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