417         }
 418
 419         return -1;
 420 }
 421
 422 /*
 423  * Use the copy approach for pipes
 424  */
 425 static int get_subbuf(struct thread_information *tip)
 426 {
 427         struct tip_subbuf *ts;
 428         int ret;
 429
 430         ts = malloc(sizeof(*ts));
 431         ts->buf = malloc(buf_size);
 432         ts->max_len = buf_size;
 433
 434         ret = read_data(tip, ts->buf, ts->max_len);
 435         if (ret > 0) {
 436                 ts->len = ret;
 437                 return subbuf_fifo_queue(tip, ts);
 438         }
 439
 440         free(ts->buf);
 441         free(ts);
 442         return ret;
 443 }
 444
 445 static void close_thread(struct thread_information *tip)
 446 {
 447         if (tip->fd != -1)
 448                 close(tip->fd);
 449         if (tip->ofile)
 450                 fclose(tip->ofile);
 451         if (tip->ofile_buffer)
 452                 free(tip->ofile_buffer);
 453         if (tip->fd_buf)
 454                 free(tip->fd_buf);
 455
 456         tip->fd = -1;
 457         tip->ofile = NULL;
 458         tip->ofile_buffer = NULL;
 459         tip->fd_buf = NULL;
 460 }
 461
 462 static void *thread_main(void *arg)
 463 {
 464         struct thread_information *tip = arg;
 465         pid_t pid = getpid();
 466         cpu_set_t cpu_mask;
 467
 468         CPU_ZERO(&cpu_mask);
 469         CPU_SET((tip->cpu), &cpu_mask);
 470
 471         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 472                 perror("sched_setaffinity");
 473                 exit_trace(1);
 474         }
 475
 476         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 477                         relay_path, tip->device->buts_name, tip->cpu);
 478         tip->fd = open(tip->fn, O_RDONLY);
 479         if (tip->fd < 0) {
 480                 perror(tip->fn);
 481                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 482                         tip->fn);
 483                 exit_trace(1);
 484         }
 485
 486         while (!is_done()) {
 487                 if (tip->ofile_stdout) {
 488                         if (get_subbuf(tip))
 489                                 break;
 490                 } else {
 491                         if (mmap_subbuf(tip))
 492                                 break;
 493                 }
 494         }
 495
 496         /*
 497          * truncate to right size and cleanup mmap
 498          */
 499         if (!tip->ofile_stdout) {
 500                 int ofd = fileno(tip->ofile);
 501
 502                 if (tip->fs_buf)
 503                         munmap(tip->fs_buf, tip->fs_buf_len);
 504
 505                 ftruncate(ofd, tip->fs_size);
 506         }
 507
 508         tip->exited = 1;
 509         return NULL;
 510 }
 511
 512 static int write_data(struct thread_information *tip,
 513                       void *buf, unsigned int buf_len)
 514 {
 515         int ret;
 516
 517         if (!buf_len)
 518                 return 0;
 519
 520         while (1) {
 521                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 522                 if (ret == 1)
 523                         break;
 524
 525                 if (ret < 0) {
 526                         perror("write");
 527                         return 1;
 528                 }
 529         }
 530
 531         if (tip->ofile_stdout)
 532                 fflush(tip->ofile);
 533
 534         return 0;
 535 }
 536
 537 static int flush_subbuf(struct thread_information *tip, struct tip_subbuf *ts)
 538 {
 539         unsigned int offset = 0;
 540         struct blk_io_trace *t;
 541         int pdu_len, events = 0;
 542
 543         /*
 544          * surplus from last run
 545          */
 546         if (tip->leftover_ts) {
 547                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 548
 549                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 550                         prev_ts->max_len += ts->len;
 551                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 552                 }
 553
 554                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 555                 prev_ts->len += ts->len;
 556
 557                 free(ts->buf);
 558                 free(ts);
 559
 560                 ts = prev_ts;
 561                 tip->leftover_ts = NULL;
 562         }
 563
 564         while (offset + sizeof(*t) <= ts->len) {
 565                 t = ts->buf + offset;
 566
 567                 if (verify_trace(t)) {
 568                         write_data(tip, ts->buf, offset);
 569                         return -1;
 570                 }
 571
 572                 pdu_len = t->pdu_len;
 573
 574                 if (offset + sizeof(*t) + pdu_len > ts->len)
 575                         break;
 576
 577                 offset += sizeof(*t) + pdu_len;
 578                 tip->events_processed++;
 579                 tip->data_read += sizeof(*t) + pdu_len;
 580                 events++;
 581         }
 582
 583         if (write_data(tip, ts->buf, offset))
 584                 return -1;
 585
 586         /*
 587          * leftover bytes, save them for next time
 588          */
 589         if (offset != ts->len) {
 590                 tip->leftover_ts = ts;
 591                 ts->len -= offset;
 592                 memmove(ts->buf, ts->buf + offset, ts->len);
 593         } else {
 594                 free(ts->buf);
 595                 free(ts);
 596         }
 597
 598         return events;
 599 }
 600
 601 static int write_tip_events(struct thread_information *tip)
 602 {
 603         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 604
 605         if (ts)
 606                 return flush_subbuf(tip, ts);
 607
 608         return 0;
 609 }
 610
 611 /*
 612  * scans the tips we know and writes out the subbuffers we accumulate
 613  */
 614 static void get_and_write_events(void)
 615 {
 616         struct device_information *dip;
 617         struct thread_information *tip;
 618         int i, j, events, ret, tips_running;
 619
 620         while (!is_done()) {
 621                 events = 0;
 622
 623                 for_each_dip(dip, i) {
 624                         for_each_tip(dip, tip, j) {
 625                                 ret = write_tip_events(tip);
 626                                 if (ret > 0)
 627                                         events += ret;
 628                         }
 629                 }
 630
 631                 if (!events)
 632                         usleep(10);
 633         }
 634
 635         /*
 636          * reap stored events
 637          */
 638         do {
 639                 events = 0;
 640                 tips_running = 0;
 641                 for_each_dip(dip, i) {
 642                         for_each_tip(dip, tip, j) {
 643                                 ret = write_tip_events(tip);
 644                                 if (ret > 0)
 645                                         events += ret;
 646                                 tips_running += !tip->exited;
 647                         }
 648                 }
 649                 usleep(10);
 650         } while (events || tips_running);
 651 }
 652
 653 static void wait_for_threads(void)
 654 {
 655         /*
 656          * for piped output, poll and fetch data for writeout. for files,
 657          * we just wait around for trace threads to exit
 658          */
 659         if (output_name && !strcmp(output_name, "-"))
 660                 get_and_write_events();
 661         else {
 662                 struct device_information *dip;
 663                 struct thread_information *tip;
 664                 int i, j, tips_running;
 665
 666                 do {
 667                         tips_running = 0;
 668                         usleep(1000);
 669
 670                         for_each_dip(dip, i)
 671                                 for_each_tip(dip, tip, j)
 672                                         tips_running += !tip->exited;
 673                 } while (tips_running);
 674         }
 675 }
 676
 677 static int start_threads(struct device_information *dip)
 678 {
 679         struct thread_information *tip;
 680         char op[64];
 681         int j, pipeline = output_name && !strcmp(output_name, "-");
 682         int len, mode, vbuf_size;
 683
 684         for_each_tip(dip, tip, j) {
 685                 tip->cpu = j;
 686                 tip->device = dip;
 687                 tip->events_processed = 0;
 688                 memset(&tip->fifo, 0, sizeof(tip->fifo));
 689                 tip->leftover_ts = NULL;
 690
 691                 if (pipeline) {
 692                         tip->ofile = fdopen(STDOUT_FILENO, "w");
 693                         tip->ofile_stdout = 1;
 694                         mode = _IOLBF;
 695                         vbuf_size = 512;
 696                 } else {
 697                         len = 0;
 698
 699                         if (output_dir)
 700                                 len = sprintf(op, "%s/", output_dir);
 701
 702                         if (output_name) {
 703                                 sprintf(op + len, "%s.blktrace.%d", output_name,
 704                                         tip->cpu);
 705                         } else {
 706                                 sprintf(op + len, "%s.blktrace.%d",
 707                                         dip->buts_name, tip->cpu);
 708                         }
 709                         tip->ofile = fopen(op, "w+");
 710                         tip->ofile_stdout = 0;
 711                         mode = _IOFBF;
 712                         vbuf_size = OFILE_BUF;
 713                 }
 714
 715                 if (tip->ofile == NULL) {
 716                         perror(op);
 717                         return 1;
 718                 }
 719
 720                 tip->ofile_buffer = malloc(vbuf_size);
 721                 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
 722                         perror("setvbuf");
 723                         close_thread(tip);
 724                         return 1;
 725                 }
 726
 727                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
 728                         perror("pthread_create");
 729                         close_thread(tip);
 730                         return 1;
 731                 }
 732         }
 733
 734         return 0;
 735 }
 736
 737 static void stop_threads(struct device_information *dip)
 738 {
 739         struct thread_information *tip;
 740         unsigned long ret;
 741         int i;
 742
 743         for_each_tip(dip, tip, i) {
 744                 (void) pthread_join(tip->thread, (void *) &ret);
 745                 close_thread(tip);
 746         }
 747 }
 748
 749 static void stop_all_threads(void)
 750 {
 751         struct device_information *dip;
 752         int i;
 753
 754         for_each_dip(dip, i)
 755                 stop_threads(dip);
 756 }
 757
 758 static void stop_all_tracing(void)
 759 {
 760         struct device_information *dip;
 761         int i;
 762
 763         for_each_dip(dip, i)
 764                 stop_trace(dip);
 765 }
 766
 767 static void exit_trace(int status)
 768 {
 769         if (!is_trace_stopped()) {
 770                 trace_stopped = 1;
 771                 stop_all_threads();
 772                 stop_all_tracing();
 773         }
 774
 775         exit(status);
 776 }
 777
 778 static int resize_devices(char *path)
 779 {
 780         int size = (ndevs + 1) * sizeof(struct device_information);
 781
 782         device_information = realloc(device_information, size);
 783         if (!device_information) {
 784                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
 785                 return 1;
 786         }
 787         device_information[ndevs].path = path;
 788         ndevs++;
 789         return 0;
 790 }
 791
 792 static int open_devices(void)
 793 {
 794         struct device_information *dip;
 795         int i;
 796
 797         for_each_dip(dip, i) {
 798                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
 799                 if (dip->fd < 0) {
 800                         perror(dip->path);
 801                         return 1;
 802                 }
 803         }
 804
 805         return 0;
 806 }
 807
 808 static int start_devices(void)
 809 {
 810         struct device_information *dip;
 811         int i, j, size;
 812
 813         size = ncpus * sizeof(struct thread_information);
 814         thread_information = malloc(size * ndevs);
 815         if (!thread_information) {
 816                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
 817                 return 1;
 818         }
 819
 820         for_each_dip(dip, i) {
 821                 if (start_trace(dip)) {
 822                         close(dip->fd);
 823                         fprintf(stderr, "Failed to start trace on %s\n",
 824                                 dip->path);
 825                         break;
 826                 }
 827         }
 828
 829         if (i != ndevs) {
 830                 __for_each_dip(dip, j, i)
 831                         stop_trace(dip);
 832
 833                 return 1;
 834         }
 835
 836         for_each_dip(dip, i) {
 837                 dip->threads = thread_information + (i * ncpus);
 838                 if (start_threads(dip)) {
 839                         fprintf(stderr, "Failed to start worker threads\n");
 840                         break;
 841                 }
 842         }
 843
 844         if (i != ndevs) {
 845                 __for_each_dip(dip, j, i)
 846                         stop_threads(dip);
 847                 for_each_dip(dip, i)
 848                         stop_trace(dip);
 849
 850                 return 1;
 851         }
 852
 853         return 0;
 854 }
 855
 856 static void show_stats(void)
 857 {
 858         struct device_information *dip;
 859         struct thread_information *tip;
 860         unsigned long long events_processed, data_read;
 861         unsigned long total_drops;
 862         int i, j, no_stdout = 0;
 863
 864         if (is_stat_shown())
 865                 return;
 866
 867         if (output_name && !strcmp(output_name, "-"))
 868                 no_stdout = 1;
 869
 870         stat_shown = 1;
 871
 872         total_drops = 0;
 873         for_each_dip(dip, i) {
 874                 if (!no_stdout)
 875                         printf("Device: %s\n", dip->path);
 876                 events_processed = 0;
 877                 data_read = 0;
 878                 for_each_tip(dip, tip, j) {
 879                         if (!no_stdout)
 880                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
 881                                         tip->cpu, tip->events_processed,
 882                                         tip->data_read >> 10);
 883                         events_processed += tip->events_processed;
 884                         data_read += tip->data_read;
 885                 }
 886                 total_drops += dip->drop_count;
 887                 if (!no_stdout)
 888                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
 889                                         events_processed, dip->drop_count,
 890                                         data_read >> 10);
 891         }
 892
 893         if (total_drops)
 894                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
 895 }
 896
 897 static char usage_str[] = \
 898         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
 899         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
 900         "\t-d Use specified device. May also be given last after options\n" \
 901         "\t-r Path to mounted relayfs, defaults to /relay\n" \
 902         "\t-o File(s) to send output to\n" \
 903         "\t-D Directory to prepend to output file names\n" \
 904         "\t-k Kill a running trace\n" \
 905         "\t-w Stop after defined time, in seconds\n" \
 906         "\t-a Only trace specified actions. See documentation\n" \
 907         "\t-A Give trace mask as a single value. See documentation\n" \
 908         "\t-b Sub buffer size in KiB\n" \
 909         "\t-n Number of sub buffers\n" \
 910         "\t-v Print program version info\n\n";
 911
 912 static void show_usage(char *program)
 913 {
 914         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
 915 }
 916 static void handle_sigint(__attribute__((__unused__)) int sig)
 917 {
 918         done = 1;
 919 }
 920
 921 int main(int argc, char *argv[])
 922 {
 923         static char default_relay_path[] = "/relay";
 924         struct statfs st;
 925         int i, c;
 926         int stop_watch = 0;
 927         int act_mask_tmp = 0;
 928
 929         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
 930                 switch (c) {
 931                 case 'a':
 932                         i = find_mask_map(optarg);
 933                         if (i < 0) {
 934                                 fprintf(stderr,"Invalid action mask %s\n",
 935                                         optarg);
 936                                 return 1;
 937                         }
 938                         act_mask_tmp |= i;
 939                         break;
 940
 941                 case 'A':
 942                         if ((sscanf(optarg, "%x", &i) != 1) ||
 943                                                         !valid_act_opt(i)) {
 944                                 fprintf(stderr,
 945                                         "Invalid set action mask %s/0x%x\n",
 946                                         optarg, i);
 947                                 return 1;
 948                         }
 949                         act_mask_tmp = i;
 950                         break;
 951
 952                 case 'd':
 953                         if (resize_devices(optarg) != 0)
 954                                 return 1;
 955                         break;
 956
 957                 case 'r':
 958                         relay_path = optarg;
 959                         break;
 960
 961                 case 'o':
 962                         output_name = optarg;
 963                         break;
 964                 case 'k':
 965                         kill_running_trace = 1;
 966                         break;
 967                 case 'w':
 968                         stop_watch = atoi(optarg);
 969                         if (stop_watch <= 0) {
 970                                 fprintf(stderr,
 971                                         "Invalid stopwatch value (%d secs)\n",
 972                                         stop_watch);
 973                                 return 1;
 974                         }
 975                         break;
 976                 case 'V':
 977                         printf("%s version %s\n", argv[0], blktrace_version);
 978                         return 0;
 979                 case 'b':
 980                         buf_size = strtoul(optarg, NULL, 10);
 981                         if (buf_size <= 0 || buf_size > 16*1024) {
 982                                 fprintf(stderr,
 983                                         "Invalid buffer size (%lu)\n",buf_size);
 984                                 return 1;
 985                         }
 986                         buf_size <<= 10;
 987                         break;
 988                 case 'n':
 989                         buf_nr = strtoul(optarg, NULL, 10);
 990                         if (buf_nr <= 0) {
 991                                 fprintf(stderr,
 992                                         "Invalid buffer nr (%lu)\n", buf_nr);
 993                                 return 1;
 994                         }
 995                         break;
 996                 case 'D':
 997                         output_dir = optarg;
 998                         break;
 999                 default:
1000                         show_usage(argv[0]);
1001                         return 1;
1002                 }
1003         }
1004
1005         while (optind < argc) {
1006                 if (resize_devices(argv[optind++]) != 0)
1007                         return 1;
1008         }
1009
1010         if (ndevs == 0) {
1011                 show_usage(argv[0]);
1012                 return 1;
1013         }
1014
1015         if (!relay_path)
1016                 relay_path = default_relay_path;
1017
1018         if (act_mask_tmp != 0)
1019                 act_mask = act_mask_tmp;
1020
1021         if (statfs(relay_path, &st) < 0) {
1022                 perror("statfs");
1023                 fprintf(stderr,"%s does not appear to be a valid path\n",
1024                         relay_path);
1025                 return 1;
1026         } else if (st.f_type != (long) RELAYFS_TYPE) {
1027                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1028                         relay_path);
1029                 return 1;
1030         }
1031
1032         if (open_devices() != 0)
1033                 return 1;
1034
1035         if (kill_running_trace) {
1036                 stop_all_traces();
1037                 return 0;
1038         }
1039
1040         setlocale(LC_NUMERIC, "en_US");
1041
1042         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1043         if (ncpus < 0) {
1044                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1045                 return 1;
1046         }
1047
1048         page_size = getpagesize();
1049
1050         if (start_devices() != 0)
1051                 return 1;
1052
1053         signal(SIGINT, handle_sigint);
1054         signal(SIGHUP, handle_sigint);
1055         signal(SIGTERM, handle_sigint);
1056         signal(SIGALRM, handle_sigint);
1057
1058         atexit(stop_all_tracing);
1059
1060         if (stop_watch)
1061                 alarm(stop_watch);
1062
1063         wait_for_threads();
1064
1065         if (!is_trace_stopped()) {
1066                 trace_stopped = 1;
1067                 stop_all_threads();
1068                 stop_all_traces();
1069         }
1070
1071         show_stats();
1072
1073         return 0;
1074 }
1075