X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fcommon%2Fconsumer.c;h=008bf8e54904d71158400216393f0fc6c5786e63;hp=9c54b84cf963bb5f0ae853248023216f0d7e1cdb;hb=f73fabfda365d22e7dd180fb1614e37c446fbd9e;hpb=4c462e790c62ed5f6c5d61b3a182762fe02f7e9a diff --git a/src/common/consumer.c b/src/common/consumer.c index 9c54b84cf..008bf8e54 100644 --- a/src/common/consumer.c +++ b/src/common/consumer.c @@ -1,25 +1,24 @@ /* * Copyright (C) 2011 - Julien Desfossez * Mathieu Desnoyers + * 2012 - David Goulet * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public License - * as published by the Free Software Foundation; only version 2 - * of the License. + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License, version 2 only, + * as published by the Free Software Foundation. * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * This program is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. + * You should have received a copy of the GNU General Public License along + * with this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ #define _GNU_SOURCE #include -#include #include #include #include @@ -28,11 +27,14 @@ #include #include #include +#include #include #include +#include #include #include +#include #include #include "consumer.h" @@ -86,9 +88,18 @@ static void consumer_steal_stream_key(int key) { struct lttng_consumer_stream *stream; + rcu_read_lock(); stream = consumer_find_stream(key); - if (stream) + if (stream) { stream->key = -1; + /* + * 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; + } + rcu_read_unlock(); } static struct lttng_consumer_channel *consumer_find_channel(int key) @@ -119,9 +130,94 @@ static void consumer_steal_channel_key(int key) { struct lttng_consumer_channel *channel; + rcu_read_lock(); channel = consumer_find_channel(key); - if (channel) + if (channel) { channel->key = -1; + /* + * We don't want the lookup to match, but we still need + * to iterate on this channel when iterating over the hash table. Just + * change the node key. + */ + channel->node.key = -1; + } + rcu_read_unlock(); +} + +static +void consumer_free_stream(struct rcu_head *head) +{ + struct lttng_ht_node_ulong *node = + caa_container_of(head, struct lttng_ht_node_ulong, head); + struct lttng_consumer_stream *stream = + caa_container_of(node, struct lttng_consumer_stream, node); + + free(stream); +} + +/* + * RCU protected relayd socket pair free. + */ +static void consumer_rcu_free_relayd(struct rcu_head *head) +{ + struct lttng_ht_node_ulong *node = + caa_container_of(head, struct lttng_ht_node_ulong, head); + struct consumer_relayd_sock_pair *relayd = + caa_container_of(node, struct consumer_relayd_sock_pair, node); + + free(relayd); +} + +/* + * Destroy and free relayd socket pair object. + * + * This function MUST be called with the consumer_data lock acquired. + */ +void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) +{ + int ret; + struct lttng_ht_iter iter; + + if (relayd == NULL) { + return; + } + + DBG("Consumer destroy and close relayd socket pair"); + + iter.iter.node = &relayd->node.node; + ret = lttng_ht_del(consumer_data.relayd_ht, &iter); + if (ret != 0) { + /* We assume the relayd was already destroyed */ + return; + } + + /* Close all sockets */ + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + (void) relayd_close(&relayd->control_sock); + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + (void) relayd_close(&relayd->data_sock); + + /* RCU free() call */ + call_rcu(&relayd->node.head, consumer_rcu_free_relayd); +} + +/* + * Flag a relayd socket pair for destruction. Destroy it if the refcount + * reaches zero. + * + * RCU read side lock MUST be aquired before calling this function. + */ +void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) +{ + assert(relayd); + + /* Set destroy flag for this object */ + uatomic_set(&relayd->destroy_flag, 1); + + /* Destroy the relayd if refcount is 0 */ + if (uatomic_read(&relayd->refcount) == 0) { + consumer_destroy_relayd(relayd); + } } /* @@ -133,6 +229,9 @@ void consumer_del_stream(struct lttng_consumer_stream *stream) int ret; struct lttng_ht_iter iter; struct lttng_consumer_channel *free_chan = NULL; + struct consumer_relayd_sock_pair *relayd; + + assert(stream); pthread_mutex_lock(&consumer_data.lock); @@ -156,11 +255,7 @@ void consumer_del_stream(struct lttng_consumer_stream *stream) } rcu_read_lock(); - - /* Get stream node from hash table */ - lttng_ht_lookup(consumer_data.stream_ht, - (void *)((unsigned long) stream->key), &iter); - /* Remove stream node from hash table */ + iter.iter.node = &stream->node.node; ret = lttng_ht_del(consumer_data.stream_ht, &iter); assert(!ret); @@ -191,9 +286,43 @@ void consumer_del_stream(struct lttng_consumer_stream *stream) PERROR("close"); } } - if (!--stream->chan->refcount) + + /* 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)) { + consumer_destroy_relayd(relayd); + } + } + rcu_read_unlock(); + + if (!--stream->chan->refcount) { free_chan = stream->chan; - free(stream); + } + + + call_rcu(&stream->node.head, consumer_free_stream); end: consumer_data.need_update = 1; pthread_mutex_unlock(&consumer_data.lock); @@ -202,16 +331,6 @@ end: consumer_del_channel(free_chan); } -static void consumer_del_stream_rcu(struct rcu_head *head) -{ - struct lttng_ht_node_ulong *node = - caa_container_of(head, struct lttng_ht_node_ulong, head); - struct lttng_consumer_stream *stream = - caa_container_of(node, struct lttng_consumer_stream, node); - - consumer_del_stream(stream); -} - struct lttng_consumer_stream *consumer_allocate_stream( int channel_key, int stream_key, int shm_fd, int wait_fd, @@ -220,7 +339,9 @@ struct lttng_consumer_stream *consumer_allocate_stream( enum lttng_event_output output, const char *path_name, uid_t uid, - gid_t gid) + gid_t gid, + int net_index, + int metadata_flag) { struct lttng_consumer_stream *stream; int ret; @@ -247,9 +368,12 @@ struct lttng_consumer_stream *consumer_allocate_stream( stream->output = output; stream->uid = uid; stream->gid = gid; - strncpy(stream->path_name, path_name, PATH_MAX - 1); - stream->path_name[PATH_MAX - 1] = '\0'; + stream->net_seq_idx = net_index; + stream->metadata_flag = metadata_flag; + strncpy(stream->path_name, path_name, sizeof(stream->path_name)); + stream->path_name[sizeof(stream->path_name) - 1] = '\0'; lttng_ht_node_init_ulong(&stream->node, stream->key); + lttng_ht_node_init_ulong(&stream->waitfd_node, stream->wait_fd); switch (consumer_data.type) { case LTTNG_CONSUMER_KERNEL: @@ -268,12 +392,13 @@ struct lttng_consumer_stream *consumer_allocate_stream( assert(0); goto end; } - DBG("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, out_fd %d)", + DBG("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, out_fd %d, net_seq_idx %d)", stream->path_name, stream->key, stream->shm_fd, stream->wait_fd, (unsigned long long) stream->mmap_len, - stream->out_fd); + stream->out_fd, + stream->net_seq_idx); end: return stream; } @@ -284,13 +409,34 @@ end: int consumer_add_stream(struct lttng_consumer_stream *stream) { int ret = 0; + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter iter; + struct consumer_relayd_sock_pair *relayd; pthread_mutex_lock(&consumer_data.lock); /* Steal stream identifier, for UST */ consumer_steal_stream_key(stream->key); + rcu_read_lock(); + lttng_ht_lookup(consumer_data.stream_ht, + (void *)((unsigned long) stream->key), &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node != NULL) { + rcu_read_unlock(); + /* Stream already exist. Ignore the insertion */ + goto end; + } + lttng_ht_add_unique_ulong(consumer_data.stream_ht, &stream->node); + + /* Check and cleanup relayd */ + relayd = consumer_find_relayd(stream->net_seq_idx); + if (relayd != NULL) { + uatomic_inc(&relayd->refcount); + } rcu_read_unlock(); + + /* Update consumer data */ consumer_data.stream_count++; consumer_data.need_update = 1; @@ -310,9 +456,146 @@ int consumer_add_stream(struct lttng_consumer_stream *stream) end: pthread_mutex_unlock(&consumer_data.lock); + + return ret; +} + +/* + * Add relayd socket to global consumer data hashtable. RCU read side lock MUST + * be acquired before calling this. + */ + +int consumer_add_relayd(struct consumer_relayd_sock_pair *relayd) +{ + int ret = 0; + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter iter; + + if (relayd == NULL) { + ret = -1; + goto end; + } + + lttng_ht_lookup(consumer_data.relayd_ht, + (void *)((unsigned long) relayd->net_seq_idx), &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node != NULL) { + /* Relayd already exist. Ignore the insertion */ + goto end; + } + lttng_ht_add_unique_ulong(consumer_data.relayd_ht, &relayd->node); + +end: return ret; } +/* + * Allocate and return a consumer relayd socket. + */ +struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( + int net_seq_idx) +{ + struct consumer_relayd_sock_pair *obj = NULL; + + /* Negative net sequence index is a failure */ + if (net_seq_idx < 0) { + goto error; + } + + obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); + if (obj == NULL) { + PERROR("zmalloc relayd sock"); + goto error; + } + + 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); + pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); + +error: + return obj; +} + +/* + * Find a relayd socket pair in the global consumer data. + * + * Return the object if found else NULL. + * 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 lttng_ht_iter iter; + struct lttng_ht_node_ulong *node; + struct consumer_relayd_sock_pair *relayd = NULL; + + /* Negative keys are lookup failures */ + if (key < 0) { + goto error; + } + + lttng_ht_lookup(consumer_data.relayd_ht, (void *)((unsigned long) key), + &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node != NULL) { + relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); + } + +error: + return relayd; +} + +/* + * Handle stream for relayd transmission if the stream applies for network + * streaming where the net sequence index is set. + * + * Return destination file descriptor or negative value on error. + */ +static int write_relayd_stream_header(struct lttng_consumer_stream *stream, + size_t data_size, struct consumer_relayd_sock_pair *relayd) +{ + int outfd = -1, ret; + struct lttcomm_relayd_data_hdr data_hdr; + + /* Safety net */ + assert(stream); + assert(relayd); + + /* Reset data header */ + memset(&data_hdr, 0, sizeof(data_hdr)); + + if (stream->metadata_flag) { + /* Caller MUST acquire the relayd control socket lock */ + ret = relayd_send_metadata(&relayd->control_sock, data_size); + if (ret < 0) { + goto error; + } + + /* Metadata are always sent on the control socket. */ + outfd = relayd->control_sock.fd; + } else { + /* Set header with stream information */ + data_hdr.stream_id = htobe64(stream->relayd_stream_id); + data_hdr.data_size = htobe32(data_size); + data_hdr.net_seq_num = htobe64(stream->next_net_seq_num++); + /* Other fields are zeroed previously */ + + ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, + sizeof(data_hdr)); + if (ret < 0) { + goto error; + } + + /* Set to go on data socket */ + outfd = relayd->data_sock.fd; + } + +error: + return outfd; +} + /* * Update a stream according to what we just received. */ @@ -330,6 +613,17 @@ void consumer_change_stream_state(int stream_key, pthread_mutex_unlock(&consumer_data.lock); } +static +void consumer_free_channel(struct rcu_head *head) +{ + struct lttng_ht_node_ulong *node = + caa_container_of(head, struct lttng_ht_node_ulong, head); + struct lttng_consumer_channel *channel = + caa_container_of(node, struct lttng_consumer_channel, node); + + free(channel); +} + /* * Remove a channel from the global list protected by a mutex. This * function is also responsible for freeing its data structures. @@ -355,12 +649,9 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) } rcu_read_lock(); - - lttng_ht_lookup(consumer_data.channel_ht, - (void *)((unsigned long) channel->key), &iter); + iter.iter.node = &channel->node.node; ret = lttng_ht_del(consumer_data.channel_ht, &iter); assert(!ret); - rcu_read_unlock(); if (channel->mmap_base != NULL) { @@ -381,21 +672,12 @@ void consumer_del_channel(struct lttng_consumer_channel *channel) PERROR("close"); } } - free(channel); + + call_rcu(&channel->node.head, consumer_free_channel); end: pthread_mutex_unlock(&consumer_data.lock); } -static void consumer_del_channel_rcu(struct rcu_head *head) -{ - struct lttng_ht_node_ulong *node = - caa_container_of(head, struct lttng_ht_node_ulong, head); - struct lttng_consumer_channel *channel= - caa_container_of(node, struct lttng_consumer_channel, node); - - consumer_del_channel(channel); -} - struct lttng_consumer_channel *consumer_allocate_channel( int channel_key, int shm_fd, int wait_fd, @@ -438,9 +720,7 @@ struct lttng_consumer_channel *consumer_allocate_channel( goto end; } DBG("Allocated channel (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, max_sb_size %llu)", - channel->key, - channel->shm_fd, - channel->wait_fd, + channel->key, channel->shm_fd, channel->wait_fd, (unsigned long long) channel->mmap_len, (unsigned long long) channel->max_sb_size); end: @@ -452,13 +732,28 @@ end: */ int consumer_add_channel(struct lttng_consumer_channel *channel) { + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter iter; + pthread_mutex_lock(&consumer_data.lock); /* Steal channel identifier, for UST */ consumer_steal_channel_key(channel->key); rcu_read_lock(); + + lttng_ht_lookup(consumer_data.channel_ht, + (void *)((unsigned long) channel->key), &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node != NULL) { + /* Channel already exist. Ignore the insertion */ + goto end; + } + lttng_ht_add_unique_ulong(consumer_data.channel_ht, &channel->node); + +end: rcu_read_unlock(); pthread_mutex_unlock(&consumer_data.lock); + return 0; } @@ -471,13 +766,15 @@ int consumer_add_channel(struct lttng_consumer_channel *channel) */ int consumer_update_poll_array( struct lttng_consumer_local_data *ctx, struct pollfd **pollfd, - struct lttng_consumer_stream **local_stream) + struct lttng_consumer_stream **local_stream, + struct lttng_ht *metadata_ht) { int i = 0; struct lttng_ht_iter iter; struct lttng_consumer_stream *stream; DBG("Updating poll fd array"); + rcu_read_lock(); cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, stream, node.node) { if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM) { @@ -486,16 +783,21 @@ int consumer_update_poll_array( DBG("Active FD %d", stream->wait_fd); (*pollfd)[i].fd = stream->wait_fd; (*pollfd)[i].events = POLLIN | POLLPRI; + if (stream->metadata_flag && metadata_ht) { + lttng_ht_add_unique_ulong(metadata_ht, &stream->waitfd_node); + DBG("Active FD added to metadata hash table"); + } local_stream[i] = stream; i++; } + rcu_read_unlock(); /* * Insert the consumer_poll_pipe at the end of the array and don't * increment i so nb_fd is the number of real FD. */ (*pollfd)[i].fd = ctx->consumer_poll_pipe[0]; - (*pollfd)[i].events = POLLIN; + (*pollfd)[i].events = POLLIN | POLLPRI; return i; } @@ -519,7 +821,7 @@ restart: perror("Poll error"); goto exit; } - if (consumer_sockpoll[0].revents == POLLIN) { + if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { DBG("consumer_should_quit wake up"); goto exit; } @@ -541,7 +843,6 @@ void lttng_consumer_set_error_sock( /* * Set the command socket path. */ - void lttng_consumer_set_command_sock_path( struct lttng_consumer_local_data *ctx, char *sock) { @@ -569,7 +870,6 @@ int lttng_consumer_send_error( */ void lttng_consumer_cleanup(void) { - int ret; struct lttng_ht_iter iter; struct lttng_ht_node_ulong *node; @@ -581,19 +881,22 @@ void lttng_consumer_cleanup(void) */ cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, node, node) { - ret = lttng_ht_del(consumer_data.stream_ht, &iter); - assert(!ret); - call_rcu(&node->head, consumer_del_stream_rcu); + struct lttng_consumer_stream *stream = + caa_container_of(node, struct lttng_consumer_stream, node); + consumer_del_stream(stream); } cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, node, node) { - ret = lttng_ht_del(consumer_data.channel_ht, &iter); - assert(!ret); - call_rcu(&node->head, consumer_del_channel_rcu); + struct lttng_consumer_channel *channel = + caa_container_of(node, struct lttng_consumer_channel, node); + consumer_del_channel(channel); } rcu_read_unlock(); + + lttng_ht_destroy(consumer_data.stream_ht); + lttng_ht_destroy(consumer_data.channel_ht); } /* @@ -603,14 +906,16 @@ void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) { int ret; consumer_quit = 1; - ret = write(ctx->consumer_should_quit[1], "4", 1); + do { + ret = write(ctx->consumer_should_quit[1], "4", 1); + } while (ret < 0 && errno == EINTR); if (ret < 0) { perror("write consumer quit"); } } -void lttng_consumer_sync_trace_file( - struct lttng_consumer_stream *stream, off_t orig_offset) +void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, + off_t orig_offset) { int outfd = stream->out_fd; @@ -623,7 +928,7 @@ void lttng_consumer_sync_trace_file( if (orig_offset < stream->chan->max_sb_size) { return; } - sync_file_range(outfd, orig_offset - stream->chan->max_sb_size, + lttng_sync_file_range(outfd, orig_offset - stream->chan->max_sb_size, stream->chan->max_sb_size, SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE @@ -694,6 +999,20 @@ struct lttng_consumer_local_data *lttng_consumer_create( goto error_poll_pipe; } + /* set read end of the pipe to non-blocking */ + ret = fcntl(ctx->consumer_poll_pipe[0], F_SETFL, O_NONBLOCK); + if (ret < 0) { + perror("fcntl O_NONBLOCK"); + goto error_poll_fcntl; + } + + /* set write end of the pipe to non-blocking */ + ret = fcntl(ctx->consumer_poll_pipe[1], F_SETFL, O_NONBLOCK); + if (ret < 0) { + perror("fcntl O_NONBLOCK"); + goto error_poll_fcntl; + } + ret = pipe(ctx->consumer_should_quit); if (ret < 0) { perror("Error creating recv pipe"); @@ -718,6 +1037,7 @@ error_thread_pipe: PERROR("close"); } } +error_poll_fcntl: error_quit_pipe: for (i = 0; i < 2; i++) { int err; @@ -773,7 +1093,37 @@ void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) } /* - * Mmap the ring buffer, read it and write the data to the tracefile. + * Write the metadata stream id on the specified file descriptor. + */ +static int write_relayd_metadata_id(int fd, + struct lttng_consumer_stream *stream, + struct consumer_relayd_sock_pair *relayd) +{ + int ret; + uint64_t metadata_id; + + metadata_id = htobe64(stream->relayd_stream_id); + do { + ret = write(fd, (void *) &metadata_id, + sizeof(stream->relayd_stream_id)); + } while (ret < 0 && errno == EINTR); + if (ret < 0) { + PERROR("write metadata stream id"); + goto end; + } + DBG("Metadata stream id %" PRIu64 " written before data", + stream->relayd_stream_id); + +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 + * the network. + * + * Careful review MUST be put if any changes occur! * * Returns the number of bytes written */ @@ -781,16 +1131,115 @@ ssize_t lttng_consumer_on_read_subbuffer_mmap( struct lttng_consumer_local_data *ctx, struct lttng_consumer_stream *stream, unsigned long len) { + unsigned long mmap_offset; + ssize_t ret = 0, written = 0; + off_t orig_offset = stream->out_fd_offset; + /* Default is on the disk */ + int outfd = stream->out_fd; + struct consumer_relayd_sock_pair *relayd = NULL; + + /* RCU lock for the relayd pointer */ + rcu_read_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); + if (relayd == NULL) { + goto end; + } + } + + /* get the offset inside the fd to mmap */ switch (consumer_data.type) { case LTTNG_CONSUMER_KERNEL: - return lttng_kconsumer_on_read_subbuffer_mmap(ctx, stream, len); + ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); + break; case LTTNG_CONSUMER32_UST: case LTTNG_CONSUMER64_UST: - return lttng_ustconsumer_on_read_subbuffer_mmap(ctx, stream, len); + ret = lttng_ustctl_get_mmap_read_offset(stream->chan->handle, + stream->buf, &mmap_offset); + break; default: ERR("Unknown consumer_data type"); assert(0); } + if (ret != 0) { + errno = -ret; + PERROR("tracer ctl get_mmap_read_offset"); + written = ret; + goto end; + } + + /* Handle stream on the relayd if the output is on the network */ + if (relayd) { + unsigned long netlen = len; + + /* + * Lock the control socket for the complete duration of the function + * since from this point on we will use the socket. + */ + if (stream->metadata_flag) { + /* Metadata requires the control socket. */ + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + netlen += sizeof(stream->relayd_stream_id); + } + + ret = write_relayd_stream_header(stream, netlen, relayd); + if (ret >= 0) { + /* Use the returned socket. */ + outfd = ret; + + /* Write metadata stream id before payload */ + if (stream->metadata_flag) { + ret = write_relayd_metadata_id(outfd, stream, relayd); + if (ret < 0) { + written = ret; + goto end; + } + } + } + /* Else, use the default set before which is the filesystem. */ + } + + while (len > 0) { + do { + ret = write(outfd, stream->mmap_base + mmap_offset, len); + } while (ret < 0 && errno == EINTR); + if (ret < 0) { + PERROR("Error in file write"); + if (written == 0) { + written = ret; + } + goto end; + } else if (ret > len) { + PERROR("Error in file write (ret %zd > len %lu)", ret, len); + written += ret; + goto end; + } else { + len -= ret; + mmap_offset += ret; + } + DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); + + /* This call is useless on a socket so better save a syscall. */ + if (!relayd) { + /* This won't block, but will start writeout asynchronously */ + lttng_sync_file_range(outfd, stream->out_fd_offset, ret, + SYNC_FILE_RANGE_WRITE); + stream->out_fd_offset += ret; + } + written += ret; + } + lttng_consumer_sync_trace_file(stream, orig_offset); + +end: + /* Unlock only if ctrl socket used */ + if (relayd && stream->metadata_flag) { + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + } + + rcu_read_unlock(); + return written; } /* @@ -802,18 +1251,151 @@ ssize_t lttng_consumer_on_read_subbuffer_splice( struct lttng_consumer_local_data *ctx, struct lttng_consumer_stream *stream, unsigned long len) { + ssize_t ret = 0, written = 0, ret_splice = 0; + loff_t offset = 0; + off_t orig_offset = stream->out_fd_offset; + int fd = stream->wait_fd; + /* Default is on the disk */ + int outfd = stream->out_fd; + struct consumer_relayd_sock_pair *relayd = NULL; + switch (consumer_data.type) { case LTTNG_CONSUMER_KERNEL: - return lttng_kconsumer_on_read_subbuffer_splice(ctx, stream, len); + break; case LTTNG_CONSUMER32_UST: case LTTNG_CONSUMER64_UST: + /* Not supported for user space tracing */ return -ENOSYS; default: ERR("Unknown consumer_data type"); assert(0); - return -ENOSYS; } + /* RCU lock for the relayd pointer */ + rcu_read_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); + if (relayd == NULL) { + goto end; + } + } + + /* Write metadata stream id before payload */ + if (stream->metadata_flag && relayd) { + /* + * Lock the control socket for the complete duration of the function + * since from this point on we will use the socket. + */ + pthread_mutex_lock(&relayd->ctrl_sock_mutex); + + ret = write_relayd_metadata_id(ctx->consumer_thread_pipe[1], + stream, relayd); + if (ret < 0) { + written = ret; + goto end; + } + } + + while (len > 0) { + DBG("splice chan to pipe offset %lu of len %lu (fd : %d)", + (unsigned long)offset, len, fd); + ret_splice = splice(fd, &offset, ctx->consumer_thread_pipe[1], NULL, len, + SPLICE_F_MOVE | SPLICE_F_MORE); + DBG("splice chan to pipe, ret %zd", ret_splice); + if (ret_splice < 0) { + PERROR("Error in relay splice"); + if (written == 0) { + written = ret_splice; + } + ret = errno; + goto splice_error; + } + + /* Handle stream on the relayd if the output is on the network */ + if (relayd) { + if (stream->metadata_flag) { + /* Update counter to fit the spliced data */ + ret_splice += sizeof(stream->relayd_stream_id); + len += sizeof(stream->relayd_stream_id); + /* + * We do this so the return value can match the len passed as + * argument to this function. + */ + written -= sizeof(stream->relayd_stream_id); + } + + ret = write_relayd_stream_header(stream, ret_splice, relayd); + if (ret >= 0) { + /* Use the returned socket. */ + outfd = ret; + } else { + ERR("Remote relayd disconnected. Stopping"); + goto end; + } + } + + /* Splice data out */ + ret_splice = splice(ctx->consumer_thread_pipe[0], NULL, outfd, NULL, + ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); + DBG("Kernel consumer splice pipe to file, ret %zd", ret_splice); + if (ret_splice < 0) { + PERROR("Error in file splice"); + if (written == 0) { + written = ret_splice; + } + ret = errno; + goto splice_error; + } else if (ret_splice > len) { + errno = EINVAL; + PERROR("Wrote more data than requested %zd (len: %lu)", + ret_splice, len); + written += ret_splice; + ret = errno; + goto splice_error; + } + len -= ret_splice; + + /* This call is useless on a socket so better save a syscall. */ + if (!relayd) { + /* This won't block, but will start writeout asynchronously */ + lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, + SYNC_FILE_RANGE_WRITE); + stream->out_fd_offset += ret_splice; + } + written += ret_splice; + } + lttng_consumer_sync_trace_file(stream, orig_offset); + + ret = ret_splice; + + goto end; + +splice_error: + /* send the appropriate error description to sessiond */ + switch (ret) { + case EBADF: + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EBADF); + break; + case EINVAL: + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); + break; + case ENOMEM: + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); + break; + case ESPIPE: + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); + break; + } + +end: + if (relayd && stream->metadata_flag) { + pthread_mutex_unlock(&relayd->ctrl_sock_mutex); + } + + rcu_read_unlock(); + return written; } /* @@ -889,9 +1471,14 @@ void *lttng_consumer_thread_poll_fds(void *data) struct lttng_consumer_stream **local_stream = NULL; /* local view of consumer_data.fds_count */ int nb_fd = 0; - char tmp; - int tmp2; struct lttng_consumer_local_data *ctx = data; + struct lttng_ht *metadata_ht; + struct lttng_ht_iter iter; + struct lttng_ht_node_ulong *node; + struct lttng_consumer_stream *metadata_stream; + ssize_t len; + + metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); rcu_register_thread(); @@ -932,10 +1519,11 @@ void *lttng_consumer_thread_poll_fds(void *data) pthread_mutex_unlock(&consumer_data.lock); goto end; } - ret = consumer_update_poll_array(ctx, &pollfd, local_stream); + ret = consumer_update_poll_array(ctx, &pollfd, local_stream, + metadata_ht); if (ret < 0) { ERR("Error in allocating pollfd or local_outfds"); - lttng_consumer_send_error(ctx, CONSUMERD_POLL_ERROR); + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); pthread_mutex_unlock(&consumer_data.lock); goto end; } @@ -961,7 +1549,7 @@ void *lttng_consumer_thread_poll_fds(void *data) goto restart; } perror("Poll error"); - lttng_consumer_send_error(ctx, CONSUMERD_POLL_ERROR); + lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); goto end; } else if (num_rdy == 0) { DBG("Polling thread timed out"); @@ -969,25 +1557,42 @@ void *lttng_consumer_thread_poll_fds(void *data) } /* - * If the consumer_poll_pipe triggered poll go - * directly to the beginning of the loop to update the - * array. We want to prioritize array update over - * low-priority reads. + * If the consumer_poll_pipe triggered poll go directly to the + * beginning of the loop to update the array. We want to prioritize + * array update over low-priority reads. */ - if (pollfd[nb_fd].revents & POLLIN) { + if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { + size_t pipe_readlen; + char tmp; + DBG("consumer_poll_pipe wake up"); - tmp2 = read(ctx->consumer_poll_pipe[0], &tmp, 1); - if (tmp2 < 0) { - perror("read consumer poll"); - } + /* Consume 1 byte of pipe data */ + do { + pipe_readlen = read(ctx->consumer_poll_pipe[0], &tmp, 1); + } while (pipe_readlen == -1 && errno == EINTR); continue; } /* Take care of high priority channels first. */ for (i = 0; i < nb_fd; i++) { - if (pollfd[i].revents & POLLPRI) { - ssize_t len; - + /* Lookup for metadata which is the highest priority */ + lttng_ht_lookup(metadata_ht, + (void *)((unsigned long) pollfd[i].fd), &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node != NULL && + (pollfd[i].revents & (POLLIN | POLLPRI))) { + DBG("Urgent metadata read on fd %d", pollfd[i].fd); + metadata_stream = caa_container_of(node, + struct lttng_consumer_stream, waitfd_node); + high_prio = 1; + len = ctx->on_buffer_ready(metadata_stream, ctx); + /* it's ok to have an unavailable sub-buffer */ + if (len < 0 && len != -EAGAIN) { + goto end; + } else if (len > 0) { + metadata_stream->data_read = 1; + } + } else if (pollfd[i].revents & POLLPRI) { DBG("Urgent read on fd %d", pollfd[i].fd); high_prio = 1; len = ctx->on_buffer_ready(local_stream[i], ctx); @@ -1012,10 +1617,6 @@ void *lttng_consumer_thread_poll_fds(void *data) for (i = 0; i < nb_fd; i++) { if ((pollfd[i].revents & POLLIN) || local_stream[i]->hangup_flush_done) { - ssize_t len; - - assert(!(pollfd[i].revents & POLLERR)); - assert(!(pollfd[i].revents & POLLNVAL)); DBG("Normal read on fd %d", pollfd[i].fd); len = ctx->on_buffer_ready(local_stream[i], ctx); /* it's ok to have an unavailable sub-buffer */ @@ -1047,25 +1648,34 @@ void *lttng_consumer_thread_poll_fds(void *data) if ((pollfd[i].revents & POLLHUP)) { DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); if (!local_stream[i]->data_read) { - rcu_read_lock(); - consumer_del_stream_rcu(&local_stream[i]->node.head); - rcu_read_unlock(); + if (local_stream[i]->metadata_flag) { + iter.iter.node = &local_stream[i]->waitfd_node.node; + ret = lttng_ht_del(metadata_ht, &iter); + assert(!ret); + } + consumer_del_stream(local_stream[i]); num_hup++; } } else if (pollfd[i].revents & POLLERR) { ERR("Error returned in polling fd %d.", pollfd[i].fd); if (!local_stream[i]->data_read) { - rcu_read_lock(); - consumer_del_stream_rcu(&local_stream[i]->node.head); - rcu_read_unlock(); + if (local_stream[i]->metadata_flag) { + iter.iter.node = &local_stream[i]->waitfd_node.node; + ret = lttng_ht_del(metadata_ht, &iter); + assert(!ret); + } + consumer_del_stream(local_stream[i]); 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) { - rcu_read_lock(); - consumer_del_stream_rcu(&local_stream[i]->node.head); - rcu_read_unlock(); + if (local_stream[i]->metadata_flag) { + iter.iter.node = &local_stream[i]->waitfd_node.node; + ret = lttng_ht_del(metadata_ht, &iter); + assert(!ret); + } + consumer_del_stream(local_stream[i]); num_hup++; } } @@ -1116,7 +1726,7 @@ void *lttng_consumer_thread_receive_fds(void *data) } DBG("Sending ready command to lttng-sessiond"); - ret = lttng_consumer_send_error(ctx, CONSUMERD_COMMAND_SOCK_READY); + ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); /* return < 0 on error, but == 0 is not fatal */ if (ret < 0) { ERR("Error sending ready command to lttng-sessiond"); @@ -1192,11 +1802,20 @@ end: */ consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT; - /* wake up the polling thread */ - ret = write(ctx->consumer_poll_pipe[1], "4", 1); - if (ret < 0) { - perror("poll pipe write"); - } + /* + * Wake-up the other end by writing a null byte in the pipe + * (non-blocking). Important note: Because writing into the + * pipe is non-blocking (and therefore we allow dropping wakeup + * data, as long as there is wakeup data present in the pipe + * buffer to wake up the other end), the other end should + * perform the following sequence for waiting: + * 1) empty the pipe (reads). + * 2) perform update operation. + * 3) wait on the pipe (poll). + */ + do { + ret = write(ctx->consumer_poll_pipe[1], "", 1); + } while (ret < 0 && errno == EINTR); rcu_unregister_thread(); return NULL; } @@ -1239,5 +1858,97 @@ void lttng_consumer_init(void) { consumer_data.stream_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); } +/* + * Process the ADD_RELAYD command receive by a consumer. + * + * 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, + struct lttng_consumer_local_data *ctx, int sock, + struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock) +{ + int fd, ret = -1; + struct consumer_relayd_sock_pair *relayd; + + DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx); + + /* 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); + goto error; + } + } + + /* Poll on consumer socket. */ + if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { + ret = -EINTR; + goto error; + } + + /* 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; + goto error; + } + + /* Copy socket information and received FD */ + switch (sock_type) { + case LTTNG_STREAM_CONTROL: + /* Copy received lttcomm socket */ + lttcomm_copy_sock(&relayd->control_sock, relayd_sock); + ret = lttcomm_create_sock(&relayd->control_sock); + if (ret < 0) { + goto error; + } + + /* Close the created socket fd which is useless */ + close(relayd->control_sock.fd); + + /* Assign new file descriptor */ + relayd->control_sock.fd = fd; + break; + case LTTNG_STREAM_DATA: + /* Copy received lttcomm socket */ + lttcomm_copy_sock(&relayd->data_sock, relayd_sock); + ret = lttcomm_create_sock(&relayd->data_sock); + if (ret < 0) { + goto error; + } + + /* Close the created socket fd which is useless */ + close(relayd->data_sock.fd); + + /* Assign new file descriptor */ + relayd->data_sock.fd = fd; + break; + default: + ERR("Unknown relayd socket type (%d)", sock_type); + goto error; + } + + DBG("Consumer %s socket created successfully with net idx %d (fd: %d)", + sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", + relayd->net_seq_idx, fd); + + /* + * Add relayd socket pair to consumer data hashtable. If object already + * exists or on error, the function gracefully returns. + */ + consumer_add_relayd(relayd); + + /* All good! */ + ret = 0; + +error: + return ret; +}