| 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/relayd/relayd.h> |
| 38 | |
| 39 | #include "kernel-consumer.h" |
| 40 | |
| 41 | extern struct lttng_consumer_global_data consumer_data; |
| 42 | extern int consumer_poll_timeout; |
| 43 | extern volatile int consumer_quit; |
| 44 | |
| 45 | /* |
| 46 | * Take a snapshot for a specific fd |
| 47 | * |
| 48 | * Returns 0 on success, < 0 on error |
| 49 | */ |
| 50 | int lttng_kconsumer_take_snapshot(struct lttng_consumer_local_data *ctx, |
| 51 | struct lttng_consumer_stream *stream) |
| 52 | { |
| 53 | int ret = 0; |
| 54 | int infd = stream->wait_fd; |
| 55 | |
| 56 | ret = kernctl_snapshot(infd); |
| 57 | if (ret != 0) { |
| 58 | errno = -ret; |
| 59 | perror("Getting sub-buffer snapshot."); |
| 60 | } |
| 61 | |
| 62 | return ret; |
| 63 | } |
| 64 | |
| 65 | /* |
| 66 | * Get the produced position |
| 67 | * |
| 68 | * Returns 0 on success, < 0 on error |
| 69 | */ |
| 70 | int lttng_kconsumer_get_produced_snapshot( |
| 71 | struct lttng_consumer_local_data *ctx, |
| 72 | 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 | int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
| 88 | int sock, struct pollfd *consumer_sockpoll) |
| 89 | { |
| 90 | ssize_t ret; |
| 91 | enum lttng_error_code ret_code = LTTNG_OK; |
| 92 | struct lttcomm_consumer_msg msg; |
| 93 | |
| 94 | ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); |
| 95 | if (ret != sizeof(msg)) { |
| 96 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); |
| 97 | return ret; |
| 98 | } |
| 99 | if (msg.cmd_type == LTTNG_CONSUMER_STOP) { |
| 100 | /* |
| 101 | * Notify the session daemon that the command is completed. |
| 102 | * |
| 103 | * On transport layer error, the function call will print an error |
| 104 | * message so handling the returned code is a bit useless since we |
| 105 | * return an error code anyway. |
| 106 | */ |
| 107 | (void) consumer_send_status_msg(sock, ret_code); |
| 108 | return -ENOENT; |
| 109 | } |
| 110 | |
| 111 | /* relayd needs RCU read-side protection */ |
| 112 | rcu_read_lock(); |
| 113 | |
| 114 | switch (msg.cmd_type) { |
| 115 | case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: |
| 116 | { |
| 117 | /* Session daemon status message are handled in the following call. */ |
| 118 | ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index, |
| 119 | msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, |
| 120 | &msg.u.relayd_sock.sock); |
| 121 | goto end_nosignal; |
| 122 | } |
| 123 | case LTTNG_CONSUMER_ADD_CHANNEL: |
| 124 | { |
| 125 | struct lttng_consumer_channel *new_channel; |
| 126 | |
| 127 | /* First send a status message before receiving the fds. */ |
| 128 | ret = consumer_send_status_msg(sock, ret_code); |
| 129 | if (ret < 0) { |
| 130 | /* Somehow, the session daemon is not responding anymore. */ |
| 131 | goto end_nosignal; |
| 132 | } |
| 133 | |
| 134 | DBG("consumer_add_channel %d", msg.u.channel.channel_key); |
| 135 | new_channel = consumer_allocate_channel(msg.u.channel.channel_key, |
| 136 | -1, -1, |
| 137 | msg.u.channel.mmap_len, |
| 138 | msg.u.channel.max_sb_size, |
| 139 | msg.u.channel.nb_init_streams); |
| 140 | if (new_channel == NULL) { |
| 141 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
| 142 | goto end_nosignal; |
| 143 | } |
| 144 | if (ctx->on_recv_channel != NULL) { |
| 145 | ret = ctx->on_recv_channel(new_channel); |
| 146 | if (ret == 0) { |
| 147 | consumer_add_channel(new_channel); |
| 148 | } else if (ret < 0) { |
| 149 | goto end_nosignal; |
| 150 | } |
| 151 | } else { |
| 152 | consumer_add_channel(new_channel); |
| 153 | } |
| 154 | goto end_nosignal; |
| 155 | } |
| 156 | case LTTNG_CONSUMER_ADD_STREAM: |
| 157 | { |
| 158 | int fd, stream_pipe; |
| 159 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 160 | struct lttng_consumer_stream *new_stream; |
| 161 | int alloc_ret = 0; |
| 162 | |
| 163 | /* First send a status message before receiving the fds. */ |
| 164 | ret = consumer_send_status_msg(sock, ret_code); |
| 165 | if (ret < 0) { |
| 166 | /* Somehow, the session daemon is not responding anymore. */ |
| 167 | goto end_nosignal; |
| 168 | } |
| 169 | |
| 170 | /* block */ |
| 171 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { |
| 172 | rcu_read_unlock(); |
| 173 | return -EINTR; |
| 174 | } |
| 175 | |
| 176 | /* Get stream file descriptor from socket */ |
| 177 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); |
| 178 | if (ret != sizeof(fd)) { |
| 179 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
| 180 | rcu_read_unlock(); |
| 181 | return ret; |
| 182 | } |
| 183 | |
| 184 | /* |
| 185 | * Send status code to session daemon only if the recv works. If the |
| 186 | * above recv() failed, the session daemon is notified through the |
| 187 | * error socket and the teardown is eventually done. |
| 188 | */ |
| 189 | ret = consumer_send_status_msg(sock, ret_code); |
| 190 | if (ret < 0) { |
| 191 | /* Somehow, the session daemon is not responding anymore. */ |
| 192 | goto end_nosignal; |
| 193 | } |
| 194 | |
| 195 | new_stream = consumer_allocate_stream(msg.u.stream.channel_key, |
| 196 | msg.u.stream.stream_key, |
| 197 | fd, fd, |
| 198 | msg.u.stream.state, |
| 199 | msg.u.stream.mmap_len, |
| 200 | msg.u.stream.output, |
| 201 | msg.u.stream.path_name, |
| 202 | msg.u.stream.uid, |
| 203 | msg.u.stream.gid, |
| 204 | msg.u.stream.net_index, |
| 205 | msg.u.stream.metadata_flag, |
| 206 | msg.u.stream.session_id, |
| 207 | &alloc_ret); |
| 208 | if (new_stream == NULL) { |
| 209 | switch (alloc_ret) { |
| 210 | case -ENOMEM: |
| 211 | case -EINVAL: |
| 212 | default: |
| 213 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
| 214 | break; |
| 215 | case -ENOENT: |
| 216 | /* |
| 217 | * We could not find the channel. Can happen if cpu hotplug |
| 218 | * happens while tearing down. |
| 219 | */ |
| 220 | DBG3("Could not find channel"); |
| 221 | break; |
| 222 | } |
| 223 | goto end_nosignal; |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * The buffer flush is done on the session daemon side for the kernel |
| 228 | * so no need for the stream "hangup_flush_done" variable to be |
| 229 | * tracked. This is important for a kernel stream since we don't rely |
| 230 | * on the flush state of the stream to read data. It's not the case for |
| 231 | * user space tracing. |
| 232 | */ |
| 233 | new_stream->hangup_flush_done = 0; |
| 234 | |
| 235 | /* The stream is not metadata. Get relayd reference if exists. */ |
| 236 | relayd = consumer_find_relayd(msg.u.stream.net_index); |
| 237 | if (relayd != NULL) { |
| 238 | /* Add stream on the relayd */ |
| 239 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 240 | ret = relayd_add_stream(&relayd->control_sock, |
| 241 | msg.u.stream.name, msg.u.stream.path_name, |
| 242 | &new_stream->relayd_stream_id); |
| 243 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 244 | if (ret < 0) { |
| 245 | consumer_del_stream(new_stream, NULL); |
| 246 | goto end_nosignal; |
| 247 | } |
| 248 | } else if (msg.u.stream.net_index != -1) { |
| 249 | ERR("Network sequence index %d unknown. Not adding stream.", |
| 250 | msg.u.stream.net_index); |
| 251 | consumer_del_stream(new_stream, NULL); |
| 252 | goto end_nosignal; |
| 253 | } |
| 254 | |
| 255 | if (ctx->on_recv_stream) { |
| 256 | ret = ctx->on_recv_stream(new_stream); |
| 257 | if (ret < 0) { |
| 258 | consumer_del_stream(new_stream, NULL); |
| 259 | goto end_nosignal; |
| 260 | } |
| 261 | } |
| 262 | |
| 263 | /* Get the right pipe where the stream will be sent. */ |
| 264 | if (new_stream->metadata_flag) { |
| 265 | stream_pipe = ctx->consumer_metadata_pipe[1]; |
| 266 | } else { |
| 267 | stream_pipe = ctx->consumer_data_pipe[1]; |
| 268 | } |
| 269 | |
| 270 | do { |
| 271 | ret = write(stream_pipe, &new_stream, sizeof(new_stream)); |
| 272 | } while (ret < 0 && errno == EINTR); |
| 273 | if (ret < 0) { |
| 274 | PERROR("Consumer write %s stream to pipe %d", |
| 275 | new_stream->metadata_flag ? "metadata" : "data", |
| 276 | stream_pipe); |
| 277 | consumer_del_stream(new_stream, NULL); |
| 278 | goto end_nosignal; |
| 279 | } |
| 280 | |
| 281 | DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64, |
| 282 | msg.u.stream.path_name, fd, new_stream->relayd_stream_id); |
| 283 | break; |
| 284 | } |
| 285 | case LTTNG_CONSUMER_UPDATE_STREAM: |
| 286 | { |
| 287 | rcu_read_unlock(); |
| 288 | return -ENOSYS; |
| 289 | } |
| 290 | case LTTNG_CONSUMER_DESTROY_RELAYD: |
| 291 | { |
| 292 | uint64_t index = msg.u.destroy_relayd.net_seq_idx; |
| 293 | struct consumer_relayd_sock_pair *relayd; |
| 294 | |
| 295 | DBG("Kernel consumer destroying relayd %" PRIu64, index); |
| 296 | |
| 297 | /* Get relayd reference if exists. */ |
| 298 | relayd = consumer_find_relayd(index); |
| 299 | if (relayd == NULL) { |
| 300 | ERR("Unable to find relayd %" PRIu64, index); |
| 301 | ret_code = LTTNG_ERR_NO_CONSUMER; |
| 302 | } |
| 303 | |
| 304 | /* |
| 305 | * Each relayd socket pair has a refcount of stream attached to it |
| 306 | * which tells if the relayd is still active or not depending on the |
| 307 | * refcount value. |
| 308 | * |
| 309 | * This will set the destroy flag of the relayd object and destroy it |
| 310 | * if the refcount reaches zero when called. |
| 311 | * |
| 312 | * The destroy can happen either here or when a stream fd hangs up. |
| 313 | */ |
| 314 | if (relayd) { |
| 315 | consumer_flag_relayd_for_destroy(relayd); |
| 316 | } |
| 317 | |
| 318 | ret = consumer_send_status_msg(sock, ret_code); |
| 319 | if (ret < 0) { |
| 320 | /* Somehow, the session daemon is not responding anymore. */ |
| 321 | goto end_nosignal; |
| 322 | } |
| 323 | |
| 324 | goto end_nosignal; |
| 325 | } |
| 326 | case LTTNG_CONSUMER_DATA_PENDING: |
| 327 | { |
| 328 | int32_t ret; |
| 329 | uint64_t id = msg.u.data_pending.session_id; |
| 330 | |
| 331 | DBG("Kernel consumer data pending command for id %" PRIu64, id); |
| 332 | |
| 333 | ret = consumer_data_pending(id); |
| 334 | |
| 335 | /* Send back returned value to session daemon */ |
| 336 | ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret)); |
| 337 | if (ret < 0) { |
| 338 | PERROR("send data pending ret code"); |
| 339 | } |
| 340 | |
| 341 | /* |
| 342 | * No need to send back a status message since the data pending |
| 343 | * returned value is the response. |
| 344 | */ |
| 345 | break; |
| 346 | } |
| 347 | default: |
| 348 | goto end_nosignal; |
| 349 | } |
| 350 | |
| 351 | end_nosignal: |
| 352 | rcu_read_unlock(); |
| 353 | |
| 354 | /* |
| 355 | * Return 1 to indicate success since the 0 value can be a socket |
| 356 | * shutdown during the recv() or send() call. |
| 357 | */ |
| 358 | return 1; |
| 359 | } |
| 360 | |
| 361 | /* |
| 362 | * Consume data on a file descriptor and write it on a trace file. |
| 363 | */ |
| 364 | ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, |
| 365 | struct lttng_consumer_local_data *ctx) |
| 366 | { |
| 367 | unsigned long len, subbuf_size, padding; |
| 368 | int err; |
| 369 | ssize_t ret = 0; |
| 370 | int infd = stream->wait_fd; |
| 371 | |
| 372 | DBG("In read_subbuffer (infd : %d)", infd); |
| 373 | /* Get the next subbuffer */ |
| 374 | err = kernctl_get_next_subbuf(infd); |
| 375 | if (err != 0) { |
| 376 | ret = err; |
| 377 | /* |
| 378 | * This is a debug message even for single-threaded consumer, |
| 379 | * because poll() have more relaxed criterions than get subbuf, |
| 380 | * so get_subbuf may fail for short race windows where poll() |
| 381 | * would issue wakeups. |
| 382 | */ |
| 383 | DBG("Reserving sub buffer failed (everything is normal, " |
| 384 | "it is due to concurrency)"); |
| 385 | goto end; |
| 386 | } |
| 387 | |
| 388 | /* Get the full subbuffer size including padding */ |
| 389 | err = kernctl_get_padded_subbuf_size(infd, &len); |
| 390 | if (err != 0) { |
| 391 | errno = -err; |
| 392 | perror("Getting sub-buffer len failed."); |
| 393 | ret = err; |
| 394 | goto end; |
| 395 | } |
| 396 | |
| 397 | switch (stream->output) { |
| 398 | case LTTNG_EVENT_SPLICE: |
| 399 | |
| 400 | /* |
| 401 | * XXX: The lttng-modules splice "actor" does not handle copying |
| 402 | * partial pages hence only using the subbuffer size without the |
| 403 | * padding makes the splice fail. |
| 404 | */ |
| 405 | subbuf_size = len; |
| 406 | padding = 0; |
| 407 | |
| 408 | /* splice the subbuffer to the tracefile */ |
| 409 | ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size, |
| 410 | padding); |
| 411 | /* |
| 412 | * XXX: Splice does not support network streaming so the return value |
| 413 | * is simply checked against subbuf_size and not like the mmap() op. |
| 414 | */ |
| 415 | if (ret != subbuf_size) { |
| 416 | /* |
| 417 | * display the error but continue processing to try |
| 418 | * to release the subbuffer |
| 419 | */ |
| 420 | ERR("Error splicing to tracefile (ret: %zd != len: %lu)", |
| 421 | ret, subbuf_size); |
| 422 | } |
| 423 | break; |
| 424 | case LTTNG_EVENT_MMAP: |
| 425 | /* Get subbuffer size without padding */ |
| 426 | err = kernctl_get_subbuf_size(infd, &subbuf_size); |
| 427 | if (err != 0) { |
| 428 | errno = -err; |
| 429 | perror("Getting sub-buffer len failed."); |
| 430 | ret = err; |
| 431 | goto end; |
| 432 | } |
| 433 | |
| 434 | /* Make sure the tracer is not gone mad on us! */ |
| 435 | assert(len >= subbuf_size); |
| 436 | |
| 437 | padding = len - subbuf_size; |
| 438 | |
| 439 | /* write the subbuffer to the tracefile */ |
| 440 | ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size, |
| 441 | padding); |
| 442 | /* |
| 443 | * The mmap operation should write subbuf_size amount of data when |
| 444 | * network streaming or the full padding (len) size when we are _not_ |
| 445 | * streaming. |
| 446 | */ |
| 447 | if ((ret != subbuf_size && stream->net_seq_idx != -1) || |
| 448 | (ret != len && stream->net_seq_idx == -1)) { |
| 449 | /* |
| 450 | * Display the error but continue processing to try to release the |
| 451 | * subbuffer |
| 452 | */ |
| 453 | ERR("Error writing to tracefile " |
| 454 | "(ret: %zd != len: %lu != subbuf_size: %lu)", |
| 455 | ret, len, subbuf_size); |
| 456 | } |
| 457 | break; |
| 458 | default: |
| 459 | ERR("Unknown output method"); |
| 460 | ret = -1; |
| 461 | } |
| 462 | |
| 463 | err = kernctl_put_next_subbuf(infd); |
| 464 | if (err != 0) { |
| 465 | errno = -err; |
| 466 | if (errno == EFAULT) { |
| 467 | perror("Error in unreserving sub buffer\n"); |
| 468 | } else if (errno == EIO) { |
| 469 | /* Should never happen with newer LTTng versions */ |
| 470 | perror("Reader has been pushed by the writer, last sub-buffer corrupted."); |
| 471 | } |
| 472 | |
| 473 | ret = -err; |
| 474 | goto end; |
| 475 | } |
| 476 | |
| 477 | end: |
| 478 | return ret; |
| 479 | } |
| 480 | |
| 481 | int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) |
| 482 | { |
| 483 | int ret; |
| 484 | |
| 485 | /* Opening the tracefile in write mode */ |
| 486 | if (strlen(stream->path_name) > 0 && stream->net_seq_idx == -1) { |
| 487 | ret = run_as_open(stream->path_name, |
| 488 | O_WRONLY|O_CREAT|O_TRUNC, |
| 489 | S_IRWXU|S_IRWXG|S_IRWXO, |
| 490 | stream->uid, stream->gid); |
| 491 | if (ret < 0) { |
| 492 | ERR("Opening %s", stream->path_name); |
| 493 | perror("open"); |
| 494 | goto error; |
| 495 | } |
| 496 | stream->out_fd = ret; |
| 497 | } |
| 498 | |
| 499 | if (stream->output == LTTNG_EVENT_MMAP) { |
| 500 | /* get the len of the mmap region */ |
| 501 | unsigned long mmap_len; |
| 502 | |
| 503 | ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len); |
| 504 | if (ret != 0) { |
| 505 | errno = -ret; |
| 506 | perror("kernctl_get_mmap_len"); |
| 507 | goto error_close_fd; |
| 508 | } |
| 509 | stream->mmap_len = (size_t) mmap_len; |
| 510 | |
| 511 | stream->mmap_base = mmap(NULL, stream->mmap_len, |
| 512 | PROT_READ, MAP_PRIVATE, stream->wait_fd, 0); |
| 513 | if (stream->mmap_base == MAP_FAILED) { |
| 514 | perror("Error mmaping"); |
| 515 | ret = -1; |
| 516 | goto error_close_fd; |
| 517 | } |
| 518 | } |
| 519 | |
| 520 | /* we return 0 to let the library handle the FD internally */ |
| 521 | return 0; |
| 522 | |
| 523 | error_close_fd: |
| 524 | { |
| 525 | int err; |
| 526 | |
| 527 | err = close(stream->out_fd); |
| 528 | assert(!err); |
| 529 | } |
| 530 | error: |
| 531 | return ret; |
| 532 | } |
| 533 | |
| 534 | /* |
| 535 | * Check if data is still being extracted from the buffers for a specific |
| 536 | * stream. Consumer data lock MUST be acquired before calling this function |
| 537 | * and the stream lock. |
| 538 | * |
| 539 | * Return 1 if the traced data are still getting read else 0 meaning that the |
| 540 | * data is available for trace viewer reading. |
| 541 | */ |
| 542 | int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream) |
| 543 | { |
| 544 | int ret; |
| 545 | |
| 546 | assert(stream); |
| 547 | |
| 548 | ret = kernctl_get_next_subbuf(stream->wait_fd); |
| 549 | if (ret == 0) { |
| 550 | /* There is still data so let's put back this subbuffer. */ |
| 551 | ret = kernctl_put_subbuf(stream->wait_fd); |
| 552 | assert(ret == 0); |
| 553 | ret = 1; /* Data is pending */ |
| 554 | goto end; |
| 555 | } |
| 556 | |
| 557 | /* Data is NOT pending and ready to be read. */ |
| 558 | ret = 0; |
| 559 | |
| 560 | end: |
| 561 | return ret; |
| 562 | } |