X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fconsumer.c;h=b6e440a486fc5995f9a7caca0eac39631bf3340c;hp=09b3bee330457d46483a85d1f2f5402e25241b9e;hb=1624d5b7ba86633d36f3a5c86ea1df5d308c4360;hpb=ffe600149a7608221985751e1bf293234bf2545c diff --git a/src/common/consumer.c b/src/common/consumer.c index 09b3bee33..b6e440a48 100644 --- a/src/common/consumer.c +++ b/src/common/consumer.c @@ -28,6 +28,7 @@ #include #include #include +#include #include #include @@ -47,6 +48,16 @@ struct lttng_consumer_global_data consumer_data = { .type = LTTNG_CONSUMER_UNKNOWN, }; +enum consumer_channel_action { + CONSUMER_CHANNEL_ADD, + CONSUMER_CHANNEL_QUIT, +}; + +struct consumer_channel_msg { + enum consumer_channel_action action; + struct lttng_consumer_channel *chan; +}; + /* * Flag to inform the polling thread to quit when all fd hung up. Updated by * the consumer_thread_receive_fds when it notices that all fds has hung up. @@ -78,28 +89,59 @@ static void notify_thread_pipe(int wpipe) } while (ret < 0 && errno == EINTR); } +static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, + struct lttng_consumer_channel *chan, + enum consumer_channel_action action) +{ + struct consumer_channel_msg msg; + int ret; + + msg.action = action; + msg.chan = chan; + do { + ret = write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); + } while (ret < 0 && errno == EINTR); +} + +static int read_channel_pipe(struct lttng_consumer_local_data *ctx, + struct lttng_consumer_channel **chan, + enum consumer_channel_action *action) +{ + struct consumer_channel_msg msg; + int ret; + + do { + ret = read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); + } while (ret < 0 && errno == EINTR); + if (ret > 0) { + *action = msg.action; + *chan = msg.chan; + } + return ret; +} + /* * Find a stream. The consumer_data.lock must be locked during this * call. */ -static struct lttng_consumer_stream *find_stream(int key, +static struct lttng_consumer_stream *find_stream(uint64_t key, struct lttng_ht *ht) { struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; struct lttng_consumer_stream *stream = NULL; assert(ht); - /* Negative keys are lookup failures */ - if (key < 0) { + /* -1ULL keys are lookup failures */ + if (key == (uint64_t) -1ULL) { return NULL; } rcu_read_lock(); - lttng_ht_lookup(ht, (void *)((unsigned long) key), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + lttng_ht_lookup(ht, &key, &iter); + node = lttng_ht_iter_get_node_u64(&iter); if (node != NULL) { stream = caa_container_of(node, struct lttng_consumer_stream, node); } @@ -116,13 +158,13 @@ static void steal_stream_key(int key, struct lttng_ht *ht) rcu_read_lock(); stream = find_stream(key, ht); if (stream) { - stream->key = -1; + stream->key = -1ULL; /* * We don't want the lookup to match, but we still need * to iterate on this stream when iterating over the hash table. Just * change the node key. */ - stream->node.key = -1; + stream->node.key = -1ULL; } rcu_read_unlock(); } @@ -133,19 +175,19 @@ static void steal_stream_key(int key, struct lttng_ht *ht) * RCU read side lock MUST be acquired before calling this function and * protects the channel ptr. */ -struct lttng_consumer_channel *consumer_find_channel(unsigned long key) +struct lttng_consumer_channel *consumer_find_channel(uint64_t key) { struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; struct lttng_consumer_channel *channel = NULL; - /* Negative keys are lookup failures */ - if (key < 0) { + /* -1ULL keys are lookup failures */ + if (key == (uint64_t) -1ULL) { return NULL; } - lttng_ht_lookup(consumer_data.channel_ht, (void *) key, &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); + node = lttng_ht_iter_get_node_u64(&iter); if (node != NULL) { channel = caa_container_of(node, struct lttng_consumer_channel, node); } @@ -155,8 +197,8 @@ struct lttng_consumer_channel *consumer_find_channel(unsigned long key) static void free_stream_rcu(struct rcu_head *head) { - struct lttng_ht_node_ulong *node = - caa_container_of(head, struct lttng_ht_node_ulong, head); + struct lttng_ht_node_u64 *node = + caa_container_of(head, struct lttng_ht_node_u64, head); struct lttng_consumer_stream *stream = caa_container_of(node, struct lttng_consumer_stream, node); @@ -165,8 +207,8 @@ static void free_stream_rcu(struct rcu_head *head) static void free_channel_rcu(struct rcu_head *head) { - struct lttng_ht_node_ulong *node = - caa_container_of(head, struct lttng_ht_node_ulong, head); + struct lttng_ht_node_u64 *node = + caa_container_of(head, struct lttng_ht_node_u64, head); struct lttng_consumer_channel *channel = caa_container_of(node, struct lttng_consumer_channel, node); @@ -178,8 +220,8 @@ static void free_channel_rcu(struct rcu_head *head) */ static void free_relayd_rcu(struct rcu_head *head) { - struct lttng_ht_node_ulong *node = - caa_container_of(head, struct lttng_ht_node_ulong, head); + struct lttng_ht_node_u64 *node = + caa_container_of(head, struct lttng_ht_node_u64, head); struct consumer_relayd_sock_pair *relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); @@ -233,7 +275,7 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) int ret; struct lttng_ht_iter iter; - DBG("Consumer delete channel key %d", channel->key); + DBG("Consumer delete channel key %" PRIu64, channel->key); pthread_mutex_lock(&consumer_data.lock); @@ -425,7 +467,10 @@ void consumer_del_stream(struct lttng_consumer_stream *stream, ret = lttng_ht_del(ht, &iter); assert(!ret); - /* Remove node session id from the consumer_data stream ht */ + iter.iter.node = &stream->node_channel_id.node; + ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter); + assert(!ret); + iter.iter.node = &stream->node_session_id.node; ret = lttng_ht_del(consumer_data.stream_list_ht, &iter); assert(!ret); @@ -490,8 +535,8 @@ free_stream_rcu: call_rcu(&stream->node.head, free_stream_rcu); } -struct lttng_consumer_stream *consumer_allocate_stream(int channel_key, - int stream_key, +struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, + uint64_t stream_key, enum lttng_consumer_stream_state state, const char *channel_name, uid_t uid, @@ -540,13 +585,16 @@ struct lttng_consumer_stream *consumer_allocate_stream(int channel_key, } /* Key is always the wait_fd for streams. */ - lttng_ht_node_init_ulong(&stream->node, stream->key); + lttng_ht_node_init_u64(&stream->node, stream->key); + + /* Init node per channel id key */ + lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); /* Init session id node with the stream session id */ - lttng_ht_node_init_ulong(&stream->node_session_id, stream->session_id); + lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); - DBG3("Allocated stream %s (key %d, relayd_id %d, session_id %" PRIu64, - stream->name, stream->key, stream->net_seq_idx, stream->session_id); + DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 " relayd_id %" PRIu64 ", session_id %" PRIu64, + stream->name, stream->key, channel_key, stream->net_seq_idx, stream->session_id); rcu_read_unlock(); return stream; @@ -573,7 +621,7 @@ static int add_stream(struct lttng_consumer_stream *stream, assert(stream); assert(ht); - DBG3("Adding consumer stream %d", stream->key); + DBG3("Adding consumer stream %" PRIu64, stream->key); pthread_mutex_lock(&consumer_data.lock); pthread_mutex_lock(&stream->lock); @@ -582,14 +630,17 @@ static int add_stream(struct lttng_consumer_stream *stream, /* Steal stream identifier to avoid having streams with the same key */ steal_stream_key(stream->key, ht); - lttng_ht_add_unique_ulong(ht, &stream->node); + lttng_ht_add_unique_u64(ht, &stream->node); + + lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, + &stream->node_channel_id); /* * Add stream to the stream_list_ht of the consumer data. No need to steal * the key since the HT does not use it and we allow to add redundant keys * into this table. */ - lttng_ht_add_ulong(consumer_data.stream_list_ht, &stream->node_session_id); + lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); /* Check and cleanup relayd */ relayd = consumer_find_relayd(stream->net_seq_idx); @@ -629,18 +680,18 @@ static int add_stream(struct lttng_consumer_stream *stream, static int add_relayd(struct consumer_relayd_sock_pair *relayd) { int ret = 0; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; struct lttng_ht_iter iter; assert(relayd); lttng_ht_lookup(consumer_data.relayd_ht, - (void *)((unsigned long) relayd->net_seq_idx), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + &relayd->net_seq_idx, &iter); + node = lttng_ht_iter_get_node_u64(&iter); if (node != NULL) { goto end; } - lttng_ht_add_unique_ulong(consumer_data.relayd_ht, &relayd->node); + lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); end: return ret; @@ -668,7 +719,7 @@ struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( obj->net_seq_idx = net_seq_idx; obj->refcount = 0; obj->destroy_flag = 0; - lttng_ht_node_init_ulong(&obj->node, obj->net_seq_idx); + lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); error: @@ -682,20 +733,20 @@ error: * RCU read-side lock must be held across this call and while using the * returned object. */ -struct consumer_relayd_sock_pair *consumer_find_relayd(int key) +struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) { struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; struct consumer_relayd_sock_pair *relayd = NULL; /* Negative keys are lookup failures */ - if (key < 0) { + if (key == (uint64_t) -1ULL) { goto error; } - lttng_ht_lookup(consumer_data.relayd_ht, (void *)((unsigned long) key), + lttng_ht_lookup(consumer_data.relayd_ht, &key, &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + node = lttng_ht_iter_get_node_u64(&iter); if (node != NULL) { relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); } @@ -770,14 +821,16 @@ error: * * On error, return NULL. */ -struct lttng_consumer_channel *consumer_allocate_channel(unsigned long key, +struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, uint64_t session_id, const char *pathname, const char *name, uid_t uid, gid_t gid, int relayd_id, - enum lttng_event_output output) + enum lttng_event_output output, + uint64_t tracefile_size, + uint64_t tracefile_count) { struct lttng_consumer_channel *channel; @@ -794,6 +847,8 @@ struct lttng_consumer_channel *consumer_allocate_channel(unsigned long key, channel->gid = gid; channel->relayd_id = relayd_id; channel->output = output; + channel->tracefile_size = tracefile_size; + channel->tracefile_count = tracefile_count; strncpy(channel->pathname, pathname, sizeof(channel->pathname)); channel->pathname[sizeof(channel->pathname) - 1] = '\0'; @@ -801,10 +856,13 @@ struct lttng_consumer_channel *consumer_allocate_channel(unsigned long key, strncpy(channel->name, name, sizeof(channel->name)); channel->name[sizeof(channel->name) - 1] = '\0'; - lttng_ht_node_init_ulong(&channel->node, channel->key); + lttng_ht_node_init_u64(&channel->node, channel->key); + + channel->wait_fd = -1; + CDS_INIT_LIST_HEAD(&channel->streams.head); - DBG("Allocated channel (key %d)", channel->key) + DBG("Allocated channel (key %" PRIu64 ")", channel->key) end: return channel; @@ -813,31 +871,36 @@ end: /* * Add a channel to the global list protected by a mutex. */ -int consumer_add_channel(struct lttng_consumer_channel *channel) +int consumer_add_channel(struct lttng_consumer_channel *channel, + struct lttng_consumer_local_data *ctx) { int ret = 0; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; struct lttng_ht_iter iter; pthread_mutex_lock(&consumer_data.lock); rcu_read_lock(); - lttng_ht_lookup(consumer_data.channel_ht, - (void *)((unsigned long) channel->key), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + lttng_ht_lookup(consumer_data.channel_ht, &channel->key, &iter); + node = lttng_ht_iter_get_node_u64(&iter); if (node != NULL) { /* Channel already exist. Ignore the insertion */ - ERR("Consumer add channel key %d already exists!", channel->key); + ERR("Consumer add channel key %" PRIu64 " already exists!", + channel->key); ret = -1; goto end; } - lttng_ht_add_unique_ulong(consumer_data.channel_ht, &channel->node); + lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); end: rcu_read_unlock(); pthread_mutex_unlock(&consumer_data.lock); + if (!ret && channel->wait_fd != -1 && + channel->metadata_stream == NULL) { + notify_channel_pipe(ctx, channel, CONSUMER_CHANNEL_ADD); + } return ret; } @@ -877,7 +940,12 @@ static int update_poll_array(struct lttng_consumer_local_data *ctx, stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { continue; } - DBG("Active FD %d", stream->wait_fd); + /* + * This clobbers way too much the debug output. Uncomment that if you + * need it for debugging purposes. + * + * DBG("Active FD %d", stream->wait_fd); + */ (*pollfd)[i].fd = stream->wait_fd; (*pollfd)[i].events = POLLIN | POLLPRI; local_stream[i] = stream; @@ -978,6 +1046,8 @@ void lttng_consumer_cleanup(void) cleanup_relayd_ht(); + lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); + /* * This HT contains streams that are freed by either the metadata thread or * the data thread so we do *nothing* on the hash table and simply destroy @@ -1062,7 +1132,7 @@ struct lttng_consumer_local_data *lttng_consumer_create( int (*recv_stream)(struct lttng_consumer_stream *stream), int (*update_stream)(int stream_key, uint32_t state)) { - int ret, i; + int ret; struct lttng_consumer_local_data *ctx; assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || @@ -1076,6 +1146,7 @@ struct lttng_consumer_local_data *lttng_consumer_create( } ctx->consumer_error_socket = -1; + ctx->consumer_metadata_socket = -1; /* assign the callbacks */ ctx->on_buffer_ready = buffer_ready; ctx->on_recv_channel = recv_channel; @@ -1114,6 +1185,12 @@ struct lttng_consumer_local_data *lttng_consumer_create( goto error_thread_pipe; } + ret = pipe(ctx->consumer_channel_pipe); + if (ret < 0) { + PERROR("Error creating channel pipe"); + goto error_channel_pipe; + } + ret = utils_create_pipe(ctx->consumer_metadata_pipe); if (ret < 0) { goto error_metadata_pipe; @@ -1129,26 +1206,14 @@ struct lttng_consumer_local_data *lttng_consumer_create( error_splice_pipe: utils_close_pipe(ctx->consumer_metadata_pipe); error_metadata_pipe: + utils_close_pipe(ctx->consumer_channel_pipe); +error_channel_pipe: utils_close_pipe(ctx->consumer_thread_pipe); error_thread_pipe: - for (i = 0; i < 2; i++) { - int err; - - err = close(ctx->consumer_should_quit[i]); - if (err) { - PERROR("close"); - } - } + utils_close_pipe(ctx->consumer_should_quit); error_poll_fcntl: error_quit_pipe: - for (i = 0; i < 2; i++) { - int err; - - err = close(ctx->consumer_data_pipe[i]); - if (err) { - PERROR("close"); - } - } + utils_close_pipe(ctx->consumer_data_pipe); error_poll_pipe: free(ctx); error: @@ -1168,30 +1233,14 @@ void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) if (ret) { PERROR("close"); } - ret = close(ctx->consumer_thread_pipe[0]); - if (ret) { - PERROR("close"); - } - ret = close(ctx->consumer_thread_pipe[1]); - if (ret) { - PERROR("close"); - } - ret = close(ctx->consumer_data_pipe[0]); - if (ret) { - PERROR("close"); - } - ret = close(ctx->consumer_data_pipe[1]); - if (ret) { - PERROR("close"); - } - ret = close(ctx->consumer_should_quit[0]); - if (ret) { - PERROR("close"); - } - ret = close(ctx->consumer_should_quit[1]); + ret = close(ctx->consumer_metadata_socket); if (ret) { PERROR("close"); } + utils_close_pipe(ctx->consumer_thread_pipe); + utils_close_pipe(ctx->consumer_channel_pipe); + utils_close_pipe(ctx->consumer_data_pipe); + utils_close_pipe(ctx->consumer_should_quit); utils_close_pipe(ctx->consumer_splice_metadata_pipe); unlink(ctx->consumer_command_sock_path); @@ -1238,6 +1287,93 @@ end: return ret; } +/* + * Create the tracefile on disk. + * + * Return 0 on success or else a negative value. + */ +int lttng_create_output_file(struct lttng_consumer_stream *stream) +{ + int ret; + char full_path[PATH_MAX]; + char *path_name_id = NULL; + char *path; + + assert(stream); + assert(stream->net_seq_idx == (uint64_t) -1ULL); + + ret = snprintf(full_path, sizeof(full_path), "%s/%s", + stream->chan->pathname, stream->name); + if (ret < 0) { + PERROR("snprintf create output file"); + goto error; + } + + /* + * If we split the trace in multiple files, we have to add the tracefile + * current count at the end of the tracefile name + */ + if (stream->chan->tracefile_size > 0) { + ret = asprintf(&path_name_id, "%s_%" PRIu64, full_path, + stream->tracefile_count_current); + if (ret < 0) { + PERROR("Allocating path name ID"); + goto error; + } + path = path_name_id; + } else { + path = full_path; + } + + ret = run_as_open(path, O_WRONLY | O_CREAT | O_TRUNC, + S_IRWXU | S_IRWXG | S_IRWXO, stream->uid, stream->gid); + if (ret < 0) { + PERROR("open stream path %s", path); + goto error_open; + } + stream->out_fd = ret; + stream->tracefile_size_current = 0; + +error_open: + free(path_name_id); +error: + return ret; +} + +/* + * Change the output tracefile according to the tracefile_size and + * tracefile_count parameters. The stream lock MUST be held before calling this + * function because we are modifying the stream status. + * + * Return 0 on success or else a negative value. + */ +static int rotate_output_file(struct lttng_consumer_stream *stream) +{ + int ret; + + assert(stream); + assert(stream->tracefile_size_current); + + ret = close(stream->out_fd); + if (ret < 0) { + PERROR("Closing tracefile"); + goto end; + } + + if (stream->chan->tracefile_count > 0) { + stream->tracefile_count_current = + (stream->tracefile_count_current + 1) % + stream->chan->tracefile_count; + } else { + stream->tracefile_count_current++; + } + + return lttng_create_output_file(stream); + +end: + return ret; +} + /* * Mmap the ring buffer, read it and write the data to the tracefile. This is a * core function for writing trace buffers to either the local filesystem or @@ -1289,6 +1425,7 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( goto end; } ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); + break; default: ERR("Unknown consumer_data type"); @@ -1344,6 +1481,21 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( } else { /* No streaming, we have to set the len with the full padding */ len += padding; + + /* + * Check if we need to change the tracefile before writing the packet. + */ + if (stream->chan->tracefile_size > 0 && + (stream->tracefile_size_current + len) > + stream->chan->tracefile_size) { + ret = rotate_output_file(stream); + if (ret < 0) { + ERR("Rotating output file"); + goto end; + } + outfd = stream->out_fd; + } + stream->tracefile_size_current += len; } while (len > 0) { @@ -1506,6 +1658,21 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( } else { /* No streaming, we have to set the len with the full padding */ len += padding; + + /* + * Check if we need to change the tracefile before writing the packet. + */ + if (stream->chan->tracefile_size > 0 && + (stream->tracefile_size_current + len) > + stream->chan->tracefile_size) { + ret = rotate_output_file(stream); + if (ret < 0) { + ERR("Rotating output file"); + goto end; + } + outfd = stream->out_fd; + } + stream->tracefile_size_current += len; } while (len > 0) { @@ -1634,7 +1801,6 @@ int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) assert(0); return -ENOSYS; } - } /* @@ -1728,6 +1894,32 @@ static void destroy_stream_ht(struct lttng_ht *ht) lttng_ht_destroy(ht); } +void lttng_consumer_close_metadata(void) +{ + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + /* + * The Kernel consumer has a different metadata scheme so we don't + * close anything because the stream will be closed by the session + * daemon. + */ + break; + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + /* + * Close all metadata streams. The metadata hash table is passed and + * this call iterates over it by closing all wakeup fd. This is safe + * because at this point we are sure that the metadata producer is + * either dead or blocked. + */ + lttng_ustconsumer_close_metadata(metadata_ht); + break; + default: + ERR("Unknown consumer_data type"); + assert(0); + } +} + /* * Clean up a metadata stream and free its memory. */ @@ -1780,7 +1972,10 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, ret = lttng_ht_del(ht, &iter); assert(!ret); - /* Remove node session id from the consumer_data stream ht */ + iter.iter.node = &stream->node_channel_id.node; + ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter); + assert(!ret); + iter.iter.node = &stream->node_session_id.node; ret = lttng_ht_del(consumer_data.stream_list_ht, &iter); assert(!ret); @@ -1852,12 +2047,12 @@ static int add_metadata_stream(struct lttng_consumer_stream *stream, int ret = 0; struct consumer_relayd_sock_pair *relayd; struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; assert(stream); assert(ht); - DBG3("Adding metadata stream %d to hash table", stream->key); + DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); pthread_mutex_lock(&consumer_data.lock); pthread_mutex_lock(&stream->lock); @@ -1873,8 +2068,8 @@ static int add_metadata_stream(struct lttng_consumer_stream *stream, * Lookup the stream just to make sure it does not exist in our internal * state. This should NEVER happen. */ - lttng_ht_lookup(ht, (void *)((unsigned long) stream->key), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + lttng_ht_lookup(ht, &stream->key, &iter); + node = lttng_ht_iter_get_node_u64(&iter); assert(!node); /* Find relayd and, if one is found, increment refcount. */ @@ -1897,14 +2092,17 @@ static int add_metadata_stream(struct lttng_consumer_stream *stream, uatomic_dec(&stream->chan->nb_init_stream_left); } - lttng_ht_add_unique_ulong(ht, &stream->node); + lttng_ht_add_unique_u64(ht, &stream->node); + + lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, + &stream->node_channel_id); /* * Add stream to the stream_list_ht of the consumer data. No need to steal * the key since the HT does not use it and we allow to add redundant keys * into this table. */ - lttng_ht_add_ulong(consumer_data.stream_list_ht, &stream->node_session_id); + lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); rcu_read_unlock(); @@ -1976,17 +2174,17 @@ void *consumer_thread_metadata_poll(void *data) uint32_t revents, nb_fd; struct lttng_consumer_stream *stream = NULL; struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; struct lttng_poll_event events; struct lttng_consumer_local_data *ctx = data; ssize_t len; rcu_register_thread(); - metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); if (!metadata_ht) { /* ENOMEM at this point. Better to bail out. */ - goto error; + goto end_ht; } DBG("Thread metadata poll started"); @@ -1995,7 +2193,7 @@ void *consumer_thread_metadata_poll(void *data) ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); if (ret < 0) { ERR("Poll set creation failed"); - goto end; + goto end_poll; } ret = lttng_poll_add(&events, ctx->consumer_metadata_pipe[0], LPOLLIN); @@ -2092,9 +2290,12 @@ restart: } rcu_read_lock(); - lttng_ht_lookup(metadata_ht, (void *)((unsigned long) pollfd), - &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + { + uint64_t tmp_id = (uint64_t) pollfd; + + lttng_ht_lookup(metadata_ht, &tmp_id, &iter); + } + node = lttng_ht_iter_get_node_u64(&iter); assert(node); stream = caa_container_of(node, struct lttng_consumer_stream, @@ -2151,10 +2352,11 @@ restart: error: end: DBG("Metadata poll thread exiting"); - lttng_poll_clean(&events); + lttng_poll_clean(&events); +end_poll: destroy_stream_ht(metadata_ht); - +end_ht: rcu_unregister_thread(); return NULL; } @@ -2176,7 +2378,7 @@ void *consumer_thread_data_poll(void *data) rcu_register_thread(); - data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); if (data_ht == NULL) { /* ENOMEM at this point. Better to bail out. */ goto end; @@ -2285,7 +2487,7 @@ void *consumer_thread_data_poll(void *data) ret = add_stream(new_stream, data_ht); if (ret) { - ERR("Consumer add stream %d failed. Continuing", + ERR("Consumer add stream %" PRIu64 " failed. Continuing", new_stream->key); /* * At this point, if the add_stream fails, it is not in the @@ -2417,6 +2619,249 @@ end: return NULL; } +/* + * Close wake-up end of each stream belonging to the channel. This will + * allow the poll() on the stream read-side to detect when the + * write-side (application) finally closes them. + */ +static +void consumer_close_channel_streams(struct lttng_consumer_channel *channel) +{ + struct lttng_ht *ht; + struct lttng_consumer_stream *stream; + struct lttng_ht_iter iter; + + ht = consumer_data.stream_per_chan_id_ht; + + rcu_read_lock(); + cds_lfht_for_each_entry_duplicate(ht->ht, + ht->hash_fct(&channel->key, lttng_ht_seed), + ht->match_fct, &channel->key, + &iter.iter, stream, node_channel_id.node) { + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + break; + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + /* + * Note: a mutex is taken internally within + * liblttng-ust-ctl to protect timer wakeup_fd + * use from concurrent close. + */ + lttng_ustconsumer_close_stream_wakeup(stream); + break; + default: + ERR("Unknown consumer_data type"); + assert(0); + } + } + rcu_read_unlock(); +} + +static void destroy_channel_ht(struct lttng_ht *ht) +{ + struct lttng_ht_iter iter; + struct lttng_consumer_channel *channel; + int ret; + + if (ht == NULL) { + return; + } + + rcu_read_lock(); + cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { + ret = lttng_ht_del(ht, &iter); + assert(ret != 0); + } + rcu_read_unlock(); + + lttng_ht_destroy(ht); +} + +/* + * This thread polls the channel fds to detect when they are being + * closed. It closes all related streams if the channel is detected as + * closed. It is currently only used as a shim layer for UST because the + * consumerd needs to keep the per-stream wakeup end of pipes open for + * periodical flush. + */ +void *consumer_thread_channel_poll(void *data) +{ + int ret, i, pollfd; + uint32_t revents, nb_fd; + struct lttng_consumer_channel *chan = NULL; + struct lttng_ht_iter iter; + struct lttng_ht_node_u64 *node; + struct lttng_poll_event events; + struct lttng_consumer_local_data *ctx = data; + struct lttng_ht *channel_ht; + + rcu_register_thread(); + + channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + if (!channel_ht) { + /* ENOMEM at this point. Better to bail out. */ + goto end_ht; + } + + DBG("Thread channel poll started"); + + /* Size is set to 1 for the consumer_channel pipe */ + ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); + if (ret < 0) { + ERR("Poll set creation failed"); + goto end_poll; + } + + ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); + if (ret < 0) { + goto end; + } + + /* Main loop */ + DBG("Channel main loop started"); + + while (1) { + /* Only the channel pipe is set */ + if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { + goto end; + } + +restart: + DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); + ret = lttng_poll_wait(&events, -1); + DBG("Channel event catched in thread"); + if (ret < 0) { + if (errno == EINTR) { + ERR("Poll EINTR catched"); + goto restart; + } + goto end; + } + + nb_fd = ret; + + /* From here, the event is a channel wait fd */ + for (i = 0; i < nb_fd; i++) { + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); + + /* Just don't waste time if no returned events for the fd */ + if (!revents) { + continue; + } + if (pollfd == ctx->consumer_channel_pipe[0]) { + if (revents & (LPOLLERR | LPOLLHUP)) { + DBG("Channel thread pipe hung up"); + /* + * Remove the pipe from the poll set and continue the loop + * since their might be data to consume. + */ + lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); + continue; + } else if (revents & LPOLLIN) { + enum consumer_channel_action action; + + ret = read_channel_pipe(ctx, &chan, &action); + if (ret <= 0) { + ERR("Error reading channel pipe"); + continue; + } + + switch (action) { + case CONSUMER_CHANNEL_ADD: + DBG("Adding channel %d to poll set", + chan->wait_fd); + + lttng_ht_node_init_u64(&chan->wait_fd_node, + chan->wait_fd); + lttng_ht_add_unique_u64(channel_ht, + &chan->wait_fd_node); + /* Add channel to the global poll events list */ + lttng_poll_add(&events, chan->wait_fd, + LPOLLIN | LPOLLPRI); + break; + case CONSUMER_CHANNEL_QUIT: + /* + * Remove the pipe from the poll set and continue the loop + * since their might be data to consume. + */ + lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); + continue; + default: + ERR("Unknown action"); + break; + } + } + + /* Handle other stream */ + continue; + } + + rcu_read_lock(); + { + uint64_t tmp_id = (uint64_t) pollfd; + + lttng_ht_lookup(channel_ht, &tmp_id, &iter); + } + node = lttng_ht_iter_get_node_u64(&iter); + assert(node); + + chan = caa_container_of(node, struct lttng_consumer_channel, + wait_fd_node); + + /* Check for error event */ + if (revents & (LPOLLERR | LPOLLHUP)) { + DBG("Channel fd %d is hup|err.", pollfd); + + lttng_poll_del(&events, chan->wait_fd); + ret = lttng_ht_del(channel_ht, &iter); + assert(ret == 0); + consumer_close_channel_streams(chan); + } + + /* Release RCU lock for the channel looked up */ + rcu_read_unlock(); + } + } + +end: + lttng_poll_clean(&events); +end_poll: + destroy_channel_ht(channel_ht); +end_ht: + DBG("Channel poll thread exiting"); + rcu_unregister_thread(); + return NULL; +} + +static int set_metadata_socket(struct lttng_consumer_local_data *ctx, + struct pollfd *sockpoll, int client_socket) +{ + int ret; + + assert(ctx); + assert(sockpoll); + + if (lttng_consumer_poll_socket(sockpoll) < 0) { + ret = -1; + goto error; + } + DBG("Metadata connection on client_socket"); + + /* Blocking call, waiting for transmission */ + ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); + if (ctx->consumer_metadata_socket < 0) { + WARN("On accept metadata"); + ret = -1; + goto error; + } + ret = 0; + +error: + return ret; +} + /* * This thread listens on the consumerd socket and receives the file * descriptors from the session daemon. @@ -2483,6 +2928,15 @@ void *consumer_thread_sessiond_poll(void *data) goto end; } + /* + * Setup metadata socket which is the second socket connection on the + * command unix socket. + */ + ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); + if (ret < 0) { + goto end; + } + /* This socket is not useful anymore. */ ret = close(client_socket); if (ret < 0) { @@ -2521,6 +2975,14 @@ void *consumer_thread_sessiond_poll(void *data) end: DBG("Consumer thread sessiond poll exiting"); + /* + * Close metadata streams since the producer is the session daemon which + * just died. + * + * NOTE: for now, this only applies to the UST tracer. + */ + lttng_consumer_close_metadata(); + /* * when all fds have hung up, the polling thread * can exit cleanly @@ -2533,6 +2995,8 @@ end: */ notify_thread_pipe(ctx->consumer_data_pipe[1]); + notify_channel_pipe(ctx, NULL, CONSUMER_CHANNEL_QUIT); + /* Cleaning up possibly open sockets. */ if (sock >= 0) { ret = close(sock); @@ -2597,9 +3061,10 @@ int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) */ void lttng_consumer_init(void) { - consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); - consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); - consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); } /* @@ -2730,7 +3195,7 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, goto error; } - DBG("Consumer %s socket created successfully with net idx %d (fd: %d)", + DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", relayd->net_seq_idx, fd); @@ -2867,8 +3332,8 @@ int consumer_data_pending(uint64_t id) } cds_lfht_for_each_entry_duplicate(ht->ht, - ht->hash_fct((void *)((unsigned long) id), lttng_ht_seed), - ht->match_fct, (void *)((unsigned long) id), + ht->hash_fct(&id, lttng_ht_seed), + ht->match_fct, &id, &iter.iter, stream, node_session_id.node) { /* If this call fails, the stream is being used hence data pending. */ ret = stream_try_lock(stream);