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         free(ts->buf);
 771         free(ts);
 772         return 0;
 773 }
 774
 775 static int flush_subbuf_sendfile(struct thread_information *tip,
 776                                  struct tip_subbuf *ts)
 777 {
 778         size_t padding;
 779         unsigned subbuf;
 780         unsigned len;
 781
 782         /*
 783          * currently we cannot use sendfile() on the last bytes read, as they
 784          * may not be a full subbuffer. get_subbuf_sendfile() falls back to
 785          * the read approach for those, so use send() to ship them out
 786          */
 787         if (ts->buf)
 788                 return flush_subbuf_net(tip, ts);
 789
 790         subbuf = (ts->offset / buf_size) % buf_nr;
 791         padding = get_subbuf_padding(tip, subbuf);
 792         len = ts->len - padding;
 793
 794         if (net_send_header(tip, len))
 795                 return 1;
 796         if (sendfile(net_out_fd, tip->fd, &ts->offset, len) < 0) {
 797                 perror("sendfile");
 798                 return 1;
 799         }
 800
 801         tip->data_read += len;
 802         free(ts);
 803         return 0;
 804 }
 805
 806 static int write_data(struct thread_information *tip, void *buf,
 807                       unsigned int buf_len)
 808 {
 809         int ret;
 810
 811         if (!buf_len)
 812                 return 0;
 813
 814         while (1) {
 815                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 816                 if (ret == 1)
 817                         break;
 818
 819                 if (ret < 0) {
 820                         perror("write");
 821                         return 1;
 822                 }
 823         }
 824
 825         if (tip->ofile_stdout)
 826                 fflush(tip->ofile);
 827
 828         return 0;
 829 }
 830
 831 static int flush_subbuf_file(struct thread_information *tip,
 832                              struct tip_subbuf *ts)
 833 {
 834         unsigned int offset = 0;
 835         struct blk_io_trace *t;
 836         int pdu_len, events = 0;
 837
 838         /*
 839          * surplus from last run
 840          */
 841         if (tip->leftover_ts) {
 842                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 843
 844                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 845                         prev_ts->max_len += ts->len;
 846                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 847                 }
 848
 849                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 850                 prev_ts->len += ts->len;
 851
 852                 free(ts->buf);
 853                 free(ts);
 854
 855                 ts = prev_ts;
 856                 tip->leftover_ts = NULL;
 857         }
 858
 859         while (offset + sizeof(*t) <= ts->len) {
 860                 t = ts->buf + offset;
 861
 862                 if (verify_trace(t)) {
 863                         write_data(tip, ts->buf, offset);
 864                         return -1;
 865                 }
 866
 867                 pdu_len = t->pdu_len;
 868
 869                 if (offset + sizeof(*t) + pdu_len > ts->len)
 870                         break;
 871
 872                 offset += sizeof(*t) + pdu_len;
 873                 tip->events_processed++;
 874                 tip->data_read += sizeof(*t) + pdu_len;
 875                 events++;
 876         }
 877
 878         if (write_data(tip, ts->buf, offset))
 879                 return -1;
 880
 881         /*
 882          * leftover bytes, save them for next time
 883          */
 884         if (offset != ts->len) {
 885                 tip->leftover_ts = ts;
 886                 ts->len -= offset;
 887                 memmove(ts->buf, ts->buf + offset, ts->len);
 888         } else {
 889                 free(ts->buf);
 890                 free(ts);
 891         }
 892
 893         return events;
 894 }
 895
 896 static int write_tip_events(struct thread_information *tip)
 897 {
 898         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 899
 900         if (ts)
 901                 return tip->flush_subbuf(tip, ts);
 902
 903         return 0;
 904 }
 905
 906 /*
 907  * scans the tips we know and writes out the subbuffers we accumulate
 908  */
 909 static void get_and_write_events(void)
 910 {
 911         struct device_information *dip;
 912         struct thread_information *tip;
 913         int i, j, events, ret, tips_running;
 914
 915         while (!is_done()) {
 916                 events = 0;
 917
 918                 for_each_dip(dip, i) {
 919                         for_each_tip(dip, tip, j) {
 920                                 ret = write_tip_events(tip);
 921                                 if (ret > 0)
 922                                         events += ret;
 923                         }
 924                 }
 925
 926                 if (!events)
 927                         usleep(10);
 928         }
 929
 930         /*
 931          * reap stored events
 932          */
 933         do {
 934                 events = 0;
 935                 tips_running = 0;
 936                 for_each_dip(dip, i) {
 937                         for_each_tip(dip, tip, j) {
 938                                 ret = write_tip_events(tip);
 939                                 if (ret > 0)
 940                                         events += ret;
 941                                 tips_running += !tip->exited;
 942                         }
 943                 }
 944                 usleep(10);
 945         } while (events || tips_running);
 946 }
 947
 948 static void wait_for_threads(void)
 949 {
 950         /*
 951          * for piped or network output, poll and fetch data for writeout.
 952          * for files, we just wait around for trace threads to exit
 953          */
 954         if ((output_name && !strcmp(output_name, "-")) ||
 955             net_mode == Net_client)
 956                 get_and_write_events();
 957         else {
 958                 struct device_information *dip;
 959                 struct thread_information *tip;
 960                 int i, j, tips_running;
 961
 962                 do {
 963                         tips_running = 0;
 964                         usleep(1000);
 965
 966                         for_each_dip(dip, i)
 967                                 for_each_tip(dip, tip, j)
 968                                         tips_running += !tip->exited;
 969                 } while (tips_running);
 970         }
 971
 972         if (net_mode == Net_client)
 973                 net_client_send_close();
 974 }
 975
 976 static void fill_ofname(char *dst, char *buts_name, int cpu)
 977 {
 978         int len = 0;
 979
 980         if (output_dir)
 981                 len = sprintf(dst, "%s/", output_dir);
 982
 983         if (output_name)
 984                 sprintf(dst + len, "%s.blktrace.%d", output_name, cpu);
 985         else
 986                 sprintf(dst + len, "%s.blktrace.%d", buts_name, cpu);
 987 }
 988
 989 static void fill_ops(struct thread_information *tip)
 990 {
 991         /*
 992          * setup ops
 993          */
 994         if (net_mode == Net_client) {
 995                 if (net_sendfile) {
 996                         tip->get_subbuf = get_subbuf_sendfile;
 997                         tip->flush_subbuf = flush_subbuf_sendfile;
 998                 } else {
 999                         tip->get_subbuf = get_subbuf;
1000                         tip->flush_subbuf = flush_subbuf_net;
1001                 }
1002         } else {
1003                 if (tip->ofile_mmap)
1004                         tip->get_subbuf = mmap_subbuf;
1005                 else
1006                         tip->get_subbuf = get_subbuf;
1007
1008                 tip->flush_subbuf = flush_subbuf_file;
1009         }
1010
1011         if (net_mode == Net_server)
1012                 tip->read_data = read_data_net;
1013         else
1014                 tip->read_data = read_data_file;
1015 }
1016
1017 static int tip_open_output(struct device_information *dip,
1018                            struct thread_information *tip)
1019 {
1020         int pipeline = output_name && !strcmp(output_name, "-");
1021         int mode, vbuf_size;
1022         char op[64];
1023
1024         if (net_mode == Net_client) {
1025                 tip->ofile = NULL;
1026                 tip->ofile_stdout = 0;
1027                 tip->ofile_mmap = 0;
1028                 vbuf_size = 0;
1029                 mode = 0; /* gcc 4.x issues a bogus warning */
1030         } else if (pipeline) {
1031                 tip->ofile = fdopen(STDOUT_FILENO, "w");
1032                 tip->ofile_stdout = 1;
1033                 tip->ofile_mmap = 0;
1034                 mode = _IOLBF;
1035                 vbuf_size = 512;
1036         } else {
1037                 fill_ofname(op, dip->buts_name, tip->cpu);
1038                 tip->ofile = fopen(op, "w+");
1039                 tip->ofile_stdout = 0;
1040                 tip->ofile_mmap = 1;
1041                 mode = _IOFBF;
1042                 vbuf_size = OFILE_BUF;
1043         }
1044
1045         if (net_mode != Net_client && tip->ofile == NULL) {
1046                 perror(op);
1047                 return 1;
1048         }
1049
1050         if (vbuf_size) {
1051                 tip->ofile_buffer = malloc(vbuf_size);
1052                 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
1053                         perror("setvbuf");
1054                         close_thread(tip);
1055                         return 1;
1056                 }
1057         }
1058
1059         fill_ops(tip);
1060         return 0;
1061 }
1062
1063 static int start_threads(struct device_information *dip)
1064 {
1065         struct thread_information *tip;
1066         int j;
1067
1068         for_each_tip(dip, tip, j) {
1069                 tip->cpu = j;
1070                 tip->device = dip;
1071                 tip->events_processed = 0;
1072                 tip->fd = -1;
1073                 tip->pfd = -1;
1074                 memset(&tip->fifo, 0, sizeof(tip->fifo));
1075                 tip->leftover_ts = NULL;
1076
1077                 if (tip_open_output(dip, tip))
1078                         return 1;
1079
1080                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
1081                         perror("pthread_create");
1082                         close_thread(tip);
1083                         return 1;
1084                 }
1085         }
1086
1087         return 0;
1088 }
1089
1090 static void stop_threads(struct device_information *dip)
1091 {
1092         struct thread_information *tip;
1093         unsigned long ret;
1094         int i;
1095
1096         for_each_tip(dip, tip, i) {
1097                 (void) pthread_join(tip->thread, (void *) &ret);
1098                 close_thread(tip);
1099         }
1100 }
1101
1102 static void stop_all_threads(void)
1103 {
1104         struct device_information *dip;
1105         int i;
1106
1107         for_each_dip(dip, i)
1108                 stop_threads(dip);
1109 }
1110
1111 static void stop_all_tracing(void)
1112 {
1113         struct device_information *dip;
1114         int i;
1115
1116         for_each_dip(dip, i)
1117                 stop_trace(dip);
1118 }
1119
1120 static void exit_trace(int status)
1121 {
1122         if (!is_trace_stopped()) {
1123                 trace_stopped = 1;
1124                 stop_all_threads();
1125                 stop_all_tracing();
1126         }
1127
1128         exit(status);
1129 }
1130
1131 static int resize_devices(char *path)
1132 {
1133         int size = (ndevs + 1) * sizeof(struct device_information);
1134
1135         device_information = realloc(device_information, size);
1136         if (!device_information) {
1137                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
1138                 return 1;
1139         }
1140         device_information[ndevs].path = path;
1141         ndevs++;
1142         return 0;
1143 }
1144
1145 static int open_devices(void)
1146 {
1147         struct device_information *dip;
1148         int i;
1149
1150         for_each_dip(dip, i) {
1151                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
1152                 if (dip->fd < 0) {
1153                         perror(dip->path);
1154                         return 1;
1155                 }
1156         }
1157
1158         return 0;
1159 }
1160
1161 static int start_devices(void)
1162 {
1163         struct device_information *dip;
1164         int i, j, size;
1165
1166         size = ncpus * sizeof(struct thread_information);
1167         thread_information = malloc(size * ndevs);
1168         if (!thread_information) {
1169                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
1170                 return 1;
1171         }
1172
1173         for_each_dip(dip, i) {
1174                 if (start_trace(dip)) {
1175                         close(dip->fd);
1176                         fprintf(stderr, "Failed to start trace on %s\n",
1177                                 dip->path);
1178                         break;
1179                 }
1180         }
1181
1182         if (i != ndevs) {
1183                 __for_each_dip(dip, j, i)
1184                         stop_trace(dip);
1185
1186                 return 1;
1187         }
1188
1189         for_each_dip(dip, i) {
1190                 dip->threads = thread_information + (i * ncpus);
1191                 if (start_threads(dip)) {
1192                         fprintf(stderr, "Failed to start worker threads\n");
1193                         break;
1194                 }
1195         }
1196
1197         if (i != ndevs) {
1198                 __for_each_dip(dip, j, i)
1199                         stop_threads(dip);
1200                 for_each_dip(dip, i)
1201                         stop_trace(dip);
1202
1203                 return 1;
1204         }
1205
1206         return 0;
1207 }
1208
1209 static void show_stats(void)
1210 {
1211         struct device_information *dip;
1212         struct thread_information *tip;
1213         unsigned long long events_processed, data_read;
1214         unsigned long total_drops;
1215         int i, j, no_stdout = 0;
1216
1217         if (is_stat_shown())
1218                 return;
1219
1220         if (output_name && !strcmp(output_name, "-"))
1221                 no_stdout = 1;
1222
1223         stat_shown = 1;
1224
1225         total_drops = 0;
1226         for_each_dip(dip, i) {
1227                 if (!no_stdout)
1228                         printf("Device: %s\n", dip->path);
1229                 events_processed = 0;
1230                 data_read = 0;
1231                 for_each_tip(dip, tip, j) {
1232                         if (!no_stdout)
1233                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
1234                                         tip->cpu, tip->events_processed,
1235                                         (tip->data_read + 1023) >> 10);
1236                         events_processed += tip->events_processed;
1237                         data_read += tip->data_read;
1238                 }
1239                 total_drops += dip->drop_count;
1240                 if (!no_stdout)
1241                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
1242                                         events_processed, dip->drop_count,
1243                                         (data_read + 1023) >> 10);
1244         }
1245
1246         if (total_drops)
1247                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1248 }
1249
1250 static struct device_information *net_get_dip(char *buts_name)
1251 {
1252         struct device_information *dip;
1253         int i;
1254
1255         for (i = 0; i < ndevs; i++) {
1256                 dip = &device_information[i];
1257
1258                 if (!strcmp(dip->buts_name, buts_name))
1259                         return dip;
1260         }
1261
1262         device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
1263         dip = &device_information[ndevs];
1264         strcpy(dip->buts_name, buts_name);
1265         strcpy(dip->path, buts_name);
1266         ndevs++;
1267         dip->threads = malloc(ncpus * sizeof(struct thread_information));
1268         memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
1269
1270         /*
1271          * open all files
1272          */
1273         for (i = 0; i < ncpus; i++) {
1274                 struct thread_information *tip = &dip->threads[i];
1275
1276                 tip->cpu = i;
1277                 tip->device = dip;
1278                 tip->fd = -1;
1279                 tip->pfd = -1;
1280
1281                 if (tip_open_output(dip, tip))
1282                         return NULL;
1283         }
1284
1285         return dip;
1286 }
1287
1288 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh)
1289 {
1290         struct device_information *dip;
1291
1292         ncpus = bnh->max_cpus;
1293         dip = net_get_dip(bnh->buts_name);
1294         return &dip->threads[bnh->cpu];
1295 }
1296
1297 static int net_get_header(struct blktrace_net_hdr *bnh)
1298 {
1299         int fl = fcntl(net_in_fd, F_GETFL);
1300         int bytes_left, ret;
1301         void *p = bnh;
1302
1303         fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1304         bytes_left = sizeof(*bnh);
1305         while (bytes_left && !is_done()) {
1306                 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1307                 if (ret < 0) {
1308                         if (errno != EAGAIN) {
1309                                 perror("recv header");
1310                                 return 1;
1311                         }
1312                         usleep(100);
1313                         continue;
1314                 } else if (!ret) {
1315                         usleep(100);
1316                         continue;
1317                 } else {
1318                         p += ret;
1319                         bytes_left -= ret;
1320                 }
1321         }
1322         fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1323         return 0;
1324 }
1325
1326 static int net_server_loop(void)
1327 {
1328         struct thread_information *tip;
1329         struct blktrace_net_hdr bnh;
1330
1331         if (net_get_header(&bnh))
1332                 return 1;
1333
1334         if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1335                 fprintf(stderr, "server: received data is bad\n");
1336                 return 1;
1337         }
1338
1339         if (!data_is_native) {
1340                 bnh.cpu = be32_to_cpu(bnh.cpu);
1341                 bnh.len = be32_to_cpu(bnh.len);
1342         }
1343
1344         /*
1345          * len == 0 means that the other end signalled end-of-run
1346          */
1347         if (!bnh.len) {
1348                 fprintf(stderr, "server: end of run\n");
1349                 return 1;
1350         }
1351
1352         tip = net_get_tip(&bnh);
1353         if (!tip)
1354                 return 1;
1355
1356         if (mmap_subbuf(tip, bnh.len))
1357                 return 1;
1358
1359         return 0;
1360 }
1361
1362 /*
1363  * Start here when we are in server mode - just fetch data from the network
1364  * and dump to files
1365  */
1366 static int net_server(void)
1367 {
1368         struct device_information *dip;
1369         struct thread_information *tip;
1370         struct sockaddr_in addr;
1371         socklen_t socklen;
1372         int fd, opt, i, j;
1373
1374         fd = socket(AF_INET, SOCK_STREAM, 0);
1375         if (fd < 0) {
1376                 perror("server: socket");
1377                 return 1;
1378         }
1379
1380         opt = 1;
1381         if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1382                 perror("setsockopt");
1383                 return 1;
1384         }
1385
1386         memset(&addr, 0, sizeof(addr));
1387         addr.sin_family = AF_INET;
1388         addr.sin_addr.s_addr = htonl(INADDR_ANY);
1389         addr.sin_port = htons(net_port);
1390
1391         if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1392                 perror("bind");
1393                 return 1;
1394         }
1395
1396         if (listen(fd, 1) < 0) {
1397                 perror("listen");
1398                 return 1;
1399         }
1400
1401 repeat:
1402         signal(SIGINT, NULL);
1403         signal(SIGHUP, NULL);
1404         signal(SIGTERM, NULL);
1405         signal(SIGALRM, NULL);
1406
1407         printf("blktrace: waiting for incoming connection...\n");
1408
1409         socklen = sizeof(addr);
1410         net_in_fd = accept(fd, (struct sockaddr *) &addr, &socklen);
1411         if (net_in_fd < 0) {
1412                 perror("accept");
1413                 return 1;
1414         }
1415
1416         signal(SIGINT, handle_sigint);
1417         signal(SIGHUP, handle_sigint);
1418         signal(SIGTERM, handle_sigint);
1419         signal(SIGALRM, handle_sigint);
1420
1421         printf("blktrace: connection from %s\n", inet_ntoa(addr.sin_addr));
1422
1423         while (!is_done()) {
1424                 if (net_server_loop())
1425                         break;
1426         }
1427
1428         for_each_dip(dip, i)
1429                 for_each_tip(dip, tip, j)
1430                         tip_ftrunc_final(tip);
1431
1432         show_stats();
1433
1434         if (is_done())
1435                 return 0;
1436
1437         /*
1438          * cleanup for next run
1439          */
1440         for_each_dip(dip, i) {
1441                 for_each_tip(dip, tip, j)
1442                         fclose(tip->ofile);
1443
1444                 free(dip->threads);
1445         }
1446
1447         free(device_information);
1448         device_information = NULL;
1449         ncpus = ndevs = 0;
1450
1451         close(net_in_fd);
1452         net_in_fd = -1;
1453         stat_shown = 0;
1454         goto repeat;
1455 }
1456
1457 /*
1458  * Setup outgoing network connection where we will transmit data
1459  */
1460 static int net_setup_client(void)
1461 {
1462         struct sockaddr_in addr;
1463         int fd;
1464
1465         fd = socket(AF_INET, SOCK_STREAM, 0);
1466         if (fd < 0) {
1467                 perror("client: socket");
1468                 return 1;
1469         }
1470
1471         memset(&addr, 0, sizeof(addr));
1472         addr.sin_family = AF_INET;
1473         addr.sin_port = htons(net_port);
1474
1475         if (inet_aton(hostname, &addr.sin_addr) != 1) {
1476                 struct hostent *hent = gethostbyname(hostname);
1477                 if (!hent) {
1478                         perror("gethostbyname");
1479                         return 1;
1480                 }
1481
1482                 memcpy(&addr.sin_addr, hent->h_addr, 4);
1483                 strcpy(hostname, hent->h_name);
1484         }
1485
1486         printf("blktrace: connecting to %s\n", hostname);
1487
1488         if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1489                 perror("client: connect");
1490                 return 1;
1491         }
1492
1493         printf("blktrace: connected!\n");
1494         net_out_fd = fd;
1495         return 0;
1496 }
1497
1498 static char usage_str[] = \
1499         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1500         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1501         "\t-d Use specified device. May also be given last after options\n" \
1502         "\t-r Path to mounted relayfs, defaults to /relay\n" \
1503         "\t-o File(s) to send output to\n" \
1504         "\t-D Directory to prepend to output file names\n" \
1505         "\t-k Kill a running trace\n" \
1506         "\t-w Stop after defined time, in seconds\n" \
1507         "\t-a Only trace specified actions. See documentation\n" \
1508         "\t-A Give trace mask as a single value. See documentation\n" \
1509         "\t-b Sub buffer size in KiB\n" \
1510         "\t-n Number of sub buffers\n" \
1511         "\t-l Run in network listen mode (blktrace server)\n" \
1512         "\t-h Run in network client mode, connecting to the given host\n" \
1513         "\t-p Network port to use (default 8462)\n" \
1514         "\t-s Make the network client use sendfile() to transfer data\n" \
1515         "\t-V Print program version info\n\n";
1516
1517 static void show_usage(char *program)
1518 {
1519         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1520 }
1521
1522 int main(int argc, char *argv[])
1523 {
1524         static char default_relay_path[] = "/relay";
1525         struct statfs st;
1526         int i, c;
1527         int stop_watch = 0;
1528         int act_mask_tmp = 0;
1529
1530         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1531                 switch (c) {
1532                 case 'a':
1533                         i = find_mask_map(optarg);
1534                         if (i < 0) {
1535                                 fprintf(stderr,"Invalid action mask %s\n",
1536                                         optarg);
1537                                 return 1;
1538                         }
1539                         act_mask_tmp |= i;
1540                         break;
1541
1542                 case 'A':
1543                         if ((sscanf(optarg, "%x", &i) != 1) ||
1544                                                         !valid_act_opt(i)) {
1545                                 fprintf(stderr,
1546                                         "Invalid set action mask %s/0x%x\n",
1547                                         optarg, i);
1548                                 return 1;
1549                         }
1550                         act_mask_tmp = i;
1551                         break;
1552
1553                 case 'd':
1554                         if (resize_devices(optarg) != 0)
1555                                 return 1;
1556                         break;
1557
1558                 case 'r':
1559                         relay_path = optarg;
1560                         break;
1561
1562                 case 'o':
1563                         output_name = optarg;
1564                         break;
1565                 case 'k':
1566                         kill_running_trace = 1;
1567                         break;
1568                 case 'w':
1569                         stop_watch = atoi(optarg);
1570                         if (stop_watch <= 0) {
1571                                 fprintf(stderr,
1572                                         "Invalid stopwatch value (%d secs)\n",
1573                                         stop_watch);
1574                                 return 1;
1575                         }
1576                         break;
1577                 case 'V':
1578                         printf("%s version %s\n", argv[0], blktrace_version);
1579                         return 0;
1580                 case 'b':
1581                         buf_size = strtoul(optarg, NULL, 10);
1582                         if (buf_size <= 0 || buf_size > 16*1024) {
1583                                 fprintf(stderr,
1584                                         "Invalid buffer size (%lu)\n",buf_size);
1585                                 return 1;
1586                         }
1587                         buf_size <<= 10;
1588                         break;
1589                 case 'n':
1590                         buf_nr = strtoul(optarg, NULL, 10);
1591                         if (buf_nr <= 0) {
1592                                 fprintf(stderr,
1593                                         "Invalid buffer nr (%lu)\n", buf_nr);
1594                                 return 1;
1595                         }
1596                         break;
1597                 case 'D':
1598                         output_dir = optarg;
1599                         break;
1600                 case 'h':
1601                         net_mode = Net_client;
1602                         strcpy(hostname, optarg);
1603                         break;
1604                 case 'l':
1605                         net_mode = Net_server;
1606                         break;
1607                 case 'p':
1608                         net_port = atoi(optarg);
1609                         break;
1610                 case 's':
1611                         net_sendfile = 1;
1612                         break;
1613                 default:
1614                         show_usage(argv[0]);
1615                         return 1;
1616                 }
1617         }
1618
1619         setlocale(LC_NUMERIC, "en_US");
1620
1621         page_size = getpagesize();
1622
1623         if (net_mode == Net_server)
1624                 return net_server();
1625
1626         while (optind < argc) {
1627                 if (resize_devices(argv[optind++]) != 0)
1628                         return 1;
1629         }
1630
1631         if (ndevs == 0) {
1632                 show_usage(argv[0]);
1633                 return 1;
1634         }
1635
1636         if (!relay_path)
1637                 relay_path = default_relay_path;
1638
1639         if (act_mask_tmp != 0)
1640                 act_mask = act_mask_tmp;
1641
1642         if (statfs(relay_path, &st) < 0) {
1643                 perror("statfs");
1644                 fprintf(stderr,"%s does not appear to be a valid path\n",
1645                         relay_path);
1646                 return 1;
1647         } else if (st.f_type != (long) RELAYFS_TYPE) {
1648                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1649                         relay_path);
1650                 return 1;
1651         }
1652
1653         if (open_devices() != 0)
1654                 return 1;
1655
1656         if (kill_running_trace) {
1657                 stop_all_traces();
1658                 return 0;
1659         }
1660
1661         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1662         if (ncpus < 0) {
1663                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1664                 return 1;
1665         }
1666
1667         signal(SIGINT, handle_sigint);
1668         signal(SIGHUP, handle_sigint);
1669         signal(SIGTERM, handle_sigint);
1670         signal(SIGALRM, handle_sigint);
1671
1672         if (net_mode == Net_client && net_setup_client())
1673                 return 1;
1674
1675         if (start_devices() != 0)
1676                 return 1;
1677
1678         atexit(stop_all_tracing);
1679
1680         if (stop_watch)
1681                 alarm(stop_watch);
1682
1683         wait_for_threads();
1684
1685         if (!is_trace_stopped()) {
1686                 trace_stopped = 1;
1687                 stop_all_threads();
1688                 stop_all_traces();
1689         }
1690
1691         show_stats();
1692
1693         return 0;
1694 }
1695