static struct relay_conn_queue relay_conn_queue;
/* Cap of file desriptors to be in simultaneous use by the relay daemon. */
-static unsigned int lttng_opt_fd_cap;
+static unsigned int lttng_opt_fd_pool_size = -1;
/* Global relay stream hash table. */
struct lttng_ht *relay_streams_ht;
{ "daemonize", 0, 0, 'd', },
{ "background", 0, 0, 'b', },
{ "group", 1, 0, 'g', },
- { "fd-cap", 1, 0, '\0', },
+ { "fd-pool-size", 1, 0, '\0', },
{ "help", 0, 0, 'h', },
{ "output", 1, 0, 'o', },
{ "verbose", 0, 0, 'v', },
switch (opt) {
case 0:
- if (!strcmp(optname, "fd-cap")) {
+ if (!strcmp(optname, "fd-pool-size")) {
unsigned long v;
errno = 0;
v = strtoul(arg, NULL, 0);
if (errno != 0 || !isdigit(arg[0])) {
- ERR("Wrong value in --fd-cap parameter: %s", arg);
+ ERR("Wrong value in --fd-pool-size parameter: %s", arg);
ret = -1;
goto end;
}
- if (v < DEFAULT_RELAYD_MINIMAL_FD_CAP) {
- ERR("File descriptor cap must be set to at least %d",
- DEFAULT_RELAYD_MINIMAL_FD_CAP);
- }
if (v >= UINT_MAX) {
- ERR("File descriptor cap overflow in --fd-cap parameter: %s", arg);
+ ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg);
ret = -1;
goto end;
}
- lttng_opt_fd_cap = (unsigned int) v;
- DBG3("File descriptor cap set to %u", lttng_opt_fd_cap);
-
+ lttng_opt_fd_pool_size = (unsigned int) v;
} else {
fprintf(stderr, "unknown option %s", optname);
if (arg) {
}
}
+static int set_fd_pool_size(void)
+{
+ int ret = 0;
+ struct rlimit rlimit;
+
+ ret = getrlimit(RLIMIT_NOFILE, &rlimit);
+ if (ret) {
+ PERROR("Failed to get file descriptor limit");
+ ret = -1;
+ goto end;
+ }
+
+ DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
+ (uint64_t) rlimit.rlim_cur, (uint64_t) rlimit.rlim_max);
+ if (lttng_opt_fd_pool_size == -1) {
+ /* Use default value (soft limit - reserve). */
+ if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
+ ERR("The process' file number limit is too low (%" PRIu64 "). The process' file number limit must be set to at least %i.",
+ (uint64_t) rlimit.rlim_cur,
+ DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
+ ret = -1;
+ goto end;
+ }
+ lttng_opt_fd_pool_size = rlimit.rlim_cur -
+ DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
+ goto end;
+ }
+
+ if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
+ ERR("File descriptor pool size must be set to at least %d",
+ DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
+ ret = -1;
+ goto end;
+ }
+
+ if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
+ ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%lu).",
+ lttng_opt_fd_pool_size, rlimit.rlim_cur);
+ ret = -1;
+ goto end;
+ }
+
+
+ DBG("File descriptor pool size argument (%u) adjusted to %u to accomodate transient fd uses",
+ lttng_opt_fd_pool_size,
+ lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
+ lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
+end:
+ return ret;
+}
+
static int set_options(int argc, char **argv)
{
int c, ret = 0, option_index = 0, retval = 0;
goto exit;
}
}
- if (lttng_opt_fd_cap == 0) {
- int ret;
- struct rlimit rlimit;
-
- ret = getrlimit(RLIMIT_NOFILE, &rlimit);
- if (ret) {
- PERROR("Failed to get file descriptor limit");
- retval = -1;
- }
-
- lttng_opt_fd_cap = rlimit.rlim_cur;
+ ret = set_fd_pool_size();
+ if (ret) {
+ retval = -1;
+ goto exit;
}
if (!opt_group_output_by_session && !opt_group_output_by_host) {
goto end_unlock;
}
- ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
- 0, 0, -1, -1, stream->stream_fd->fd, NULL,
- &stream->stream_fd->fd);
+ ret = stream_fd_rotate(stream->stream_fd,
+ stream->path_name, stream->channel_name, 0, 0, NULL);
if (ret < 0) {
ERR("Failed to rotate metadata file %s of channel %s",
stream->path_name, stream->channel_name);
struct lttcomm_relayd_metadata_payload metadata_payload_header;
struct relay_stream *metadata_stream;
uint64_t metadata_payload_size;
+ int metadata_fd = -1;
if (!session) {
ERR("Metadata sent before version check");
pthread_mutex_lock(&metadata_stream->lock);
- size_ret = lttng_write(metadata_stream->stream_fd->fd,
+ metadata_fd = stream_fd_get_fd(metadata_stream->stream_fd);
+ if (metadata_fd < 0) {
+ goto end_put;
+ }
+ size_ret = lttng_write(metadata_fd,
payload->data + sizeof(metadata_payload_header),
metadata_payload_size);
if (size_ret < metadata_payload_size) {
ERR("Relay error writing metadata on file");
ret = -1;
- goto end_put;
+ goto end_put_fd;
}
- size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
+ size_ret = write_padding_to_file(metadata_fd,
metadata_payload_header.padding_size);
if (size_ret < (int64_t) metadata_payload_header.padding_size) {
ret = -1;
- goto end_put;
+ goto end_put_fd;
}
metadata_stream->metadata_received +=
DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
metadata_stream->metadata_received);
+end_put_fd:
+ stream_fd_put_fd(metadata_stream->stream_fd);
end_put:
pthread_mutex_unlock(&metadata_stream->lock);
stream_put(metadata_stream);
/* no flush. */
ret = 0;
} else {
+ /*
+ * ret < 0
+ *
+ * relay_index_try_flush is responsible for the self-reference
+ * put of the index object on error.
+ */
ERR("relay_index_try_flush error %d", ret);
- relay_index_put(index);
ret = -1;
}
/* Put ref on previous index_file. */
if (stream->index_file) {
- lttng_index_file_put(stream->index_file);
+ relay_index_file_put(stream->index_file);
stream->index_file = NULL;
}
major = stream->trace->session->major;
minor = stream->trace->session->minor;
- stream->index_file = lttng_index_file_create(stream->path_name,
+ stream->index_file = relay_index_file_create(stream->path_name,
stream->channel_name,
- -1, -1, stream->tracefile_size,
+ stream->tracefile_size,
tracefile_array_get_file_index_head(stream->tfa),
lttng_to_index_major(major, minor),
lttng_to_index_minor(major, minor));
/* No flush. */
ret = 0;
} else {
- /* Put self-ref for this index due to error. */
- relay_index_put(index);
- index = NULL;
+ /*
+ * ret < 0
+ *
+ * relay_index_try_flush is responsible for the self-reference
+ * put of the index object on error.
+ */
+ ERR("relay_index_try_flush error %d", ret);
ret = -1;
}
end:
/* new_id is updated by utils_rotate_stream_file. */
new_id = old_id;
- ret = utils_rotate_stream_file(stream->path_name,
+ ret = stream_fd_rotate(stream->stream_fd, stream->path_name,
stream->channel_name, stream->tracefile_size,
- stream->tracefile_count, -1,
- -1, stream->stream_fd->fd,
- &new_id, &stream->stream_fd->fd);
+ stream->tracefile_count, &new_id);
if (ret < 0) {
ERR("Failed to rotate stream output file");
status = RELAY_CONNECTION_STATUS_ERROR;
bool new_stream = false, close_requested = false;
uint64_t left_to_receive = state->left_to_receive;
struct relay_session *session;
+ int stream_fd = -1;
DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive",
state->header.stream_id, state->header.net_seq_num,
}
}
+ stream_fd = stream_fd_get_fd(stream->stream_fd);
+ if (stream_fd < 0) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end_stream_unlock;
+ }
+
/*
* The size of the "chunk" received on any iteration is bounded by:
* - the data left to receive,
PERROR("Socket %d error", conn->sock->fd);
status = RELAY_CONNECTION_STATUS_ERROR;
}
- goto end_stream_unlock;
+ goto end_put_fd;
} else if (ret == 0) {
/* No more data ready to be consumed on socket. */
DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
recv_size = ret;
/* Write data to stream output fd. */
- write_ret = lttng_write(stream->stream_fd->fd, data_buffer,
+ write_ret = lttng_write(stream_fd, data_buffer,
recv_size);
if (write_ret < (ssize_t) recv_size) {
ERR("Relay error writing data to file");
status = RELAY_CONNECTION_STATUS_ERROR;
- goto end_stream_unlock;
+ goto end_put_fd;
}
left_to_receive -= recv_size;
DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive",
state->header.stream_id, state->received,
state->left_to_receive);
- goto end_stream_unlock;
+ goto end_put_fd;
}
- ret = write_padding_to_file(stream->stream_fd->fd,
+ ret = write_padding_to_file(stream_fd,
state->header.padding_size);
if ((int64_t) ret < (int64_t) state->header.padding_size) {
ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
stream->stream_handle,
state->header.net_seq_num, ret);
status = RELAY_CONNECTION_STATUS_ERROR;
- goto end_stream_unlock;
+ goto end_put_fd;
}
stream->stream_handle,
state->header.net_seq_num, ret);
status = RELAY_CONNECTION_STATUS_ERROR;
- goto end_stream_unlock;
+ goto end_put_fd;
}
}
connection_reset_protocol_state(conn);
state = NULL;
+end_put_fd:
+ stream_fd_put_fd(stream->stream_fd);
end_stream_unlock:
close_requested = stream->close_requested;
pthread_mutex_unlock(&stream->lock);
*/
rcu_register_thread();
- the_fd_tracker = fd_tracker_create(lttng_opt_fd_cap);
+ the_fd_tracker = fd_tracker_create(lttng_opt_fd_pool_size);
if (!the_fd_tracker) {
retval = -1;
goto exit_options;
ret = track_stdio();
if (ret) {
retval = -1;
- goto exit_options;
+ goto exit_tracker;
}
/* Initialize thread health monitoring */
exit_init_data:
health_app_destroy(health_relayd);
exit_health_app_create:
-exit_options:
+
/*
* Wait for all pending call_rcu work to complete before tearing
* down data structures. call_rcu worker may be trying to
/* Ensure all prior call_rcu are done. */
rcu_barrier();
-
+exit_tracker:
untrack_stdio();
/*
* fd_tracker_destroy() will log the contents of the fd-tracker
*/
fd_tracker_destroy(the_fd_tracker);
rcu_unregister_thread();
-
+exit_options:
if (!retval) {
exit(EXIT_SUCCESS);
} else {
projects / lttng-tools.git / blobdiff