lttd/lttd.c

   1 /* lttd
   2  *
   3  * Linux Trace Toolkit Daemon
   4  *
   5  * This is a simple daemon that reads a few relay+debugfs channels and save
   6  * them in a trace.
   7  *
   8  * CPU hot-plugging is supported using inotify.
   9  *
  10  * Copyright 2005 -
  11  *      Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
  12  */
  13
  14 #ifdef HAVE_CONFIG_H
  15 #include <config.h>
  16 #endif
  17
  18 #define _REENTRANT
  19 #define _GNU_SOURCE
  20 #include <features.h>
  21 #include <stdio.h>
  22 #include <unistd.h>
  23 #include <errno.h>
  24 #include <sys/types.h>
  25 #include <sys/stat.h>
  26 #include <stdlib.h>
  27 #include <dirent.h>
  28 #include <string.h>
  29 #include <fcntl.h>
  30 #include <sys/poll.h>
  31 #include <sys/mman.h>
  32 #include <signal.h>
  33 #include <pthread.h>
  34 #include <sys/syscall.h>
  35 #include <unistd.h>
  36 #include <asm/ioctls.h>
  37
  38 #include <linux/version.h>
  39
  40 /* Relayfs IOCTL */
  41 #include <asm/ioctl.h>
  42 #include <asm/types.h>
  43
  44 /* Get the next sub buffer that can be read. */
  45 #define RELAY_GET_SB            _IOR(0xF5, 0x00,__u32)
  46 /* Release the oldest reserved (by "get") sub buffer. */
  47 #define RELAY_PUT_SB            _IOW(0xF5, 0x01,__u32)
  48 /* returns the number of sub buffers in the per cpu channel. */
  49 #define RELAY_GET_N_SB          _IOR(0xF5, 0x02,__u32)
  50 /* returns the size of the current sub buffer. */
  51 #define RELAY_GET_SB_SIZE       _IOR(0xF5, 0x03, __u32)
  52 /* returns the size of data to consume in the current sub-buffer. */
  53 #define RELAY_GET_MAX_SB_SIZE   _IOR(0xF5, 0x04, __u32)
  54
  55
  56 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
  57 #include <sys/inotify.h>
  58 #if 0   /* should now be provided by libc. */
  59 /* From the inotify-tools 2.6 package */
  60 static inline int inotify_init (void)
  61 {
  62         return syscall (__NR_inotify_init);
  63 }
  64
  65 static inline int inotify_add_watch (int fd, const char *name, __u32 mask)
  66 {
  67         return syscall (__NR_inotify_add_watch, fd, name, mask);
  68 }
  69
  70 static inline int inotify_rm_watch (int fd, __u32 wd)
  71 {
  72         return syscall (__NR_inotify_rm_watch, fd, wd);
  73 }
  74 #endif //0
  75 #define HAS_INOTIFY
  76 #else
  77 static inline int inotify_init (void)
  78 {
  79         return -1;
  80 }
  81
  82 static inline int inotify_add_watch (int fd, const char *name, __u32 mask)
  83 {
  84         return 0;
  85 }
  86
  87 static inline int inotify_rm_watch (int fd, __u32 wd)
  88 {
  89         return 0;
  90 }
  91 #undef HAS_INOTIFY
  92 #endif
  93
  94 struct fd_pair {
  95         int channel;
  96         int trace;
  97         unsigned int n_sb;
  98         unsigned int max_sb_size;
  99         void *mmap;
 100         pthread_mutex_t mutex;
 101 };
 102
 103 struct channel_trace_fd {
 104         struct fd_pair *pair;
 105         int num_pairs;
 106 };
 107
 108 struct inotify_watch {
 109         int wd;
 110         char path_channel[PATH_MAX];
 111         char path_trace[PATH_MAX];
 112 };
 113
 114 struct inotify_watch_array {
 115         struct inotify_watch *elem;
 116         int num;
 117 };
 118
 119 static __thread int thread_pipe[2];
 120
 121 struct channel_trace_fd fd_pairs = { NULL, 0 };
 122 int inotify_fd = -1;
 123 struct inotify_watch_array inotify_watch_array = { NULL, 0 };
 124
 125 /* protects fd_pairs and inotify_watch_array */
 126 pthread_rwlock_t fd_pairs_lock = PTHREAD_RWLOCK_INITIALIZER;
 127
 128
 129 static char             *trace_name = NULL;
 130 static char             *channel_name = NULL;
 131 static int              daemon_mode = 0;
 132 static int              append_mode = 0;
 133 static unsigned long    num_threads = 1;
 134 volatile static int     quit_program = 0;       /* For signal handler */
 135 static int              dump_flight_only = 0;
 136 static int              dump_normal_only = 0;
 137 static int              verbose_mode = 0;
 138
 139 #define printf_verbose(fmt, args...) \
 140   do {                               \
 141     if (verbose_mode)                \
 142       printf(fmt, ##args);           \
 143   } while (0)
 144
 145 /* Args :
 146  *
 147  * -t directory         Directory name of the trace to write to. Will be created.
 148  * -c directory         Root directory of the debugfs trace channels.
 149  * -d                   Run in background (daemon).
 150  * -a                   Trace append mode.
 151  * -s                   Send SIGUSR1 to parent when ready for IO.
 152  */
 153 void show_arguments(void)
 154 {
 155         printf("Please use the following arguments :\n");
 156         printf("\n");
 157         printf("-t directory  Directory name of the trace to write to.\n"
 158                                  "              It will be created.\n");
 159         printf("-c directory  Root directory of the debugfs trace channels.\n");
 160         printf("-d            Run in background (daemon).\n");
 161         printf("-a            Append to an possibly existing trace.\n");
 162         printf("-N            Number of threads to start.\n");
 163         printf("-f            Dump only flight recorder channels.\n");
 164         printf("-n            Dump only normal channels.\n");
 165         printf("-v            Verbose mode.\n");
 166         printf("\n");
 167 }
 168
 169
 170 /* parse_arguments
 171  *
 172  * Parses the command line arguments.
 173  *
 174  * Returns 1 if the arguments were correct, but doesn't ask for program
 175  * continuation. Returns -1 if the arguments are incorrect, or 0 if OK.
 176  */
 177 int parse_arguments(int argc, char **argv)
 178 {
 179         int ret = 0;
 180         int argn = 1;
 181
 182         if(argc == 2) {
 183                 if(strcmp(argv[1], "-h") == 0) {
 184                         return 1;
 185                 }
 186         }
 187
 188         while(argn < argc) {
 189
 190                 switch(argv[argn][0]) {
 191                         case '-':
 192                                 switch(argv[argn][1]) {
 193                                         case 't':
 194                                                 if(argn+1 < argc) {
 195                                                         trace_name = argv[argn+1];
 196                                                         argn++;
 197                                                 }
 198                                                 break;
 199                                         case 'c':
 200                                                 if(argn+1 < argc) {
 201                                                         channel_name = argv[argn+1];
 202                                                         argn++;
 203                                                 }
 204                                                 break;
 205                                         case 'd':
 206                                                 daemon_mode = 1;
 207                                                 break;
 208                                         case 'a':
 209                                                 append_mode = 1;
 210                                                 break;
 211                                         case 'N':
 212                                                 if(argn+1 < argc) {
 213                                                         num_threads = strtoul(argv[argn+1], NULL, 0);
 214                                                         argn++;
 215                                                 }
 216                                                 break;
 217                                         case 'f':
 218                                                 dump_flight_only = 1;
 219                                                 break;
 220                                         case 'n':
 221                                                 dump_normal_only = 1;
 222                                                 break;
 223                                         case 'v':
 224                                                 verbose_mode = 1;
 225                                                 break;
 226                                         default:
 227                                                 printf("Invalid argument '%s'.\n", argv[argn]);
 228                                                 printf("\n");
 229                                                 ret = -1;
 230                                 }
 231                                 break;
 232                         default:
 233                                 printf("Invalid argument '%s'.\n", argv[argn]);
 234                                 printf("\n");
 235                                 ret = -1;
 236                 }
 237                 argn++;
 238         }
 239
 240         if(trace_name == NULL) {
 241                 printf("Please specify a trace name.\n");
 242                 printf("\n");
 243                 ret = -1;
 244         }
 245
 246         if(channel_name == NULL) {
 247                 printf("Please specify a channel name.\n");
 248                 printf("\n");
 249                 ret = -1;
 250         }
 251
 252         return ret;
 253 }
 254
 255 void show_info(void)
 256 {
 257         printf("Linux Trace Toolkit Trace Daemon " VERSION "\n");
 258         printf("\n");
 259         printf("Reading from debugfs directory : %s\n", channel_name);
 260         printf("Writing to trace directory : %s\n", trace_name);
 261         printf("\n");
 262 }
 263
 264
 265 /* signal handling */
 266
 267 static void handler(int signo)
 268 {
 269         printf("Signal %d received : exiting cleanly\n", signo);
 270         quit_program = 1;
 271 }
 272
 273
 274 int open_buffer_file(char *filename, char *path_channel, char *path_trace,
 275         struct channel_trace_fd *fd_pairs)
 276 {
 277         int open_ret = 0;
 278         int ret = 0;
 279         struct stat stat_buf;
 280
 281         if(strncmp(filename, "flight-", sizeof("flight-")-1) != 0) {
 282                 if(dump_flight_only) {
 283                         printf_verbose("Skipping normal channel %s\n",
 284                                 path_channel);
 285                         return 0;
 286                 }
 287         } else {
 288                 if(dump_normal_only) {
 289                         printf_verbose("Skipping flight channel %s\n",
 290                                 path_channel);
 291                         return 0;
 292                 }
 293         }
 294         printf_verbose("Opening file.\n");
 295
 296         fd_pairs->pair = realloc(fd_pairs->pair,
 297                         ++fd_pairs->num_pairs * sizeof(struct fd_pair));
 298
 299         /* Open the channel in read mode */
 300         fd_pairs->pair[fd_pairs->num_pairs-1].channel =
 301                 open(path_channel, O_RDONLY | O_NONBLOCK);
 302         if(fd_pairs->pair[fd_pairs->num_pairs-1].channel == -1) {
 303                 perror(path_channel);
 304                 fd_pairs->num_pairs--;
 305                 return 0;       /* continue */
 306         }
 307         /* Open the trace in write mode, only append if append_mode */
 308         ret = stat(path_trace, &stat_buf);
 309         if(ret == 0) {
 310                 if(append_mode) {
 311                         printf_verbose("Appending to file %s as requested\n",
 312                                 path_trace);
 313
 314                         fd_pairs->pair[fd_pairs->num_pairs-1].trace =
 315                                 open(path_trace, O_WRONLY,
 316                                                 S_IRWXU|S_IRWXG|S_IRWXO);
 317                         if(fd_pairs->pair[fd_pairs->num_pairs-1].trace == -1) {
 318                                 perror(path_trace);
 319                                 open_ret = -1;
 320                                 close(fd_pairs->pair[fd_pairs->num_pairs-1].channel);
 321                                 fd_pairs->num_pairs--;
 322                                 goto end;
 323                         }
 324                         ret = lseek(fd_pairs->pair[fd_pairs->num_pairs-1].trace,
 325                                     0, SEEK_END);
 326                         if (ret < 0) {
 327                                 perror(path_trace);
 328                                 open_ret = -1;
 329                                 close(fd_pairs->pair[fd_pairs->num_pairs-1].channel);
 330                                 close(fd_pairs->pair[fd_pairs->num_pairs-1].trace);
 331                                 fd_pairs->num_pairs--;
 332                                 goto end;
 333                         }
 334                 } else {
 335                         printf("File %s exists, cannot open. Try append mode.\n", path_trace);
 336                         open_ret = -1;
 337                         close(fd_pairs->pair[fd_pairs->num_pairs-1].channel);
 338                         fd_pairs->num_pairs--;
 339                         goto end;
 340                 }
 341         } else {
 342                 if(errno == ENOENT) {
 343                         fd_pairs->pair[fd_pairs->num_pairs-1].trace =
 344                                 open(path_trace, O_WRONLY|O_CREAT|O_EXCL,
 345                                                 S_IRWXU|S_IRWXG|S_IRWXO);
 346                         if(fd_pairs->pair[fd_pairs->num_pairs-1].trace == -1) {
 347                                 perror(path_trace);
 348                                 open_ret = -1;
 349                                 close(fd_pairs->pair[fd_pairs->num_pairs-1].channel);
 350                                 fd_pairs->num_pairs--;
 351                                 goto end;
 352                         }
 353                 }
 354         }
 355 end:
 356         return open_ret;
 357 }
 358
 359 int open_channel_trace_pairs(char *subchannel_name, char *subtrace_name,
 360                 struct channel_trace_fd *fd_pairs, int *inotify_fd,
 361                 struct inotify_watch_array *iwatch_array)
 362 {
 363         DIR *channel_dir = opendir(subchannel_name);
 364         struct dirent *entry;
 365         struct stat stat_buf;
 366         int ret;
 367         char path_channel[PATH_MAX];
 368         int path_channel_len;
 369         char *path_channel_ptr;
 370         char path_trace[PATH_MAX];
 371         int path_trace_len;
 372         char *path_trace_ptr;
 373         int open_ret = 0;
 374
 375         if(channel_dir == NULL) {
 376                 perror(subchannel_name);
 377                 open_ret = ENOENT;
 378                 goto end;
 379         }
 380
 381         printf_verbose("Creating trace subdirectory %s\n", subtrace_name);
 382         ret = mkdir(subtrace_name, S_IRWXU|S_IRWXG|S_IRWXO);
 383         if(ret == -1) {
 384                 if(errno != EEXIST) {
 385                         perror(subtrace_name);
 386                         open_ret = -1;
 387                         goto end;
 388                 }
 389         }
 390
 391         strncpy(path_channel, subchannel_name, PATH_MAX-1);
 392         path_channel_len = strlen(path_channel);
 393         path_channel[path_channel_len] = '/';
 394         path_channel_len++;
 395         path_channel_ptr = path_channel + path_channel_len;
 396
 397         strncpy(path_trace, subtrace_name, PATH_MAX-1);
 398         path_trace_len = strlen(path_trace);
 399         path_trace[path_trace_len] = '/';
 400         path_trace_len++;
 401         path_trace_ptr = path_trace + path_trace_len;
 402
 403 #ifdef HAS_INOTIFY
 404         iwatch_array->elem = realloc(iwatch_array->elem,
 405                 ++iwatch_array->num * sizeof(struct inotify_watch));
 406
 407         printf_verbose("Adding inotify for channel %s\n", path_channel);
 408         iwatch_array->elem[iwatch_array->num-1].wd = inotify_add_watch(*inotify_fd, path_channel, IN_CREATE);
 409         strcpy(iwatch_array->elem[iwatch_array->num-1].path_channel, path_channel);
 410         strcpy(iwatch_array->elem[iwatch_array->num-1].path_trace, path_trace);
 411         printf_verbose("Added inotify for channel %s, wd %u\n",
 412                 iwatch_array->elem[iwatch_array->num-1].path_channel,
 413                 iwatch_array->elem[iwatch_array->num-1].wd);
 414 #endif
 415
 416         while((entry = readdir(channel_dir)) != NULL) {
 417
 418                 if(entry->d_name[0] == '.') continue;
 419
 420                 strncpy(path_channel_ptr, entry->d_name, PATH_MAX - path_channel_len);
 421                 strncpy(path_trace_ptr, entry->d_name, PATH_MAX - path_trace_len);
 422
 423                 ret = stat(path_channel, &stat_buf);
 424                 if(ret == -1) {
 425                         perror(path_channel);
 426                         continue;
 427                 }
 428
 429                 printf_verbose("Channel file : %s\n", path_channel);
 430
 431                 if(S_ISDIR(stat_buf.st_mode)) {
 432
 433                         printf_verbose("Entering channel subdirectory...\n");
 434                         ret = open_channel_trace_pairs(path_channel, path_trace, fd_pairs,
 435                                 inotify_fd, iwatch_array);
 436                         if(ret < 0) continue;
 437                 } else if(S_ISREG(stat_buf.st_mode)) {
 438                         open_ret = open_buffer_file(entry->d_name, path_channel, path_trace,
 439                                 fd_pairs);
 440                         if(open_ret)
 441                                 goto end;
 442                 }
 443         }
 444
 445 end:
 446         closedir(channel_dir);
 447
 448         return open_ret;
 449 }
 450
 451
 452 int read_subbuffer(struct fd_pair *pair)
 453 {
 454         unsigned int consumed_old, len;
 455         int err;
 456         long ret;
 457         off_t offset;
 458
 459
 460         err = ioctl(pair->channel, RELAY_GET_SB, &consumed_old);
 461         printf_verbose("cookie : %u\n", consumed_old);
 462         if(err != 0) {
 463                 ret = errno;
 464                 perror("Reserving sub buffer failed (everything is normal, it is due to concurrency)");
 465                 goto get_error;
 466         }
 467 #if 0
 468         err = TEMP_FAILURE_RETRY(write(pair->trace,
 469                                 pair->mmap
 470                                         + (consumed_old & ((pair->n_subbufs * pair->subbuf_size)-1)),
 471                                 pair->subbuf_size));
 472
 473         if(err < 0) {
 474                 ret = errno;
 475                 perror("Error in writing to file");
 476                 goto write_error;
 477         }
 478 #endif //0
 479         err = ioctl(pair->channel, RELAY_GET_SB_SIZE, &len);
 480         if(err != 0) {
 481                 ret = errno;
 482                 perror("Getting sub-buffer len failed.");
 483                 goto get_error;
 484         }
 485
 486         offset = 0;
 487         while (len > 0) {
 488                 printf_verbose("splice chan to pipe offset %lu\n",
 489                         (unsigned long)offset);
 490                 ret = splice(pair->channel, &offset, thread_pipe[1], NULL,
 491                         len, SPLICE_F_MOVE | SPLICE_F_MORE);
 492                 printf_verbose("splice chan to pipe ret %ld\n", ret);
 493                 if (ret < 0) {
 494                         perror("Error in relay splice");
 495                         goto write_error;
 496                 }
 497                 ret = splice(thread_pipe[0], NULL, pair->trace, NULL,
 498                         ret, SPLICE_F_MOVE | SPLICE_F_MORE);
 499                 printf_verbose("splice pipe to file %ld\n", ret);
 500                 if (ret < 0) {
 501                         perror("Error in file splice");
 502                         goto write_error;
 503                 }
 504                 len -= ret;
 505         }
 506
 507 #if 0
 508         err = fsync(pair->trace);
 509         if(err < 0) {
 510                 ret = errno;
 511                 perror("Error in writing to file");
 512                 goto write_error;
 513         }
 514 #endif //0
 515 write_error:
 516         ret = 0;
 517         err = ioctl(pair->channel, RELAY_PUT_SB, &consumed_old);
 518         if(err != 0) {
 519                 ret = errno;
 520                 if(errno == EFAULT) {
 521                         perror("Error in unreserving sub buffer\n");
 522                 } else if(errno == EIO) {
 523                         /* Should never happen with newer LTTng versions */
 524                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 525                 }
 526                 goto get_error;
 527         }
 528
 529 get_error:
 530         return ret;
 531 }
 532
 533
 534 int map_channels(struct channel_trace_fd *fd_pairs,
 535         int idx_begin, int idx_end)
 536 {
 537         int i,j;
 538         int ret=0;
 539
 540         if(fd_pairs->num_pairs <= 0) {
 541                 printf("No channel to read\n");
 542                 goto end;
 543         }
 544
 545         /* Get the subbuf sizes and number */
 546
 547         for(i=idx_begin;i<idx_end;i++) {
 548                 struct fd_pair *pair = &fd_pairs->pair[i];
 549
 550                 ret = ioctl(pair->channel, RELAY_GET_N_SB, &pair->n_sb);
 551                 if(ret != 0) {
 552                         perror("Error in getting the number of sub-buffers");
 553                         goto end;
 554                 }
 555                 ret = ioctl(pair->channel, RELAY_GET_MAX_SB_SIZE,
 556                             &pair->max_sb_size);
 557                 if(ret != 0) {
 558                         perror("Error in getting the max sub-buffer size");
 559                         goto end;
 560                 }
 561                 ret = pthread_mutex_init(&pair->mutex, NULL);   /* Fast mutex */
 562                 if(ret != 0) {
 563                         perror("Error in mutex init");
 564                         goto end;
 565                 }
 566         }
 567
 568 #if 0
 569         /* Mmap each FD */
 570         for(i=idx_begin;i<idx_end;i++) {
 571                 struct fd_pair *pair = &fd_pairs->pair[i];
 572
 573                 pair->mmap = mmap(0, pair->subbuf_size * pair->n_subbufs, PROT_READ,
 574                                 MAP_SHARED, pair->channel, 0);
 575                 if(pair->mmap == MAP_FAILED) {
 576                         perror("Mmap error");
 577                         goto munmap;
 578                 }
 579         }
 580
 581         goto end; /* success */
 582
 583         /* Error handling */
 584         /* munmap only the successfully mmapped indexes */
 585 munmap:
 586                 /* Munmap each FD */
 587         for(j=idx_begin;j<i;j++) {
 588                 struct fd_pair *pair = &fd_pairs->pair[j];
 589                 int err_ret;
 590
 591                 err_ret = munmap(pair->mmap, pair->subbuf_size * pair->n_subbufs);
 592                 if(err_ret != 0) {
 593                         perror("Error in munmap");
 594                 }
 595                 ret |= err_ret;
 596         }
 597
 598 #endif //0
 599 end:
 600         return ret;
 601 }
 602
 603 int unmap_channels(struct channel_trace_fd *fd_pairs)
 604 {
 605         int j;
 606         int ret=0;
 607
 608         /* Munmap each FD */
 609         for(j=0;j<fd_pairs->num_pairs;j++) {
 610                 struct fd_pair *pair = &fd_pairs->pair[j];
 611                 int err_ret;
 612
 613 #if 0
 614                 err_ret = munmap(pair->mmap, pair->subbuf_size * pair->n_subbufs);
 615                 if(err_ret != 0) {
 616                         perror("Error in munmap");
 617                 }
 618                 ret |= err_ret;
 619 #endif //0
 620                 err_ret = pthread_mutex_destroy(&pair->mutex);
 621                 if(err_ret != 0) {
 622                         perror("Error in mutex destroy");
 623                 }
 624                 ret |= err_ret;
 625         }
 626
 627         return ret;
 628 }
 629
 630 #ifdef HAS_INOTIFY
 631 /* Inotify event arrived.
 632  *
 633  * Only support add file for now.
 634  */
 635
 636 int read_inotify(int inotify_fd,
 637         struct channel_trace_fd *fd_pairs,
 638         struct inotify_watch_array *iwatch_array)
 639 {
 640         char buf[sizeof(struct inotify_event) + PATH_MAX];
 641         char path_channel[PATH_MAX];
 642         char path_trace[PATH_MAX];
 643         ssize_t len;
 644         struct inotify_event *ievent;
 645         size_t offset;
 646         unsigned int i;
 647         int ret;
 648         int old_num;
 649
 650         offset = 0;
 651         len = read(inotify_fd, buf, sizeof(struct inotify_event) + PATH_MAX);
 652         if(len < 0) {
 653
 654                 if(errno == EAGAIN)
 655                         return 0;  /* another thread got the data before us */
 656
 657                 printf("Error in read from inotify FD %s.\n", strerror(len));
 658                 return -1;
 659         }
 660         while(offset < len) {
 661                 ievent = (struct inotify_event *)&(buf[offset]);
 662                 for(i=0; i<iwatch_array->num; i++) {
 663                         if(iwatch_array->elem[i].wd == ievent->wd &&
 664                                 ievent->mask == IN_CREATE) {
 665                                 printf_verbose(
 666                                         "inotify wd %u event mask : %u for %s%s\n",
 667                                         ievent->wd, ievent->mask,
 668                                         iwatch_array->elem[i].path_channel,
 669                                         ievent->name);
 670                                 old_num = fd_pairs->num_pairs;
 671                                 strcpy(path_channel, iwatch_array->elem[i].path_channel);
 672                                 strcat(path_channel, ievent->name);
 673                                 strcpy(path_trace, iwatch_array->elem[i].path_trace);
 674                                 strcat(path_trace, ievent->name);
 675                                 if(ret = open_buffer_file(ievent->name, path_channel,
 676                                         path_trace, fd_pairs)) {
 677                                         printf("Error opening buffer file\n");
 678                                         return -1;
 679                                 }
 680                                 if(ret = map_channels(fd_pairs, old_num, fd_pairs->num_pairs)) {
 681                                         printf("Error mapping channel\n");
 682                                         return -1;
 683                                 }
 684
 685                         }
 686                 }
 687                 offset += sizeof(*ievent) + ievent->len;
 688         }
 689 }
 690 #endif //HAS_INOTIFY
 691
 692 /* read_channels
 693  *
 694  * Thread worker.
 695  *
 696  * Read the debugfs channels and write them in the paired tracefiles.
 697  *
 698  * @fd_pairs : paired channels and trace files.
 699  *
 700  * returns 0 on success, -1 on error.
 701  *
 702  * Note that the high priority polled channels are consumed first. We then poll
 703  * again to see if these channels are still in priority. Only when no
 704  * high priority channel is left, we start reading low priority channels.
 705  *
 706  * Note that a channel is considered high priority when the buffer is almost
 707  * full.
 708  */
 709
 710 int read_channels(unsigned long thread_num, struct channel_trace_fd *fd_pairs,
 711         int inotify_fd, struct inotify_watch_array *iwatch_array)
 712 {
 713         struct pollfd *pollfd = NULL;
 714         int num_pollfd;
 715         int i,j;
 716         int num_rdy, num_hup;
 717         int high_prio;
 718         int ret = 0;
 719         int inotify_fds;
 720         unsigned int old_num;
 721
 722 #ifdef HAS_INOTIFY
 723         inotify_fds = 1;
 724 #else
 725         inotify_fds = 0;
 726 #endif
 727
 728         pthread_rwlock_rdlock(&fd_pairs_lock);
 729
 730         /* Start polling the FD. Keep one fd for inotify */
 731         pollfd = malloc((inotify_fds + fd_pairs->num_pairs) * sizeof(struct pollfd));
 732
 733 #ifdef HAS_INOTIFY
 734         pollfd[0].fd = inotify_fd;
 735         pollfd[0].events = POLLIN|POLLPRI;
 736 #endif
 737
 738         for(i=0;i<fd_pairs->num_pairs;i++) {
 739                 pollfd[inotify_fds+i].fd = fd_pairs->pair[i].channel;
 740                 pollfd[inotify_fds+i].events = POLLIN|POLLPRI;
 741         }
 742         num_pollfd = inotify_fds + fd_pairs->num_pairs;
 743
 744
 745         pthread_rwlock_unlock(&fd_pairs_lock);
 746
 747         while(1) {
 748                 high_prio = 0;
 749                 num_hup = 0;
 750 #ifdef DEBUG
 751                 printf("Press a key for next poll...\n");
 752                 char buf[1];
 753                 read(STDIN_FILENO, &buf, 1);
 754                 printf("Next poll (polling %d fd) :\n", num_pollfd);
 755 #endif //DEBUG
 756
 757                 /* Have we received a signal ? */
 758                 if(quit_program) break;
 759
 760                 num_rdy = poll(pollfd, num_pollfd, -1);
 761
 762                 if(num_rdy == -1) {
 763                         perror("Poll error");
 764                         goto free_fd;
 765                 }
 766
 767                 printf_verbose("Data received\n");
 768 #ifdef HAS_INOTIFY
 769                 switch(pollfd[0].revents) {
 770                         case POLLERR:
 771                                 printf_verbose(
 772                                         "Error returned in polling inotify fd %d.\n",
 773                                         pollfd[0].fd);
 774                                 break;
 775                         case POLLHUP:
 776                                 printf_verbose(
 777                                         "Polling inotify fd %d tells it has hung up.\n",
 778                                         pollfd[0].fd);
 779                                 break;
 780                         case POLLNVAL:
 781                                 printf_verbose(
 782                                         "Polling inotify fd %d tells fd is not open.\n",
 783                                         pollfd[0].fd);
 784                                 break;
 785                         case POLLPRI:
 786                         case POLLIN:
 787                                 printf_verbose(
 788                                         "Polling inotify fd %d : data ready.\n",
 789                                         pollfd[0].fd);
 790
 791                                 pthread_rwlock_wrlock(&fd_pairs_lock);
 792                                 read_inotify(inotify_fd, fd_pairs, iwatch_array);
 793                                 pthread_rwlock_unlock(&fd_pairs_lock);
 794
 795                         break;
 796                 }
 797 #endif
 798
 799                 for(i=inotify_fds;i<num_pollfd;i++) {
 800                         switch(pollfd[i].revents) {
 801                                 case POLLERR:
 802                                         printf_verbose(
 803                                                 "Error returned in polling fd %d.\n",
 804                                                 pollfd[i].fd);
 805                                         num_hup++;
 806                                         break;
 807                                 case POLLHUP:
 808                                         printf_verbose(
 809                                                 "Polling fd %d tells it has hung up.\n",
 810                                                 pollfd[i].fd);
 811                                         num_hup++;
 812                                         break;
 813                                 case POLLNVAL:
 814                                         printf_verbose(
 815                                                 "Polling fd %d tells fd is not open.\n",
 816                                                 pollfd[i].fd);
 817                                         num_hup++;
 818                                         break;
 819                                 case POLLPRI:
 820                                         pthread_rwlock_rdlock(&fd_pairs_lock);
 821                                         if(pthread_mutex_trylock(&fd_pairs->pair[i-inotify_fds].mutex) == 0) {
 822                                                 printf_verbose(
 823                                                         "Urgent read on fd %d\n",
 824                                                         pollfd[i].fd);
 825                                                 /* Take care of high priority channels first. */
 826                                                 high_prio = 1;
 827                                                 /* it's ok to have an unavailable sub-buffer */
 828                                                 ret = read_subbuffer(&fd_pairs->pair[i-inotify_fds]);
 829                                                 if(ret == EAGAIN) ret = 0;
 830
 831                                                 ret = pthread_mutex_unlock(&fd_pairs->pair[i-inotify_fds].mutex);
 832                                                 if(ret)
 833                                                         printf("Error in mutex unlock : %s\n", strerror(ret));
 834                                         }
 835                                         pthread_rwlock_unlock(&fd_pairs_lock);
 836                                         break;
 837                         }
 838                 }
 839                 /* If every buffer FD has hung up, we end the read loop here */
 840                 if(num_hup == num_pollfd - inotify_fds) break;
 841
 842                 if(!high_prio) {
 843                         for(i=inotify_fds;i<num_pollfd;i++) {
 844                                 switch(pollfd[i].revents) {
 845                                         case POLLIN:
 846                                                 pthread_rwlock_rdlock(&fd_pairs_lock);
 847                                                 if(pthread_mutex_trylock(&fd_pairs->pair[i-inotify_fds].mutex) == 0) {
 848                                                         /* Take care of low priority channels. */
 849                                                         printf_verbose(
 850                                                                 "Normal read on fd %d\n",
 851                                                                 pollfd[i].fd);
 852                                                         /* it's ok to have an unavailable subbuffer */
 853                                                         ret = read_subbuffer(&fd_pairs->pair[i-inotify_fds]);
 854                                                         if(ret == EAGAIN) ret = 0;
 855
 856                                                         ret = pthread_mutex_unlock(&fd_pairs->pair[i-inotify_fds].mutex);
 857                                                         if(ret)
 858                                                                 printf("Error in mutex unlock : %s\n", strerror(ret));
 859                                                 }
 860                                                 pthread_rwlock_unlock(&fd_pairs_lock);
 861                                                 break;
 862                                 }
 863                         }
 864                 }
 865
 866                 /* Update pollfd array if an entry was added to fd_pairs */
 867                 pthread_rwlock_rdlock(&fd_pairs_lock);
 868                 if((inotify_fds + fd_pairs->num_pairs) != num_pollfd) {
 869                         pollfd = realloc(pollfd,
 870                                         (inotify_fds + fd_pairs->num_pairs) * sizeof(struct pollfd));
 871                         for(i=num_pollfd-inotify_fds;i<fd_pairs->num_pairs;i++) {
 872                                 pollfd[inotify_fds+i].fd = fd_pairs->pair[i].channel;
 873                                 pollfd[inotify_fds+i].events = POLLIN|POLLPRI;
 874                         }
 875                         num_pollfd = fd_pairs->num_pairs + inotify_fds;
 876                 }
 877                 pthread_rwlock_unlock(&fd_pairs_lock);
 878
 879                 /* NB: If the fd_pairs structure is updated by another thread from this
 880                  *     point forward, the current thread will wait in the poll without
 881                  *     monitoring the new channel. However, this thread will add the
 882                  *     new channel on next poll (and this should not take too much time
 883                  *     on a loaded system).
 884                  *
 885                  *     This event is quite unlikely and can only occur if a CPU is
 886                  *     hot-plugged while multple lttd threads are running.
 887                  */
 888         }
 889
 890 free_fd:
 891         free(pollfd);
 892
 893 end:
 894         return ret;
 895 }
 896
 897
 898 void close_channel_trace_pairs(struct channel_trace_fd *fd_pairs, int inotify_fd,
 899         struct inotify_watch_array *iwatch_array)
 900 {
 901         int i;
 902         int ret;
 903
 904         for(i=0;i<fd_pairs->num_pairs;i++) {
 905                 ret = close(fd_pairs->pair[i].channel);
 906                 if(ret == -1) perror("Close error on channel");
 907                 ret = close(fd_pairs->pair[i].trace);
 908                 if(ret == -1) perror("Close error on trace");
 909         }
 910         free(fd_pairs->pair);
 911         free(iwatch_array->elem);
 912 }
 913
 914 /* Thread worker */
 915 void * thread_main(void *arg)
 916 {
 917         long ret;
 918         unsigned long thread_num = (unsigned long)arg;
 919
 920         ret = pipe(thread_pipe);
 921         if (ret < 0) {
 922                 perror("Error creating pipe");
 923                 return (void*)ret;
 924         }
 925         ret = read_channels(thread_num, &fd_pairs, inotify_fd, &inotify_watch_array);
 926         close(thread_pipe[0]);  /* close read end */
 927         close(thread_pipe[1]);  /* close write end */
 928         return (void*)ret;
 929 }
 930
 931
 932 int channels_init()
 933 {
 934         int ret = 0;
 935
 936         inotify_fd = inotify_init();
 937         fcntl(inotify_fd, F_SETFL, O_NONBLOCK);
 938
 939         if(ret = open_channel_trace_pairs(channel_name, trace_name, &fd_pairs,
 940                         &inotify_fd, &inotify_watch_array))
 941                 goto close_channel;
 942         if (fd_pairs.num_pairs == 0) {
 943                 printf("No channel available for reading, exiting\n");
 944                 ret = -ENOENT;
 945                 goto close_channel;
 946         }
 947         if(ret = map_channels(&fd_pairs, 0, fd_pairs.num_pairs))
 948                 goto close_channel;
 949         return 0;
 950
 951 close_channel:
 952         close_channel_trace_pairs(&fd_pairs, inotify_fd, &inotify_watch_array);
 953         if(inotify_fd >= 0)
 954                 close(inotify_fd);
 955         return ret;
 956 }
 957
 958
 959 int main(int argc, char ** argv)
 960 {
 961         int ret = 0;
 962         struct sigaction act;
 963         pthread_t *tids;
 964         unsigned long i;
 965         void *tret;
 966
 967         ret = parse_arguments(argc, argv);
 968
 969         if(ret != 0) show_arguments();
 970         if(ret < 0) return EINVAL;
 971         if(ret > 0) return 0;
 972
 973         show_info();
 974
 975         /* Connect the signal handlers */
 976         act.sa_handler = handler;
 977         act.sa_flags = 0;
 978         sigemptyset(&(act.sa_mask));
 979         sigaddset(&(act.sa_mask), SIGTERM);
 980         sigaddset(&(act.sa_mask), SIGQUIT);
 981         sigaddset(&(act.sa_mask), SIGINT);
 982         sigaction(SIGTERM, &act, NULL);
 983         sigaction(SIGQUIT, &act, NULL);
 984         sigaction(SIGINT, &act, NULL);
 985
 986         if(ret = channels_init())
 987                 return ret;
 988
 989         if(daemon_mode) {
 990                 ret = daemon(0, 0);
 991
 992                 if(ret == -1) {
 993                         perror("An error occured while daemonizing.");
 994                         exit(-1);
 995                 }
 996         }
 997
 998         tids = malloc(sizeof(pthread_t) * num_threads);
 999         for(i=0; i<num_threads; i++) {
1000
1001                 ret = pthread_create(&tids[i], NULL, thread_main, (void*)i);
1002                 if(ret) {
1003                         perror("Error creating thread");
1004                         break;
1005                 }
1006         }
1007
1008         for(i=0; i<num_threads; i++) {
1009                 ret = pthread_join(tids[i], &tret);
1010                 if(ret) {
1011                         perror("Error joining thread");
1012                         break;
1013                 }
1014                 if((long)tret != 0) {
1015                         printf("Error %s occured in thread %u\n",
1016                                 strerror((long)tret), i);
1017                 }
1018         }
1019
1020         free(tids);
1021         ret = unmap_channels(&fd_pairs);
1022         close_channel_trace_pairs(&fd_pairs, inotify_fd, &inotify_watch_array);
1023         if(inotify_fd >= 0)
1024                 close(inotify_fd);
1025
1026         return ret;
1027 }