X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fconsumer.c;h=51f861e23695c642a56a943aa272fb24c89c7a5d;hp=41f8ae53fe67adde7552b54dfb31aa89d4abedcb;hb=d56db448a421370a0f33d6737cb366488d134b26;hpb=6d805429e9cb049eb0c9205fcf742a53e3166caf diff --git a/src/common/consumer.c b/src/common/consumer.c index 41f8ae53f..51f861e23 100644 --- a/src/common/consumer.c +++ b/src/common/consumer.c @@ -47,9 +47,6 @@ struct lttng_consumer_global_data consumer_data = { .type = LTTNG_CONSUMER_UNKNOWN, }; -/* timeout parameter, to control the polling thread grace period. */ -int consumer_poll_timeout = -1; - /* * 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. @@ -59,15 +56,12 @@ int consumer_poll_timeout = -1; volatile int consumer_quit; /* - * The following two hash tables are visible by all threads which are separated - * in different source files. - * * Global hash table containing respectively metadata and data streams. The * stream element in this ht should only be updated by the metadata poll thread * for the metadata and the data poll thread for the data. */ -struct lttng_ht *metadata_ht; -struct lttng_ht *data_ht; +static struct lttng_ht *metadata_ht; +static struct lttng_ht *data_ht; /* * Notify a thread pipe to poll back again. This usually means that some global @@ -133,6 +127,12 @@ void consumer_steal_stream_key(int key, struct lttng_ht *ht) rcu_read_unlock(); } +/* + * Return a channel object for the given key. + * + * RCU read side lock MUST be acquired before calling this function and + * protects the channel ptr. + */ static struct lttng_consumer_channel *consumer_find_channel(int key) { struct lttng_ht_iter iter; @@ -144,8 +144,6 @@ static struct lttng_consumer_channel *consumer_find_channel(int key) return NULL; } - rcu_read_lock(); - lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key), &iter); node = lttng_ht_iter_get_node_ulong(&iter); @@ -153,8 +151,6 @@ static struct lttng_consumer_channel *consumer_find_channel(int key) channel = caa_container_of(node, struct lttng_consumer_channel, node); } - rcu_read_unlock(); - return channel; } @@ -288,6 +284,8 @@ static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, assert(relayd); + DBG("Cleaning up relayd sockets"); + /* Save the net sequence index before destroying the object */ netidx = relayd->net_seq_idx; @@ -352,8 +350,8 @@ void consumer_del_stream(struct lttng_consumer_stream *stream, goto free_stream; } - pthread_mutex_lock(&stream->lock); pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&stream->lock); switch (consumer_data.type) { case LTTNG_CONSUMER_KERNEL: @@ -445,8 +443,8 @@ void consumer_del_stream(struct lttng_consumer_stream *stream, end: consumer_data.need_update = 1; - pthread_mutex_unlock(&consumer_data.lock); pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&consumer_data.lock); if (free_chan) { consumer_del_channel(free_chan); @@ -479,6 +477,8 @@ struct lttng_consumer_stream *consumer_allocate_stream( goto end; } + rcu_read_lock(); + /* * Get stream's channel reference. Needed when adding the stream to the * global hash table. @@ -535,9 +535,12 @@ struct lttng_consumer_stream *consumer_allocate_stream( stream->path_name, stream->key, stream->shm_fd, stream->wait_fd, (unsigned long long) stream->mmap_len, stream->out_fd, stream->net_seq_idx, stream->session_id); + + rcu_read_unlock(); return stream; error: + rcu_read_unlock(); free(stream); end: return NULL; @@ -558,6 +561,7 @@ static int consumer_add_stream(struct lttng_consumer_stream *stream, DBG3("Adding consumer stream %d", stream->key); pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&stream->lock); rcu_read_lock(); /* Steal stream identifier to avoid having streams with the same key */ @@ -597,6 +601,7 @@ static int consumer_add_stream(struct lttng_consumer_stream *stream, consumer_data.need_update = 1; rcu_read_unlock(); + pthread_mutex_unlock(&stream->lock); pthread_mutex_unlock(&consumer_data.lock); return ret; @@ -1282,8 +1287,6 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( /* RCU lock for the relayd pointer */ rcu_read_lock(); - pthread_mutex_lock(&stream->lock); - /* Flag that the current stream if set for network streaming. */ if (stream->net_seq_idx != -1) { relayd = consumer_find_relayd(stream->net_seq_idx); @@ -1408,7 +1411,6 @@ end: if (relayd && stream->metadata_flag) { pthread_mutex_unlock(&relayd->ctrl_sock_mutex); } - pthread_mutex_unlock(&stream->lock); rcu_read_unlock(); return written; @@ -1449,8 +1451,6 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( /* RCU lock for the relayd pointer */ rcu_read_lock(); - pthread_mutex_lock(&stream->lock); - /* Flag that the current stream if set for network streaming. */ if (stream->net_seq_idx != -1) { relayd = consumer_find_relayd(stream->net_seq_idx); @@ -1557,7 +1557,7 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( written = ret_splice; } /* Socket operation failed. We consider the relayd dead */ - if (errno == EBADF) { + if (errno == EBADF || errno == EPIPE) { WARN("Remote relayd disconnected. Stopping"); relayd_hang_up = 1; goto write_error; @@ -1618,7 +1618,6 @@ end: if (relayd && stream->metadata_flag) { pthread_mutex_unlock(&relayd->ctrl_sock_mutex); } - pthread_mutex_unlock(&stream->lock); rcu_read_unlock(); return written; @@ -1764,9 +1763,9 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, goto free_stream; } + pthread_mutex_lock(&consumer_data.lock); pthread_mutex_lock(&stream->lock); - pthread_mutex_lock(&consumer_data.lock); switch (consumer_data.type) { case LTTNG_CONSUMER_KERNEL: if (stream->mmap_base != NULL) { @@ -1856,8 +1855,8 @@ void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, } end: - pthread_mutex_unlock(&consumer_data.lock); pthread_mutex_unlock(&stream->lock); + pthread_mutex_unlock(&consumer_data.lock); if (free_chan) { consumer_del_channel(free_chan); @@ -1876,6 +1875,8 @@ static int consumer_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; assert(stream); assert(ht); @@ -1883,6 +1884,7 @@ static int consumer_add_metadata_stream(struct lttng_consumer_stream *stream, DBG3("Adding metadata stream %d to hash table", stream->wait_fd); pthread_mutex_lock(&consumer_data.lock); + pthread_mutex_lock(&stream->lock); /* * From here, refcounts are updated so be _careful_ when returning an error @@ -1890,6 +1892,15 @@ static int consumer_add_metadata_stream(struct lttng_consumer_stream *stream, */ rcu_read_lock(); + + /* + * 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->wait_fd), &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + assert(!node); + /* Find relayd and, if one is found, increment refcount. */ relayd = consumer_find_relayd(stream->net_seq_idx); if (relayd != NULL) { @@ -1910,9 +1921,6 @@ static int consumer_add_metadata_stream(struct lttng_consumer_stream *stream, uatomic_dec(&stream->chan->nb_init_streams); } - /* Steal stream identifier to avoid having streams with the same key */ - consumer_steal_stream_key(stream->key, ht); - lttng_ht_add_unique_ulong(ht, &stream->node); /* @@ -1924,6 +1932,7 @@ static int consumer_add_metadata_stream(struct lttng_consumer_stream *stream, rcu_read_unlock(); + pthread_mutex_unlock(&stream->lock); pthread_mutex_unlock(&consumer_data.lock); return ret; } @@ -1941,7 +1950,7 @@ static void validate_endpoint_status_data_stream(void) rcu_read_lock(); cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { /* Validate delete flag of the stream */ - if (!stream->endpoint_status) { + if (stream->endpoint_status != CONSUMER_ENDPOINT_INACTIVE) { continue; } /* Delete it right now */ @@ -2055,7 +2064,10 @@ restart: * since their might be data to consume. */ lttng_poll_del(&events, ctx->consumer_metadata_pipe[0]); - close(ctx->consumer_metadata_pipe[0]); + ret = close(ctx->consumer_metadata_pipe[0]); + if (ret < 0) { + PERROR("close metadata pipe"); + } continue; } else if (revents & LPOLLIN) { do { @@ -2248,7 +2260,7 @@ void *consumer_thread_data_poll(void *data) /* poll on the array of fds */ restart: DBG("polling on %d fd", nb_fd + 1); - num_rdy = poll(pollfd, nb_fd + 1, consumer_poll_timeout); + num_rdy = poll(pollfd, nb_fd + 1, -1); DBG("poll num_rdy : %d", num_rdy); if (num_rdy == -1) { /* @@ -2307,6 +2319,9 @@ void *consumer_thread_data_poll(void *data) /* Take care of high priority channels first. */ for (i = 0; i < nb_fd; i++) { + if (local_stream[i] == NULL) { + continue; + } if (pollfd[i].revents & POLLPRI) { DBG("Urgent read on fd %d", pollfd[i].fd); high_prio = 1; @@ -2315,6 +2330,7 @@ void *consumer_thread_data_poll(void *data) if (len < 0 && len != -EAGAIN && len != -ENODATA) { /* Clean the stream and free it. */ consumer_del_stream(local_stream[i], data_ht); + local_stream[i] = NULL; } else if (len > 0) { local_stream[i]->data_read = 1; } @@ -2331,6 +2347,9 @@ void *consumer_thread_data_poll(void *data) /* Take care of low priority channels. */ for (i = 0; i < nb_fd; i++) { + if (local_stream[i] == NULL) { + continue; + } if ((pollfd[i].revents & POLLIN) || local_stream[i]->hangup_flush_done) { DBG("Normal read on fd %d", pollfd[i].fd); @@ -2339,6 +2358,7 @@ void *consumer_thread_data_poll(void *data) if (len < 0 && len != -EAGAIN && len != -ENODATA) { /* Clean the stream and free it. */ consumer_del_stream(local_stream[i], data_ht); + local_stream[i] = NULL; } else if (len > 0) { local_stream[i]->data_read = 1; } @@ -2347,12 +2367,15 @@ void *consumer_thread_data_poll(void *data) /* Handle hangup and errors */ for (i = 0; i < nb_fd; i++) { + if (local_stream[i] == NULL) { + continue; + } if (!local_stream[i]->hangup_flush_done && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) && (consumer_data.type == LTTNG_CONSUMER32_UST || consumer_data.type == LTTNG_CONSUMER64_UST)) { DBG("fd %d is hup|err|nval. Attempting flush and read.", - pollfd[i].fd); + pollfd[i].fd); lttng_ustconsumer_on_stream_hangup(local_stream[i]); /* Attempt read again, for the data we just flushed. */ local_stream[i]->data_read = 1; @@ -2366,22 +2389,27 @@ void *consumer_thread_data_poll(void *data) DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); if (!local_stream[i]->data_read) { consumer_del_stream(local_stream[i], data_ht); + local_stream[i] = NULL; num_hup++; } } else if (pollfd[i].revents & POLLERR) { ERR("Error returned in polling fd %d.", pollfd[i].fd); if (!local_stream[i]->data_read) { consumer_del_stream(local_stream[i], data_ht); + local_stream[i] = NULL; num_hup++; } } else if (pollfd[i].revents & POLLNVAL) { ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); if (!local_stream[i]->data_read) { consumer_del_stream(local_stream[i], data_ht); + local_stream[i] = NULL; num_hup++; } } - local_stream[i]->data_read = 0; + if (local_stream[i] != NULL) { + local_stream[i]->data_read = 0; + } } } end: @@ -2403,7 +2431,10 @@ end: * only tracked fd in the poll set. The thread will take care of closing * the read side. */ - close(ctx->consumer_metadata_pipe[1]); + ret = close(ctx->consumer_metadata_pipe[1]); + if (ret < 0) { + PERROR("close data pipe"); + } if (data_ht) { destroy_data_stream_ht(data_ht); @@ -2516,13 +2547,6 @@ end: */ consumer_quit = 1; - /* - * 2s of grace period, if no polling events occur during - * this period, the polling thread will exit even if there - * are still open FDs (should not happen, but safety mechanism). - */ - consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT; - /* * Notify the data poll thread to poll back again and test the * consumer_quit state that we just set so to quit gracefully. @@ -2536,17 +2560,27 @@ end: ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, struct lttng_consumer_local_data *ctx) { + ssize_t ret; + + pthread_mutex_lock(&stream->lock); + switch (consumer_data.type) { case LTTNG_CONSUMER_KERNEL: - return lttng_kconsumer_read_subbuffer(stream, ctx); + ret = lttng_kconsumer_read_subbuffer(stream, ctx); + break; case LTTNG_CONSUMER32_UST: case LTTNG_CONSUMER64_UST: - return lttng_ustconsumer_read_subbuffer(stream, ctx); + ret = lttng_ustconsumer_read_subbuffer(stream, ctx); + break; default: ERR("Unknown consumer_data type"); assert(0); - return -ENOSYS; + ret = -ENOSYS; + break; } + + pthread_mutex_unlock(&stream->lock); + return ret; } int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) @@ -2630,7 +2664,10 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, } /* Close the created socket fd which is useless */ - close(relayd->control_sock.fd); + ret = close(relayd->control_sock.fd); + if (ret < 0) { + PERROR("close relayd control socket"); + } /* Assign new file descriptor */ relayd->control_sock.fd = fd; @@ -2644,7 +2681,10 @@ int consumer_add_relayd_socket(int net_seq_idx, int sock_type, } /* Close the created socket fd which is useless */ - close(relayd->data_sock.fd); + ret = close(relayd->data_sock.fd); + if (ret < 0) { + PERROR("close relayd control socket"); + } /* Assign new file descriptor */ relayd->data_sock.fd = fd; @@ -2736,7 +2776,7 @@ int consumer_data_pending(uint64_t id) ht = consumer_data.stream_list_ht; cds_lfht_for_each_entry_duplicate(ht->ht, - ht->hash_fct((void *)((unsigned long) id), 0x42UL), + ht->hash_fct((void *)((unsigned long) id), lttng_ht_seed), ht->match_fct, (void *)((unsigned long) id), &iter.iter, stream, node_session_id.node) { /* If this call fails, the stream is being used hence data pending. */