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                         munlock(tip->fs_buf, tip->fs_buf_len);
 391                         munmap(tip->fs_buf, tip->fs_buf_len);
 392                         tip->fs_buf = NULL;
 393                 }
 394
 395                 tip->fs_off = tip->fs_size & (page_size - 1);
 396                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 397                 tip->fs_max_size += tip->fs_buf_len;
 398
 399                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 400                         perror("ftruncate");
 401                         return -1;
 402                 }
 403
 404                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 405                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 406                 if (tip->fs_buf == MAP_FAILED) {
 407                         perror("mmap");
 408                         return -1;
 409                 }
 410                 mlock(tip->fs_buf, tip->fs_buf_len);
 411         }
 412
 413         ret = read_data(tip, tip->fs_buf + tip->fs_off, buf_size);
 414         if (ret >= 0) {
 415                 tip->data_read += ret;
 416                 tip->fs_size += ret;
 417                 tip->fs_off += ret;
 418                 return 0;
 419         }
 420
 421         return -1;
 422 }
 423
 424 /*
 425  * Use the copy approach for pipes
 426  */
 427 static int get_subbuf(struct thread_information *tip)
 428 {
 429         struct tip_subbuf *ts;
 430         int ret;
 431
 432         ts = malloc(sizeof(*ts));
 433         ts->buf = malloc(buf_size);
 434         ts->max_len = buf_size;
 435
 436         ret = read_data(tip, ts->buf, ts->max_len);
 437         if (ret > 0) {
 438                 ts->len = ret;
 439                 return subbuf_fifo_queue(tip, ts);
 440         }
 441
 442         free(ts->buf);
 443         free(ts);
 444         return ret;
 445 }
 446
 447 static void close_thread(struct thread_information *tip)
 448 {
 449         if (tip->fd != -1)
 450                 close(tip->fd);
 451         if (tip->ofile)
 452                 fclose(tip->ofile);
 453         if (tip->ofile_buffer)
 454                 free(tip->ofile_buffer);
 455         if (tip->fd_buf)
 456                 free(tip->fd_buf);
 457
 458         tip->fd = -1;
 459         tip->ofile = NULL;
 460         tip->ofile_buffer = NULL;
 461         tip->fd_buf = NULL;
 462 }
 463
 464 static void *thread_main(void *arg)
 465 {
 466         struct thread_information *tip = arg;
 467         pid_t pid = getpid();
 468         cpu_set_t cpu_mask;
 469
 470         CPU_ZERO(&cpu_mask);
 471         CPU_SET((tip->cpu), &cpu_mask);
 472
 473         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 474                 perror("sched_setaffinity");
 475                 exit_trace(1);
 476         }
 477
 478         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 479                         relay_path, tip->device->buts_name, tip->cpu);
 480         tip->fd = open(tip->fn, O_RDONLY);
 481         if (tip->fd < 0) {
 482                 perror(tip->fn);
 483                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 484                         tip->fn);
 485                 exit_trace(1);
 486         }
 487
 488         while (!is_done()) {
 489                 if (tip->ofile_stdout) {
 490                         if (get_subbuf(tip))
 491                                 break;
 492                 } else {
 493                         if (mmap_subbuf(tip))
 494                                 break;
 495                 }
 496         }
 497
 498         /*
 499          * truncate to right size and cleanup mmap
 500          */
 501         if (!tip->ofile_stdout) {
 502                 int ofd = fileno(tip->ofile);
 503
 504                 if (tip->fs_buf)
 505                         munmap(tip->fs_buf, tip->fs_buf_len);
 506
 507                 ftruncate(ofd, tip->fs_size);
 508         }
 509
 510         tip->exited = 1;
 511         return NULL;
 512 }
 513
 514 static int write_data(struct thread_information *tip,
 515                       void *buf, unsigned int buf_len)
 516 {
 517         int ret;
 518
 519         if (!buf_len)
 520                 return 0;
 521
 522         while (1) {
 523                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 524                 if (ret == 1)
 525                         break;
 526
 527                 if (ret < 0) {
 528                         perror("write");
 529                         return 1;
 530                 }
 531         }
 532
 533         if (tip->ofile_stdout)
 534                 fflush(tip->ofile);
 535
 536         return 0;
 537 }
 538
 539 static int flush_subbuf(struct thread_information *tip, struct tip_subbuf *ts)
 540 {
 541         unsigned int offset = 0;
 542         struct blk_io_trace *t;
 543         int pdu_len, events = 0;
 544
 545         /*
 546          * surplus from last run
 547          */
 548         if (tip->leftover_ts) {
 549                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 550
 551                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 552                         prev_ts->max_len += ts->len;
 553                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 554                 }
 555
 556                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 557                 prev_ts->len += ts->len;
 558
 559                 free(ts->buf);
 560                 free(ts);
 561
 562                 ts = prev_ts;
 563                 tip->leftover_ts = NULL;
 564         }
 565
 566         while (offset + sizeof(*t) <= ts->len) {
 567                 t = ts->buf + offset;
 568
 569                 if (verify_trace(t)) {
 570                         write_data(tip, ts->buf, offset);
 571                         return -1;
 572                 }
 573
 574                 pdu_len = t->pdu_len;
 575
 576                 if (offset + sizeof(*t) + pdu_len > ts->len)
 577                         break;
 578
 579                 offset += sizeof(*t) + pdu_len;
 580                 tip->events_processed++;
 581                 tip->data_read += sizeof(*t) + pdu_len;
 582                 events++;
 583         }
 584
 585         if (write_data(tip, ts->buf, offset))
 586                 return -1;
 587
 588         /*
 589          * leftover bytes, save them for next time
 590          */
 591         if (offset != ts->len) {
 592                 tip->leftover_ts = ts;
 593                 ts->len -= offset;
 594                 memmove(ts->buf, ts->buf + offset, ts->len);
 595         } else {
 596                 free(ts->buf);
 597                 free(ts);
 598         }
 599
 600         return events;
 601 }
 602
 603 static int write_tip_events(struct thread_information *tip)
 604 {
 605         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 606
 607         if (ts)
 608                 return flush_subbuf(tip, ts);
 609
 610         return 0;
 611 }
 612
 613 /*
 614  * scans the tips we know and writes out the subbuffers we accumulate
 615  */
 616 static void get_and_write_events(void)
 617 {
 618         struct device_information *dip;
 619         struct thread_information *tip;
 620         int i, j, events, ret, tips_running;
 621
 622         while (!is_done()) {
 623                 events = 0;
 624
 625                 for_each_dip(dip, i) {
 626                         for_each_tip(dip, tip, j) {
 627                                 ret = write_tip_events(tip);
 628                                 if (ret > 0)
 629                                         events += ret;
 630                         }
 631                 }
 632
 633                 if (!events)
 634                         usleep(10);
 635         }
 636
 637         /*
 638          * reap stored events
 639          */
 640         do {
 641                 events = 0;
 642                 tips_running = 0;
 643                 for_each_dip(dip, i) {
 644                         for_each_tip(dip, tip, j) {
 645                                 ret = write_tip_events(tip);
 646                                 if (ret > 0)
 647                                         events += ret;
 648                                 tips_running += !tip->exited;
 649                         }
 650                 }
 651                 usleep(10);
 652         } while (events || tips_running);
 653 }
 654
 655 static void wait_for_threads(void)
 656 {
 657         /*
 658          * for piped output, poll and fetch data for writeout. for files,
 659          * we just wait around for trace threads to exit
 660          */
 661         if (output_name && !strcmp(output_name, "-"))
 662                 get_and_write_events();
 663         else {
 664                 struct device_information *dip;
 665                 struct thread_information *tip;
 666                 int i, j, tips_running;
 667
 668                 do {
 669                         tips_running = 0;
 670                         usleep(1000);
 671
 672                         for_each_dip(dip, i)
 673                                 for_each_tip(dip, tip, j)
 674                                         tips_running += !tip->exited;
 675                 } while (tips_running);
 676         }
 677 }
 678
 679 static int start_threads(struct device_information *dip)
 680 {
 681         struct thread_information *tip;
 682         char op[64];
 683         int j, pipeline = output_name && !strcmp(output_name, "-");
 684         int len, mode, vbuf_size;
 685
 686         for_each_tip(dip, tip, j) {
 687                 tip->cpu = j;
 688                 tip->device = dip;
 689                 tip->events_processed = 0;
 690                 memset(&tip->fifo, 0, sizeof(tip->fifo));
 691                 tip->leftover_ts = NULL;
 692
 693                 if (pipeline) {
 694                         tip->ofile = fdopen(STDOUT_FILENO, "w");
 695                         tip->ofile_stdout = 1;
 696                         mode = _IOLBF;
 697                         vbuf_size = 512;
 698                 } else {
 699                         len = 0;
 700
 701                         if (output_dir)
 702                                 len = sprintf(op, "%s/", output_dir);
 703
 704                         if (output_name) {
 705                                 sprintf(op + len, "%s.blktrace.%d", output_name,
 706                                         tip->cpu);
 707                         } else {
 708                                 sprintf(op + len, "%s.blktrace.%d",
 709                                         dip->buts_name, tip->cpu);
 710                         }
 711                         tip->ofile = fopen(op, "w+");
 712                         tip->ofile_stdout = 0;
 713                         mode = _IOFBF;
 714                         vbuf_size = OFILE_BUF;
 715                 }
 716
 717                 if (tip->ofile == NULL) {
 718                         perror(op);
 719                         return 1;
 720                 }
 721
 722                 tip->ofile_buffer = malloc(vbuf_size);
 723                 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
 724                         perror("setvbuf");
 725                         close_thread(tip);
 726                         return 1;
 727                 }
 728
 729                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
 730                         perror("pthread_create");
 731                         close_thread(tip);
 732                         return 1;
 733                 }
 734         }
 735
 736         return 0;
 737 }
 738
 739 static void stop_threads(struct device_information *dip)
 740 {
 741         struct thread_information *tip;
 742         unsigned long ret;
 743         int i;
 744
 745         for_each_tip(dip, tip, i) {
 746                 (void) pthread_join(tip->thread, (void *) &ret);
 747                 close_thread(tip);
 748         }
 749 }
 750
 751 static void stop_all_threads(void)
 752 {
 753         struct device_information *dip;
 754         int i;
 755
 756         for_each_dip(dip, i)
 757                 stop_threads(dip);
 758 }
 759
 760 static void stop_all_tracing(void)
 761 {
 762         struct device_information *dip;
 763         int i;
 764
 765         for_each_dip(dip, i)
 766                 stop_trace(dip);
 767 }
 768
 769 static void exit_trace(int status)
 770 {
 771         if (!is_trace_stopped()) {
 772                 trace_stopped = 1;
 773                 stop_all_threads();
 774                 stop_all_tracing();
 775         }
 776
 777         exit(status);
 778 }
 779
 780 static int resize_devices(char *path)
 781 {
 782         int size = (ndevs + 1) * sizeof(struct device_information);
 783
 784         device_information = realloc(device_information, size);
 785         if (!device_information) {
 786                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
 787                 return 1;
 788         }
 789         device_information[ndevs].path = path;
 790         ndevs++;
 791         return 0;
 792 }
 793
 794 static int open_devices(void)
 795 {
 796         struct device_information *dip;
 797         int i;
 798
 799         for_each_dip(dip, i) {
 800                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
 801                 if (dip->fd < 0) {
 802                         perror(dip->path);
 803                         return 1;
 804                 }
 805         }
 806
 807         return 0;
 808 }
 809
 810 static int start_devices(void)
 811 {
 812         struct device_information *dip;
 813         int i, j, size;
 814
 815         size = ncpus * sizeof(struct thread_information);
 816         thread_information = malloc(size * ndevs);
 817         if (!thread_information) {
 818                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
 819                 return 1;
 820         }
 821
 822         for_each_dip(dip, i) {
 823                 if (start_trace(dip)) {
 824                         close(dip->fd);
 825                         fprintf(stderr, "Failed to start trace on %s\n",
 826                                 dip->path);
 827                         break;
 828                 }
 829         }
 830
 831         if (i != ndevs) {
 832                 __for_each_dip(dip, j, i)
 833                         stop_trace(dip);
 834
 835                 return 1;
 836         }
 837
 838         for_each_dip(dip, i) {
 839                 dip->threads = thread_information + (i * ncpus);
 840                 if (start_threads(dip)) {
 841                         fprintf(stderr, "Failed to start worker threads\n");
 842                         break;
 843                 }
 844         }
 845
 846         if (i != ndevs) {
 847                 __for_each_dip(dip, j, i)
 848                         stop_threads(dip);
 849                 for_each_dip(dip, i)
 850                         stop_trace(dip);
 851
 852                 return 1;
 853         }
 854
 855         return 0;
 856 }
 857
 858 static void show_stats(void)
 859 {
 860         struct device_information *dip;
 861         struct thread_information *tip;
 862         unsigned long long events_processed, data_read;
 863         unsigned long total_drops;
 864         int i, j, no_stdout = 0;
 865
 866         if (is_stat_shown())
 867                 return;
 868
 869         if (output_name && !strcmp(output_name, "-"))
 870                 no_stdout = 1;
 871
 872         stat_shown = 1;
 873
 874         total_drops = 0;
 875         for_each_dip(dip, i) {
 876                 if (!no_stdout)
 877                         printf("Device: %s\n", dip->path);
 878                 events_processed = 0;
 879                 data_read = 0;
 880                 for_each_tip(dip, tip, j) {
 881                         if (!no_stdout)
 882                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
 883                                         tip->cpu, tip->events_processed,
 884                                         tip->data_read >> 10);
 885                         events_processed += tip->events_processed;
 886                         data_read += tip->data_read;
 887                 }
 888                 total_drops += dip->drop_count;
 889                 if (!no_stdout)
 890                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
 891                                         events_processed, dip->drop_count,
 892                                         data_read >> 10);
 893         }
 894
 895         if (total_drops)
 896                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
 897 }
 898
 899 static char usage_str[] = \
 900         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
 901         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
 902         "\t-d Use specified device. May also be given last after options\n" \
 903         "\t-r Path to mounted relayfs, defaults to /relay\n" \
 904         "\t-o File(s) to send output to\n" \
 905         "\t-D Directory to prepend to output file names\n" \
 906         "\t-k Kill a running trace\n" \
 907         "\t-w Stop after defined time, in seconds\n" \
 908         "\t-a Only trace specified actions. See documentation\n" \
 909         "\t-A Give trace mask as a single value. See documentation\n" \
 910         "\t-b Sub buffer size in KiB\n" \
 911         "\t-n Number of sub buffers\n" \
 912         "\t-v Print program version info\n\n";
 913
 914 static void show_usage(char *program)
 915 {
 916         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
 917 }
 918 static void handle_sigint(__attribute__((__unused__)) int sig)
 919 {
 920         done = 1;
 921 }
 922
 923 int main(int argc, char *argv[])
 924 {
 925         static char default_relay_path[] = "/relay";
 926         struct statfs st;
 927         int i, c;
 928         int stop_watch = 0;
 929         int act_mask_tmp = 0;
 930
 931         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
 932                 switch (c) {
 933                 case 'a':
 934                         i = find_mask_map(optarg);
 935                         if (i < 0) {
 936                                 fprintf(stderr,"Invalid action mask %s\n",
 937                                         optarg);
 938                                 return 1;
 939                         }
 940                         act_mask_tmp |= i;
 941                         break;
 942
 943                 case 'A':
 944                         if ((sscanf(optarg, "%x", &i) != 1) ||
 945                                                         !valid_act_opt(i)) {
 946                                 fprintf(stderr,
 947                                         "Invalid set action mask %s/0x%x\n",
 948                                         optarg, i);
 949                                 return 1;
 950                         }
 951                         act_mask_tmp = i;
 952                         break;
 953
 954                 case 'd':
 955                         if (resize_devices(optarg) != 0)
 956                                 return 1;
 957                         break;
 958
 959                 case 'r':
 960                         relay_path = optarg;
 961                         break;
 962
 963                 case 'o':
 964                         output_name = optarg;
 965                         break;
 966                 case 'k':
 967                         kill_running_trace = 1;
 968                         break;
 969                 case 'w':
 970                         stop_watch = atoi(optarg);
 971                         if (stop_watch <= 0) {
 972                                 fprintf(stderr,
 973                                         "Invalid stopwatch value (%d secs)\n",
 974                                         stop_watch);
 975                                 return 1;
 976                         }
 977                         break;
 978                 case 'V':
 979                         printf("%s version %s\n", argv[0], blktrace_version);
 980                         return 0;
 981                 case 'b':
 982                         buf_size = strtoul(optarg, NULL, 10);
 983                         if (buf_size <= 0 || buf_size > 16*1024) {
 984                                 fprintf(stderr,
 985                                         "Invalid buffer size (%lu)\n",buf_size);
 986                                 return 1;
 987                         }
 988                         buf_size <<= 10;
 989                         break;
 990                 case 'n':
 991                         buf_nr = strtoul(optarg, NULL, 10);
 992                         if (buf_nr <= 0) {
 993                                 fprintf(stderr,
 994                                         "Invalid buffer nr (%lu)\n", buf_nr);
 995                                 return 1;
 996                         }
 997                         break;
 998                 case 'D':
 999                         output_dir = optarg;
1000                         break;
1001                 default:
1002                         show_usage(argv[0]);
1003                         return 1;
1004                 }
1005         }
1006
1007         while (optind < argc) {
1008                 if (resize_devices(argv[optind++]) != 0)
1009                         return 1;
1010         }
1011
1012         if (ndevs == 0) {
1013                 show_usage(argv[0]);
1014                 return 1;
1015         }
1016
1017         if (!relay_path)
1018                 relay_path = default_relay_path;
1019
1020         if (act_mask_tmp != 0)
1021                 act_mask = act_mask_tmp;
1022
1023         if (statfs(relay_path, &st) < 0) {
1024                 perror("statfs");
1025                 fprintf(stderr,"%s does not appear to be a valid path\n",
1026                         relay_path);
1027                 return 1;
1028         } else if (st.f_type != (long) RELAYFS_TYPE) {
1029                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1030                         relay_path);
1031                 return 1;
1032         }
1033
1034         if (open_devices() != 0)
1035                 return 1;
1036
1037         if (kill_running_trace) {
1038                 stop_all_traces();
1039                 return 0;
1040         }
1041
1042         setlocale(LC_NUMERIC, "en_US");
1043
1044         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1045         if (ncpus < 0) {
1046                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1047                 return 1;
1048         }
1049
1050         page_size = getpagesize();
1051
1052         if (start_devices() != 0)
1053                 return 1;
1054
1055         signal(SIGINT, handle_sigint);
1056         signal(SIGHUP, handle_sigint);
1057         signal(SIGTERM, handle_sigint);
1058         signal(SIGALRM, handle_sigint);
1059
1060         atexit(stop_all_tracing);
1061
1062         if (stop_watch)
1063                 alarm(stop_watch);
1064
1065         wait_for_threads();
1066
1067         if (!is_trace_stopped()) {
1068                 trace_stopped = 1;
1069                 stop_all_threads();
1070                 stop_all_traces();
1071         }
1072
1073         show_stats();
1074
1075         return 0;
1076 }
1077