Fix: sessiond: consumer.c: rotation error handling
[lttng-tools.git] / src / bin / lttng-sessiond / consumer.c
index f0870e6b0c0abcf2ae50e83c32df5ef2b162a33e..b6c9d3055705b04bea1e848750f76cf1a1d86912 100644 (file)
@@ -1,5 +1,6 @@
 /*
  * Copyright (C) 2012 - David Goulet <dgoulet@efficios.com>
+ *               2018 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
  *
  * 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
@@ -15,7 +16,7 @@
  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  */
 
-#define _GNU_SOURCE
+#define _LGPL_SOURCE
 #include <assert.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -31,7 +32,7 @@
 #include <common/relayd/relayd.h>
 
 #include "consumer.h"
-#include "health.h"
+#include "health-sessiond.h"
 #include "ust-app.h"
 #include "utils.h"
 
@@ -142,7 +143,7 @@ int consumer_recv_status_reply(struct consumer_socket *sock)
                goto end;
        }
 
-       if (reply.ret_code == LTTNG_OK) {
+       if (reply.ret_code == LTTCOMM_CONSUMERD_SUCCESS) {
                /* All good. */
                ret = 0;
        } else {
@@ -178,13 +179,14 @@ int consumer_recv_status_channel(struct consumer_socket *sock,
        }
 
        /* An error is possible so don't touch the key and stream_count. */
-       if (reply.ret_code != LTTNG_OK) {
+       if (reply.ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
                ret = -1;
                goto end;
        }
 
        *key = reply.key;
        *stream_count = reply.stream_count;
+       ret = 0;
 
 end:
        return ret;
@@ -206,6 +208,7 @@ int consumer_send_destroy_relayd(struct consumer_socket *sock,
 
        DBG2("Sending destroy relayd command to consumer sock %d", *sock->fd_ptr);
 
+       memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_DESTROY_RELAYD;
        msg.u.destroy_relayd.net_seq_idx = consumer->net_seq_index;
 
@@ -470,6 +473,7 @@ struct consumer_output *consumer_create_output(enum consumer_dst_type type)
        output->enabled = 1;
        output->type = type;
        output->net_seq_index = (uint64_t) -1ULL;
+       urcu_ref_init(&output->ref);
 
        output->socks = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
 
@@ -504,11 +508,10 @@ void consumer_destroy_output_sockets(struct consumer_output *obj)
  *
  * Should *NOT* be called with RCU read-side lock held.
  */
-void consumer_destroy_output(struct consumer_output *obj)
+static void consumer_release_output(struct urcu_ref *ref)
 {
-       if (obj == NULL) {
-               return;
-       }
+       struct consumer_output *obj =
+               caa_container_of(ref, struct consumer_output, ref);
 
        consumer_destroy_output_sockets(obj);
 
@@ -520,6 +523,27 @@ void consumer_destroy_output(struct consumer_output *obj)
        free(obj);
 }
 
+/*
+ * Get the consumer_output object.
+ */
+void consumer_output_get(struct consumer_output *obj)
+{
+       urcu_ref_get(&obj->ref);
+}
+
+/*
+ * Put the consumer_output object.
+ *
+ * Should *NOT* be called with RCU read-side lock held.
+ */
+void consumer_output_put(struct consumer_output *obj)
+{
+       if (!obj) {
+               return;
+       }
+       urcu_ref_put(&obj->ref, consumer_release_output);
+}
+
 /*
  * Copy consumer output and returned the newly allocated copy.
  *
@@ -528,33 +552,30 @@ void consumer_destroy_output(struct consumer_output *obj)
 struct consumer_output *consumer_copy_output(struct consumer_output *obj)
 {
        int ret;
-       struct lttng_ht *tmp_ht_ptr;
        struct consumer_output *output;
 
        assert(obj);
 
        output = consumer_create_output(obj->type);
        if (output == NULL) {
-               goto error;
+               goto end;
        }
-       /* Avoid losing the HT reference after the memcpy() */
-       tmp_ht_ptr = output->socks;
-
-       memcpy(output, obj, sizeof(struct consumer_output));
-
-       /* Putting back the HT pointer and start copying socket(s). */
-       output->socks = tmp_ht_ptr;
-
+       output->enabled = obj->enabled;
+       output->net_seq_index = obj->net_seq_index;
+       memcpy(output->subdir, obj->subdir, sizeof(output->subdir));
+       output->snapshot = obj->snapshot;
+       output->relay_major_version = obj->relay_major_version;
+       output->relay_minor_version = obj->relay_minor_version;
+       memcpy(&output->dst, &obj->dst, sizeof(output->dst));
        ret = consumer_copy_sockets(output, obj);
        if (ret < 0) {
-               goto malloc_error;
+               goto error_put;
        }
-
-error:
+end:
        return output;
 
-malloc_error:
-       consumer_destroy_output(output);
+error_put:
+       consumer_output_put(output);
        return NULL;
 }
 
@@ -695,8 +716,12 @@ int consumer_set_network_uri(struct consumer_output *obj,
                        goto error;
                }
 
-               strncpy(obj->subdir, tmp_path, sizeof(obj->subdir));
-               DBG3("Consumer set network uri subdir path %s", tmp_path);
+               if (lttng_strncpy(obj->dst.net.base_dir, tmp_path,
+                               sizeof(obj->dst.net.base_dir))) {
+                       ret = -LTTNG_ERR_INVALID;
+                       goto error;
+               }
+               DBG3("Consumer set network uri base_dir path %s", tmp_path);
        }
 
        return 0;
@@ -708,14 +733,18 @@ error:
 
 /*
  * Send file descriptor to consumer via sock.
+ *
+ * The consumer socket lock must be held by the caller.
  */
-int consumer_send_fds(struct consumer_socket *sock, int *fds, size_t nb_fd)
+int consumer_send_fds(struct consumer_socket *sock, const int *fds,
+               size_t nb_fd)
 {
        int ret;
 
        assert(fds);
        assert(sock);
        assert(nb_fd > 0);
+       assert(pthread_mutex_trylock(sock->lock) == EBUSY);
 
        ret = lttcomm_send_fds_unix_sock(*sock->fd_ptr, fds, nb_fd);
        if (ret < 0) {
@@ -725,13 +754,14 @@ int consumer_send_fds(struct consumer_socket *sock, int *fds, size_t nb_fd)
        }
 
        ret = consumer_recv_status_reply(sock);
-
 error:
        return ret;
 }
 
 /*
  * Consumer send communication message structure to consumer.
+ *
+ * The consumer socket lock must be held by the caller.
  */
 int consumer_send_msg(struct consumer_socket *sock,
                struct lttcomm_consumer_msg *msg)
@@ -740,6 +770,7 @@ int consumer_send_msg(struct consumer_socket *sock,
 
        assert(msg);
        assert(sock);
+       assert(pthread_mutex_trylock(sock->lock) == EBUSY);
 
        ret = consumer_socket_send(sock, msg, sizeof(struct lttcomm_consumer_msg));
        if (ret < 0) {
@@ -754,6 +785,8 @@ error:
 
 /*
  * Consumer send channel communication message structure to consumer.
+ *
+ * The consumer socket lock must be held by the caller.
  */
 int consumer_send_channel(struct consumer_socket *sock,
                struct lttcomm_consumer_msg *msg)
@@ -783,6 +816,7 @@ void consumer_init_ask_channel_comm_msg(struct lttcomm_consumer_msg *msg,
                unsigned int switch_timer_interval,
                unsigned int read_timer_interval,
                unsigned int live_timer_interval,
+               unsigned int monitor_timer_interval,
                int output,
                int type,
                uint64_t session_id,
@@ -798,7 +832,11 @@ void consumer_init_ask_channel_comm_msg(struct lttcomm_consumer_msg *msg,
                uint64_t tracefile_count,
                uint64_t session_id_per_pid,
                unsigned int monitor,
-               uint32_t ust_app_uid)
+               uint32_t ust_app_uid,
+               int64_t blocking_timeout,
+               const char *root_shm_path,
+               const char *shm_path,
+               uint64_t trace_archive_id)
 {
        assert(msg);
 
@@ -812,6 +850,7 @@ void consumer_init_ask_channel_comm_msg(struct lttcomm_consumer_msg *msg,
        msg->u.ask_channel.switch_timer_interval = switch_timer_interval;
        msg->u.ask_channel.read_timer_interval = read_timer_interval;
        msg->u.ask_channel.live_timer_interval = live_timer_interval;
+       msg->u.ask_channel.monitor_timer_interval = monitor_timer_interval;
        msg->u.ask_channel.output = output;
        msg->u.ask_channel.type = type;
        msg->u.ask_channel.session_id = session_id;
@@ -825,6 +864,8 @@ void consumer_init_ask_channel_comm_msg(struct lttcomm_consumer_msg *msg,
        msg->u.ask_channel.tracefile_count = tracefile_count;
        msg->u.ask_channel.monitor = monitor;
        msg->u.ask_channel.ust_app_uid = ust_app_uid;
+       msg->u.ask_channel.blocking_timeout = blocking_timeout;
+       msg->u.ask_channel.trace_archive_id = trace_archive_id;
 
        memcpy(msg->u.ask_channel.uuid, uuid, sizeof(msg->u.ask_channel.uuid));
 
@@ -836,13 +877,23 @@ void consumer_init_ask_channel_comm_msg(struct lttcomm_consumer_msg *msg,
 
        strncpy(msg->u.ask_channel.name, name, sizeof(msg->u.ask_channel.name));
        msg->u.ask_channel.name[sizeof(msg->u.ask_channel.name) - 1] = '\0';
+
+       if (root_shm_path) {
+               strncpy(msg->u.ask_channel.root_shm_path, root_shm_path,
+                       sizeof(msg->u.ask_channel.root_shm_path));
+               msg->u.ask_channel.root_shm_path[sizeof(msg->u.ask_channel.root_shm_path) - 1] = '\0';
+       }
+       if (shm_path) {
+               strncpy(msg->u.ask_channel.shm_path, shm_path,
+                       sizeof(msg->u.ask_channel.shm_path));
+               msg->u.ask_channel.shm_path[sizeof(msg->u.ask_channel.shm_path) - 1] = '\0';
+       }
 }
 
 /*
  * Init channel communication message structure.
  */
-void consumer_init_channel_comm_msg(struct lttcomm_consumer_msg *msg,
-               enum lttng_consumer_command cmd,
+void consumer_init_add_channel_comm_msg(struct lttcomm_consumer_msg *msg,
                uint64_t channel_key,
                uint64_t session_id,
                const char *pathname,
@@ -856,7 +907,8 @@ void consumer_init_channel_comm_msg(struct lttcomm_consumer_msg *msg,
                uint64_t tracefile_size,
                uint64_t tracefile_count,
                unsigned int monitor,
-               unsigned int live_timer_interval)
+               unsigned int live_timer_interval,
+               unsigned int monitor_timer_interval)
 {
        assert(msg);
 
@@ -864,7 +916,7 @@ void consumer_init_channel_comm_msg(struct lttcomm_consumer_msg *msg,
        memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
 
        /* Send channel */
-       msg->cmd_type = cmd;
+       msg->cmd_type = LTTNG_CONSUMER_ADD_CHANNEL;
        msg->u.channel.channel_key = channel_key;
        msg->u.channel.session_id = session_id;
        msg->u.channel.uid = uid;
@@ -877,6 +929,7 @@ void consumer_init_channel_comm_msg(struct lttcomm_consumer_msg *msg,
        msg->u.channel.tracefile_count = tracefile_count;
        msg->u.channel.monitor = monitor;
        msg->u.channel.live_timer_interval = live_timer_interval;
+       msg->u.channel.monitor_timer_interval = monitor_timer_interval;
 
        strncpy(msg->u.channel.pathname, pathname,
                        sizeof(msg->u.channel.pathname));
@@ -889,20 +942,34 @@ void consumer_init_channel_comm_msg(struct lttcomm_consumer_msg *msg,
 /*
  * Init stream communication message structure.
  */
-void consumer_init_stream_comm_msg(struct lttcomm_consumer_msg *msg,
-               enum lttng_consumer_command cmd,
+void consumer_init_add_stream_comm_msg(struct lttcomm_consumer_msg *msg,
                uint64_t channel_key,
                uint64_t stream_key,
-               int cpu)
+               int32_t cpu,
+               uint64_t trace_archive_id)
 {
        assert(msg);
 
        memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
 
-       msg->cmd_type = cmd;
+       msg->cmd_type = LTTNG_CONSUMER_ADD_STREAM;
        msg->u.stream.channel_key = channel_key;
        msg->u.stream.stream_key = stream_key;
        msg->u.stream.cpu = cpu;
+       msg->u.stream.trace_archive_id = trace_archive_id;
+}
+
+void consumer_init_streams_sent_comm_msg(struct lttcomm_consumer_msg *msg,
+               enum lttng_consumer_command cmd,
+               uint64_t channel_key, uint64_t net_seq_idx)
+{
+       assert(msg);
+
+       memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
+
+       msg->cmd_type = cmd;
+       msg->u.sent_streams.channel_key = channel_key;
+       msg->u.sent_streams.net_seq_idx = net_seq_idx;
 }
 
 /*
@@ -910,7 +977,7 @@ void consumer_init_stream_comm_msg(struct lttcomm_consumer_msg *msg,
  */
 int consumer_send_stream(struct consumer_socket *sock,
                struct consumer_output *dst, struct lttcomm_consumer_msg *msg,
-               int *fds, size_t nb_fd)
+               const int *fds, size_t nb_fd)
 {
        int ret;
 
@@ -936,6 +1003,8 @@ error:
 /*
  * Send relayd socket to consumer associated with a session name.
  *
+ * The consumer socket lock must be held by the caller.
+ *
  * On success return positive value. On error, negative value.
  */
 int consumer_send_relayd_socket(struct consumer_socket *consumer_sock,
@@ -951,6 +1020,7 @@ int consumer_send_relayd_socket(struct consumer_socket *consumer_sock,
        assert(consumer);
        assert(consumer_sock);
 
+       memset(&msg, 0, sizeof(msg));
        /* Bail out if consumer is disabled */
        if (!consumer->enabled) {
                ret = LTTNG_OK;
@@ -960,7 +1030,8 @@ int consumer_send_relayd_socket(struct consumer_socket *consumer_sock,
        if (type == LTTNG_STREAM_CONTROL) {
                ret = relayd_create_session(rsock,
                                &msg.u.relayd_sock.relayd_session_id,
-                               session_name, hostname, session_live_timer);
+                               session_name, hostname, session_live_timer,
+                               consumer->snapshot);
                if (ret < 0) {
                        /* Close the control socket. */
                        (void) relayd_close(rsock);
@@ -997,6 +1068,59 @@ error:
        return ret;
 }
 
+static
+int consumer_send_pipe(struct consumer_socket *consumer_sock,
+               enum lttng_consumer_command cmd, int pipe)
+{
+       int ret;
+       struct lttcomm_consumer_msg msg;
+       const char *pipe_name;
+       const char *command_name;
+
+       switch (cmd) {
+       case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
+               pipe_name = "channel monitor";
+               command_name = "SET_CHANNEL_MONITOR_PIPE";
+               break;
+       default:
+               ERR("Unexpected command received in %s (cmd = %d)", __func__,
+                               (int) cmd);
+               abort();
+       }
+
+       /* Code flow error. Safety net. */
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = cmd;
+
+       pthread_mutex_lock(consumer_sock->lock);
+       DBG3("Sending %s command to consumer", command_name);
+       ret = consumer_send_msg(consumer_sock, &msg);
+       if (ret < 0) {
+               goto error;
+       }
+
+       DBG3("Sending %s pipe %d to consumer on socket %d",
+                       pipe_name,
+                       pipe, *consumer_sock->fd_ptr);
+       ret = consumer_send_fds(consumer_sock, &pipe, 1);
+       if (ret < 0) {
+               goto error;
+       }
+
+       DBG2("%s pipe successfully sent", pipe_name);
+error:
+       pthread_mutex_unlock(consumer_sock->lock);
+       return ret;
+}
+
+int consumer_send_channel_monitor_pipe(struct consumer_socket *consumer_sock,
+               int pipe)
+{
+       return consumer_send_pipe(consumer_sock,
+                       LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE, pipe);
+}
+
 /*
  * Set consumer subdirectory using the session name and a generated datetime if
  * needed. This is appended to the current subdirectory.
@@ -1038,7 +1162,11 @@ int consumer_set_subdir(struct consumer_output *consumer,
                goto error;
        }
 
-       strncpy(consumer->subdir, tmp_path, sizeof(consumer->subdir));
+       if (lttng_strncpy(consumer->subdir, tmp_path,
+                       sizeof(consumer->subdir))) {
+               ret = -EINVAL;
+               goto error;
+       }
        DBG2("Consumer subdir set to %s", consumer->subdir);
 
 error:
@@ -1046,11 +1174,8 @@ error:
 }
 
 /*
- * Ask the consumer if the data is ready to read (NOT pending) for the specific
- * session id.
- *
- * This function has a different behavior with the consumer i.e. that it waits
- * for a reply from the consumer if yes or no the data is pending.
+ * Ask the consumer if the data is pending for the specific session id.
+ * Returns 1 if data is pending, 0 otherwise, or < 0 on error.
  */
 int consumer_is_data_pending(uint64_t session_id,
                struct consumer_output *consumer)
@@ -1063,12 +1188,12 @@ int consumer_is_data_pending(uint64_t session_id,
 
        assert(consumer);
 
-       msg.cmd_type = LTTNG_CONSUMER_DATA_PENDING;
+       DBG3("Consumer data pending for id %" PRIu64, session_id);
 
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_DATA_PENDING;
        msg.u.data_pending.session_id = session_id;
 
-       DBG3("Consumer data pending for id %" PRIu64, session_id);
-
        /* Send command for each consumer */
        rcu_read_lock();
        cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
@@ -1121,6 +1246,7 @@ int consumer_flush_channel(struct consumer_socket *socket, uint64_t key)
 
        DBG2("Consumer flush channel key %" PRIu64, key);
 
+       memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_FLUSH_CHANNEL;
        msg.u.flush_channel.key = key;
 
@@ -1139,7 +1265,40 @@ end:
 }
 
 /*
- * Send a close metdata command to consumer using the given channel key.
+ * Send a clear quiescent command to consumer using the given channel key.
+ *
+ * Return 0 on success else a negative value.
+ */
+int consumer_clear_quiescent_channel(struct consumer_socket *socket, uint64_t key)
+{
+       int ret;
+       struct lttcomm_consumer_msg msg;
+
+       assert(socket);
+
+       DBG2("Consumer clear quiescent channel key %" PRIu64, key);
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL;
+       msg.u.clear_quiescent_channel.key = key;
+
+       pthread_mutex_lock(socket->lock);
+       health_code_update();
+
+       ret = consumer_send_msg(socket, &msg);
+       if (ret < 0) {
+               goto end;
+       }
+
+end:
+       health_code_update();
+       pthread_mutex_unlock(socket->lock);
+       return ret;
+}
+
+/*
+ * Send a close metadata command to consumer using the given channel key.
+ * Called with registry lock held.
  *
  * Return 0 on success else a negative value.
  */
@@ -1153,6 +1312,7 @@ int consumer_close_metadata(struct consumer_socket *socket,
 
        DBG2("Consumer close metadata channel key %" PRIu64, metadata_key);
 
+       memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_CLOSE_METADATA;
        msg.u.close_metadata.key = metadata_key;
 
@@ -1185,6 +1345,7 @@ int consumer_setup_metadata(struct consumer_socket *socket,
 
        DBG2("Consumer setup metadata channel key %" PRIu64, metadata_key);
 
+       memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_SETUP_METADATA;
        msg.u.setup_metadata.key = metadata_key;
 
@@ -1203,13 +1364,14 @@ end:
 }
 
 /*
- * Send metadata string to consumer. Socket lock MUST be acquired.
+ * Send metadata string to consumer.
+ * RCU read-side lock must be held to guarantee existence of socket.
  *
  * Return 0 on success else a negative value.
  */
 int consumer_push_metadata(struct consumer_socket *socket,
                uint64_t metadata_key, char *metadata_str, size_t len,
-               size_t target_offset)
+               size_t target_offset, uint64_t version)
 {
        int ret;
        struct lttcomm_consumer_msg msg;
@@ -1218,10 +1380,14 @@ int consumer_push_metadata(struct consumer_socket *socket,
 
        DBG2("Consumer push metadata to consumer socket %d", *socket->fd_ptr);
 
+       pthread_mutex_lock(socket->lock);
+
+       memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_PUSH_METADATA;
        msg.u.push_metadata.key = metadata_key;
        msg.u.push_metadata.target_offset = target_offset;
        msg.u.push_metadata.len = len;
+       msg.u.push_metadata.version = version;
 
        health_code_update();
        ret = consumer_send_msg(socket, &msg);
@@ -1244,6 +1410,7 @@ int consumer_push_metadata(struct consumer_socket *socket,
        }
 
 end:
+       pthread_mutex_unlock(socket->lock);
        health_code_update();
        return ret;
 }
@@ -1255,7 +1422,8 @@ end:
  */
 int consumer_snapshot_channel(struct consumer_socket *socket, uint64_t key,
                struct snapshot_output *output, int metadata, uid_t uid, gid_t gid,
-               const char *session_path, int wait, int max_stream_size)
+               const char *session_path, int wait, uint64_t nb_packets_per_stream,
+               uint64_t trace_archive_id)
 {
        int ret;
        struct lttcomm_consumer_msg msg;
@@ -1269,47 +1437,451 @@ int consumer_snapshot_channel(struct consumer_socket *socket, uint64_t key,
        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_SNAPSHOT_CHANNEL;
        msg.u.snapshot_channel.key = key;
-       msg.u.snapshot_channel.max_stream_size = max_stream_size;
+       msg.u.snapshot_channel.nb_packets_per_stream = nb_packets_per_stream;
        msg.u.snapshot_channel.metadata = metadata;
+       msg.u.snapshot_channel.trace_archive_id = trace_archive_id;
 
        if (output->consumer->type == CONSUMER_DST_NET) {
                msg.u.snapshot_channel.relayd_id = output->consumer->net_seq_index;
                msg.u.snapshot_channel.use_relayd = 1;
                ret = snprintf(msg.u.snapshot_channel.pathname,
                                sizeof(msg.u.snapshot_channel.pathname),
-                               "%s/%s-%s-%" PRIu64 "%s", output->consumer->subdir,
-                               output->name, output->datetime, output->nb_snapshot,
+                               "%s/%s/%s-%s-%" PRIu64 "%s",
+                               output->consumer->dst.net.base_dir,
+                               output->consumer->subdir,
+                               output->name, output->datetime,
+                               output->nb_snapshot,
                                session_path);
                if (ret < 0) {
                        ret = -LTTNG_ERR_NOMEM;
                        goto error;
+               } else if (ret >= sizeof(msg.u.snapshot_channel.pathname)) {
+                       ERR("Snapshot path exceeds the maximal allowed length of %zu bytes (%i bytes required) with path \"%s/%s/%s-%s-%" PRIu64 "%s\"",
+                                       sizeof(msg.u.snapshot_channel.pathname),
+                                       ret, output->consumer->dst.net.base_dir,
+                                       output->consumer->subdir,
+                                       output->name, output->datetime,
+                                       output->nb_snapshot,
+                                       session_path);
+                       ret = -LTTNG_ERR_SNAPSHOT_FAIL;
+                       goto error;
                }
        } else {
                ret = snprintf(msg.u.snapshot_channel.pathname,
                                sizeof(msg.u.snapshot_channel.pathname),
-                               "%s/%s-%s-%" PRIu64 "%s", output->consumer->dst.trace_path,
-                               output->name, output->datetime, output->nb_snapshot,
+                               "%s/%s-%s-%" PRIu64 "%s",
+                               output->consumer->dst.session_root_path,
+                               output->name, output->datetime,
+                               output->nb_snapshot,
                                session_path);
                if (ret < 0) {
                        ret = -LTTNG_ERR_NOMEM;
                        goto error;
+               } else if (ret >= sizeof(msg.u.snapshot_channel.pathname)) {
+                       ERR("Snapshot path exceeds the maximal allowed length of %zu bytes (%i bytes required) with path \"%s/%s-%s-%" PRIu64 "%s\"",
+                                       sizeof(msg.u.snapshot_channel.pathname),
+                                       ret, output->consumer->dst.session_root_path,
+                                       output->name, output->datetime, output->nb_snapshot,
+                                       session_path);
+                       ret = -LTTNG_ERR_SNAPSHOT_FAIL;
+                       goto error;
                }
+
                msg.u.snapshot_channel.relayd_id = (uint64_t) -1ULL;
 
                /* Create directory. Ignore if exist. */
                ret = run_as_mkdir_recursive(msg.u.snapshot_channel.pathname,
                                S_IRWXU | S_IRWXG, uid, gid);
                if (ret < 0) {
-                       if (ret != -EEXIST) {
+                       if (errno != EEXIST) {
                                ERR("Trace directory creation error");
                                goto error;
                        }
                }
        }
 
+       health_code_update();
+       pthread_mutex_lock(socket->lock);
+       ret = consumer_send_msg(socket, &msg);
+       pthread_mutex_unlock(socket->lock);
+       if (ret < 0) {
+               goto error;
+       }
+
+error:
+       health_code_update();
+       return ret;
+}
+
+/*
+ * Ask the consumer the number of discarded events for a channel.
+ */
+int consumer_get_discarded_events(uint64_t session_id, uint64_t channel_key,
+               struct consumer_output *consumer, uint64_t *discarded)
+{
+       int ret;
+       struct consumer_socket *socket;
+       struct lttng_ht_iter iter;
+       struct lttcomm_consumer_msg msg;
+
+       assert(consumer);
+
+       DBG3("Consumer discarded events id %" PRIu64, session_id);
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_DISCARDED_EVENTS;
+       msg.u.discarded_events.session_id = session_id;
+       msg.u.discarded_events.channel_key = channel_key;
+
+       *discarded = 0;
+
+       /* Send command for each consumer */
+       rcu_read_lock();
+       cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
+                       node.node) {
+               uint64_t consumer_discarded = 0;
+               pthread_mutex_lock(socket->lock);
+               ret = consumer_socket_send(socket, &msg, sizeof(msg));
+               if (ret < 0) {
+                       pthread_mutex_unlock(socket->lock);
+                       goto end;
+               }
+
+               /*
+                * No need for a recv reply status because the answer to the
+                * command is the reply status message.
+                */
+               ret = consumer_socket_recv(socket, &consumer_discarded,
+                               sizeof(consumer_discarded));
+               if (ret < 0) {
+                       ERR("get discarded events");
+                       pthread_mutex_unlock(socket->lock);
+                       goto end;
+               }
+               pthread_mutex_unlock(socket->lock);
+               *discarded += consumer_discarded;
+       }
+       ret = 0;
+       DBG("Consumer discarded %" PRIu64 " events in session id %" PRIu64,
+                       *discarded, session_id);
+
+end:
+       rcu_read_unlock();
+       return ret;
+}
+
+/*
+ * Ask the consumer the number of lost packets for a channel.
+ */
+int consumer_get_lost_packets(uint64_t session_id, uint64_t channel_key,
+               struct consumer_output *consumer, uint64_t *lost)
+{
+       int ret;
+       struct consumer_socket *socket;
+       struct lttng_ht_iter iter;
+       struct lttcomm_consumer_msg msg;
+
+       assert(consumer);
+
+       DBG3("Consumer lost packets id %" PRIu64, session_id);
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_LOST_PACKETS;
+       msg.u.lost_packets.session_id = session_id;
+       msg.u.lost_packets.channel_key = channel_key;
+
+       *lost = 0;
+
+       /* Send command for each consumer */
+       rcu_read_lock();
+       cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
+                       node.node) {
+               uint64_t consumer_lost = 0;
+               pthread_mutex_lock(socket->lock);
+               ret = consumer_socket_send(socket, &msg, sizeof(msg));
+               if (ret < 0) {
+                       pthread_mutex_unlock(socket->lock);
+                       goto end;
+               }
+
+               /*
+                * No need for a recv reply status because the answer to the
+                * command is the reply status message.
+                */
+               ret = consumer_socket_recv(socket, &consumer_lost,
+                               sizeof(consumer_lost));
+               if (ret < 0) {
+                       ERR("get lost packets");
+                       pthread_mutex_unlock(socket->lock);
+                       goto end;
+               }
+               pthread_mutex_unlock(socket->lock);
+               *lost += consumer_lost;
+       }
+       ret = 0;
+       DBG("Consumer lost %" PRIu64 " packets in session id %" PRIu64,
+                       *lost, session_id);
+
+end:
+       rcu_read_unlock();
+       return ret;
+}
+
+/*
+ * Ask the consumer to rotate a channel.
+ * domain_path contains "/kernel" for kernel or the complete path for UST
+ * (ex: /ust/uid/1000/64-bit);
+ *
+ * The new_chunk_id is the session->rotate_count that has been incremented
+ * when the rotation started. On the relay, this allows to keep track in which
+ * chunk each stream is currently writing to (for the rotate_pending operation).
+ */
+int consumer_rotate_channel(struct consumer_socket *socket, uint64_t key,
+               uid_t uid, gid_t gid, struct consumer_output *output,
+               char *domain_path, bool is_metadata_channel,
+               uint64_t new_chunk_id)
+{
+       int ret;
+       struct lttcomm_consumer_msg msg;
+
+       assert(socket);
+
+       DBG("Consumer rotate channel key %" PRIu64, key);
+
+       pthread_mutex_lock(socket->lock);
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_ROTATE_CHANNEL;
+       msg.u.rotate_channel.key = key;
+       msg.u.rotate_channel.metadata = !!is_metadata_channel;
+       msg.u.rotate_channel.new_chunk_id = new_chunk_id;
+
+       if (output->type == CONSUMER_DST_NET) {
+               msg.u.rotate_channel.relayd_id = output->net_seq_index;
+               ret = snprintf(msg.u.rotate_channel.pathname,
+                               sizeof(msg.u.rotate_channel.pathname), "%s%s%s",
+                               output->dst.net.base_dir,
+                               output->chunk_path, domain_path);
+               if (ret < 0 || ret == sizeof(msg.u.rotate_channel.pathname)) {
+                       ERR("Failed to format channel path name when asking consumer to rotate channel");
+                       ret = -LTTNG_ERR_INVALID;
+                       goto error;
+               }
+       } else {
+               msg.u.rotate_channel.relayd_id = (uint64_t) -1ULL;
+               ret = snprintf(msg.u.rotate_channel.pathname,
+                               sizeof(msg.u.rotate_channel.pathname), "%s%s%s",
+                               output->dst.session_root_path,
+                               output->chunk_path, domain_path);
+               if (ret < 0 || ret == sizeof(msg.u.rotate_channel.pathname)) {
+                       ERR("Failed to format channel path name when asking consumer to rotate channel");
+                       ret = -LTTNG_ERR_INVALID;
+                       goto error;
+               }
+       }
+
+       health_code_update();
+       ret = consumer_send_msg(socket, &msg);
+       if (ret < 0) {
+               switch (-ret) {
+               case LTTCOMM_CONSUMERD_CHAN_NOT_FOUND:
+                       ret = -LTTNG_ERR_CHAN_NOT_FOUND;
+                       break;
+               default:
+                       ret = -LTTNG_ERR_ROTATION_FAIL_CONSUMER;
+                       break;
+               }
+               goto error;
+       }
+error:
+       pthread_mutex_unlock(socket->lock);
+       health_code_update();
+       return ret;
+}
+
+int consumer_rotate_rename(struct consumer_socket *socket, uint64_t session_id,
+               const struct consumer_output *output, const char *old_path,
+               const char *new_path, uid_t uid, gid_t gid)
+{
+       int ret;
+       struct lttcomm_consumer_msg msg;
+       size_t old_path_length, new_path_length;
+
+       assert(socket);
+       assert(old_path);
+       assert(new_path);
+
+       DBG("Consumer rotate rename session %" PRIu64 ", old path = \"%s\", new_path = \"%s\"",
+                       session_id, old_path, new_path);
+
+       old_path_length = strlen(old_path);
+       if (old_path_length >= sizeof(msg.u.rotate_rename.old_path)) {
+               ERR("consumer_rotate_rename: old path length (%zu bytes) exceeds the maximal length allowed by the consumer protocol (%zu bytes)",
+                               old_path_length + 1, sizeof(msg.u.rotate_rename.old_path));
+               ret = -LTTNG_ERR_INVALID;
+               goto error;
+       }
+
+       new_path_length = strlen(new_path);
+       if (new_path_length >= sizeof(msg.u.rotate_rename.new_path)) {
+               ERR("consumer_rotate_rename: new path length (%zu bytes) exceeds the maximal length allowed by the consumer protocol (%zu bytes)",
+                               new_path_length + 1, sizeof(msg.u.rotate_rename.new_path));
+               ret = -LTTNG_ERR_INVALID;
+               goto error;
+       }
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_ROTATE_RENAME;
+       msg.u.rotate_rename.session_id = session_id;
+       msg.u.rotate_rename.uid = uid;
+       msg.u.rotate_rename.gid = gid;
+       strcpy(msg.u.rotate_rename.old_path, old_path);
+       strcpy(msg.u.rotate_rename.new_path, new_path);
+
+       if (output->type == CONSUMER_DST_NET) {
+               msg.u.rotate_rename.relayd_id = output->net_seq_index;
+       } else {
+               msg.u.rotate_rename.relayd_id = -1ULL;
+       }
+
+       health_code_update();
+       ret = consumer_send_msg(socket, &msg);
+       if (ret < 0) {
+               ret = -LTTNG_ERR_ROTATE_RENAME_FAIL_CONSUMER;
+               goto error;
+       }
+
+error:
+       health_code_update();
+       return ret;
+}
+
+/*
+ * Ask the consumer if a rotation is locally pending. Must be called with the
+ * socket lock held.
+ *
+ * Return 1 if the rotation is still pending, 0 if finished, a negative value
+ * on error.
+ */
+int consumer_check_rotation_pending_local(struct consumer_socket *socket,
+               uint64_t session_id, uint64_t chunk_id)
+{
+       int ret;
+       struct lttcomm_consumer_msg msg;
+       uint32_t pending = 0;
+
+       assert(socket);
+
+       DBG("Asking consumer to locally check for pending rotation for session %" PRIu64 ", chunk id %" PRIu64,
+                       session_id, chunk_id);
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_CHECK_ROTATION_PENDING_LOCAL;
+       msg.u.check_rotation_pending_local.session_id = session_id;
+       msg.u.check_rotation_pending_local.chunk_id = chunk_id;
+
+       health_code_update();
+       ret = consumer_send_msg(socket, &msg);
+       if (ret < 0) {
+               ret = -LTTNG_ERR_ROTATION_PENDING_LOCAL_FAIL_CONSUMER;
+               goto error;
+       }
+
+       ret = consumer_socket_recv(socket, &pending, sizeof(pending));
+       if (ret < 0) {
+               goto error;
+       }
+
+       ret = pending;
+
+error:
+       health_code_update();
+       return ret;
+}
+
+/*
+ * Ask the consumer if a rotation is pending on the relayd. Must be called with
+ * the socket lock held.
+ *
+ * Return 1 if the rotation is still pending, 0 if finished, a negative value
+ * on error.
+ */
+int consumer_check_rotation_pending_relay(struct consumer_socket *socket,
+               const struct consumer_output *output, uint64_t session_id,
+               uint64_t chunk_id)
+{
+       int ret;
+       struct lttcomm_consumer_msg msg;
+       uint32_t pending = 0;
+
+       assert(socket);
+
+       DBG("Asking consumer to check for pending rotation on relay for session %" PRIu64 ", chunk id %" PRIu64,
+                       session_id, chunk_id);
+       assert(output->type == CONSUMER_DST_NET);
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_CHECK_ROTATION_PENDING_RELAY;
+       msg.u.check_rotation_pending_relay.session_id = session_id;
+       msg.u.check_rotation_pending_relay.relayd_id = output->net_seq_index;
+       msg.u.check_rotation_pending_relay.chunk_id = chunk_id;
+
+       health_code_update();
+       ret = consumer_send_msg(socket, &msg);
+       if (ret < 0) {
+               ret = -LTTNG_ERR_ROTATION_PENDING_RELAY_FAIL_CONSUMER;
+               goto error;
+       }
+
+       ret = consumer_socket_recv(socket, &pending, sizeof(pending));
+       if (ret < 0) {
+               goto error;
+       }
+
+       ret = pending;
+
+error:
+       health_code_update();
+       return ret;
+}
+
+/*
+ * Ask the consumer to create a directory.
+ *
+ * Called with the consumer socket lock held.
+ */
+int consumer_mkdir(struct consumer_socket *socket, uint64_t session_id,
+               const struct consumer_output *output, const char *path,
+               uid_t uid, gid_t gid)
+{
+       int ret;
+       struct lttcomm_consumer_msg msg;
+
+       assert(socket);
+
+       DBG("Consumer mkdir %s in session %" PRIu64, path, session_id);
+
+       memset(&msg, 0, sizeof(msg));
+       msg.cmd_type = LTTNG_CONSUMER_MKDIR;
+       msg.u.mkdir.session_id = session_id;
+       msg.u.mkdir.uid = uid;
+       msg.u.mkdir.gid = gid;
+       ret = snprintf(msg.u.mkdir.path, sizeof(msg.u.mkdir.path), "%s", path);
+       if (ret < 0 || ret >= sizeof(msg.u.mkdir.path)) {
+               ERR("Format path");
+               ret = -LTTNG_ERR_INVALID;
+               goto error;
+       }
+
+       if (output->type == CONSUMER_DST_NET) {
+               msg.u.mkdir.relayd_id = output->net_seq_index;
+       } else {
+               msg.u.mkdir.relayd_id = -1ULL;
+       }
+
        health_code_update();
        ret = consumer_send_msg(socket, &msg);
        if (ret < 0) {
+               ret = -LTTNG_ERR_MKDIR_FAIL_CONSUMER;
                goto error;
        }
 
This page took 0.036182 seconds and 4 git commands to generate.