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:"
  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 = NULL,
 150         }
 151 };
 152
 153 struct tip_subbuf {
 154         void *buf;
 155         unsigned int len;
 156         unsigned int max_len;
 157         off_t offset;
 158 };
 159
 160 #define FIFO_SIZE       (1024)  /* should be plenty big! */
 161 #define CL_SIZE         (128)   /* cache line, any bigger? */
 162
 163 struct tip_subbuf_fifo {
 164         int tail __attribute__((aligned(CL_SIZE)));
 165         int head __attribute__((aligned(CL_SIZE)));
 166         struct tip_subbuf *q[FIFO_SIZE];
 167 };
 168
 169 struct thread_information {
 170         int cpu;
 171         pthread_t thread;
 172
 173         int fd;
 174         void *fd_buf;
 175         char fn[MAXPATHLEN + 64];
 176
 177         char ofname[64];
 178
 179         FILE *ofile;
 180         off_t ofile_offset;
 181         char *ofile_buffer;
 182         int ofile_stdout;
 183         int ofile_mmap;
 184
 185         unsigned long events_processed;
 186         unsigned long long data_read;
 187         struct device_information *device;
 188
 189         int exited;
 190
 191         /*
 192          * piped fifo buffers
 193          */
 194         struct tip_subbuf_fifo fifo;
 195         struct tip_subbuf *leftover_ts;
 196
 197         /*
 198          * mmap controlled output files
 199          */
 200         unsigned long long fs_size;
 201         unsigned long long fs_max_size;
 202         unsigned long fs_off;
 203         void *fs_buf;
 204         unsigned long fs_buf_len;
 205 };
 206
 207 struct device_information {
 208         int fd;
 209         char *path;
 210         char buts_name[32];
 211         volatile int trace_started;
 212         unsigned long drop_count;
 213         struct thread_information *threads;
 214 };
 215
 216 static int ncpus;
 217 static struct thread_information *thread_information;
 218 static int ndevs;
 219 static struct device_information *device_information;
 220
 221 /* command line option globals */
 222 static char *relay_path;
 223 static char *output_name;
 224 static char *output_dir;
 225 static int act_mask = ~0U;
 226 static int kill_running_trace;
 227 static unsigned long buf_size = BUF_SIZE;
 228 static unsigned long buf_nr = BUF_NR;
 229 static unsigned int page_size;
 230
 231 #define is_done()       (*(volatile int *)(&done))
 232 static volatile int done;
 233
 234 #define is_trace_stopped()      (*(volatile int *)(&trace_stopped))
 235 static volatile int trace_stopped;
 236
 237 #define is_stat_shown() (*(volatile int *)(&stat_shown))
 238 static volatile int stat_shown;
 239
 240 int data_is_native = -1;
 241
 242 static void exit_trace(int status);
 243
 244 #define dip_tracing(dip)        (*(volatile int *)(&(dip)->trace_started))
 245 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
 246
 247 #define __for_each_dip(__d, __i, __e)   \
 248         for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
 249
 250 #define for_each_dip(__d, __i)  __for_each_dip(__d, __i, ndevs)
 251 #define for_each_tip(__d, __t, __j)     \
 252         for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
 253
 254 /*
 255  * networking stuff follows. we include a magic number so we know whether
 256  * to endianness convert or not
 257  */
 258 struct blktrace_net_hdr {
 259         u32 magic;              /* same as trace magic */
 260         char ofname[64];        /* trace name */
 261         u32 cpu;                /* for which cpu */
 262         u32 len;                /* length of following trace data */
 263 };
 264
 265 #define TRACE_NET_PORT          (8462)
 266
 267 enum {
 268         Net_none = 0,
 269         Net_server,
 270         Net_client,
 271 };
 272
 273 /*
 274  * network cmd line params
 275  */
 276 static char hostname[MAXHOSTNAMELEN];
 277 static int net_port = TRACE_NET_PORT;
 278 static int net_mode = 0;
 279
 280 static int net_in_fd = -1;
 281 static int net_out_fd = -1;
 282
 283 static void handle_sigint(__attribute__((__unused__)) int sig)
 284 {
 285         done = 1;
 286 }
 287
 288 static int get_dropped_count(const char *buts_name)
 289 {
 290         int fd;
 291         char tmp[MAXPATHLEN + 64];
 292
 293         snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
 294                  relay_path, buts_name);
 295
 296         fd = open(tmp, O_RDONLY);
 297         if (fd < 0) {
 298                 /*
 299                  * this may be ok, if the kernel doesn't support dropped counts
 300                  */
 301                 if (errno == ENOENT)
 302                         return 0;
 303
 304                 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
 305                 return -1;
 306         }
 307
 308         if (read(fd, tmp, sizeof(tmp)) < 0) {
 309                 perror(tmp);
 310                 close(fd);
 311                 return -1;
 312         }
 313
 314         close(fd);
 315
 316         return atoi(tmp);
 317 }
 318
 319 static int start_trace(struct device_information *dip)
 320 {
 321         struct blk_user_trace_setup buts;
 322
 323         memset(&buts, 0, sizeof(buts));
 324         buts.buf_size = buf_size;
 325         buts.buf_nr = buf_nr;
 326         buts.act_mask = act_mask;
 327
 328         if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
 329                 perror("BLKTRACESETUP");
 330                 return 1;
 331         }
 332
 333         if (ioctl(dip->fd, BLKTRACESTART) < 0) {
 334                 perror("BLKTRACESTART");
 335                 return 1;
 336         }
 337
 338         memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
 339         dip_set_tracing(dip, 1);
 340         return 0;
 341 }
 342
 343 static void stop_trace(struct device_information *dip)
 344 {
 345         if (dip_tracing(dip) || kill_running_trace) {
 346                 dip_set_tracing(dip, 0);
 347
 348                 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
 349                         perror("BLKTRACESTOP");
 350                 if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
 351                         perror("BLKTRACETEARDOWN");
 352
 353                 close(dip->fd);
 354                 dip->fd = -1;
 355         }
 356 }
 357
 358 static void stop_all_traces(void)
 359 {
 360         struct device_information *dip;
 361         int i;
 362
 363         for_each_dip(dip, i) {
 364                 dip->drop_count = get_dropped_count(dip->buts_name);
 365                 stop_trace(dip);
 366         }
 367 }
 368
 369 static void wait_for_data(struct thread_information *tip)
 370 {
 371         struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
 372
 373         do {
 374                 poll(&pfd, 1, 100);
 375                 if (pfd.revents & POLLIN)
 376                         break;
 377                 if (tip->ofile_stdout)
 378                         break;
 379         } while (!is_done());
 380 }
 381
 382 static int read_data_file(struct thread_information *tip, void *buf, int len)
 383 {
 384         int ret = 0;
 385
 386         do {
 387                 wait_for_data(tip);
 388
 389                 ret = read(tip->fd, buf, len);
 390                 if (!ret)
 391                         continue;
 392                 else if (ret > 0)
 393                         return ret;
 394                 else {
 395                         if (errno != EAGAIN) {
 396                                 perror(tip->fn);
 397                                 fprintf(stderr,"Thread %d failed read of %s\n",
 398                                         tip->cpu, tip->fn);
 399                                 break;
 400                         }
 401                         continue;
 402                 }
 403         } while (!is_done());
 404
 405         return ret;
 406
 407 }
 408
 409 static int read_data_net(struct thread_information *tip, void *buf, int len)
 410 {
 411         unsigned int bytes_left = len;
 412         int ret = 0;
 413
 414         do {
 415                 ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
 416
 417                 if (!ret)
 418                         continue;
 419                 else if (ret < 0) {
 420                         if (errno != EAGAIN) {
 421                                 perror(tip->fn);
 422                                 fprintf(stderr, "server: failed read\n");
 423                                 return 0;
 424                         }
 425                         continue;
 426                 } else {
 427                         buf += ret;
 428                         bytes_left -= ret;
 429                 }
 430         } while (!is_done() && bytes_left);
 431
 432         return len;
 433 }
 434
 435 static int read_data(struct thread_information *tip, void *buf, int len)
 436 {
 437         if (net_mode == Net_server)
 438                 return read_data_net(tip, buf, len);
 439
 440         return read_data_file(tip, buf, len);
 441 }
 442
 443 static inline struct tip_subbuf *
 444 subbuf_fifo_dequeue(struct thread_information *tip)
 445 {
 446         const int head = tip->fifo.head;
 447         const int next = (head + 1) & (FIFO_SIZE - 1);
 448
 449         if (head != tip->fifo.tail) {
 450                 struct tip_subbuf *ts = tip->fifo.q[head];
 451
 452                 store_barrier();
 453                 tip->fifo.head = next;
 454                 return ts;
 455         }
 456
 457         return NULL;
 458 }
 459
 460 static inline int subbuf_fifo_queue(struct thread_information *tip,
 461                                     struct tip_subbuf *ts)
 462 {
 463         const int tail = tip->fifo.tail;
 464         const int next = (tail + 1) & (FIFO_SIZE - 1);
 465
 466         if (next != tip->fifo.head) {
 467                 tip->fifo.q[tail] = ts;
 468                 store_barrier();
 469                 tip->fifo.tail = next;
 470                 return 0;
 471         }
 472
 473         fprintf(stderr, "fifo too small!\n");
 474         return 1;
 475 }
 476
 477 /*
 478  * For file output, truncate and mmap the file appropriately
 479  */
 480 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
 481 {
 482         int ofd = fileno(tip->ofile);
 483         int ret;
 484
 485         /*
 486          * extend file, if we have to. use chunks of 16 subbuffers.
 487          */
 488         if (tip->fs_off + buf_size > tip->fs_buf_len) {
 489                 if (tip->fs_buf) {
 490                         munlock(tip->fs_buf, tip->fs_buf_len);
 491                         munmap(tip->fs_buf, tip->fs_buf_len);
 492                         tip->fs_buf = NULL;
 493                 }
 494
 495                 tip->fs_off = tip->fs_size & (page_size - 1);
 496                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 497                 tip->fs_max_size += tip->fs_buf_len;
 498
 499                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 500                         perror("ftruncate");
 501                         return -1;
 502                 }
 503
 504                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 505                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 506                 if (tip->fs_buf == MAP_FAILED) {
 507                         perror("mmap");
 508                         return -1;
 509                 }
 510                 mlock(tip->fs_buf, tip->fs_buf_len);
 511         }
 512
 513         ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
 514         if (ret >= 0) {
 515                 tip->data_read += ret;
 516                 tip->fs_size += ret;
 517                 tip->fs_off += ret;
 518                 return 0;
 519         }
 520
 521         return -1;
 522 }
 523
 524 /*
 525  * Use the copy approach for pipes
 526  */
 527 static int get_subbuf(struct thread_information *tip)
 528 {
 529         struct tip_subbuf *ts = malloc(sizeof(*ts));
 530         int ret = 1;
 531
 532         /*
 533          * for the net client, don't actually read the data in as we will
 534          * use sendfile to transmit it. just mark how much data we have
 535          */
 536         if (net_mode == Net_client) {
 537                 struct stat sb;
 538
 539                 ts->buf = NULL;
 540
 541                 if (fstat(tip->fd, &sb) < 0) {
 542                         perror("trace stat");
 543                         goto out;
 544                 }
 545
 546                 ts->len = sb.st_size - tip->ofile_offset;
 547                 ts->max_len = ts->len;
 548                 ts->offset = tip->ofile_offset;
 549                 tip->ofile_offset += ts->len;
 550                 ret = subbuf_fifo_queue(tip, ts);
 551         } else {
 552                 ts->buf = malloc(buf_size);
 553                 ts->max_len = buf_size;
 554
 555                 ret = read_data(tip, ts->buf, ts->max_len);
 556                 if (ret > 0) {
 557                         ts->len = ret;
 558                         ret = subbuf_fifo_queue(tip, ts);
 559                 }
 560         }
 561
 562 out:
 563         if (ret) {
 564                 if (ts->buf)
 565                         free(ts->buf);
 566                 free(ts);
 567         }
 568
 569         return ret;
 570 }
 571
 572 static void close_thread(struct thread_information *tip)
 573 {
 574         if (tip->fd != -1)
 575                 close(tip->fd);
 576         if (tip->ofile)
 577                 fclose(tip->ofile);
 578         if (tip->ofile_buffer)
 579                 free(tip->ofile_buffer);
 580         if (tip->fd_buf)
 581                 free(tip->fd_buf);
 582
 583         tip->fd = -1;
 584         tip->ofile = NULL;
 585         tip->ofile_buffer = NULL;
 586         tip->fd_buf = NULL;
 587 }
 588
 589 static void tip_ftrunc_final(struct thread_information *tip)
 590 {
 591         /*
 592          * truncate to right size and cleanup mmap
 593          */
 594         if (tip->ofile_mmap) {
 595                 int ofd = fileno(tip->ofile);
 596
 597                 if (tip->fs_buf)
 598                         munmap(tip->fs_buf, tip->fs_buf_len);
 599
 600                 ftruncate(ofd, tip->fs_size);
 601         }
 602 }
 603
 604 static void *thread_main(void *arg)
 605 {
 606         struct thread_information *tip = arg;
 607         pid_t pid = getpid();
 608         cpu_set_t cpu_mask;
 609
 610         CPU_ZERO(&cpu_mask);
 611         CPU_SET((tip->cpu), &cpu_mask);
 612
 613         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 614                 perror("sched_setaffinity");
 615                 exit_trace(1);
 616         }
 617
 618         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 619                         relay_path, tip->device->buts_name, tip->cpu);
 620         tip->fd = open(tip->fn, O_RDONLY);
 621         if (tip->fd < 0) {
 622                 perror(tip->fn);
 623                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 624                         tip->fn);
 625                 exit_trace(1);
 626         }
 627
 628         while (!is_done()) {
 629                 if (tip->ofile_mmap && net_mode != Net_client) {
 630                         if (mmap_subbuf(tip, buf_size))
 631                                 break;
 632                 } else {
 633                         if (get_subbuf(tip))
 634                                 break;
 635                 }
 636         }
 637
 638         tip_ftrunc_final(tip);
 639         tip->exited = 1;
 640         return NULL;
 641 }
 642
 643 static int write_data_net(int fd, void *buf, unsigned int buf_len)
 644 {
 645         unsigned int bytes_left = buf_len;
 646         int ret;
 647
 648         while (bytes_left) {
 649                 ret = send(fd, buf, bytes_left, 0);
 650                 if (ret < 0) {
 651                         perror("send");
 652                         return 1;
 653                 }
 654
 655                 buf += ret;
 656                 bytes_left -= ret;
 657         }
 658
 659         return 0;
 660 }
 661
 662 static int flush_subbuf_net(struct thread_information *tip,
 663                             struct tip_subbuf *ts)
 664 {
 665         struct blktrace_net_hdr hdr;
 666         int ret = 0;
 667
 668         hdr.magic = BLK_IO_TRACE_MAGIC;
 669         strcpy(hdr.ofname, tip->ofname);
 670         hdr.cpu = tip->cpu;
 671         hdr.len = ts->len;
 672
 673         if (write_data_net(net_out_fd, &hdr, sizeof(hdr)))
 674                 return 1;
 675
 676         if (sendfile(net_out_fd, tip->fd, &ts->offset, ts->len) < 0) {
 677                 perror("sendfile");
 678                 ret = 1;
 679         }
 680
 681         free(ts);
 682         return 0;
 683 }
 684
 685 static int write_data(struct thread_information *tip, void *buf,
 686                       unsigned int buf_len)
 687 {
 688         int ret;
 689
 690         if (!buf_len)
 691                 return 0;
 692
 693         while (1) {
 694                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 695                 if (ret == 1)
 696                         break;
 697
 698                 if (ret < 0) {
 699                         perror("write");
 700                         return 1;
 701                 }
 702         }
 703
 704         if (tip->ofile_stdout)
 705                 fflush(tip->ofile);
 706
 707         return 0;
 708 }
 709
 710 static int flush_subbuf_file(struct thread_information *tip,
 711                              struct tip_subbuf *ts)
 712 {
 713         unsigned int offset = 0;
 714         struct blk_io_trace *t;
 715         int pdu_len, events = 0;
 716
 717         /*
 718          * surplus from last run
 719          */
 720         if (tip->leftover_ts) {
 721                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 722
 723                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 724                         prev_ts->max_len += ts->len;
 725                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 726                 }
 727
 728                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 729                 prev_ts->len += ts->len;
 730
 731                 free(ts->buf);
 732                 free(ts);
 733
 734                 ts = prev_ts;
 735                 tip->leftover_ts = NULL;
 736         }
 737
 738         while (offset + sizeof(*t) <= ts->len) {
 739                 t = ts->buf + offset;
 740
 741                 if (verify_trace(t)) {
 742                         write_data(tip, ts->buf, offset);
 743                         return -1;
 744                 }
 745
 746                 pdu_len = t->pdu_len;
 747
 748                 if (offset + sizeof(*t) + pdu_len > ts->len)
 749                         break;
 750
 751                 offset += sizeof(*t) + pdu_len;
 752                 tip->events_processed++;
 753                 tip->data_read += sizeof(*t) + pdu_len;
 754                 events++;
 755         }
 756
 757         if (write_data(tip, ts->buf, offset))
 758                 return -1;
 759
 760         /*
 761          * leftover bytes, save them for next time
 762          */
 763         if (offset != ts->len) {
 764                 tip->leftover_ts = ts;
 765                 ts->len -= offset;
 766                 memmove(ts->buf, ts->buf + offset, ts->len);
 767         } else {
 768                 free(ts->buf);
 769                 free(ts);
 770         }
 771
 772         return events;
 773 }
 774
 775 static int write_tip_events(struct thread_information *tip)
 776 {
 777         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 778
 779         if (ts) {
 780                 if (net_mode == Net_client)
 781                         return flush_subbuf_net(tip, ts);
 782
 783                 return flush_subbuf_file(tip, ts);
 784         }
 785
 786         return 0;
 787 }
 788
 789 /*
 790  * scans the tips we know and writes out the subbuffers we accumulate
 791  */
 792 static void get_and_write_events(void)
 793 {
 794         struct device_information *dip;
 795         struct thread_information *tip;
 796         int i, j, events, ret, tips_running;
 797
 798         while (!is_done()) {
 799                 events = 0;
 800
 801                 for_each_dip(dip, i) {
 802                         for_each_tip(dip, tip, j) {
 803                                 ret = write_tip_events(tip);
 804                                 if (ret > 0)
 805                                         events += ret;
 806                         }
 807                 }
 808
 809                 if (!events)
 810                         usleep(10);
 811         }
 812
 813         /*
 814          * reap stored events
 815          */
 816         do {
 817                 events = 0;
 818                 tips_running = 0;
 819                 for_each_dip(dip, i) {
 820                         for_each_tip(dip, tip, j) {
 821                                 ret = write_tip_events(tip);
 822                                 if (ret > 0)
 823                                         events += ret;
 824                                 tips_running += !tip->exited;
 825                         }
 826                 }
 827                 usleep(10);
 828         } while (events || tips_running);
 829 }
 830
 831 static void wait_for_threads(void)
 832 {
 833         /*
 834          * for piped or network output, poll and fetch data for writeout.
 835          * for files, we just wait around for trace threads to exit
 836          */
 837         if ((output_name && !strcmp(output_name, "-")) ||
 838             net_mode == Net_client)
 839                 get_and_write_events();
 840         else {
 841                 struct device_information *dip;
 842                 struct thread_information *tip;
 843                 int i, j, tips_running;
 844
 845                 do {
 846                         tips_running = 0;
 847                         usleep(1000);
 848
 849                         for_each_dip(dip, i)
 850                                 for_each_tip(dip, tip, j)
 851                                         tips_running += !tip->exited;
 852                 } while (tips_running);
 853         }
 854 }
 855
 856 static void fill_ofname(char *dst, char *buts_name, int cpu)
 857 {
 858         int len = 0;
 859
 860         if (output_dir)
 861                 len = sprintf(dst, "%s/", output_dir);
 862
 863         if (output_name)
 864                 sprintf(dst + len, "%s.blktrace.%d", output_name, cpu);
 865         else
 866                 sprintf(dst + len, "%s.blktrace.%d", buts_name, cpu);
 867 }
 868
 869 static int start_threads(struct device_information *dip)
 870 {
 871         struct thread_information *tip;
 872         int j, pipeline = output_name && !strcmp(output_name, "-");
 873         int mode, vbuf_size;
 874
 875         for_each_tip(dip, tip, j) {
 876                 tip->cpu = j;
 877                 tip->device = dip;
 878                 tip->events_processed = 0;
 879                 memset(&tip->fifo, 0, sizeof(tip->fifo));
 880                 tip->leftover_ts = NULL;
 881
 882                 if (pipeline) {
 883                         tip->ofile = fdopen(STDOUT_FILENO, "w");
 884                         tip->ofile_stdout = 1;
 885                         tip->ofile_mmap = 0;
 886                         mode = _IOLBF;
 887                         vbuf_size = 512;
 888                 } else {
 889                         fill_ofname(tip->ofname, dip->buts_name, tip->cpu);
 890                         tip->ofile = fopen(tip->ofname, "w+");
 891                         tip->ofile_stdout = 0;
 892                         tip->ofile_mmap = 1;
 893                         mode = _IOFBF;
 894                         vbuf_size = OFILE_BUF;
 895                 }
 896
 897                 if (tip->ofile == NULL) {
 898                         perror(tip->ofname);
 899                         return 1;
 900                 }
 901
 902                 tip->ofile_buffer = malloc(vbuf_size);
 903                 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
 904                         perror("setvbuf");
 905                         close_thread(tip);
 906                         return 1;
 907                 }
 908
 909                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
 910                         perror("pthread_create");
 911                         close_thread(tip);
 912                         return 1;
 913                 }
 914         }
 915
 916         return 0;
 917 }
 918
 919 static void stop_threads(struct device_information *dip)
 920 {
 921         struct thread_information *tip;
 922         unsigned long ret;
 923         int i;
 924
 925         for_each_tip(dip, tip, i) {
 926                 (void) pthread_join(tip->thread, (void *) &ret);
 927                 close_thread(tip);
 928         }
 929 }
 930
 931 static void stop_all_threads(void)
 932 {
 933         struct device_information *dip;
 934         int i;
 935
 936         for_each_dip(dip, i)
 937                 stop_threads(dip);
 938 }
 939
 940 static void stop_all_tracing(void)
 941 {
 942         struct device_information *dip;
 943         int i;
 944
 945         for_each_dip(dip, i)
 946                 stop_trace(dip);
 947 }
 948
 949 static void exit_trace(int status)
 950 {
 951         if (!is_trace_stopped()) {
 952                 trace_stopped = 1;
 953                 stop_all_threads();
 954                 stop_all_tracing();
 955         }
 956
 957         exit(status);
 958 }
 959
 960 static int resize_devices(char *path)
 961 {
 962         int size = (ndevs + 1) * sizeof(struct device_information);
 963
 964         device_information = realloc(device_information, size);
 965         if (!device_information) {
 966                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
 967                 return 1;
 968         }
 969         device_information[ndevs].path = path;
 970         ndevs++;
 971         return 0;
 972 }
 973
 974 static int open_devices(void)
 975 {
 976         struct device_information *dip;
 977         int i;
 978
 979         for_each_dip(dip, i) {
 980                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
 981                 if (dip->fd < 0) {
 982                         perror(dip->path);
 983                         return 1;
 984                 }
 985         }
 986
 987         return 0;
 988 }
 989
 990 static int start_devices(void)
 991 {
 992         struct device_information *dip;
 993         int i, j, size;
 994
 995         size = ncpus * sizeof(struct thread_information);
 996         thread_information = malloc(size * ndevs);
 997         if (!thread_information) {
 998                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
 999                 return 1;
1000         }
1001
1002         for_each_dip(dip, i) {
1003                 if (start_trace(dip)) {
1004                         close(dip->fd);
1005                         fprintf(stderr, "Failed to start trace on %s\n",
1006                                 dip->path);
1007                         break;
1008                 }
1009         }
1010
1011         if (i != ndevs) {
1012                 __for_each_dip(dip, j, i)
1013                         stop_trace(dip);
1014
1015                 return 1;
1016         }
1017
1018         for_each_dip(dip, i) {
1019                 dip->threads = thread_information + (i * ncpus);
1020                 if (start_threads(dip)) {
1021                         fprintf(stderr, "Failed to start worker threads\n");
1022                         break;
1023                 }
1024         }
1025
1026         if (i != ndevs) {
1027                 __for_each_dip(dip, j, i)
1028                         stop_threads(dip);
1029                 for_each_dip(dip, i)
1030                         stop_trace(dip);
1031
1032                 return 1;
1033         }
1034
1035         return 0;
1036 }
1037
1038 static void show_stats(void)
1039 {
1040         struct device_information *dip;
1041         struct thread_information *tip;
1042         unsigned long long events_processed, data_read;
1043         unsigned long total_drops;
1044         int i, j, no_stdout = 0;
1045
1046         if (is_stat_shown())
1047                 return;
1048
1049         if (output_name && !strcmp(output_name, "-"))
1050                 no_stdout = 1;
1051
1052         stat_shown = 1;
1053
1054         total_drops = 0;
1055         for_each_dip(dip, i) {
1056                 if (!no_stdout)
1057                         printf("Device: %s\n", dip->path);
1058                 events_processed = 0;
1059                 data_read = 0;
1060                 for_each_tip(dip, tip, j) {
1061                         if (!no_stdout)
1062                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
1063                                         tip->cpu, tip->events_processed,
1064                                         tip->data_read >> 10);
1065                         events_processed += tip->events_processed;
1066                         data_read += tip->data_read;
1067                 }
1068                 total_drops += dip->drop_count;
1069                 if (!no_stdout)
1070                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
1071                                         events_processed, dip->drop_count,
1072                                         data_read >> 10);
1073         }
1074
1075         if (total_drops)
1076                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1077 }
1078
1079 static struct thread_information *net_tip_list;
1080 static int net_tip_len;
1081
1082 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh)
1083 {
1084         struct thread_information *tip;
1085         int i;
1086
1087         for (i = 0; i < net_tip_len; i++) {
1088                 tip = &net_tip_list[i];
1089
1090                 if (!strcmp(bnh->ofname, tip->ofname))
1091                         return tip;
1092         }
1093
1094         net_tip_list = realloc(net_tip_list, (net_tip_len + 1) * sizeof(*tip));
1095         tip = &net_tip_list[net_tip_len];
1096         net_tip_len++;
1097
1098         memset(tip, 0, sizeof(*tip));
1099
1100         tip->cpu = bnh->cpu;
1101         strcpy(tip->ofname, bnh->ofname);
1102
1103         tip->ofile = fopen(tip->ofname, "w+");
1104         if (!tip->ofile) {
1105                 perror("fopen");
1106                 return NULL;
1107         }
1108
1109         tip->ofile_stdout = 0;
1110         tip->ofile_mmap = 1;
1111         return tip;
1112 }
1113
1114 static int net_get_header(struct blktrace_net_hdr *bnh)
1115 {
1116         int fl = fcntl(net_in_fd, F_GETFL);
1117         int bytes_left, ret;
1118         void *p = bnh;
1119
1120         fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1121         bytes_left = sizeof(*bnh);
1122         while (bytes_left && !is_done()) {
1123                 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1124                 if (ret < 0) {
1125                         if (errno != EAGAIN) {
1126                                 perror("recv header");
1127                                 return 1;
1128                         }
1129                         usleep(100);
1130                         continue;
1131                 } else if (!ret) {
1132                         usleep(100);
1133                         continue;
1134                 } else {
1135                         p += ret;
1136                         bytes_left -= ret;
1137                 }
1138         }
1139         fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1140         return 0;
1141 }
1142
1143 static int net_server_loop(void)
1144 {
1145         struct thread_information *tip;
1146         struct blktrace_net_hdr bnh;
1147
1148         if (net_get_header(&bnh))
1149                 return 1;
1150
1151         if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1152                 fprintf(stderr, "server: received data is bad\n");
1153                 return 1;
1154         }
1155
1156         if (!data_is_native) {
1157                 bnh.cpu = be32_to_cpu(bnh.cpu);
1158                 bnh.len = be32_to_cpu(bnh.len);
1159         }
1160
1161         tip = net_get_tip(&bnh);
1162         if (!tip)
1163                 return 1;
1164
1165         if (mmap_subbuf(tip, bnh.len))
1166                 return 1;
1167
1168         return 0;
1169 }
1170
1171 /*
1172  * Start here when we are in server mode - just fetch data from the network
1173  * and dump to files
1174  */
1175 static int net_server(void)
1176 {
1177         struct sockaddr_in addr;
1178         socklen_t socklen;
1179         int fd, opt, i;
1180
1181         fd = socket(AF_INET, SOCK_STREAM, 0);
1182         if (fd < 0) {
1183                 perror("server: socket");
1184                 return 1;
1185         }
1186
1187         opt = 1;
1188         if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1189                 perror("setsockopt");
1190                 return 1;
1191         }
1192
1193         memset(&addr, 0, sizeof(addr));
1194         addr.sin_family = AF_INET;
1195         addr.sin_addr.s_addr = htonl(INADDR_ANY);
1196         addr.sin_port = htons(net_port);
1197
1198         if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1199                 perror("bind");
1200                 return 1;
1201         }
1202
1203         if (listen(fd, 1) < 0) {
1204                 perror("listen");
1205                 return 1;
1206         }
1207
1208         printf("blktrace: waiting for incoming connection...\n");
1209
1210         socklen = sizeof(addr);
1211         net_in_fd = accept(fd, (struct sockaddr *) &addr, &socklen);
1212         if (net_in_fd < 0) {
1213                 perror("accept");
1214                 return 1;
1215         }
1216
1217         signal(SIGINT, handle_sigint);
1218         signal(SIGHUP, handle_sigint);
1219         signal(SIGTERM, handle_sigint);
1220         signal(SIGALRM, handle_sigint);
1221
1222         printf("blktrace: connected!\n");
1223
1224         while (!is_done()) {
1225                 if (net_server_loop())
1226                         break;
1227         }
1228
1229         for (i = 0; i < net_tip_len; i++)
1230                 tip_ftrunc_final(&net_tip_list[i]);
1231
1232         return 0;
1233 }
1234
1235 /*
1236  * Setup outgoing network connection where we will transmit data
1237  */
1238 static int net_setup_client(void)
1239 {
1240         struct sockaddr_in addr;
1241         int fd;
1242
1243         fd = socket(AF_INET, SOCK_STREAM, 0);
1244         if (fd < 0) {
1245                 perror("client: socket");
1246                 return 1;
1247         }
1248
1249         memset(&addr, 0, sizeof(addr));
1250         addr.sin_family = AF_INET;
1251         addr.sin_port = htons(net_port);
1252
1253         if (inet_aton(hostname, &addr.sin_addr) != 1) {
1254                 struct hostent *hent = gethostbyname(hostname);
1255                 if (!hent) {
1256                         perror("gethostbyname");
1257                         return 1;
1258                 }
1259
1260                 memcpy(&addr.sin_addr, hent->h_addr, 4);
1261                 strcpy(hostname, hent->h_name);
1262         }
1263
1264         printf("blktrace: connecting to %s\n", hostname);
1265
1266         if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1267                 perror("client: connect");
1268                 return 1;
1269         }
1270
1271         printf("blktrace: connected!\n");
1272         net_out_fd = fd;
1273         return 0;
1274 }
1275
1276 static char usage_str[] = \
1277         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1278         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1279         "\t-d Use specified device. May also be given last after options\n" \
1280         "\t-r Path to mounted relayfs, defaults to /relay\n" \
1281         "\t-o File(s) to send output to\n" \
1282         "\t-D Directory to prepend to output file names\n" \
1283         "\t-k Kill a running trace\n" \
1284         "\t-w Stop after defined time, in seconds\n" \
1285         "\t-a Only trace specified actions. See documentation\n" \
1286         "\t-A Give trace mask as a single value. See documentation\n" \
1287         "\t-b Sub buffer size in KiB\n" \
1288         "\t-n Number of sub buffers\n" \
1289         "\t-v Print program version info\n\n";
1290
1291 static void show_usage(char *program)
1292 {
1293         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1294 }
1295
1296 int main(int argc, char *argv[])
1297 {
1298         static char default_relay_path[] = "/relay";
1299         struct statfs st;
1300         int i, c;
1301         int stop_watch = 0;
1302         int act_mask_tmp = 0;
1303
1304         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1305                 switch (c) {
1306                 case 'a':
1307                         i = find_mask_map(optarg);
1308                         if (i < 0) {
1309                                 fprintf(stderr,"Invalid action mask %s\n",
1310                                         optarg);
1311                                 return 1;
1312                         }
1313                         act_mask_tmp |= i;
1314                         break;
1315
1316                 case 'A':
1317                         if ((sscanf(optarg, "%x", &i) != 1) ||
1318                                                         !valid_act_opt(i)) {
1319                                 fprintf(stderr,
1320                                         "Invalid set action mask %s/0x%x\n",
1321                                         optarg, i);
1322                                 return 1;
1323                         }
1324                         act_mask_tmp = i;
1325                         break;
1326
1327                 case 'd':
1328                         if (resize_devices(optarg) != 0)
1329                                 return 1;
1330                         break;
1331
1332                 case 'r':
1333                         relay_path = optarg;
1334                         break;
1335
1336                 case 'o':
1337                         output_name = optarg;
1338                         break;
1339                 case 'k':
1340                         kill_running_trace = 1;
1341                         break;
1342                 case 'w':
1343                         stop_watch = atoi(optarg);
1344                         if (stop_watch <= 0) {
1345                                 fprintf(stderr,
1346                                         "Invalid stopwatch value (%d secs)\n",
1347                                         stop_watch);
1348                                 return 1;
1349                         }
1350                         break;
1351                 case 'V':
1352                         printf("%s version %s\n", argv[0], blktrace_version);
1353                         return 0;
1354                 case 'b':
1355                         buf_size = strtoul(optarg, NULL, 10);
1356                         if (buf_size <= 0 || buf_size > 16*1024) {
1357                                 fprintf(stderr,
1358                                         "Invalid buffer size (%lu)\n",buf_size);
1359                                 return 1;
1360                         }
1361                         buf_size <<= 10;
1362                         break;
1363                 case 'n':
1364                         buf_nr = strtoul(optarg, NULL, 10);
1365                         if (buf_nr <= 0) {
1366                                 fprintf(stderr,
1367                                         "Invalid buffer nr (%lu)\n", buf_nr);
1368                                 return 1;
1369                         }
1370                         break;
1371                 case 'D':
1372                         output_dir = optarg;
1373                         break;
1374                 case 'h':
1375                         net_mode = Net_client;
1376                         strcpy(hostname, optarg);
1377                         break;
1378                 case 'l':
1379                         net_mode = Net_server;
1380                         break;
1381                 case 'p':
1382                         net_port = atoi(optarg);
1383                         break;
1384                 default:
1385                         show_usage(argv[0]);
1386                         return 1;
1387                 }
1388         }
1389
1390         while (optind < argc) {
1391                 if (resize_devices(argv[optind++]) != 0)
1392                         return 1;
1393         }
1394
1395         setlocale(LC_NUMERIC, "en_US");
1396
1397         page_size = getpagesize();
1398
1399         if (net_mode == Net_server)
1400                 return net_server();
1401
1402         if (ndevs == 0) {
1403                 show_usage(argv[0]);
1404                 return 1;
1405         }
1406
1407         if (!relay_path)
1408                 relay_path = default_relay_path;
1409
1410         if (act_mask_tmp != 0)
1411                 act_mask = act_mask_tmp;
1412
1413         if (statfs(relay_path, &st) < 0) {
1414                 perror("statfs");
1415                 fprintf(stderr,"%s does not appear to be a valid path\n",
1416                         relay_path);
1417                 return 1;
1418         } else if (st.f_type != (long) RELAYFS_TYPE) {
1419                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1420                         relay_path);
1421                 return 1;
1422         }
1423
1424         if (open_devices() != 0)
1425                 return 1;
1426
1427         if (kill_running_trace) {
1428                 stop_all_traces();
1429                 return 0;
1430         }
1431
1432         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1433         if (ncpus < 0) {
1434                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1435                 return 1;
1436         }
1437
1438         signal(SIGINT, handle_sigint);
1439         signal(SIGHUP, handle_sigint);
1440         signal(SIGTERM, handle_sigint);
1441         signal(SIGALRM, handle_sigint);
1442
1443         if (net_mode == Net_client && net_setup_client())
1444                 return 1;
1445
1446         if (start_devices() != 0)
1447                 return 1;
1448
1449         atexit(stop_all_tracing);
1450
1451         if (stop_watch)
1452                 alarm(stop_watch);
1453
1454         wait_for_threads();
1455
1456         if (!is_trace_stopped()) {
1457                 trace_stopped = 1;
1458                 stop_all_threads();
1459                 stop_all_traces();
1460         }
1461
1462         show_stats();
1463
1464         return 0;
1465 }
1466