X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fconsumer.c;h=4b657f33249f7a4f777b7b3ff784b379b15f5878;hp=b5fe9831640217455c40c11b9e85e36b470ec05c;hb=6a00837f8cb0431a2ad90974d67fae138ea97dd5;hpb=f21dae485bbd4a21104eff49a80b59ce25054bc2 diff --git a/src/common/consumer.c b/src/common/consumer.c index b5fe98316..4b657f332 100644 --- a/src/common/consumer.c +++ b/src/common/consumer.c @@ -41,6 +41,7 @@ #include #include "consumer.h" +#include "consumer-stream.h" struct lttng_consumer_global_data consumer_data = { .stream_count = 0, @@ -164,20 +165,20 @@ static struct lttng_consumer_stream *find_stream(uint64_t key, return stream; } -static void steal_stream_key(int key, struct lttng_ht *ht) +static void steal_stream_key(uint64_t key, struct lttng_ht *ht) { struct lttng_consumer_stream *stream; rcu_read_lock(); stream = find_stream(key, ht); if (stream) { - stream->key = -1ULL; + stream->key = (uint64_t) -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 = -1ULL; + stream->node.key = (uint64_t) -1ULL; } rcu_read_unlock(); } @@ -254,10 +255,8 @@ static void free_relayd_rcu(struct rcu_head *head) /* * Destroy and free relayd socket pair object. - * - * This function MUST be called with the consumer_data lock acquired. */ -static void destroy_relayd(struct consumer_relayd_sock_pair *relayd) +void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) { int ret; struct lttng_ht_iter iter; @@ -287,10 +286,23 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) { int ret; struct lttng_ht_iter iter; + struct lttng_consumer_stream *stream, *stmp; DBG("Consumer delete channel key %" PRIu64, channel->key); pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&channel->lock); + + /* Delete streams that might have been left in the stream list. */ + cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, + send_node) { + cds_list_del(&stream->send_node); + /* + * Once a stream is added to this list, the buffers were created so + * we have a guarantee that this call will succeed. + */ + consumer_stream_destroy(stream, NULL); + } switch (consumer_data.type) { case LTTNG_CONSUMER_KERNEL: @@ -313,6 +325,7 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) call_rcu(&channel->node.head, free_channel_rcu); end: + pthread_mutex_unlock(&channel->lock); pthread_mutex_unlock(&consumer_data.lock); } @@ -329,7 +342,7 @@ static void cleanup_relayd_ht(void) cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, node.node) { - destroy_relayd(relayd); + consumer_destroy_relayd(relayd); } rcu_read_unlock(); @@ -344,13 +357,13 @@ static void cleanup_relayd_ht(void) * It's atomically set without having the stream mutex locked which is fine * because we handle the write/read race with a pipe wakeup for each thread. */ -static void update_endpoint_status_by_netidx(int net_seq_idx, +static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, enum consumer_endpoint_status status) { struct lttng_ht_iter iter; struct lttng_consumer_stream *stream; - DBG("Consumer set delete flag on stream by idx %d", net_seq_idx); + DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); rcu_read_lock(); @@ -383,7 +396,7 @@ static void update_endpoint_status_by_netidx(int net_seq_idx, static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, struct lttng_consumer_local_data *ctx) { - int netidx; + uint64_t netidx; assert(relayd); @@ -396,7 +409,7 @@ static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, * Delete the relayd from the relayd hash table, close the sockets and free * the object in a RCU call. */ - destroy_relayd(relayd); + consumer_destroy_relayd(relayd); /* Set inactive endpoint to all streams */ update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); @@ -428,130 +441,33 @@ void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) /* Destroy the relayd if refcount is 0 */ if (uatomic_read(&relayd->refcount) == 0) { - destroy_relayd(relayd); + consumer_destroy_relayd(relayd); } } /* - * Remove a stream from the global list protected by a mutex. This - * function is also responsible for freeing its data structures. + * Completly destroy stream from every visiable data structure and the given + * hash table if one. + * + * One this call returns, the stream object is not longer usable nor visible. */ void consumer_del_stream(struct lttng_consumer_stream *stream, struct lttng_ht *ht) { - int ret; - struct lttng_ht_iter iter; - struct lttng_consumer_channel *free_chan = NULL; - struct consumer_relayd_sock_pair *relayd; - - assert(stream); - - DBG("Consumer del stream %d", stream->wait_fd); - - if (ht == NULL) { - /* Means the stream was allocated but not successfully added */ - goto free_stream_rcu; - } - - pthread_mutex_lock(&consumer_data.lock); - pthread_mutex_lock(&stream->lock); - - switch (consumer_data.type) { - case LTTNG_CONSUMER_KERNEL: - if (stream->mmap_base != NULL) { - ret = munmap(stream->mmap_base, stream->mmap_len); - if (ret != 0) { - PERROR("munmap"); - } - } - - if (stream->wait_fd >= 0) { - ret = close(stream->wait_fd); - if (ret) { - PERROR("close"); - } - } - break; - case LTTNG_CONSUMER32_UST: - case LTTNG_CONSUMER64_UST: - lttng_ustconsumer_del_stream(stream); - break; - default: - ERR("Unknown consumer_data type"); - assert(0); - goto end; - } - - rcu_read_lock(); - iter.iter.node = &stream->node.node; - ret = lttng_ht_del(ht, &iter); - assert(!ret); - - 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); - rcu_read_unlock(); - - assert(consumer_data.stream_count > 0); - consumer_data.stream_count--; - - if (stream->out_fd >= 0) { - ret = close(stream->out_fd); - if (ret) { - PERROR("close"); - } - } - - /* Check and cleanup relayd */ - rcu_read_lock(); - relayd = consumer_find_relayd(stream->net_seq_idx); - if (relayd != NULL) { - uatomic_dec(&relayd->refcount); - assert(uatomic_read(&relayd->refcount) >= 0); - - /* Closing streams requires to lock the control socket. */ - pthread_mutex_lock(&relayd->ctrl_sock_mutex); - ret = relayd_send_close_stream(&relayd->control_sock, - stream->relayd_stream_id, - stream->next_net_seq_num - 1); - pthread_mutex_unlock(&relayd->ctrl_sock_mutex); - if (ret < 0) { - DBG("Unable to close stream on the relayd. Continuing"); - /* - * Continue here. There is nothing we can do for the relayd. - * Chances are that the relayd has closed the socket so we just - * continue cleaning up. - */ - } - - /* Both conditions are met, we destroy the relayd. */ - if (uatomic_read(&relayd->refcount) == 0 && - uatomic_read(&relayd->destroy_flag)) { - destroy_relayd(relayd); - } - } - rcu_read_unlock(); - - if (!uatomic_sub_return(&stream->chan->refcount, 1) - && !uatomic_read(&stream->chan->nb_init_stream_left)) { - free_chan = stream->chan; - } - -end: - consumer_data.need_update = 1; - pthread_mutex_unlock(&stream->lock); - pthread_mutex_unlock(&consumer_data.lock); + consumer_stream_destroy(stream, ht); +} - if (free_chan) { - consumer_del_channel(free_chan); - } +/* + * XXX naming of del vs destroy is all mixed up. + */ +void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) +{ + consumer_stream_destroy(stream, data_ht); +} -free_stream_rcu: - call_rcu(&stream->node.head, free_stream_rcu); +void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) +{ + consumer_stream_destroy(stream, metadata_ht); } struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, @@ -560,11 +476,12 @@ struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, const char *channel_name, uid_t uid, gid_t gid, - int relayd_id, + uint64_t relayd_id, uint64_t session_id, int cpu, int *alloc_ret, - enum consumer_channel_type type) + enum consumer_channel_type type, + unsigned int monitor) { int ret; struct lttng_consumer_stream *stream; @@ -586,6 +503,8 @@ struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, stream->gid = gid; stream->net_seq_idx = relayd_id; stream->session_id = session_id; + stream->monitor = monitor; + stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; pthread_mutex_init(&stream->lock, NULL); /* If channel is the metadata, flag this stream as metadata. */ @@ -612,8 +531,10 @@ struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, /* Init session id node with the stream session id */ lttng_ht_node_init_u64(&stream->node_session_id, 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); + 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; @@ -631,11 +552,10 @@ end: /* * Add a stream to the global list protected by a mutex. */ -static int add_stream(struct lttng_consumer_stream *stream, - struct lttng_ht *ht) +int consumer_add_data_stream(struct lttng_consumer_stream *stream) { + struct lttng_ht *ht = data_ht; int ret = 0; - struct consumer_relayd_sock_pair *relayd; assert(stream); assert(ht); @@ -643,6 +563,8 @@ static int add_stream(struct lttng_consumer_stream *stream, DBG3("Adding consumer stream %" PRIu64, stream->key); pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&stream->chan->lock); + pthread_mutex_lock(&stream->chan->timer_lock); pthread_mutex_lock(&stream->lock); rcu_read_lock(); @@ -661,12 +583,6 @@ static int add_stream(struct lttng_consumer_stream *stream, */ 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); - if (relayd != NULL) { - uatomic_inc(&relayd->refcount); - } - /* * When nb_init_stream_left reaches 0, we don't need to trigger any action * in terms of destroying the associated channel, because the action that @@ -686,11 +602,18 @@ static int add_stream(struct lttng_consumer_stream *stream, rcu_read_unlock(); pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&stream->chan->timer_lock); + pthread_mutex_unlock(&stream->chan->lock); pthread_mutex_unlock(&consumer_data.lock); return ret; } +void consumer_del_data_stream(struct lttng_consumer_stream *stream) +{ + consumer_del_stream(stream, data_ht); +} + /* * Add relayd socket to global consumer data hashtable. RCU read side lock MUST * be acquired before calling this. @@ -719,12 +642,12 @@ end: * Allocate and return a consumer relayd socket. */ struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( - int net_seq_idx) + uint64_t net_seq_idx) { struct consumer_relayd_sock_pair *obj = NULL; - /* Negative net sequence index is a failure */ - if (net_seq_idx < 0) { + /* net sequence index of -1 is a failure */ + if (net_seq_idx == (uint64_t) -1ULL) { goto error; } @@ -775,6 +698,67 @@ error: return relayd; } +/* + * Find a relayd and send the stream + * + * Returns 0 on success, < 0 on error + */ +int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, + char *path) +{ + int ret = 0; + struct consumer_relayd_sock_pair *relayd; + + assert(stream); + assert(stream->net_seq_idx != -1ULL); + assert(path); + + /* The stream is not metadata. Get relayd reference if exists. */ + rcu_read_lock(); + relayd = consumer_find_relayd(stream->net_seq_idx); + if (relayd != NULL) { + /* Add stream on the relayd */ + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + ret = relayd_add_stream(&relayd->control_sock, stream->name, + path, &stream->relayd_stream_id, + stream->chan->tracefile_size, stream->chan->tracefile_count); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + if (ret < 0) { + goto end; + } + uatomic_inc(&relayd->refcount); + stream->sent_to_relayd = 1; + } else { + ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", + stream->key, stream->net_seq_idx); + ret = -1; + goto end; + } + + DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, + stream->name, stream->key, stream->net_seq_idx); + +end: + rcu_read_unlock(); + return ret; +} + +/* + * Find a relayd and close the stream + */ +void close_relayd_stream(struct lttng_consumer_stream *stream) +{ + struct consumer_relayd_sock_pair *relayd; + + /* The stream is not metadata. Get relayd reference if exists. */ + rcu_read_lock(); + relayd = consumer_find_relayd(stream->net_seq_idx); + if (relayd) { + consumer_stream_relayd_close(stream, relayd); + } + rcu_read_unlock(); +} + /* * Handle stream for relayd transmission if the stream applies for network * streaming where the net sequence index is set. @@ -847,10 +831,12 @@ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, const char *name, uid_t uid, gid_t gid, - int relayd_id, + uint64_t relayd_id, enum lttng_event_output output, uint64_t tracefile_size, - uint64_t tracefile_count) + uint64_t tracefile_count, + uint64_t session_id_per_pid, + unsigned int monitor) { struct lttng_consumer_channel *channel; @@ -863,12 +849,29 @@ struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, channel->key = key; channel->refcount = 0; channel->session_id = session_id; + channel->session_id_per_pid = session_id_per_pid; channel->uid = uid; channel->gid = gid; channel->relayd_id = relayd_id; channel->output = output; channel->tracefile_size = tracefile_size; channel->tracefile_count = tracefile_count; + channel->monitor = monitor; + pthread_mutex_init(&channel->lock, NULL); + pthread_mutex_init(&channel->timer_lock, NULL); + + /* + * In monitor mode, the streams associated with the channel will be put in + * a special list ONLY owned by this channel. So, the refcount is set to 1 + * here meaning that the channel itself has streams that are referenced. + * + * On a channel deletion, once the channel is no longer visible, the + * refcount is decremented and checked for a zero value to delete it. With + * streams in no monitor mode, it will now be safe to destroy the channel. + */ + if (!channel->monitor) { + channel->refcount = 1; + } strncpy(channel->pathname, pathname, sizeof(channel->pathname)); channel->pathname[sizeof(channel->pathname) - 1] = '\0'; @@ -890,6 +893,8 @@ end: /* * Add a channel to the global list protected by a mutex. + * + * On success 0 is returned else a negative value. */ int consumer_add_channel(struct lttng_consumer_channel *channel, struct lttng_consumer_local_data *ctx) @@ -899,6 +904,8 @@ int consumer_add_channel(struct lttng_consumer_channel *channel, struct lttng_ht_iter iter; pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&channel->lock); + pthread_mutex_lock(&channel->timer_lock); rcu_read_lock(); lttng_ht_lookup(consumer_data.channel_ht, &channel->key, &iter); @@ -907,7 +914,7 @@ int consumer_add_channel(struct lttng_consumer_channel *channel, /* Channel already exist. Ignore the insertion */ ERR("Consumer add channel key %" PRIu64 " already exists!", channel->key); - ret = LTTNG_ERR_KERN_CHAN_EXIST; + ret = -EEXIST; goto end; } @@ -915,10 +922,12 @@ int consumer_add_channel(struct lttng_consumer_channel *channel, end: rcu_read_unlock(); + pthread_mutex_unlock(&channel->timer_lock); + pthread_mutex_unlock(&channel->lock); pthread_mutex_unlock(&consumer_data.lock); if (!ret && channel->wait_fd != -1 && - channel->metadata_stream == NULL) { + channel->type == CONSUMER_CHANNEL_TYPE_DATA) { notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); } return ret; @@ -1150,7 +1159,7 @@ struct lttng_consumer_local_data *lttng_consumer_create( struct lttng_consumer_local_data *ctx), int (*recv_channel)(struct lttng_consumer_channel *channel), int (*recv_stream)(struct lttng_consumer_stream *stream), - int (*update_stream)(int stream_key, uint32_t state)) + int (*update_stream)(uint64_t stream_key, uint32_t state)) { int ret; struct lttng_consumer_local_data *ctx; @@ -1167,6 +1176,7 @@ struct lttng_consumer_local_data *lttng_consumer_create( ctx->consumer_error_socket = -1; ctx->consumer_metadata_socket = -1; + pthread_mutex_init(&ctx->metadata_socket_lock, NULL); /* assign the callbacks */ ctx->on_buffer_ready = buffer_ready; ctx->on_recv_channel = recv_channel; @@ -1321,7 +1331,7 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( rcu_read_lock(); /* Flag that the current stream if set for network streaming. */ - if (stream->net_seq_idx != -1) { + if (stream->net_seq_idx != (uint64_t) -1ULL) { relayd = consumer_find_relayd(stream->net_seq_idx); if (relayd == NULL) { goto end; @@ -1520,7 +1530,7 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( rcu_read_lock(); /* Flag that the current stream if set for network streaming. */ - if (stream->net_seq_idx != -1) { + if (stream->net_seq_idx != (uint64_t) -1ULL) { relayd = consumer_find_relayd(stream->net_seq_idx); if (relayd == NULL) { goto end; @@ -1874,6 +1884,7 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, } pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&stream->chan->lock); pthread_mutex_lock(&stream->lock); switch (consumer_data.type) { @@ -1884,7 +1895,6 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, PERROR("munmap metadata stream"); } } - if (stream->wait_fd >= 0) { ret = close(stream->wait_fd); if (ret < 0) { @@ -1894,6 +1904,13 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, break; case LTTNG_CONSUMER32_UST: case LTTNG_CONSUMER64_UST: + if (stream->monitor) { + /* close the write-side in close_metadata */ + ret = close(stream->ust_metadata_poll_pipe[0]); + if (ret < 0) { + PERROR("Close UST metadata read-side poll pipe"); + } + } lttng_ustconsumer_del_stream(stream); break; default: @@ -1947,7 +1964,7 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, /* Both conditions are met, we destroy the relayd. */ if (uatomic_read(&relayd->refcount) == 0 && uatomic_read(&relayd->destroy_flag)) { - destroy_relayd(relayd); + consumer_destroy_relayd(relayd); } } rcu_read_unlock(); @@ -1960,7 +1977,15 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, } end: + /* + * Nullify the stream reference so it is not used after deletion. The + * channel lock MUST be acquired before being able to check for + * a NULL pointer value. + */ + stream->chan->metadata_stream = NULL; + pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&stream->chan->lock); pthread_mutex_unlock(&consumer_data.lock); if (free_chan) { @@ -1975,11 +2000,10 @@ free_stream_rcu: * Action done with the metadata stream when adding it to the consumer internal * data structures to handle it. */ -static int add_metadata_stream(struct lttng_consumer_stream *stream, - struct lttng_ht *ht) +int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) { + struct lttng_ht *ht = metadata_ht; int ret = 0; - struct consumer_relayd_sock_pair *relayd; struct lttng_ht_iter iter; struct lttng_ht_node_u64 *node; @@ -1989,6 +2013,8 @@ static int add_metadata_stream(struct lttng_consumer_stream *stream, DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&stream->chan->lock); + pthread_mutex_lock(&stream->chan->timer_lock); pthread_mutex_lock(&stream->lock); /* @@ -2006,15 +2032,6 @@ static int add_metadata_stream(struct lttng_consumer_stream *stream, node = lttng_ht_iter_get_node_u64(&iter); assert(!node); - /* Find relayd and, if one is found, increment refcount. */ - relayd = consumer_find_relayd(stream->net_seq_idx); - if (relayd != NULL) { - uatomic_inc(&relayd->refcount); - } - - /* Update channel refcount once added without error(s). */ - uatomic_inc(&stream->chan->refcount); - /* * When nb_init_stream_left reaches 0, we don't need to trigger any action * in terms of destroying the associated channel, because the action that @@ -2043,6 +2060,8 @@ static int add_metadata_stream(struct lttng_consumer_stream *stream, rcu_read_unlock(); pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&stream->chan->lock); + pthread_mutex_unlock(&stream->chan->timer_lock); pthread_mutex_unlock(&consumer_data.lock); return ret; } @@ -2188,7 +2207,7 @@ restart: pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, &stream, sizeof(stream)); if (pipe_len < 0) { - ERR("read metadata stream, ret: %ld", pipe_len); + ERR("read metadata stream, ret: %zd", pipe_len); /* * Continue here to handle the rest of the streams. */ @@ -2205,14 +2224,6 @@ restart: DBG("Adding metadata stream %d to poll set", stream->wait_fd); - ret = add_metadata_stream(stream, metadata_ht); - if (ret) { - ERR("Unable to add metadata stream"); - /* Stream was not setup properly. Continuing. */ - consumer_del_metadata_stream(stream, NULL); - continue; - } - /* Add metadata stream to the global poll events list */ lttng_poll_add(&events, stream->wait_fd, LPOLLIN | LPOLLPRI); @@ -2266,14 +2277,21 @@ restart: DBG("Metadata available on fd %d", pollfd); assert(stream->wait_fd == pollfd); - len = ctx->on_buffer_ready(stream, ctx); + do { + len = ctx->on_buffer_ready(stream, ctx); + /* + * We don't check the return value here since if we get + * a negative len, it means an error occured thus we + * simply remove it from the poll set and free the + * stream. + */ + } while (len > 0); + /* It's ok to have an unavailable sub-buffer */ if (len < 0 && len != -EAGAIN && len != -ENODATA) { /* Clean up stream from consumer and free it. */ lttng_poll_del(&events, stream->wait_fd); consumer_del_metadata_stream(stream, metadata_ht); - } else if (len > 0) { - stream->data_read = 1; } } @@ -2317,7 +2335,11 @@ void *consumer_thread_data_poll(void *data) goto end; } - local_stream = zmalloc(sizeof(struct lttng_consumer_stream)); + local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); + if (local_stream == NULL) { + PERROR("local_stream malloc"); + goto end; + } while (1) { high_prio = 0; @@ -2345,7 +2367,7 @@ void *consumer_thread_data_poll(void *data) /* allocate for all fds + 1 for the consumer_data_pipe */ local_stream = zmalloc((consumer_data.stream_count + 1) * - sizeof(struct lttng_consumer_stream)); + sizeof(struct lttng_consumer_stream *)); if (local_stream == NULL) { PERROR("local_stream malloc"); pthread_mutex_unlock(&consumer_data.lock); @@ -2400,7 +2422,7 @@ void *consumer_thread_data_poll(void *data) pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, &new_stream, sizeof(new_stream)); if (pipe_readlen < 0) { - ERR("Consumer data pipe ret %ld", pipe_readlen); + ERR("Consumer data pipe ret %zd", pipe_readlen); /* Continue so we can at least handle the current stream(s). */ continue; } @@ -2415,17 +2437,6 @@ void *consumer_thread_data_poll(void *data) continue; } - ret = add_stream(new_stream, data_ht); - if (ret) { - ERR("Consumer add stream %" PRIu64 " failed. Continuing", - new_stream->key); - /* - * At this point, if the add_stream fails, it is not in the - * hash table thus passing the NULL value here. - */ - consumer_del_stream(new_stream, NULL); - } - /* Continue to update the local streams and handle prio ones */ continue; } @@ -2712,24 +2723,51 @@ restart: lttng_ht_node_init_u64(&chan->wait_fd_node, chan->wait_fd); + rcu_read_lock(); lttng_ht_add_unique_u64(channel_ht, &chan->wait_fd_node); + rcu_read_unlock(); /* Add channel to the global poll events list */ lttng_poll_add(&events, chan->wait_fd, LPOLLIN | LPOLLPRI); break; case CONSUMER_CHANNEL_DEL: { + struct lttng_consumer_stream *stream, *stmp; + + rcu_read_lock(); chan = consumer_find_channel(key); if (!chan) { + rcu_read_unlock(); ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); break; } lttng_poll_del(&events, chan->wait_fd); + iter.iter.node = &chan->wait_fd_node.node; ret = lttng_ht_del(channel_ht, &iter); assert(ret == 0); consumer_close_channel_streams(chan); + switch (consumer_data.type) { + case LTTNG_CONSUMER_KERNEL: + break; + case LTTNG_CONSUMER32_UST: + case LTTNG_CONSUMER64_UST: + /* Delete streams that might have been left in the stream list. */ + cds_list_for_each_entry_safe(stream, stmp, &chan->streams.head, + send_node) { + cds_list_del(&stream->send_node); + lttng_ustconsumer_del_stream(stream); + uatomic_sub(&stream->chan->refcount, 1); + assert(&chan->refcount); + free(stream); + } + break; + default: + ERR("Unknown consumer_data type"); + assert(0); + } + /* * Release our own refcount. Force channel deletion even if * streams were not initialized. @@ -2737,6 +2775,7 @@ restart: if (!uatomic_sub_return(&chan->refcount, 1)) { consumer_del_channel(chan); } + rcu_read_unlock(); goto restart; } case CONSUMER_CHANNEL_QUIT: @@ -2863,12 +2902,6 @@ void *consumer_thread_sessiond_poll(void *data) goto end; } - ret = fcntl(client_socket, F_SETFL, O_NONBLOCK); - if (ret < 0) { - PERROR("fcntl O_NONBLOCK"); - goto end; - } - /* prepare the FDs to poll : to client socket and the should_quit pipe */ consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; consumer_sockpoll[0].events = POLLIN | POLLPRI; @@ -2886,11 +2919,6 @@ void *consumer_thread_sessiond_poll(void *data) WARN("On accept"); goto end; } - ret = fcntl(sock, F_SETFL, O_NONBLOCK); - if (ret < 0) { - PERROR("fcntl O_NONBLOCK"); - goto end; - } /* * Setup metadata socket which is the second socket connection on the @@ -2969,7 +2997,7 @@ end: } } if (client_socket >= 0) { - ret = close(sock); + ret = close(client_socket); if (ret < 0) { PERROR("close client_socket sessiond poll"); } @@ -3037,10 +3065,10 @@ void lttng_consumer_init(void) * This will create a relayd socket pair and add it to the relayd hash table. * The caller MUST acquire a RCU read side lock before calling it. */ -int consumer_add_relayd_socket(int net_seq_idx, int sock_type, +int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, struct lttng_consumer_local_data *ctx, int sock, struct pollfd *consumer_sockpoll, - struct lttcomm_relayd_sock *relayd_sock, unsigned int sessiond_id) + struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id) { int fd = -1, ret = -1, relayd_created = 0; enum lttng_error_code ret_code = LTTNG_OK; @@ -3049,48 +3077,67 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, assert(ctx); assert(relayd_sock); - DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx); - - /* 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 error; - } + DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); /* Get relayd reference if exists. */ relayd = consumer_find_relayd(net_seq_idx); if (relayd == NULL) { + assert(sock_type == LTTNG_STREAM_CONTROL); /* Not found. Allocate one. */ relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); if (relayd == NULL) { - lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); - ret = -1; + ret = -ENOMEM; + ret_code = LTTCOMM_CONSUMERD_ENOMEM; goto error; + } else { + relayd->sessiond_session_id = sessiond_id; + relayd_created = 1; } - relayd->sessiond_session_id = (uint64_t) sessiond_id; - relayd_created = 1; + + /* + * This code path MUST continue to the consumer send status message to + * we can notify the session daemon and continue our work without + * killing everything. + */ + } else { + /* + * relayd key should never be found for control socket. + */ + assert(sock_type != LTTNG_STREAM_CONTROL); + } + + /* First send a status message before receiving the fds. */ + ret = consumer_send_status_msg(sock, LTTNG_OK); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); + goto error_nosignal; } /* Poll on consumer socket. */ if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); ret = -EINTR; - goto error; + goto error_nosignal; } /* Get relayd socket from session daemon */ ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); if (ret != sizeof(fd)) { - lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); ret = -1; fd = -1; /* Just in case it gets set with an invalid value. */ - goto error_close; - } - /* We have the fds without error. Send status back. */ - ret = consumer_send_status_msg(sock, ret_code); - if (ret < 0) { - /* Somehow, the session daemon is not responding anymore. */ + /* + * Failing to receive FDs might indicate a major problem such as + * reaching a fd limit during the receive where the kernel returns a + * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we + * don't take any chances and stop everything. + * + * XXX: Feature request #558 will fix that and avoid this possible + * issue when reaching the fd limit. + */ + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); + ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; goto error; } @@ -3100,19 +3147,23 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, /* Copy received lttcomm socket */ lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); ret = lttcomm_create_sock(&relayd->control_sock.sock); - /* Immediately try to close the created socket if valid. */ - if (relayd->control_sock.sock.fd >= 0) { - if (close(relayd->control_sock.sock.fd)) { - PERROR("close relayd control socket"); - } - } /* Handle create_sock error. */ if (ret < 0) { + ret_code = LTTCOMM_CONSUMERD_ENOMEM; goto error; } + /* + * Close the socket created internally by + * lttcomm_create_sock, so we can replace it by the one + * received from sessiond. + */ + if (close(relayd->control_sock.sock.fd)) { + PERROR("close"); + } /* Assign new file descriptor */ relayd->control_sock.sock.fd = fd; + fd = -1; /* For error path */ /* Assign version values. */ relayd->control_sock.major = relayd_sock->major; relayd->control_sock.minor = relayd_sock->minor; @@ -3137,6 +3188,7 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, */ (void) relayd_close(&relayd->control_sock); (void) relayd_close(&relayd->data_sock); + ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; goto error; } @@ -3145,19 +3197,23 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, /* Copy received lttcomm socket */ lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); ret = lttcomm_create_sock(&relayd->data_sock.sock); - /* Immediately try to close the created socket if valid. */ - if (relayd->data_sock.sock.fd >= 0) { - if (close(relayd->data_sock.sock.fd)) { - PERROR("close relayd data socket"); - } - } /* Handle create_sock error. */ if (ret < 0) { + ret_code = LTTCOMM_CONSUMERD_ENOMEM; goto error; } + /* + * Close the socket created internally by + * lttcomm_create_sock, so we can replace it by the one + * received from sessiond. + */ + if (close(relayd->data_sock.sock.fd)) { + PERROR("close"); + } /* Assign new file descriptor */ relayd->data_sock.sock.fd = fd; + fd = -1; /* for eventual error paths */ /* Assign version values. */ relayd->data_sock.major = relayd_sock->major; relayd->data_sock.minor = relayd_sock->minor; @@ -3165,6 +3221,7 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, default: ERR("Unknown relayd socket type (%d)", sock_type); ret = -1; + ret_code = LTTCOMM_CONSUMERD_FATAL; goto error; } @@ -3172,6 +3229,14 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", relayd->net_seq_idx, fd); + /* We successfully added the socket. Send status back. */ + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0) { + /* Somehow, the session daemon is not responding anymore. */ + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); + goto error_nosignal; + } + /* * Add relayd socket pair to consumer data hashtable. If object already * exists or on error, the function gracefully returns. @@ -3182,6 +3247,11 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, return 0; error: + if (consumer_send_status_msg(sock, ret_code) < 0) { + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); + } + +error_nosignal: /* Close received socket if valid. */ if (fd >= 0) { if (close(fd)) { @@ -3189,7 +3259,6 @@ error: } } -error_close: if (relayd_created) { free(relayd); } @@ -3421,3 +3490,23 @@ int consumer_send_status_channel(int sock, return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); } + +/* + * Using a maximum stream size with the produced and consumed position of a + * stream, computes the new consumed position to be as close as possible to the + * maximum possible stream size. + * + * If maximum stream size is lower than the possible buffer size (produced - + * consumed), the consumed_pos given is returned untouched else the new value + * is returned. + */ +unsigned long consumer_get_consumed_maxsize(unsigned long consumed_pos, + unsigned long produced_pos, uint64_t max_stream_size) +{ + if (max_stream_size && max_stream_size < (produced_pos - consumed_pos)) { + /* Offset from the produced position to get the latest buffers. */ + return produced_pos - max_stream_size; + } + + return consumed_pos; +}