blktrace.c

   1 /*
   2  * block queue tracing application
   3  *
   4  * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; either version 2 of the License, or
   9  *  (at your option) any later version.
  10  *
  11  *  This program is distributed in the hope that it will be useful,
  12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  *  GNU General Public License for more details.
  15  *
  16  *  You should have received a copy of the GNU General Public License
  17  *  along with this program; if not, write to the Free Software
  18  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19  *
  20  */
  21 #include <pthread.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <unistd.h>
  25 #include <locale.h>
  26 #include <signal.h>
  27 #include <fcntl.h>
  28 #include <string.h>
  29 #include <sys/ioctl.h>
  30 #include <sys/param.h>
  31 #include <sys/statfs.h>
  32 #include <sys/poll.h>
  33 #include <sys/mman.h>
  34 #include <sys/socket.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <sched.h>
  38 #include <ctype.h>
  39 #include <getopt.h>
  40 #include <errno.h>
  41 #include <netinet/in.h>
  42 #include <arpa/inet.h>
  43 #include <netdb.h>
  44 #include <sys/sendfile.h>
  45
  46 #include "blktrace.h"
  47 #include "barrier.h"
  48
  49 static char blktrace_version[] = "0.99";
  50
  51 /*
  52  * You may want to increase this even more, if you are logging at a high
  53  * rate and see skipped/missed events
  54  */
  55 #define BUF_SIZE        (512 * 1024)
  56 #define BUF_NR          (4)
  57
  58 #define OFILE_BUF       (128 * 1024)
  59
  60 #define RELAYFS_TYPE    0xF0B4A981
  61
  62 #define S_OPTS  "d:a:A:r:o:kw:Vb:n:D:lh:p:s"
  63 static struct option l_opts[] = {
  64         {
  65                 .name = "dev",
  66                 .has_arg = required_argument,
  67                 .flag = NULL,
  68                 .val = 'd'
  69         },
  70         {
  71                 .name = "act-mask",
  72                 .has_arg = required_argument,
  73                 .flag = NULL,
  74                 .val = 'a'
  75         },
  76         {
  77                 .name = "set-mask",
  78                 .has_arg = required_argument,
  79                 .flag = NULL,
  80                 .val = 'A'
  81         },
  82         {
  83                 .name = "relay",
  84                 .has_arg = required_argument,
  85                 .flag = NULL,
  86                 .val = 'r'
  87         },
  88         {
  89                 .name = "output",
  90                 .has_arg = required_argument,
  91                 .flag = NULL,
  92                 .val = 'o'
  93         },
  94         {
  95                 .name = "kill",
  96                 .has_arg = no_argument,
  97                 .flag = NULL,
  98                 .val = 'k'
  99         },
 100         {
 101                 .name = "stopwatch",
 102                 .has_arg = required_argument,
 103                 .flag = NULL,
 104                 .val = 'w'
 105         },
 106         {
 107                 .name = "version",
 108                 .has_arg = no_argument,
 109                 .flag = NULL,
 110                 .val = 'V'
 111         },
 112         {
 113                 .name = "buffer-size",
 114                 .has_arg = required_argument,
 115                 .flag = NULL,
 116                 .val = 'b'
 117         },
 118         {
 119                 .name = "num-sub-buffers",
 120                 .has_arg = required_argument,
 121                 .flag = NULL,
 122                 .val = 'n'
 123         },
 124         {
 125                 .name = "output-dir",
 126                 .has_arg = required_argument,
 127                 .flag = NULL,
 128                 .val = 'D'
 129         },
 130         {
 131                 .name = "listen",
 132                 .has_arg = no_argument,
 133                 .flag = NULL,
 134                 .val = 'l'
 135         },
 136         {
 137                 .name = "host",
 138                 .has_arg = required_argument,
 139                 .flag = NULL,
 140                 .val = 'h'
 141         },
 142         {
 143                 .name = "port",
 144                 .has_arg = required_argument,
 145                 .flag = NULL,
 146                 .val = 'p'
 147         },
 148         {
 149                 .name = "sendfile",
 150                 .has_arg = no_argument,
 151                 .flag = NULL,
 152                 .val = 's'
 153         },
 154         {
 155                 .name = NULL,
 156         }
 157 };
 158
 159 struct tip_subbuf {
 160         void *buf;
 161         unsigned int len;
 162         unsigned int max_len;
 163 };
 164
 165 #define FIFO_SIZE       (1024)  /* should be plenty big! */
 166 #define CL_SIZE         (128)   /* cache line, any bigger? */
 167
 168 struct tip_subbuf_fifo {
 169         int tail __attribute__((aligned(CL_SIZE)));
 170         int head __attribute__((aligned(CL_SIZE)));
 171         struct tip_subbuf *q[FIFO_SIZE];
 172 };
 173
 174 struct thread_information {
 175         int cpu;
 176         pthread_t thread;
 177
 178         int fd;
 179         void *fd_buf;
 180         char fn[MAXPATHLEN + 64];
 181
 182         struct in_addr cl_in_addr;
 183
 184         FILE *ofile;
 185         char *ofile_buffer;
 186         off_t ofile_offset;
 187         int ofile_stdout;
 188         int ofile_mmap;
 189         volatile int sendfile_pending;
 190
 191         int (*get_subbuf)(struct thread_information *, unsigned int);
 192         int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *);
 193         int (*read_data)(struct thread_information *, void *, unsigned int);
 194
 195         unsigned long events_processed;
 196         unsigned long long data_read;
 197         unsigned long long data_queued;
 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_use_sendfile;
 292
 293 static int net_in_fd = -1;
 294 static int net_out_fd = -1;
 295
 296 static void handle_sigint(__attribute__((__unused__)) int sig)
 297 {
 298         struct device_information *dip;
 299         int i;
 300
 301         /*
 302          * stop trace so we can reap currently produced data
 303          */
 304         for_each_dip(dip, i) {
 305                 if (dip->fd == -1)
 306                         continue;
 307                 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
 308                         perror("BLKTRACESTOP");
 309         }
 310
 311         done = 1;
 312 }
 313
 314 static int get_dropped_count(const char *buts_name)
 315 {
 316         int fd;
 317         char tmp[MAXPATHLEN + 64];
 318
 319         snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
 320                  relay_path, buts_name);
 321
 322         fd = open(tmp, O_RDONLY);
 323         if (fd < 0) {
 324                 /*
 325                  * this may be ok, if the kernel doesn't support dropped counts
 326                  */
 327                 if (errno == ENOENT)
 328                         return 0;
 329
 330                 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
 331                 return -1;
 332         }
 333
 334         if (read(fd, tmp, sizeof(tmp)) < 0) {
 335                 perror(tmp);
 336                 close(fd);
 337                 return -1;
 338         }
 339
 340         close(fd);
 341
 342         return atoi(tmp);
 343 }
 344
 345 static int start_trace(struct device_information *dip)
 346 {
 347         struct blk_user_trace_setup buts;
 348
 349         memset(&buts, 0, sizeof(buts));
 350         buts.buf_size = buf_size;
 351         buts.buf_nr = buf_nr;
 352         buts.act_mask = act_mask;
 353
 354         if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
 355                 perror("BLKTRACESETUP");
 356                 return 1;
 357         }
 358
 359         if (ioctl(dip->fd, BLKTRACESTART) < 0) {
 360                 perror("BLKTRACESTART");
 361                 return 1;
 362         }
 363
 364         memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
 365         dip_set_tracing(dip, 1);
 366         return 0;
 367 }
 368
 369 static void stop_trace(struct device_information *dip)
 370 {
 371         if (dip_tracing(dip) || kill_running_trace) {
 372                 dip_set_tracing(dip, 0);
 373
 374                 /*
 375                  * should be stopped, just don't complain if it isn't
 376                  */
 377                 ioctl(dip->fd, BLKTRACESTOP);
 378
 379                 if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
 380                         perror("BLKTRACETEARDOWN");
 381
 382                 close(dip->fd);
 383                 dip->fd = -1;
 384         }
 385 }
 386
 387 static void stop_all_traces(void)
 388 {
 389         struct device_information *dip;
 390         int i;
 391
 392         for_each_dip(dip, i) {
 393                 dip->drop_count = get_dropped_count(dip->buts_name);
 394                 stop_trace(dip);
 395         }
 396 }
 397
 398 static void wait_for_data(struct thread_information *tip)
 399 {
 400         struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
 401
 402         do {
 403                 poll(&pfd, 1, 100);
 404                 if (pfd.revents & POLLIN)
 405                         break;
 406                 if (tip->ofile_stdout)
 407                         break;
 408         } while (!is_done());
 409 }
 410
 411 static int read_data_file(struct thread_information *tip, void *buf,
 412                           unsigned int len)
 413 {
 414         int ret = 0;
 415
 416         do {
 417                 wait_for_data(tip);
 418
 419                 ret = read(tip->fd, buf, len);
 420                 if (!ret)
 421                         continue;
 422                 else if (ret > 0)
 423                         return ret;
 424                 else {
 425                         if (errno != EAGAIN) {
 426                                 perror(tip->fn);
 427                                 fprintf(stderr,"Thread %d failed read of %s\n",
 428                                         tip->cpu, tip->fn);
 429                                 break;
 430                         }
 431                         continue;
 432                 }
 433         } while (!is_done());
 434
 435         return ret;
 436
 437 }
 438
 439 static int read_data_net(struct thread_information *tip, void *buf,
 440                          unsigned int len)
 441 {
 442         unsigned int bytes_left = len;
 443         int ret = 0;
 444
 445         do {
 446                 ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
 447
 448                 if (!ret)
 449                         continue;
 450                 else if (ret < 0) {
 451                         if (errno != EAGAIN) {
 452                                 perror(tip->fn);
 453                                 fprintf(stderr, "server: failed read\n");
 454                                 return 0;
 455                         }
 456                         continue;
 457                 } else {
 458                         buf += ret;
 459                         bytes_left -= ret;
 460                 }
 461         } while (!is_done() && bytes_left);
 462
 463         return len - bytes_left;
 464 }
 465
 466 static int read_data(struct thread_information *tip, void *buf,
 467                      unsigned int len)
 468 {
 469         return tip->read_data(tip, buf, len);
 470 }
 471
 472 static inline struct tip_subbuf *
 473 subbuf_fifo_dequeue(struct thread_information *tip)
 474 {
 475         const int head = tip->fifo.head;
 476         const int next = (head + 1) & (FIFO_SIZE - 1);
 477
 478         if (head != tip->fifo.tail) {
 479                 struct tip_subbuf *ts = tip->fifo.q[head];
 480
 481                 store_barrier();
 482                 tip->fifo.head = next;
 483                 return ts;
 484         }
 485
 486         return NULL;
 487 }
 488
 489 static inline int subbuf_fifo_queue(struct thread_information *tip,
 490                                     struct tip_subbuf *ts)
 491 {
 492         const int tail = tip->fifo.tail;
 493         const int next = (tail + 1) & (FIFO_SIZE - 1);
 494
 495         if (next != tip->fifo.head) {
 496                 tip->fifo.q[tail] = ts;
 497                 store_barrier();
 498                 tip->fifo.tail = next;
 499                 return 0;
 500         }
 501
 502         fprintf(stderr, "fifo too small!\n");
 503         return 1;
 504 }
 505
 506 /*
 507  * For file output, truncate and mmap the file appropriately
 508  */
 509 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
 510 {
 511         int ofd = fileno(tip->ofile);
 512         int ret;
 513
 514         /*
 515          * extend file, if we have to. use chunks of 16 subbuffers.
 516          */
 517         if (tip->fs_off + buf_size > tip->fs_buf_len) {
 518                 if (tip->fs_buf) {
 519                         munlock(tip->fs_buf, tip->fs_buf_len);
 520                         munmap(tip->fs_buf, tip->fs_buf_len);
 521                         tip->fs_buf = NULL;
 522                 }
 523
 524                 tip->fs_off = tip->fs_size & (page_size - 1);
 525                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 526                 tip->fs_max_size += tip->fs_buf_len;
 527
 528                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 529                         perror("ftruncate");
 530                         return -1;
 531                 }
 532
 533                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 534                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 535                 if (tip->fs_buf == MAP_FAILED) {
 536                         perror("mmap");
 537                         return -1;
 538                 }
 539                 mlock(tip->fs_buf, tip->fs_buf_len);
 540         }
 541
 542         ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
 543         if (ret >= 0) {
 544                 tip->data_read += ret;
 545                 tip->fs_size += ret;
 546                 tip->fs_off += ret;
 547                 return 0;
 548         }
 549
 550         return -1;
 551 }
 552
 553 /*
 554  * Use the copy approach for pipes and network
 555  */
 556 static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
 557 {
 558         struct tip_subbuf *ts = malloc(sizeof(*ts));
 559         int ret;
 560
 561         ts->buf = malloc(buf_size);
 562         ts->max_len = maxlen;
 563
 564         ret = read_data(tip, ts->buf, ts->max_len);
 565         if (ret > 0) {
 566                 ts->len = ret;
 567                 tip->data_read += ret;
 568                 if (subbuf_fifo_queue(tip, ts))
 569                         return -1;
 570         }
 571
 572         return ret;
 573 }
 574
 575 static int get_subbuf_sendfile(struct thread_information *tip,
 576                                unsigned int maxlen)
 577 {
 578         struct tip_subbuf *ts;
 579         struct stat sb;
 580         unsigned int ready;
 581
 582         wait_for_data(tip);
 583
 584         /*
 585          * hack to get last data out, we can't use sendfile for that
 586          */
 587         if (is_done())
 588                 return get_subbuf(tip, maxlen);
 589
 590         if (tip->sendfile_pending) {
 591                 usleep(100);
 592                 return 0;
 593         }
 594
 595         if (fstat(tip->fd, &sb) < 0) {
 596                 perror("trace stat");
 597                 return -1;
 598         }
 599         ready = sb.st_size - tip->data_queued;
 600         if (!ready)
 601                 return 0;
 602
 603         ts = malloc(sizeof(*ts));
 604         ts->buf = NULL;
 605         ts->max_len = 0;
 606         ts->len = ready;
 607         tip->data_queued += ready;
 608
 609         if (subbuf_fifo_queue(tip, ts))
 610                 return -1;
 611
 612         tip->sendfile_pending++;
 613         return ready;
 614 }
 615
 616 static void close_thread(struct thread_information *tip)
 617 {
 618         if (tip->fd != -1)
 619                 close(tip->fd);
 620         if (tip->ofile)
 621                 fclose(tip->ofile);
 622         if (tip->ofile_buffer)
 623                 free(tip->ofile_buffer);
 624         if (tip->fd_buf)
 625                 free(tip->fd_buf);
 626
 627         tip->fd = -1;
 628         tip->ofile = NULL;
 629         tip->ofile_buffer = NULL;
 630         tip->fd_buf = NULL;
 631 }
 632
 633 static void tip_ftrunc_final(struct thread_information *tip)
 634 {
 635         /*
 636          * truncate to right size and cleanup mmap
 637          */
 638         if (tip->ofile_mmap) {
 639                 int ofd = fileno(tip->ofile);
 640
 641                 if (tip->fs_buf)
 642                         munmap(tip->fs_buf, tip->fs_buf_len);
 643
 644                 ftruncate(ofd, tip->fs_size);
 645         }
 646 }
 647
 648 static void *thread_main(void *arg)
 649 {
 650         struct thread_information *tip = arg;
 651         pid_t pid = getpid();
 652         cpu_set_t cpu_mask;
 653
 654         CPU_ZERO(&cpu_mask);
 655         CPU_SET((tip->cpu), &cpu_mask);
 656
 657         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 658                 perror("sched_setaffinity");
 659                 exit_trace(1);
 660         }
 661
 662         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 663                         relay_path, tip->device->buts_name, tip->cpu);
 664         tip->fd = open(tip->fn, O_RDONLY);
 665         if (tip->fd < 0) {
 666                 perror(tip->fn);
 667                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 668                         tip->fn);
 669                 exit_trace(1);
 670         }
 671
 672         while (!is_done()) {
 673                 if (tip->get_subbuf(tip, buf_size) < 0)
 674                         break;
 675         }
 676
 677         /*
 678          * trace is stopped, pull data until we get a short read
 679          */
 680         while (tip->get_subbuf(tip, buf_size) > 0)
 681                 ;
 682
 683         tip_ftrunc_final(tip);
 684         tip->exited = 1;
 685         return NULL;
 686 }
 687
 688 static int write_data_net(int fd, void *buf, unsigned int buf_len)
 689 {
 690         unsigned int bytes_left = buf_len;
 691         int ret;
 692
 693         while (bytes_left) {
 694                 ret = send(fd, buf, bytes_left, 0);
 695                 if (ret < 0) {
 696                         perror("send");
 697                         return 1;
 698                 }
 699
 700                 buf += ret;
 701                 bytes_left -= ret;
 702         }
 703
 704         return 0;
 705 }
 706
 707 static int net_send_header(struct thread_information *tip, unsigned int len)
 708 {
 709         struct blktrace_net_hdr hdr;
 710
 711         hdr.magic = BLK_IO_TRACE_MAGIC;
 712         strcpy(hdr.buts_name, tip->device->buts_name);
 713         hdr.cpu = tip->cpu;
 714         hdr.max_cpus = ncpus;
 715         hdr.len = len;
 716
 717         return write_data_net(net_out_fd, &hdr, sizeof(hdr));
 718 }
 719
 720 /*
 721  * send header with 0 length to signal end-of-run
 722  */
 723 static void net_client_send_close(void)
 724 {
 725         struct blktrace_net_hdr hdr;
 726
 727         hdr.magic = BLK_IO_TRACE_MAGIC;
 728         hdr.cpu = 0;
 729         hdr.max_cpus = ncpus;
 730         hdr.len = 0;
 731
 732         write_data_net(net_out_fd, &hdr, sizeof(hdr));
 733 }
 734
 735 static int flush_subbuf_net(struct thread_information *tip,
 736                             struct tip_subbuf *ts)
 737 {
 738         if (net_send_header(tip, ts->len))
 739                 return 1;
 740         if (write_data_net(net_out_fd, ts->buf, ts->len))
 741                 return 1;
 742
 743         free(ts->buf);
 744         free(ts);
 745         return 0;
 746 }
 747
 748 static int net_sendfile(struct thread_information *tip, struct tip_subbuf *ts)
 749 {
 750         int ret = sendfile(net_out_fd, tip->fd, NULL, ts->len);
 751
 752         if (ret < 0) {
 753                 perror("sendfile");
 754                 return 1;
 755         } else if (ret < (int) ts->len) {
 756                 fprintf(stderr, "short sendfile send (%d of %d)\n", ret, ts->len);
 757                 return 1;
 758         }
 759
 760         return 0;
 761 }
 762
 763 static int flush_subbuf_sendfile(struct thread_information *tip,
 764                                  struct tip_subbuf *ts)
 765 {
 766         int ret = 1;
 767
 768         /*
 769          * currently we cannot use sendfile() on the last bytes read, as they
 770          * may not be a full subbuffer. get_subbuf_sendfile() falls back to
 771          * the read approach for those, so use send() to ship them out
 772          */
 773         if (ts->buf)
 774                 return flush_subbuf_net(tip, ts);
 775
 776         if (net_send_header(tip, ts->len))
 777                 goto err;
 778         if (net_sendfile(tip, ts))
 779                 goto err;
 780
 781         tip->data_read += ts->len;
 782         tip->ofile_offset += buf_size;
 783         ret = 0;
 784 err:
 785         tip->sendfile_pending--;
 786         free(ts);
 787         return ret;
 788 }
 789
 790 static int write_data(struct thread_information *tip, void *buf,
 791                       unsigned int buf_len)
 792 {
 793         int ret;
 794
 795         if (!buf_len)
 796                 return 0;
 797
 798         while (1) {
 799                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 800                 if (ret == 1)
 801                         break;
 802
 803                 if (ret < 0) {
 804                         perror("write");
 805                         return 1;
 806                 }
 807         }
 808
 809         if (tip->ofile_stdout)
 810                 fflush(tip->ofile);
 811
 812         return 0;
 813 }
 814
 815 static int flush_subbuf_file(struct thread_information *tip,
 816                              struct tip_subbuf *ts)
 817 {
 818         unsigned int offset = 0;
 819         struct blk_io_trace *t;
 820         int pdu_len, events = 0;
 821
 822         /*
 823          * surplus from last run
 824          */
 825         if (tip->leftover_ts) {
 826                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 827
 828                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 829                         prev_ts->max_len += ts->len;
 830                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 831                 }
 832
 833                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 834                 prev_ts->len += ts->len;
 835
 836                 free(ts->buf);
 837                 free(ts);
 838
 839                 ts = prev_ts;
 840                 tip->leftover_ts = NULL;
 841         }
 842
 843         while (offset + sizeof(*t) <= ts->len) {
 844                 t = ts->buf + offset;
 845
 846                 if (verify_trace(t)) {
 847                         write_data(tip, ts->buf, offset);
 848                         return -1;
 849                 }
 850
 851                 pdu_len = t->pdu_len;
 852
 853                 if (offset + sizeof(*t) + pdu_len > ts->len)
 854                         break;
 855
 856                 offset += sizeof(*t) + pdu_len;
 857                 tip->events_processed++;
 858                 tip->data_read += sizeof(*t) + pdu_len;
 859                 events++;
 860         }
 861
 862         if (write_data(tip, ts->buf, offset))
 863                 return -1;
 864
 865         /*
 866          * leftover bytes, save them for next time
 867          */
 868         if (offset != ts->len) {
 869                 tip->leftover_ts = ts;
 870                 ts->len -= offset;
 871                 memmove(ts->buf, ts->buf + offset, ts->len);
 872         } else {
 873                 free(ts->buf);
 874                 free(ts);
 875         }
 876
 877         return events;
 878 }
 879
 880 static int write_tip_events(struct thread_information *tip)
 881 {
 882         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 883
 884         if (ts)
 885                 return tip->flush_subbuf(tip, ts);
 886
 887         return 0;
 888 }
 889
 890 /*
 891  * scans the tips we know and writes out the subbuffers we accumulate
 892  */
 893 static void get_and_write_events(void)
 894 {
 895         struct device_information *dip;
 896         struct thread_information *tip;
 897         int i, j, events, ret, tips_running;
 898
 899         while (!is_done()) {
 900                 events = 0;
 901
 902                 for_each_dip(dip, i) {
 903                         for_each_tip(dip, tip, j) {
 904                                 ret = write_tip_events(tip);
 905                                 if (ret > 0)
 906                                         events += ret;
 907                         }
 908                 }
 909
 910                 if (!events)
 911                         usleep(10);
 912         }
 913
 914         /*
 915          * reap stored events
 916          */
 917         do {
 918                 events = 0;
 919                 tips_running = 0;
 920                 for_each_dip(dip, i) {
 921                         for_each_tip(dip, tip, j) {
 922                                 ret = write_tip_events(tip);
 923                                 if (ret > 0)
 924                                         events += ret;
 925                                 tips_running += !tip->exited;
 926                         }
 927                 }
 928                 usleep(10);
 929         } while (events || tips_running);
 930 }
 931
 932 static void wait_for_threads(void)
 933 {
 934         /*
 935          * for piped or network output, poll and fetch data for writeout.
 936          * for files, we just wait around for trace threads to exit
 937          */
 938         if ((output_name && !strcmp(output_name, "-")) ||
 939             net_mode == Net_client)
 940                 get_and_write_events();
 941         else {
 942                 struct device_information *dip;
 943                 struct thread_information *tip;
 944                 int i, j, tips_running;
 945
 946                 do {
 947                         tips_running = 0;
 948                         usleep(1000);
 949
 950                         for_each_dip(dip, i)
 951                                 for_each_tip(dip, tip, j)
 952                                         tips_running += !tip->exited;
 953                 } while (tips_running);
 954         }
 955
 956         if (net_mode == Net_client)
 957                 net_client_send_close();
 958 }
 959
 960 static int fill_ofname(struct thread_information *tip, char *dst,
 961                        char *buts_name)
 962 {
 963         struct stat sb;
 964         int len = 0;
 965         time_t t;
 966
 967         if (output_dir)
 968                 len = sprintf(dst, "%s/", output_dir);
 969
 970         if (net_mode == Net_server) {
 971                 len += sprintf(dst + len, "%s-", inet_ntoa(tip->cl_in_addr));
 972                 time(&t);
 973                 len += strftime(dst + len, 64, "%F-%T/", gmtime(&t));
 974         }
 975
 976         if (stat(dst, &sb) < 0) {
 977                 if (errno != ENOENT) {
 978                         perror("stat");
 979                         return 1;
 980                 }
 981                 if (mkdir(dst, 0755) < 0) {
 982                         perror(dst);
 983                         fprintf(stderr, "Can't make output dir\n");
 984                         return 1;
 985                 }
 986         }
 987
 988         if (output_name)
 989                 sprintf(dst + len, "%s.blktrace.%d", output_name, tip->cpu);
 990         else
 991                 sprintf(dst + len, "%s.blktrace.%d", buts_name, tip->cpu);
 992
 993         return 0;
 994 }
 995
 996 static void fill_ops(struct thread_information *tip)
 997 {
 998         /*
 999          * setup ops
1000          */
1001         if (net_mode == Net_client) {
1002                 if (net_use_sendfile) {
1003                         tip->get_subbuf = get_subbuf_sendfile;
1004                         tip->flush_subbuf = flush_subbuf_sendfile;
1005                 } else {
1006                         tip->get_subbuf = get_subbuf;
1007                         tip->flush_subbuf = flush_subbuf_net;
1008                 }
1009         } else {
1010                 if (tip->ofile_mmap)
1011                         tip->get_subbuf = mmap_subbuf;
1012                 else
1013                         tip->get_subbuf = get_subbuf;
1014
1015                 tip->flush_subbuf = flush_subbuf_file;
1016         }
1017
1018         if (net_mode == Net_server)
1019                 tip->read_data = read_data_net;
1020         else
1021                 tip->read_data = read_data_file;
1022 }
1023
1024 static int tip_open_output(struct device_information *dip,
1025                            struct thread_information *tip)
1026 {
1027         int pipeline = output_name && !strcmp(output_name, "-");
1028         int mode, vbuf_size;
1029         char op[128];
1030
1031         if (net_mode == Net_client) {
1032                 tip->ofile = NULL;
1033                 tip->ofile_stdout = 0;
1034                 tip->ofile_mmap = 0;
1035                 goto done;
1036         } else if (pipeline) {
1037                 tip->ofile = fdopen(STDOUT_FILENO, "w");
1038                 tip->ofile_stdout = 1;
1039                 tip->ofile_mmap = 0;
1040                 mode = _IOLBF;
1041                 vbuf_size = 512;
1042         } else {
1043                 if (fill_ofname(tip, op, dip->buts_name))
1044                         return 1;
1045                 tip->ofile = fopen(op, "w+");
1046                 tip->ofile_stdout = 0;
1047                 tip->ofile_mmap = 1;
1048                 mode = _IOFBF;
1049                 vbuf_size = OFILE_BUF;
1050         }
1051
1052         if (tip->ofile == NULL) {
1053                 perror(op);
1054                 return 1;
1055         }
1056
1057         tip->ofile_buffer = malloc(vbuf_size);
1058         if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
1059                 perror("setvbuf");
1060                 close_thread(tip);
1061                 return 1;
1062         }
1063
1064 done:
1065         fill_ops(tip);
1066         return 0;
1067 }
1068
1069 static int start_threads(struct device_information *dip)
1070 {
1071         struct thread_information *tip;
1072         int j;
1073
1074         for_each_tip(dip, tip, j) {
1075                 tip->cpu = j;
1076                 tip->device = dip;
1077                 tip->events_processed = 0;
1078                 tip->fd = -1;
1079                 memset(&tip->fifo, 0, sizeof(tip->fifo));
1080                 tip->leftover_ts = NULL;
1081
1082                 if (tip_open_output(dip, tip))
1083                         return 1;
1084
1085                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
1086                         perror("pthread_create");
1087                         close_thread(tip);
1088                         return 1;
1089                 }
1090         }
1091
1092         return 0;
1093 }
1094
1095 static void stop_threads(struct device_information *dip)
1096 {
1097         struct thread_information *tip;
1098         unsigned long ret;
1099         int i;
1100
1101         for_each_tip(dip, tip, i) {
1102                 (void) pthread_join(tip->thread, (void *) &ret);
1103                 close_thread(tip);
1104         }
1105 }
1106
1107 static void stop_all_threads(void)
1108 {
1109         struct device_information *dip;
1110         int i;
1111
1112         for_each_dip(dip, i)
1113                 stop_threads(dip);
1114 }
1115
1116 static void stop_all_tracing(void)
1117 {
1118         struct device_information *dip;
1119         int i;
1120
1121         for_each_dip(dip, i)
1122                 stop_trace(dip);
1123 }
1124
1125 static void exit_trace(int status)
1126 {
1127         if (!is_trace_stopped()) {
1128                 trace_stopped = 1;
1129                 stop_all_threads();
1130                 stop_all_tracing();
1131         }
1132
1133         exit(status);
1134 }
1135
1136 static int resize_devices(char *path)
1137 {
1138         int size = (ndevs + 1) * sizeof(struct device_information);
1139
1140         device_information = realloc(device_information, size);
1141         if (!device_information) {
1142                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
1143                 return 1;
1144         }
1145         device_information[ndevs].path = path;
1146         ndevs++;
1147         return 0;
1148 }
1149
1150 static int open_devices(void)
1151 {
1152         struct device_information *dip;
1153         int i;
1154
1155         for_each_dip(dip, i) {
1156                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
1157                 if (dip->fd < 0) {
1158                         perror(dip->path);
1159                         return 1;
1160                 }
1161         }
1162
1163         return 0;
1164 }
1165
1166 static int start_devices(void)
1167 {
1168         struct device_information *dip;
1169         int i, j, size;
1170
1171         size = ncpus * sizeof(struct thread_information);
1172         thread_information = malloc(size * ndevs);
1173         if (!thread_information) {
1174                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
1175                 return 1;
1176         }
1177
1178         for_each_dip(dip, i) {
1179                 if (start_trace(dip)) {
1180                         close(dip->fd);
1181                         fprintf(stderr, "Failed to start trace on %s\n",
1182                                 dip->path);
1183                         break;
1184                 }
1185         }
1186
1187         if (i != ndevs) {
1188                 __for_each_dip(dip, j, i)
1189                         stop_trace(dip);
1190
1191                 return 1;
1192         }
1193
1194         for_each_dip(dip, i) {
1195                 dip->threads = thread_information + (i * ncpus);
1196                 if (start_threads(dip)) {
1197                         fprintf(stderr, "Failed to start worker threads\n");
1198                         break;
1199                 }
1200         }
1201
1202         if (i != ndevs) {
1203                 __for_each_dip(dip, j, i)
1204                         stop_threads(dip);
1205                 for_each_dip(dip, i)
1206                         stop_trace(dip);
1207
1208                 return 1;
1209         }
1210
1211         return 0;
1212 }
1213
1214 static void show_stats(void)
1215 {
1216         struct device_information *dip;
1217         struct thread_information *tip;
1218         unsigned long long events_processed, data_read;
1219         unsigned long total_drops;
1220         int i, j, no_stdout = 0;
1221
1222         if (is_stat_shown())
1223                 return;
1224
1225         if (output_name && !strcmp(output_name, "-"))
1226                 no_stdout = 1;
1227
1228         stat_shown = 1;
1229
1230         total_drops = 0;
1231         for_each_dip(dip, i) {
1232                 if (!no_stdout)
1233                         printf("Device: %s\n", dip->path);
1234                 events_processed = 0;
1235                 data_read = 0;
1236                 for_each_tip(dip, tip, j) {
1237                         if (!no_stdout)
1238                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
1239                                         tip->cpu, tip->events_processed,
1240                                         (tip->data_read + 1023) >> 10);
1241                         events_processed += tip->events_processed;
1242                         data_read += tip->data_read;
1243                 }
1244                 total_drops += dip->drop_count;
1245                 if (!no_stdout)
1246                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
1247                                         events_processed, dip->drop_count,
1248                                         (data_read + 1023) >> 10);
1249         }
1250
1251         if (total_drops)
1252                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1253 }
1254
1255 static struct device_information *net_get_dip(char *buts_name,
1256                                               struct in_addr *cl_in_addr)
1257 {
1258         struct device_information *dip;
1259         int i;
1260
1261         for (i = 0; i < ndevs; i++) {
1262                 dip = &device_information[i];
1263
1264                 if (!strcmp(dip->buts_name, buts_name))
1265                         return dip;
1266         }
1267
1268         device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
1269         dip = &device_information[ndevs];
1270         memset(dip, 0, sizeof(*dip));
1271         dip->fd = -1;
1272         strcpy(dip->buts_name, buts_name);
1273         dip->path = strdup(buts_name);
1274         ndevs++;
1275         dip->threads = malloc(ncpus * sizeof(struct thread_information));
1276         memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
1277
1278         /*
1279          * open all files
1280          */
1281         for (i = 0; i < ncpus; i++) {
1282                 struct thread_information *tip = &dip->threads[i];
1283
1284                 tip->cpu = i;
1285                 tip->device = dip;
1286                 tip->fd = -1;
1287                 tip->cl_in_addr = *cl_in_addr;
1288
1289                 if (tip_open_output(dip, tip))
1290                         return NULL;
1291         }
1292
1293         return dip;
1294 }
1295
1296 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh,
1297                                               struct in_addr *cl_in_addr)
1298 {
1299         struct device_information *dip;
1300
1301         ncpus = bnh->max_cpus;
1302         dip = net_get_dip(bnh->buts_name, cl_in_addr);
1303         return &dip->threads[bnh->cpu];
1304 }
1305
1306 static int net_get_header(struct blktrace_net_hdr *bnh)
1307 {
1308         int fl = fcntl(net_in_fd, F_GETFL);
1309         int bytes_left, ret;
1310         void *p = bnh;
1311
1312         fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1313         bytes_left = sizeof(*bnh);
1314         while (bytes_left && !is_done()) {
1315                 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1316                 if (ret < 0) {
1317                         if (errno != EAGAIN) {
1318                                 perror("recv header");
1319                                 return 1;
1320                         }
1321                         usleep(100);
1322                         continue;
1323                 } else if (!ret) {
1324                         usleep(100);
1325                         continue;
1326                 } else {
1327                         p += ret;
1328                         bytes_left -= ret;
1329                 }
1330         }
1331         fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1332         return bytes_left;
1333 }
1334
1335 static int net_server_loop(struct in_addr *cl_in_addr)
1336 {
1337         struct thread_information *tip;
1338         struct blktrace_net_hdr bnh;
1339
1340         if (net_get_header(&bnh))
1341                 return 1;
1342
1343         if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1344                 fprintf(stderr, "server: received data is bad\n");
1345                 return 1;
1346         }
1347
1348         if (!data_is_native) {
1349                 bnh.magic = be32_to_cpu(bnh.magic);
1350                 bnh.cpu = be32_to_cpu(bnh.cpu);
1351                 bnh.len = be32_to_cpu(bnh.len);
1352         }
1353
1354         if ((bnh.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1355                 fprintf(stderr, "server: bad data magic\n");
1356                 return 1;
1357         }
1358
1359         /*
1360          * len == 0 means that the other end signalled end-of-run
1361          */
1362         if (!bnh.len) {
1363                 fprintf(stderr, "server: end of run\n");
1364                 return 1;
1365         }
1366
1367         tip = net_get_tip(&bnh, cl_in_addr);
1368         if (!tip)
1369                 return 1;
1370
1371         if (mmap_subbuf(tip, bnh.len))
1372                 return 1;
1373
1374         return 0;
1375 }
1376
1377 static int get_connection(int fd, struct sockaddr_in *addr)
1378 {
1379         struct pollfd pfd = { .fd = fd, .events = POLLIN };
1380         socklen_t socklen;
1381
1382         printf("blktrace: waiting for incoming connection...\n");
1383
1384         if (poll(&pfd, 1, -1) < 0) {
1385                 perror("poll for connection");
1386                 return 1;
1387         }
1388         if ((pfd.revents & POLLIN) == 0)
1389                 return 1;
1390
1391         socklen = sizeof(*addr);
1392         net_in_fd = accept(fd, (struct sockaddr *) addr, &socklen);
1393         if (net_in_fd < 0) {
1394                 perror("accept");
1395                 return 1;
1396         }
1397
1398         printf("blktrace: connection from %s\n", inet_ntoa(addr->sin_addr));
1399         return 0;
1400 }
1401
1402 /*
1403  * Start here when we are in server mode - just fetch data from the network
1404  * and dump to files
1405  */
1406 static int net_server(void)
1407 {
1408         struct device_information *dip;
1409         struct thread_information *tip;
1410         struct sockaddr_in addr;
1411         int fd, opt, i, j;
1412
1413         fd = socket(AF_INET, SOCK_STREAM, 0);
1414         if (fd < 0) {
1415                 perror("server: socket");
1416                 return 1;
1417         }
1418
1419         opt = 1;
1420         if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1421                 perror("setsockopt");
1422                 return 1;
1423         }
1424
1425         memset(&addr, 0, sizeof(addr));
1426         addr.sin_family = AF_INET;
1427         addr.sin_addr.s_addr = htonl(INADDR_ANY);
1428         addr.sin_port = htons(net_port);
1429
1430         if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1431                 perror("bind");
1432                 return 1;
1433         }
1434
1435         if (listen(fd, 1) < 0) {
1436                 perror("listen");
1437                 return 1;
1438         }
1439
1440 repeat:
1441         if (get_connection(fd, &addr))
1442                 return 0;
1443
1444         while (!is_done()) {
1445                 if (net_server_loop(&addr.sin_addr))
1446                         break;
1447         }
1448
1449         for_each_dip(dip, i)
1450                 for_each_tip(dip, tip, j)
1451                         tip_ftrunc_final(tip);
1452
1453         show_stats();
1454
1455         if (is_done())
1456                 return 0;
1457
1458         /*
1459          * cleanup for next run
1460          */
1461         for_each_dip(dip, i) {
1462                 for_each_tip(dip, tip, j)
1463                         fclose(tip->ofile);
1464
1465                 free(dip->threads);
1466                 free(dip->path);
1467         }
1468
1469         free(device_information);
1470         device_information = NULL;
1471         ncpus = ndevs = 0;
1472
1473         close(net_in_fd);
1474         net_in_fd = -1;
1475         stat_shown = 0;
1476         goto repeat;
1477 }
1478
1479 /*
1480  * Setup outgoing network connection where we will transmit data
1481  */
1482 static int net_setup_client(void)
1483 {
1484         struct sockaddr_in addr;
1485         int fd;
1486
1487         fd = socket(AF_INET, SOCK_STREAM, 0);
1488         if (fd < 0) {
1489                 perror("client: socket");
1490                 return 1;
1491         }
1492
1493         memset(&addr, 0, sizeof(addr));
1494         addr.sin_family = AF_INET;
1495         addr.sin_port = htons(net_port);
1496
1497         if (inet_aton(hostname, &addr.sin_addr) != 1) {
1498                 struct hostent *hent = gethostbyname(hostname);
1499                 if (!hent) {
1500                         perror("gethostbyname");
1501                         return 1;
1502                 }
1503
1504                 memcpy(&addr.sin_addr, hent->h_addr, 4);
1505                 strcpy(hostname, hent->h_name);
1506         }
1507
1508         printf("blktrace: connecting to %s\n", hostname);
1509
1510         if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1511                 perror("client: connect");
1512                 return 1;
1513         }
1514
1515         printf("blktrace: connected!\n");
1516         net_out_fd = fd;
1517         return 0;
1518 }
1519
1520 static char usage_str[] = \
1521         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1522         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1523         "\t-d Use specified device. May also be given last after options\n" \
1524         "\t-r Path to mounted relayfs, defaults to /relay\n" \
1525         "\t-o File(s) to send output to\n" \
1526         "\t-D Directory to prepend to output file names\n" \
1527         "\t-k Kill a running trace\n" \
1528         "\t-w Stop after defined time, in seconds\n" \
1529         "\t-a Only trace specified actions. See documentation\n" \
1530         "\t-A Give trace mask as a single value. See documentation\n" \
1531         "\t-b Sub buffer size in KiB\n" \
1532         "\t-n Number of sub buffers\n" \
1533         "\t-l Run in network listen mode (blktrace server)\n" \
1534         "\t-h Run in network client mode, connecting to the given host\n" \
1535         "\t-p Network port to use (default 8462)\n" \
1536         "\t-s Make the network client use sendfile() to transfer data\n" \
1537         "\t-V Print program version info\n\n";
1538
1539 static void show_usage(char *program)
1540 {
1541         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1542 }
1543
1544 int main(int argc, char *argv[])
1545 {
1546         static char default_relay_path[] = "/relay";
1547         struct statfs st;
1548         int i, c;
1549         int stop_watch = 0;
1550         int act_mask_tmp = 0;
1551
1552         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1553                 switch (c) {
1554                 case 'a':
1555                         i = find_mask_map(optarg);
1556                         if (i < 0) {
1557                                 fprintf(stderr,"Invalid action mask %s\n",
1558                                         optarg);
1559                                 return 1;
1560                         }
1561                         act_mask_tmp |= i;
1562                         break;
1563
1564                 case 'A':
1565                         if ((sscanf(optarg, "%x", &i) != 1) ||
1566                                                         !valid_act_opt(i)) {
1567                                 fprintf(stderr,
1568                                         "Invalid set action mask %s/0x%x\n",
1569                                         optarg, i);
1570                                 return 1;
1571                         }
1572                         act_mask_tmp = i;
1573                         break;
1574
1575                 case 'd':
1576                         if (resize_devices(optarg) != 0)
1577                                 return 1;
1578                         break;
1579
1580                 case 'r':
1581                         relay_path = optarg;
1582                         break;
1583
1584                 case 'o':
1585                         output_name = optarg;
1586                         break;
1587                 case 'k':
1588                         kill_running_trace = 1;
1589                         break;
1590                 case 'w':
1591                         stop_watch = atoi(optarg);
1592                         if (stop_watch <= 0) {
1593                                 fprintf(stderr,
1594                                         "Invalid stopwatch value (%d secs)\n",
1595                                         stop_watch);
1596                                 return 1;
1597                         }
1598                         break;
1599                 case 'V':
1600                         printf("%s version %s\n", argv[0], blktrace_version);
1601                         return 0;
1602                 case 'b':
1603                         buf_size = strtoul(optarg, NULL, 10);
1604                         if (buf_size <= 0 || buf_size > 16*1024) {
1605                                 fprintf(stderr,
1606                                         "Invalid buffer size (%lu)\n",buf_size);
1607                                 return 1;
1608                         }
1609                         buf_size <<= 10;
1610                         break;
1611                 case 'n':
1612                         buf_nr = strtoul(optarg, NULL, 10);
1613                         if (buf_nr <= 0) {
1614                                 fprintf(stderr,
1615                                         "Invalid buffer nr (%lu)\n", buf_nr);
1616                                 return 1;
1617                         }
1618                         break;
1619                 case 'D':
1620                         output_dir = optarg;
1621                         break;
1622                 case 'h':
1623                         net_mode = Net_client;
1624                         strcpy(hostname, optarg);
1625                         break;
1626                 case 'l':
1627                         net_mode = Net_server;
1628                         break;
1629                 case 'p':
1630                         net_port = atoi(optarg);
1631                         break;
1632                 case 's':
1633                         net_use_sendfile = 1;
1634                         break;
1635                 default:
1636                         show_usage(argv[0]);
1637                         return 1;
1638                 }
1639         }
1640
1641         setlocale(LC_NUMERIC, "en_US");
1642
1643         page_size = getpagesize();
1644
1645         if (net_mode == Net_server)
1646                 return net_server();
1647
1648         while (optind < argc) {
1649                 if (resize_devices(argv[optind++]) != 0)
1650                         return 1;
1651         }
1652
1653         if (ndevs == 0) {
1654                 show_usage(argv[0]);
1655                 return 1;
1656         }
1657
1658         if (!relay_path)
1659                 relay_path = default_relay_path;
1660
1661         if (act_mask_tmp != 0)
1662                 act_mask = act_mask_tmp;
1663
1664         if (statfs(relay_path, &st) < 0) {
1665                 perror("statfs");
1666                 fprintf(stderr,"%s does not appear to be a valid path\n",
1667                         relay_path);
1668                 return 1;
1669         } else if (st.f_type != (long) RELAYFS_TYPE) {
1670                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1671                         relay_path);
1672                 return 1;
1673         }
1674
1675         if (open_devices() != 0)
1676                 return 1;
1677
1678         if (kill_running_trace) {
1679                 stop_all_traces();
1680                 return 0;
1681         }
1682
1683         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1684         if (ncpus < 0) {
1685                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1686                 return 1;
1687         }
1688
1689         signal(SIGINT, handle_sigint);
1690         signal(SIGHUP, handle_sigint);
1691         signal(SIGTERM, handle_sigint);
1692         signal(SIGALRM, handle_sigint);
1693
1694         if (net_mode == Net_client && net_setup_client())
1695                 return 1;
1696
1697         if (start_devices() != 0)
1698                 return 1;
1699
1700         atexit(stop_all_tracing);
1701
1702         if (stop_watch)
1703                 alarm(stop_watch);
1704
1705         wait_for_threads();
1706
1707         if (!is_trace_stopped()) {
1708                 trace_stopped = 1;
1709                 stop_all_threads();
1710                 stop_all_traces();
1711         }
1712
1713         show_stats();
1714
1715         return 0;
1716 }
1717