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
  45 #include "blktrace.h"
  46 #include "barrier.h"
  47
  48 static char blktrace_version[] = "0.99";
  49
  50 /*
  51  * You may want to increase this even more, if you are logging at a high
  52  * rate and see skipped/missed events
  53  */
  54 #define BUF_SIZE        (512 * 1024)
  55 #define BUF_NR          (4)
  56
  57 #define OFILE_BUF       (128 * 1024)
  58
  59 #define RELAYFS_TYPE    0xF0B4A981
  60
  61 #define S_OPTS  "d:a:A:r:o:kw:Vb:n:D:lh:p:"
  62 static struct option l_opts[] = {
  63         {
  64                 .name = "dev",
  65                 .has_arg = required_argument,
  66                 .flag = NULL,
  67                 .val = 'd'
  68         },
  69         {
  70                 .name = "act-mask",
  71                 .has_arg = required_argument,
  72                 .flag = NULL,
  73                 .val = 'a'
  74         },
  75         {
  76                 .name = "set-mask",
  77                 .has_arg = required_argument,
  78                 .flag = NULL,
  79                 .val = 'A'
  80         },
  81         {
  82                 .name = "relay",
  83                 .has_arg = required_argument,
  84                 .flag = NULL,
  85                 .val = 'r'
  86         },
  87         {
  88                 .name = "output",
  89                 .has_arg = required_argument,
  90                 .flag = NULL,
  91                 .val = 'o'
  92         },
  93         {
  94                 .name = "kill",
  95                 .has_arg = no_argument,
  96                 .flag = NULL,
  97                 .val = 'k'
  98         },
  99         {
 100                 .name = "stopwatch",
 101                 .has_arg = required_argument,
 102                 .flag = NULL,
 103                 .val = 'w'
 104         },
 105         {
 106                 .name = "version",
 107                 .has_arg = no_argument,
 108                 .flag = NULL,
 109                 .val = 'V'
 110         },
 111         {
 112                 .name = "buffer-size",
 113                 .has_arg = required_argument,
 114                 .flag = NULL,
 115                 .val = 'b'
 116         },
 117         {
 118                 .name = "num-sub-buffers",
 119                 .has_arg = required_argument,
 120                 .flag = NULL,
 121                 .val = 'n'
 122         },
 123         {
 124                 .name = "output-dir",
 125                 .has_arg = required_argument,
 126                 .flag = NULL,
 127                 .val = 'D'
 128         },
 129         {
 130                 .name = "listen",
 131                 .has_arg = no_argument,
 132                 .flag = NULL,
 133                 .val = 'l'
 134         },
 135         {
 136                 .name = "host",
 137                 .has_arg = required_argument,
 138                 .flag = NULL,
 139                 .val = 'h'
 140         },
 141         {
 142                 .name = "port",
 143                 .has_arg = required_argument,
 144                 .flag = NULL,
 145                 .val = 'p'
 146         },
 147         {
 148                 .name = NULL,
 149         }
 150 };
 151
 152 struct tip_subbuf {
 153         void *buf;
 154         unsigned int len;
 155         unsigned int max_len;
 156 };
 157
 158 #define FIFO_SIZE       (1024)  /* should be plenty big! */
 159 #define CL_SIZE         (128)   /* cache line, any bigger? */
 160
 161 struct tip_subbuf_fifo {
 162         int tail __attribute__((aligned(CL_SIZE)));
 163         int head __attribute__((aligned(CL_SIZE)));
 164         struct tip_subbuf *q[FIFO_SIZE];
 165 };
 166
 167 struct thread_information {
 168         int cpu;
 169         pthread_t thread;
 170
 171         int fd;
 172         void *fd_buf;
 173         char fn[MAXPATHLEN + 64];
 174
 175         FILE *ofile;
 176         char *ofile_buffer;
 177         int ofile_stdout;
 178         int ofile_mmap;
 179
 180         int (*get_subbuf)(struct thread_information *, unsigned int);
 181         int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *);
 182         int (*read_data)(struct thread_information *, void *, unsigned int);
 183
 184         unsigned long events_processed;
 185         unsigned long long data_read;
 186         struct device_information *device;
 187
 188         int exited;
 189
 190         /*
 191          * piped fifo buffers
 192          */
 193         struct tip_subbuf_fifo fifo;
 194         struct tip_subbuf *leftover_ts;
 195
 196         /*
 197          * mmap controlled output files
 198          */
 199         unsigned long long fs_size;
 200         unsigned long long fs_max_size;
 201         unsigned long fs_off;
 202         void *fs_buf;
 203         unsigned long fs_buf_len;
 204 };
 205
 206 struct device_information {
 207         int fd;
 208         char *path;
 209         char buts_name[32];
 210         volatile int trace_started;
 211         unsigned long drop_count;
 212         struct thread_information *threads;
 213 };
 214
 215 static int ncpus;
 216 static struct thread_information *thread_information;
 217 static int ndevs;
 218 static struct device_information *device_information;
 219
 220 /* command line option globals */
 221 static char *relay_path;
 222 static char *output_name;
 223 static char *output_dir;
 224 static int act_mask = ~0U;
 225 static int kill_running_trace;
 226 static unsigned long buf_size = BUF_SIZE;
 227 static unsigned long buf_nr = BUF_NR;
 228 static unsigned int page_size;
 229
 230 #define is_done()       (*(volatile int *)(&done))
 231 static volatile int done;
 232
 233 #define is_trace_stopped()      (*(volatile int *)(&trace_stopped))
 234 static volatile int trace_stopped;
 235
 236 #define is_stat_shown() (*(volatile int *)(&stat_shown))
 237 static volatile int stat_shown;
 238
 239 int data_is_native = -1;
 240
 241 static void exit_trace(int status);
 242
 243 #define dip_tracing(dip)        (*(volatile int *)(&(dip)->trace_started))
 244 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
 245
 246 #define __for_each_dip(__d, __i, __e)   \
 247         for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
 248
 249 #define for_each_dip(__d, __i)  __for_each_dip(__d, __i, ndevs)
 250 #define for_each_tip(__d, __t, __j)     \
 251         for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
 252
 253 /*
 254  * networking stuff follows. we include a magic number so we know whether
 255  * to endianness convert or not
 256  */
 257 struct blktrace_net_hdr {
 258         u32 magic;              /* same as trace magic */
 259         char buts_name[32];     /* trace name */
 260         u32 cpu;                /* for which cpu */
 261         u32 max_cpus;
 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,
 383                           unsigned int len)
 384 {
 385         int ret = 0;
 386
 387         do {
 388                 wait_for_data(tip);
 389
 390                 ret = read(tip->fd, buf, len);
 391                 if (!ret)
 392                         continue;
 393                 else if (ret > 0)
 394                         return ret;
 395                 else {
 396                         if (errno != EAGAIN) {
 397                                 perror(tip->fn);
 398                                 fprintf(stderr,"Thread %d failed read of %s\n",
 399                                         tip->cpu, tip->fn);
 400                                 break;
 401                         }
 402                         continue;
 403                 }
 404         } while (!is_done());
 405
 406         return ret;
 407
 408 }
 409
 410 static int read_data_net(struct thread_information *tip, void *buf,
 411                          unsigned int len)
 412 {
 413         unsigned int bytes_left = len;
 414         int ret = 0;
 415
 416         do {
 417                 ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
 418
 419                 if (!ret)
 420                         continue;
 421                 else if (ret < 0) {
 422                         if (errno != EAGAIN) {
 423                                 perror(tip->fn);
 424                                 fprintf(stderr, "server: failed read\n");
 425                                 return 0;
 426                         }
 427                         continue;
 428                 } else {
 429                         buf += ret;
 430                         bytes_left -= ret;
 431                 }
 432         } while (!is_done() && bytes_left);
 433
 434         return len - bytes_left;
 435 }
 436
 437 static int read_data(struct thread_information *tip, void *buf,
 438                      unsigned int len)
 439 {
 440         int ret = tip->read_data(tip, buf, len);
 441
 442         if (ret > 0)
 443                 tip->data_read += ret;
 444
 445         return ret;
 446 }
 447
 448 static inline struct tip_subbuf *
 449 subbuf_fifo_dequeue(struct thread_information *tip)
 450 {
 451         const int head = tip->fifo.head;
 452         const int next = (head + 1) & (FIFO_SIZE - 1);
 453
 454         if (head != tip->fifo.tail) {
 455                 struct tip_subbuf *ts = tip->fifo.q[head];
 456
 457                 store_barrier();
 458                 tip->fifo.head = next;
 459                 return ts;
 460         }
 461
 462         return NULL;
 463 }
 464
 465 static inline int subbuf_fifo_queue(struct thread_information *tip,
 466                                     struct tip_subbuf *ts)
 467 {
 468         const int tail = tip->fifo.tail;
 469         const int next = (tail + 1) & (FIFO_SIZE - 1);
 470
 471         if (next != tip->fifo.head) {
 472                 tip->fifo.q[tail] = ts;
 473                 store_barrier();
 474                 tip->fifo.tail = next;
 475                 return 0;
 476         }
 477
 478         fprintf(stderr, "fifo too small!\n");
 479         return 1;
 480 }
 481
 482 /*
 483  * For file output, truncate and mmap the file appropriately
 484  */
 485 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
 486 {
 487         int ofd = fileno(tip->ofile);
 488         int ret;
 489
 490         /*
 491          * extend file, if we have to. use chunks of 16 subbuffers.
 492          */
 493         if (tip->fs_off + buf_size > tip->fs_buf_len) {
 494                 if (tip->fs_buf) {
 495                         munlock(tip->fs_buf, tip->fs_buf_len);
 496                         munmap(tip->fs_buf, tip->fs_buf_len);
 497                         tip->fs_buf = NULL;
 498                 }
 499
 500                 tip->fs_off = tip->fs_size & (page_size - 1);
 501                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 502                 tip->fs_max_size += tip->fs_buf_len;
 503
 504                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 505                         perror("ftruncate");
 506                         return -1;
 507                 }
 508
 509                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 510                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 511                 if (tip->fs_buf == MAP_FAILED) {
 512                         perror("mmap");
 513                         return -1;
 514                 }
 515                 mlock(tip->fs_buf, tip->fs_buf_len);
 516         }
 517
 518         ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
 519         if (ret >= 0) {
 520                 tip->fs_size += ret;
 521                 tip->fs_off += ret;
 522                 return 0;
 523         }
 524
 525         return -1;
 526 }
 527
 528 /*
 529  * Use the copy approach for pipes and network
 530  */
 531 static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
 532 {
 533         struct tip_subbuf *ts = malloc(sizeof(*ts));
 534         int ret;
 535
 536         ts->buf = malloc(buf_size);
 537         ts->max_len = maxlen;
 538
 539         ret = read_data(tip, ts->buf, ts->max_len);
 540         if (ret > 0) {
 541                 ts->len = ret;
 542                 return subbuf_fifo_queue(tip, ts);
 543         }
 544
 545         return ret;
 546 }
 547
 548 static void close_thread(struct thread_information *tip)
 549 {
 550         if (tip->fd != -1)
 551                 close(tip->fd);
 552         if (tip->ofile)
 553                 fclose(tip->ofile);
 554         if (tip->ofile_buffer)
 555                 free(tip->ofile_buffer);
 556         if (tip->fd_buf)
 557                 free(tip->fd_buf);
 558
 559         tip->fd = -1;
 560         tip->ofile = NULL;
 561         tip->ofile_buffer = NULL;
 562         tip->fd_buf = NULL;
 563 }
 564
 565 static void tip_ftrunc_final(struct thread_information *tip)
 566 {
 567         /*
 568          * truncate to right size and cleanup mmap
 569          */
 570         if (tip->ofile_mmap) {
 571                 int ofd = fileno(tip->ofile);
 572
 573                 if (tip->fs_buf)
 574                         munmap(tip->fs_buf, tip->fs_buf_len);
 575
 576                 ftruncate(ofd, tip->fs_size);
 577         }
 578 }
 579
 580 static void *thread_main(void *arg)
 581 {
 582         struct thread_information *tip = arg;
 583         pid_t pid = getpid();
 584         cpu_set_t cpu_mask;
 585
 586         CPU_ZERO(&cpu_mask);
 587         CPU_SET((tip->cpu), &cpu_mask);
 588
 589         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 590                 perror("sched_setaffinity");
 591                 exit_trace(1);
 592         }
 593
 594         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 595                         relay_path, tip->device->buts_name, tip->cpu);
 596         tip->fd = open(tip->fn, O_RDONLY);
 597         if (tip->fd < 0) {
 598                 perror(tip->fn);
 599                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 600                         tip->fn);
 601                 exit_trace(1);
 602         }
 603
 604         while (!is_done()) {
 605                 if (tip->get_subbuf(tip, buf_size))
 606                         break;
 607         }
 608
 609         tip_ftrunc_final(tip);
 610         tip->exited = 1;
 611         return NULL;
 612 }
 613
 614 static int write_data_net(int fd, void *buf, unsigned int buf_len)
 615 {
 616         unsigned int bytes_left = buf_len;
 617         int ret;
 618
 619         while (bytes_left) {
 620                 ret = send(fd, buf, bytes_left, 0);
 621                 if (ret < 0) {
 622                         perror("send");
 623                         return 1;
 624                 }
 625
 626                 buf += ret;
 627                 bytes_left -= ret;
 628         }
 629
 630         return 0;
 631 }
 632
 633 static int flush_subbuf_net(struct thread_information *tip,
 634                             struct tip_subbuf *ts)
 635 {
 636         struct blktrace_net_hdr hdr;
 637
 638         hdr.magic = BLK_IO_TRACE_MAGIC;
 639         strcpy(hdr.buts_name, tip->device->buts_name);
 640         hdr.cpu = tip->cpu;
 641         hdr.max_cpus = ncpus;
 642         hdr.len = ts->len;
 643
 644         if (write_data_net(net_out_fd, &hdr, sizeof(hdr)))
 645                 return 1;
 646
 647         if (write_data_net(net_out_fd, ts->buf, ts->len))
 648                 return 1;
 649
 650         free(ts->buf);
 651         free(ts);
 652         return 0;
 653 }
 654
 655 static int write_data(struct thread_information *tip, void *buf,
 656                       unsigned int buf_len)
 657 {
 658         int ret;
 659
 660         if (!buf_len)
 661                 return 0;
 662
 663         while (1) {
 664                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 665                 if (ret == 1)
 666                         break;
 667
 668                 if (ret < 0) {
 669                         perror("write");
 670                         return 1;
 671                 }
 672         }
 673
 674         if (tip->ofile_stdout)
 675                 fflush(tip->ofile);
 676
 677         return 0;
 678 }
 679
 680 static int flush_subbuf_file(struct thread_information *tip,
 681                              struct tip_subbuf *ts)
 682 {
 683         unsigned int offset = 0;
 684         struct blk_io_trace *t;
 685         int pdu_len, events = 0;
 686
 687         /*
 688          * surplus from last run
 689          */
 690         if (tip->leftover_ts) {
 691                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 692
 693                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 694                         prev_ts->max_len += ts->len;
 695                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 696                 }
 697
 698                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 699                 prev_ts->len += ts->len;
 700
 701                 free(ts->buf);
 702                 free(ts);
 703
 704                 ts = prev_ts;
 705                 tip->leftover_ts = NULL;
 706         }
 707
 708         while (offset + sizeof(*t) <= ts->len) {
 709                 t = ts->buf + offset;
 710
 711                 if (verify_trace(t)) {
 712                         write_data(tip, ts->buf, offset);
 713                         return -1;
 714                 }
 715
 716                 pdu_len = t->pdu_len;
 717
 718                 if (offset + sizeof(*t) + pdu_len > ts->len)
 719                         break;
 720
 721                 offset += sizeof(*t) + pdu_len;
 722                 tip->events_processed++;
 723                 tip->data_read += sizeof(*t) + pdu_len;
 724                 events++;
 725         }
 726
 727         if (write_data(tip, ts->buf, offset))
 728                 return -1;
 729
 730         /*
 731          * leftover bytes, save them for next time
 732          */
 733         if (offset != ts->len) {
 734                 tip->leftover_ts = ts;
 735                 ts->len -= offset;
 736                 memmove(ts->buf, ts->buf + offset, ts->len);
 737         } else {
 738                 free(ts->buf);
 739                 free(ts);
 740         }
 741
 742         return events;
 743 }
 744
 745 static int write_tip_events(struct thread_information *tip)
 746 {
 747         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 748
 749         if (ts)
 750                 return tip->flush_subbuf(tip, ts);
 751
 752         return 0;
 753 }
 754
 755 /*
 756  * scans the tips we know and writes out the subbuffers we accumulate
 757  */
 758 static void get_and_write_events(void)
 759 {
 760         struct device_information *dip;
 761         struct thread_information *tip;
 762         int i, j, events, ret, tips_running;
 763
 764         while (!is_done()) {
 765                 events = 0;
 766
 767                 for_each_dip(dip, i) {
 768                         for_each_tip(dip, tip, j) {
 769                                 ret = write_tip_events(tip);
 770                                 if (ret > 0)
 771                                         events += ret;
 772                         }
 773                 }
 774
 775                 if (!events)
 776                         usleep(10);
 777         }
 778
 779         /*
 780          * reap stored events
 781          */
 782         do {
 783                 events = 0;
 784                 tips_running = 0;
 785                 for_each_dip(dip, i) {
 786                         for_each_tip(dip, tip, j) {
 787                                 ret = write_tip_events(tip);
 788                                 if (ret > 0)
 789                                         events += ret;
 790                                 tips_running += !tip->exited;
 791                         }
 792                 }
 793                 usleep(10);
 794         } while (events || tips_running);
 795 }
 796
 797 static void wait_for_threads(void)
 798 {
 799         /*
 800          * for piped or network output, poll and fetch data for writeout.
 801          * for files, we just wait around for trace threads to exit
 802          */
 803         if ((output_name && !strcmp(output_name, "-")) ||
 804             net_mode == Net_client)
 805                 get_and_write_events();
 806         else {
 807                 struct device_information *dip;
 808                 struct thread_information *tip;
 809                 int i, j, tips_running;
 810
 811                 do {
 812                         tips_running = 0;
 813                         usleep(1000);
 814
 815                         for_each_dip(dip, i)
 816                                 for_each_tip(dip, tip, j)
 817                                         tips_running += !tip->exited;
 818                 } while (tips_running);
 819         }
 820 }
 821
 822 static void fill_ofname(char *dst, char *buts_name, int cpu)
 823 {
 824         int len = 0;
 825
 826         if (output_dir)
 827                 len = sprintf(dst, "%s/", output_dir);
 828
 829         if (output_name)
 830                 sprintf(dst + len, "%s.blktrace.%d", output_name, cpu);
 831         else
 832                 sprintf(dst + len, "%s.blktrace.%d", buts_name, cpu);
 833 }
 834
 835 static void fill_ops(struct thread_information *tip)
 836 {
 837         /*
 838          * setup ops
 839          */
 840         if (tip->ofile_mmap && net_mode != Net_client)
 841                 tip->get_subbuf = mmap_subbuf;
 842         else
 843                 tip->get_subbuf = get_subbuf;
 844
 845         if (net_mode == Net_client)
 846                 tip->flush_subbuf = flush_subbuf_net;
 847         else
 848                 tip->flush_subbuf = flush_subbuf_file;
 849
 850         if (net_mode == Net_server)
 851                 tip->read_data = read_data_net;
 852         else
 853                 tip->read_data = read_data_file;
 854 }
 855
 856 static int start_threads(struct device_information *dip)
 857 {
 858         struct thread_information *tip;
 859         int j, pipeline = output_name && !strcmp(output_name, "-");
 860         int mode, vbuf_size;
 861         char op[64];
 862
 863         for_each_tip(dip, tip, j) {
 864                 tip->cpu = j;
 865                 tip->device = dip;
 866                 tip->events_processed = 0;
 867                 memset(&tip->fifo, 0, sizeof(tip->fifo));
 868                 tip->leftover_ts = NULL;
 869
 870                 if (pipeline) {
 871                         tip->ofile = fdopen(STDOUT_FILENO, "w");
 872                         tip->ofile_stdout = 1;
 873                         tip->ofile_mmap = 0;
 874                         mode = _IOLBF;
 875                         vbuf_size = 512;
 876                 } else {
 877                         fill_ofname(op, dip->buts_name, tip->cpu);
 878                         tip->ofile = fopen(op, "w+");
 879                         tip->ofile_stdout = 0;
 880                         tip->ofile_mmap = 1;
 881                         mode = _IOFBF;
 882                         vbuf_size = OFILE_BUF;
 883                 }
 884
 885                 if (tip->ofile == NULL) {
 886                         perror(op);
 887                         return 1;
 888                 }
 889
 890                 tip->ofile_buffer = malloc(vbuf_size);
 891                 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
 892                         perror("setvbuf");
 893                         close_thread(tip);
 894                         return 1;
 895                 }
 896
 897                 fill_ops(tip);
 898
 899                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
 900                         perror("pthread_create");
 901                         close_thread(tip);
 902                         return 1;
 903                 }
 904         }
 905
 906         return 0;
 907 }
 908
 909 static void stop_threads(struct device_information *dip)
 910 {
 911         struct thread_information *tip;
 912         unsigned long ret;
 913         int i;
 914
 915         for_each_tip(dip, tip, i) {
 916                 (void) pthread_join(tip->thread, (void *) &ret);
 917                 close_thread(tip);
 918         }
 919 }
 920
 921 static void stop_all_threads(void)
 922 {
 923         struct device_information *dip;
 924         int i;
 925
 926         for_each_dip(dip, i)
 927                 stop_threads(dip);
 928 }
 929
 930 static void stop_all_tracing(void)
 931 {
 932         struct device_information *dip;
 933         int i;
 934
 935         for_each_dip(dip, i)
 936                 stop_trace(dip);
 937 }
 938
 939 static void exit_trace(int status)
 940 {
 941         if (!is_trace_stopped()) {
 942                 trace_stopped = 1;
 943                 stop_all_threads();
 944                 stop_all_tracing();
 945         }
 946
 947         exit(status);
 948 }
 949
 950 static int resize_devices(char *path)
 951 {
 952         int size = (ndevs + 1) * sizeof(struct device_information);
 953
 954         device_information = realloc(device_information, size);
 955         if (!device_information) {
 956                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
 957                 return 1;
 958         }
 959         device_information[ndevs].path = path;
 960         ndevs++;
 961         return 0;
 962 }
 963
 964 static int open_devices(void)
 965 {
 966         struct device_information *dip;
 967         int i;
 968
 969         for_each_dip(dip, i) {
 970                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
 971                 if (dip->fd < 0) {
 972                         perror(dip->path);
 973                         return 1;
 974                 }
 975         }
 976
 977         return 0;
 978 }
 979
 980 static int start_devices(void)
 981 {
 982         struct device_information *dip;
 983         int i, j, size;
 984
 985         size = ncpus * sizeof(struct thread_information);
 986         thread_information = malloc(size * ndevs);
 987         if (!thread_information) {
 988                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
 989                 return 1;
 990         }
 991
 992         for_each_dip(dip, i) {
 993                 if (start_trace(dip)) {
 994                         close(dip->fd);
 995                         fprintf(stderr, "Failed to start trace on %s\n",
 996                                 dip->path);
 997                         break;
 998                 }
 999         }
1000
1001         if (i != ndevs) {
1002                 __for_each_dip(dip, j, i)
1003                         stop_trace(dip);
1004
1005                 return 1;
1006         }
1007
1008         for_each_dip(dip, i) {
1009                 dip->threads = thread_information + (i * ncpus);
1010                 if (start_threads(dip)) {
1011                         fprintf(stderr, "Failed to start worker threads\n");
1012                         break;
1013                 }
1014         }
1015
1016         if (i != ndevs) {
1017                 __for_each_dip(dip, j, i)
1018                         stop_threads(dip);
1019                 for_each_dip(dip, i)
1020                         stop_trace(dip);
1021
1022                 return 1;
1023         }
1024
1025         return 0;
1026 }
1027
1028 static void show_stats(void)
1029 {
1030         struct device_information *dip;
1031         struct thread_information *tip;
1032         unsigned long long events_processed, data_read;
1033         unsigned long total_drops;
1034         int i, j, no_stdout = 0;
1035
1036         if (is_stat_shown())
1037                 return;
1038
1039         if (output_name && !strcmp(output_name, "-"))
1040                 no_stdout = 1;
1041
1042         stat_shown = 1;
1043
1044         total_drops = 0;
1045         for_each_dip(dip, i) {
1046                 if (!no_stdout)
1047                         printf("Device: %s\n", dip->path);
1048                 events_processed = 0;
1049                 data_read = 0;
1050                 for_each_tip(dip, tip, j) {
1051                         if (!no_stdout)
1052                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
1053                                         tip->cpu, tip->events_processed,
1054                                         tip->data_read >> 10);
1055                         events_processed += tip->events_processed;
1056                         data_read += tip->data_read;
1057                 }
1058                 total_drops += dip->drop_count;
1059                 if (!no_stdout)
1060                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
1061                                         events_processed, dip->drop_count,
1062                                         data_read >> 10);
1063         }
1064
1065         if (total_drops)
1066                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1067 }
1068
1069 static struct device_information *net_get_dip(char *buts_name)
1070 {
1071         struct device_information *dip;
1072         int i;
1073
1074         for (i = 0; i < ndevs; i++) {
1075                 dip = &device_information[i];
1076
1077                 if (!strcmp(dip->buts_name, buts_name))
1078                         return dip;
1079         }
1080
1081         device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
1082         dip = &device_information[ndevs];
1083         strcpy(dip->buts_name, buts_name);
1084         strcpy(dip->path, buts_name);
1085         ndevs++;
1086         dip->threads = malloc(ncpus * sizeof(struct thread_information));
1087         memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
1088
1089         /*
1090          * open all files
1091          */
1092         for (i = 0; i < ncpus; i++) {
1093                 struct thread_information *tip = &dip->threads[i];
1094                 char op[64];
1095
1096                 tip->cpu = i;
1097                 tip->ofile_stdout = 0;
1098                 tip->ofile_mmap = 1;
1099                 tip->device = dip;
1100
1101                 fill_ofname(op, dip->buts_name, tip->cpu);
1102
1103                 tip->ofile = fopen(op, "w+");
1104                 if (!tip->ofile) {
1105                         perror("fopen");
1106                         return NULL;
1107                 }
1108         }
1109
1110         return dip;
1111 }
1112
1113 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh)
1114 {
1115         struct device_information *dip;
1116
1117         ncpus = bnh->max_cpus;
1118         dip = net_get_dip(bnh->buts_name);
1119         return &dip->threads[bnh->cpu];
1120 }
1121
1122 static int net_get_header(struct blktrace_net_hdr *bnh)
1123 {
1124         int fl = fcntl(net_in_fd, F_GETFL);
1125         int bytes_left, ret;
1126         void *p = bnh;
1127
1128         fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1129         bytes_left = sizeof(*bnh);
1130         while (bytes_left && !is_done()) {
1131                 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1132                 if (ret < 0) {
1133                         if (errno != EAGAIN) {
1134                                 perror("recv header");
1135                                 return 1;
1136                         }
1137                         usleep(100);
1138                         continue;
1139                 } else if (!ret) {
1140                         usleep(100);
1141                         continue;
1142                 } else {
1143                         p += ret;
1144                         bytes_left -= ret;
1145                 }
1146         }
1147         fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1148         return 0;
1149 }
1150
1151 static int net_server_loop(void)
1152 {
1153         struct thread_information *tip;
1154         struct blktrace_net_hdr bnh;
1155
1156         if (net_get_header(&bnh))
1157                 return 1;
1158
1159         if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1160                 fprintf(stderr, "server: received data is bad\n");
1161                 return 1;
1162         }
1163
1164         if (!data_is_native) {
1165                 bnh.cpu = be32_to_cpu(bnh.cpu);
1166                 bnh.len = be32_to_cpu(bnh.len);
1167         }
1168
1169         tip = net_get_tip(&bnh);
1170         if (!tip)
1171                 return 1;
1172
1173         if (mmap_subbuf(tip, bnh.len))
1174                 return 1;
1175
1176         return 0;
1177 }
1178
1179 /*
1180  * Start here when we are in server mode - just fetch data from the network
1181  * and dump to files
1182  */
1183 static int net_server(void)
1184 {
1185         struct sockaddr_in addr;
1186         socklen_t socklen;
1187         int fd, opt, i, j;
1188
1189         fd = socket(AF_INET, SOCK_STREAM, 0);
1190         if (fd < 0) {
1191                 perror("server: socket");
1192                 return 1;
1193         }
1194
1195         opt = 1;
1196         if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1197                 perror("setsockopt");
1198                 return 1;
1199         }
1200
1201         memset(&addr, 0, sizeof(addr));
1202         addr.sin_family = AF_INET;
1203         addr.sin_addr.s_addr = htonl(INADDR_ANY);
1204         addr.sin_port = htons(net_port);
1205
1206         if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1207                 perror("bind");
1208                 return 1;
1209         }
1210
1211         if (listen(fd, 1) < 0) {
1212                 perror("listen");
1213                 return 1;
1214         }
1215
1216         printf("blktrace: waiting for incoming connection...\n");
1217
1218         socklen = sizeof(addr);
1219         net_in_fd = accept(fd, (struct sockaddr *) &addr, &socklen);
1220         if (net_in_fd < 0) {
1221                 perror("accept");
1222                 return 1;
1223         }
1224
1225         signal(SIGINT, handle_sigint);
1226         signal(SIGHUP, handle_sigint);
1227         signal(SIGTERM, handle_sigint);
1228         signal(SIGALRM, handle_sigint);
1229
1230         printf("blktrace: connected!\n");
1231
1232         while (!is_done()) {
1233                 if (net_server_loop())
1234                         break;
1235         }
1236
1237         for (i = 0; i < ndevs; i++) {
1238                 struct device_information *dip = &device_information[i];
1239
1240                 for (j = 0; j < ncpus; j++)
1241                         tip_ftrunc_final(&dip->threads[j]);
1242         }
1243
1244         show_stats();
1245         return 0;
1246 }
1247
1248 /*
1249  * Setup outgoing network connection where we will transmit data
1250  */
1251 static int net_setup_client(void)
1252 {
1253         struct sockaddr_in addr;
1254         int fd;
1255
1256         fd = socket(AF_INET, SOCK_STREAM, 0);
1257         if (fd < 0) {
1258                 perror("client: socket");
1259                 return 1;
1260         }
1261
1262         memset(&addr, 0, sizeof(addr));
1263         addr.sin_family = AF_INET;
1264         addr.sin_port = htons(net_port);
1265
1266         if (inet_aton(hostname, &addr.sin_addr) != 1) {
1267                 struct hostent *hent = gethostbyname(hostname);
1268                 if (!hent) {
1269                         perror("gethostbyname");
1270                         return 1;
1271                 }
1272
1273                 memcpy(&addr.sin_addr, hent->h_addr, 4);
1274                 strcpy(hostname, hent->h_name);
1275         }
1276
1277         printf("blktrace: connecting to %s\n", hostname);
1278
1279         if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1280                 perror("client: connect");
1281                 return 1;
1282         }
1283
1284         printf("blktrace: connected!\n");
1285         net_out_fd = fd;
1286         return 0;
1287 }
1288
1289 static char usage_str[] = \
1290         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1291         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1292         "\t-d Use specified device. May also be given last after options\n" \
1293         "\t-r Path to mounted relayfs, defaults to /relay\n" \
1294         "\t-o File(s) to send output to\n" \
1295         "\t-D Directory to prepend to output file names\n" \
1296         "\t-k Kill a running trace\n" \
1297         "\t-w Stop after defined time, in seconds\n" \
1298         "\t-a Only trace specified actions. See documentation\n" \
1299         "\t-A Give trace mask as a single value. See documentation\n" \
1300         "\t-b Sub buffer size in KiB\n" \
1301         "\t-n Number of sub buffers\n" \
1302         "\t-v Print program version info\n\n";
1303
1304 static void show_usage(char *program)
1305 {
1306         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1307 }
1308
1309 int main(int argc, char *argv[])
1310 {
1311         static char default_relay_path[] = "/relay";
1312         struct statfs st;
1313         int i, c;
1314         int stop_watch = 0;
1315         int act_mask_tmp = 0;
1316
1317         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1318                 switch (c) {
1319                 case 'a':
1320                         i = find_mask_map(optarg);
1321                         if (i < 0) {
1322                                 fprintf(stderr,"Invalid action mask %s\n",
1323                                         optarg);
1324                                 return 1;
1325                         }
1326                         act_mask_tmp |= i;
1327                         break;
1328
1329                 case 'A':
1330                         if ((sscanf(optarg, "%x", &i) != 1) ||
1331                                                         !valid_act_opt(i)) {
1332                                 fprintf(stderr,
1333                                         "Invalid set action mask %s/0x%x\n",
1334                                         optarg, i);
1335                                 return 1;
1336                         }
1337                         act_mask_tmp = i;
1338                         break;
1339
1340                 case 'd':
1341                         if (resize_devices(optarg) != 0)
1342                                 return 1;
1343                         break;
1344
1345                 case 'r':
1346                         relay_path = optarg;
1347                         break;
1348
1349                 case 'o':
1350                         output_name = optarg;
1351                         break;
1352                 case 'k':
1353                         kill_running_trace = 1;
1354                         break;
1355                 case 'w':
1356                         stop_watch = atoi(optarg);
1357                         if (stop_watch <= 0) {
1358                                 fprintf(stderr,
1359                                         "Invalid stopwatch value (%d secs)\n",
1360                                         stop_watch);
1361                                 return 1;
1362                         }
1363                         break;
1364                 case 'V':
1365                         printf("%s version %s\n", argv[0], blktrace_version);
1366                         return 0;
1367                 case 'b':
1368                         buf_size = strtoul(optarg, NULL, 10);
1369                         if (buf_size <= 0 || buf_size > 16*1024) {
1370                                 fprintf(stderr,
1371                                         "Invalid buffer size (%lu)\n",buf_size);
1372                                 return 1;
1373                         }
1374                         buf_size <<= 10;
1375                         break;
1376                 case 'n':
1377                         buf_nr = strtoul(optarg, NULL, 10);
1378                         if (buf_nr <= 0) {
1379                                 fprintf(stderr,
1380                                         "Invalid buffer nr (%lu)\n", buf_nr);
1381                                 return 1;
1382                         }
1383                         break;
1384                 case 'D':
1385                         output_dir = optarg;
1386                         break;
1387                 case 'h':
1388                         net_mode = Net_client;
1389                         strcpy(hostname, optarg);
1390                         break;
1391                 case 'l':
1392                         net_mode = Net_server;
1393                         break;
1394                 case 'p':
1395                         net_port = atoi(optarg);
1396                         break;
1397                 default:
1398                         show_usage(argv[0]);
1399                         return 1;
1400                 }
1401         }
1402
1403         setlocale(LC_NUMERIC, "en_US");
1404
1405         page_size = getpagesize();
1406
1407         if (net_mode == Net_server)
1408                 return net_server();
1409
1410         while (optind < argc) {
1411                 if (resize_devices(argv[optind++]) != 0)
1412                         return 1;
1413         }
1414
1415         if (ndevs == 0) {
1416                 show_usage(argv[0]);
1417                 return 1;
1418         }
1419
1420         if (!relay_path)
1421                 relay_path = default_relay_path;
1422
1423         if (act_mask_tmp != 0)
1424                 act_mask = act_mask_tmp;
1425
1426         if (statfs(relay_path, &st) < 0) {
1427                 perror("statfs");
1428                 fprintf(stderr,"%s does not appear to be a valid path\n",
1429                         relay_path);
1430                 return 1;
1431         } else if (st.f_type != (long) RELAYFS_TYPE) {
1432                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1433                         relay_path);
1434                 return 1;
1435         }
1436
1437         if (open_devices() != 0)
1438                 return 1;
1439
1440         if (kill_running_trace) {
1441                 stop_all_traces();
1442                 return 0;
1443         }
1444
1445         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1446         if (ncpus < 0) {
1447                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1448                 return 1;
1449         }
1450
1451         signal(SIGINT, handle_sigint);
1452         signal(SIGHUP, handle_sigint);
1453         signal(SIGTERM, handle_sigint);
1454         signal(SIGALRM, handle_sigint);
1455
1456         if (net_mode == Net_client && net_setup_client())
1457                 return 1;
1458
1459         if (start_devices() != 0)
1460                 return 1;
1461
1462         atexit(stop_all_tracing);
1463
1464         if (stop_watch)
1465                 alarm(stop_watch);
1466
1467         wait_for_threads();
1468
1469         if (!is_trace_stopped()) {
1470                 trace_stopped = 1;
1471                 stop_all_threads();
1472                 stop_all_traces();
1473         }
1474
1475         show_stats();
1476
1477         return 0;
1478 }
1479