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 <sys/socket.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <sched.h>
  38 #include <ctype.h>
  39 #include <getopt.h>
  40 #include <errno.h>
  41 #include <netinet/in.h>
  42 #include <arpa/inet.h>
  43 #include <netdb.h>
  44 #include <sys/sendfile.h>
  45
  46 #include "blktrace.h"
  47 #include "barrier.h"
  48
  49 static char blktrace_version[] = "0.99";
  50
  51 /*
  52  * You may want to increase this even more, if you are logging at a high
  53  * rate and see skipped/missed events
  54  */
  55 #define BUF_SIZE        (512 * 1024)
  56 #define BUF_NR          (4)
  57
  58 #define OFILE_BUF       (128 * 1024)
  59
  60 #define RELAYFS_TYPE    0xF0B4A981
  61
  62 #define S_OPTS  "d:a:A:r:o:kw:Vb:n:D:lh:p:s"
  63 static struct option l_opts[] = {
  64         {
  65                 .name = "dev",
  66                 .has_arg = required_argument,
  67                 .flag = NULL,
  68                 .val = 'd'
  69         },
  70         {
  71                 .name = "act-mask",
  72                 .has_arg = required_argument,
  73                 .flag = NULL,
  74                 .val = 'a'
  75         },
  76         {
  77                 .name = "set-mask",
  78                 .has_arg = required_argument,
  79                 .flag = NULL,
  80                 .val = 'A'
  81         },
  82         {
  83                 .name = "relay",
  84                 .has_arg = required_argument,
  85                 .flag = NULL,
  86                 .val = 'r'
  87         },
  88         {
  89                 .name = "output",
  90                 .has_arg = required_argument,
  91                 .flag = NULL,
  92                 .val = 'o'
  93         },
  94         {
  95                 .name = "kill",
  96                 .has_arg = no_argument,
  97                 .flag = NULL,
  98                 .val = 'k'
  99         },
 100         {
 101                 .name = "stopwatch",
 102                 .has_arg = required_argument,
 103                 .flag = NULL,
 104                 .val = 'w'
 105         },
 106         {
 107                 .name = "version",
 108                 .has_arg = no_argument,
 109                 .flag = NULL,
 110                 .val = 'V'
 111         },
 112         {
 113                 .name = "buffer-size",
 114                 .has_arg = required_argument,
 115                 .flag = NULL,
 116                 .val = 'b'
 117         },
 118         {
 119                 .name = "num-sub-buffers",
 120                 .has_arg = required_argument,
 121                 .flag = NULL,
 122                 .val = 'n'
 123         },
 124         {
 125                 .name = "output-dir",
 126                 .has_arg = required_argument,
 127                 .flag = NULL,
 128                 .val = 'D'
 129         },
 130         {
 131                 .name = "listen",
 132                 .has_arg = no_argument,
 133                 .flag = NULL,
 134                 .val = 'l'
 135         },
 136         {
 137                 .name = "host",
 138                 .has_arg = required_argument,
 139                 .flag = NULL,
 140                 .val = 'h'
 141         },
 142         {
 143                 .name = "port",
 144                 .has_arg = required_argument,
 145                 .flag = NULL,
 146                 .val = 'p'
 147         },
 148         {
 149                 .name = "sendfile",
 150                 .has_arg = no_argument,
 151                 .flag = NULL,
 152                 .val = 's'
 153         },
 154         {
 155                 .name = NULL,
 156         }
 157 };
 158
 159 struct tip_subbuf {
 160         void *buf;
 161         unsigned int len;
 162         unsigned int max_len;
 163 };
 164
 165 #define FIFO_SIZE       (1024)  /* should be plenty big! */
 166 #define CL_SIZE         (128)   /* cache line, any bigger? */
 167
 168 struct tip_subbuf_fifo {
 169         int tail __attribute__((aligned(CL_SIZE)));
 170         int head __attribute__((aligned(CL_SIZE)));
 171         struct tip_subbuf *q[FIFO_SIZE];
 172 };
 173
 174 struct thread_information {
 175         int cpu;
 176         pthread_t thread;
 177
 178         int fd;
 179         void *fd_buf;
 180         char fn[MAXPATHLEN + 64];
 181
 182         struct in_addr cl_in_addr;
 183
 184         FILE *ofile;
 185         char *ofile_buffer;
 186         off_t ofile_offset;
 187         int ofile_stdout;
 188         int ofile_mmap;
 189
 190         int (*get_subbuf)(struct thread_information *, unsigned int);
 191         int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *);
 192         int (*read_data)(struct thread_information *, void *, unsigned int);
 193
 194         unsigned long events_processed;
 195         unsigned long long data_read;
 196         unsigned long long data_queued;
 197         struct device_information *device;
 198
 199         int exited;
 200
 201         /*
 202          * piped fifo buffers
 203          */
 204         struct tip_subbuf_fifo fifo;
 205         struct tip_subbuf *leftover_ts;
 206
 207         /*
 208          * mmap controlled output files
 209          */
 210         unsigned long long fs_size;
 211         unsigned long long fs_max_size;
 212         unsigned long fs_off;
 213         void *fs_buf;
 214         unsigned long fs_buf_len;
 215 };
 216
 217 struct device_information {
 218         int fd;
 219         char *path;
 220         char buts_name[32];
 221         volatile int trace_started;
 222         unsigned long drop_count;
 223         struct thread_information *threads;
 224 };
 225
 226 static int ncpus;
 227 static struct thread_information *thread_information;
 228 static int ndevs;
 229 static struct device_information *device_information;
 230 static int ndevs_running;
 231
 232 /* command line option globals */
 233 static char *relay_path;
 234 static char *output_name;
 235 static char *output_dir;
 236 static int act_mask = ~0U;
 237 static int kill_running_trace;
 238 static unsigned long buf_size = BUF_SIZE;
 239 static unsigned long buf_nr = BUF_NR;
 240 static unsigned int page_size;
 241
 242 #define is_done()       (*(volatile int *)(&done))
 243 static volatile int done;
 244
 245 #define is_trace_stopped()      (*(volatile int *)(&trace_stopped))
 246 static volatile int trace_stopped;
 247
 248 #define is_stat_shown() (*(volatile int *)(&stat_shown))
 249 static volatile int stat_shown;
 250
 251 int data_is_native = -1;
 252
 253 static void exit_trace(int status);
 254
 255 #define dip_tracing(dip)        (*(volatile int *)(&(dip)->trace_started))
 256 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
 257
 258 #define __for_each_dip(__d, __i, __e)   \
 259         for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
 260
 261 #define for_each_dip(__d, __i)  __for_each_dip(__d, __i, ndevs)
 262 #define for_each_tip(__d, __t, __j)     \
 263         for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
 264
 265 /*
 266  * networking stuff follows. we include a magic number so we know whether
 267  * to endianness convert or not
 268  */
 269 struct blktrace_net_hdr {
 270         u32 magic;              /* same as trace magic */
 271         char buts_name[32];     /* trace name */
 272         u32 cpu;                /* for which cpu */
 273         u32 max_cpus;
 274         u32 len;                /* length of following trace data */
 275 };
 276
 277 #define TRACE_NET_PORT          (8462)
 278
 279 enum {
 280         Net_none = 0,
 281         Net_server,
 282         Net_client,
 283 };
 284
 285 /*
 286  * network cmd line params
 287  */
 288 static char hostname[MAXHOSTNAMELEN];
 289 static int net_port = TRACE_NET_PORT;
 290 static int net_mode = 0;
 291 static int net_use_sendfile;
 292
 293 static int net_in_fd = -1;
 294 static int net_out_fd = -1;
 295
 296 static void handle_sigint(__attribute__((__unused__)) int sig)
 297 {
 298         struct device_information *dip;
 299         int i;
 300
 301         /*
 302          * stop trace so we can reap currently produced data
 303          */
 304         for_each_dip(dip, i) {
 305                 if (dip->fd == -1)
 306                         continue;
 307                 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
 308                         perror("BLKTRACESTOP");
 309         }
 310
 311         done = 1;
 312 }
 313
 314 static int get_dropped_count(const char *buts_name)
 315 {
 316         int fd;
 317         char tmp[MAXPATHLEN + 64];
 318
 319         snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
 320                  relay_path, buts_name);
 321
 322         fd = open(tmp, O_RDONLY);
 323         if (fd < 0) {
 324                 /*
 325                  * this may be ok, if the kernel doesn't support dropped counts
 326                  */
 327                 if (errno == ENOENT)
 328                         return 0;
 329
 330                 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
 331                 return -1;
 332         }
 333
 334         if (read(fd, tmp, sizeof(tmp)) < 0) {
 335                 perror(tmp);
 336                 close(fd);
 337                 return -1;
 338         }
 339
 340         close(fd);
 341
 342         return atoi(tmp);
 343 }
 344
 345 static int start_trace(struct device_information *dip)
 346 {
 347         struct blk_user_trace_setup buts;
 348
 349         memset(&buts, 0, sizeof(buts));
 350         buts.buf_size = buf_size;
 351         buts.buf_nr = buf_nr;
 352         buts.act_mask = act_mask;
 353
 354         if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
 355                 perror("BLKTRACESETUP");
 356                 return 1;
 357         }
 358
 359         if (ioctl(dip->fd, BLKTRACESTART) < 0) {
 360                 perror("BLKTRACESTART");
 361                 return 1;
 362         }
 363
 364         memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
 365         dip_set_tracing(dip, 1);
 366         return 0;
 367 }
 368
 369 static void stop_trace(struct device_information *dip)
 370 {
 371         if (dip_tracing(dip) || kill_running_trace) {
 372                 dip_set_tracing(dip, 0);
 373
 374                 /*
 375                  * should be stopped, just don't complain if it isn't
 376                  */
 377                 ioctl(dip->fd, BLKTRACESTOP);
 378
 379                 if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
 380                         perror("BLKTRACETEARDOWN");
 381
 382                 close(dip->fd);
 383                 dip->fd = -1;
 384         }
 385 }
 386
 387 static void stop_all_traces(void)
 388 {
 389         struct device_information *dip;
 390         int i;
 391
 392         for_each_dip(dip, i) {
 393                 dip->drop_count = get_dropped_count(dip->buts_name);
 394                 stop_trace(dip);
 395         }
 396 }
 397
 398 static void wait_for_data(struct thread_information *tip, int events)
 399 {
 400         struct pollfd pfd = { .fd = tip->fd, .events = events };
 401
 402         do {
 403                 if (poll(&pfd, 1, 100) < 0) {
 404                         perror("poll");
 405                         break;
 406                 }
 407                 if (pfd.revents & events)
 408                         break;
 409                 if (tip->ofile_stdout)
 410                         break;
 411         } while (!is_done());
 412 }
 413
 414 static int read_data_file(struct thread_information *tip, void *buf,
 415                           unsigned int len)
 416 {
 417         int ret = 0;
 418
 419         do {
 420                 wait_for_data(tip, POLLIN);
 421
 422                 ret = read(tip->fd, buf, len);
 423                 if (!ret)
 424                         continue;
 425                 else if (ret > 0)
 426                         return ret;
 427                 else {
 428                         if (errno != EAGAIN) {
 429                                 perror(tip->fn);
 430                                 fprintf(stderr,"Thread %d failed read of %s\n",
 431                                         tip->cpu, tip->fn);
 432                                 break;
 433                         }
 434                         continue;
 435                 }
 436         } while (!is_done());
 437
 438         return ret;
 439
 440 }
 441
 442 static int read_data_net(struct thread_information *tip, void *buf,
 443                          unsigned int len)
 444 {
 445         unsigned int bytes_left = len;
 446         int ret = 0;
 447
 448         do {
 449                 ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
 450
 451                 if (!ret)
 452                         continue;
 453                 else if (ret < 0) {
 454                         if (errno != EAGAIN) {
 455                                 perror(tip->fn);
 456                                 fprintf(stderr, "server: failed read\n");
 457                                 return 0;
 458                         }
 459                         continue;
 460                 } else {
 461                         buf += ret;
 462                         bytes_left -= ret;
 463                 }
 464         } while (!is_done() && bytes_left);
 465
 466         return len - bytes_left;
 467 }
 468
 469 static inline struct tip_subbuf *
 470 subbuf_fifo_dequeue(struct thread_information *tip)
 471 {
 472         const int head = tip->fifo.head;
 473         const int next = (head + 1) & (FIFO_SIZE - 1);
 474
 475         if (head != tip->fifo.tail) {
 476                 struct tip_subbuf *ts = tip->fifo.q[head];
 477
 478                 store_barrier();
 479                 tip->fifo.head = next;
 480                 return ts;
 481         }
 482
 483         return NULL;
 484 }
 485
 486 static inline int subbuf_fifo_queue(struct thread_information *tip,
 487                                     struct tip_subbuf *ts)
 488 {
 489         const int tail = tip->fifo.tail;
 490         const int next = (tail + 1) & (FIFO_SIZE - 1);
 491
 492         if (next != tip->fifo.head) {
 493                 tip->fifo.q[tail] = ts;
 494                 store_barrier();
 495                 tip->fifo.tail = next;
 496                 return 0;
 497         }
 498
 499         fprintf(stderr, "fifo too small!\n");
 500         return 1;
 501 }
 502
 503 /*
 504  * For file output, truncate and mmap the file appropriately
 505  */
 506 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
 507 {
 508         int ofd = fileno(tip->ofile);
 509         int ret;
 510
 511         /*
 512          * extend file, if we have to. use chunks of 16 subbuffers.
 513          */
 514         if (tip->fs_off + buf_size > tip->fs_buf_len) {
 515                 if (tip->fs_buf) {
 516                         munlock(tip->fs_buf, tip->fs_buf_len);
 517                         munmap(tip->fs_buf, tip->fs_buf_len);
 518                         tip->fs_buf = NULL;
 519                 }
 520
 521                 tip->fs_off = tip->fs_size & (page_size - 1);
 522                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 523                 tip->fs_max_size += tip->fs_buf_len;
 524
 525                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 526                         perror("ftruncate");
 527                         return -1;
 528                 }
 529
 530                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 531                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 532                 if (tip->fs_buf == MAP_FAILED) {
 533                         perror("mmap");
 534                         return -1;
 535                 }
 536                 mlock(tip->fs_buf, tip->fs_buf_len);
 537         }
 538
 539         ret = tip->read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
 540         if (ret >= 0) {
 541                 tip->data_read += ret;
 542                 tip->fs_size += ret;
 543                 tip->fs_off += ret;
 544                 return 0;
 545         }
 546
 547         return -1;
 548 }
 549
 550 /*
 551  * Use the copy approach for pipes and network
 552  */
 553 static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
 554 {
 555         struct tip_subbuf *ts = malloc(sizeof(*ts));
 556         int ret;
 557
 558         ts->buf = malloc(buf_size);
 559         ts->max_len = maxlen;
 560
 561         ret = tip->read_data(tip, ts->buf, ts->max_len);
 562         if (ret > 0) {
 563                 ts->len = ret;
 564                 tip->data_read += ret;
 565                 if (subbuf_fifo_queue(tip, ts))
 566                         return -1;
 567         }
 568
 569         return ret;
 570 }
 571
 572 static int get_subbuf_sendfile(struct thread_information *tip,
 573                                unsigned int maxlen)
 574 {
 575         struct tip_subbuf *ts;
 576         struct stat sb;
 577         unsigned int ready;
 578
 579         wait_for_data(tip, POLLMSG);
 580
 581         /*
 582          * hack to get last data out, we can't use sendfile for that
 583          */
 584         if (is_done())
 585                 return get_subbuf(tip, maxlen);
 586
 587         if (fstat(tip->fd, &sb) < 0) {
 588                 perror("trace stat");
 589                 return -1;
 590         }
 591         ready = sb.st_size - tip->data_queued;
 592         if (!ready) {
 593                 usleep(1000);
 594                 return 0;
 595         }
 596
 597         ts = malloc(sizeof(*ts));
 598         ts->buf = NULL;
 599         ts->max_len = 0;
 600         ts->len = ready;
 601         tip->data_queued += ready;
 602
 603         if (subbuf_fifo_queue(tip, ts))
 604                 return -1;
 605
 606         return ready;
 607 }
 608
 609 static void close_thread(struct thread_information *tip)
 610 {
 611         if (tip->fd != -1)
 612                 close(tip->fd);
 613         if (tip->ofile)
 614                 fclose(tip->ofile);
 615         if (tip->ofile_buffer)
 616                 free(tip->ofile_buffer);
 617         if (tip->fd_buf)
 618                 free(tip->fd_buf);
 619
 620         tip->fd = -1;
 621         tip->ofile = NULL;
 622         tip->ofile_buffer = NULL;
 623         tip->fd_buf = NULL;
 624 }
 625
 626 static void tip_ftrunc_final(struct thread_information *tip)
 627 {
 628         /*
 629          * truncate to right size and cleanup mmap
 630          */
 631         if (tip->ofile_mmap) {
 632                 int ofd = fileno(tip->ofile);
 633
 634                 if (tip->fs_buf)
 635                         munmap(tip->fs_buf, tip->fs_buf_len);
 636
 637                 ftruncate(ofd, tip->fs_size);
 638         }
 639 }
 640
 641 static void *thread_main(void *arg)
 642 {
 643         struct thread_information *tip = arg;
 644         pid_t pid = getpid();
 645         cpu_set_t cpu_mask;
 646
 647         CPU_ZERO(&cpu_mask);
 648         CPU_SET((tip->cpu), &cpu_mask);
 649
 650         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 651                 perror("sched_setaffinity");
 652                 exit_trace(1);
 653         }
 654
 655         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 656                         relay_path, tip->device->buts_name, tip->cpu);
 657         tip->fd = open(tip->fn, O_RDONLY);
 658         if (tip->fd < 0) {
 659                 perror(tip->fn);
 660                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 661                         tip->fn);
 662                 exit_trace(1);
 663         }
 664
 665         while (!is_done()) {
 666                 if (tip->get_subbuf(tip, buf_size) < 0)
 667                         break;
 668         }
 669
 670         /*
 671          * trace is stopped, pull data until we get a short read
 672          */
 673         while (tip->get_subbuf(tip, buf_size) > 0)
 674                 ;
 675
 676         tip_ftrunc_final(tip);
 677         tip->exited = 1;
 678         return NULL;
 679 }
 680
 681 static int write_data_net(int fd, void *buf, unsigned int buf_len)
 682 {
 683         unsigned int bytes_left = buf_len;
 684         int ret;
 685
 686         while (bytes_left) {
 687                 ret = send(fd, buf, bytes_left, 0);
 688                 if (ret < 0) {
 689                         perror("send");
 690                         return 1;
 691                 }
 692
 693                 buf += ret;
 694                 bytes_left -= ret;
 695         }
 696
 697         return 0;
 698 }
 699
 700 static int net_send_header(struct thread_information *tip, unsigned int len)
 701 {
 702         struct blktrace_net_hdr hdr;
 703
 704         hdr.magic = BLK_IO_TRACE_MAGIC;
 705         strcpy(hdr.buts_name, tip->device->buts_name);
 706         hdr.cpu = tip->cpu;
 707         hdr.max_cpus = ncpus;
 708         hdr.len = len;
 709
 710         return write_data_net(net_out_fd, &hdr, sizeof(hdr));
 711 }
 712
 713 /*
 714  * send header with 0 length to signal end-of-run
 715  */
 716 static void net_client_send_close(void)
 717 {
 718         struct device_information *dip;
 719         struct blktrace_net_hdr hdr;
 720         int i;
 721
 722         for_each_dip(dip, i) {
 723                 hdr.magic = BLK_IO_TRACE_MAGIC;
 724                 hdr.max_cpus = ncpus;
 725                 hdr.len = 0;
 726                 strcpy(hdr.buts_name, dip->buts_name);
 727                 hdr.cpu = get_dropped_count(dip->buts_name);
 728
 729                 write_data_net(net_out_fd, &hdr, sizeof(hdr));
 730         }
 731
 732 }
 733
 734 static int flush_subbuf_net(struct thread_information *tip,
 735                             struct tip_subbuf *ts)
 736 {
 737         if (net_send_header(tip, ts->len))
 738                 return -1;
 739         if (write_data_net(net_out_fd, ts->buf, ts->len))
 740                 return -1;
 741
 742         free(ts->buf);
 743         free(ts);
 744         return 1;
 745 }
 746
 747 static int net_sendfile(struct thread_information *tip, struct tip_subbuf *ts)
 748 {
 749         int ret = sendfile(net_out_fd, tip->fd, NULL, ts->len);
 750
 751         if (ret < 0) {
 752                 perror("sendfile");
 753                 return 1;
 754         } else if (ret < (int) ts->len) {
 755                 fprintf(stderr, "short sendfile send (%d of %d)\n", ret, ts->len);
 756                 return 1;
 757         }
 758
 759         return 0;
 760 }
 761
 762 static int flush_subbuf_sendfile(struct thread_information *tip,
 763                                  struct tip_subbuf *ts)
 764 {
 765         int ret = -1;
 766
 767         /*
 768          * currently we cannot use sendfile() on the last bytes read, as they
 769          * may not be a full subbuffer. get_subbuf_sendfile() falls back to
 770          * the read approach for those, so use send() to ship them out
 771          */
 772         if (ts->buf)
 773                 return flush_subbuf_net(tip, ts);
 774
 775         if (net_send_header(tip, ts->len))
 776                 goto err;
 777         if (net_sendfile(tip, ts))
 778                 goto err;
 779
 780         tip->data_read += ts->len;
 781         tip->ofile_offset += buf_size;
 782         ret = 1;
 783 err:
 784         free(ts);
 785         return ret;
 786 }
 787
 788 static int write_data(struct thread_information *tip, void *buf,
 789                       unsigned int buf_len)
 790 {
 791         int ret;
 792
 793         if (!buf_len)
 794                 return 0;
 795
 796         while (1) {
 797                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 798                 if (ret == 1)
 799                         break;
 800
 801                 if (ret < 0) {
 802                         perror("write");
 803                         return 1;
 804                 }
 805         }
 806
 807         if (tip->ofile_stdout)
 808                 fflush(tip->ofile);
 809
 810         return 0;
 811 }
 812
 813 static int flush_subbuf_file(struct thread_information *tip,
 814                              struct tip_subbuf *ts)
 815 {
 816         unsigned int offset = 0;
 817         struct blk_io_trace *t;
 818         int pdu_len, events = 0;
 819
 820         /*
 821          * surplus from last run
 822          */
 823         if (tip->leftover_ts) {
 824                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 825
 826                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 827                         prev_ts->max_len += ts->len;
 828                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 829                 }
 830
 831                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 832                 prev_ts->len += ts->len;
 833
 834                 free(ts->buf);
 835                 free(ts);
 836
 837                 ts = prev_ts;
 838                 tip->leftover_ts = NULL;
 839         }
 840
 841         while (offset + sizeof(*t) <= ts->len) {
 842                 t = ts->buf + offset;
 843
 844                 if (verify_trace(t)) {
 845                         write_data(tip, ts->buf, offset);
 846                         return -1;
 847                 }
 848
 849                 pdu_len = t->pdu_len;
 850
 851                 if (offset + sizeof(*t) + pdu_len > ts->len)
 852                         break;
 853
 854                 offset += sizeof(*t) + pdu_len;
 855                 tip->events_processed++;
 856                 tip->data_read += sizeof(*t) + pdu_len;
 857                 events++;
 858         }
 859
 860         if (write_data(tip, ts->buf, offset))
 861                 return -1;
 862
 863         /*
 864          * leftover bytes, save them for next time
 865          */
 866         if (offset != ts->len) {
 867                 tip->leftover_ts = ts;
 868                 ts->len -= offset;
 869                 memmove(ts->buf, ts->buf + offset, ts->len);
 870         } else {
 871                 free(ts->buf);
 872                 free(ts);
 873         }
 874
 875         return events;
 876 }
 877
 878 static int write_tip_events(struct thread_information *tip)
 879 {
 880         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 881
 882         if (ts)
 883                 return tip->flush_subbuf(tip, ts);
 884
 885         return 0;
 886 }
 887
 888 /*
 889  * scans the tips we know and writes out the subbuffers we accumulate
 890  */
 891 static void get_and_write_events(void)
 892 {
 893         struct device_information *dip;
 894         struct thread_information *tip;
 895         int i, j, events, ret, tips_running;
 896
 897         while (!is_done()) {
 898                 events = 0;
 899
 900                 for_each_dip(dip, i) {
 901                         for_each_tip(dip, tip, j) {
 902                                 ret = write_tip_events(tip);
 903                                 if (ret > 0)
 904                                         events += ret;
 905                         }
 906                 }
 907
 908                 if (!events)
 909                         usleep(100000);
 910         }
 911
 912         /*
 913          * reap stored events
 914          */
 915         do {
 916                 events = 0;
 917                 tips_running = 0;
 918                 for_each_dip(dip, i) {
 919                         for_each_tip(dip, tip, j) {
 920                                 ret = write_tip_events(tip);
 921                                 if (ret > 0)
 922                                         events += ret;
 923                                 tips_running += !tip->exited;
 924                         }
 925                 }
 926                 usleep(10);
 927         } while (events || tips_running);
 928 }
 929
 930 static void wait_for_threads(void)
 931 {
 932         /*
 933          * for piped or network output, poll and fetch data for writeout.
 934          * for files, we just wait around for trace threads to exit
 935          */
 936         if ((output_name && !strcmp(output_name, "-")) ||
 937             net_mode == Net_client)
 938                 get_and_write_events();
 939         else {
 940                 struct device_information *dip;
 941                 struct thread_information *tip;
 942                 int i, j, tips_running;
 943
 944                 do {
 945                         tips_running = 0;
 946                         usleep(100000);
 947
 948                         for_each_dip(dip, i)
 949                                 for_each_tip(dip, tip, j)
 950                                         tips_running += !tip->exited;
 951                 } while (tips_running);
 952         }
 953
 954         if (net_mode == Net_client)
 955                 net_client_send_close();
 956 }
 957
 958 static int fill_ofname(struct thread_information *tip, char *dst,
 959                        char *buts_name)
 960 {
 961         struct stat sb;
 962         int len = 0;
 963         time_t t;
 964
 965         if (output_dir)
 966                 len = sprintf(dst, "%s/", output_dir);
 967
 968         if (net_mode == Net_server) {
 969                 len += sprintf(dst + len, "%s-", inet_ntoa(tip->cl_in_addr));
 970                 time(&t);
 971                 len += strftime(dst + len, 64, "%F-%T/", gmtime(&t));
 972         }
 973
 974         if (stat(dst, &sb) < 0) {
 975                 if (errno != ENOENT) {
 976                         perror("stat");
 977                         return 1;
 978                 }
 979                 if (mkdir(dst, 0755) < 0) {
 980                         perror(dst);
 981                         fprintf(stderr, "Can't make output dir\n");
 982                         return 1;
 983                 }
 984         }
 985
 986         if (output_name)
 987                 sprintf(dst + len, "%s.blktrace.%d", output_name, tip->cpu);
 988         else
 989                 sprintf(dst + len, "%s.blktrace.%d", buts_name, tip->cpu);
 990
 991         return 0;
 992 }
 993
 994 static void fill_ops(struct thread_information *tip)
 995 {
 996         /*
 997          * setup ops
 998          */
 999         if (net_mode == Net_client) {
1000                 if (net_use_sendfile) {
1001                         tip->get_subbuf = get_subbuf_sendfile;
1002                         tip->flush_subbuf = flush_subbuf_sendfile;
1003                 } else {
1004                         tip->get_subbuf = get_subbuf;
1005                         tip->flush_subbuf = flush_subbuf_net;
1006                 }
1007         } else {
1008                 if (tip->ofile_mmap)
1009                         tip->get_subbuf = mmap_subbuf;
1010                 else
1011                         tip->get_subbuf = get_subbuf;
1012
1013                 tip->flush_subbuf = flush_subbuf_file;
1014         }
1015
1016         if (net_mode == Net_server)
1017                 tip->read_data = read_data_net;
1018         else
1019                 tip->read_data = read_data_file;
1020 }
1021
1022 static int tip_open_output(struct device_information *dip,
1023                            struct thread_information *tip)
1024 {
1025         int pipeline = output_name && !strcmp(output_name, "-");
1026         int mode, vbuf_size;
1027         char op[128];
1028
1029         if (net_mode == Net_client) {
1030                 tip->ofile = NULL;
1031                 tip->ofile_stdout = 0;
1032                 tip->ofile_mmap = 0;
1033                 goto done;
1034         } else if (pipeline) {
1035                 tip->ofile = fdopen(STDOUT_FILENO, "w");
1036                 tip->ofile_stdout = 1;
1037                 tip->ofile_mmap = 0;
1038                 mode = _IOLBF;
1039                 vbuf_size = 512;
1040         } else {
1041                 if (fill_ofname(tip, op, dip->buts_name))
1042                         return 1;
1043                 tip->ofile = fopen(op, "w+");
1044                 tip->ofile_stdout = 0;
1045                 tip->ofile_mmap = 1;
1046                 mode = _IOFBF;
1047                 vbuf_size = OFILE_BUF;
1048         }
1049
1050         if (tip->ofile == NULL) {
1051                 perror(op);
1052                 return 1;
1053         }
1054
1055         tip->ofile_buffer = malloc(vbuf_size);
1056         if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
1057                 perror("setvbuf");
1058                 close_thread(tip);
1059                 return 1;
1060         }
1061
1062 done:
1063         fill_ops(tip);
1064         return 0;
1065 }
1066
1067 static int start_threads(struct device_information *dip)
1068 {
1069         struct thread_information *tip;
1070         int j;
1071
1072         for_each_tip(dip, tip, j) {
1073                 tip->cpu = j;
1074                 tip->device = dip;
1075                 tip->events_processed = 0;
1076                 tip->fd = -1;
1077                 memset(&tip->fifo, 0, sizeof(tip->fifo));
1078                 tip->leftover_ts = NULL;
1079
1080                 if (tip_open_output(dip, tip))
1081                         return 1;
1082
1083                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
1084                         perror("pthread_create");
1085                         close_thread(tip);
1086                         return 1;
1087                 }
1088         }
1089
1090         return 0;
1091 }
1092
1093 static void stop_threads(struct device_information *dip)
1094 {
1095         struct thread_information *tip;
1096         unsigned long ret;
1097         int i;
1098
1099         for_each_tip(dip, tip, i) {
1100                 (void) pthread_join(tip->thread, (void *) &ret);
1101                 close_thread(tip);
1102         }
1103 }
1104
1105 static void stop_all_threads(void)
1106 {
1107         struct device_information *dip;
1108         int i;
1109
1110         for_each_dip(dip, i)
1111                 stop_threads(dip);
1112 }
1113
1114 static void stop_all_tracing(void)
1115 {
1116         struct device_information *dip;
1117         int i;
1118
1119         for_each_dip(dip, i)
1120                 stop_trace(dip);
1121 }
1122
1123 static void exit_trace(int status)
1124 {
1125         if (!is_trace_stopped()) {
1126                 trace_stopped = 1;
1127                 stop_all_threads();
1128                 stop_all_tracing();
1129         }
1130
1131         exit(status);
1132 }
1133
1134 static int resize_devices(char *path)
1135 {
1136         int size = (ndevs + 1) * sizeof(struct device_information);
1137
1138         device_information = realloc(device_information, size);
1139         if (!device_information) {
1140                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
1141                 return 1;
1142         }
1143         device_information[ndevs].path = path;
1144         ndevs++;
1145         return 0;
1146 }
1147
1148 static int open_devices(void)
1149 {
1150         struct device_information *dip;
1151         int i;
1152
1153         for_each_dip(dip, i) {
1154                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
1155                 if (dip->fd < 0) {
1156                         perror(dip->path);
1157                         return 1;
1158                 }
1159         }
1160
1161         return 0;
1162 }
1163
1164 static int start_devices(void)
1165 {
1166         struct device_information *dip;
1167         int i, j, size;
1168
1169         size = ncpus * sizeof(struct thread_information);
1170         thread_information = malloc(size * ndevs);
1171         if (!thread_information) {
1172                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
1173                 return 1;
1174         }
1175
1176         for_each_dip(dip, i) {
1177                 if (start_trace(dip)) {
1178                         close(dip->fd);
1179                         fprintf(stderr, "Failed to start trace on %s\n",
1180                                 dip->path);
1181                         break;
1182                 }
1183         }
1184
1185         if (i != ndevs) {
1186                 __for_each_dip(dip, j, i)
1187                         stop_trace(dip);
1188
1189                 return 1;
1190         }
1191
1192         for_each_dip(dip, i) {
1193                 dip->threads = thread_information + (i * ncpus);
1194                 if (start_threads(dip)) {
1195                         fprintf(stderr, "Failed to start worker threads\n");
1196                         break;
1197                 }
1198         }
1199
1200         if (i != ndevs) {
1201                 __for_each_dip(dip, j, i)
1202                         stop_threads(dip);
1203                 for_each_dip(dip, i)
1204                         stop_trace(dip);
1205
1206                 return 1;
1207         }
1208
1209         return 0;
1210 }
1211
1212 static void show_stats(void)
1213 {
1214         struct device_information *dip;
1215         struct thread_information *tip;
1216         unsigned long long events_processed, data_read;
1217         unsigned long total_drops;
1218         int i, j, no_stdout = 0;
1219
1220         if (is_stat_shown())
1221                 return;
1222
1223         if (output_name && !strcmp(output_name, "-"))
1224                 no_stdout = 1;
1225
1226         stat_shown = 1;
1227
1228         total_drops = 0;
1229         for_each_dip(dip, i) {
1230                 if (!no_stdout)
1231                         printf("Device: %s\n", dip->path);
1232                 events_processed = 0;
1233                 data_read = 0;
1234                 for_each_tip(dip, tip, j) {
1235                         if (!no_stdout)
1236                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
1237                                         tip->cpu, tip->events_processed,
1238                                         (tip->data_read + 1023) >> 10);
1239                         events_processed += tip->events_processed;
1240                         data_read += tip->data_read;
1241                 }
1242                 total_drops += dip->drop_count;
1243                 if (!no_stdout)
1244                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
1245                                         events_processed, dip->drop_count,
1246                                         (data_read + 1023) >> 10);
1247         }
1248
1249         if (total_drops)
1250                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1251 }
1252
1253 static struct device_information *net_get_dip(char *buts_name,
1254                                               struct in_addr *cl_in_addr)
1255 {
1256         struct device_information *dip;
1257         int i;
1258
1259         for (i = 0; i < ndevs; i++) {
1260                 dip = &device_information[i];
1261
1262                 if (!strcmp(dip->buts_name, buts_name))
1263                         return dip;
1264         }
1265
1266         device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
1267         dip = &device_information[ndevs];
1268         memset(dip, 0, sizeof(*dip));
1269         dip->fd = -1;
1270         strcpy(dip->buts_name, buts_name);
1271         dip->path = strdup(buts_name);
1272         dip->trace_started = 1;
1273         ndevs++;
1274         ndevs_running++;
1275         dip->threads = malloc(ncpus * sizeof(struct thread_information));
1276         memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
1277
1278         /*
1279          * open all files
1280          */
1281         for (i = 0; i < ncpus; i++) {
1282                 struct thread_information *tip = &dip->threads[i];
1283
1284                 tip->cpu = i;
1285                 tip->device = dip;
1286                 tip->fd = -1;
1287                 tip->cl_in_addr = *cl_in_addr;
1288
1289                 if (tip_open_output(dip, tip))
1290                         return NULL;
1291         }
1292
1293         return dip;
1294 }
1295
1296 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh,
1297                                               struct in_addr *cl_in_addr)
1298 {
1299         struct device_information *dip;
1300
1301         ncpus = bnh->max_cpus;
1302         dip = net_get_dip(bnh->buts_name, cl_in_addr);
1303         if (!dip->trace_started) {
1304                 fprintf(stderr, "Events for closed devices %s\n", dip->buts_name);
1305                 return NULL;
1306         }
1307
1308         return &dip->threads[bnh->cpu];
1309 }
1310
1311 static int net_get_header(struct blktrace_net_hdr *bnh)
1312 {
1313         int fl = fcntl(net_in_fd, F_GETFL);
1314         int bytes_left, ret;
1315         void *p = bnh;
1316
1317         fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1318         bytes_left = sizeof(*bnh);
1319         while (bytes_left && !is_done()) {
1320                 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1321                 if (ret < 0) {
1322                         if (errno != EAGAIN) {
1323                                 perror("recv header");
1324                                 return 1;
1325                         }
1326                         usleep(1000);
1327                         continue;
1328                 } else if (!ret) {
1329                         usleep(1000);
1330                         continue;
1331                 } else {
1332                         p += ret;
1333                         bytes_left -= ret;
1334                 }
1335         }
1336         fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1337         return bytes_left;
1338 }
1339
1340 static int net_server_loop(struct in_addr *cl_in_addr)
1341 {
1342         struct thread_information *tip;
1343         struct blktrace_net_hdr bnh;
1344
1345         if (net_get_header(&bnh))
1346                 return 1;
1347
1348         if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1349                 fprintf(stderr, "server: received data is bad\n");
1350                 return 1;
1351         }
1352
1353         if (!data_is_native) {
1354                 bnh.magic = be32_to_cpu(bnh.magic);
1355                 bnh.cpu = be32_to_cpu(bnh.cpu);
1356                 bnh.len = be32_to_cpu(bnh.len);
1357         }
1358
1359         if ((bnh.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1360                 fprintf(stderr, "server: bad data magic\n");
1361                 return 1;
1362         }
1363
1364         /*
1365          * len == 0 means that the other end signalled end-of-run
1366          */
1367         if (!bnh.len) {
1368                 /*
1369                  * overload cpu count with dropped events
1370                  */
1371                 struct device_information *dip;
1372
1373                 dip = net_get_dip(bnh.buts_name, cl_in_addr);
1374                 dip->drop_count = bnh.cpu;
1375                 dip->trace_started = 0;
1376                 ndevs_running--;
1377
1378                 fprintf(stderr, "server: end of run for %s\n", dip->buts_name);
1379                 return !ndevs_running;
1380         }
1381
1382         tip = net_get_tip(&bnh, cl_in_addr);
1383         if (!tip)
1384                 return 1;
1385
1386         if (mmap_subbuf(tip, bnh.len))
1387                 return 1;
1388
1389         return 0;
1390 }
1391
1392 static int get_connection(int fd, struct sockaddr_in *addr)
1393 {
1394         struct pollfd pfd = { .fd = fd, .events = POLLIN };
1395         socklen_t socklen;
1396
1397         printf("blktrace: waiting for incoming connection...\n");
1398
1399         if (poll(&pfd, 1, -1) < 0) {
1400                 perror("poll for connection");
1401                 return 1;
1402         }
1403         if ((pfd.revents & POLLIN) == 0)
1404                 return 1;
1405
1406         socklen = sizeof(*addr);
1407         net_in_fd = accept(fd, (struct sockaddr *) addr, &socklen);
1408         if (net_in_fd < 0) {
1409                 perror("accept");
1410                 return 1;
1411         }
1412
1413         printf("blktrace: connection from %s\n", inet_ntoa(addr->sin_addr));
1414         return 0;
1415 }
1416
1417 /*
1418  * Start here when we are in server mode - just fetch data from the network
1419  * and dump to files
1420  */
1421 static int net_server(void)
1422 {
1423         struct device_information *dip;
1424         struct thread_information *tip;
1425         struct sockaddr_in addr;
1426         int fd, opt, i, j;
1427
1428         fd = socket(AF_INET, SOCK_STREAM, 0);
1429         if (fd < 0) {
1430                 perror("server: socket");
1431                 return 1;
1432         }
1433
1434         opt = 1;
1435         if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1436                 perror("setsockopt");
1437                 return 1;
1438         }
1439
1440         memset(&addr, 0, sizeof(addr));
1441         addr.sin_family = AF_INET;
1442         addr.sin_addr.s_addr = htonl(INADDR_ANY);
1443         addr.sin_port = htons(net_port);
1444
1445         if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1446                 perror("bind");
1447                 return 1;
1448         }
1449
1450         if (listen(fd, 1) < 0) {
1451                 perror("listen");
1452                 return 1;
1453         }
1454
1455 repeat:
1456         if (get_connection(fd, &addr))
1457                 return 0;
1458
1459         while (!is_done()) {
1460                 if (net_server_loop(&addr.sin_addr))
1461                         break;
1462         }
1463
1464         for_each_dip(dip, i)
1465                 for_each_tip(dip, tip, j)
1466                         tip_ftrunc_final(tip);
1467
1468         show_stats();
1469
1470         if (is_done())
1471                 return 0;
1472
1473         /*
1474          * cleanup for next run
1475          */
1476         for_each_dip(dip, i) {
1477                 for_each_tip(dip, tip, j)
1478                         fclose(tip->ofile);
1479
1480                 free(dip->threads);
1481                 free(dip->path);
1482         }
1483
1484         free(device_information);
1485         device_information = NULL;
1486         ncpus = ndevs = 0;
1487
1488         close(net_in_fd);
1489         net_in_fd = -1;
1490         stat_shown = 0;
1491         goto repeat;
1492 }
1493
1494 /*
1495  * Setup outgoing network connection where we will transmit data
1496  */
1497 static int net_setup_client(void)
1498 {
1499         struct sockaddr_in addr;
1500         int fd;
1501
1502         fd = socket(AF_INET, SOCK_STREAM, 0);
1503         if (fd < 0) {
1504                 perror("client: socket");
1505                 return 1;
1506         }
1507
1508         memset(&addr, 0, sizeof(addr));
1509         addr.sin_family = AF_INET;
1510         addr.sin_port = htons(net_port);
1511
1512         if (inet_aton(hostname, &addr.sin_addr) != 1) {
1513                 struct hostent *hent = gethostbyname(hostname);
1514                 if (!hent) {
1515                         perror("gethostbyname");
1516                         return 1;
1517                 }
1518
1519                 memcpy(&addr.sin_addr, hent->h_addr, 4);
1520                 strcpy(hostname, hent->h_name);
1521         }
1522
1523         printf("blktrace: connecting to %s\n", hostname);
1524
1525         if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1526                 perror("client: connect");
1527                 return 1;
1528         }
1529
1530         printf("blktrace: connected!\n");
1531         net_out_fd = fd;
1532         return 0;
1533 }
1534
1535 static char usage_str[] = \
1536         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1537         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1538         "\t-d Use specified device. May also be given last after options\n" \
1539         "\t-r Path to mounted relayfs, defaults to /relay\n" \
1540         "\t-o File(s) to send output to\n" \
1541         "\t-D Directory to prepend to output file names\n" \
1542         "\t-k Kill a running trace\n" \
1543         "\t-w Stop after defined time, in seconds\n" \
1544         "\t-a Only trace specified actions. See documentation\n" \
1545         "\t-A Give trace mask as a single value. See documentation\n" \
1546         "\t-b Sub buffer size in KiB\n" \
1547         "\t-n Number of sub buffers\n" \
1548         "\t-l Run in network listen mode (blktrace server)\n" \
1549         "\t-h Run in network client mode, connecting to the given host\n" \
1550         "\t-p Network port to use (default 8462)\n" \
1551         "\t-s Make the network client use sendfile() to transfer data\n" \
1552         "\t-V Print program version info\n\n";
1553
1554 static void show_usage(char *program)
1555 {
1556         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1557 }
1558
1559 int main(int argc, char *argv[])
1560 {
1561         static char default_relay_path[] = "/relay";
1562         struct statfs st;
1563         int i, c;
1564         int stop_watch = 0;
1565         int act_mask_tmp = 0;
1566
1567         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1568                 switch (c) {
1569                 case 'a':
1570                         i = find_mask_map(optarg);
1571                         if (i < 0) {
1572                                 fprintf(stderr,"Invalid action mask %s\n",
1573                                         optarg);
1574                                 return 1;
1575                         }
1576                         act_mask_tmp |= i;
1577                         break;
1578
1579                 case 'A':
1580                         if ((sscanf(optarg, "%x", &i) != 1) ||
1581                                                         !valid_act_opt(i)) {
1582                                 fprintf(stderr,
1583                                         "Invalid set action mask %s/0x%x\n",
1584                                         optarg, i);
1585                                 return 1;
1586                         }
1587                         act_mask_tmp = i;
1588                         break;
1589
1590                 case 'd':
1591                         if (resize_devices(optarg) != 0)
1592                                 return 1;
1593                         break;
1594
1595                 case 'r':
1596                         relay_path = optarg;
1597                         break;
1598
1599                 case 'o':
1600                         output_name = optarg;
1601                         break;
1602                 case 'k':
1603                         kill_running_trace = 1;
1604                         break;
1605                 case 'w':
1606                         stop_watch = atoi(optarg);
1607                         if (stop_watch <= 0) {
1608                                 fprintf(stderr,
1609                                         "Invalid stopwatch value (%d secs)\n",
1610                                         stop_watch);
1611                                 return 1;
1612                         }
1613                         break;
1614                 case 'V':
1615                         printf("%s version %s\n", argv[0], blktrace_version);
1616                         return 0;
1617                 case 'b':
1618                         buf_size = strtoul(optarg, NULL, 10);
1619                         if (buf_size <= 0 || buf_size > 16*1024) {
1620                                 fprintf(stderr,
1621                                         "Invalid buffer size (%lu)\n",buf_size);
1622                                 return 1;
1623                         }
1624                         buf_size <<= 10;
1625                         break;
1626                 case 'n':
1627                         buf_nr = strtoul(optarg, NULL, 10);
1628                         if (buf_nr <= 0) {
1629                                 fprintf(stderr,
1630                                         "Invalid buffer nr (%lu)\n", buf_nr);
1631                                 return 1;
1632                         }
1633                         break;
1634                 case 'D':
1635                         output_dir = optarg;
1636                         break;
1637                 case 'h':
1638                         net_mode = Net_client;
1639                         strcpy(hostname, optarg);
1640                         break;
1641                 case 'l':
1642                         net_mode = Net_server;
1643                         break;
1644                 case 'p':
1645                         net_port = atoi(optarg);
1646                         break;
1647                 case 's':
1648                         net_use_sendfile = 1;
1649                         break;
1650                 default:
1651                         show_usage(argv[0]);
1652                         return 1;
1653                 }
1654         }
1655
1656         setlocale(LC_NUMERIC, "en_US");
1657
1658         page_size = getpagesize();
1659
1660         if (net_mode == Net_server)
1661                 return net_server();
1662
1663         while (optind < argc) {
1664                 if (resize_devices(argv[optind++]) != 0)
1665                         return 1;
1666         }
1667
1668         if (ndevs == 0) {
1669                 show_usage(argv[0]);
1670                 return 1;
1671         }
1672
1673         if (!relay_path)
1674                 relay_path = default_relay_path;
1675
1676         if (act_mask_tmp != 0)
1677                 act_mask = act_mask_tmp;
1678
1679         if (statfs(relay_path, &st) < 0) {
1680                 perror("statfs");
1681                 fprintf(stderr,"%s does not appear to be a valid path\n",
1682                         relay_path);
1683                 return 1;
1684         } else if (st.f_type != (long) RELAYFS_TYPE) {
1685                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1686                         relay_path);
1687                 return 1;
1688         }
1689
1690         if (open_devices() != 0)
1691                 return 1;
1692
1693         if (kill_running_trace) {
1694                 stop_all_traces();
1695                 return 0;
1696         }
1697
1698         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1699         if (ncpus < 0) {
1700                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1701                 return 1;
1702         }
1703
1704         signal(SIGINT, handle_sigint);
1705         signal(SIGHUP, handle_sigint);
1706         signal(SIGTERM, handle_sigint);
1707         signal(SIGALRM, handle_sigint);
1708
1709         if (net_mode == Net_client && net_setup_client())
1710                 return 1;
1711
1712         if (start_devices() != 0)
1713                 return 1;
1714
1715         atexit(stop_all_tracing);
1716
1717         if (stop_watch)
1718                 alarm(stop_watch);
1719
1720         wait_for_threads();
1721
1722         if (!is_trace_stopped()) {
1723                 trace_stopped = 1;
1724                 stop_all_threads();
1725                 stop_all_traces();
1726         }
1727
1728         show_stats();
1729
1730         return 0;
1731 }
1732