src/common/kernel-consumer/kernel-consumer.c

   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                                 msg.u.channel.nb_init_streams);
 123                 if (new_channel == NULL) {
 124                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 125                         goto end_nosignal;
 126                 }
 127                 if (ctx->on_recv_channel != NULL) {
 128                         ret = ctx->on_recv_channel(new_channel);
 129                         if (ret == 0) {
 130                                 consumer_add_channel(new_channel);
 131                         } else if (ret < 0) {
 132                                 goto end_nosignal;
 133                         }
 134                 } else {
 135                         consumer_add_channel(new_channel);
 136                 }
 137                 goto end_nosignal;
 138         }
 139         case LTTNG_CONSUMER_ADD_STREAM:
 140         {
 141                 int fd;
 142                 struct consumer_relayd_sock_pair *relayd = NULL;
 143                 struct lttng_consumer_stream *new_stream;
 144                 int alloc_ret = 0;
 145
 146                 /* block */
 147                 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
 148                         rcu_read_unlock();
 149                         return -EINTR;
 150                 }
 151
 152                 /* Get stream file descriptor from socket */
 153                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 154                 if (ret != sizeof(fd)) {
 155                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 156                         rcu_read_unlock();
 157                         return ret;
 158                 }
 159
 160                 new_stream = consumer_allocate_stream(msg.u.stream.channel_key,
 161                                 msg.u.stream.stream_key,
 162                                 fd, fd,
 163                                 msg.u.stream.state,
 164                                 msg.u.stream.mmap_len,
 165                                 msg.u.stream.output,
 166                                 msg.u.stream.path_name,
 167                                 msg.u.stream.uid,
 168                                 msg.u.stream.gid,
 169                                 msg.u.stream.net_index,
 170                                 msg.u.stream.metadata_flag,
 171                                 &alloc_ret);
 172                 if (new_stream == NULL) {
 173                         switch (alloc_ret) {
 174                         case -ENOMEM:
 175                         case -EINVAL:
 176                         default:
 177                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 178                                 break;
 179                         case -ENOENT:
 180                                 /*
 181                                  * We could not find the channel. Can happen if cpu hotplug
 182                                  * happens while tearing down.
 183                                  */
 184                                 DBG3("Could not find channel");
 185                                 break;
 186                         }
 187                         goto end_nosignal;
 188                 }
 189
 190                 /*
 191                  * The buffer flush is done on the session daemon side for the kernel
 192                  * so no need for the stream "hangup_flush_done" variable to be
 193                  * tracked. This is important for a kernel stream since we don't rely
 194                  * on the flush state of the stream to read data. It's not the case for
 195                  * user space tracing.
 196                  */
 197                 new_stream->hangup_flush_done = 0;
 198
 199                 /* The stream is not metadata. Get relayd reference if exists. */
 200                 relayd = consumer_find_relayd(msg.u.stream.net_index);
 201                 if (relayd != NULL) {
 202                         /* Add stream on the relayd */
 203                         pthread_mutex_lock(&relayd->ctrl_sock_mutex);
 204                         ret = relayd_add_stream(&relayd->control_sock,
 205                                         msg.u.stream.name, msg.u.stream.path_name,
 206                                         &new_stream->relayd_stream_id);
 207                         pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 208                         if (ret < 0) {
 209                                 consumer_del_stream(new_stream, NULL);
 210                                 goto end_nosignal;
 211                         }
 212                 } else if (msg.u.stream.net_index != -1) {
 213                         ERR("Network sequence index %d unknown. Not adding stream.",
 214                                         msg.u.stream.net_index);
 215                         consumer_del_stream(new_stream, NULL);
 216                         goto end_nosignal;
 217                 }
 218
 219                 if (ctx->on_recv_stream) {
 220                         ret = ctx->on_recv_stream(new_stream);
 221                         if (ret < 0) {
 222                                 consumer_del_stream(new_stream, NULL);
 223                                 goto end_nosignal;
 224                         }
 225                 }
 226
 227                 /* Send stream to the metadata thread */
 228                 if (new_stream->metadata_flag) {
 229                         do {
 230                                 ret = write(ctx->consumer_metadata_pipe[1], &new_stream,
 231                                                 sizeof(new_stream));
 232                         } while (ret < 0 && errno == EINTR);
 233                         if (ret < 0) {
 234                                 PERROR("write metadata pipe");
 235                                 consumer_del_stream(new_stream, NULL);
 236                         }
 237                 } else {
 238                         ret = consumer_add_stream(new_stream);
 239                         if (ret) {
 240                                 ERR("Consumer add stream %d failed. Continuing",
 241                                                 new_stream->key);
 242                                 consumer_del_stream(new_stream, NULL);
 243                                 goto end_nosignal;
 244                         }
 245                 }
 246
 247                 DBG("Kernel consumer_add_stream (%d)", fd);
 248                 break;
 249         }
 250         case LTTNG_CONSUMER_UPDATE_STREAM:
 251         {
 252                 rcu_read_unlock();
 253                 return -ENOSYS;
 254         }
 255         case LTTNG_CONSUMER_DESTROY_RELAYD:
 256         {
 257                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 258                 struct consumer_relayd_sock_pair *relayd;
 259
 260                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 261
 262                 /* Get relayd reference if exists. */
 263                 relayd = consumer_find_relayd(index);
 264                 if (relayd == NULL) {
 265                         ERR("Unable to find relayd %" PRIu64, index);
 266                         goto end_nosignal;
 267                 }
 268
 269                 /*
 270                  * Each relayd socket pair has a refcount of stream attached to it
 271                  * which tells if the relayd is still active or not depending on the
 272                  * refcount value.
 273                  *
 274                  * This will set the destroy flag of the relayd object and destroy it
 275                  * if the refcount reaches zero when called.
 276                  *
 277                  * The destroy can happen either here or when a stream fd hangs up.
 278                  */
 279                 consumer_flag_relayd_for_destroy(relayd);
 280
 281                 goto end_nosignal;
 282         }
 283         default:
 284                 goto end_nosignal;
 285         }
 286
 287         /*
 288          * Wake-up the other end by writing a null byte in the pipe (non-blocking).
 289          * Important note: Because writing into the pipe is non-blocking (and
 290          * therefore we allow dropping wakeup data, as long as there is wakeup data
 291          * present in the pipe buffer to wake up the other end), the other end
 292          * should perform the following sequence for waiting:
 293          *
 294          * 1) empty the pipe (reads).
 295          * 2) perform update operation.
 296          * 3) wait on the pipe (poll).
 297          */
 298         do {
 299                 ret = write(ctx->consumer_poll_pipe[1], "", 1);
 300         } while (ret < 0 && errno == EINTR);
 301 end_nosignal:
 302         rcu_read_unlock();
 303
 304         /*
 305          * Return 1 to indicate success since the 0 value can be a socket
 306          * shutdown during the recv() or send() call.
 307          */
 308         return 1;
 309 }
 310
 311 /*
 312  * Consume data on a file descriptor and write it on a trace file.
 313  */
 314 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
 315                 struct lttng_consumer_local_data *ctx)
 316 {
 317         unsigned long len, subbuf_size, padding;
 318         int err;
 319         ssize_t ret = 0;
 320         int infd = stream->wait_fd;
 321
 322         DBG("In read_subbuffer (infd : %d)", infd);
 323         /* Get the next subbuffer */
 324         err = kernctl_get_next_subbuf(infd);
 325         if (err != 0) {
 326                 ret = err;
 327                 /*
 328                  * This is a debug message even for single-threaded consumer,
 329                  * because poll() have more relaxed criterions than get subbuf,
 330                  * so get_subbuf may fail for short race windows where poll()
 331                  * would issue wakeups.
 332                  */
 333                 DBG("Reserving sub buffer failed (everything is normal, "
 334                                 "it is due to concurrency)");
 335                 goto end;
 336         }
 337
 338         /* Get the full subbuffer size including padding */
 339         err = kernctl_get_padded_subbuf_size(infd, &len);
 340         if (err != 0) {
 341                 errno = -err;
 342                 perror("Getting sub-buffer len failed.");
 343                 ret = err;
 344                 goto end;
 345         }
 346
 347         switch (stream->output) {
 348         case LTTNG_EVENT_SPLICE:
 349
 350                 /*
 351                  * XXX: The lttng-modules splice "actor" does not handle copying
 352                  * partial pages hence only using the subbuffer size without the
 353                  * padding makes the splice fail.
 354                  */
 355                 subbuf_size = len;
 356                 padding = 0;
 357
 358                 /* splice the subbuffer to the tracefile */
 359                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
 360                                 padding);
 361                 /*
 362                  * XXX: Splice does not support network streaming so the return value
 363                  * is simply checked against subbuf_size and not like the mmap() op.
 364                  */
 365                 if (ret != subbuf_size) {
 366                         /*
 367                          * display the error but continue processing to try
 368                          * to release the subbuffer
 369                          */
 370                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
 371                                         ret, subbuf_size);
 372                 }
 373                 break;
 374         case LTTNG_EVENT_MMAP:
 375                 /* Get subbuffer size without padding */
 376                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
 377                 if (err != 0) {
 378                         errno = -err;
 379                         perror("Getting sub-buffer len failed.");
 380                         ret = err;
 381                         goto end;
 382                 }
 383
 384                 /* Make sure the tracer is not gone mad on us! */
 385                 assert(len >= subbuf_size);
 386
 387                 padding = len - subbuf_size;
 388
 389                 /* write the subbuffer to the tracefile */
 390                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
 391                                 padding);
 392                 /*
 393                  * The mmap operation should write subbuf_size amount of data when
 394                  * network streaming or the full padding (len) size when we are _not_
 395                  * streaming.
 396                  */
 397                 if ((ret != subbuf_size && stream->net_seq_idx != -1) ||
 398                                 (ret != len && stream->net_seq_idx == -1)) {
 399                         /*
 400                          * Display the error but continue processing to try to release the
 401                          * subbuffer
 402                          */
 403                         ERR("Error writing to tracefile "
 404                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
 405                                         ret, len, subbuf_size);
 406                 }
 407                 break;
 408         default:
 409                 ERR("Unknown output method");
 410                 ret = -1;
 411         }
 412
 413         err = kernctl_put_next_subbuf(infd);
 414         if (err != 0) {
 415                 errno = -err;
 416                 if (errno == EFAULT) {
 417                         perror("Error in unreserving sub buffer\n");
 418                 } else if (errno == EIO) {
 419                         /* Should never happen with newer LTTng versions */
 420                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 421                 }
 422
 423                 ret = -err;
 424                 goto end;
 425         }
 426
 427 end:
 428         return ret;
 429 }
 430
 431 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
 432 {
 433         int ret;
 434
 435         /* Opening the tracefile in write mode */
 436         if (strlen(stream->path_name) > 0 && stream->net_seq_idx == -1) {
 437                 ret = run_as_open(stream->path_name,
 438                                 O_WRONLY|O_CREAT|O_TRUNC,
 439                                 S_IRWXU|S_IRWXG|S_IRWXO,
 440                                 stream->uid, stream->gid);
 441                 if (ret < 0) {
 442                         ERR("Opening %s", stream->path_name);
 443                         perror("open");
 444                         goto error;
 445                 }
 446                 stream->out_fd = ret;
 447         }
 448
 449         if (stream->output == LTTNG_EVENT_MMAP) {
 450                 /* get the len of the mmap region */
 451                 unsigned long mmap_len;
 452
 453                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
 454                 if (ret != 0) {
 455                         errno = -ret;
 456                         perror("kernctl_get_mmap_len");
 457                         goto error_close_fd;
 458                 }
 459                 stream->mmap_len = (size_t) mmap_len;
 460
 461                 stream->mmap_base = mmap(NULL, stream->mmap_len,
 462                                 PROT_READ, MAP_PRIVATE, stream->wait_fd, 0);
 463                 if (stream->mmap_base == MAP_FAILED) {
 464                         perror("Error mmaping");
 465                         ret = -1;
 466                         goto error_close_fd;
 467                 }
 468         }
 469
 470         /* we return 0 to let the library handle the FD internally */
 471         return 0;
 472
 473 error_close_fd:
 474         {
 475                 int err;
 476
 477                 err = close(stream->out_fd);
 478                 assert(!err);
 479         }
 480 error:
 481         return ret;
 482 }
 483