blktrace.c

   1 /*
   2  * block queue tracing application
   3  *
   4  * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; either version 2 of the License, or
   9  *  (at your option) any later version.
  10  *
  11  *  This program is distributed in the hope that it will be useful,
  12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  *  GNU General Public License for more details.
  15  *
  16  *  You should have received a copy of the GNU General Public License
  17  *  along with this program; if not, write to the Free Software
  18  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19  *
  20  */
  21 #include <pthread.h>
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <unistd.h>
  25 #include <locale.h>
  26 #include <signal.h>
  27 #include <fcntl.h>
  28 #include <string.h>
  29 #include <sys/ioctl.h>
  30 #include <sys/param.h>
  31 #include <sys/statfs.h>
  32 #include <sys/poll.h>
  33 #include <sys/mman.h>
  34 #include <sys/socket.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <sched.h>
  38 #include <ctype.h>
  39 #include <getopt.h>
  40 #include <errno.h>
  41 #include <netinet/in.h>
  42 #include <arpa/inet.h>
  43 #include <netdb.h>
  44 #include <sys/sendfile.h>
  45
  46 #include "blktrace.h"
  47 #include "barrier.h"
  48
  49 static char blktrace_version[] = "0.99";
  50
  51 /*
  52  * You may want to increase this even more, if you are logging at a high
  53  * rate and see skipped/missed events
  54  */
  55 #define BUF_SIZE        (512 * 1024)
  56 #define BUF_NR          (4)
  57
  58 #define OFILE_BUF       (128 * 1024)
  59
  60 #define RELAYFS_TYPE    0xF0B4A981
  61
  62 #define S_OPTS  "d:a:A:r:o:kw:Vb:n:D:lh:p:s"
  63 static struct option l_opts[] = {
  64         {
  65                 .name = "dev",
  66                 .has_arg = required_argument,
  67                 .flag = NULL,
  68                 .val = 'd'
  69         },
  70         {
  71                 .name = "act-mask",
  72                 .has_arg = required_argument,
  73                 .flag = NULL,
  74                 .val = 'a'
  75         },
  76         {
  77                 .name = "set-mask",
  78                 .has_arg = required_argument,
  79                 .flag = NULL,
  80                 .val = 'A'
  81         },
  82         {
  83                 .name = "relay",
  84                 .has_arg = required_argument,
  85                 .flag = NULL,
  86                 .val = 'r'
  87         },
  88         {
  89                 .name = "output",
  90                 .has_arg = required_argument,
  91                 .flag = NULL,
  92                 .val = 'o'
  93         },
  94         {
  95                 .name = "kill",
  96                 .has_arg = no_argument,
  97                 .flag = NULL,
  98                 .val = 'k'
  99         },
 100         {
 101                 .name = "stopwatch",
 102                 .has_arg = required_argument,
 103                 .flag = NULL,
 104                 .val = 'w'
 105         },
 106         {
 107                 .name = "version",
 108                 .has_arg = no_argument,
 109                 .flag = NULL,
 110                 .val = 'V'
 111         },
 112         {
 113                 .name = "buffer-size",
 114                 .has_arg = required_argument,
 115                 .flag = NULL,
 116                 .val = 'b'
 117         },
 118         {
 119                 .name = "num-sub-buffers",
 120                 .has_arg = required_argument,
 121                 .flag = NULL,
 122                 .val = 'n'
 123         },
 124         {
 125                 .name = "output-dir",
 126                 .has_arg = required_argument,
 127                 .flag = NULL,
 128                 .val = 'D'
 129         },
 130         {
 131                 .name = "listen",
 132                 .has_arg = no_argument,
 133                 .flag = NULL,
 134                 .val = 'l'
 135         },
 136         {
 137                 .name = "host",
 138                 .has_arg = required_argument,
 139                 .flag = NULL,
 140                 .val = 'h'
 141         },
 142         {
 143                 .name = "port",
 144                 .has_arg = required_argument,
 145                 .flag = NULL,
 146                 .val = 'p'
 147         },
 148         {
 149                 .name = "sendfile",
 150                 .has_arg = no_argument,
 151                 .flag = NULL,
 152                 .val = 's'
 153         },
 154         {
 155                 .name = NULL,
 156         }
 157 };
 158
 159 struct tip_subbuf {
 160         void *buf;
 161         unsigned int len;
 162         unsigned int max_len;
 163         off_t offset;
 164 };
 165
 166 #define FIFO_SIZE       (1024)  /* should be plenty big! */
 167 #define CL_SIZE         (128)   /* cache line, any bigger? */
 168
 169 struct tip_subbuf_fifo {
 170         int tail __attribute__((aligned(CL_SIZE)));
 171         int head __attribute__((aligned(CL_SIZE)));
 172         struct tip_subbuf *q[FIFO_SIZE];
 173 };
 174
 175 struct thread_information {
 176         int cpu;
 177         pthread_t thread;
 178
 179         int fd;
 180         void *fd_buf;
 181         char fn[MAXPATHLEN + 64];
 182
 183         FILE *ofile;
 184         char *ofile_buffer;
 185         off_t ofile_offset;
 186         int ofile_stdout;
 187         int ofile_mmap;
 188
 189         int (*get_subbuf)(struct thread_information *, unsigned int);
 190         int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *);
 191         int (*read_data)(struct thread_information *, void *, unsigned int);
 192
 193         unsigned long events_processed;
 194         unsigned long long data_read;
 195         struct device_information *device;
 196
 197         int exited;
 198
 199         /*
 200          * piped fifo buffers
 201          */
 202         struct tip_subbuf_fifo fifo;
 203         struct tip_subbuf *leftover_ts;
 204
 205         /*
 206          * mmap controlled output files
 207          */
 208         unsigned long long fs_size;
 209         unsigned long long fs_max_size;
 210         unsigned long fs_off;
 211         void *fs_buf;
 212         unsigned long fs_buf_len;
 213 };
 214
 215 struct device_information {
 216         int fd;
 217         char *path;
 218         char buts_name[32];
 219         volatile int trace_started;
 220         unsigned long drop_count;
 221         struct thread_information *threads;
 222 };
 223
 224 static int ncpus;
 225 static struct thread_information *thread_information;
 226 static int ndevs;
 227 static struct device_information *device_information;
 228
 229 /* command line option globals */
 230 static char *relay_path;
 231 static char *output_name;
 232 static char *output_dir;
 233 static int act_mask = ~0U;
 234 static int kill_running_trace;
 235 static unsigned long buf_size = BUF_SIZE;
 236 static unsigned long buf_nr = BUF_NR;
 237 static unsigned int page_size;
 238
 239 #define is_done()       (*(volatile int *)(&done))
 240 static volatile int done;
 241
 242 #define is_trace_stopped()      (*(volatile int *)(&trace_stopped))
 243 static volatile int trace_stopped;
 244
 245 #define is_stat_shown() (*(volatile int *)(&stat_shown))
 246 static volatile int stat_shown;
 247
 248 int data_is_native = -1;
 249
 250 static void exit_trace(int status);
 251
 252 #define dip_tracing(dip)        (*(volatile int *)(&(dip)->trace_started))
 253 #define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
 254
 255 #define __for_each_dip(__d, __i, __e)   \
 256         for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
 257
 258 #define for_each_dip(__d, __i)  __for_each_dip(__d, __i, ndevs)
 259 #define for_each_tip(__d, __t, __j)     \
 260         for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
 261
 262 /*
 263  * networking stuff follows. we include a magic number so we know whether
 264  * to endianness convert or not
 265  */
 266 struct blktrace_net_hdr {
 267         u32 magic;              /* same as trace magic */
 268         char buts_name[32];     /* trace name */
 269         u32 cpu;                /* for which cpu */
 270         u32 max_cpus;
 271         u32 len;                /* length of following trace data */
 272 };
 273
 274 #define TRACE_NET_PORT          (8462)
 275
 276 enum {
 277         Net_none = 0,
 278         Net_server,
 279         Net_client,
 280 };
 281
 282 /*
 283  * network cmd line params
 284  */
 285 static char hostname[MAXHOSTNAMELEN];
 286 static int net_port = TRACE_NET_PORT;
 287 static int net_mode = 0;
 288 static int net_sendfile;
 289
 290 static int net_in_fd = -1;
 291 static int net_out_fd = -1;
 292
 293 static void handle_sigint(__attribute__((__unused__)) int sig)
 294 {
 295         done = 1;
 296 }
 297
 298 static int get_dropped_count(const char *buts_name)
 299 {
 300         int fd;
 301         char tmp[MAXPATHLEN + 64];
 302
 303         snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
 304                  relay_path, buts_name);
 305
 306         fd = open(tmp, O_RDONLY);
 307         if (fd < 0) {
 308                 /*
 309                  * this may be ok, if the kernel doesn't support dropped counts
 310                  */
 311                 if (errno == ENOENT)
 312                         return 0;
 313
 314                 fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
 315                 return -1;
 316         }
 317
 318         if (read(fd, tmp, sizeof(tmp)) < 0) {
 319                 perror(tmp);
 320                 close(fd);
 321                 return -1;
 322         }
 323
 324         close(fd);
 325
 326         return atoi(tmp);
 327 }
 328
 329 static int start_trace(struct device_information *dip)
 330 {
 331         struct blk_user_trace_setup buts;
 332
 333         memset(&buts, 0, sizeof(buts));
 334         buts.buf_size = buf_size;
 335         buts.buf_nr = buf_nr;
 336         buts.act_mask = act_mask;
 337
 338         if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
 339                 perror("BLKTRACESETUP");
 340                 return 1;
 341         }
 342
 343         if (ioctl(dip->fd, BLKTRACESTART) < 0) {
 344                 perror("BLKTRACESTART");
 345                 return 1;
 346         }
 347
 348         memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
 349         dip_set_tracing(dip, 1);
 350         return 0;
 351 }
 352
 353 static void stop_trace(struct device_information *dip)
 354 {
 355         if (dip_tracing(dip) || kill_running_trace) {
 356                 dip_set_tracing(dip, 0);
 357
 358                 if (ioctl(dip->fd, BLKTRACESTOP) < 0)
 359                         perror("BLKTRACESTOP");
 360                 if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
 361                         perror("BLKTRACETEARDOWN");
 362
 363                 close(dip->fd);
 364                 dip->fd = -1;
 365         }
 366 }
 367
 368 static void stop_all_traces(void)
 369 {
 370         struct device_information *dip;
 371         int i;
 372
 373         for_each_dip(dip, i) {
 374                 dip->drop_count = get_dropped_count(dip->buts_name);
 375                 stop_trace(dip);
 376         }
 377 }
 378
 379 static void wait_for_data(struct thread_information *tip)
 380 {
 381         struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
 382
 383         do {
 384                 poll(&pfd, 1, 100);
 385                 if (pfd.revents & POLLIN)
 386                         break;
 387                 if (tip->ofile_stdout)
 388                         break;
 389         } while (!is_done());
 390 }
 391
 392 static int read_data_file(struct thread_information *tip, void *buf,
 393                           unsigned int len)
 394 {
 395         int ret = 0;
 396
 397         do {
 398                 wait_for_data(tip);
 399
 400                 ret = read(tip->fd, buf, len);
 401                 if (!ret)
 402                         continue;
 403                 else if (ret > 0)
 404                         return ret;
 405                 else {
 406                         if (errno != EAGAIN) {
 407                                 perror(tip->fn);
 408                                 fprintf(stderr,"Thread %d failed read of %s\n",
 409                                         tip->cpu, tip->fn);
 410                                 break;
 411                         }
 412                         continue;
 413                 }
 414         } while (!is_done());
 415
 416         return ret;
 417
 418 }
 419
 420 static int read_data_net(struct thread_information *tip, void *buf,
 421                          unsigned int len)
 422 {
 423         unsigned int bytes_left = len;
 424         int ret = 0;
 425
 426         do {
 427                 ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
 428
 429                 if (!ret)
 430                         continue;
 431                 else if (ret < 0) {
 432                         if (errno != EAGAIN) {
 433                                 perror(tip->fn);
 434                                 fprintf(stderr, "server: failed read\n");
 435                                 return 0;
 436                         }
 437                         continue;
 438                 } else {
 439                         buf += ret;
 440                         bytes_left -= ret;
 441                 }
 442         } while (!is_done() && bytes_left);
 443
 444         return len - bytes_left;
 445 }
 446
 447 static int read_data(struct thread_information *tip, void *buf,
 448                      unsigned int len)
 449 {
 450         int ret = tip->read_data(tip, buf, len);
 451
 452         if (ret > 0)
 453                 tip->data_read += ret;
 454
 455         return ret;
 456 }
 457
 458 static inline struct tip_subbuf *
 459 subbuf_fifo_dequeue(struct thread_information *tip)
 460 {
 461         const int head = tip->fifo.head;
 462         const int next = (head + 1) & (FIFO_SIZE - 1);
 463
 464         if (head != tip->fifo.tail) {
 465                 struct tip_subbuf *ts = tip->fifo.q[head];
 466
 467                 store_barrier();
 468                 tip->fifo.head = next;
 469                 return ts;
 470         }
 471
 472         return NULL;
 473 }
 474
 475 static inline int subbuf_fifo_queue(struct thread_information *tip,
 476                                     struct tip_subbuf *ts)
 477 {
 478         const int tail = tip->fifo.tail;
 479         const int next = (tail + 1) & (FIFO_SIZE - 1);
 480
 481         if (next != tip->fifo.head) {
 482                 tip->fifo.q[tail] = ts;
 483                 store_barrier();
 484                 tip->fifo.tail = next;
 485                 return 0;
 486         }
 487
 488         fprintf(stderr, "fifo too small!\n");
 489         return 1;
 490 }
 491
 492 /*
 493  * For file output, truncate and mmap the file appropriately
 494  */
 495 static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
 496 {
 497         int ofd = fileno(tip->ofile);
 498         int ret;
 499
 500         /*
 501          * extend file, if we have to. use chunks of 16 subbuffers.
 502          */
 503         if (tip->fs_off + buf_size > tip->fs_buf_len) {
 504                 if (tip->fs_buf) {
 505                         munlock(tip->fs_buf, tip->fs_buf_len);
 506                         munmap(tip->fs_buf, tip->fs_buf_len);
 507                         tip->fs_buf = NULL;
 508                 }
 509
 510                 tip->fs_off = tip->fs_size & (page_size - 1);
 511                 tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
 512                 tip->fs_max_size += tip->fs_buf_len;
 513
 514                 if (ftruncate(ofd, tip->fs_max_size) < 0) {
 515                         perror("ftruncate");
 516                         return -1;
 517                 }
 518
 519                 tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
 520                                    MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
 521                 if (tip->fs_buf == MAP_FAILED) {
 522                         perror("mmap");
 523                         return -1;
 524                 }
 525                 mlock(tip->fs_buf, tip->fs_buf_len);
 526         }
 527
 528         ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
 529         if (ret >= 0) {
 530                 tip->fs_size += ret;
 531                 tip->fs_off += ret;
 532                 return 0;
 533         }
 534
 535         return -1;
 536 }
 537
 538 static int get_subbuf_sendfile(struct thread_information *tip,
 539                                unsigned int maxlen)
 540 {
 541         struct tip_subbuf *ts = malloc(sizeof(*ts));
 542         struct stat sb;
 543
 544         ts->max_len = maxlen;
 545         ts->buf = NULL;
 546
 547         if (fstat(tip->fd, &sb) < 0) {
 548                 perror("trace stat");
 549                 return 1;
 550         }
 551
 552         ts->len = sb.st_size - tip->ofile_offset;
 553         ts->max_len = ts->len;
 554         ts->offset = tip->ofile_offset;
 555         tip->ofile_offset += ts->len;
 556         return subbuf_fifo_queue(tip, ts);
 557 }
 558
 559 /*
 560  * Use the copy approach for pipes and network
 561  */
 562 static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
 563 {
 564         struct tip_subbuf *ts = malloc(sizeof(*ts));
 565         int ret;
 566
 567         ts->buf = malloc(buf_size);
 568         ts->max_len = maxlen;
 569
 570         ret = read_data(tip, ts->buf, ts->max_len);
 571         if (ret > 0) {
 572                 ts->len = ret;
 573                 return subbuf_fifo_queue(tip, ts);
 574         }
 575
 576         return ret;
 577 }
 578
 579 static void close_thread(struct thread_information *tip)
 580 {
 581         if (tip->fd != -1)
 582                 close(tip->fd);
 583         if (tip->ofile)
 584                 fclose(tip->ofile);
 585         if (tip->ofile_buffer)
 586                 free(tip->ofile_buffer);
 587         if (tip->fd_buf)
 588                 free(tip->fd_buf);
 589
 590         tip->fd = -1;
 591         tip->ofile = NULL;
 592         tip->ofile_buffer = NULL;
 593         tip->fd_buf = NULL;
 594 }
 595
 596 static void tip_ftrunc_final(struct thread_information *tip)
 597 {
 598         /*
 599          * truncate to right size and cleanup mmap
 600          */
 601         if (tip->ofile_mmap) {
 602                 int ofd = fileno(tip->ofile);
 603
 604                 if (tip->fs_buf)
 605                         munmap(tip->fs_buf, tip->fs_buf_len);
 606
 607                 ftruncate(ofd, tip->fs_size);
 608         }
 609 }
 610
 611 static void *thread_main(void *arg)
 612 {
 613         struct thread_information *tip = arg;
 614         pid_t pid = getpid();
 615         cpu_set_t cpu_mask;
 616
 617         CPU_ZERO(&cpu_mask);
 618         CPU_SET((tip->cpu), &cpu_mask);
 619
 620         if (sched_setaffinity(pid, sizeof(cpu_mask), &cpu_mask) == -1) {
 621                 perror("sched_setaffinity");
 622                 exit_trace(1);
 623         }
 624
 625         snprintf(tip->fn, sizeof(tip->fn), "%s/block/%s/trace%d",
 626                         relay_path, tip->device->buts_name, tip->cpu);
 627         tip->fd = open(tip->fn, O_RDONLY);
 628         if (tip->fd < 0) {
 629                 perror(tip->fn);
 630                 fprintf(stderr,"Thread %d failed open of %s\n", tip->cpu,
 631                         tip->fn);
 632                 exit_trace(1);
 633         }
 634
 635         while (!is_done()) {
 636                 if (tip->get_subbuf(tip, buf_size))
 637                         break;
 638         }
 639
 640         tip_ftrunc_final(tip);
 641         tip->exited = 1;
 642         return NULL;
 643 }
 644
 645 static int write_data_net(int fd, void *buf, unsigned int buf_len)
 646 {
 647         unsigned int bytes_left = buf_len;
 648         int ret;
 649
 650         while (bytes_left) {
 651                 ret = send(fd, buf, bytes_left, 0);
 652                 if (ret < 0) {
 653                         perror("send");
 654                         return 1;
 655                 }
 656
 657                 buf += ret;
 658                 bytes_left -= ret;
 659         }
 660
 661         return 0;
 662 }
 663
 664 static int net_send_header(struct thread_information *tip, unsigned int len)
 665 {
 666         struct blktrace_net_hdr hdr;
 667
 668         hdr.magic = BLK_IO_TRACE_MAGIC;
 669         strcpy(hdr.buts_name, tip->device->buts_name);
 670         hdr.cpu = tip->cpu;
 671         hdr.max_cpus = ncpus;
 672         hdr.len = len;
 673
 674         return write_data_net(net_out_fd, &hdr, sizeof(hdr));
 675 }
 676
 677 /*
 678  * send header with 0 length to signal end-of-run
 679  */
 680 static void net_client_send_close(void)
 681 {
 682         struct blktrace_net_hdr hdr;
 683
 684         hdr.magic = BLK_IO_TRACE_MAGIC;
 685         hdr.cpu = 0;
 686         hdr.max_cpus = ncpus;
 687         hdr.len = 0;
 688
 689         write_data_net(net_out_fd, &hdr, sizeof(hdr));
 690 }
 691
 692 static int flush_subbuf_net(struct thread_information *tip,
 693                             struct tip_subbuf *ts)
 694 {
 695         if (net_send_header(tip, ts->len))
 696                 return 1;
 697         if (write_data_net(net_out_fd, ts->buf, ts->len))
 698                 return 1;
 699
 700         free(ts->buf);
 701         free(ts);
 702         return 0;
 703 }
 704
 705 static int flush_subbuf_sendfile(struct thread_information *tip,
 706                                  struct tip_subbuf *ts)
 707 {
 708         if (net_send_header(tip, ts->len))
 709                 return 1;
 710         if (sendfile(net_out_fd, tip->fd, &ts->offset, ts->len) < 0) {
 711                 perror("sendfile");
 712                 return 1;
 713         }
 714
 715         free(ts);
 716         return 0;
 717 }
 718
 719 static int write_data(struct thread_information *tip, void *buf,
 720                       unsigned int buf_len)
 721 {
 722         int ret;
 723
 724         if (!buf_len)
 725                 return 0;
 726
 727         while (1) {
 728                 ret = fwrite(buf, buf_len, 1, tip->ofile);
 729                 if (ret == 1)
 730                         break;
 731
 732                 if (ret < 0) {
 733                         perror("write");
 734                         return 1;
 735                 }
 736         }
 737
 738         if (tip->ofile_stdout)
 739                 fflush(tip->ofile);
 740
 741         return 0;
 742 }
 743
 744 static int flush_subbuf_file(struct thread_information *tip,
 745                              struct tip_subbuf *ts)
 746 {
 747         unsigned int offset = 0;
 748         struct blk_io_trace *t;
 749         int pdu_len, events = 0;
 750
 751         /*
 752          * surplus from last run
 753          */
 754         if (tip->leftover_ts) {
 755                 struct tip_subbuf *prev_ts = tip->leftover_ts;
 756
 757                 if (prev_ts->len + ts->len > prev_ts->max_len) {
 758                         prev_ts->max_len += ts->len;
 759                         prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
 760                 }
 761
 762                 memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
 763                 prev_ts->len += ts->len;
 764
 765                 free(ts->buf);
 766                 free(ts);
 767
 768                 ts = prev_ts;
 769                 tip->leftover_ts = NULL;
 770         }
 771
 772         while (offset + sizeof(*t) <= ts->len) {
 773                 t = ts->buf + offset;
 774
 775                 if (verify_trace(t)) {
 776                         write_data(tip, ts->buf, offset);
 777                         return -1;
 778                 }
 779
 780                 pdu_len = t->pdu_len;
 781
 782                 if (offset + sizeof(*t) + pdu_len > ts->len)
 783                         break;
 784
 785                 offset += sizeof(*t) + pdu_len;
 786                 tip->events_processed++;
 787                 tip->data_read += sizeof(*t) + pdu_len;
 788                 events++;
 789         }
 790
 791         if (write_data(tip, ts->buf, offset))
 792                 return -1;
 793
 794         /*
 795          * leftover bytes, save them for next time
 796          */
 797         if (offset != ts->len) {
 798                 tip->leftover_ts = ts;
 799                 ts->len -= offset;
 800                 memmove(ts->buf, ts->buf + offset, ts->len);
 801         } else {
 802                 free(ts->buf);
 803                 free(ts);
 804         }
 805
 806         return events;
 807 }
 808
 809 static int write_tip_events(struct thread_information *tip)
 810 {
 811         struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
 812
 813         if (ts)
 814                 return tip->flush_subbuf(tip, ts);
 815
 816         return 0;
 817 }
 818
 819 /*
 820  * scans the tips we know and writes out the subbuffers we accumulate
 821  */
 822 static void get_and_write_events(void)
 823 {
 824         struct device_information *dip;
 825         struct thread_information *tip;
 826         int i, j, events, ret, tips_running;
 827
 828         while (!is_done()) {
 829                 events = 0;
 830
 831                 for_each_dip(dip, i) {
 832                         for_each_tip(dip, tip, j) {
 833                                 ret = write_tip_events(tip);
 834                                 if (ret > 0)
 835                                         events += ret;
 836                         }
 837                 }
 838
 839                 if (!events)
 840                         usleep(10);
 841         }
 842
 843         /*
 844          * reap stored events
 845          */
 846         do {
 847                 events = 0;
 848                 tips_running = 0;
 849                 for_each_dip(dip, i) {
 850                         for_each_tip(dip, tip, j) {
 851                                 ret = write_tip_events(tip);
 852                                 if (ret > 0)
 853                                         events += ret;
 854                                 tips_running += !tip->exited;
 855                         }
 856                 }
 857                 usleep(10);
 858         } while (events || tips_running);
 859 }
 860
 861 static void wait_for_threads(void)
 862 {
 863         /*
 864          * for piped or network output, poll and fetch data for writeout.
 865          * for files, we just wait around for trace threads to exit
 866          */
 867         if ((output_name && !strcmp(output_name, "-")) ||
 868             net_mode == Net_client)
 869                 get_and_write_events();
 870         else {
 871                 struct device_information *dip;
 872                 struct thread_information *tip;
 873                 int i, j, tips_running;
 874
 875                 do {
 876                         tips_running = 0;
 877                         usleep(1000);
 878
 879                         for_each_dip(dip, i)
 880                                 for_each_tip(dip, tip, j)
 881                                         tips_running += !tip->exited;
 882                 } while (tips_running);
 883         }
 884
 885         if (net_mode == Net_client)
 886                 net_client_send_close();
 887 }
 888
 889 static void fill_ofname(char *dst, char *buts_name, int cpu)
 890 {
 891         int len = 0;
 892
 893         if (output_dir)
 894                 len = sprintf(dst, "%s/", output_dir);
 895
 896         if (output_name)
 897                 sprintf(dst + len, "%s.blktrace.%d", output_name, cpu);
 898         else
 899                 sprintf(dst + len, "%s.blktrace.%d", buts_name, cpu);
 900 }
 901
 902 static void fill_ops(struct thread_information *tip)
 903 {
 904         /*
 905          * setup ops
 906          */
 907         if (net_mode == Net_client) {
 908                 if (net_sendfile) {
 909                         tip->get_subbuf = get_subbuf_sendfile;
 910                         tip->flush_subbuf = flush_subbuf_sendfile;
 911                 } else {
 912                         tip->get_subbuf = get_subbuf;
 913                         tip->flush_subbuf = flush_subbuf_net;
 914                 }
 915         } else {
 916                 if (tip->ofile_mmap)
 917                         tip->get_subbuf = mmap_subbuf;
 918                 else
 919                         tip->get_subbuf = get_subbuf;
 920
 921                 tip->flush_subbuf = flush_subbuf_file;
 922         }
 923
 924         if (net_mode == Net_server)
 925                 tip->read_data = read_data_net;
 926         else
 927                 tip->read_data = read_data_file;
 928 }
 929
 930 static int tip_open_output(struct device_information *dip,
 931                            struct thread_information *tip)
 932 {
 933         int pipeline = output_name && !strcmp(output_name, "-");
 934         int mode, vbuf_size;
 935         char op[64];
 936
 937         if (net_mode == Net_client) {
 938                 tip->ofile = NULL;
 939                 tip->ofile_stdout = 0;
 940                 tip->ofile_mmap = 0;
 941                 vbuf_size = 0;
 942         } else if (pipeline) {
 943                 tip->ofile = fdopen(STDOUT_FILENO, "w");
 944                 tip->ofile_stdout = 1;
 945                 tip->ofile_mmap = 0;
 946                 mode = _IOLBF;
 947                 vbuf_size = 512;
 948         } else {
 949                 fill_ofname(op, dip->buts_name, tip->cpu);
 950                 tip->ofile = fopen(op, "w+");
 951                 tip->ofile_stdout = 0;
 952                 tip->ofile_mmap = 1;
 953                 mode = _IOFBF;
 954                 vbuf_size = OFILE_BUF;
 955         }
 956
 957         if (net_mode != Net_client && tip->ofile == NULL) {
 958                 perror(op);
 959                 return 1;
 960         }
 961
 962         if (vbuf_size) {
 963                 tip->ofile_buffer = malloc(vbuf_size);
 964                 if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
 965                         perror("setvbuf");
 966                         close_thread(tip);
 967                         return 1;
 968                 }
 969         }
 970
 971         fill_ops(tip);
 972         return 0;
 973 }
 974
 975 static int start_threads(struct device_information *dip)
 976 {
 977         struct thread_information *tip;
 978         int j;
 979
 980         for_each_tip(dip, tip, j) {
 981                 tip->cpu = j;
 982                 tip->device = dip;
 983                 tip->events_processed = 0;
 984                 memset(&tip->fifo, 0, sizeof(tip->fifo));
 985                 tip->leftover_ts = NULL;
 986
 987                 if (tip_open_output(dip, tip))
 988                         return 1;
 989
 990                 if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
 991                         perror("pthread_create");
 992                         close_thread(tip);
 993                         return 1;
 994                 }
 995         }
 996
 997         return 0;
 998 }
 999
1000 static void stop_threads(struct device_information *dip)
1001 {
1002         struct thread_information *tip;
1003         unsigned long ret;
1004         int i;
1005
1006         for_each_tip(dip, tip, i) {
1007                 (void) pthread_join(tip->thread, (void *) &ret);
1008                 close_thread(tip);
1009         }
1010 }
1011
1012 static void stop_all_threads(void)
1013 {
1014         struct device_information *dip;
1015         int i;
1016
1017         for_each_dip(dip, i)
1018                 stop_threads(dip);
1019 }
1020
1021 static void stop_all_tracing(void)
1022 {
1023         struct device_information *dip;
1024         int i;
1025
1026         for_each_dip(dip, i)
1027                 stop_trace(dip);
1028 }
1029
1030 static void exit_trace(int status)
1031 {
1032         if (!is_trace_stopped()) {
1033                 trace_stopped = 1;
1034                 stop_all_threads();
1035                 stop_all_tracing();
1036         }
1037
1038         exit(status);
1039 }
1040
1041 static int resize_devices(char *path)
1042 {
1043         int size = (ndevs + 1) * sizeof(struct device_information);
1044
1045         device_information = realloc(device_information, size);
1046         if (!device_information) {
1047                 fprintf(stderr, "Out of memory, device %s (%d)\n", path, size);
1048                 return 1;
1049         }
1050         device_information[ndevs].path = path;
1051         ndevs++;
1052         return 0;
1053 }
1054
1055 static int open_devices(void)
1056 {
1057         struct device_information *dip;
1058         int i;
1059
1060         for_each_dip(dip, i) {
1061                 dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
1062                 if (dip->fd < 0) {
1063                         perror(dip->path);
1064                         return 1;
1065                 }
1066         }
1067
1068         return 0;
1069 }
1070
1071 static int start_devices(void)
1072 {
1073         struct device_information *dip;
1074         int i, j, size;
1075
1076         size = ncpus * sizeof(struct thread_information);
1077         thread_information = malloc(size * ndevs);
1078         if (!thread_information) {
1079                 fprintf(stderr, "Out of memory, threads (%d)\n", size * ndevs);
1080                 return 1;
1081         }
1082
1083         for_each_dip(dip, i) {
1084                 if (start_trace(dip)) {
1085                         close(dip->fd);
1086                         fprintf(stderr, "Failed to start trace on %s\n",
1087                                 dip->path);
1088                         break;
1089                 }
1090         }
1091
1092         if (i != ndevs) {
1093                 __for_each_dip(dip, j, i)
1094                         stop_trace(dip);
1095
1096                 return 1;
1097         }
1098
1099         for_each_dip(dip, i) {
1100                 dip->threads = thread_information + (i * ncpus);
1101                 if (start_threads(dip)) {
1102                         fprintf(stderr, "Failed to start worker threads\n");
1103                         break;
1104                 }
1105         }
1106
1107         if (i != ndevs) {
1108                 __for_each_dip(dip, j, i)
1109                         stop_threads(dip);
1110                 for_each_dip(dip, i)
1111                         stop_trace(dip);
1112
1113                 return 1;
1114         }
1115
1116         return 0;
1117 }
1118
1119 static void show_stats(void)
1120 {
1121         struct device_information *dip;
1122         struct thread_information *tip;
1123         unsigned long long events_processed, data_read;
1124         unsigned long total_drops;
1125         int i, j, no_stdout = 0;
1126
1127         if (is_stat_shown())
1128                 return;
1129
1130         if (output_name && !strcmp(output_name, "-"))
1131                 no_stdout = 1;
1132
1133         stat_shown = 1;
1134
1135         total_drops = 0;
1136         for_each_dip(dip, i) {
1137                 if (!no_stdout)
1138                         printf("Device: %s\n", dip->path);
1139                 events_processed = 0;
1140                 data_read = 0;
1141                 for_each_tip(dip, tip, j) {
1142                         if (!no_stdout)
1143                                 printf("  CPU%3d: %20lu events, %8llu KiB data\n",
1144                                         tip->cpu, tip->events_processed,
1145                                         (tip->data_read + 1023) >> 10);
1146                         events_processed += tip->events_processed;
1147                         data_read += tip->data_read;
1148                 }
1149                 total_drops += dip->drop_count;
1150                 if (!no_stdout)
1151                         printf("  Total:  %20llu events (dropped %lu), %8llu KiB data\n",
1152                                         events_processed, dip->drop_count,
1153                                         (data_read + 1023) >> 10);
1154         }
1155
1156         if (total_drops)
1157                 fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
1158 }
1159
1160 static struct device_information *net_get_dip(char *buts_name)
1161 {
1162         struct device_information *dip;
1163         int i;
1164
1165         for (i = 0; i < ndevs; i++) {
1166                 dip = &device_information[i];
1167
1168                 if (!strcmp(dip->buts_name, buts_name))
1169                         return dip;
1170         }
1171
1172         device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
1173         dip = &device_information[ndevs];
1174         strcpy(dip->buts_name, buts_name);
1175         strcpy(dip->path, buts_name);
1176         ndevs++;
1177         dip->threads = malloc(ncpus * sizeof(struct thread_information));
1178         memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
1179
1180         /*
1181          * open all files
1182          */
1183         for (i = 0; i < ncpus; i++) {
1184                 struct thread_information *tip = &dip->threads[i];
1185
1186                 tip->cpu = i;
1187                 tip->device = dip;
1188
1189                 if (tip_open_output(dip, tip))
1190                         return NULL;
1191         }
1192
1193         return dip;
1194 }
1195
1196 static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh)
1197 {
1198         struct device_information *dip;
1199
1200         ncpus = bnh->max_cpus;
1201         dip = net_get_dip(bnh->buts_name);
1202         return &dip->threads[bnh->cpu];
1203 }
1204
1205 static int net_get_header(struct blktrace_net_hdr *bnh)
1206 {
1207         int fl = fcntl(net_in_fd, F_GETFL);
1208         int bytes_left, ret;
1209         void *p = bnh;
1210
1211         fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
1212         bytes_left = sizeof(*bnh);
1213         while (bytes_left && !is_done()) {
1214                 ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
1215                 if (ret < 0) {
1216                         if (errno != EAGAIN) {
1217                                 perror("recv header");
1218                                 return 1;
1219                         }
1220                         usleep(100);
1221                         continue;
1222                 } else if (!ret) {
1223                         usleep(100);
1224                         continue;
1225                 } else {
1226                         p += ret;
1227                         bytes_left -= ret;
1228                 }
1229         }
1230         fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
1231         return 0;
1232 }
1233
1234 static int net_server_loop(void)
1235 {
1236         struct thread_information *tip;
1237         struct blktrace_net_hdr bnh;
1238
1239         if (net_get_header(&bnh))
1240                 return 1;
1241
1242         if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
1243                 fprintf(stderr, "server: received data is bad\n");
1244                 return 1;
1245         }
1246
1247         if (!data_is_native) {
1248                 bnh.cpu = be32_to_cpu(bnh.cpu);
1249                 bnh.len = be32_to_cpu(bnh.len);
1250         }
1251
1252         /*
1253          * len == 0 means that the other end signalled end-of-run
1254          */
1255         if (!bnh.len) {
1256                 fprintf(stderr, "server: end of run\n");
1257                 return 1;
1258         }
1259
1260         tip = net_get_tip(&bnh);
1261         if (!tip)
1262                 return 1;
1263
1264         if (mmap_subbuf(tip, bnh.len))
1265                 return 1;
1266
1267         return 0;
1268 }
1269
1270 /*
1271  * Start here when we are in server mode - just fetch data from the network
1272  * and dump to files
1273  */
1274 static int net_server(void)
1275 {
1276         struct device_information *dip;
1277         struct thread_information *tip;
1278         struct sockaddr_in addr;
1279         socklen_t socklen;
1280         int fd, opt, i, j;
1281
1282         fd = socket(AF_INET, SOCK_STREAM, 0);
1283         if (fd < 0) {
1284                 perror("server: socket");
1285                 return 1;
1286         }
1287
1288         opt = 1;
1289         if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
1290                 perror("setsockopt");
1291                 return 1;
1292         }
1293
1294         memset(&addr, 0, sizeof(addr));
1295         addr.sin_family = AF_INET;
1296         addr.sin_addr.s_addr = htonl(INADDR_ANY);
1297         addr.sin_port = htons(net_port);
1298
1299         if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1300                 perror("bind");
1301                 return 1;
1302         }
1303
1304         if (listen(fd, 1) < 0) {
1305                 perror("listen");
1306                 return 1;
1307         }
1308
1309 repeat:
1310         signal(SIGINT, NULL);
1311         signal(SIGHUP, NULL);
1312         signal(SIGTERM, NULL);
1313         signal(SIGALRM, NULL);
1314
1315         printf("blktrace: waiting for incoming connection...\n");
1316
1317         socklen = sizeof(addr);
1318         net_in_fd = accept(fd, (struct sockaddr *) &addr, &socklen);
1319         if (net_in_fd < 0) {
1320                 perror("accept");
1321                 return 1;
1322         }
1323
1324         signal(SIGINT, handle_sigint);
1325         signal(SIGHUP, handle_sigint);
1326         signal(SIGTERM, handle_sigint);
1327         signal(SIGALRM, handle_sigint);
1328
1329         printf("blktrace: connection from %s\n", inet_ntoa(addr.sin_addr));
1330
1331         while (!is_done()) {
1332                 if (net_server_loop())
1333                         break;
1334         }
1335
1336         for_each_dip(dip, i)
1337                 for_each_tip(dip, tip, j)
1338                         tip_ftrunc_final(tip);
1339
1340         show_stats();
1341
1342         if (is_done())
1343                 return 0;
1344
1345         /*
1346          * cleanup for next run
1347          */
1348         for_each_dip(dip, i) {
1349                 for_each_tip(dip, tip, j)
1350                         fclose(tip->ofile);
1351
1352                 free(dip->threads);
1353         }
1354
1355         free(device_information);
1356         device_information = NULL;
1357         ncpus = ndevs = 0;
1358
1359         close(net_in_fd);
1360         net_in_fd = -1;
1361         goto repeat;
1362 }
1363
1364 /*
1365  * Setup outgoing network connection where we will transmit data
1366  */
1367 static int net_setup_client(void)
1368 {
1369         struct sockaddr_in addr;
1370         int fd;
1371
1372         fd = socket(AF_INET, SOCK_STREAM, 0);
1373         if (fd < 0) {
1374                 perror("client: socket");
1375                 return 1;
1376         }
1377
1378         memset(&addr, 0, sizeof(addr));
1379         addr.sin_family = AF_INET;
1380         addr.sin_port = htons(net_port);
1381
1382         if (inet_aton(hostname, &addr.sin_addr) != 1) {
1383                 struct hostent *hent = gethostbyname(hostname);
1384                 if (!hent) {
1385                         perror("gethostbyname");
1386                         return 1;
1387                 }
1388
1389                 memcpy(&addr.sin_addr, hent->h_addr, 4);
1390                 strcpy(hostname, hent->h_name);
1391         }
1392
1393         printf("blktrace: connecting to %s\n", hostname);
1394
1395         if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
1396                 perror("client: connect");
1397                 return 1;
1398         }
1399
1400         printf("blktrace: connected!\n");
1401         net_out_fd = fd;
1402         return 0;
1403 }
1404
1405 static char usage_str[] = \
1406         "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
1407         "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
1408         "\t-d Use specified device. May also be given last after options\n" \
1409         "\t-r Path to mounted relayfs, defaults to /relay\n" \
1410         "\t-o File(s) to send output to\n" \
1411         "\t-D Directory to prepend to output file names\n" \
1412         "\t-k Kill a running trace\n" \
1413         "\t-w Stop after defined time, in seconds\n" \
1414         "\t-a Only trace specified actions. See documentation\n" \
1415         "\t-A Give trace mask as a single value. See documentation\n" \
1416         "\t-b Sub buffer size in KiB\n" \
1417         "\t-n Number of sub buffers\n" \
1418         "\t-l Run in network listen mode (blktrace server)\n" \
1419         "\t-h Run in network client mode, connecting to the given host\n" \
1420         "\t-p Network port to use (default 8462)\n" \
1421         "\t-s Make the network client use sendfile() to transfer data\n" \
1422         "\t-V Print program version info\n\n";
1423
1424 static void show_usage(char *program)
1425 {
1426         fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
1427 }
1428
1429 int main(int argc, char *argv[])
1430 {
1431         static char default_relay_path[] = "/relay";
1432         struct statfs st;
1433         int i, c;
1434         int stop_watch = 0;
1435         int act_mask_tmp = 0;
1436
1437         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) >= 0) {
1438                 switch (c) {
1439                 case 'a':
1440                         i = find_mask_map(optarg);
1441                         if (i < 0) {
1442                                 fprintf(stderr,"Invalid action mask %s\n",
1443                                         optarg);
1444                                 return 1;
1445                         }
1446                         act_mask_tmp |= i;
1447                         break;
1448
1449                 case 'A':
1450                         if ((sscanf(optarg, "%x", &i) != 1) ||
1451                                                         !valid_act_opt(i)) {
1452                                 fprintf(stderr,
1453                                         "Invalid set action mask %s/0x%x\n",
1454                                         optarg, i);
1455                                 return 1;
1456                         }
1457                         act_mask_tmp = i;
1458                         break;
1459
1460                 case 'd':
1461                         if (resize_devices(optarg) != 0)
1462                                 return 1;
1463                         break;
1464
1465                 case 'r':
1466                         relay_path = optarg;
1467                         break;
1468
1469                 case 'o':
1470                         output_name = optarg;
1471                         break;
1472                 case 'k':
1473                         kill_running_trace = 1;
1474                         break;
1475                 case 'w':
1476                         stop_watch = atoi(optarg);
1477                         if (stop_watch <= 0) {
1478                                 fprintf(stderr,
1479                                         "Invalid stopwatch value (%d secs)\n",
1480                                         stop_watch);
1481                                 return 1;
1482                         }
1483                         break;
1484                 case 'V':
1485                         printf("%s version %s\n", argv[0], blktrace_version);
1486                         return 0;
1487                 case 'b':
1488                         buf_size = strtoul(optarg, NULL, 10);
1489                         if (buf_size <= 0 || buf_size > 16*1024) {
1490                                 fprintf(stderr,
1491                                         "Invalid buffer size (%lu)\n",buf_size);
1492                                 return 1;
1493                         }
1494                         buf_size <<= 10;
1495                         break;
1496                 case 'n':
1497                         buf_nr = strtoul(optarg, NULL, 10);
1498                         if (buf_nr <= 0) {
1499                                 fprintf(stderr,
1500                                         "Invalid buffer nr (%lu)\n", buf_nr);
1501                                 return 1;
1502                         }
1503                         break;
1504                 case 'D':
1505                         output_dir = optarg;
1506                         break;
1507                 case 'h':
1508                         net_mode = Net_client;
1509                         strcpy(hostname, optarg);
1510                         break;
1511                 case 'l':
1512                         net_mode = Net_server;
1513                         break;
1514                 case 'p':
1515                         net_port = atoi(optarg);
1516                         break;
1517                 case 's':
1518                         net_sendfile = 1;
1519                         break;
1520                 default:
1521                         show_usage(argv[0]);
1522                         return 1;
1523                 }
1524         }
1525
1526         setlocale(LC_NUMERIC, "en_US");
1527
1528         page_size = getpagesize();
1529
1530         if (net_mode == Net_server)
1531                 return net_server();
1532
1533         while (optind < argc) {
1534                 if (resize_devices(argv[optind++]) != 0)
1535                         return 1;
1536         }
1537
1538         if (ndevs == 0) {
1539                 show_usage(argv[0]);
1540                 return 1;
1541         }
1542
1543         if (!relay_path)
1544                 relay_path = default_relay_path;
1545
1546         if (act_mask_tmp != 0)
1547                 act_mask = act_mask_tmp;
1548
1549         if (statfs(relay_path, &st) < 0) {
1550                 perror("statfs");
1551                 fprintf(stderr,"%s does not appear to be a valid path\n",
1552                         relay_path);
1553                 return 1;
1554         } else if (st.f_type != (long) RELAYFS_TYPE) {
1555                 fprintf(stderr,"%s does not appear to be a relay filesystem\n",
1556                         relay_path);
1557                 return 1;
1558         }
1559
1560         if (open_devices() != 0)
1561                 return 1;
1562
1563         if (kill_running_trace) {
1564                 stop_all_traces();
1565                 return 0;
1566         }
1567
1568         ncpus = sysconf(_SC_NPROCESSORS_ONLN);
1569         if (ncpus < 0) {
1570                 fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
1571                 return 1;
1572         }
1573
1574         signal(SIGINT, handle_sigint);
1575         signal(SIGHUP, handle_sigint);
1576         signal(SIGTERM, handle_sigint);
1577         signal(SIGALRM, handle_sigint);
1578
1579         if (net_mode == Net_client && net_setup_client())
1580                 return 1;
1581
1582         if (start_devices() != 0)
1583                 return 1;
1584
1585         atexit(stop_all_tracing);
1586
1587         if (stop_watch)
1588                 alarm(stop_watch);
1589
1590         wait_for_threads();
1591
1592         if (!is_trace_stopped()) {
1593                 trace_stopped = 1;
1594                 stop_all_threads();
1595                 stop_all_traces();
1596         }
1597
1598         show_stats();
1599
1600         return 0;
1601 }
1602