| 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 modify |
| 6 | * it under the terms of the GNU General Public License, version 2 only, |
| 7 | * as published by the Free Software Foundation. |
| 8 | * |
| 9 | * This program is distributed in the hope that it will be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
| 13 | * |
| 14 | * You should have received a copy of the GNU General Public License along |
| 15 | * with this program; if not, write to the Free Software Foundation, Inc., |
| 16 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
| 17 | */ |
| 18 | |
| 19 | #define _GNU_SOURCE |
| 20 | #include <assert.h> |
| 21 | #include <poll.h> |
| 22 | #include <pthread.h> |
| 23 | #include <stdlib.h> |
| 24 | #include <string.h> |
| 25 | #include <sys/mman.h> |
| 26 | #include <sys/socket.h> |
| 27 | #include <sys/types.h> |
| 28 | #include <inttypes.h> |
| 29 | #include <unistd.h> |
| 30 | #include <sys/stat.h> |
| 31 | |
| 32 | #include <common/common.h> |
| 33 | #include <common/kernel-ctl/kernel-ctl.h> |
| 34 | #include <common/sessiond-comm/sessiond-comm.h> |
| 35 | #include <common/sessiond-comm/relayd.h> |
| 36 | #include <common/compat/fcntl.h> |
| 37 | #include <common/pipe.h> |
| 38 | #include <common/relayd/relayd.h> |
| 39 | #include <common/utils.h> |
| 40 | #include <common/consumer-stream.h> |
| 41 | |
| 42 | #include "kernel-consumer.h" |
| 43 | |
| 44 | extern struct lttng_consumer_global_data consumer_data; |
| 45 | extern int consumer_poll_timeout; |
| 46 | extern volatile int consumer_quit; |
| 47 | |
| 48 | /* |
| 49 | * Take a snapshot for a specific fd |
| 50 | * |
| 51 | * Returns 0 on success, < 0 on error |
| 52 | */ |
| 53 | int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream) |
| 54 | { |
| 55 | int ret = 0; |
| 56 | int infd = stream->wait_fd; |
| 57 | |
| 58 | ret = kernctl_snapshot(infd); |
| 59 | if (ret != 0) { |
| 60 | errno = -ret; |
| 61 | perror("Getting sub-buffer snapshot."); |
| 62 | } |
| 63 | |
| 64 | return ret; |
| 65 | } |
| 66 | |
| 67 | /* |
| 68 | * Get the produced position |
| 69 | * |
| 70 | * Returns 0 on success, < 0 on error |
| 71 | */ |
| 72 | int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
| 73 | unsigned long *pos) |
| 74 | { |
| 75 | int ret; |
| 76 | int infd = stream->wait_fd; |
| 77 | |
| 78 | ret = kernctl_snapshot_get_produced(infd, pos); |
| 79 | if (ret != 0) { |
| 80 | errno = -ret; |
| 81 | perror("kernctl_snapshot_get_produced"); |
| 82 | } |
| 83 | |
| 84 | return ret; |
| 85 | } |
| 86 | |
| 87 | /* |
| 88 | * Get the consumerd position |
| 89 | * |
| 90 | * Returns 0 on success, < 0 on error |
| 91 | */ |
| 92 | int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, |
| 93 | unsigned long *pos) |
| 94 | { |
| 95 | int ret; |
| 96 | int infd = stream->wait_fd; |
| 97 | |
| 98 | ret = kernctl_snapshot_get_consumed(infd, pos); |
| 99 | if (ret != 0) { |
| 100 | errno = -ret; |
| 101 | perror("kernctl_snapshot_get_consumed"); |
| 102 | } |
| 103 | |
| 104 | return ret; |
| 105 | } |
| 106 | |
| 107 | /* |
| 108 | * Find a relayd and send the stream |
| 109 | * |
| 110 | * Returns 0 on success, < 0 on error |
| 111 | */ |
| 112 | static |
| 113 | int send_relayd_stream(struct lttng_consumer_stream *stream, char *path) |
| 114 | { |
| 115 | struct consumer_relayd_sock_pair *relayd; |
| 116 | int ret = 0; |
| 117 | char *stream_path; |
| 118 | |
| 119 | if (path != NULL) { |
| 120 | stream_path = path; |
| 121 | } else { |
| 122 | stream_path = stream->chan->pathname; |
| 123 | } |
| 124 | /* The stream is not metadata. Get relayd reference if exists. */ |
| 125 | rcu_read_lock(); |
| 126 | relayd = consumer_find_relayd(stream->net_seq_idx); |
| 127 | if (relayd != NULL) { |
| 128 | /* Add stream on the relayd */ |
| 129 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 130 | ret = relayd_add_stream(&relayd->control_sock, |
| 131 | stream->name, stream_path, |
| 132 | &stream->relayd_stream_id, |
| 133 | stream->chan->tracefile_size, |
| 134 | stream->chan->tracefile_count); |
| 135 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 136 | if (ret < 0) { |
| 137 | goto end; |
| 138 | } |
| 139 | uatomic_inc(&relayd->refcount); |
| 140 | } else if (stream->net_seq_idx != (uint64_t) -1ULL) { |
| 141 | ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.", |
| 142 | stream->net_seq_idx); |
| 143 | ret = -1; |
| 144 | goto end; |
| 145 | } |
| 146 | |
| 147 | end: |
| 148 | rcu_read_unlock(); |
| 149 | return ret; |
| 150 | } |
| 151 | |
| 152 | /* |
| 153 | * Find a relayd and close the stream |
| 154 | */ |
| 155 | static |
| 156 | void close_relayd_stream(struct lttng_consumer_stream *stream) |
| 157 | { |
| 158 | struct consumer_relayd_sock_pair *relayd; |
| 159 | |
| 160 | /* The stream is not metadata. Get relayd reference if exists. */ |
| 161 | rcu_read_lock(); |
| 162 | relayd = consumer_find_relayd(stream->net_seq_idx); |
| 163 | if (relayd != NULL) { |
| 164 | consumer_stream_relayd_close(stream, relayd); |
| 165 | } |
| 166 | rcu_read_unlock(); |
| 167 | } |
| 168 | |
| 169 | /* |
| 170 | * Take a snapshot of all the stream of a channel |
| 171 | * |
| 172 | * Returns 0 on success, < 0 on error |
| 173 | */ |
| 174 | int lttng_kconsumer_snapshot_channel(uint64_t key, char *path, |
| 175 | uint64_t relayd_id, struct lttng_consumer_local_data *ctx) |
| 176 | { |
| 177 | int ret; |
| 178 | unsigned long consumed_pos, produced_pos; |
| 179 | struct lttng_consumer_channel *channel; |
| 180 | struct lttng_consumer_stream *stream; |
| 181 | |
| 182 | DBG("Kernel consumer snapshot channel %lu", key); |
| 183 | |
| 184 | rcu_read_lock(); |
| 185 | |
| 186 | channel = consumer_find_channel(key); |
| 187 | if (!channel) { |
| 188 | ERR("No channel found for key %lu", key); |
| 189 | ret = -1; |
| 190 | goto end; |
| 191 | } |
| 192 | |
| 193 | /* Splice is not supported yet for channel snapshot. */ |
| 194 | if (channel->output != CONSUMER_CHANNEL_MMAP) { |
| 195 | ERR("Unsupported output %d", channel->output); |
| 196 | ret = -1; |
| 197 | goto end; |
| 198 | } |
| 199 | |
| 200 | cds_list_for_each_entry(stream, &channel->stream_no_monitor_list.head, |
| 201 | no_monitor_node) { |
| 202 | /* |
| 203 | * Lock stream because we are about to change its state. |
| 204 | */ |
| 205 | pthread_mutex_lock(&stream->lock); |
| 206 | |
| 207 | stream->net_seq_idx = relayd_id; |
| 208 | channel->relayd_id = relayd_id; |
| 209 | if (relayd_id != (uint64_t) -1ULL) { |
| 210 | ret = send_relayd_stream(stream, path); |
| 211 | if (ret < 0) { |
| 212 | ERR("sending stream to relayd"); |
| 213 | goto end_unlock; |
| 214 | } |
| 215 | DBG("Stream %s sent to the relayd", stream->name); |
| 216 | } else { |
| 217 | ret = utils_create_stream_file(path, stream->name, |
| 218 | stream->chan->tracefile_size, stream->tracefile_count_current, |
| 219 | stream->uid, stream->gid); |
| 220 | if (ret < 0) { |
| 221 | ERR("utils_create_stream_file"); |
| 222 | goto end_unlock; |
| 223 | } |
| 224 | |
| 225 | stream->out_fd = ret; |
| 226 | stream->tracefile_size_current = 0; |
| 227 | |
| 228 | DBG("Kernel consumer snapshot stream %s/%s (%lu)", path, |
| 229 | stream->name, stream->key); |
| 230 | } |
| 231 | |
| 232 | ret = kernctl_buffer_flush(stream->wait_fd); |
| 233 | if (ret < 0) { |
| 234 | ERR("Failed to flush kernel metadata stream"); |
| 235 | goto end_unlock; |
| 236 | } |
| 237 | |
| 238 | ret = lttng_kconsumer_take_snapshot(stream); |
| 239 | if (ret < 0) { |
| 240 | ERR("Taking kernel snapshot"); |
| 241 | goto end_unlock; |
| 242 | } |
| 243 | |
| 244 | ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos); |
| 245 | if (ret < 0) { |
| 246 | ERR("Produced kernel snapshot position"); |
| 247 | goto end_unlock; |
| 248 | } |
| 249 | |
| 250 | ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos); |
| 251 | if (ret < 0) { |
| 252 | ERR("Consumerd kernel snapshot position"); |
| 253 | goto end_unlock; |
| 254 | } |
| 255 | |
| 256 | if (stream->max_sb_size == 0) { |
| 257 | ret = kernctl_get_max_subbuf_size(stream->wait_fd, |
| 258 | &stream->max_sb_size); |
| 259 | if (ret < 0) { |
| 260 | ERR("Getting kernel max_sb_size"); |
| 261 | goto end_unlock; |
| 262 | } |
| 263 | } |
| 264 | |
| 265 | while (consumed_pos < produced_pos) { |
| 266 | ssize_t read_len; |
| 267 | unsigned long len, padded_len; |
| 268 | |
| 269 | DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos); |
| 270 | |
| 271 | ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos); |
| 272 | if (ret < 0) { |
| 273 | if (errno != EAGAIN) { |
| 274 | PERROR("kernctl_get_subbuf snapshot"); |
| 275 | goto end_unlock; |
| 276 | } |
| 277 | DBG("Kernel consumer get subbuf failed. Skipping it."); |
| 278 | consumed_pos += stream->max_sb_size; |
| 279 | continue; |
| 280 | } |
| 281 | |
| 282 | ret = kernctl_get_subbuf_size(stream->wait_fd, &len); |
| 283 | if (ret < 0) { |
| 284 | ERR("Snapshot kernctl_get_subbuf_size"); |
| 285 | goto end_unlock; |
| 286 | } |
| 287 | |
| 288 | ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len); |
| 289 | if (ret < 0) { |
| 290 | ERR("Snapshot kernctl_get_padded_subbuf_size"); |
| 291 | goto end_unlock; |
| 292 | } |
| 293 | |
| 294 | read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len, |
| 295 | padded_len - len); |
| 296 | /* |
| 297 | * We write the padded len in local tracefiles but the |
| 298 | * data len when using a relay. |
| 299 | * Display the error but continue processing to try to |
| 300 | * release the subbuffer. |
| 301 | */ |
| 302 | if (relayd_id != (uint64_t) -1ULL) { |
| 303 | if (read_len != len) { |
| 304 | ERR("Error sending to the relay (ret: %zd != len: %lu)", |
| 305 | read_len, len); |
| 306 | } |
| 307 | } else { |
| 308 | if (read_len != padded_len) { |
| 309 | ERR("Error writing to tracefile (ret: %zd != len: %lu)", |
| 310 | read_len, padded_len); |
| 311 | } |
| 312 | } |
| 313 | |
| 314 | ret = kernctl_put_subbuf(stream->wait_fd); |
| 315 | if (ret < 0) { |
| 316 | ERR("Snapshot kernctl_put_subbuf"); |
| 317 | goto end_unlock; |
| 318 | } |
| 319 | consumed_pos += stream->max_sb_size; |
| 320 | } |
| 321 | |
| 322 | if (relayd_id == (uint64_t) -1ULL) { |
| 323 | ret = close(stream->out_fd); |
| 324 | if (ret < 0) { |
| 325 | PERROR("Kernel consumer snapshot close out_fd"); |
| 326 | goto end_unlock; |
| 327 | } |
| 328 | stream->out_fd = -1; |
| 329 | } else { |
| 330 | close_relayd_stream(stream); |
| 331 | stream->net_seq_idx = (uint64_t) -1ULL; |
| 332 | } |
| 333 | pthread_mutex_unlock(&stream->lock); |
| 334 | } |
| 335 | |
| 336 | /* All good! */ |
| 337 | ret = 0; |
| 338 | goto end; |
| 339 | |
| 340 | end_unlock: |
| 341 | pthread_mutex_unlock(&stream->lock); |
| 342 | end: |
| 343 | rcu_read_unlock(); |
| 344 | return ret; |
| 345 | } |
| 346 | |
| 347 | /* |
| 348 | * Read the whole metadata available for a snapshot. |
| 349 | * |
| 350 | * Returns 0 on success, < 0 on error |
| 351 | */ |
| 352 | int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path, |
| 353 | uint64_t relayd_id, struct lttng_consumer_local_data *ctx) |
| 354 | { |
| 355 | struct lttng_consumer_channel *metadata_channel; |
| 356 | struct lttng_consumer_stream *metadata_stream; |
| 357 | int ret; |
| 358 | |
| 359 | DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s", |
| 360 | key, path); |
| 361 | |
| 362 | rcu_read_lock(); |
| 363 | |
| 364 | metadata_channel = consumer_find_channel(key); |
| 365 | if (!metadata_channel) { |
| 366 | ERR("Snapshot kernel metadata channel not found for key %lu", key); |
| 367 | ret = -1; |
| 368 | goto end; |
| 369 | } |
| 370 | |
| 371 | metadata_stream = metadata_channel->metadata_stream; |
| 372 | assert(metadata_stream); |
| 373 | |
| 374 | if (relayd_id != (uint64_t) -1ULL) { |
| 375 | ret = send_relayd_stream(metadata_stream, path); |
| 376 | if (ret < 0) { |
| 377 | ERR("sending stream to relayd"); |
| 378 | } |
| 379 | DBG("Stream %s sent to the relayd", metadata_stream->name); |
| 380 | } else { |
| 381 | ret = utils_create_stream_file(path, metadata_stream->name, |
| 382 | metadata_stream->chan->tracefile_size, |
| 383 | metadata_stream->tracefile_count_current, |
| 384 | metadata_stream->uid, metadata_stream->gid); |
| 385 | if (ret < 0) { |
| 386 | goto end; |
| 387 | } |
| 388 | metadata_stream->out_fd = ret; |
| 389 | } |
| 390 | |
| 391 | ret = 0; |
| 392 | while (ret >= 0) { |
| 393 | ret = lttng_kconsumer_read_subbuffer(metadata_stream, ctx); |
| 394 | if (ret < 0) { |
| 395 | if (ret != -EPERM) { |
| 396 | ERR("Kernel snapshot reading subbuffer"); |
| 397 | goto end; |
| 398 | } |
| 399 | /* "ret" is negative at this point so we will exit the loop. */ |
| 400 | continue; |
| 401 | } |
| 402 | } |
| 403 | |
| 404 | if (relayd_id == (uint64_t) -1ULL) { |
| 405 | ret = close(metadata_stream->out_fd); |
| 406 | if (ret < 0) { |
| 407 | PERROR("Kernel consumer snapshot close out_fd"); |
| 408 | goto end; |
| 409 | } |
| 410 | metadata_stream->out_fd = -1; |
| 411 | } else { |
| 412 | close_relayd_stream(metadata_stream); |
| 413 | metadata_stream->net_seq_idx = (uint64_t) -1ULL; |
| 414 | } |
| 415 | |
| 416 | ret = 0; |
| 417 | end: |
| 418 | rcu_read_unlock(); |
| 419 | return ret; |
| 420 | } |
| 421 | |
| 422 | /* |
| 423 | * Receive command from session daemon and process it. |
| 424 | * |
| 425 | * Return 1 on success else a negative value or 0. |
| 426 | */ |
| 427 | int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
| 428 | int sock, struct pollfd *consumer_sockpoll) |
| 429 | { |
| 430 | ssize_t ret; |
| 431 | enum lttng_error_code ret_code = LTTNG_OK; |
| 432 | struct lttcomm_consumer_msg msg; |
| 433 | |
| 434 | ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); |
| 435 | if (ret != sizeof(msg)) { |
| 436 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); |
| 437 | if (ret > 0) { |
| 438 | ret = -1; |
| 439 | } |
| 440 | return ret; |
| 441 | } |
| 442 | if (msg.cmd_type == LTTNG_CONSUMER_STOP) { |
| 443 | /* |
| 444 | * Notify the session daemon that the command is completed. |
| 445 | * |
| 446 | * On transport layer error, the function call will print an error |
| 447 | * message so handling the returned code is a bit useless since we |
| 448 | * return an error code anyway. |
| 449 | */ |
| 450 | (void) consumer_send_status_msg(sock, ret_code); |
| 451 | return -ENOENT; |
| 452 | } |
| 453 | |
| 454 | /* relayd needs RCU read-side protection */ |
| 455 | rcu_read_lock(); |
| 456 | |
| 457 | switch (msg.cmd_type) { |
| 458 | case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: |
| 459 | { |
| 460 | /* Session daemon status message are handled in the following call. */ |
| 461 | ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index, |
| 462 | msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, |
| 463 | &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id); |
| 464 | goto end_nosignal; |
| 465 | } |
| 466 | case LTTNG_CONSUMER_ADD_CHANNEL: |
| 467 | { |
| 468 | struct lttng_consumer_channel *new_channel; |
| 469 | int ret_recv; |
| 470 | |
| 471 | /* First send a status message before receiving the fds. */ |
| 472 | ret = consumer_send_status_msg(sock, ret_code); |
| 473 | if (ret < 0) { |
| 474 | /* Somehow, the session daemon is not responding anymore. */ |
| 475 | goto error_fatal; |
| 476 | } |
| 477 | DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key); |
| 478 | new_channel = consumer_allocate_channel(msg.u.channel.channel_key, |
| 479 | msg.u.channel.session_id, msg.u.channel.pathname, |
| 480 | msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid, |
| 481 | msg.u.channel.relayd_id, msg.u.channel.output, |
| 482 | msg.u.channel.tracefile_size, |
| 483 | msg.u.channel.tracefile_count, |
| 484 | msg.u.channel.monitor); |
| 485 | if (new_channel == NULL) { |
| 486 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
| 487 | goto end_nosignal; |
| 488 | } |
| 489 | new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams; |
| 490 | |
| 491 | /* Translate and save channel type. */ |
| 492 | switch (msg.u.channel.type) { |
| 493 | case CONSUMER_CHANNEL_TYPE_DATA: |
| 494 | case CONSUMER_CHANNEL_TYPE_METADATA: |
| 495 | new_channel->type = msg.u.channel.type; |
| 496 | break; |
| 497 | default: |
| 498 | assert(0); |
| 499 | goto end_nosignal; |
| 500 | }; |
| 501 | |
| 502 | if (ctx->on_recv_channel != NULL) { |
| 503 | ret_recv = ctx->on_recv_channel(new_channel); |
| 504 | if (ret_recv == 0) { |
| 505 | ret = consumer_add_channel(new_channel, ctx); |
| 506 | } else if (ret_recv < 0) { |
| 507 | goto end_nosignal; |
| 508 | } |
| 509 | } else { |
| 510 | ret = consumer_add_channel(new_channel, ctx); |
| 511 | } |
| 512 | |
| 513 | /* If we received an error in add_channel, we need to report it. */ |
| 514 | if (ret < 0) { |
| 515 | ret = consumer_send_status_msg(sock, ret); |
| 516 | if (ret < 0) { |
| 517 | goto error_fatal; |
| 518 | } |
| 519 | goto end_nosignal; |
| 520 | } |
| 521 | |
| 522 | goto end_nosignal; |
| 523 | } |
| 524 | case LTTNG_CONSUMER_ADD_STREAM: |
| 525 | { |
| 526 | int fd; |
| 527 | struct lttng_pipe *stream_pipe; |
| 528 | struct lttng_consumer_stream *new_stream; |
| 529 | struct lttng_consumer_channel *channel; |
| 530 | int alloc_ret = 0; |
| 531 | |
| 532 | /* |
| 533 | * Get stream's channel reference. Needed when adding the stream to the |
| 534 | * global hash table. |
| 535 | */ |
| 536 | channel = consumer_find_channel(msg.u.stream.channel_key); |
| 537 | if (!channel) { |
| 538 | /* |
| 539 | * We could not find the channel. Can happen if cpu hotplug |
| 540 | * happens while tearing down. |
| 541 | */ |
| 542 | ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key); |
| 543 | ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND; |
| 544 | } |
| 545 | |
| 546 | /* First send a status message before receiving the fds. */ |
| 547 | ret = consumer_send_status_msg(sock, ret_code); |
| 548 | if (ret < 0) { |
| 549 | /* |
| 550 | * Somehow, the session daemon is not responding |
| 551 | * anymore. |
| 552 | */ |
| 553 | goto error_fatal; |
| 554 | } |
| 555 | if (ret_code != LTTNG_OK) { |
| 556 | /* |
| 557 | * Channel was not found. |
| 558 | */ |
| 559 | goto end_nosignal; |
| 560 | } |
| 561 | |
| 562 | /* block */ |
| 563 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { |
| 564 | rcu_read_unlock(); |
| 565 | return -EINTR; |
| 566 | } |
| 567 | |
| 568 | /* Get stream file descriptor from socket */ |
| 569 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); |
| 570 | if (ret != sizeof(fd)) { |
| 571 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
| 572 | rcu_read_unlock(); |
| 573 | return ret; |
| 574 | } |
| 575 | |
| 576 | /* |
| 577 | * Send status code to session daemon only if the recv works. If the |
| 578 | * above recv() failed, the session daemon is notified through the |
| 579 | * error socket and the teardown is eventually done. |
| 580 | */ |
| 581 | ret = consumer_send_status_msg(sock, ret_code); |
| 582 | if (ret < 0) { |
| 583 | /* Somehow, the session daemon is not responding anymore. */ |
| 584 | goto end_nosignal; |
| 585 | } |
| 586 | |
| 587 | new_stream = consumer_allocate_stream(channel->key, |
| 588 | fd, |
| 589 | LTTNG_CONSUMER_ACTIVE_STREAM, |
| 590 | channel->name, |
| 591 | channel->uid, |
| 592 | channel->gid, |
| 593 | channel->relayd_id, |
| 594 | channel->session_id, |
| 595 | msg.u.stream.cpu, |
| 596 | &alloc_ret, |
| 597 | channel->type); |
| 598 | if (new_stream == NULL) { |
| 599 | switch (alloc_ret) { |
| 600 | case -ENOMEM: |
| 601 | case -EINVAL: |
| 602 | default: |
| 603 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
| 604 | break; |
| 605 | } |
| 606 | goto end_nosignal; |
| 607 | } |
| 608 | new_stream->chan = channel; |
| 609 | new_stream->wait_fd = fd; |
| 610 | switch (channel->output) { |
| 611 | case CONSUMER_CHANNEL_SPLICE: |
| 612 | new_stream->output = LTTNG_EVENT_SPLICE; |
| 613 | break; |
| 614 | case CONSUMER_CHANNEL_MMAP: |
| 615 | new_stream->output = LTTNG_EVENT_MMAP; |
| 616 | break; |
| 617 | default: |
| 618 | ERR("Stream output unknown %d", channel->output); |
| 619 | goto end_nosignal; |
| 620 | } |
| 621 | |
| 622 | /* |
| 623 | * We've just assigned the channel to the stream so increment the |
| 624 | * refcount right now. We don't need to increment the refcount for |
| 625 | * streams in no monitor because we handle manually the cleanup of |
| 626 | * those. It is very important to make sure there is NO prior |
| 627 | * consumer_del_stream() calls or else the refcount will be unbalanced. |
| 628 | */ |
| 629 | if (channel->monitor) { |
| 630 | uatomic_inc(&new_stream->chan->refcount); |
| 631 | } |
| 632 | |
| 633 | /* |
| 634 | * The buffer flush is done on the session daemon side for the kernel |
| 635 | * so no need for the stream "hangup_flush_done" variable to be |
| 636 | * tracked. This is important for a kernel stream since we don't rely |
| 637 | * on the flush state of the stream to read data. It's not the case for |
| 638 | * user space tracing. |
| 639 | */ |
| 640 | new_stream->hangup_flush_done = 0; |
| 641 | |
| 642 | if (ctx->on_recv_stream) { |
| 643 | ret = ctx->on_recv_stream(new_stream); |
| 644 | if (ret < 0) { |
| 645 | consumer_del_stream(new_stream, NULL); |
| 646 | goto end_nosignal; |
| 647 | } |
| 648 | } |
| 649 | |
| 650 | if (new_stream->metadata_flag) { |
| 651 | channel->metadata_stream = new_stream; |
| 652 | } |
| 653 | |
| 654 | /* Do not monitor this stream. */ |
| 655 | if (!channel->monitor) { |
| 656 | DBG("Kernel consumer add stream %s in no monitor mode with" |
| 657 | "relayd id %" PRIu64, new_stream->name, |
| 658 | new_stream->relayd_stream_id); |
| 659 | cds_list_add(&new_stream->no_monitor_node, |
| 660 | &channel->stream_no_monitor_list.head); |
| 661 | break; |
| 662 | } |
| 663 | |
| 664 | ret = send_relayd_stream(new_stream, NULL); |
| 665 | if (ret < 0) { |
| 666 | consumer_del_stream(new_stream, NULL); |
| 667 | goto end_nosignal; |
| 668 | } |
| 669 | |
| 670 | /* Get the right pipe where the stream will be sent. */ |
| 671 | if (new_stream->metadata_flag) { |
| 672 | stream_pipe = ctx->consumer_metadata_pipe; |
| 673 | } else { |
| 674 | stream_pipe = ctx->consumer_data_pipe; |
| 675 | } |
| 676 | |
| 677 | ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream)); |
| 678 | if (ret < 0) { |
| 679 | ERR("Consumer write %s stream to pipe %d", |
| 680 | new_stream->metadata_flag ? "metadata" : "data", |
| 681 | lttng_pipe_get_writefd(stream_pipe)); |
| 682 | consumer_del_stream(new_stream, NULL); |
| 683 | goto end_nosignal; |
| 684 | } |
| 685 | |
| 686 | DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64, |
| 687 | new_stream->name, fd, new_stream->relayd_stream_id); |
| 688 | break; |
| 689 | } |
| 690 | case LTTNG_CONSUMER_UPDATE_STREAM: |
| 691 | { |
| 692 | rcu_read_unlock(); |
| 693 | return -ENOSYS; |
| 694 | } |
| 695 | case LTTNG_CONSUMER_DESTROY_RELAYD: |
| 696 | { |
| 697 | uint64_t index = msg.u.destroy_relayd.net_seq_idx; |
| 698 | struct consumer_relayd_sock_pair *relayd; |
| 699 | |
| 700 | DBG("Kernel consumer destroying relayd %" PRIu64, index); |
| 701 | |
| 702 | /* Get relayd reference if exists. */ |
| 703 | relayd = consumer_find_relayd(index); |
| 704 | if (relayd == NULL) { |
| 705 | DBG("Unable to find relayd %" PRIu64, index); |
| 706 | ret_code = LTTNG_ERR_NO_CONSUMER; |
| 707 | } |
| 708 | |
| 709 | /* |
| 710 | * Each relayd socket pair has a refcount of stream attached to it |
| 711 | * which tells if the relayd is still active or not depending on the |
| 712 | * refcount value. |
| 713 | * |
| 714 | * This will set the destroy flag of the relayd object and destroy it |
| 715 | * if the refcount reaches zero when called. |
| 716 | * |
| 717 | * The destroy can happen either here or when a stream fd hangs up. |
| 718 | */ |
| 719 | if (relayd) { |
| 720 | consumer_flag_relayd_for_destroy(relayd); |
| 721 | } |
| 722 | |
| 723 | ret = consumer_send_status_msg(sock, ret_code); |
| 724 | if (ret < 0) { |
| 725 | /* Somehow, the session daemon is not responding anymore. */ |
| 726 | goto error_fatal; |
| 727 | } |
| 728 | |
| 729 | goto end_nosignal; |
| 730 | } |
| 731 | case LTTNG_CONSUMER_DATA_PENDING: |
| 732 | { |
| 733 | int32_t ret; |
| 734 | uint64_t id = msg.u.data_pending.session_id; |
| 735 | |
| 736 | DBG("Kernel consumer data pending command for id %" PRIu64, id); |
| 737 | |
| 738 | ret = consumer_data_pending(id); |
| 739 | |
| 740 | /* Send back returned value to session daemon */ |
| 741 | ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret)); |
| 742 | if (ret < 0) { |
| 743 | PERROR("send data pending ret code"); |
| 744 | goto error_fatal; |
| 745 | } |
| 746 | |
| 747 | /* |
| 748 | * No need to send back a status message since the data pending |
| 749 | * returned value is the response. |
| 750 | */ |
| 751 | break; |
| 752 | } |
| 753 | case LTTNG_CONSUMER_SNAPSHOT_CHANNEL: |
| 754 | { |
| 755 | if (msg.u.snapshot_channel.metadata == 1) { |
| 756 | ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key, |
| 757 | msg.u.snapshot_channel.pathname, |
| 758 | msg.u.snapshot_channel.relayd_id, ctx); |
| 759 | if (ret < 0) { |
| 760 | ERR("Snapshot metadata failed"); |
| 761 | ret_code = LTTNG_ERR_KERN_META_FAIL; |
| 762 | } |
| 763 | } else { |
| 764 | ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key, |
| 765 | msg.u.snapshot_channel.pathname, |
| 766 | msg.u.snapshot_channel.relayd_id, ctx); |
| 767 | if (ret < 0) { |
| 768 | ERR("Snapshot channel failed"); |
| 769 | ret_code = LTTNG_ERR_KERN_CHAN_FAIL; |
| 770 | } |
| 771 | } |
| 772 | |
| 773 | ret = consumer_send_status_msg(sock, ret_code); |
| 774 | if (ret < 0) { |
| 775 | /* Somehow, the session daemon is not responding anymore. */ |
| 776 | goto end_nosignal; |
| 777 | } |
| 778 | break; |
| 779 | } |
| 780 | case LTTNG_CONSUMER_DESTROY_CHANNEL: |
| 781 | { |
| 782 | uint64_t key = msg.u.destroy_channel.key; |
| 783 | struct lttng_consumer_channel *channel; |
| 784 | |
| 785 | channel = consumer_find_channel(key); |
| 786 | if (!channel) { |
| 787 | ERR("Kernel consumer destroy channel %" PRIu64 " not found", key); |
| 788 | ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND; |
| 789 | } |
| 790 | |
| 791 | ret = consumer_send_status_msg(sock, ret_code); |
| 792 | if (ret < 0) { |
| 793 | /* Somehow, the session daemon is not responding anymore. */ |
| 794 | goto end_nosignal; |
| 795 | } |
| 796 | |
| 797 | /* |
| 798 | * This command should ONLY be issued for channel with streams set in |
| 799 | * no monitor mode. |
| 800 | */ |
| 801 | assert(!channel->monitor); |
| 802 | |
| 803 | /* |
| 804 | * The refcount should ALWAYS be 0 in the case of a channel in no |
| 805 | * monitor mode. |
| 806 | */ |
| 807 | assert(!uatomic_sub_return(&channel->refcount, 1)); |
| 808 | |
| 809 | consumer_del_channel(channel); |
| 810 | |
| 811 | goto end_nosignal; |
| 812 | } |
| 813 | default: |
| 814 | goto end_nosignal; |
| 815 | } |
| 816 | |
| 817 | end_nosignal: |
| 818 | rcu_read_unlock(); |
| 819 | |
| 820 | /* |
| 821 | * Return 1 to indicate success since the 0 value can be a socket |
| 822 | * shutdown during the recv() or send() call. |
| 823 | */ |
| 824 | return 1; |
| 825 | |
| 826 | error_fatal: |
| 827 | rcu_read_unlock(); |
| 828 | /* This will issue a consumer stop. */ |
| 829 | return -1; |
| 830 | } |
| 831 | |
| 832 | /* |
| 833 | * Consume data on a file descriptor and write it on a trace file. |
| 834 | */ |
| 835 | ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, |
| 836 | struct lttng_consumer_local_data *ctx) |
| 837 | { |
| 838 | unsigned long len, subbuf_size, padding; |
| 839 | int err; |
| 840 | ssize_t ret = 0; |
| 841 | int infd = stream->wait_fd; |
| 842 | |
| 843 | DBG("In read_subbuffer (infd : %d)", infd); |
| 844 | /* Get the next subbuffer */ |
| 845 | err = kernctl_get_next_subbuf(infd); |
| 846 | if (err != 0) { |
| 847 | ret = err; |
| 848 | /* |
| 849 | * This is a debug message even for single-threaded consumer, |
| 850 | * because poll() have more relaxed criterions than get subbuf, |
| 851 | * so get_subbuf may fail for short race windows where poll() |
| 852 | * would issue wakeups. |
| 853 | */ |
| 854 | DBG("Reserving sub buffer failed (everything is normal, " |
| 855 | "it is due to concurrency)"); |
| 856 | goto end; |
| 857 | } |
| 858 | |
| 859 | /* Get the full subbuffer size including padding */ |
| 860 | err = kernctl_get_padded_subbuf_size(infd, &len); |
| 861 | if (err != 0) { |
| 862 | errno = -err; |
| 863 | perror("Getting sub-buffer len failed."); |
| 864 | ret = err; |
| 865 | goto end; |
| 866 | } |
| 867 | |
| 868 | switch (stream->chan->output) { |
| 869 | case CONSUMER_CHANNEL_SPLICE: |
| 870 | /* |
| 871 | * XXX: The lttng-modules splice "actor" does not handle copying |
| 872 | * partial pages hence only using the subbuffer size without the |
| 873 | * padding makes the splice fail. |
| 874 | */ |
| 875 | subbuf_size = len; |
| 876 | padding = 0; |
| 877 | |
| 878 | /* splice the subbuffer to the tracefile */ |
| 879 | ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size, |
| 880 | padding); |
| 881 | /* |
| 882 | * XXX: Splice does not support network streaming so the return value |
| 883 | * is simply checked against subbuf_size and not like the mmap() op. |
| 884 | */ |
| 885 | if (ret != subbuf_size) { |
| 886 | /* |
| 887 | * display the error but continue processing to try |
| 888 | * to release the subbuffer |
| 889 | */ |
| 890 | ERR("Error splicing to tracefile (ret: %zd != len: %lu)", |
| 891 | ret, subbuf_size); |
| 892 | } |
| 893 | break; |
| 894 | case CONSUMER_CHANNEL_MMAP: |
| 895 | /* Get subbuffer size without padding */ |
| 896 | err = kernctl_get_subbuf_size(infd, &subbuf_size); |
| 897 | if (err != 0) { |
| 898 | errno = -err; |
| 899 | perror("Getting sub-buffer len failed."); |
| 900 | ret = err; |
| 901 | goto end; |
| 902 | } |
| 903 | |
| 904 | /* Make sure the tracer is not gone mad on us! */ |
| 905 | assert(len >= subbuf_size); |
| 906 | |
| 907 | padding = len - subbuf_size; |
| 908 | |
| 909 | /* write the subbuffer to the tracefile */ |
| 910 | ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, |
| 911 | padding); |
| 912 | /* |
| 913 | * The mmap operation should write subbuf_size amount of data when |
| 914 | * network streaming or the full padding (len) size when we are _not_ |
| 915 | * streaming. |
| 916 | */ |
| 917 | if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) || |
| 918 | (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) { |
| 919 | /* |
| 920 | * Display the error but continue processing to try to release the |
| 921 | * subbuffer |
| 922 | */ |
| 923 | ERR("Error writing to tracefile " |
| 924 | "(ret: %zd != len: %lu != subbuf_size: %lu)", |
| 925 | ret, len, subbuf_size); |
| 926 | } |
| 927 | break; |
| 928 | default: |
| 929 | ERR("Unknown output method"); |
| 930 | ret = -1; |
| 931 | } |
| 932 | |
| 933 | err = kernctl_put_next_subbuf(infd); |
| 934 | if (err != 0) { |
| 935 | errno = -err; |
| 936 | if (errno == EFAULT) { |
| 937 | perror("Error in unreserving sub buffer\n"); |
| 938 | } else if (errno == EIO) { |
| 939 | /* Should never happen with newer LTTng versions */ |
| 940 | perror("Reader has been pushed by the writer, last sub-buffer corrupted."); |
| 941 | } |
| 942 | |
| 943 | ret = -err; |
| 944 | goto end; |
| 945 | } |
| 946 | |
| 947 | end: |
| 948 | return ret; |
| 949 | } |
| 950 | |
| 951 | int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) |
| 952 | { |
| 953 | int ret; |
| 954 | |
| 955 | assert(stream); |
| 956 | |
| 957 | /* |
| 958 | * Don't create anything if this is set for streaming or should not be |
| 959 | * monitored. |
| 960 | */ |
| 961 | if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) { |
| 962 | ret = utils_create_stream_file(stream->chan->pathname, stream->name, |
| 963 | stream->chan->tracefile_size, stream->tracefile_count_current, |
| 964 | stream->uid, stream->gid); |
| 965 | if (ret < 0) { |
| 966 | goto error; |
| 967 | } |
| 968 | stream->out_fd = ret; |
| 969 | stream->tracefile_size_current = 0; |
| 970 | } |
| 971 | |
| 972 | if (stream->output == LTTNG_EVENT_MMAP) { |
| 973 | /* get the len of the mmap region */ |
| 974 | unsigned long mmap_len; |
| 975 | |
| 976 | ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len); |
| 977 | if (ret != 0) { |
| 978 | errno = -ret; |
| 979 | PERROR("kernctl_get_mmap_len"); |
| 980 | goto error_close_fd; |
| 981 | } |
| 982 | stream->mmap_len = (size_t) mmap_len; |
| 983 | |
| 984 | stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ, |
| 985 | MAP_PRIVATE, stream->wait_fd, 0); |
| 986 | if (stream->mmap_base == MAP_FAILED) { |
| 987 | PERROR("Error mmaping"); |
| 988 | ret = -1; |
| 989 | goto error_close_fd; |
| 990 | } |
| 991 | } |
| 992 | |
| 993 | /* we return 0 to let the library handle the FD internally */ |
| 994 | return 0; |
| 995 | |
| 996 | error_close_fd: |
| 997 | { |
| 998 | int err; |
| 999 | |
| 1000 | err = close(stream->out_fd); |
| 1001 | assert(!err); |
| 1002 | } |
| 1003 | error: |
| 1004 | return ret; |
| 1005 | } |
| 1006 | |
| 1007 | /* |
| 1008 | * Check if data is still being extracted from the buffers for a specific |
| 1009 | * stream. Consumer data lock MUST be acquired before calling this function |
| 1010 | * and the stream lock. |
| 1011 | * |
| 1012 | * Return 1 if the traced data are still getting read else 0 meaning that the |
| 1013 | * data is available for trace viewer reading. |
| 1014 | */ |
| 1015 | int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream) |
| 1016 | { |
| 1017 | int ret; |
| 1018 | |
| 1019 | assert(stream); |
| 1020 | |
| 1021 | ret = kernctl_get_next_subbuf(stream->wait_fd); |
| 1022 | if (ret == 0) { |
| 1023 | /* There is still data so let's put back this subbuffer. */ |
| 1024 | ret = kernctl_put_subbuf(stream->wait_fd); |
| 1025 | assert(ret == 0); |
| 1026 | ret = 1; /* Data is pending */ |
| 1027 | goto end; |
| 1028 | } |
| 1029 | |
| 1030 | /* Data is NOT pending and ready to be read. */ |
| 1031 | ret = 0; |
| 1032 | |
| 1033 | end: |
| 1034 | return ret; |
| 1035 | } |