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 RING_INIT_NR    (2)
  58 #define RING_MAX_NR     (16UL)
  59
  60 #define S_OPTS  "d:a:A:r:o:kw:Vb:n:D:"
  61 static struct option l_opts[] = {
  62         {
  63                 .name = "dev",
  64                 .has_arg = required_argument,
  65                 .flag = NULL,
  66                 .val = 'd'
  67         },
  68         {
  69                 .name = "act-mask",
  70                 .has_arg = required_argument,
  71                 .flag = NULL,
  72                 .val = 'a'
  73         },
  74         {
  75                 .name = "set-mask",
  76                 .has_arg = required_argument,
  77                 .flag = NULL,
  78                 .val = 'A'
  79         },
  80         {
  81                 .name = "relay",
  82                 .has_arg = required_argument,
  83                 .flag = NULL,
  84                 .val = 'r'
  85         },
  86         {
  87                 .name = "output",
  88                 .has_arg = required_argument,
  89                 .flag = NULL,
  90                 .val = 'o'
  91         },
  92         {
  93                 .name = "kill",
  94                 .has_arg = no_argument,
  95                 .flag = NULL,
  96                 .val = 'k'
  97         },
  98         {
  99                 .name = "stopwatch",
 100                 .has_arg = required_argument,
 101                 .flag = NULL,
 102                 .val = 'w'
 103         },
 104         {
 105                 .name = "version",
 106                 .has_arg = no_argument,
 107                 .flag = NULL,
 108                 .val = 'V'
 109         },
 110         {
 111                 .name = "buffer-size",
 112                 .has_arg = required_argument,
 113                 .flag = NULL,
 114                 .val = 'b'
 115         },
 116         {
 117                 .name = "num-sub-buffers",
 118                 .has_arg = required_argument,
 119                 .flag = NULL,
 120                 .val = 'n'
 121         },
 122         {
 123                 .name = "output-dir",
 124                 .has_arg = required_argument,
 125                 .flag = NULL,
 126                 .val = 'D'
 127         },
 128         {
 129                 .name = NULL,
 130         }
 131 };
 132
 133 struct tip_subbuf {
 134         void *buf;
 135         unsigned int len;
 136         unsigned int max_len;
 137 };
 138
 139 #define FIFO_SIZE       (1024)  /* should be plenty big! */
 140 #define CL_SIZE         (128)   /* cache line, any bigger? */
 141
 142 struct tip_subbuf_fifo {
 143         int tail __attribute__((aligned(CL_SIZE)));
 144         int head __attribute__((aligned(CL_SIZE)));
 145         struct tip_subbuf *q[FIFO_SIZE];
 146 };
 147
 148 struct thread_information {
 149         int cpu;
 150         pthread_t thread;
 151
 152         int fd;
 153         void *fd_buf;
 154         char fn[MAXPATHLEN + 64];
 155
 156         FILE *ofile;
 157         char *ofile_buffer;
 158         int ofile_stdout;
 159
 160         unsigned long events_processed;
 161         unsigned long long data_read;
 162         struct device_information *device;
 163
 164         int exited;
 165
 166         /*
 167          * piped fifo buffers
 168          */
 169         struct tip_subbuf_fifo fifo;
 170         struct tip_subbuf *leftover_ts;
 171
 172         /*
 173          * mmap controlled output files
 174          */
 175         unsigned long long fs_size;
 176         unsigned long long fs_max_size;
 177         unsigned long fs_off;
 178         void *fs_buf;
 179         unsigned long fs_buf_len;
 180 };
 181
 182 struct device_information {
 183         int fd;
 184         char *path;
 185         char buts_name[32];
 186         volatile int trace_started;
 187         unsigned long drop_count;
 188         struct thread_information *threads;
 189 };
 190
 191 static int ncpus;
 192 static struct thread_information *thread_information;
 193 static int ndevs;
 194 static struct device_information *device_information;
 195
 196 /* command line option globals */
 197 static char *relay_path;
 198 static char *output_name;
 199 static char *output_dir;
 200 static int act_mask = ~0U;
 201 static int kill_running_trace;
 202 static unsigned long buf_size = BUF_SIZE;
 203 static unsigned long buf_nr = BUF_NR;
 204 static unsigned int page_size;
 205
 206 #define is_done()       (*(volatile int *)(&done))
 207 static volatile int done;
 208
 209 #define is_trace_stopped()      (*(volatile int *)(&trace_stopped))
 210 static volatile int trace_stopped;
 211
 212 #define is_stat_shown() (*(volatile int *)(&stat_shown))
 213 static volatile int stat_shown;
 214
 215 static void exit_trace(int status);
 216
 217 #define dip_tracing(dip)        (*(volatile int *)(&(dip)->trace_started))
 218 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
 219
 220 #define __for_each_dip(__d, __i, __e)   \
 221         for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
 222
 223 #define for_each_dip(__d, __i)  __for_each_dip(__d, __i, ndevs)
 224 #define for_each_tip(__d, __t, __j)     \
 225         for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
 226
 227 static int get_dropped_count(const char *buts_name)
 228 {
 229         int fd;
 230         char tmp[MAXPATHLEN + 64];
 231
 232         snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
 233                  relay_path, buts_name);
 234
 235         fd = open(tmp, O_RDONLY);
 236         if (fd < 0) {
 237                 /*
 238                  * this may be ok, if the kernel doesn't support dropped counts
 239                  */
 240                 if (errno == ENOENT)
 241                         return 0;
 242
 243                 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
 244                 return -1;
 245         }
 246
 247         if (read(fd, tmp, sizeof(tmp)) < 0) {
 248                 perror(tmp);
 249                 close(fd);
 250                 return -1;
 251         }
 252
 253         close(fd);
 254
 255         return atoi(tmp);
 256 }
 257
 258 static int start_trace(struct device_information *dip)
 259 {
 260         struct blk_user_trace_setup buts;
 261
 262         memset(&buts, 0, sizeof(buts));
 263         buts.buf_size = buf_size;
 264         buts.buf_nr = buf_nr;
 265         buts.act_mask = act_mask;
 266
 267         if (ioctl(dip->fd, BLKSTARTTRACE, &buts) < 0) {
 268                 perror("BLKSTARTTRACE");
 269                 return 1;
 270         }
 271
 272         memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
 273         dip_set_tracing(dip, 1);
 274         return 0;
 275 }
 276
 277 static void stop_trace(struct device_information *dip)
 278 {
 279         if (dip_tracing(dip) || kill_running_trace) {
 280                 dip_set_tracing(dip, 0);
 281
 282                 if (ioctl(dip->fd, BLKSTOPTRACE) < 0)
 283                         perror("BLKSTOPTRACE");
 284
 285                 close(dip->fd);
 286                 dip->fd = -1;
 287         }
 288 }
 289
 290 static void stop_all_traces(void)
 291 {
 292         struct device_information *dip;
 293         int i;
 294
 295         for_each_dip(dip, i) {
 296                 dip->drop_count = get_dropped_count(dip->buts_name);
 297                 stop_trace(dip);
 298         }
 299 }
 300
 301 static void wait_for_data(struct thread_information *tip)
 302 {
 303         struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
 304
 305         do {
 306                 poll(&pfd, 1, 100);
 307                 if (pfd.revents & POLLIN)
 308                         break;
 309                 if (tip->ofile_stdout)
 310                         break;
 311         } while (!is_done());
 312 }
 313
 314 static int read_data(struct thread_information *tip, void *buf, int len)
 315 {
 316         int ret = 0;
 317
 318         do {
 319                 wait_for_data(tip);
 320
 321                 ret = read(tip->fd, buf, len);
 322                 if (!ret)
 323                         continue;
 324                 else if (ret > 0)
 325                         return ret;
 326                 else {
 327                         if (errno != EAGAIN) {
 328                                 perror(tip->fn);
 329                                 fprintf(stderr,"Thread %d failed read of %s\n",
 330                                         tip->cpu, tip->fn);
 331                                 break;
 332                         }
 333                         continue;
 334                 }
 335         } while (!is_done());
 336
 337         return ret;
 338 }
 339
 340 static inline struct tip_subbuf *subbuf_fifo_dequeue(struct thread_information *tip)
 341 {
 342         const int head = tip->fifo.head;
 343         const int next = (head + 1) & (FIFO_SIZE - 1);
 344
 345         if (head != tip->fifo.tail) {
 346                 struct tip_subbuf *ts = tip->fifo.q[head];
 347
 348                 store_barrier();
 349                 tip->fifo.head = next;
 350                 return ts;
 351         }
 352
 353         return NULL;
 354 }
 355
 356 static inline int subbuf_fifo_queue(struct thread_information *tip,
 357                                     struct tip_subbuf *ts)
 358 {
 359         const int tail = tip->fifo.tail;
 360         const int next = (tail + 1) & (FIFO_SIZE - 1);
 361
 362         if (next != tip->fifo.head) {
 363                 tip->fifo.q[tail] = ts;
 364                 store_barrier();
 365                 tip->fifo.tail = next;
 366                 return 0;
 367         }
 368
 369         fprintf(stderr, "fifo too small!\n");
 370         return 1;
 371 }
 372
 373 /*
 374  * For file output, truncate and mmap the file appropriately
 375  */
 376 static int mmap_subbuf(struct thread_information *tip)
 377 {
 378         int ofd = fileno(tip->ofile);
 379         int ret;
 380
 381         /*
 382          * extend file, if we have to. use chunks of 16 subbuffers.
 383          */
 384         if (tip->fs_off + buf_size > tip->fs_buf_len) {
 385                 if (tip->fs_buf) {
 386                         munmap(tip->fs_buf, tip->fs_buf_len);
 387                         tip->fs_buf = NULL;
 388                 }
 389
 390                 tip->fs_off = tip->fs_size & (page_size - 1);
 391                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 392                 tip->fs_max_size += tip->fs_buf_len;
 393
 394                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 395                         perror("ftruncate");
 396                         return -1;
 397                 }
 398
 399                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 400                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 401                 if (tip->fs_buf == MAP_FAILED) {
 402                         perror("mmap");
 403                         return -1;
 404                 }
 405         }
 406
 407         ret = read_data(tip, tip->fs_buf + tip->fs_off, buf_size);
 408         if (ret >= 0) {
 409                 tip->data_read += ret;
 410                 tip->fs_size += ret;
 411                 tip->fs_off += ret;
 412                 return 0;
 413         }
 414
 415         return -1;
 416 }
 417
 418 /*
 419  * Use the copy approach for pipes
 420  */
 421 static int get_subbuf(struct thread_information *tip)
 422 {
 423         struct tip_subbuf *ts;
 424         int ret;
 425
 426         ts = malloc(sizeof(*ts));
 427         ts->buf = malloc(buf_size);
 428         ts->max_len = buf_size;
 429
 430         ret = read_data(tip, ts->buf, ts->max_len);
 431         if (ret > 0) {
 432                 ts->len = ret;
 433                 return subbuf_fifo_queue(tip, ts);
 434         }
 435
 436         free(ts->buf);
 437         free(ts);
 438         return ret;
 439 }
 440
 441 static void close_thread(struct thread_information *tip)
 442 {
 443         if (tip->fd != -1)
 444                 close(tip->fd);
 445         if (tip->ofile)
 446                 fclose(tip->ofile);
 447         if (tip->ofile_buffer)
 448                 free(tip->ofile_buffer);
 449         if (tip->fd_buf)
 450                 free(tip->fd_buf);
 451
 452         tip->fd = -1;
 453         tip->ofile = NULL;
 454         tip->ofile_buffer = NULL;
 455         tip->fd_buf = NULL;
 456 }
 457
 458 static void *thread_main(void *arg)
 459 {
 460         struct thread_information *tip = arg;
 461         pid_t pid = getpid();
 462         cpu_set_t cpu_mask;
 463
 464         CPU_ZERO(&cpu_mask);
 465         CPU_SET((tip->cpu), &cpu_mask);
 466
 467         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 468                 perror("sched_setaffinity");
 469                 exit_trace(1);
 470         }
 471
 472         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 473                         relay_path, tip->device->buts_name, tip->cpu);
 474         tip->fd = open(tip->fn, O_RDONLY);
 475         if (tip->fd < 0) {
 476                 perror(tip->fn);
 477                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 478                         tip->fn);
 479                 exit_trace(1);
 480         }
 481
 482         while (!is_done()) {
 483                 if (tip->ofile_stdout) {
 484                         if (get_subbuf(tip))
 485                                 break;
 486                 } else {
 487                         if (mmap_subbuf(tip))
 488                                 break;
 489                 }
 490         }
 491
 492         /*
 493          * truncate to right size and cleanup mmap
 494          */
 495         if (!tip->ofile_stdout) {
 496                 int ofd = fileno(tip->ofile);
 497
 498                 if (tip->fs_buf)
 499                         munmap(tip->fs_buf, tip->fs_buf_len);
 500
 501                 ftruncate(ofd, tip->fs_size);
 502         }
 503
 504         tip->exited = 1;
 505         return NULL;
 506 }
 507
 508 static int write_data(struct thread_information *tip,
 509                       void *buf, unsigned int buf_len)
 510 {
 511         int ret;
 512
 513         if (!buf_len)
 514                 return 0;
 515
 516         while (1) {
 517                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 518                 if (ret == 1)
 519                         break;
 520
 521                 if (ret < 0) {
 522                         perror("write");
 523                         return 1;
 524                 }
 525         }
 526
 527         if (tip->ofile_stdout)
 528                 fflush(tip->ofile);
 529
 530         return 0;
 531 }
 532
 533 static int flush_subbuf(struct thread_information *tip, struct tip_subbuf *ts)
 534 {
 535         unsigned int offset = 0;
 536         struct blk_io_trace *t;
 537         int pdu_len, events = 0;
 538
 539         /*
 540          * surplus from last run
 541          */
 542         if (tip->leftover_ts) {
 543                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 544
 545                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 546                         prev_ts->max_len += ts->len;
 547                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 548                 }
 549
 550                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 551                 prev_ts->len += ts->len;
 552
 553                 free(ts->buf);
 554                 free(ts);
 555
 556                 ts = prev_ts;
 557                 tip->leftover_ts = NULL;
 558         }
 559
 560         while (offset + sizeof(*t) <= ts->len) {
 561                 t = ts->buf + offset;
 562
 563                 if (verify_trace(t)) {
 564                         write_data(tip, ts->buf, offset);
 565                         return -1;
 566                 }
 567
 568                 pdu_len = t->pdu_len;
 569
 570                 if (offset + sizeof(*t) + pdu_len > ts->len)
 571                         break;
 572
 573                 offset += sizeof(*t) + pdu_len;
 574                 tip->events_processed++;
 575                 tip->data_read += sizeof(*t) + pdu_len;
 576                 events++;
 577         }
 578
 579         if (write_data(tip, ts->buf, offset))
 580                 return -1;
 581
 582         /*
 583          * leftover bytes, save them for next time
 584          */
 585         if (offset != ts->len) {
 586                 tip->leftover_ts = ts;
 587                 ts->len -= offset;
 588                 memmove(ts->buf, ts->buf + offset, ts->len);
 589         } else {
 590                 free(ts->buf);
 591                 free(ts);
 592         }
 593
 594         return events;
 595 }
 596
 597 static int write_tip_events(struct thread_information *tip)
 598 {
 599         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 600
 601         if (ts)
 602                 return flush_subbuf(tip, ts);
 603
 604         return 0;
 605 }
 606
 607 /*
 608  * scans the tips we know and writes out the subbuffers we accumulate
 609  */
 610 static void get_and_write_events(void)
 611 {
 612         struct device_information *dip;
 613         struct thread_information *tip;
 614         int i, j, events, ret, tips_running;
 615
 616         while (!is_done()) {
 617                 events = 0;
 618
 619                 for_each_dip(dip, i) {
 620                         for_each_tip(dip, tip, j) {
 621                                 ret = write_tip_events(tip);
 622                                 if (ret > 0)
 623                                         events += ret;
 624                         }
 625                 }
 626
 627                 if (!events)
 628                         usleep(10);
 629         }
 630
 631         /*
 632          * reap stored events
 633          */
 634         do {
 635                 events = 0;
 636                 tips_running = 0;
 637                 for_each_dip(dip, i) {
 638                         for_each_tip(dip, tip, j) {
 639                                 ret = write_tip_events(tip);
 640                                 if (ret > 0)
 641                                         events += ret;
 642                                 tips_running += !tip->exited;
 643                         }
 644                 }
 645                 usleep(10);
 646         } while (events || tips_running);
 647 }
 648
 649 static void wait_for_threads(void)
 650 {
 651         /*
 652          * for piped output, poll and fetch data for writeout. for files,
 653          * we just wait around for trace threads to exit
 654          */
 655         if (output_name && !strcmp(output_name, "-"))
 656                 get_and_write_events();
 657         else {
 658                 struct device_information *dip;
 659                 struct thread_information *tip;
 660                 int i, j, tips_running;
 661
 662                 do {
 663                         tips_running = 0;
 664                         usleep(1000);
 665
 666                         for_each_dip(dip, i)
 667                                 for_each_tip(dip, tip, j)
 668                                         tips_running += !tip->exited;
 669                 } while (tips_running);
 670         }
 671 }
 672
 673 static int start_threads(struct device_information *dip)
 674 {
 675         struct thread_information *tip;
 676         char op[64];
 677         int j, pipeline = output_name && !strcmp(output_name, "-");
 678         int len, mode, vbuf_size;
 679
 680         for_each_tip(dip, tip, j) {
 681                 tip->cpu = j;
 682                 tip->device = dip;
 683                 tip->events_processed = 0;
 684                 memset(&tip->fifo, 0, sizeof(tip->fifo));
 685                 tip->leftover_ts = NULL;
 686
 687                 if (pipeline) {
 688                         tip->ofile = fdopen(STDOUT_FILENO, "w");
 689                         tip->ofile_stdout = 1;
 690                         mode = _IOLBF;
 691                         vbuf_size = 512;
 692                 } else {
 693                         len = 0;
 694
 695                         if (output_dir)
 696                                 len = sprintf(op, "%s/", output_dir);
 697
 698                         if (output_name) {
 699                                 sprintf(op + len, "%s.blktrace.%d", output_name,
 700                                         tip->cpu);
 701                         } else {
 702                                 sprintf(op + len, "%s.blktrace.%d",
 703                                         dip->buts_name, tip->cpu);
 704                         }
 705                         tip->ofile = fopen(op, "w+");
 706                         tip->ofile_stdout = 0;
 707                         mode = _IOFBF;
 708                         vbuf_size = OFILE_BUF;
 709                 }
 710
 711                 if (tip->ofile == NULL) {
 712                         perror(op);
 713                         return 1;
 714                 }
 715
 716                 tip->ofile_buffer = malloc(vbuf_size);
 717                 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
 718                         perror("setvbuf");
 719                         close_thread(tip);
 720                         return 1;
 721                 }
 722
 723                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
 724                         perror("pthread_create");
 725                         close_thread(tip);
 726                         return 1;
 727                 }
 728         }
 729
 730         return 0;
 731 }
 732
 733 static void stop_threads(struct device_information *dip)
 734 {
 735         struct thread_information *tip;
 736         unsigned long ret;
 737         int i;
 738
 739         for_each_tip(dip, tip, i) {
 740                 (void) pthread_join(tip->thread, (void *) &ret);
 741                 close_thread(tip);
 742         }
 743 }
 744
 745 static void stop_all_threads(void)
 746 {
 747         struct device_information *dip;
 748         int i;
 749
 750         for_each_dip(dip, i)
 751                 stop_threads(dip);
 752 }
 753
 754 static void stop_all_tracing(void)
 755 {
 756         struct device_information *dip;
 757         int i;
 758
 759         for_each_dip(dip, i)
 760                 stop_trace(dip);
 761 }
 762
 763 static void exit_trace(int status)
 764 {
 765         if (!is_trace_stopped()) {
 766                 trace_stopped = 1;
 767                 stop_all_threads();
 768                 stop_all_tracing();
 769         }
 770
 771         exit(status);
 772 }
 773
 774 static int resize_devices(char *path)
 775 {
 776         int size = (ndevs + 1) * sizeof(struct device_information);
 777
 778         device_information = realloc(device_information, size);
 779         if (!device_information) {
 780                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
 781                 return 1;
 782         }
 783         device_information[ndevs].path = path;
 784         ndevs++;
 785         return 0;
 786 }
 787
 788 static int open_devices(void)
 789 {
 790         struct device_information *dip;
 791         int i;
 792
 793         for_each_dip(dip, i) {
 794                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
 795                 if (dip->fd < 0) {
 796                         perror(dip->path);
 797                         return 1;
 798                 }
 799         }
 800
 801         return 0;
 802 }
 803
 804 static int start_devices(void)
 805 {
 806         struct device_information *dip;
 807         int i, j, size;
 808
 809         size = ncpus * sizeof(struct thread_information);
 810         thread_information = malloc(size * ndevs);
 811         if (!thread_information) {
 812                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
 813                 return 1;
 814         }
 815
 816         for_each_dip(dip, i) {
 817                 if (start_trace(dip)) {
 818                         close(dip->fd);
 819                         fprintf(stderr, "Failed to start trace on %s\n",
 820                                 dip->path);
 821                         break;
 822                 }
 823         }
 824
 825         if (i != ndevs) {
 826                 __for_each_dip(dip, j, i)
 827                         stop_trace(dip);
 828
 829                 return 1;
 830         }
 831
 832         for_each_dip(dip, i) {
 833                 dip->threads = thread_information + (i * ncpus);
 834                 if (start_threads(dip)) {
 835                         fprintf(stderr, "Failed to start worker threads\n");
 836                         break;
 837                 }
 838         }
 839
 840         if (i != ndevs) {
 841                 __for_each_dip(dip, j, i)
 842                         stop_threads(dip);
 843                 for_each_dip(dip, i)
 844                         stop_trace(dip);
 845
 846                 return 1;
 847         }
 848
 849         return 0;
 850 }
 851
 852 static void show_stats(void)
 853 {
 854         struct device_information *dip;
 855         struct thread_information *tip;
 856         unsigned long long events_processed, data_read;
 857         unsigned long total_drops;
 858         int i, j, no_stdout = 0;
 859
 860         if (is_stat_shown())
 861                 return;
 862
 863         if (output_name && !strcmp(output_name, "-"))
 864                 no_stdout = 1;
 865
 866         stat_shown = 1;
 867
 868         total_drops = 0;
 869         for_each_dip(dip, i) {
 870                 if (!no_stdout)
 871                         printf("Device: %s\n", dip->path);
 872                 events_processed = 0;
 873                 data_read = 0;
 874                 for_each_tip(dip, tip, j) {
 875                         if (!no_stdout)
 876                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
 877                                         tip->cpu, tip->events_processed,
 878                                         tip->data_read >> 10);
 879                         events_processed += tip->events_processed;
 880                         data_read += tip->data_read;
 881                 }
 882                 total_drops += dip->drop_count;
 883                 if (!no_stdout)
 884                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
 885                                         events_processed, dip->drop_count,
 886                                         data_read >> 10);
 887         }
 888
 889         if (total_drops)
 890                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
 891 }
 892
 893 static char usage_str[] = \
 894         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
 895         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
 896         "\t-d Use specified device. May also be given last after options\n" \
 897         "\t-r Path to mounted relayfs, defaults to /relay\n" \
 898         "\t-o File(s) to send output to\n" \
 899         "\t-D Directory to prepend to output file names\n" \
 900         "\t-k Kill a running trace\n" \
 901         "\t-w Stop after defined time, in seconds\n" \
 902         "\t-a Only trace specified actions. See documentation\n" \
 903         "\t-A Give trace mask as a single value. See documentation\n" \
 904         "\t-b Sub buffer size in KiB\n" \
 905         "\t-n Number of sub buffers\n" \
 906         "\t-v Print program version info\n\n";
 907
 908 static void show_usage(char *program)
 909 {
 910         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
 911 }
 912 static void handle_sigint(__attribute__((__unused__)) int sig)
 913 {
 914         done = 1;
 915 }
 916
 917 int main(int argc, char *argv[])
 918 {
 919         static char default_relay_path[] = "/relay";
 920         struct statfs st;
 921         int i, c;
 922         int stop_watch = 0;
 923         int act_mask_tmp = 0;
 924
 925         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
 926                 switch (c) {
 927                 case 'a':
 928                         i = find_mask_map(optarg);
 929                         if (i < 0) {
 930                                 fprintf(stderr,"Invalid action mask %s\n",
 931                                         optarg);
 932                                 return 1;
 933                         }
 934                         act_mask_tmp |= i;
 935                         break;
 936
 937                 case 'A':
 938                         if ((sscanf(optarg, "%x", &i) != 1) ||
 939                                                         !valid_act_opt(i)) {
 940                                 fprintf(stderr,
 941                                         "Invalid set action mask %s/0x%x\n",
 942                                         optarg, i);
 943                                 return 1;
 944                         }
 945                         act_mask_tmp = i;
 946                         break;
 947
 948                 case 'd':
 949                         if (resize_devices(optarg) != 0)
 950                                 return 1;
 951                         break;
 952
 953                 case 'r':
 954                         relay_path = optarg;
 955                         break;
 956
 957                 case 'o':
 958                         output_name = optarg;
 959                         break;
 960                 case 'k':
 961                         kill_running_trace = 1;
 962                         break;
 963                 case 'w':
 964                         stop_watch = atoi(optarg);
 965                         if (stop_watch <= 0) {
 966                                 fprintf(stderr,
 967                                         "Invalid stopwatch value (%d secs)\n",
 968                                         stop_watch);
 969                                 return 1;
 970                         }
 971                         break;
 972                 case 'V':
 973                         printf("%s version %s\n", argv[0], blktrace_version);
 974                         return 0;
 975                 case 'b':
 976                         buf_size = strtoul(optarg, NULL, 10);
 977                         if (buf_size <= 0 || buf_size > 16*1024) {
 978                                 fprintf(stderr,
 979                                         "Invalid buffer size (%lu)\n",buf_size);
 980                                 return 1;
 981                         }
 982                         buf_size <<= 10;
 983                         break;
 984                 case 'n':
 985                         buf_nr = strtoul(optarg, NULL, 10);
 986                         if (buf_nr <= 0) {
 987                                 fprintf(stderr,
 988                                         "Invalid buffer nr (%lu)\n", buf_nr);
 989                                 return 1;
 990                         }
 991                         break;
 992                 case 'D':
 993                         output_dir = optarg;
 994                         break;
 995                 default:
 996                         show_usage(argv[0]);
 997                         return 1;
 998                 }
 999         }
1000
1001         while (optind < argc) {
1002                 if (resize_devices(argv[optind++]) != 0)
1003                         return 1;
1004         }
1005
1006         if (ndevs == 0) {
1007                 show_usage(argv[0]);
1008                 return 1;
1009         }
1010
1011         if (!relay_path)
1012                 relay_path = default_relay_path;
1013
1014         if (act_mask_tmp != 0)
1015                 act_mask = act_mask_tmp;
1016
1017         if (statfs(relay_path, &st) < 0) {
1018                 perror("statfs");
1019                 fprintf(stderr,"%s does not appear to be a valid path\n",
1020                         relay_path);
1021                 return 1;
1022         } else if (st.f_type != (long) RELAYFS_TYPE) {
1023                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1024                         relay_path);
1025                 return 1;
1026         }
1027
1028         if (open_devices() != 0)
1029                 return 1;
1030
1031         if (kill_running_trace) {
1032                 stop_all_traces();
1033                 return 0;
1034         }
1035
1036         setlocale(LC_NUMERIC, "en_US");
1037
1038         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1039         if (ncpus < 0) {
1040                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1041                 return 1;
1042         }
1043
1044         page_size = getpagesize();
1045
1046         if (start_devices() != 0)
1047                 return 1;
1048
1049         signal(SIGINT, handle_sigint);
1050         signal(SIGHUP, handle_sigint);
1051         signal(SIGTERM, handle_sigint);
1052         signal(SIGALRM, handle_sigint);
1053
1054         atexit(stop_all_tracing);
1055
1056         if (stop_watch)
1057                 alarm(stop_watch);
1058
1059         wait_for_threads();
1060
1061         if (!is_trace_stopped()) {
1062                 trace_stopped = 1;
1063                 stop_all_threads();
1064                 stop_all_traces();
1065         }
1066
1067         show_stats();
1068
1069         return 0;
1070 }
1071