X-Git-Url: https://git.lttng.org/?a=blobdiff_plain;f=src%2Fcommon%2Fconsumer.c;h=37a4e01978dc05324fb5739e3e681e84e82d1ad8;hb=0d08d75e73e92edb002f3d480004ccb687cf1eeb;hp=044a504cec512443039e15eac186df07e841f896;hpb=13886d2d433762b7cf54f8e812f747ec2829de57;p=lttng-tools.git diff --git a/src/common/consumer.c b/src/common/consumer.c index 044a504ce..37a4e0197 100644 --- a/src/common/consumer.c +++ b/src/common/consumer.c @@ -102,6 +102,7 @@ static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, msg.action = action; msg.chan = chan; + msg.key = key; do { ret = write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); } while (ret < 0 && errno == EINTR); @@ -286,6 +287,7 @@ 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); @@ -296,6 +298,13 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) 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, &channel->streams.head, + send_node) { + cds_list_del(&stream->send_node); + lttng_ustconsumer_del_stream(stream); + free(stream); + } lttng_ustconsumer_del_channel(channel); break; default: @@ -463,6 +472,13 @@ void consumer_del_stream(struct lttng_consumer_stream *stream, 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: @@ -882,6 +898,8 @@ 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) @@ -899,7 +917,7 @@ 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 = -1; + ret = -EEXIST; goto end; } @@ -1876,6 +1894,13 @@ 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) { + PERROR("close kernel metadata wait_fd"); + } + } break; case LTTNG_CONSUMER32_UST: case LTTNG_CONSUMER64_UST: @@ -1945,6 +1970,13 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, } end: + /* + * Nullify the stream reference so it is not used after deletion. The + * consumer data 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(&consumer_data.lock); @@ -1997,9 +2029,6 @@ static int add_metadata_stream(struct lttng_consumer_stream *stream, 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 @@ -2330,7 +2359,7 @@ 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); @@ -2697,24 +2726,51 @@ 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. @@ -2722,6 +2778,7 @@ restart: if (!uatomic_sub_return(&chan->refcount, 1)) { consumer_del_channel(chan); } + rcu_read_unlock(); goto restart; } case CONSUMER_CHANNEL_QUIT: @@ -2760,6 +2817,7 @@ restart: lttng_poll_del(&events, chan->wait_fd); ret = lttng_ht_del(channel_ht, &iter); assert(ret == 0); + assert(cds_list_empty(&chan->streams.head)); consumer_close_channel_streams(chan); /* Release our own refcount */ @@ -2848,12 +2906,6 @@ 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; @@ -2871,11 +2923,6 @@ 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 @@ -2954,7 +3001,7 @@ end: } } if (client_socket >= 0) { - ret = close(sock); + ret = close(client_socket); if (ret < 0) { PERROR("close client_socket sessiond poll"); } @@ -3036,29 +3083,36 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, 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; - } - /* Get relayd reference if exists. */ relayd = consumer_find_relayd(net_seq_idx); if (relayd == NULL) { /* 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; - goto error; + ret_code = LTTCOMM_CONSUMERD_ENOMEM; + ret = -ENOMEM; + } else { + relayd->sessiond_session_id = (uint64_t) 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. + */ + } + + /* First send a status message before receiving the fds. */ + ret = consumer_send_status_msg(sock, ret_code); + if (ret < 0 || ret_code != LTTNG_OK) { + /* Somehow, the session daemon is not responding anymore. */ + goto error; } /* 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; } @@ -3066,15 +3120,31 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, /* 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_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; ret = -1; fd = -1; /* Just in case it gets set with an invalid value. */ - goto error_close; + + /* + * 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); + + /* + * This code path MUST continue to the consumer send status message so + * we can send the error to the thread expecting a reply. The above + * call will make everything stop. + */ } /* We have the fds without error. Send status back. */ ret = consumer_send_status_msg(sock, ret_code); - if (ret < 0) { + if (ret < 0 || ret_code != LTTNG_OK) { /* Somehow, the session daemon is not responding anymore. */ goto error; } @@ -3174,7 +3244,6 @@ error: } } -error_close: if (relayd_created) { free(relayd); }