| 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 | struct lttcomm_consumer_msg msg; |
| 92 | |
| 93 | ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg)); |
| 94 | if (ret != sizeof(msg)) { |
| 95 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD); |
| 96 | return ret; |
| 97 | } |
| 98 | if (msg.cmd_type == LTTNG_CONSUMER_STOP) { |
| 99 | return -ENOENT; |
| 100 | } |
| 101 | |
| 102 | /* relayd needs RCU read-side protection */ |
| 103 | rcu_read_lock(); |
| 104 | |
| 105 | switch (msg.cmd_type) { |
| 106 | case LTTNG_CONSUMER_ADD_RELAYD_SOCKET: |
| 107 | { |
| 108 | ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index, |
| 109 | msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll, |
| 110 | &msg.u.relayd_sock.sock); |
| 111 | goto end_nosignal; |
| 112 | } |
| 113 | case LTTNG_CONSUMER_ADD_CHANNEL: |
| 114 | { |
| 115 | struct lttng_consumer_channel *new_channel; |
| 116 | |
| 117 | DBG("consumer_add_channel %d", msg.u.channel.channel_key); |
| 118 | new_channel = consumer_allocate_channel(msg.u.channel.channel_key, |
| 119 | -1, -1, |
| 120 | msg.u.channel.mmap_len, |
| 121 | msg.u.channel.max_sb_size); |
| 122 | if (new_channel == NULL) { |
| 123 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
| 124 | goto end_nosignal; |
| 125 | } |
| 126 | if (ctx->on_recv_channel != NULL) { |
| 127 | ret = ctx->on_recv_channel(new_channel); |
| 128 | if (ret == 0) { |
| 129 | consumer_add_channel(new_channel); |
| 130 | } else if (ret < 0) { |
| 131 | goto end_nosignal; |
| 132 | } |
| 133 | } else { |
| 134 | consumer_add_channel(new_channel); |
| 135 | } |
| 136 | goto end_nosignal; |
| 137 | } |
| 138 | case LTTNG_CONSUMER_ADD_STREAM: |
| 139 | { |
| 140 | int fd; |
| 141 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 142 | struct lttng_consumer_stream *new_stream; |
| 143 | |
| 144 | /* block */ |
| 145 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { |
| 146 | rcu_read_unlock(); |
| 147 | return -EINTR; |
| 148 | } |
| 149 | |
| 150 | /* Get stream file descriptor from socket */ |
| 151 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); |
| 152 | if (ret != sizeof(fd)) { |
| 153 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
| 154 | rcu_read_unlock(); |
| 155 | return ret; |
| 156 | } |
| 157 | |
| 158 | new_stream = consumer_allocate_stream(msg.u.stream.channel_key, |
| 159 | msg.u.stream.stream_key, |
| 160 | fd, fd, |
| 161 | msg.u.stream.state, |
| 162 | msg.u.stream.mmap_len, |
| 163 | msg.u.stream.output, |
| 164 | msg.u.stream.path_name, |
| 165 | msg.u.stream.uid, |
| 166 | msg.u.stream.gid, |
| 167 | msg.u.stream.net_index, |
| 168 | msg.u.stream.metadata_flag); |
| 169 | if (new_stream == NULL) { |
| 170 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
| 171 | goto end_nosignal; |
| 172 | } |
| 173 | |
| 174 | /* |
| 175 | * The buffer flush is done on the session daemon side for the kernel |
| 176 | * so no need for the stream "hangup_flush_done" variable to be |
| 177 | * tracked. This is important for a kernel stream since we don't rely |
| 178 | * on the flush state of the stream to read data. It's not the case for |
| 179 | * user space tracing. |
| 180 | */ |
| 181 | new_stream->hangup_flush_done = 0; |
| 182 | |
| 183 | /* The stream is not metadata. Get relayd reference if exists. */ |
| 184 | relayd = consumer_find_relayd(msg.u.stream.net_index); |
| 185 | if (relayd != NULL) { |
| 186 | /* Add stream on the relayd */ |
| 187 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 188 | ret = relayd_add_stream(&relayd->control_sock, |
| 189 | msg.u.stream.name, msg.u.stream.path_name, |
| 190 | &new_stream->relayd_stream_id); |
| 191 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 192 | if (ret < 0) { |
| 193 | goto end_nosignal; |
| 194 | } |
| 195 | } else if (msg.u.stream.net_index != -1) { |
| 196 | ERR("Network sequence index %d unknown. Not adding stream.", |
| 197 | msg.u.stream.net_index); |
| 198 | free(new_stream); |
| 199 | goto end_nosignal; |
| 200 | } |
| 201 | |
| 202 | /* Send stream to the metadata thread */ |
| 203 | if (new_stream->metadata_flag) { |
| 204 | if (ctx->on_recv_stream) { |
| 205 | ret = ctx->on_recv_stream(new_stream); |
| 206 | if (ret < 0) { |
| 207 | goto end_nosignal; |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | do { |
| 212 | ret = write(ctx->consumer_metadata_pipe[1], new_stream, |
| 213 | sizeof(struct lttng_consumer_stream)); |
| 214 | } while (ret < 0 && errno == EINTR); |
| 215 | if (ret < 0) { |
| 216 | PERROR("write metadata pipe"); |
| 217 | } |
| 218 | } else { |
| 219 | if (ctx->on_recv_stream) { |
| 220 | ret = ctx->on_recv_stream(new_stream); |
| 221 | if (ret < 0) { |
| 222 | goto end_nosignal; |
| 223 | } |
| 224 | } |
| 225 | consumer_add_stream(new_stream); |
| 226 | } |
| 227 | |
| 228 | DBG("Kernel consumer_add_stream (%d)", fd); |
| 229 | break; |
| 230 | } |
| 231 | case LTTNG_CONSUMER_UPDATE_STREAM: |
| 232 | { |
| 233 | rcu_read_unlock(); |
| 234 | return -ENOSYS; |
| 235 | } |
| 236 | case LTTNG_CONSUMER_DESTROY_RELAYD: |
| 237 | { |
| 238 | uint64_t index = msg.u.destroy_relayd.net_seq_idx; |
| 239 | struct consumer_relayd_sock_pair *relayd; |
| 240 | |
| 241 | DBG("Kernel consumer destroying relayd %" PRIu64, index); |
| 242 | |
| 243 | /* Get relayd reference if exists. */ |
| 244 | relayd = consumer_find_relayd(index); |
| 245 | if (relayd == NULL) { |
| 246 | ERR("Unable to find relayd %" PRIu64, index); |
| 247 | goto end_nosignal; |
| 248 | } |
| 249 | |
| 250 | /* |
| 251 | * Each relayd socket pair has a refcount of stream attached to it |
| 252 | * which tells if the relayd is still active or not depending on the |
| 253 | * refcount value. |
| 254 | * |
| 255 | * This will set the destroy flag of the relayd object and destroy it |
| 256 | * if the refcount reaches zero when called. |
| 257 | * |
| 258 | * The destroy can happen either here or when a stream fd hangs up. |
| 259 | */ |
| 260 | consumer_flag_relayd_for_destroy(relayd); |
| 261 | |
| 262 | goto end_nosignal; |
| 263 | } |
| 264 | default: |
| 265 | goto end_nosignal; |
| 266 | } |
| 267 | |
| 268 | /* |
| 269 | * Wake-up the other end by writing a null byte in the pipe (non-blocking). |
| 270 | * Important note: Because writing into the pipe is non-blocking (and |
| 271 | * therefore we allow dropping wakeup data, as long as there is wakeup data |
| 272 | * present in the pipe buffer to wake up the other end), the other end |
| 273 | * should perform the following sequence for waiting: |
| 274 | * |
| 275 | * 1) empty the pipe (reads). |
| 276 | * 2) perform update operation. |
| 277 | * 3) wait on the pipe (poll). |
| 278 | */ |
| 279 | do { |
| 280 | ret = write(ctx->consumer_poll_pipe[1], "", 1); |
| 281 | } while (ret < 0 && errno == EINTR); |
| 282 | end_nosignal: |
| 283 | rcu_read_unlock(); |
| 284 | |
| 285 | /* |
| 286 | * Return 1 to indicate success since the 0 value can be a socket |
| 287 | * shutdown during the recv() or send() call. |
| 288 | */ |
| 289 | return 1; |
| 290 | } |
| 291 | |
| 292 | /* |
| 293 | * Consume data on a file descriptor and write it on a trace file. |
| 294 | */ |
| 295 | ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream, |
| 296 | struct lttng_consumer_local_data *ctx) |
| 297 | { |
| 298 | unsigned long len; |
| 299 | int err; |
| 300 | ssize_t ret = 0; |
| 301 | int infd = stream->wait_fd; |
| 302 | |
| 303 | DBG("In read_subbuffer (infd : %d)", infd); |
| 304 | /* Get the next subbuffer */ |
| 305 | err = kernctl_get_next_subbuf(infd); |
| 306 | if (err != 0) { |
| 307 | /* |
| 308 | * This is a debug message even for single-threaded consumer, |
| 309 | * because poll() have more relaxed criterions than get subbuf, |
| 310 | * so get_subbuf may fail for short race windows where poll() |
| 311 | * would issue wakeups. |
| 312 | */ |
| 313 | DBG("Reserving sub buffer failed (everything is normal, " |
| 314 | "it is due to concurrency)"); |
| 315 | goto end; |
| 316 | } |
| 317 | |
| 318 | switch (stream->output) { |
| 319 | case LTTNG_EVENT_SPLICE: |
| 320 | /* read the whole subbuffer */ |
| 321 | err = kernctl_get_padded_subbuf_size(infd, &len); |
| 322 | if (err != 0) { |
| 323 | errno = -ret; |
| 324 | perror("Getting sub-buffer len failed."); |
| 325 | goto end; |
| 326 | } |
| 327 | |
| 328 | /* splice the subbuffer to the tracefile */ |
| 329 | ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, len); |
| 330 | if (ret != len) { |
| 331 | /* |
| 332 | * display the error but continue processing to try |
| 333 | * to release the subbuffer |
| 334 | */ |
| 335 | ERR("Error splicing to tracefile (ret: %zd != len: %lu)", |
| 336 | ret, len); |
| 337 | } |
| 338 | |
| 339 | break; |
| 340 | case LTTNG_EVENT_MMAP: |
| 341 | /* read the used subbuffer size */ |
| 342 | err = kernctl_get_padded_subbuf_size(infd, &len); |
| 343 | if (err != 0) { |
| 344 | errno = -ret; |
| 345 | perror("Getting sub-buffer len failed."); |
| 346 | goto end; |
| 347 | } |
| 348 | /* write the subbuffer to the tracefile */ |
| 349 | ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len); |
| 350 | if (ret != len) { |
| 351 | /* |
| 352 | * display the error but continue processing to try |
| 353 | * to release the subbuffer |
| 354 | */ |
| 355 | ERR("Error writing to tracefile"); |
| 356 | } |
| 357 | break; |
| 358 | default: |
| 359 | ERR("Unknown output method"); |
| 360 | ret = -1; |
| 361 | } |
| 362 | |
| 363 | err = kernctl_put_next_subbuf(infd); |
| 364 | if (err != 0) { |
| 365 | errno = -ret; |
| 366 | if (errno == EFAULT) { |
| 367 | perror("Error in unreserving sub buffer\n"); |
| 368 | } else if (errno == EIO) { |
| 369 | /* Should never happen with newer LTTng versions */ |
| 370 | perror("Reader has been pushed by the writer, last sub-buffer corrupted."); |
| 371 | } |
| 372 | goto end; |
| 373 | } |
| 374 | |
| 375 | end: |
| 376 | return ret; |
| 377 | } |
| 378 | |
| 379 | int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream) |
| 380 | { |
| 381 | int ret; |
| 382 | |
| 383 | /* Opening the tracefile in write mode */ |
| 384 | if (strlen(stream->path_name) > 0 && stream->net_seq_idx == -1) { |
| 385 | ret = run_as_open(stream->path_name, |
| 386 | O_WRONLY|O_CREAT|O_TRUNC, |
| 387 | S_IRWXU|S_IRWXG|S_IRWXO, |
| 388 | stream->uid, stream->gid); |
| 389 | if (ret < 0) { |
| 390 | ERR("Opening %s", stream->path_name); |
| 391 | perror("open"); |
| 392 | goto error; |
| 393 | } |
| 394 | stream->out_fd = ret; |
| 395 | } |
| 396 | |
| 397 | if (stream->output == LTTNG_EVENT_MMAP) { |
| 398 | /* get the len of the mmap region */ |
| 399 | unsigned long mmap_len; |
| 400 | |
| 401 | ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len); |
| 402 | if (ret != 0) { |
| 403 | errno = -ret; |
| 404 | perror("kernctl_get_mmap_len"); |
| 405 | goto error_close_fd; |
| 406 | } |
| 407 | stream->mmap_len = (size_t) mmap_len; |
| 408 | |
| 409 | stream->mmap_base = mmap(NULL, stream->mmap_len, |
| 410 | PROT_READ, MAP_PRIVATE, stream->wait_fd, 0); |
| 411 | if (stream->mmap_base == MAP_FAILED) { |
| 412 | perror("Error mmaping"); |
| 413 | ret = -1; |
| 414 | goto error_close_fd; |
| 415 | } |
| 416 | } |
| 417 | |
| 418 | /* we return 0 to let the library handle the FD internally */ |
| 419 | return 0; |
| 420 | |
| 421 | error_close_fd: |
| 422 | { |
| 423 | int err; |
| 424 | |
| 425 | err = close(stream->out_fd); |
| 426 | assert(!err); |
| 427 | } |
| 428 | error: |
| 429 | return ret; |
| 430 | } |
| 431 | |