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