int sock, struct pollfd *consumer_sockpoll)
{
ssize_t ret;
+ enum lttng_error_code ret_code = LTTNG_OK;
struct lttcomm_consumer_msg msg;
ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
if (ret != sizeof(msg)) {
- lttng_consumer_send_error(ctx, CONSUMERD_ERROR_RECV_CMD);
+ lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
return ret;
}
if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
+ /*
+ * Notify the session daemon that the command is completed.
+ *
+ * On transport layer error, the function call will print an error
+ * message so handling the returned code is a bit useless since we
+ * return an error code anyway.
+ */
+ (void) consumer_send_status_msg(sock, ret_code);
return -ENOENT;
}
switch (msg.cmd_type) {
case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
{
+ /* Session daemon status message are handled in the following call. */
ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
- &msg.u.relayd_sock.sock);
+ &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
goto end_nosignal;
}
case LTTNG_CONSUMER_ADD_CHANNEL:
{
struct lttng_consumer_channel *new_channel;
+ /* First send a status message before receiving the fds. */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto end_nosignal;
+ }
+
DBG("consumer_add_channel %d", msg.u.channel.channel_key);
new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
-1, -1,
msg.u.channel.mmap_len,
- msg.u.channel.max_sb_size);
+ msg.u.channel.max_sb_size,
+ msg.u.channel.nb_init_streams);
if (new_channel == NULL) {
- lttng_consumer_send_error(ctx, CONSUMERD_OUTFD_ERROR);
+ lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
goto end_nosignal;
}
if (ctx->on_recv_channel != NULL) {
}
case LTTNG_CONSUMER_ADD_STREAM:
{
- int fd;
+ int fd, stream_pipe;
struct consumer_relayd_sock_pair *relayd = NULL;
struct lttng_consumer_stream *new_stream;
+ int alloc_ret = 0;
+
+ /* First send a status message before receiving the fds. */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto end_nosignal;
+ }
/* block */
if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
/* Get stream file descriptor from socket */
ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
if (ret != sizeof(fd)) {
- lttng_consumer_send_error(ctx, CONSUMERD_ERROR_RECV_FD);
+ lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
rcu_read_unlock();
return ret;
}
+ /*
+ * Send status code to session daemon only if the recv works. If the
+ * above recv() failed, the session daemon is notified through the
+ * error socket and the teardown is eventually done.
+ */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto end_nosignal;
+ }
+
new_stream = consumer_allocate_stream(msg.u.stream.channel_key,
msg.u.stream.stream_key,
fd, fd,
msg.u.stream.uid,
msg.u.stream.gid,
msg.u.stream.net_index,
- msg.u.stream.metadata_flag);
+ msg.u.stream.metadata_flag,
+ msg.u.stream.session_id,
+ &alloc_ret);
if (new_stream == NULL) {
- lttng_consumer_send_error(ctx, CONSUMERD_OUTFD_ERROR);
+ switch (alloc_ret) {
+ case -ENOMEM:
+ case -EINVAL:
+ default:
+ lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
+ break;
+ case -ENOENT:
+ /*
+ * We could not find the channel. Can happen if cpu hotplug
+ * happens while tearing down.
+ */
+ DBG3("Could not find channel");
+ break;
+ }
goto end_nosignal;
}
+ /*
+ * The buffer flush is done on the session daemon side for the kernel
+ * so no need for the stream "hangup_flush_done" variable to be
+ * tracked. This is important for a kernel stream since we don't rely
+ * on the flush state of the stream to read data. It's not the case for
+ * user space tracing.
+ */
+ new_stream->hangup_flush_done = 0;
+
/* The stream is not metadata. Get relayd reference if exists. */
relayd = consumer_find_relayd(msg.u.stream.net_index);
if (relayd != NULL) {
&new_stream->relayd_stream_id);
pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
if (ret < 0) {
+ consumer_del_stream(new_stream, NULL);
goto end_nosignal;
}
} else if (msg.u.stream.net_index != -1) {
ERR("Network sequence index %d unknown. Not adding stream.",
msg.u.stream.net_index);
- free(new_stream);
+ consumer_del_stream(new_stream, NULL);
goto end_nosignal;
}
- if (ctx->on_recv_stream != NULL) {
+ if (ctx->on_recv_stream) {
ret = ctx->on_recv_stream(new_stream);
- if (ret == 0) {
- consumer_add_stream(new_stream);
- } else if (ret < 0) {
+ if (ret < 0) {
+ consumer_del_stream(new_stream, NULL);
goto end_nosignal;
}
+ }
+
+ /* Get the right pipe where the stream will be sent. */
+ if (new_stream->metadata_flag) {
+ stream_pipe = ctx->consumer_metadata_pipe[1];
} else {
- consumer_add_stream(new_stream);
+ stream_pipe = ctx->consumer_data_pipe[1];
+ }
+
+ do {
+ ret = write(stream_pipe, &new_stream, sizeof(new_stream));
+ } while (ret < 0 && errno == EINTR);
+ if (ret < 0) {
+ PERROR("Consumer write %s stream to pipe %d",
+ new_stream->metadata_flag ? "metadata" : "data",
+ stream_pipe);
+ consumer_del_stream(new_stream, NULL);
+ goto end_nosignal;
}
- DBG("Kernel consumer_add_stream (%d)", fd);
+ DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
+ msg.u.stream.path_name, fd, new_stream->relayd_stream_id);
break;
}
case LTTNG_CONSUMER_UPDATE_STREAM:
/* Get relayd reference if exists. */
relayd = consumer_find_relayd(index);
if (relayd == NULL) {
- ERR("Unable to find relayd %" PRIu64, index);
- goto end_nosignal;
+ DBG("Unable to find relayd %" PRIu64, index);
+ ret_code = LTTNG_ERR_NO_CONSUMER;
}
/*
*
* The destroy can happen either here or when a stream fd hangs up.
*/
- consumer_flag_relayd_for_destroy(relayd);
+ if (relayd) {
+ consumer_flag_relayd_for_destroy(relayd);
+ }
+
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto end_nosignal;
+ }
goto end_nosignal;
}
+ case LTTNG_CONSUMER_DATA_PENDING:
+ {
+ int32_t ret;
+ uint64_t id = msg.u.data_pending.session_id;
+
+ DBG("Kernel consumer data pending command for id %" PRIu64, id);
+
+ ret = consumer_data_pending(id);
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
+ if (ret < 0) {
+ PERROR("send data pending ret code");
+ }
+
+ /*
+ * No need to send back a status message since the data pending
+ * returned value is the response.
+ */
+ break;
+ }
default:
goto end_nosignal;
}
- /*
- * Wake-up the other end by writing a null byte in the pipe (non-blocking).
- * Important note: Because writing into the pipe is non-blocking (and
- * therefore we allow dropping wakeup data, as long as there is wakeup data
- * present in the pipe buffer to wake up the other end), the other end
- * should perform the following sequence for waiting:
- *
- * 1) empty the pipe (reads).
- * 2) perform update operation.
- * 3) wait on the pipe (poll).
- */
- do {
- ret = write(ctx->consumer_poll_pipe[1], "", 1);
- } while (ret < 0 && errno == EINTR);
end_nosignal:
rcu_read_unlock();
- return 0;
+
+ /*
+ * Return 1 to indicate success since the 0 value can be a socket
+ * shutdown during the recv() or send() call.
+ */
+ return 1;
}
/*
ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
struct lttng_consumer_local_data *ctx)
{
- unsigned long len;
+ unsigned long len, subbuf_size, padding;
int err;
ssize_t ret = 0;
int infd = stream->wait_fd;
/* Get the next subbuffer */
err = kernctl_get_next_subbuf(infd);
if (err != 0) {
+ ret = err;
/*
* This is a debug message even for single-threaded consumer,
* because poll() have more relaxed criterions than get subbuf,
goto end;
}
+ /* Get the full subbuffer size including padding */
+ err = kernctl_get_padded_subbuf_size(infd, &len);
+ if (err != 0) {
+ errno = -err;
+ perror("Getting sub-buffer len failed.");
+ ret = err;
+ goto end;
+ }
+
switch (stream->output) {
- case LTTNG_EVENT_SPLICE:
- /* read the whole subbuffer */
- err = kernctl_get_padded_subbuf_size(infd, &len);
- if (err != 0) {
- errno = -ret;
- perror("Getting sub-buffer len failed.");
- goto end;
- }
+ case LTTNG_EVENT_SPLICE:
- /* splice the subbuffer to the tracefile */
- ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, len);
- if (ret != len) {
- /*
- * display the error but continue processing to try
- * to release the subbuffer
- */
- ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
- ret, len);
- }
+ /*
+ * XXX: The lttng-modules splice "actor" does not handle copying
+ * partial pages hence only using the subbuffer size without the
+ * padding makes the splice fail.
+ */
+ subbuf_size = len;
+ padding = 0;
- break;
- case LTTNG_EVENT_MMAP:
- /* read the used subbuffer size */
- err = kernctl_get_padded_subbuf_size(infd, &len);
- if (err != 0) {
- errno = -ret;
- perror("Getting sub-buffer len failed.");
- goto end;
- }
- /* write the subbuffer to the tracefile */
- ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len);
- if (ret != len) {
- /*
- * display the error but continue processing to try
- * to release the subbuffer
- */
- ERR("Error writing to tracefile");
- }
- break;
- default:
- ERR("Unknown output method");
- ret = -1;
+ /* splice the subbuffer to the tracefile */
+ ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
+ padding);
+ /*
+ * XXX: Splice does not support network streaming so the return value
+ * is simply checked against subbuf_size and not like the mmap() op.
+ */
+ if (ret != subbuf_size) {
+ /*
+ * display the error but continue processing to try
+ * to release the subbuffer
+ */
+ ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
+ ret, subbuf_size);
+ }
+ break;
+ case LTTNG_EVENT_MMAP:
+ /* Get subbuffer size without padding */
+ err = kernctl_get_subbuf_size(infd, &subbuf_size);
+ if (err != 0) {
+ errno = -err;
+ perror("Getting sub-buffer len failed.");
+ ret = err;
+ goto end;
+ }
+
+ /* Make sure the tracer is not gone mad on us! */
+ assert(len >= subbuf_size);
+
+ padding = len - subbuf_size;
+
+ /* write the subbuffer to the tracefile */
+ ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
+ padding);
+ /*
+ * The mmap operation should write subbuf_size amount of data when
+ * network streaming or the full padding (len) size when we are _not_
+ * streaming.
+ */
+ if ((ret != subbuf_size && stream->net_seq_idx != -1) ||
+ (ret != len && stream->net_seq_idx == -1)) {
+ /*
+ * Display the error but continue processing to try to release the
+ * subbuffer
+ */
+ ERR("Error writing to tracefile "
+ "(ret: %zd != len: %lu != subbuf_size: %lu)",
+ ret, len, subbuf_size);
+ }
+ break;
+ default:
+ ERR("Unknown output method");
+ ret = -1;
}
err = kernctl_put_next_subbuf(infd);
if (err != 0) {
- errno = -ret;
+ errno = -err;
if (errno == EFAULT) {
perror("Error in unreserving sub buffer\n");
} else if (errno == EIO) {
/* Should never happen with newer LTTng versions */
perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
}
+
+ ret = -err;
goto end;
}
return ret;
}
+/*
+ * Check if data is still being extracted from the buffers for a specific
+ * stream. Consumer data lock MUST be acquired before calling this function
+ * and the stream lock.
+ *
+ * Return 1 if the traced data are still getting read else 0 meaning that the
+ * data is available for trace viewer reading.
+ */
+int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
+{
+ int ret;
+
+ assert(stream);
+
+ ret = kernctl_get_next_subbuf(stream->wait_fd);
+ if (ret == 0) {
+ /* There is still data so let's put back this subbuffer. */
+ ret = kernctl_put_subbuf(stream->wait_fd);
+ assert(ret == 0);
+ ret = 1; /* Data is pending */
+ goto end;
+ }
+
+ /* Data is NOT pending and ready to be read. */
+ ret = 0;
+
+end:
+ return ret;
+}