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[] = "0.99.1";
  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         FILE *ofile;
 183         char *ofile_buffer;
 184         off_t ofile_offset;
 185         int ofile_stdout;
 186         int ofile_mmap;
 187
 188         int (*get_subbuf)(struct thread_information *, unsigned int);
 189         int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *);
 190         int (*read_data)(struct thread_information *, void *, unsigned int);
 191
 192         unsigned long events_processed;
 193         unsigned long long data_read;
 194         unsigned long long data_queued;
 195         struct device_information *device;
 196
 197         int exited;
 198
 199         /*
 200          * piped fifo buffers
 201          */
 202         struct tip_subbuf_fifo fifo;
 203         struct tip_subbuf *leftover_ts;
 204
 205         /*
 206          * mmap controlled output files
 207          */
 208         unsigned long long fs_size;
 209         unsigned long long fs_max_size;
 210         unsigned long fs_off;
 211         void *fs_buf;
 212         unsigned long fs_buf_len;
 213 };
 214
 215 struct device_information {
 216         int fd;
 217         char *path;
 218         char buts_name[32];
 219         volatile int trace_started;
 220         unsigned long drop_count;
 221         struct thread_information *threads;
 222         struct net_connection *nc;
 223 };
 224
 225 static int ncpus;
 226 static struct thread_information *thread_information;
 227 static int ndevs;
 228 static struct device_information *device_information;
 229
 230 /* command line option globals */
 231 static char *relay_path;
 232 static char *output_name;
 233 static char *output_dir;
 234 static int act_mask = ~0U;
 235 static int kill_running_trace;
 236 static unsigned long buf_size = BUF_SIZE;
 237 static unsigned long buf_nr = BUF_NR;
 238 static unsigned int page_size;
 239
 240 #define is_done()       (*(volatile int *)(&done))
 241 static volatile int done;
 242
 243 #define is_trace_stopped()      (*(volatile int *)(&trace_stopped))
 244 static volatile int trace_stopped;
 245
 246 #define is_stat_shown() (*(volatile int *)(&stat_shown))
 247 static volatile int stat_shown;
 248
 249 int data_is_native = -1;
 250
 251 static void exit_trace(int status);
 252
 253 #define dip_tracing(dip)        (*(volatile int *)(&(dip)->trace_started))
 254 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
 255
 256 #define __for_each_dip(__d, __di, __e, __i)     \
 257         for (__i = 0, __d = __di; __i < __e; __i++, __d++)
 258
 259 #define for_each_dip(__d, __i)          \
 260         __for_each_dip(__d, device_information, ndevs, __i)
 261 #define for_each_nc_dip(__nc, __d, __i)         \
 262         __for_each_dip(__d, (__nc)->device_information, (__nc)->ndevs, __i)
 263
 264 #define __for_each_tip(__d, __t, __ncpus, __j)  \
 265         for (__j = 0, __t = (__d)->threads; __j < __ncpus; __j++, __t++)
 266 #define for_each_tip(__d, __t, __j)     \
 267         __for_each_tip(__d, __t, ncpus, __j)
 268
 269 /*
 270  * networking stuff follows. we include a magic number so we know whether
 271  * to endianness convert or not
 272  */
 273 struct blktrace_net_hdr {
 274         u32 magic;              /* same as trace magic */
 275         char buts_name[32];     /* trace name */
 276         u32 cpu;                /* for which cpu */
 277         u32 max_cpus;
 278         u32 len;                /* length of following trace data */
 279 };
 280
 281 #define TRACE_NET_PORT          (8462)
 282
 283 enum {
 284         Net_none = 0,
 285         Net_server,
 286         Net_client,
 287 };
 288
 289 /*
 290  * network cmd line params
 291  */
 292 static char hostname[MAXHOSTNAMELEN];
 293 static int net_port = TRACE_NET_PORT;
 294 static int net_mode = 0;
 295 static int net_use_sendfile;
 296
 297 struct net_connection {
 298         int in_fd;
 299         time_t connect_time;
 300         struct in_addr cl_in_addr;
 301         struct device_information *device_information;
 302         int ndevs;
 303         int ncpus;
 304         int connection_index;
 305 };
 306
 307 #define NET_MAX_CONNECTIONS     (1024)
 308 static struct net_connection net_connections[NET_MAX_CONNECTIONS];
 309 static int net_connects;
 310 static int net_out_fd = -1;
 311
 312 static void handle_sigint(__attribute__((__unused__)) int sig)
 313 {
 314         struct device_information *dip;
 315         int i;
 316
 317         /*
 318          * stop trace so we can reap currently produced data
 319          */
 320         for_each_dip(dip, i) {
 321                 if (dip->fd == -1)
 322                         continue;
 323                 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
 324                         perror("BLKTRACESTOP");
 325         }
 326
 327         done = 1;
 328 }
 329
 330 static int get_dropped_count(const char *buts_name)
 331 {
 332         int fd;
 333         char tmp[MAXPATHLEN + 64];
 334
 335         snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
 336                  relay_path, buts_name);
 337
 338         fd = open(tmp, O_RDONLY);
 339         if (fd < 0) {
 340                 /*
 341                  * this may be ok, if the kernel doesn't support dropped counts
 342                  */
 343                 if (errno == ENOENT)
 344                         return 0;
 345
 346                 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
 347                 return -1;
 348         }
 349
 350         if (read(fd, tmp, sizeof(tmp)) < 0) {
 351                 perror(tmp);
 352                 close(fd);
 353                 return -1;
 354         }
 355
 356         close(fd);
 357
 358         return atoi(tmp);
 359 }
 360
 361 static int start_trace(struct device_information *dip)
 362 {
 363         struct blk_user_trace_setup buts;
 364
 365         memset(&buts, 0, sizeof(buts));
 366         buts.buf_size = buf_size;
 367         buts.buf_nr = buf_nr;
 368         buts.act_mask = act_mask;
 369
 370         if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
 371                 perror("BLKTRACESETUP");
 372                 return 1;
 373         }
 374
 375         if (ioctl(dip->fd, BLKTRACESTART) < 0) {
 376                 perror("BLKTRACESTART");
 377                 return 1;
 378         }
 379
 380         memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
 381         dip_set_tracing(dip, 1);
 382         return 0;
 383 }
 384
 385 static void stop_trace(struct device_information *dip)
 386 {
 387         if (dip_tracing(dip) || kill_running_trace) {
 388                 dip_set_tracing(dip, 0);
 389
 390                 /*
 391                  * should be stopped, just don't complain if it isn't
 392                  */
 393                 ioctl(dip->fd, BLKTRACESTOP);
 394
 395                 if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
 396                         perror("BLKTRACETEARDOWN");
 397
 398                 close(dip->fd);
 399                 dip->fd = -1;
 400         }
 401 }
 402
 403 static void stop_all_traces(void)
 404 {
 405         struct device_information *dip;
 406         int i;
 407
 408         for_each_dip(dip, i) {
 409                 dip->drop_count = get_dropped_count(dip->buts_name);
 410                 stop_trace(dip);
 411         }
 412 }
 413
 414 static void wait_for_data(struct thread_information *tip, int events)
 415 {
 416         struct pollfd pfd = { .fd = tip->fd, .events = events };
 417
 418         do {
 419                 if (poll(&pfd, 1, 100) < 0) {
 420                         perror("poll");
 421                         break;
 422                 }
 423                 if (pfd.revents & events)
 424                         break;
 425                 if (tip->ofile_stdout)
 426                         break;
 427         } while (!is_done());
 428 }
 429
 430 static int read_data_file(struct thread_information *tip, void *buf,
 431                           unsigned int len)
 432 {
 433         int ret = 0;
 434
 435         do {
 436                 wait_for_data(tip, POLLIN);
 437
 438                 ret = read(tip->fd, buf, len);
 439                 if (!ret)
 440                         continue;
 441                 else if (ret > 0)
 442                         return ret;
 443                 else {
 444                         if (errno != EAGAIN) {
 445                                 perror(tip->fn);
 446                                 fprintf(stderr,"Thread %d failed read of %s\n",
 447                                         tip->cpu, tip->fn);
 448                                 break;
 449                         }
 450                         continue;
 451                 }
 452         } while (!is_done());
 453
 454         return ret;
 455
 456 }
 457
 458 static int read_data_net(struct thread_information *tip, void *buf,
 459                          unsigned int len)
 460 {
 461         struct net_connection *nc = tip->device->nc;
 462         unsigned int bytes_left = len;
 463         int ret = 0;
 464
 465         do {
 466                 ret = recv(nc->in_fd, buf, bytes_left, MSG_WAITALL);
 467
 468                 if (!ret)
 469                         continue;
 470                 else if (ret < 0) {
 471                         if (errno != EAGAIN) {
 472                                 perror(tip->fn);
 473                                 fprintf(stderr, "server: failed read\n");
 474                                 return 0;
 475                         }
 476                         continue;
 477                 } else {
 478                         buf += ret;
 479                         bytes_left -= ret;
 480                 }
 481         } while (!is_done() && bytes_left);
 482
 483         return len - bytes_left;
 484 }
 485
 486 static inline struct tip_subbuf *
 487 subbuf_fifo_dequeue(struct thread_information *tip)
 488 {
 489         const int head = tip->fifo.head;
 490         const int next = (head + 1) & (FIFO_SIZE - 1);
 491
 492         if (head != tip->fifo.tail) {
 493                 struct tip_subbuf *ts = tip->fifo.q[head];
 494
 495                 store_barrier();
 496                 tip->fifo.head = next;
 497                 return ts;
 498         }
 499
 500         return NULL;
 501 }
 502
 503 static inline int subbuf_fifo_queue(struct thread_information *tip,
 504                                     struct tip_subbuf *ts)
 505 {
 506         const int tail = tip->fifo.tail;
 507         const int next = (tail + 1) & (FIFO_SIZE - 1);
 508
 509         if (next != tip->fifo.head) {
 510                 tip->fifo.q[tail] = ts;
 511                 store_barrier();
 512                 tip->fifo.tail = next;
 513                 return 0;
 514         }
 515
 516         fprintf(stderr, "fifo too small!\n");
 517         return 1;
 518 }
 519
 520 /*
 521  * For file output, truncate and mmap the file appropriately
 522  */
 523 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
 524 {
 525         int ofd = fileno(tip->ofile);
 526         int ret;
 527
 528         /*
 529          * extend file, if we have to. use chunks of 16 subbuffers.
 530          */
 531         if (tip->fs_off + buf_size > tip->fs_buf_len) {
 532                 if (tip->fs_buf) {
 533                         munlock(tip->fs_buf, tip->fs_buf_len);
 534                         munmap(tip->fs_buf, tip->fs_buf_len);
 535                         tip->fs_buf = NULL;
 536                 }
 537
 538                 tip->fs_off = tip->fs_size & (page_size - 1);
 539                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 540                 tip->fs_max_size += tip->fs_buf_len;
 541
 542                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 543                         perror("ftruncate");
 544                         return -1;
 545                 }
 546
 547                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 548                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 549                 if (tip->fs_buf == MAP_FAILED) {
 550                         perror("mmap");
 551                         return -1;
 552                 }
 553                 mlock(tip->fs_buf, tip->fs_buf_len);
 554         }
 555
 556         ret = tip->read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
 557         if (ret >= 0) {
 558                 tip->data_read += ret;
 559                 tip->fs_size += ret;
 560                 tip->fs_off += ret;
 561                 return 0;
 562         }
 563
 564         return -1;
 565 }
 566
 567 /*
 568  * Use the copy approach for pipes and network
 569  */
 570 static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
 571 {
 572         struct tip_subbuf *ts = malloc(sizeof(*ts));
 573         int ret;
 574
 575         ts->buf = malloc(buf_size);
 576         ts->max_len = maxlen;
 577
 578         ret = tip->read_data(tip, ts->buf, ts->max_len);
 579         if (ret > 0) {
 580                 ts->len = ret;
 581                 tip->data_read += ret;
 582                 if (subbuf_fifo_queue(tip, ts))
 583                         return -1;
 584         }
 585
 586         return ret;
 587 }
 588
 589 static int get_subbuf_sendfile(struct thread_information *tip,
 590                                unsigned int maxlen)
 591 {
 592         struct tip_subbuf *ts;
 593         struct stat sb;
 594         unsigned int ready;
 595
 596         wait_for_data(tip, POLLMSG);
 597
 598         /*
 599          * hack to get last data out, we can't use sendfile for that
 600          */
 601         if (is_done())
 602                 return get_subbuf(tip, maxlen);
 603
 604         if (fstat(tip->fd, &sb) < 0) {
 605                 perror("trace stat");
 606                 return -1;
 607         }
 608         ready = sb.st_size - tip->data_queued;
 609         if (!ready) {
 610                 usleep(1000);
 611                 return 0;
 612         }
 613
 614         ts = malloc(sizeof(*ts));
 615         ts->buf = NULL;
 616         ts->max_len = 0;
 617         ts->len = ready;
 618         tip->data_queued += ready;
 619
 620         if (subbuf_fifo_queue(tip, ts))
 621                 return -1;
 622
 623         return ready;
 624 }
 625
 626 static void close_thread(struct thread_information *tip)
 627 {
 628         if (tip->fd != -1)
 629                 close(tip->fd);
 630         if (tip->ofile)
 631                 fclose(tip->ofile);
 632         if (tip->ofile_buffer)
 633                 free(tip->ofile_buffer);
 634         if (tip->fd_buf)
 635                 free(tip->fd_buf);
 636
 637         tip->fd = -1;
 638         tip->ofile = NULL;
 639         tip->ofile_buffer = NULL;
 640         tip->fd_buf = NULL;
 641 }
 642
 643 static void tip_ftrunc_final(struct thread_information *tip)
 644 {
 645         /*
 646          * truncate to right size and cleanup mmap
 647          */
 648         if (tip->ofile_mmap && tip->ofile) {
 649                 int ofd = fileno(tip->ofile);
 650
 651                 if (tip->fs_buf)
 652                         munmap(tip->fs_buf, tip->fs_buf_len);
 653
 654                 ftruncate(ofd, tip->fs_size);
 655         }
 656 }
 657
 658 static void *thread_main(void *arg)
 659 {
 660         struct thread_information *tip = arg;
 661         pid_t pid = getpid();
 662         cpu_set_t cpu_mask;
 663
 664         CPU_ZERO(&cpu_mask);
 665         CPU_SET((tip->cpu), &cpu_mask);
 666
 667         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 668                 perror("sched_setaffinity");
 669                 exit_trace(1);
 670         }
 671
 672         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 673                         relay_path, tip->device->buts_name, tip->cpu);
 674         tip->fd = open(tip->fn, O_RDONLY);
 675         if (tip->fd < 0) {
 676                 perror(tip->fn);
 677                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 678                         tip->fn);
 679                 exit_trace(1);
 680         }
 681
 682         while (!is_done()) {
 683                 if (tip->get_subbuf(tip, buf_size) < 0)
 684                         break;
 685         }
 686
 687         /*
 688          * trace is stopped, pull data until we get a short read
 689          */
 690         while (tip->get_subbuf(tip, buf_size) > 0)
 691                 ;
 692
 693         tip_ftrunc_final(tip);
 694         tip->exited = 1;
 695         return NULL;
 696 }
 697
 698 static int write_data_net(int fd, void *buf, unsigned int buf_len)
 699 {
 700         unsigned int bytes_left = buf_len;
 701         int ret;
 702
 703         while (bytes_left) {
 704                 ret = send(fd, buf, bytes_left, 0);
 705                 if (ret < 0) {
 706                         perror("send");
 707                         return 1;
 708                 }
 709
 710                 buf += ret;
 711                 bytes_left -= ret;
 712         }
 713
 714         return 0;
 715 }
 716
 717 static int net_send_header(struct thread_information *tip, unsigned int len)
 718 {
 719         struct blktrace_net_hdr hdr;
 720
 721         hdr.magic = BLK_IO_TRACE_MAGIC;
 722         strcpy(hdr.buts_name, tip->device->buts_name);
 723         hdr.cpu = tip->cpu;
 724         hdr.max_cpus = ncpus;
 725         hdr.len = len;
 726
 727         return write_data_net(net_out_fd, &hdr, sizeof(hdr));
 728 }
 729
 730 /*
 731  * send header with 0 length to signal end-of-run
 732  */
 733 static void net_client_send_close(void)
 734 {
 735         struct device_information *dip;
 736         struct blktrace_net_hdr hdr;
 737         int i;
 738
 739         for_each_dip(dip, i) {
 740                 hdr.magic = BLK_IO_TRACE_MAGIC;
 741                 hdr.max_cpus = ncpus;
 742                 hdr.len = 0;
 743                 strcpy(hdr.buts_name, dip->buts_name);
 744                 hdr.cpu = get_dropped_count(dip->buts_name);
 745
 746                 write_data_net(net_out_fd, &hdr, sizeof(hdr));
 747         }
 748
 749 }
 750
 751 static int flush_subbuf_net(struct thread_information *tip,
 752                             struct tip_subbuf *ts)
 753 {
 754         if (net_send_header(tip, ts->len))
 755                 return -1;
 756         if (write_data_net(net_out_fd, ts->buf, ts->len))
 757                 return -1;
 758
 759         free(ts->buf);
 760         free(ts);
 761         return 1;
 762 }
 763
 764 static int net_sendfile(struct thread_information *tip, struct tip_subbuf *ts)
 765 {
 766         int ret = sendfile(net_out_fd, tip->fd, NULL, ts->len);
 767
 768         if (ret < 0) {
 769                 perror("sendfile");
 770                 return 1;
 771         } else if (ret < (int) ts->len) {
 772                 fprintf(stderr, "short sendfile send (%d of %d)\n", ret, ts->len);
 773                 return 1;
 774         }
 775
 776         return 0;
 777 }
 778
 779 static int flush_subbuf_sendfile(struct thread_information *tip,
 780                                  struct tip_subbuf *ts)
 781 {
 782         int ret = -1;
 783
 784         /*
 785          * currently we cannot use sendfile() on the last bytes read, as they
 786          * may not be a full subbuffer. get_subbuf_sendfile() falls back to
 787          * the read approach for those, so use send() to ship them out
 788          */
 789         if (ts->buf)
 790                 return flush_subbuf_net(tip, ts);
 791
 792         if (net_send_header(tip, ts->len))
 793                 goto err;
 794         if (net_sendfile(tip, ts))
 795                 goto err;
 796
 797         tip->data_read += ts->len;
 798         tip->ofile_offset += buf_size;
 799         ret = 1;
 800 err:
 801         free(ts);
 802         return ret;
 803 }
 804
 805 static int write_data(struct thread_information *tip, void *buf,
 806                       unsigned int buf_len)
 807 {
 808         int ret;
 809
 810         if (!buf_len)
 811                 return 0;
 812
 813         while (1) {
 814                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 815                 if (ret == 1)
 816                         break;
 817
 818                 if (ret < 0) {
 819                         perror("write");
 820                         return 1;
 821                 }
 822         }
 823
 824         if (tip->ofile_stdout)
 825                 fflush(tip->ofile);
 826
 827         return 0;
 828 }
 829
 830 static int flush_subbuf_file(struct thread_information *tip,
 831                              struct tip_subbuf *ts)
 832 {
 833         unsigned int offset = 0;
 834         struct blk_io_trace *t;
 835         int pdu_len, events = 0;
 836
 837         /*
 838          * surplus from last run
 839          */
 840         if (tip->leftover_ts) {
 841                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 842
 843                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 844                         prev_ts->max_len += ts->len;
 845                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 846                 }
 847
 848                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 849                 prev_ts->len += ts->len;
 850
 851                 free(ts->buf);
 852                 free(ts);
 853
 854                 ts = prev_ts;
 855                 tip->leftover_ts = NULL;
 856         }
 857
 858         while (offset + sizeof(*t) <= ts->len) {
 859                 t = ts->buf + offset;
 860
 861                 if (verify_trace(t)) {
 862                         write_data(tip, ts->buf, offset);
 863                         return -1;
 864                 }
 865
 866                 pdu_len = t->pdu_len;
 867
 868                 if (offset + sizeof(*t) + pdu_len > ts->len)
 869                         break;
 870
 871                 offset += sizeof(*t) + pdu_len;
 872                 tip->events_processed++;
 873                 tip->data_read += sizeof(*t) + pdu_len;
 874                 events++;
 875         }
 876
 877         if (write_data(tip, ts->buf, offset))
 878                 return -1;
 879
 880         /*
 881          * leftover bytes, save them for next time
 882          */
 883         if (offset != ts->len) {
 884                 tip->leftover_ts = ts;
 885                 ts->len -= offset;
 886                 memmove(ts->buf, ts->buf + offset, ts->len);
 887         } else {
 888                 free(ts->buf);
 889                 free(ts);
 890         }
 891
 892         return events;
 893 }
 894
 895 static int write_tip_events(struct thread_information *tip)
 896 {
 897         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 898
 899         if (ts)
 900                 return tip->flush_subbuf(tip, ts);
 901
 902         return 0;
 903 }
 904
 905 /*
 906  * scans the tips we know and writes out the subbuffers we accumulate
 907  */
 908 static void get_and_write_events(void)
 909 {
 910         struct device_information *dip;
 911         struct thread_information *tip;
 912         int i, j, events, ret, tips_running;
 913
 914         while (!is_done()) {
 915                 events = 0;
 916
 917                 for_each_dip(dip, i) {
 918                         for_each_tip(dip, tip, j) {
 919                                 ret = write_tip_events(tip);
 920                                 if (ret > 0)
 921                                         events += ret;
 922                         }
 923                 }
 924
 925                 if (!events)
 926                         usleep(100000);
 927         }
 928
 929         /*
 930          * reap stored events
 931          */
 932         do {
 933                 events = 0;
 934                 tips_running = 0;
 935                 for_each_dip(dip, i) {
 936                         for_each_tip(dip, tip, j) {
 937                                 ret = write_tip_events(tip);
 938                                 if (ret > 0)
 939                                         events += ret;
 940                                 tips_running += !tip->exited;
 941                         }
 942                 }
 943                 usleep(10);
 944         } while (events || tips_running);
 945 }
 946
 947 static void wait_for_threads(void)
 948 {
 949         /*
 950          * for piped or network output, poll and fetch data for writeout.
 951          * for files, we just wait around for trace threads to exit
 952          */
 953         if ((output_name && !strcmp(output_name, "-")) ||
 954             net_mode == Net_client)
 955                 get_and_write_events();
 956         else {
 957                 struct device_information *dip;
 958                 struct thread_information *tip;
 959                 int i, j, tips_running;
 960
 961                 do {
 962                         tips_running = 0;
 963                         usleep(100000);
 964
 965                         for_each_dip(dip, i)
 966                                 for_each_tip(dip, tip, j)
 967                                         tips_running += !tip->exited;
 968                 } while (tips_running);
 969         }
 970
 971         if (net_mode == Net_client)
 972                 net_client_send_close();
 973 }
 974
 975 static int fill_ofname(struct device_information *dip,
 976                        struct thread_information *tip, char *dst,
 977                        char *buts_name)
 978 {
 979         struct stat sb;
 980         int len = 0;
 981
 982         if (output_dir)
 983                 len = sprintf(dst, "%s/", output_dir);
 984
 985         if (net_mode == Net_server) {
 986                 struct net_connection *nc = dip->nc;
 987
 988                 len += sprintf(dst + len, "%s-", inet_ntoa(nc->cl_in_addr));
 989                 len += strftime(dst + len, 64, "%F-%T/", gmtime(&nc->connect_time));
 990         }
 991
 992         if (stat(dst, &sb) < 0) {
 993                 if (errno != ENOENT) {
 994                         perror("stat");
 995                         return 1;
 996                 }
 997                 if (mkdir(dst, 0755) < 0) {
 998                         perror(dst);
 999                         fprintf(stderr, "Can't make output dir\n");
1000                         return 1;
1001                 }
1002         }
1003
1004         if (output_name)
1005                 sprintf(dst + len, "%s.blktrace.%d", output_name, tip->cpu);
1006         else
1007                 sprintf(dst + len, "%s.blktrace.%d", buts_name, tip->cpu);
1008
1009         return 0;
1010 }
1011
1012 static void fill_ops(struct thread_information *tip)
1013 {
1014         /*
1015          * setup ops
1016          */
1017         if (net_mode == Net_client) {
1018                 if (net_use_sendfile) {
1019                         tip->get_subbuf = get_subbuf_sendfile;
1020                         tip->flush_subbuf = flush_subbuf_sendfile;
1021                 } else {
1022                         tip->get_subbuf = get_subbuf;
1023                         tip->flush_subbuf = flush_subbuf_net;
1024                 }
1025         } else {
1026                 if (tip->ofile_mmap)
1027                         tip->get_subbuf = mmap_subbuf;
1028                 else
1029                         tip->get_subbuf = get_subbuf;
1030
1031                 tip->flush_subbuf = flush_subbuf_file;
1032         }
1033
1034         if (net_mode == Net_server)
1035                 tip->read_data = read_data_net;
1036         else
1037                 tip->read_data = read_data_file;
1038 }
1039
1040 static int tip_open_output(struct device_information *dip,
1041                            struct thread_information *tip)
1042 {
1043         int pipeline = output_name && !strcmp(output_name, "-");
1044         int mode, vbuf_size;
1045         char op[128];
1046
1047         if (net_mode == Net_client) {
1048                 tip->ofile = NULL;
1049                 tip->ofile_stdout = 0;
1050                 tip->ofile_mmap = 0;
1051                 goto done;
1052         } else if (pipeline) {
1053                 tip->ofile = fdopen(STDOUT_FILENO, "w");
1054                 tip->ofile_stdout = 1;
1055                 tip->ofile_mmap = 0;
1056                 mode = _IOLBF;
1057                 vbuf_size = 512;
1058         } else {
1059                 if (fill_ofname(dip, tip, op, dip->buts_name))
1060                         return 1;
1061                 tip->ofile = fopen(op, "w+");
1062                 tip->ofile_stdout = 0;
1063                 tip->ofile_mmap = 1;
1064                 mode = _IOFBF;
1065                 vbuf_size = OFILE_BUF;
1066         }
1067
1068         if (tip->ofile == NULL) {
1069                 perror(op);
1070                 return 1;
1071         }
1072
1073         tip->ofile_buffer = malloc(vbuf_size);
1074         if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
1075                 perror("setvbuf");
1076                 close_thread(tip);
1077                 return 1;
1078         }
1079
1080 done:
1081         fill_ops(tip);
1082         return 0;
1083 }
1084
1085 static int start_threads(struct device_information *dip)
1086 {
1087         struct thread_information *tip;
1088         int j;
1089
1090         for_each_tip(dip, tip, j) {
1091                 tip->cpu = j;
1092                 tip->device = dip;
1093                 tip->events_processed = 0;
1094                 tip->fd = -1;
1095                 memset(&tip->fifo, 0, sizeof(tip->fifo));
1096                 tip->leftover_ts = NULL;
1097
1098                 if (tip_open_output(dip, tip))
1099                         return 1;
1100
1101                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
1102                         perror("pthread_create");
1103                         close_thread(tip);
1104                         return 1;
1105                 }
1106         }
1107
1108         return 0;
1109 }
1110
1111 static void stop_threads(struct device_information *dip)
1112 {
1113         struct thread_information *tip;
1114         unsigned long ret;
1115         int i;
1116
1117         for_each_tip(dip, tip, i) {
1118                 (void) pthread_join(tip->thread, (void *) &ret);
1119                 close_thread(tip);
1120         }
1121 }
1122
1123 static void stop_all_threads(void)
1124 {
1125         struct device_information *dip;
1126         int i;
1127
1128         for_each_dip(dip, i)
1129                 stop_threads(dip);
1130 }
1131
1132 static void stop_all_tracing(void)
1133 {
1134         struct device_information *dip;
1135         int i;
1136
1137         for_each_dip(dip, i)
1138                 stop_trace(dip);
1139 }
1140
1141 static void exit_trace(int status)
1142 {
1143         if (!is_trace_stopped()) {
1144                 trace_stopped = 1;
1145                 stop_all_threads();
1146                 stop_all_tracing();
1147         }
1148
1149         exit(status);
1150 }
1151
1152 static int resize_devices(char *path)
1153 {
1154         int size = (ndevs + 1) * sizeof(struct device_information);
1155
1156         device_information = realloc(device_information, size);
1157         if (!device_information) {
1158                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
1159                 return 1;
1160         }
1161         device_information[ndevs].path = path;
1162         ndevs++;
1163         return 0;
1164 }
1165
1166 static int open_devices(void)
1167 {
1168         struct device_information *dip;
1169         int i;
1170
1171         for_each_dip(dip, i) {
1172                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
1173                 if (dip->fd < 0) {
1174                         perror(dip->path);
1175                         return 1;
1176                 }
1177         }
1178
1179         return 0;
1180 }
1181
1182 static int start_devices(void)
1183 {
1184         struct device_information *dip;
1185         int i, j, size;
1186
1187         size = ncpus * sizeof(struct thread_information);
1188         thread_information = malloc(size * ndevs);
1189         if (!thread_information) {
1190                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
1191                 return 1;
1192         }
1193
1194         for_each_dip(dip, i) {
1195                 if (start_trace(dip)) {
1196                         close(dip->fd);
1197                         fprintf(stderr, "Failed to start trace on %s\n",
1198                                 dip->path);
1199                         break;
1200                 }
1201         }
1202
1203         if (i != ndevs) {
1204                 __for_each_dip(dip, device_information, i, j)
1205                         stop_trace(dip);
1206
1207                 return 1;
1208         }
1209
1210         for_each_dip(dip, i) {
1211                 dip->threads = thread_information + (i * ncpus);
1212                 if (start_threads(dip)) {
1213                         fprintf(stderr, "Failed to start worker threads\n");
1214                         break;
1215                 }
1216         }
1217
1218         if (i != ndevs) {
1219                 __for_each_dip(dip, device_information, i, j)
1220                         stop_threads(dip);
1221                 for_each_dip(dip, i)
1222                         stop_trace(dip);
1223
1224                 return 1;
1225         }
1226
1227         return 0;
1228 }
1229
1230 static void show_stats(struct device_information *dips, int ndips, int cpus)
1231 {
1232         struct device_information *dip;
1233         struct thread_information *tip;
1234         unsigned long long events_processed, data_read;
1235         unsigned long total_drops;
1236         int i, j, no_stdout = 0;
1237
1238         if (is_stat_shown())
1239                 return;
1240
1241         if (output_name && !strcmp(output_name, "-"))
1242                 no_stdout = 1;
1243
1244         stat_shown = 1;
1245
1246         total_drops = 0;
1247         __for_each_dip(dip, dips, ndips, i) {
1248                 if (!no_stdout)
1249                         printf("Device: %s\n", dip->path);
1250                 events_processed = 0;
1251                 data_read = 0;
1252                 __for_each_tip(dip, tip, cpus, j) {
1253                         if (!no_stdout)
1254                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
1255                                         tip->cpu, tip->events_processed,
1256                                         (tip->data_read + 1023) >> 10);
1257                         events_processed += tip->events_processed;
1258                         data_read += tip->data_read;
1259                 }
1260                 total_drops += dip->drop_count;
1261                 if (!no_stdout)
1262                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
1263                                         events_processed, dip->drop_count,
1264                                         (data_read + 1023) >> 10);
1265         }
1266
1267         if (total_drops)
1268                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1269 }
1270
1271 static struct device_information *net_get_dip(struct net_connection *nc,
1272                                               char *buts_name)
1273 {
1274         struct device_information *dip;
1275         int i;
1276
1277         for (i = 0; i < nc->ndevs; i++) {
1278                 dip = &nc->device_information[i];
1279
1280                 if (!strcmp(dip->buts_name, buts_name))
1281                         return dip;
1282         }
1283
1284         nc->device_information = realloc(nc->device_information, (nc->ndevs + 1) * sizeof(*dip));
1285         dip = &nc->device_information[nc->ndevs];
1286         memset(dip, 0, sizeof(*dip));
1287         dip->fd = -1;
1288         dip->nc = nc;
1289         strcpy(dip->buts_name, buts_name);
1290         dip->path = strdup(buts_name);
1291         dip->trace_started = 1;
1292         nc->ndevs++;
1293         dip->threads = malloc(nc->ncpus * sizeof(struct thread_information));
1294         memset(dip->threads, 0, nc->ncpus * sizeof(struct thread_information));
1295
1296         /*
1297          * open all files
1298          */
1299         for (i = 0; i < nc->ncpus; i++) {
1300                 struct thread_information *tip = &dip->threads[i];
1301
1302                 tip->cpu = i;
1303                 tip->device = dip;
1304                 tip->fd = -1;
1305
1306                 if (tip_open_output(dip, tip))
1307                         return NULL;
1308         }
1309
1310         return dip;
1311 }
1312
1313 static struct thread_information *net_get_tip(struct net_connection *nc,
1314                                               struct blktrace_net_hdr *bnh)
1315 {
1316         struct device_information *dip;
1317
1318         nc->ncpus = bnh->max_cpus;
1319         dip = net_get_dip(nc, bnh->buts_name);
1320         if (!dip->trace_started) {
1321                 fprintf(stderr, "Events for closed devices %s\n", dip->buts_name);
1322                 return NULL;
1323         }
1324
1325         return &dip->threads[bnh->cpu];
1326 }
1327
1328 static int net_get_header(struct net_connection *nc,
1329                           struct blktrace_net_hdr *bnh)
1330 {
1331         int fl = fcntl(nc->in_fd, F_GETFL);
1332         int bytes_left, ret;
1333         void *p = bnh;
1334
1335         fcntl(nc->in_fd, F_SETFL, fl | O_NONBLOCK);
1336         bytes_left = sizeof(*bnh);
1337         while (bytes_left && !is_done()) {
1338                 ret = recv(nc->in_fd, p, bytes_left, MSG_WAITALL);
1339                 if (ret < 0) {
1340                         if (errno != EAGAIN) {
1341                                 perror("recv header");
1342                                 return 1;
1343                         }
1344                         usleep(1000);
1345                         continue;
1346                 } else if (!ret) {
1347                         usleep(1000);
1348                         continue;
1349                 } else {
1350                         p += ret;
1351                         bytes_left -= ret;
1352                 }
1353         }
1354         fcntl(nc->in_fd, F_SETFL, fl & ~O_NONBLOCK);
1355         return bytes_left;
1356 }
1357
1358 /*
1359  * finalize a net client: truncate files, show stats, cleanup, etc
1360  */
1361 static void net_client_done(struct net_connection *nc)
1362 {
1363         struct device_information *dip;
1364         struct thread_information *tip;
1365         struct net_connection *last_nc;
1366         int i, j;
1367
1368         for_each_nc_dip(nc, dip, i)
1369                 __for_each_tip(dip, tip, nc->ncpus, j)
1370                         tip_ftrunc_final(tip);
1371
1372         show_stats(nc->device_information, nc->ndevs, nc->ncpus);
1373
1374         /*
1375          * cleanup for next run
1376          */
1377         for_each_nc_dip(nc, dip, i) {
1378                 __for_each_tip(dip, tip, nc->ncpus, j) {
1379                         if (tip->ofile)
1380                                 fclose(tip->ofile);
1381                 }
1382
1383                 free(dip->threads);
1384                 free(dip->path);
1385         }
1386
1387         free(nc->device_information);
1388         nc->device_information = NULL;
1389         nc->ncpus = nc->ndevs = 0;
1390
1391         close(nc->in_fd);
1392         nc->in_fd = -1;
1393
1394         net_connects--;
1395
1396         /*
1397          * now put last entry where this one was, a little nasty since we
1398          * need to adjust dip->nc as well
1399          */
1400         if (nc->connection_index != net_connects) {
1401                 last_nc = &net_connections[net_connects];
1402                 *nc = *last_nc;
1403                 for_each_nc_dip(nc, dip, i)
1404                         dip->nc = nc;
1405         }
1406
1407         stat_shown = 0;
1408 }
1409
1410 /*
1411  * handle incoming events from a net client
1412  */
1413 static int net_client_data(struct net_connection *nc)
1414 {
1415         struct thread_information *tip;
1416         struct blktrace_net_hdr bnh;
1417
1418         if (net_get_header(nc, &bnh))
1419                 return 1;
1420
1421         if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1422                 fprintf(stderr, "server: received data is bad\n");
1423                 return 1;
1424         }
1425
1426         if (!data_is_native) {
1427                 bnh.magic = be32_to_cpu(bnh.magic);
1428                 bnh.cpu = be32_to_cpu(bnh.cpu);
1429                 bnh.len = be32_to_cpu(bnh.len);
1430         }
1431
1432         if ((bnh.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
1433                 fprintf(stderr, "server: bad data magic\n");
1434                 return 1;
1435         }
1436
1437         /*
1438          * len == 0 means that the other end signalled end-of-run
1439          */
1440         if (!bnh.len) {
1441                 /*
1442                  * overload cpu count with dropped events
1443                  */
1444                 struct device_information *dip;
1445
1446                 dip = net_get_dip(nc, bnh.buts_name);
1447                 dip->drop_count = bnh.cpu;
1448                 dip->trace_started = 0;
1449
1450                 printf("server: end of run for %s\n", dip->buts_name);
1451                 net_client_done(nc);
1452                 return 0;
1453         }
1454
1455         tip = net_get_tip(nc, &bnh);
1456         if (!tip)
1457                 return 1;
1458
1459         if (mmap_subbuf(tip, bnh.len))
1460                 return 1;
1461
1462         return 0;
1463 }
1464
1465 static void net_add_connection(int listen_fd, struct sockaddr_in *addr)
1466 {
1467         socklen_t socklen = sizeof(*addr);
1468         struct net_connection *nc;
1469
1470         if (net_connects == NET_MAX_CONNECTIONS) {
1471                 fprintf(stderr, "server: no more connections allowed\n");
1472                 return;
1473         }
1474
1475         nc = &net_connections[net_connects];
1476         memset(nc, 0, sizeof(*nc));
1477
1478         nc->in_fd = accept(listen_fd, (struct sockaddr *) addr, &socklen);
1479         if (nc->in_fd < 0) {
1480                 perror("accept");
1481                 return;
1482         }
1483
1484         printf("server: connection from %s\n", inet_ntoa(addr->sin_addr));
1485         time(&nc->connect_time);
1486         nc->connection_index = net_connects;
1487         nc->cl_in_addr = addr->sin_addr;
1488         net_connects++;
1489 }
1490
1491 /*
1492  * event driven loop, handle new incoming connections and data from
1493  * existing connections
1494  */
1495 static void net_server_handle_connections(int listen_fd,
1496                                           struct sockaddr_in *addr)
1497 {
1498         struct pollfd pfds[NET_MAX_CONNECTIONS + 1];
1499         int i, events;
1500
1501         printf("server: waiting for connections...\n");
1502
1503         while (!is_done()) {
1504                 /*
1505                  * the zero entry is for incoming connections, remaining
1506                  * entries for clients
1507                  */
1508                 pfds[0].fd = listen_fd;
1509                 pfds[0].events = POLLIN;
1510                 for (i = 0; i < net_connects; i++) {
1511                         pfds[i + 1].fd = net_connections[i].in_fd;
1512                         pfds[i + 1].events = POLLIN;
1513                 }
1514
1515                 events = poll(pfds, 1 + net_connects, -1);
1516                 if (events < 0) {
1517                         if (errno == EINTR)
1518                                 continue;
1519
1520                         perror("poll");
1521                         break;
1522                 } else if (!events)
1523                         continue;
1524
1525                 if (pfds[0].revents & POLLIN) {
1526                         net_add_connection(listen_fd, addr);
1527                         events--;
1528                 }
1529
1530                 for (i = 0; events && i < net_connects; i++) {
1531                         if (pfds[i + 1].revents & POLLIN) {
1532                                 net_client_data(&net_connections[i]);
1533                                 events--;
1534                         }
1535                 }
1536         }
1537 }
1538
1539 /*
1540  * Start here when we are in server mode - just fetch data from the network
1541  * and dump to files
1542  */
1543 static int net_server(void)
1544 {
1545         struct sockaddr_in addr;
1546         int fd, opt;
1547
1548         fd = socket(AF_INET, SOCK_STREAM, 0);
1549         if (fd < 0) {
1550                 perror("server: socket");
1551                 return 1;
1552         }
1553
1554         opt = 1;
1555         if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1556                 perror("setsockopt");
1557                 return 1;
1558         }
1559
1560         memset(&addr, 0, sizeof(addr));
1561         addr.sin_family = AF_INET;
1562         addr.sin_addr.s_addr = htonl(INADDR_ANY);
1563         addr.sin_port = htons(net_port);
1564
1565         if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1566                 perror("bind");
1567                 return 1;
1568         }
1569
1570         if (listen(fd, 1) < 0) {
1571                 perror("listen");
1572                 return 1;
1573         }
1574
1575         net_server_handle_connections(fd, &addr);
1576         return 0;
1577 }
1578
1579 /*
1580  * Setup outgoing network connection where we will transmit data
1581  */
1582 static int net_setup_client(void)
1583 {
1584         struct sockaddr_in addr;
1585         int fd;
1586
1587         fd = socket(AF_INET, SOCK_STREAM, 0);
1588         if (fd < 0) {
1589                 perror("client: socket");
1590                 return 1;
1591         }
1592
1593         memset(&addr, 0, sizeof(addr));
1594         addr.sin_family = AF_INET;
1595         addr.sin_port = htons(net_port);
1596
1597         if (inet_aton(hostname, &addr.sin_addr) != 1) {
1598                 struct hostent *hent = gethostbyname(hostname);
1599                 if (!hent) {
1600                         perror("gethostbyname");
1601                         return 1;
1602                 }
1603
1604                 memcpy(&addr.sin_addr, hent->h_addr, 4);
1605                 strcpy(hostname, hent->h_name);
1606         }
1607
1608         printf("blktrace: connecting to %s\n", hostname);
1609
1610         if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1611                 perror("client: connect");
1612                 return 1;
1613         }
1614
1615         printf("blktrace: connected!\n");
1616         net_out_fd = fd;
1617         return 0;
1618 }
1619
1620 static char usage_str[] = \
1621         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1622         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1623         "\t-d Use specified device. May also be given last after options\n" \
1624         "\t-r Path to mounted relayfs, defaults to /relay\n" \
1625         "\t-o File(s) to send output to\n" \
1626         "\t-D Directory to prepend to output file names\n" \
1627         "\t-k Kill a running trace\n" \
1628         "\t-w Stop after defined time, in seconds\n" \
1629         "\t-a Only trace specified actions. See documentation\n" \
1630         "\t-A Give trace mask as a single value. See documentation\n" \
1631         "\t-b Sub buffer size in KiB\n" \
1632         "\t-n Number of sub buffers\n" \
1633         "\t-l Run in network listen mode (blktrace server)\n" \
1634         "\t-h Run in network client mode, connecting to the given host\n" \
1635         "\t-p Network port to use (default 8462)\n" \
1636         "\t-s Make the network client use sendfile() to transfer data\n" \
1637         "\t-V Print program version info\n\n";
1638
1639 static void show_usage(char *program)
1640 {
1641         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1642 }
1643
1644 int main(int argc, char *argv[])
1645 {
1646         static char default_relay_path[] = "/relay";
1647         struct statfs st;
1648         int i, c;
1649         int stop_watch = 0;
1650         int act_mask_tmp = 0;
1651
1652         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1653                 switch (c) {
1654                 case 'a':
1655                         i = find_mask_map(optarg);
1656                         if (i < 0) {
1657                                 fprintf(stderr,"Invalid action mask %s\n",
1658                                         optarg);
1659                                 return 1;
1660                         }
1661                         act_mask_tmp |= i;
1662                         break;
1663
1664                 case 'A':
1665                         if ((sscanf(optarg, "%x", &i) != 1) ||
1666                                                         !valid_act_opt(i)) {
1667                                 fprintf(stderr,
1668                                         "Invalid set action mask %s/0x%x\n",
1669                                         optarg, i);
1670                                 return 1;
1671                         }
1672                         act_mask_tmp = i;
1673                         break;
1674
1675                 case 'd':
1676                         if (resize_devices(optarg) != 0)
1677                                 return 1;
1678                         break;
1679
1680                 case 'r':
1681                         relay_path = optarg;
1682                         break;
1683
1684                 case 'o':
1685                         output_name = optarg;
1686                         break;
1687                 case 'k':
1688                         kill_running_trace = 1;
1689                         break;
1690                 case 'w':
1691                         stop_watch = atoi(optarg);
1692                         if (stop_watch <= 0) {
1693                                 fprintf(stderr,
1694                                         "Invalid stopwatch value (%d secs)\n",
1695                                         stop_watch);
1696                                 return 1;
1697                         }
1698                         break;
1699                 case 'V':
1700                         printf("%s version %s\n", argv[0], blktrace_version);
1701                         return 0;
1702                 case 'b':
1703                         buf_size = strtoul(optarg, NULL, 10);
1704                         if (buf_size <= 0 || buf_size > 16*1024) {
1705                                 fprintf(stderr,
1706                                         "Invalid buffer size (%lu)\n",buf_size);
1707                                 return 1;
1708                         }
1709                         buf_size <<= 10;
1710                         break;
1711                 case 'n':
1712                         buf_nr = strtoul(optarg, NULL, 10);
1713                         if (buf_nr <= 0) {
1714                                 fprintf(stderr,
1715                                         "Invalid buffer nr (%lu)\n", buf_nr);
1716                                 return 1;
1717                         }
1718                         break;
1719                 case 'D':
1720                         output_dir = optarg;
1721                         break;
1722                 case 'h':
1723                         net_mode = Net_client;
1724                         strcpy(hostname, optarg);
1725                         break;
1726                 case 'l':
1727                         net_mode = Net_server;
1728                         break;
1729                 case 'p':
1730                         net_port = atoi(optarg);
1731                         break;
1732                 case 's':
1733                         net_use_sendfile = 1;
1734                         break;
1735                 default:
1736                         show_usage(argv[0]);
1737                         return 1;
1738                 }
1739         }
1740
1741         setlocale(LC_NUMERIC, "en_US");
1742
1743         page_size = getpagesize();
1744
1745         if (net_mode == Net_server)
1746                 return net_server();
1747
1748         while (optind < argc) {
1749                 if (resize_devices(argv[optind++]) != 0)
1750                         return 1;
1751         }
1752
1753         if (ndevs == 0) {
1754                 show_usage(argv[0]);
1755                 return 1;
1756         }
1757
1758         if (!relay_path)
1759                 relay_path = default_relay_path;
1760
1761         if (act_mask_tmp != 0)
1762                 act_mask = act_mask_tmp;
1763
1764         if (statfs(relay_path, &st) < 0) {
1765                 perror("statfs");
1766                 fprintf(stderr,"%s does not appear to be a valid path\n",
1767                         relay_path);
1768                 return 1;
1769         } else if (st.f_type != (long) RELAYFS_TYPE) {
1770                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1771                         relay_path);
1772                 return 1;
1773         }
1774
1775         if (open_devices() != 0)
1776                 return 1;
1777
1778         if (kill_running_trace) {
1779                 stop_all_traces();
1780                 return 0;
1781         }
1782
1783         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1784         if (ncpus < 0) {
1785                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1786                 return 1;
1787         }
1788
1789         signal(SIGINT, handle_sigint);
1790         signal(SIGHUP, handle_sigint);
1791         signal(SIGTERM, handle_sigint);
1792         signal(SIGALRM, handle_sigint);
1793
1794         if (net_mode == Net_client && net_setup_client())
1795                 return 1;
1796
1797         if (start_devices() != 0)
1798                 return 1;
1799
1800         atexit(stop_all_tracing);
1801
1802         if (stop_watch)
1803                 alarm(stop_watch);
1804
1805         wait_for_threads();
1806
1807         if (!is_trace_stopped()) {
1808                 trace_stopped = 1;
1809                 stop_all_threads();
1810                 stop_all_traces();
1811         }
1812
1813         show_stats(device_information, ndevs, ncpus);
1814
1815         return 0;
1816 }
1817