| 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; only version 2 |
| 8 | * of the License. |
| 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 | #include <assert.h> |
| 32 | |
| 33 | #include "kernelctl.h" |
| 34 | #include "lttkconsumerd.h" |
| 35 | #include "lttngerr.h" |
| 36 | |
| 37 | static |
| 38 | struct kconsumerd_global_data { |
| 39 | /* |
| 40 | * kconsumerd_data.lock protects kconsumerd_data.fd_list, |
| 41 | * kconsumerd_data.fds_count, and kconsumerd_data.need_update. It |
| 42 | * ensures the count matches the number of items in the fd_list. |
| 43 | * It ensures the list updates *always* trigger an fd_array |
| 44 | * update (therefore need to make list update vs |
| 45 | * kconsumerd_data.need_update flag update atomic, and also flag |
| 46 | * read, fd array and flag clear atomic). |
| 47 | */ |
| 48 | pthread_mutex_t lock; |
| 49 | /* |
| 50 | * Number of element for the list below. Protected by |
| 51 | * kconsumerd_data.lock. |
| 52 | */ |
| 53 | unsigned int fds_count; |
| 54 | /* |
| 55 | * List of FDs. Protected by kconsumerd_data.lock. |
| 56 | */ |
| 57 | struct kconsumerd_fd_list fd_list; |
| 58 | /* |
| 59 | * Flag specifying if the local array of FDs needs update in the |
| 60 | * poll function. Protected by kconsumerd_data.lock. |
| 61 | */ |
| 62 | unsigned int need_update; |
| 63 | } kconsumerd_data = { |
| 64 | .fd_list.head = CDS_LIST_HEAD_INIT(kconsumerd_data.fd_list.head), |
| 65 | }; |
| 66 | |
| 67 | /* communication with splice */ |
| 68 | static int kconsumerd_thread_pipe[2]; |
| 69 | |
| 70 | /* pipe to wake the poll thread when necessary */ |
| 71 | static int kconsumerd_poll_pipe[2]; |
| 72 | |
| 73 | /* to let the signal handler wake up the fd receiver thread */ |
| 74 | static int kconsumerd_should_quit[2]; |
| 75 | |
| 76 | /* timeout parameter, to control the polling thread grace period */ |
| 77 | static int kconsumerd_poll_timeout = -1; |
| 78 | |
| 79 | /* socket to communicate errors with sessiond */ |
| 80 | static int kconsumerd_error_socket; |
| 81 | |
| 82 | /* socket to exchange commands with sessiond */ |
| 83 | static char *kconsumerd_command_sock_path; |
| 84 | |
| 85 | /* |
| 86 | * flag to inform the polling thread to quit when all fd hung up. |
| 87 | * Updated by the kconsumerd_thread_receive_fds when it notices that all |
| 88 | * fds has hung up. Also updated by the signal handler |
| 89 | * (kconsumerd_should_exit()). Read by the polling threads. |
| 90 | */ |
| 91 | static volatile int kconsumerd_quit = 0; |
| 92 | |
| 93 | /* |
| 94 | * kconsumerd_set_error_socket |
| 95 | * |
| 96 | * Set the error socket |
| 97 | */ |
| 98 | void kconsumerd_set_error_socket(int sock) |
| 99 | { |
| 100 | kconsumerd_error_socket = sock; |
| 101 | } |
| 102 | |
| 103 | /* |
| 104 | * kconsumerd_set_command_socket_path |
| 105 | * |
| 106 | * Set the command socket path |
| 107 | */ |
| 108 | void kconsumerd_set_command_socket_path(char *sock) |
| 109 | { |
| 110 | kconsumerd_command_sock_path = sock; |
| 111 | } |
| 112 | |
| 113 | /* |
| 114 | * kconsumerd_find_session_fd |
| 115 | * |
| 116 | * Find a session fd in the global list. |
| 117 | * The kconsumerd_data.lock must be locked during this call |
| 118 | * |
| 119 | * Return 1 if found else 0 |
| 120 | */ |
| 121 | static int kconsumerd_find_session_fd(int fd) |
| 122 | { |
| 123 | struct kconsumerd_fd *iter; |
| 124 | |
| 125 | cds_list_for_each_entry(iter, &kconsumerd_data.fd_list.head, list) { |
| 126 | if (iter->sessiond_fd == fd) { |
| 127 | DBG("Duplicate session fd %d", fd); |
| 128 | return 1; |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | return 0; |
| 133 | } |
| 134 | |
| 135 | /* |
| 136 | * kconsumerd_del_fd |
| 137 | * |
| 138 | * Remove a fd from the global list protected by a mutex |
| 139 | */ |
| 140 | static void kconsumerd_del_fd(struct kconsumerd_fd *lcf) |
| 141 | { |
| 142 | pthread_mutex_lock(&kconsumerd_data.lock); |
| 143 | cds_list_del(&lcf->list); |
| 144 | if (kconsumerd_data.fds_count > 0) { |
| 145 | kconsumerd_data.fds_count--; |
| 146 | if (lcf != NULL) { |
| 147 | close(lcf->out_fd); |
| 148 | close(lcf->consumerd_fd); |
| 149 | free(lcf); |
| 150 | lcf = NULL; |
| 151 | } |
| 152 | } |
| 153 | kconsumerd_data.need_update = 1; |
| 154 | pthread_mutex_unlock(&kconsumerd_data.lock); |
| 155 | } |
| 156 | |
| 157 | /* |
| 158 | * kconsumerd_add_fd |
| 159 | * |
| 160 | * Add a fd to the global list protected by a mutex |
| 161 | */ |
| 162 | static int kconsumerd_add_fd(struct lttcomm_kconsumerd_msg *buf, int consumerd_fd) |
| 163 | { |
| 164 | int ret; |
| 165 | struct kconsumerd_fd *tmp_fd; |
| 166 | |
| 167 | pthread_mutex_lock(&kconsumerd_data.lock); |
| 168 | /* Check if already exist */ |
| 169 | ret = kconsumerd_find_session_fd(buf->fd); |
| 170 | if (ret == 1) { |
| 171 | goto end; |
| 172 | } |
| 173 | |
| 174 | tmp_fd = malloc(sizeof(struct kconsumerd_fd)); |
| 175 | tmp_fd->sessiond_fd = buf->fd; |
| 176 | tmp_fd->consumerd_fd = consumerd_fd; |
| 177 | tmp_fd->state = buf->state; |
| 178 | tmp_fd->max_sb_size = buf->max_sb_size; |
| 179 | strncpy(tmp_fd->path_name, buf->path_name, PATH_MAX); |
| 180 | tmp_fd->path_name[PATH_MAX - 1] = '\0'; |
| 181 | |
| 182 | /* Opening the tracefile in write mode */ |
| 183 | ret = open(tmp_fd->path_name, |
| 184 | O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO); |
| 185 | if (ret < 0) { |
| 186 | ERR("Opening %s", tmp_fd->path_name); |
| 187 | perror("open"); |
| 188 | goto end; |
| 189 | } |
| 190 | tmp_fd->out_fd = ret; |
| 191 | tmp_fd->out_fd_offset = 0; |
| 192 | |
| 193 | DBG("Adding %s (%d, %d, %d)", tmp_fd->path_name, |
| 194 | tmp_fd->sessiond_fd, tmp_fd->consumerd_fd, tmp_fd->out_fd); |
| 195 | |
| 196 | cds_list_add(&tmp_fd->list, &kconsumerd_data.fd_list.head); |
| 197 | kconsumerd_data.fds_count++; |
| 198 | kconsumerd_data.need_update = 1; |
| 199 | end: |
| 200 | pthread_mutex_unlock(&kconsumerd_data.lock); |
| 201 | return ret; |
| 202 | } |
| 203 | |
| 204 | /* |
| 205 | * kconsumerd_change_fd_state |
| 206 | * |
| 207 | * Update a fd according to what we just received |
| 208 | */ |
| 209 | static void kconsumerd_change_fd_state(int sessiond_fd, |
| 210 | enum kconsumerd_fd_state state) |
| 211 | { |
| 212 | struct kconsumerd_fd *iter; |
| 213 | |
| 214 | pthread_mutex_lock(&kconsumerd_data.lock); |
| 215 | cds_list_for_each_entry(iter, &kconsumerd_data.fd_list.head, list) { |
| 216 | if (iter->sessiond_fd == sessiond_fd) { |
| 217 | iter->state = state; |
| 218 | break; |
| 219 | } |
| 220 | } |
| 221 | kconsumerd_data.need_update = 1; |
| 222 | pthread_mutex_unlock(&kconsumerd_data.lock); |
| 223 | } |
| 224 | |
| 225 | /* |
| 226 | * kconsumerd_update_poll_array |
| 227 | * |
| 228 | * Allocate the pollfd structure and the local view of the out fds |
| 229 | * to avoid doing a lookup in the linked list and concurrency issues |
| 230 | * when writing is needed. |
| 231 | * Returns the number of fds in the structures |
| 232 | * Called with kconsumerd_data.lock held. |
| 233 | */ |
| 234 | static int kconsumerd_update_poll_array(struct pollfd **pollfd, |
| 235 | struct kconsumerd_fd **local_kconsumerd_fd) |
| 236 | { |
| 237 | struct kconsumerd_fd *iter; |
| 238 | int i = 0; |
| 239 | |
| 240 | DBG("Updating poll fd array"); |
| 241 | |
| 242 | cds_list_for_each_entry(iter, &kconsumerd_data.fd_list.head, list) { |
| 243 | DBG("Inside for each"); |
| 244 | if (iter->state == ACTIVE_FD) { |
| 245 | DBG("Active FD %d", iter->consumerd_fd); |
| 246 | (*pollfd)[i].fd = iter->consumerd_fd; |
| 247 | (*pollfd)[i].events = POLLIN | POLLPRI; |
| 248 | local_kconsumerd_fd[i] = iter; |
| 249 | i++; |
| 250 | } |
| 251 | } |
| 252 | |
| 253 | /* |
| 254 | * insert the kconsumerd_poll_pipe at the end of the array and don't |
| 255 | * increment i so nb_fd is the number of real FD |
| 256 | */ |
| 257 | (*pollfd)[i].fd = kconsumerd_poll_pipe[0]; |
| 258 | (*pollfd)[i].events = POLLIN; |
| 259 | return i; |
| 260 | } |
| 261 | |
| 262 | |
| 263 | /* |
| 264 | * kconsumerd_on_read_subbuffer_mmap |
| 265 | * |
| 266 | * mmap the ring buffer, read it and write the data to the tracefile. |
| 267 | * Returns the number of bytes written |
| 268 | */ |
| 269 | static int kconsumerd_on_read_subbuffer_mmap( |
| 270 | struct kconsumerd_fd *kconsumerd_fd, unsigned long len) |
| 271 | { |
| 272 | unsigned long mmap_len, mmap_offset, padded_len, padding_len; |
| 273 | char *mmap_base; |
| 274 | char *padding = NULL; |
| 275 | long ret = 0; |
| 276 | off_t orig_offset = kconsumerd_fd->out_fd_offset; |
| 277 | int fd = kconsumerd_fd->consumerd_fd; |
| 278 | int outfd = kconsumerd_fd->out_fd; |
| 279 | |
| 280 | /* get the padded subbuffer size to know the padding required */ |
| 281 | ret = kernctl_get_padded_subbuf_size(fd, &padded_len); |
| 282 | if (ret != 0) { |
| 283 | ret = errno; |
| 284 | perror("kernctl_get_padded_subbuf_size"); |
| 285 | goto end; |
| 286 | } |
| 287 | padding_len = padded_len - len; |
| 288 | padding = malloc(padding_len * sizeof(char)); |
| 289 | memset(padding, '\0', padding_len); |
| 290 | |
| 291 | /* get the len of the mmap region */ |
| 292 | ret = kernctl_get_mmap_len(fd, &mmap_len); |
| 293 | if (ret != 0) { |
| 294 | ret = errno; |
| 295 | perror("kernctl_get_mmap_len"); |
| 296 | goto end; |
| 297 | } |
| 298 | |
| 299 | /* get the offset inside the fd to mmap */ |
| 300 | ret = kernctl_get_mmap_read_offset(fd, &mmap_offset); |
| 301 | if (ret != 0) { |
| 302 | ret = errno; |
| 303 | perror("kernctl_get_mmap_read_offset"); |
| 304 | goto end; |
| 305 | } |
| 306 | |
| 307 | mmap_base = mmap(NULL, mmap_len, PROT_READ, MAP_PRIVATE, fd, mmap_offset); |
| 308 | if (mmap_base == MAP_FAILED) { |
| 309 | perror("Error mmaping"); |
| 310 | ret = -1; |
| 311 | goto end; |
| 312 | } |
| 313 | |
| 314 | while (len > 0) { |
| 315 | ret = write(outfd, mmap_base, len); |
| 316 | if (ret >= len) { |
| 317 | len = 0; |
| 318 | } else if (ret < 0) { |
| 319 | ret = errno; |
| 320 | perror("Error in file write"); |
| 321 | goto end; |
| 322 | } |
| 323 | /* This won't block, but will start writeout asynchronously */ |
| 324 | sync_file_range(outfd, kconsumerd_fd->out_fd_offset, ret, |
| 325 | SYNC_FILE_RANGE_WRITE); |
| 326 | kconsumerd_fd->out_fd_offset += ret; |
| 327 | } |
| 328 | |
| 329 | /* once all the data is written, write the padding to disk */ |
| 330 | ret = write(outfd, padding, padding_len); |
| 331 | if (ret < 0) { |
| 332 | ret = errno; |
| 333 | perror("Error writing padding to file"); |
| 334 | goto end; |
| 335 | } |
| 336 | |
| 337 | /* |
| 338 | * This does a blocking write-and-wait on any page that belongs to the |
| 339 | * subbuffer prior to the one we just wrote. |
| 340 | * Don't care about error values, as these are just hints and ways to |
| 341 | * limit the amount of page cache used. |
| 342 | */ |
| 343 | if (orig_offset >= kconsumerd_fd->max_sb_size) { |
| 344 | sync_file_range(outfd, orig_offset - kconsumerd_fd->max_sb_size, |
| 345 | kconsumerd_fd->max_sb_size, |
| 346 | SYNC_FILE_RANGE_WAIT_BEFORE |
| 347 | | SYNC_FILE_RANGE_WRITE |
| 348 | | SYNC_FILE_RANGE_WAIT_AFTER); |
| 349 | |
| 350 | /* |
| 351 | * Give hints to the kernel about how we access the file: |
| 352 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after |
| 353 | * we write it. |
| 354 | * |
| 355 | * We need to call fadvise again after the file grows because the |
| 356 | * kernel does not seem to apply fadvise to non-existing parts of the |
| 357 | * file. |
| 358 | * |
| 359 | * Call fadvise _after_ having waited for the page writeback to |
| 360 | * complete because the dirty page writeback semantic is not well |
| 361 | * defined. So it can be expected to lead to lower throughput in |
| 362 | * streaming. |
| 363 | */ |
| 364 | posix_fadvise(outfd, orig_offset - kconsumerd_fd->max_sb_size, |
| 365 | kconsumerd_fd->max_sb_size, POSIX_FADV_DONTNEED); |
| 366 | } |
| 367 | goto end; |
| 368 | |
| 369 | end: |
| 370 | if (padding != NULL) { |
| 371 | free(padding); |
| 372 | } |
| 373 | return ret; |
| 374 | } |
| 375 | |
| 376 | /* |
| 377 | * kconsumerd_on_read_subbuffer |
| 378 | * |
| 379 | * Splice the data from the ring buffer to the tracefile. |
| 380 | * Returns the number of bytes spliced |
| 381 | */ |
| 382 | static int kconsumerd_on_read_subbuffer( |
| 383 | struct kconsumerd_fd *kconsumerd_fd, unsigned long len) |
| 384 | { |
| 385 | long ret = 0; |
| 386 | loff_t offset = 0; |
| 387 | off_t orig_offset = kconsumerd_fd->out_fd_offset; |
| 388 | int fd = kconsumerd_fd->consumerd_fd; |
| 389 | int outfd = kconsumerd_fd->out_fd; |
| 390 | |
| 391 | while (len > 0) { |
| 392 | DBG("splice chan to pipe offset %lu (fd : %d)", |
| 393 | (unsigned long)offset, fd); |
| 394 | ret = splice(fd, &offset, kconsumerd_thread_pipe[1], NULL, len, |
| 395 | SPLICE_F_MOVE | SPLICE_F_MORE); |
| 396 | DBG("splice chan to pipe ret %ld", ret); |
| 397 | if (ret < 0) { |
| 398 | ret = errno; |
| 399 | perror("Error in relay splice"); |
| 400 | goto splice_error; |
| 401 | } |
| 402 | |
| 403 | ret = splice(kconsumerd_thread_pipe[0], NULL, outfd, NULL, ret, |
| 404 | SPLICE_F_MOVE | SPLICE_F_MORE); |
| 405 | DBG("splice pipe to file %ld", ret); |
| 406 | if (ret < 0) { |
| 407 | ret = errno; |
| 408 | perror("Error in file splice"); |
| 409 | goto splice_error; |
| 410 | } |
| 411 | if (ret >= len) { |
| 412 | len = 0; |
| 413 | } |
| 414 | /* This won't block, but will start writeout asynchronously */ |
| 415 | sync_file_range(outfd, kconsumerd_fd->out_fd_offset, ret, |
| 416 | SYNC_FILE_RANGE_WRITE); |
| 417 | kconsumerd_fd->out_fd_offset += ret; |
| 418 | } |
| 419 | |
| 420 | /* |
| 421 | * This does a blocking write-and-wait on any page that belongs to the |
| 422 | * subbuffer prior to the one we just wrote. |
| 423 | * Don't care about error values, as these are just hints and ways to |
| 424 | * limit the amount of page cache used. |
| 425 | */ |
| 426 | if (orig_offset >= kconsumerd_fd->max_sb_size) { |
| 427 | sync_file_range(outfd, orig_offset - kconsumerd_fd->max_sb_size, |
| 428 | kconsumerd_fd->max_sb_size, |
| 429 | SYNC_FILE_RANGE_WAIT_BEFORE |
| 430 | | SYNC_FILE_RANGE_WRITE |
| 431 | | SYNC_FILE_RANGE_WAIT_AFTER); |
| 432 | /* |
| 433 | * Give hints to the kernel about how we access the file: |
| 434 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after |
| 435 | * we write it. |
| 436 | * |
| 437 | * We need to call fadvise again after the file grows because the |
| 438 | * kernel does not seem to apply fadvise to non-existing parts of the |
| 439 | * file. |
| 440 | * |
| 441 | * Call fadvise _after_ having waited for the page writeback to |
| 442 | * complete because the dirty page writeback semantic is not well |
| 443 | * defined. So it can be expected to lead to lower throughput in |
| 444 | * streaming. |
| 445 | */ |
| 446 | posix_fadvise(outfd, orig_offset - kconsumerd_fd->max_sb_size, |
| 447 | kconsumerd_fd->max_sb_size, POSIX_FADV_DONTNEED); |
| 448 | } |
| 449 | goto end; |
| 450 | |
| 451 | splice_error: |
| 452 | /* send the appropriate error description to sessiond */ |
| 453 | switch(ret) { |
| 454 | case EBADF: |
| 455 | kconsumerd_send_error(KCONSUMERD_SPLICE_EBADF); |
| 456 | break; |
| 457 | case EINVAL: |
| 458 | kconsumerd_send_error(KCONSUMERD_SPLICE_EINVAL); |
| 459 | break; |
| 460 | case ENOMEM: |
| 461 | kconsumerd_send_error(KCONSUMERD_SPLICE_ENOMEM); |
| 462 | break; |
| 463 | case ESPIPE: |
| 464 | kconsumerd_send_error(KCONSUMERD_SPLICE_ESPIPE); |
| 465 | break; |
| 466 | } |
| 467 | |
| 468 | end: |
| 469 | return ret; |
| 470 | } |
| 471 | |
| 472 | /* |
| 473 | * kconsumerd_read_subbuffer |
| 474 | * |
| 475 | * Consume data on a file descriptor and write it on a trace file |
| 476 | */ |
| 477 | static int kconsumerd_read_subbuffer(struct kconsumerd_fd *kconsumerd_fd) |
| 478 | { |
| 479 | unsigned long len; |
| 480 | int err; |
| 481 | long ret = 0; |
| 482 | int infd = kconsumerd_fd->consumerd_fd; |
| 483 | |
| 484 | DBG("In kconsumerd_read_subbuffer (infd : %d)", infd); |
| 485 | /* Get the next subbuffer */ |
| 486 | err = kernctl_get_next_subbuf(infd); |
| 487 | if (err != 0) { |
| 488 | ret = errno; |
| 489 | perror("Reserving sub buffer failed (everything is normal, " |
| 490 | "it is due to concurrency)"); |
| 491 | goto end; |
| 492 | } |
| 493 | |
| 494 | switch (DEFAULT_KERNEL_CHANNEL_OUTPUT) { |
| 495 | case LTTNG_EVENT_SPLICE: |
| 496 | /* read the whole subbuffer */ |
| 497 | err = kernctl_get_padded_subbuf_size(infd, &len); |
| 498 | if (err != 0) { |
| 499 | ret = errno; |
| 500 | perror("Getting sub-buffer len failed."); |
| 501 | goto end; |
| 502 | } |
| 503 | |
| 504 | /* splice the subbuffer to the tracefile */ |
| 505 | ret = kconsumerd_on_read_subbuffer(kconsumerd_fd, len); |
| 506 | if (ret < 0) { |
| 507 | /* |
| 508 | * display the error but continue processing to try |
| 509 | * to release the subbuffer |
| 510 | */ |
| 511 | ERR("Error splicing to tracefile"); |
| 512 | } |
| 513 | break; |
| 514 | case LTTNG_EVENT_MMAP: |
| 515 | /* read the used subbuffer size */ |
| 516 | err = kernctl_get_subbuf_size(infd, &len); |
| 517 | if (err != 0) { |
| 518 | ret = errno; |
| 519 | perror("Getting sub-buffer len failed."); |
| 520 | goto end; |
| 521 | } |
| 522 | /* write the subbuffer to the tracefile */ |
| 523 | ret = kconsumerd_on_read_subbuffer_mmap(kconsumerd_fd, len); |
| 524 | if (ret < 0) { |
| 525 | /* |
| 526 | * display the error but continue processing to try |
| 527 | * to release the subbuffer |
| 528 | */ |
| 529 | ERR("Error writing to tracefile"); |
| 530 | } |
| 531 | break; |
| 532 | default: |
| 533 | ERR("Unknown output method"); |
| 534 | ret = -1; |
| 535 | } |
| 536 | |
| 537 | err = kernctl_put_next_subbuf(infd); |
| 538 | if (err != 0) { |
| 539 | ret = errno; |
| 540 | if (errno == EFAULT) { |
| 541 | perror("Error in unreserving sub buffer\n"); |
| 542 | } else if (errno == EIO) { |
| 543 | /* Should never happen with newer LTTng versions */ |
| 544 | perror("Reader has been pushed by the writer, last sub-buffer corrupted."); |
| 545 | } |
| 546 | goto end; |
| 547 | } |
| 548 | |
| 549 | end: |
| 550 | return ret; |
| 551 | } |
| 552 | |
| 553 | /* |
| 554 | * kconsumerd_poll_socket |
| 555 | * |
| 556 | * Poll on the should_quit pipe and the command socket |
| 557 | * return -1 on error and should exit, 0 if data is |
| 558 | * available on the command socket |
| 559 | */ |
| 560 | int kconsumerd_poll_socket(struct pollfd *kconsumerd_sockpoll) |
| 561 | { |
| 562 | int num_rdy; |
| 563 | |
| 564 | num_rdy = poll(kconsumerd_sockpoll, 2, -1); |
| 565 | if (num_rdy == -1) { |
| 566 | perror("Poll error"); |
| 567 | goto exit; |
| 568 | } |
| 569 | if (kconsumerd_sockpoll[0].revents == POLLIN) { |
| 570 | DBG("kconsumerd_should_quit wake up"); |
| 571 | goto exit; |
| 572 | } |
| 573 | return 0; |
| 574 | |
| 575 | exit: |
| 576 | return -1; |
| 577 | } |
| 578 | |
| 579 | /* |
| 580 | * kconsumerd_consumerd_recv_fd |
| 581 | * |
| 582 | * Receives an array of file descriptors and the associated |
| 583 | * structures describing each fd (path name). |
| 584 | * Returns the size of received data |
| 585 | */ |
| 586 | static int kconsumerd_consumerd_recv_fd(int sfd, |
| 587 | struct pollfd *kconsumerd_sockpoll, int size, |
| 588 | enum kconsumerd_command cmd_type) |
| 589 | { |
| 590 | struct iovec iov[1]; |
| 591 | int ret = 0, i, tmp2; |
| 592 | struct cmsghdr *cmsg; |
| 593 | int nb_fd; |
| 594 | char recv_fd[CMSG_SPACE(sizeof(int))]; |
| 595 | struct lttcomm_kconsumerd_msg lkm; |
| 596 | |
| 597 | /* the number of fds we are about to receive */ |
| 598 | nb_fd = size / sizeof(struct lttcomm_kconsumerd_msg); |
| 599 | |
| 600 | /* |
| 601 | * nb_fd is the number of fds we receive. One fd per recvmsg. |
| 602 | */ |
| 603 | for (i = 0; i < nb_fd; i++) { |
| 604 | struct msghdr msg = { 0 }; |
| 605 | |
| 606 | /* Prepare to receive the structures */ |
| 607 | iov[0].iov_base = &lkm; |
| 608 | iov[0].iov_len = sizeof(lkm); |
| 609 | msg.msg_iov = iov; |
| 610 | msg.msg_iovlen = 1; |
| 611 | |
| 612 | msg.msg_control = recv_fd; |
| 613 | msg.msg_controllen = sizeof(recv_fd); |
| 614 | |
| 615 | DBG("Waiting to receive fd"); |
| 616 | if (kconsumerd_poll_socket(kconsumerd_sockpoll) < 0) { |
| 617 | goto end; |
| 618 | } |
| 619 | |
| 620 | if ((ret = recvmsg(sfd, &msg, 0)) < 0) { |
| 621 | perror("recvmsg"); |
| 622 | continue; |
| 623 | } |
| 624 | |
| 625 | if (ret != (size / nb_fd)) { |
| 626 | ERR("Received only %d, expected %d", ret, size); |
| 627 | kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD); |
| 628 | goto end; |
| 629 | } |
| 630 | |
| 631 | cmsg = CMSG_FIRSTHDR(&msg); |
| 632 | if (!cmsg) { |
| 633 | ERR("Invalid control message header"); |
| 634 | ret = -1; |
| 635 | kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD); |
| 636 | goto end; |
| 637 | } |
| 638 | |
| 639 | /* if we received fds */ |
| 640 | if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) { |
| 641 | switch (cmd_type) { |
| 642 | case ADD_STREAM: |
| 643 | DBG("kconsumerd_add_fd %s (%d)", lkm.path_name, ((int *) CMSG_DATA(cmsg))[0]); |
| 644 | ret = kconsumerd_add_fd(&lkm, ((int *) CMSG_DATA(cmsg))[0]); |
| 645 | if (ret < 0) { |
| 646 | kconsumerd_send_error(KCONSUMERD_OUTFD_ERROR); |
| 647 | goto end; |
| 648 | } |
| 649 | break; |
| 650 | case UPDATE_STREAM: |
| 651 | kconsumerd_change_fd_state(lkm.fd, lkm.state); |
| 652 | break; |
| 653 | default: |
| 654 | break; |
| 655 | } |
| 656 | /* signal the poll thread */ |
| 657 | tmp2 = write(kconsumerd_poll_pipe[1], "4", 1); |
| 658 | if (tmp2 < 0) { |
| 659 | perror("write kconsumerd poll"); |
| 660 | } |
| 661 | } else { |
| 662 | ERR("Didn't received any fd"); |
| 663 | kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD); |
| 664 | ret = -1; |
| 665 | goto end; |
| 666 | } |
| 667 | } |
| 668 | |
| 669 | end: |
| 670 | return ret; |
| 671 | } |
| 672 | |
| 673 | /* |
| 674 | * kconsumerd_thread_poll_fds |
| 675 | * |
| 676 | * This thread polls the fds in the ltt_fd_list to consume the data |
| 677 | * and write it to tracefile if necessary. |
| 678 | */ |
| 679 | void *kconsumerd_thread_poll_fds(void *data) |
| 680 | { |
| 681 | int num_rdy, num_hup, high_prio, ret, i; |
| 682 | struct pollfd *pollfd = NULL; |
| 683 | /* local view of the fds */ |
| 684 | struct kconsumerd_fd **local_kconsumerd_fd = NULL; |
| 685 | /* local view of kconsumerd_data.fds_count */ |
| 686 | int nb_fd = 0; |
| 687 | char tmp; |
| 688 | int tmp2; |
| 689 | |
| 690 | ret = pipe(kconsumerd_thread_pipe); |
| 691 | if (ret < 0) { |
| 692 | perror("Error creating pipe"); |
| 693 | goto end; |
| 694 | } |
| 695 | |
| 696 | local_kconsumerd_fd = malloc(sizeof(struct kconsumerd_fd)); |
| 697 | |
| 698 | while (1) { |
| 699 | high_prio = 0; |
| 700 | num_hup = 0; |
| 701 | |
| 702 | /* |
| 703 | * the ltt_fd_list has been updated, we need to update our |
| 704 | * local array as well |
| 705 | */ |
| 706 | pthread_mutex_lock(&kconsumerd_data.lock); |
| 707 | if (kconsumerd_data.need_update) { |
| 708 | if (pollfd != NULL) { |
| 709 | free(pollfd); |
| 710 | pollfd = NULL; |
| 711 | } |
| 712 | if (local_kconsumerd_fd != NULL) { |
| 713 | free(local_kconsumerd_fd); |
| 714 | local_kconsumerd_fd = NULL; |
| 715 | } |
| 716 | |
| 717 | /* allocate for all fds + 1 for the kconsumerd_poll_pipe */ |
| 718 | pollfd = malloc((kconsumerd_data.fds_count + 1) * sizeof(struct pollfd)); |
| 719 | if (pollfd == NULL) { |
| 720 | perror("pollfd malloc"); |
| 721 | pthread_mutex_unlock(&kconsumerd_data.lock); |
| 722 | goto end; |
| 723 | } |
| 724 | |
| 725 | /* allocate for all fds + 1 for the kconsumerd_poll_pipe */ |
| 726 | local_kconsumerd_fd = malloc((kconsumerd_data.fds_count + 1) * |
| 727 | sizeof(struct kconsumerd_fd)); |
| 728 | if (local_kconsumerd_fd == NULL) { |
| 729 | perror("local_kconsumerd_fd malloc"); |
| 730 | pthread_mutex_unlock(&kconsumerd_data.lock); |
| 731 | goto end; |
| 732 | } |
| 733 | ret = kconsumerd_update_poll_array(&pollfd, local_kconsumerd_fd); |
| 734 | if (ret < 0) { |
| 735 | ERR("Error in allocating pollfd or local_outfds"); |
| 736 | kconsumerd_send_error(KCONSUMERD_POLL_ERROR); |
| 737 | pthread_mutex_unlock(&kconsumerd_data.lock); |
| 738 | goto end; |
| 739 | } |
| 740 | nb_fd = ret; |
| 741 | kconsumerd_data.need_update = 0; |
| 742 | } |
| 743 | pthread_mutex_unlock(&kconsumerd_data.lock); |
| 744 | |
| 745 | /* poll on the array of fds */ |
| 746 | DBG("polling on %d fd", nb_fd + 1); |
| 747 | num_rdy = poll(pollfd, nb_fd + 1, kconsumerd_poll_timeout); |
| 748 | DBG("poll num_rdy : %d", num_rdy); |
| 749 | if (num_rdy == -1) { |
| 750 | perror("Poll error"); |
| 751 | kconsumerd_send_error(KCONSUMERD_POLL_ERROR); |
| 752 | goto end; |
| 753 | } else if (num_rdy == 0) { |
| 754 | DBG("Polling thread timed out"); |
| 755 | goto end; |
| 756 | } |
| 757 | |
| 758 | /* No FDs and kconsumerd_quit, kconsumerd_cleanup the thread */ |
| 759 | if (nb_fd == 0 && kconsumerd_quit == 1) { |
| 760 | goto end; |
| 761 | } |
| 762 | |
| 763 | /* |
| 764 | * If the kconsumerd_poll_pipe triggered poll go |
| 765 | * directly to the beginning of the loop to update the |
| 766 | * array. We want to prioritize array update over |
| 767 | * low-priority reads. |
| 768 | */ |
| 769 | if (pollfd[nb_fd].revents == POLLIN) { |
| 770 | DBG("kconsumerd_poll_pipe wake up"); |
| 771 | tmp2 = read(kconsumerd_poll_pipe[0], &tmp, 1); |
| 772 | if (tmp2 < 0) { |
| 773 | perror("read kconsumerd poll"); |
| 774 | } |
| 775 | continue; |
| 776 | } |
| 777 | |
| 778 | /* Take care of high priority channels first. */ |
| 779 | for (i = 0; i < nb_fd; i++) { |
| 780 | switch(pollfd[i].revents) { |
| 781 | case POLLERR: |
| 782 | ERR("Error returned in polling fd %d.", pollfd[i].fd); |
| 783 | kconsumerd_del_fd(local_kconsumerd_fd[i]); |
| 784 | num_hup++; |
| 785 | break; |
| 786 | case POLLHUP: |
| 787 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); |
| 788 | kconsumerd_del_fd(local_kconsumerd_fd[i]); |
| 789 | num_hup++; |
| 790 | break; |
| 791 | case POLLNVAL: |
| 792 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); |
| 793 | kconsumerd_del_fd(local_kconsumerd_fd[i]); |
| 794 | num_hup++; |
| 795 | break; |
| 796 | case POLLPRI: |
| 797 | DBG("Urgent read on fd %d", pollfd[i].fd); |
| 798 | high_prio = 1; |
| 799 | ret = kconsumerd_read_subbuffer(local_kconsumerd_fd[i]); |
| 800 | /* it's ok to have an unavailable sub-buffer */ |
| 801 | if (ret == EAGAIN) { |
| 802 | ret = 0; |
| 803 | } |
| 804 | break; |
| 805 | } |
| 806 | } |
| 807 | |
| 808 | /* If every buffer FD has hung up, we end the read loop here */ |
| 809 | if (nb_fd > 0 && num_hup == nb_fd) { |
| 810 | DBG("every buffer FD has hung up\n"); |
| 811 | if (kconsumerd_quit == 1) { |
| 812 | goto end; |
| 813 | } |
| 814 | continue; |
| 815 | } |
| 816 | |
| 817 | /* Take care of low priority channels. */ |
| 818 | if (high_prio == 0) { |
| 819 | for (i = 0; i < nb_fd; i++) { |
| 820 | if (pollfd[i].revents == POLLIN) { |
| 821 | DBG("Normal read on fd %d", pollfd[i].fd); |
| 822 | ret = kconsumerd_read_subbuffer(local_kconsumerd_fd[i]); |
| 823 | /* it's ok to have an unavailable subbuffer */ |
| 824 | if (ret == EAGAIN) { |
| 825 | ret = 0; |
| 826 | } |
| 827 | } |
| 828 | } |
| 829 | } |
| 830 | } |
| 831 | end: |
| 832 | DBG("polling thread exiting"); |
| 833 | if (pollfd != NULL) { |
| 834 | free(pollfd); |
| 835 | pollfd = NULL; |
| 836 | } |
| 837 | if (local_kconsumerd_fd != NULL) { |
| 838 | free(local_kconsumerd_fd); |
| 839 | local_kconsumerd_fd = NULL; |
| 840 | } |
| 841 | return NULL; |
| 842 | } |
| 843 | |
| 844 | /* |
| 845 | * kconsumerd_init(void) |
| 846 | * |
| 847 | * initialise the necessary environnement : |
| 848 | * - inform the polling thread to update the polling array |
| 849 | * - create the poll_pipe |
| 850 | * - create the should_quit pipe (for signal handler) |
| 851 | */ |
| 852 | int kconsumerd_init(void) |
| 853 | { |
| 854 | int ret; |
| 855 | |
| 856 | /* need to update the polling array at init time */ |
| 857 | kconsumerd_data.need_update = 1; |
| 858 | |
| 859 | ret = pipe(kconsumerd_poll_pipe); |
| 860 | if (ret < 0) { |
| 861 | perror("Error creating poll pipe"); |
| 862 | goto end; |
| 863 | } |
| 864 | |
| 865 | ret = pipe(kconsumerd_should_quit); |
| 866 | if (ret < 0) { |
| 867 | perror("Error creating recv pipe"); |
| 868 | goto end; |
| 869 | } |
| 870 | |
| 871 | end: |
| 872 | return ret; |
| 873 | } |
| 874 | |
| 875 | /* |
| 876 | * kconsumerd_thread_receive_fds |
| 877 | * |
| 878 | * This thread listens on the consumerd socket and |
| 879 | * receives the file descriptors from ltt-sessiond |
| 880 | */ |
| 881 | void *kconsumerd_thread_receive_fds(void *data) |
| 882 | { |
| 883 | int sock, client_socket, ret; |
| 884 | struct lttcomm_kconsumerd_header tmp; |
| 885 | /* |
| 886 | * structure to poll for incoming data on communication socket |
| 887 | * avoids making blocking sockets |
| 888 | */ |
| 889 | struct pollfd kconsumerd_sockpoll[2]; |
| 890 | |
| 891 | |
| 892 | DBG("Creating command socket %s", kconsumerd_command_sock_path); |
| 893 | unlink(kconsumerd_command_sock_path); |
| 894 | client_socket = lttcomm_create_unix_sock(kconsumerd_command_sock_path); |
| 895 | if (client_socket < 0) { |
| 896 | ERR("Cannot create command socket"); |
| 897 | goto end; |
| 898 | } |
| 899 | |
| 900 | ret = lttcomm_listen_unix_sock(client_socket); |
| 901 | if (ret < 0) { |
| 902 | goto end; |
| 903 | } |
| 904 | |
| 905 | DBG("Sending ready command to ltt-sessiond"); |
| 906 | ret = kconsumerd_send_error(KCONSUMERD_COMMAND_SOCK_READY); |
| 907 | if (ret < 0) { |
| 908 | ERR("Error sending ready command to ltt-sessiond"); |
| 909 | goto end; |
| 910 | } |
| 911 | |
| 912 | ret = fcntl(client_socket, F_SETFL, O_NONBLOCK); |
| 913 | if (ret < 0) { |
| 914 | perror("fcntl O_NONBLOCK"); |
| 915 | goto end; |
| 916 | } |
| 917 | |
| 918 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
| 919 | kconsumerd_sockpoll[0].fd = kconsumerd_should_quit[0]; |
| 920 | kconsumerd_sockpoll[0].events = POLLIN | POLLPRI; |
| 921 | kconsumerd_sockpoll[1].fd = client_socket; |
| 922 | kconsumerd_sockpoll[1].events = POLLIN | POLLPRI; |
| 923 | |
| 924 | if (kconsumerd_poll_socket(kconsumerd_sockpoll) < 0) { |
| 925 | goto end; |
| 926 | } |
| 927 | DBG("Connection on client_socket"); |
| 928 | |
| 929 | /* Blocking call, waiting for transmission */ |
| 930 | sock = lttcomm_accept_unix_sock(client_socket); |
| 931 | if (sock <= 0) { |
| 932 | WARN("On accept"); |
| 933 | goto end; |
| 934 | } |
| 935 | ret = fcntl(sock, F_SETFL, O_NONBLOCK); |
| 936 | if (ret < 0) { |
| 937 | perror("fcntl O_NONBLOCK"); |
| 938 | goto end; |
| 939 | } |
| 940 | |
| 941 | /* update the polling structure to poll on the established socket */ |
| 942 | kconsumerd_sockpoll[1].fd = sock; |
| 943 | kconsumerd_sockpoll[1].events = POLLIN | POLLPRI; |
| 944 | |
| 945 | while (1) { |
| 946 | if (kconsumerd_poll_socket(kconsumerd_sockpoll) < 0) { |
| 947 | goto end; |
| 948 | } |
| 949 | DBG("Incoming fds on sock"); |
| 950 | |
| 951 | /* We first get the number of fd we are about to receive */ |
| 952 | ret = lttcomm_recv_unix_sock(sock, &tmp, |
| 953 | sizeof(struct lttcomm_kconsumerd_header)); |
| 954 | if (ret <= 0) { |
| 955 | ERR("Communication interrupted on command socket"); |
| 956 | goto end; |
| 957 | } |
| 958 | if (tmp.cmd_type == STOP) { |
| 959 | DBG("Received STOP command"); |
| 960 | goto end; |
| 961 | } |
| 962 | if (kconsumerd_quit) { |
| 963 | DBG("kconsumerd_thread_receive_fds received quit from signal"); |
| 964 | goto end; |
| 965 | } |
| 966 | |
| 967 | /* we received a command to add or update fds */ |
| 968 | ret = kconsumerd_consumerd_recv_fd(sock, kconsumerd_sockpoll, |
| 969 | tmp.payload_size, tmp.cmd_type); |
| 970 | if (ret <= 0) { |
| 971 | ERR("Receiving the FD, exiting"); |
| 972 | goto end; |
| 973 | } |
| 974 | DBG("received fds on sock"); |
| 975 | } |
| 976 | |
| 977 | end: |
| 978 | DBG("kconsumerd_thread_receive_fds exiting"); |
| 979 | |
| 980 | /* |
| 981 | * when all fds have hung up, the polling thread |
| 982 | * can exit cleanly |
| 983 | */ |
| 984 | kconsumerd_quit = 1; |
| 985 | |
| 986 | /* |
| 987 | * 2s of grace period, if no polling events occur during |
| 988 | * this period, the polling thread will exit even if there |
| 989 | * are still open FDs (should not happen, but safety mechanism). |
| 990 | */ |
| 991 | kconsumerd_poll_timeout = KCONSUMERD_POLL_GRACE_PERIOD; |
| 992 | |
| 993 | /* wake up the polling thread */ |
| 994 | ret = write(kconsumerd_poll_pipe[1], "4", 1); |
| 995 | if (ret < 0) { |
| 996 | perror("poll pipe write"); |
| 997 | } |
| 998 | return NULL; |
| 999 | } |
| 1000 | |
| 1001 | /* |
| 1002 | * kconsumerd_cleanup |
| 1003 | * |
| 1004 | * Cleanup the daemon's socket on exit |
| 1005 | */ |
| 1006 | void kconsumerd_cleanup(void) |
| 1007 | { |
| 1008 | struct kconsumerd_fd *iter, *tmp; |
| 1009 | |
| 1010 | /* remove the socket file */ |
| 1011 | unlink(kconsumerd_command_sock_path); |
| 1012 | |
| 1013 | /* |
| 1014 | * close all outfd. Called when there are no more threads |
| 1015 | * running (after joining on the threads), no need to protect |
| 1016 | * list iteration with mutex. |
| 1017 | */ |
| 1018 | cds_list_for_each_entry_safe(iter, tmp, &kconsumerd_data.fd_list.head, list) { |
| 1019 | kconsumerd_del_fd(iter); |
| 1020 | } |
| 1021 | } |
| 1022 | |
| 1023 | /* |
| 1024 | * kconsumerd_should_exit |
| 1025 | * |
| 1026 | * Called from signal handler. |
| 1027 | */ |
| 1028 | void kconsumerd_should_exit(void) |
| 1029 | { |
| 1030 | int ret; |
| 1031 | kconsumerd_quit = 1; |
| 1032 | ret = write(kconsumerd_should_quit[1], "4", 1); |
| 1033 | if (ret < 0) { |
| 1034 | perror("write kconsumerd quit"); |
| 1035 | } |
| 1036 | } |
| 1037 | |
| 1038 | /* |
| 1039 | * kconsumerd_send_error |
| 1040 | * |
| 1041 | * send return code to ltt-sessiond |
| 1042 | */ |
| 1043 | int kconsumerd_send_error(enum lttcomm_return_code cmd) |
| 1044 | { |
| 1045 | if (kconsumerd_error_socket > 0) { |
| 1046 | return lttcomm_send_unix_sock(kconsumerd_error_socket, &cmd, |
| 1047 | sizeof(enum lttcomm_sessiond_command)); |
| 1048 | } |
| 1049 | |
| 1050 | return 0; |
| 1051 | } |