| 1 | /* |
| 2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> |
| 3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License |
| 7 | * as published by the Free Software Foundation; either version 2 |
| 8 | * of the License, or (at your option) any later version. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, |
| 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 13 | * GNU General Public License for more details. |
| 14 | * |
| 15 | * You should have received a copy of the GNU General Public License |
| 16 | * along with this program; if not, write to the Free Software |
| 17 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
| 18 | */ |
| 19 | |
| 20 | #define _GNU_SOURCE |
| 21 | #include <fcntl.h> |
| 22 | #include <poll.h> |
| 23 | #include <pthread.h> |
| 24 | #include <stdlib.h> |
| 25 | #include <string.h> |
| 26 | #include <sys/mman.h> |
| 27 | #include <sys/socket.h> |
| 28 | #include <sys/types.h> |
| 29 | #include <unistd.h> |
| 30 | #include <urcu/list.h> |
| 31 | |
| 32 | #include "libkernelctl.h" |
| 33 | #include "liblttkconsumerd.h" |
| 34 | #include "lttngerr.h" |
| 35 | |
| 36 | /* Init the list of FDs */ |
| 37 | static struct kconsumerd_fd_list kconsumerd_fd_list = { |
| 38 | .head = CDS_LIST_HEAD_INIT(kconsumerd_fd_list.head), |
| 39 | }; |
| 40 | |
| 41 | /* Number of element for the list below. */ |
| 42 | static unsigned int kconsumerd_fds_count; |
| 43 | |
| 44 | /* If the local array of FDs needs update in the poll function */ |
| 45 | static unsigned int kconsumerd_update_fd_array = 1; |
| 46 | |
| 47 | /* lock the fd array and structures */ |
| 48 | static pthread_mutex_t kconsumerd_lock_fds; |
| 49 | |
| 50 | /* communication with splice */ |
| 51 | static int kconsumerd_thread_pipe[2]; |
| 52 | |
| 53 | /* pipe to wake the poll thread when necessary */ |
| 54 | static int kconsumerd_poll_pipe[2]; |
| 55 | |
| 56 | /* timeout parameter, to control the polling thread grace period */ |
| 57 | static int kconsumerd_poll_timeout = -1; |
| 58 | |
| 59 | /* socket to communicate errors with sessiond */ |
| 60 | static int kconsumerd_error_socket; |
| 61 | |
| 62 | /* socket to exchange commands with sessiond */ |
| 63 | static char *kconsumerd_command_sock_path; |
| 64 | |
| 65 | /* flag to inform the polling thread to kconsumerd_quit when all fd hung up */ |
| 66 | static int kconsumerd_quit = 0; |
| 67 | |
| 68 | /* |
| 69 | * kconsumerd_set_error_socket |
| 70 | * |
| 71 | * Set the error socket |
| 72 | */ |
| 73 | void kconsumerd_set_error_socket(int sock) |
| 74 | { |
| 75 | kconsumerd_error_socket = sock; |
| 76 | } |
| 77 | |
| 78 | /* |
| 79 | * kconsumerd_set_command_socket_path |
| 80 | * |
| 81 | * Set the command socket path |
| 82 | */ |
| 83 | void kconsumerd_set_command_socket_path(char *sock) |
| 84 | { |
| 85 | kconsumerd_command_sock_path = sock; |
| 86 | } |
| 87 | |
| 88 | /* |
| 89 | * kconsumerd_find_session_fd |
| 90 | * |
| 91 | * Find a session fd in the global list. |
| 92 | * |
| 93 | * Return 1 if found else 0 |
| 94 | */ |
| 95 | static int kconsumerd_find_session_fd(int fd) |
| 96 | { |
| 97 | struct kconsumerd_fd *iter; |
| 98 | |
| 99 | pthread_mutex_lock(&kconsumerd_lock_fds); |
| 100 | cds_list_for_each_entry(iter, &kconsumerd_fd_list.head, list) { |
| 101 | if (iter->sessiond_fd == fd) { |
| 102 | DBG("Duplicate session fd %d", fd); |
| 103 | pthread_mutex_unlock(&kconsumerd_lock_fds); |
| 104 | return 1; |
| 105 | } |
| 106 | } |
| 107 | pthread_mutex_unlock(&kconsumerd_lock_fds); |
| 108 | |
| 109 | return 0; |
| 110 | } |
| 111 | |
| 112 | /* |
| 113 | * kconsumerd_del_fd |
| 114 | * |
| 115 | * Remove a fd from the global list protected by a mutex |
| 116 | */ |
| 117 | static void kconsumerd_del_fd(struct kconsumerd_fd *lcf) |
| 118 | { |
| 119 | pthread_mutex_lock(&kconsumerd_lock_fds); |
| 120 | cds_list_del(&lcf->list); |
| 121 | if (kconsumerd_fds_count > 0) { |
| 122 | kconsumerd_fds_count--; |
| 123 | if (lcf != NULL) { |
| 124 | close(lcf->out_fd); |
| 125 | close(lcf->consumerd_fd); |
| 126 | free(lcf); |
| 127 | lcf = NULL; |
| 128 | } |
| 129 | } |
| 130 | pthread_mutex_unlock(&kconsumerd_lock_fds); |
| 131 | } |
| 132 | |
| 133 | /* |
| 134 | * kconsumerd_add_fd |
| 135 | * |
| 136 | * Add a fd to the global list protected by a mutex |
| 137 | */ |
| 138 | static int kconsumerd_add_fd(struct lttcomm_kconsumerd_msg *buf, int consumerd_fd) |
| 139 | { |
| 140 | int ret; |
| 141 | struct kconsumerd_fd *tmp_fd; |
| 142 | |
| 143 | /* Check if already exist */ |
| 144 | ret = kconsumerd_find_session_fd(buf->fd); |
| 145 | if (ret == 1) { |
| 146 | goto end; |
| 147 | } |
| 148 | |
| 149 | tmp_fd = malloc(sizeof(struct kconsumerd_fd)); |
| 150 | tmp_fd->sessiond_fd = buf->fd; |
| 151 | tmp_fd->consumerd_fd = consumerd_fd; |
| 152 | tmp_fd->state = buf->state; |
| 153 | tmp_fd->max_sb_size = buf->max_sb_size; |
| 154 | strncpy(tmp_fd->path_name, buf->path_name, PATH_MAX); |
| 155 | |
| 156 | /* Opening the tracefile in write mode */ |
| 157 | ret = open(tmp_fd->path_name, |
| 158 | O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO); |
| 159 | if (ret < 0) { |
| 160 | ERR("Opening %s", tmp_fd->path_name); |
| 161 | perror("open"); |
| 162 | goto end; |
| 163 | } |
| 164 | tmp_fd->out_fd = ret; |
| 165 | tmp_fd->out_fd_offset = 0; |
| 166 | |
| 167 | DBG("Adding %s (%d, %d, %d)", tmp_fd->path_name, |
| 168 | tmp_fd->sessiond_fd, tmp_fd->consumerd_fd, tmp_fd->out_fd); |
| 169 | |
| 170 | pthread_mutex_lock(&kconsumerd_lock_fds); |
| 171 | cds_list_add(&tmp_fd->list, &kconsumerd_fd_list.head); |
| 172 | kconsumerd_fds_count++; |
| 173 | pthread_mutex_unlock(&kconsumerd_lock_fds); |
| 174 | |
| 175 | end: |
| 176 | return ret; |
| 177 | } |
| 178 | |
| 179 | /* |
| 180 | * kconsumerd_change_fd_state |
| 181 | * |
| 182 | * Update a fd according to what we just received |
| 183 | */ |
| 184 | static void kconsumerd_change_fd_state(int sessiond_fd, |
| 185 | enum kconsumerd_fd_state state) |
| 186 | { |
| 187 | struct kconsumerd_fd *iter; |
| 188 | |
| 189 | pthread_mutex_lock(&kconsumerd_lock_fds); |
| 190 | cds_list_for_each_entry(iter, &kconsumerd_fd_list.head, list) { |
| 191 | if (iter->sessiond_fd == sessiond_fd) { |
| 192 | iter->state = state; |
| 193 | break; |
| 194 | } |
| 195 | } |
| 196 | pthread_mutex_unlock(&kconsumerd_lock_fds); |
| 197 | } |
| 198 | |
| 199 | /* |
| 200 | * kconsumerd_update_poll_array |
| 201 | * |
| 202 | * Allocate the pollfd structure and the local view of the out fds |
| 203 | * to avoid doing a lookup in the linked list and concurrency issues |
| 204 | * when writing is needed. |
| 205 | * Returns the number of fds in the structures |
| 206 | */ |
| 207 | static int kconsumerd_update_poll_array(struct pollfd **pollfd, |
| 208 | struct kconsumerd_fd **local_kconsumerd_fd) |
| 209 | { |
| 210 | struct kconsumerd_fd *iter; |
| 211 | int i = 0; |
| 212 | |
| 213 | DBG("Updating poll fd array"); |
| 214 | |
| 215 | cds_list_for_each_entry(iter, &kconsumerd_fd_list.head, list) { |
| 216 | DBG("Inside for each"); |
| 217 | if (iter->state == ACTIVE_FD) { |
| 218 | DBG("Active FD %d", iter->consumerd_fd); |
| 219 | (*pollfd)[i].fd = iter->consumerd_fd; |
| 220 | (*pollfd)[i].events = POLLIN | POLLPRI; |
| 221 | local_kconsumerd_fd[i] = iter; |
| 222 | i++; |
| 223 | } |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * insert the kconsumerd_poll_pipe at the end of the array and don't |
| 228 | * increment i so nb_fd is the number of real FD |
| 229 | */ |
| 230 | (*pollfd)[i].fd = kconsumerd_poll_pipe[0]; |
| 231 | (*pollfd)[i].events = POLLIN; |
| 232 | |
| 233 | kconsumerd_update_fd_array = 0; |
| 234 | return i; |
| 235 | } |
| 236 | |
| 237 | |
| 238 | /* |
| 239 | * kconsumerd_on_read_subbuffer_mmap |
| 240 | * |
| 241 | * mmap the ring buffer, read it and write the data to the tracefile. |
| 242 | * Returns the number of bytes written |
| 243 | */ |
| 244 | static int kconsumerd_on_read_subbuffer_mmap( |
| 245 | struct kconsumerd_fd *kconsumerd_fd, unsigned long len) |
| 246 | { |
| 247 | unsigned long mmap_len, mmap_offset, padded_len, padding_len; |
| 248 | char *mmap_base; |
| 249 | char *padding = NULL; |
| 250 | long ret = 0; |
| 251 | off_t orig_offset = kconsumerd_fd->out_fd_offset; |
| 252 | int fd = kconsumerd_fd->consumerd_fd; |
| 253 | int outfd = kconsumerd_fd->out_fd; |
| 254 | |
| 255 | /* get the padded subbuffer size to know the padding required */ |
| 256 | ret = kernctl_get_padded_subbuf_size(fd, &padded_len); |
| 257 | if (ret != 0) { |
| 258 | ret = errno; |
| 259 | perror("kernctl_get_padded_subbuf_size"); |
| 260 | goto end; |
| 261 | } |
| 262 | padding_len = padded_len - len; |
| 263 | padding = malloc(padding_len * sizeof(char)); |
| 264 | memset(padding, '\0', padding_len); |
| 265 | |
| 266 | /* get the len of the mmap region */ |
| 267 | ret = kernctl_get_mmap_len(fd, &mmap_len); |
| 268 | if (ret != 0) { |
| 269 | ret = errno; |
| 270 | perror("kernctl_get_mmap_len"); |
| 271 | goto end; |
| 272 | } |
| 273 | |
| 274 | /* get the offset inside the fd to mmap */ |
| 275 | ret = kernctl_get_mmap_read_offset(fd, &mmap_offset); |
| 276 | if (ret != 0) { |
| 277 | ret = errno; |
| 278 | perror("kernctl_get_mmap_read_offset"); |
| 279 | goto end; |
| 280 | } |
| 281 | |
| 282 | mmap_base = mmap(NULL, mmap_len, PROT_READ, MAP_PRIVATE, fd, mmap_offset); |
| 283 | if (mmap_base == MAP_FAILED) { |
| 284 | perror("Error mmaping"); |
| 285 | ret = -1; |
| 286 | goto end; |
| 287 | } |
| 288 | |
| 289 | while (len > 0) { |
| 290 | ret = write(outfd, mmap_base, len); |
| 291 | if (ret >= len) { |
| 292 | len = 0; |
| 293 | } else if (ret < 0) { |
| 294 | ret = errno; |
| 295 | perror("Error in file write"); |
| 296 | goto end; |
| 297 | } |
| 298 | /* This won't block, but will start writeout asynchronously */ |
| 299 | sync_file_range(outfd, kconsumerd_fd->out_fd_offset, ret, |
| 300 | SYNC_FILE_RANGE_WRITE); |
| 301 | kconsumerd_fd->out_fd_offset += ret; |
| 302 | } |
| 303 | |
| 304 | /* once all the data is written, write the padding to disk */ |
| 305 | ret = write(outfd, padding, padding_len); |
| 306 | if (ret < 0) { |
| 307 | ret = errno; |
| 308 | perror("Error writing padding to file"); |
| 309 | goto end; |
| 310 | } |
| 311 | |
| 312 | /* |
| 313 | * This does a blocking write-and-wait on any page that belongs to the |
| 314 | * subbuffer prior to the one we just wrote. |
| 315 | * Don't care about error values, as these are just hints and ways to |
| 316 | * limit the amount of page cache used. |
| 317 | */ |
| 318 | if (orig_offset >= kconsumerd_fd->max_sb_size) { |
| 319 | sync_file_range(outfd, orig_offset - kconsumerd_fd->max_sb_size, |
| 320 | kconsumerd_fd->max_sb_size, |
| 321 | SYNC_FILE_RANGE_WAIT_BEFORE |
| 322 | | SYNC_FILE_RANGE_WRITE |
| 323 | | SYNC_FILE_RANGE_WAIT_AFTER); |
| 324 | |
| 325 | /* |
| 326 | * Give hints to the kernel about how we access the file: |
| 327 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after |
| 328 | * we write it. |
| 329 | * |
| 330 | * We need to call fadvise again after the file grows because the |
| 331 | * kernel does not seem to apply fadvise to non-existing parts of the |
| 332 | * file. |
| 333 | * |
| 334 | * Call fadvise _after_ having waited for the page writeback to |
| 335 | * complete because the dirty page writeback semantic is not well |
| 336 | * defined. So it can be expected to lead to lower throughput in |
| 337 | * streaming. |
| 338 | */ |
| 339 | posix_fadvise(outfd, orig_offset - kconsumerd_fd->max_sb_size, |
| 340 | kconsumerd_fd->max_sb_size, POSIX_FADV_DONTNEED); |
| 341 | } |
| 342 | goto end; |
| 343 | |
| 344 | end: |
| 345 | if (padding != NULL) { |
| 346 | free(padding); |
| 347 | } |
| 348 | return ret; |
| 349 | } |
| 350 | |
| 351 | /* |
| 352 | * kconsumerd_on_read_subbuffer |
| 353 | * |
| 354 | * Splice the data from the ring buffer to the tracefile. |
| 355 | * Returns the number of bytes spliced |
| 356 | */ |
| 357 | static int kconsumerd_on_read_subbuffer( |
| 358 | struct kconsumerd_fd *kconsumerd_fd, unsigned long len) |
| 359 | { |
| 360 | long ret = 0; |
| 361 | loff_t offset = 0; |
| 362 | off_t orig_offset = kconsumerd_fd->out_fd_offset; |
| 363 | int fd = kconsumerd_fd->consumerd_fd; |
| 364 | int outfd = kconsumerd_fd->out_fd; |
| 365 | |
| 366 | while (len > 0) { |
| 367 | DBG("splice chan to pipe offset %lu (fd : %d)", |
| 368 | (unsigned long)offset, fd); |
| 369 | ret = splice(fd, &offset, kconsumerd_thread_pipe[1], NULL, len, |
| 370 | SPLICE_F_MOVE | SPLICE_F_MORE); |
| 371 | DBG("splice chan to pipe ret %ld", ret); |
| 372 | if (ret < 0) { |
| 373 | ret = errno; |
| 374 | perror("Error in relay splice"); |
| 375 | goto splice_error; |
| 376 | } |
| 377 | |
| 378 | ret = splice(kconsumerd_thread_pipe[0], NULL, outfd, NULL, ret, |
| 379 | SPLICE_F_MOVE | SPLICE_F_MORE); |
| 380 | DBG("splice pipe to file %ld", ret); |
| 381 | if (ret < 0) { |
| 382 | ret = errno; |
| 383 | perror("Error in file splice"); |
| 384 | goto splice_error; |
| 385 | } |
| 386 | if (ret >= len) { |
| 387 | len = 0; |
| 388 | } |
| 389 | /* This won't block, but will start writeout asynchronously */ |
| 390 | sync_file_range(outfd, kconsumerd_fd->out_fd_offset, ret, |
| 391 | SYNC_FILE_RANGE_WRITE); |
| 392 | kconsumerd_fd->out_fd_offset += ret; |
| 393 | } |
| 394 | |
| 395 | /* |
| 396 | * This does a blocking write-and-wait on any page that belongs to the |
| 397 | * subbuffer prior to the one we just wrote. |
| 398 | * Don't care about error values, as these are just hints and ways to |
| 399 | * limit the amount of page cache used. |
| 400 | */ |
| 401 | if (orig_offset >= kconsumerd_fd->max_sb_size) { |
| 402 | sync_file_range(outfd, orig_offset - kconsumerd_fd->max_sb_size, |
| 403 | kconsumerd_fd->max_sb_size, |
| 404 | SYNC_FILE_RANGE_WAIT_BEFORE |
| 405 | | SYNC_FILE_RANGE_WRITE |
| 406 | | SYNC_FILE_RANGE_WAIT_AFTER); |
| 407 | /* |
| 408 | * Give hints to the kernel about how we access the file: |
| 409 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after |
| 410 | * we write it. |
| 411 | * |
| 412 | * We need to call fadvise again after the file grows because the |
| 413 | * kernel does not seem to apply fadvise to non-existing parts of the |
| 414 | * file. |
| 415 | * |
| 416 | * Call fadvise _after_ having waited for the page writeback to |
| 417 | * complete because the dirty page writeback semantic is not well |
| 418 | * defined. So it can be expected to lead to lower throughput in |
| 419 | * streaming. |
| 420 | */ |
| 421 | posix_fadvise(outfd, orig_offset - kconsumerd_fd->max_sb_size, |
| 422 | kconsumerd_fd->max_sb_size, POSIX_FADV_DONTNEED); |
| 423 | } |
| 424 | goto end; |
| 425 | |
| 426 | splice_error: |
| 427 | /* send the appropriate error description to sessiond */ |
| 428 | switch(ret) { |
| 429 | case EBADF: |
| 430 | kconsumerd_send_error(KCONSUMERD_SPLICE_EBADF); |
| 431 | break; |
| 432 | case EINVAL: |
| 433 | kconsumerd_send_error(KCONSUMERD_SPLICE_EINVAL); |
| 434 | break; |
| 435 | case ENOMEM: |
| 436 | kconsumerd_send_error(KCONSUMERD_SPLICE_ENOMEM); |
| 437 | break; |
| 438 | case ESPIPE: |
| 439 | kconsumerd_send_error(KCONSUMERD_SPLICE_ESPIPE); |
| 440 | break; |
| 441 | } |
| 442 | |
| 443 | end: |
| 444 | return ret; |
| 445 | } |
| 446 | |
| 447 | /* |
| 448 | * kconsumerd_read_subbuffer |
| 449 | * |
| 450 | * Consume data on a file descriptor and write it on a trace file |
| 451 | */ |
| 452 | static int kconsumerd_read_subbuffer(struct kconsumerd_fd *kconsumerd_fd) |
| 453 | { |
| 454 | unsigned long len; |
| 455 | int err; |
| 456 | long ret = 0; |
| 457 | int infd = kconsumerd_fd->consumerd_fd; |
| 458 | |
| 459 | DBG("In kconsumerd_read_subbuffer (infd : %d)", infd); |
| 460 | /* Get the next subbuffer */ |
| 461 | err = kernctl_get_next_subbuf(infd); |
| 462 | if (err != 0) { |
| 463 | ret = errno; |
| 464 | perror("Reserving sub buffer failed (everything is normal, " |
| 465 | "it is due to concurrency)"); |
| 466 | goto end; |
| 467 | } |
| 468 | |
| 469 | switch (DEFAULT_KERNEL_CHANNEL_OUTPUT) { |
| 470 | case LTTNG_KERNEL_SPLICE: |
| 471 | /* read the whole subbuffer */ |
| 472 | err = kernctl_get_padded_subbuf_size(infd, &len); |
| 473 | if (err != 0) { |
| 474 | ret = errno; |
| 475 | perror("Getting sub-buffer len failed."); |
| 476 | goto end; |
| 477 | } |
| 478 | |
| 479 | /* splice the subbuffer to the tracefile */ |
| 480 | ret = kconsumerd_on_read_subbuffer(kconsumerd_fd, len); |
| 481 | if (ret < 0) { |
| 482 | /* |
| 483 | * display the error but continue processing to try |
| 484 | * to release the subbuffer |
| 485 | */ |
| 486 | ERR("Error splicing to tracefile"); |
| 487 | } |
| 488 | break; |
| 489 | case LTTNG_KERNEL_MMAP: |
| 490 | /* read the used subbuffer size */ |
| 491 | err = kernctl_get_subbuf_size(infd, &len); |
| 492 | if (err != 0) { |
| 493 | ret = errno; |
| 494 | perror("Getting sub-buffer len failed."); |
| 495 | goto end; |
| 496 | } |
| 497 | /* write the subbuffer to the tracefile */ |
| 498 | ret = kconsumerd_on_read_subbuffer_mmap(kconsumerd_fd, len); |
| 499 | if (ret < 0) { |
| 500 | /* |
| 501 | * display the error but continue processing to try |
| 502 | * to release the subbuffer |
| 503 | */ |
| 504 | ERR("Error writing to tracefile"); |
| 505 | } |
| 506 | break; |
| 507 | default: |
| 508 | ERR("Unknown output method"); |
| 509 | ret = -1; |
| 510 | } |
| 511 | |
| 512 | err = kernctl_put_next_subbuf(infd); |
| 513 | if (err != 0) { |
| 514 | ret = errno; |
| 515 | if (errno == EFAULT) { |
| 516 | perror("Error in unreserving sub buffer\n"); |
| 517 | } else if (errno == EIO) { |
| 518 | /* Should never happen with newer LTTng versions */ |
| 519 | perror("Reader has been pushed by the writer, last sub-buffer corrupted."); |
| 520 | } |
| 521 | goto end; |
| 522 | } |
| 523 | |
| 524 | end: |
| 525 | return ret; |
| 526 | } |
| 527 | |
| 528 | /* |
| 529 | * kconsumerd_consumerd_recv_fd |
| 530 | * |
| 531 | * Receives an array of file descriptors and the associated |
| 532 | * structures describing each fd (path name). |
| 533 | * Returns the size of received data |
| 534 | */ |
| 535 | static int kconsumerd_consumerd_recv_fd(int sfd, int size, |
| 536 | enum kconsumerd_command cmd_type) |
| 537 | { |
| 538 | struct msghdr msg; |
| 539 | struct iovec iov[1]; |
| 540 | int ret = 0, i, tmp2; |
| 541 | struct cmsghdr *cmsg; |
| 542 | int nb_fd; |
| 543 | char recv_fd[CMSG_SPACE(sizeof(int))]; |
| 544 | struct lttcomm_kconsumerd_msg lkm; |
| 545 | |
| 546 | /* the number of fds we are about to receive */ |
| 547 | nb_fd = size / sizeof(struct lttcomm_kconsumerd_msg); |
| 548 | |
| 549 | for (i = 0; i < nb_fd; i++) { |
| 550 | memset(&msg, 0, sizeof(msg)); |
| 551 | |
| 552 | /* Prepare to receive the structures */ |
| 553 | iov[0].iov_base = &lkm; |
| 554 | iov[0].iov_len = sizeof(lkm); |
| 555 | msg.msg_iov = iov; |
| 556 | msg.msg_iovlen = 1; |
| 557 | |
| 558 | msg.msg_control = recv_fd; |
| 559 | msg.msg_controllen = sizeof(recv_fd); |
| 560 | |
| 561 | DBG("Waiting to receive fd"); |
| 562 | if ((ret = recvmsg(sfd, &msg, 0)) < 0) { |
| 563 | perror("recvmsg"); |
| 564 | continue; |
| 565 | } |
| 566 | |
| 567 | if (ret != (size / nb_fd)) { |
| 568 | ERR("Received only %d, expected %d", ret, size); |
| 569 | kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD); |
| 570 | goto end; |
| 571 | } |
| 572 | |
| 573 | cmsg = CMSG_FIRSTHDR(&msg); |
| 574 | if (!cmsg) { |
| 575 | ERR("Invalid control message header"); |
| 576 | ret = -1; |
| 577 | kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD); |
| 578 | goto end; |
| 579 | } |
| 580 | /* if we received fds */ |
| 581 | if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) { |
| 582 | switch (cmd_type) { |
| 583 | case ADD_STREAM: |
| 584 | DBG("kconsumerd_add_fd %s (%d)", lkm.path_name, (CMSG_DATA(cmsg)[0])); |
| 585 | ret = kconsumerd_add_fd(&lkm, (CMSG_DATA(cmsg)[0])); |
| 586 | if (ret < 0) { |
| 587 | kconsumerd_send_error(KCONSUMERD_OUTFD_ERROR); |
| 588 | goto end; |
| 589 | } |
| 590 | break; |
| 591 | case UPDATE_STREAM: |
| 592 | kconsumerd_change_fd_state(lkm.fd, lkm.state); |
| 593 | break; |
| 594 | default: |
| 595 | break; |
| 596 | } |
| 597 | /* flag to tell the polling thread to update its fd array */ |
| 598 | kconsumerd_update_fd_array = 1; |
| 599 | /* signal the poll thread */ |
| 600 | tmp2 = write(kconsumerd_poll_pipe[1], "4", 1); |
| 601 | } else { |
| 602 | ERR("Didn't received any fd"); |
| 603 | kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD); |
| 604 | ret = -1; |
| 605 | goto end; |
| 606 | } |
| 607 | } |
| 608 | |
| 609 | end: |
| 610 | return ret; |
| 611 | } |
| 612 | |
| 613 | /* |
| 614 | * kconsumerd_thread_poll_fds |
| 615 | * |
| 616 | * This thread polls the fds in the ltt_fd_list to consume the data |
| 617 | * and write it to tracefile if necessary. |
| 618 | */ |
| 619 | void *kconsumerd_thread_poll_fds(void *data) |
| 620 | { |
| 621 | int num_rdy, num_hup, high_prio, ret, i; |
| 622 | struct pollfd *pollfd = NULL; |
| 623 | /* local view of the fds */ |
| 624 | struct kconsumerd_fd **local_kconsumerd_fd = NULL; |
| 625 | /* local view of kconsumerd_fds_count */ |
| 626 | int nb_fd = 0; |
| 627 | char tmp; |
| 628 | int tmp2; |
| 629 | |
| 630 | ret = pipe(kconsumerd_thread_pipe); |
| 631 | if (ret < 0) { |
| 632 | perror("Error creating pipe"); |
| 633 | goto end; |
| 634 | } |
| 635 | |
| 636 | local_kconsumerd_fd = malloc(sizeof(struct kconsumerd_fd)); |
| 637 | |
| 638 | while (1) { |
| 639 | high_prio = 0; |
| 640 | num_hup = 0; |
| 641 | |
| 642 | /* |
| 643 | * the ltt_fd_list has been updated, we need to update our |
| 644 | * local array as well |
| 645 | */ |
| 646 | if (kconsumerd_update_fd_array == 1) { |
| 647 | if (pollfd != NULL) { |
| 648 | free(pollfd); |
| 649 | pollfd = NULL; |
| 650 | } |
| 651 | if (local_kconsumerd_fd != NULL) { |
| 652 | free(local_kconsumerd_fd); |
| 653 | local_kconsumerd_fd = NULL; |
| 654 | } |
| 655 | |
| 656 | /* Lock mutex for fds count */ |
| 657 | pthread_mutex_lock(&kconsumerd_lock_fds); |
| 658 | /* allocate for all fds + 1 for the kconsumerd_poll_pipe */ |
| 659 | pollfd = malloc((kconsumerd_fds_count + 1) * sizeof(struct pollfd)); |
| 660 | if (pollfd == NULL) { |
| 661 | perror("pollfd malloc"); |
| 662 | goto end; |
| 663 | } |
| 664 | |
| 665 | /* allocate for all fds + 1 for the kconsumerd_poll_pipe */ |
| 666 | local_kconsumerd_fd = malloc((kconsumerd_fds_count + 1) * |
| 667 | sizeof(struct kconsumerd_fd)); |
| 668 | if (local_kconsumerd_fd == NULL) { |
| 669 | perror("local_kconsumerd_fd malloc"); |
| 670 | goto end; |
| 671 | } |
| 672 | ret = kconsumerd_update_poll_array(&pollfd, local_kconsumerd_fd); |
| 673 | if (ret < 0) { |
| 674 | ERR("Error in allocating pollfd or local_outfds"); |
| 675 | kconsumerd_send_error(KCONSUMERD_POLL_ERROR); |
| 676 | goto end; |
| 677 | } |
| 678 | /* Unlock mutex for fds count */ |
| 679 | pthread_mutex_unlock(&kconsumerd_lock_fds); |
| 680 | |
| 681 | nb_fd = ret; |
| 682 | } |
| 683 | |
| 684 | /* poll on the array of fds */ |
| 685 | DBG("polling on %d fd", nb_fd + 1); |
| 686 | num_rdy = poll(pollfd, nb_fd + 1, kconsumerd_poll_timeout); |
| 687 | DBG("poll num_rdy : %d", num_rdy); |
| 688 | if (num_rdy == -1) { |
| 689 | perror("Poll error"); |
| 690 | kconsumerd_send_error(KCONSUMERD_POLL_ERROR); |
| 691 | goto end; |
| 692 | } else if (num_rdy == 0) { |
| 693 | DBG("Polling thread timed out"); |
| 694 | goto end; |
| 695 | } |
| 696 | |
| 697 | /* No FDs and kconsumerd_quit, kconsumerd_cleanup the thread */ |
| 698 | if (nb_fd == 0 && kconsumerd_quit == 1) { |
| 699 | goto end; |
| 700 | } |
| 701 | |
| 702 | /* |
| 703 | * if only the kconsumerd_poll_pipe triggered poll to return just |
| 704 | * return to the beginning of the loop to update the array |
| 705 | */ |
| 706 | if (num_rdy == 1 && pollfd[nb_fd].revents == POLLIN) { |
| 707 | DBG("kconsumerd_poll_pipe wake up"); |
| 708 | tmp2 = read(kconsumerd_poll_pipe[0], &tmp, 1); |
| 709 | continue; |
| 710 | } |
| 711 | |
| 712 | /* Take care of high priority channels first. */ |
| 713 | for (i = 0; i < nb_fd; i++) { |
| 714 | switch(pollfd[i].revents) { |
| 715 | case POLLERR: |
| 716 | ERR("Error returned in polling fd %d.", pollfd[i].fd); |
| 717 | kconsumerd_del_fd(local_kconsumerd_fd[i]); |
| 718 | kconsumerd_update_fd_array = 1; |
| 719 | num_hup++; |
| 720 | break; |
| 721 | case POLLHUP: |
| 722 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); |
| 723 | kconsumerd_del_fd(local_kconsumerd_fd[i]); |
| 724 | kconsumerd_update_fd_array = 1; |
| 725 | num_hup++; |
| 726 | break; |
| 727 | case POLLNVAL: |
| 728 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); |
| 729 | kconsumerd_del_fd(local_kconsumerd_fd[i]); |
| 730 | kconsumerd_update_fd_array = 1; |
| 731 | num_hup++; |
| 732 | break; |
| 733 | case POLLPRI: |
| 734 | DBG("Urgent read on fd %d", pollfd[i].fd); |
| 735 | high_prio = 1; |
| 736 | ret = kconsumerd_read_subbuffer(local_kconsumerd_fd[i]); |
| 737 | /* it's ok to have an unavailable sub-buffer */ |
| 738 | if (ret == EAGAIN) { |
| 739 | ret = 0; |
| 740 | } |
| 741 | break; |
| 742 | } |
| 743 | } |
| 744 | |
| 745 | /* If every buffer FD has hung up, we end the read loop here */ |
| 746 | if (nb_fd > 0 && num_hup == nb_fd) { |
| 747 | DBG("every buffer FD has hung up\n"); |
| 748 | if (kconsumerd_quit == 1) { |
| 749 | goto end; |
| 750 | } |
| 751 | continue; |
| 752 | } |
| 753 | |
| 754 | /* Take care of low priority channels. */ |
| 755 | if (high_prio == 0) { |
| 756 | for (i = 0; i < nb_fd; i++) { |
| 757 | if (pollfd[i].revents == POLLIN) { |
| 758 | DBG("Normal read on fd %d", pollfd[i].fd); |
| 759 | ret = kconsumerd_read_subbuffer(local_kconsumerd_fd[i]); |
| 760 | /* it's ok to have an unavailable subbuffer */ |
| 761 | if (ret == EAGAIN) { |
| 762 | ret = 0; |
| 763 | } |
| 764 | } |
| 765 | } |
| 766 | } |
| 767 | } |
| 768 | end: |
| 769 | pthread_mutex_unlock(&kconsumerd_lock_fds); |
| 770 | DBG("polling thread exiting"); |
| 771 | if (pollfd != NULL) { |
| 772 | free(pollfd); |
| 773 | pollfd = NULL; |
| 774 | } |
| 775 | if (local_kconsumerd_fd != NULL) { |
| 776 | free(local_kconsumerd_fd); |
| 777 | local_kconsumerd_fd = NULL; |
| 778 | } |
| 779 | kconsumerd_cleanup(); |
| 780 | return NULL; |
| 781 | } |
| 782 | |
| 783 | /* |
| 784 | * kconsumerd_create_poll_pipe |
| 785 | * |
| 786 | * create the pipe to wake to polling thread when needed |
| 787 | */ |
| 788 | int kconsumerd_create_poll_pipe() |
| 789 | { |
| 790 | return pipe(kconsumerd_poll_pipe); |
| 791 | } |
| 792 | |
| 793 | /* |
| 794 | * kconsumerd_thread_receive_fds |
| 795 | * |
| 796 | * This thread listens on the consumerd socket and |
| 797 | * receives the file descriptors from ltt-sessiond |
| 798 | */ |
| 799 | void *kconsumerd_thread_receive_fds(void *data) |
| 800 | { |
| 801 | int sock, client_socket, ret; |
| 802 | struct lttcomm_kconsumerd_header tmp; |
| 803 | |
| 804 | DBG("Creating command socket %s", kconsumerd_command_sock_path); |
| 805 | unlink(kconsumerd_command_sock_path); |
| 806 | client_socket = lttcomm_create_unix_sock(kconsumerd_command_sock_path); |
| 807 | if (client_socket < 0) { |
| 808 | ERR("Cannot create command socket"); |
| 809 | goto end; |
| 810 | } |
| 811 | |
| 812 | ret = lttcomm_listen_unix_sock(client_socket); |
| 813 | if (ret < 0) { |
| 814 | goto end; |
| 815 | } |
| 816 | |
| 817 | DBG("Sending ready command to ltt-sessiond"); |
| 818 | ret = kconsumerd_send_error(KCONSUMERD_COMMAND_SOCK_READY); |
| 819 | if (ret < 0) { |
| 820 | ERR("Error sending ready command to ltt-sessiond"); |
| 821 | goto end; |
| 822 | } |
| 823 | |
| 824 | /* Blocking call, waiting for transmission */ |
| 825 | sock = lttcomm_accept_unix_sock(client_socket); |
| 826 | if (sock <= 0) { |
| 827 | WARN("On accept"); |
| 828 | goto end; |
| 829 | } |
| 830 | while (1) { |
| 831 | /* We first get the number of fd we are about to receive */ |
| 832 | ret = lttcomm_recv_unix_sock(sock, &tmp, |
| 833 | sizeof(struct lttcomm_kconsumerd_header)); |
| 834 | if (ret <= 0) { |
| 835 | ERR("Communication interrupted on command socket"); |
| 836 | goto end; |
| 837 | } |
| 838 | if (tmp.cmd_type == STOP) { |
| 839 | DBG("Received STOP command"); |
| 840 | goto end; |
| 841 | } |
| 842 | /* we received a command to add or update fds */ |
| 843 | ret = kconsumerd_consumerd_recv_fd(sock, tmp.payload_size, tmp.cmd_type); |
| 844 | if (ret <= 0) { |
| 845 | ERR("Receiving the FD, exiting"); |
| 846 | goto end; |
| 847 | } |
| 848 | } |
| 849 | |
| 850 | end: |
| 851 | DBG("kconsumerd_thread_receive_fds exiting"); |
| 852 | |
| 853 | /* |
| 854 | * when all fds have hung up, the polling thread |
| 855 | * can exit cleanly |
| 856 | */ |
| 857 | kconsumerd_quit = 1; |
| 858 | |
| 859 | /* |
| 860 | * 2s of grace period, if no polling events occur during |
| 861 | * this period, the polling thread will exit even if there |
| 862 | * are still open FDs (should not happen, but safety mechanism). |
| 863 | */ |
| 864 | kconsumerd_poll_timeout = KCONSUMERD_POLL_GRACE_PERIOD; |
| 865 | |
| 866 | /* wake up the polling thread */ |
| 867 | ret = write(kconsumerd_poll_pipe[1], "4", 1); |
| 868 | if (ret < 0) { |
| 869 | perror("poll pipe write"); |
| 870 | } |
| 871 | return NULL; |
| 872 | } |
| 873 | |
| 874 | /* |
| 875 | * kconsumerd_cleanup |
| 876 | * |
| 877 | * Cleanup the daemon's socket on exit |
| 878 | */ |
| 879 | void kconsumerd_cleanup() |
| 880 | { |
| 881 | struct kconsumerd_fd *iter; |
| 882 | |
| 883 | /* remove the socket file */ |
| 884 | unlink(kconsumerd_command_sock_path); |
| 885 | |
| 886 | /* close all outfd */ |
| 887 | cds_list_for_each_entry(iter, &kconsumerd_fd_list.head, list) { |
| 888 | kconsumerd_del_fd(iter); |
| 889 | } |
| 890 | } |
| 891 | |
| 892 | /* |
| 893 | * kconsumerd_send_error |
| 894 | * |
| 895 | * send return code to ltt-sessiond |
| 896 | */ |
| 897 | int kconsumerd_send_error(enum lttcomm_return_code cmd) |
| 898 | { |
| 899 | if (kconsumerd_error_socket > 0) { |
| 900 | return lttcomm_send_unix_sock(kconsumerd_error_socket, &cmd, |
| 901 | sizeof(enum lttcomm_sessiond_command)); |
| 902 | } |
| 903 | |
| 904 | return 0; |
| 905 | } |