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