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