X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fcmd.c;h=be14b6bd333cce161ebe6b3c0fad8a807cd1d18c;hb=c1b2f2e24cda9b60bce9380a3967b9207293a32d;hp=64fa5c0bb5c1e870cd11d1f7e6cd3a69f2b7c6da;hpb=31ea48462f529be003231ca0e14c5b882cd7d66e;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/cmd.c b/src/bin/lttng-sessiond/cmd.c index 64fa5c0bb..be14b6bd3 100644 --- a/src/bin/lttng-sessiond/cmd.c +++ b/src/bin/lttng-sessiond/cmd.c @@ -17,6 +17,7 @@ #define _GNU_SOURCE #include +#include #include #include #include @@ -30,7 +31,7 @@ #include "channel.h" #include "consumer.h" #include "event.h" -#include "health.h" +#include "health-sessiond.h" #include "kernel.h" #include "kernel-consumer.h" #include "lttng-sessiond.h" @@ -182,6 +183,55 @@ static void list_lttng_channels(int domain, struct ltt_session *session, } } +/* + * Create a list of JUL domain events. + * + * Return number of events in list on success or else a negative value. + */ +static int list_lttng_jul_events(struct jul_domain *dom, + struct lttng_event **events) +{ + int i = 0, ret = 0; + unsigned int nb_event = 0; + struct jul_event *event; + struct lttng_event *tmp_events; + struct lttng_ht_iter iter; + + assert(dom); + assert(events); + + DBG3("Listing JUL events"); + + nb_event = lttng_ht_get_count(dom->events); + if (nb_event == 0) { + ret = nb_event; + goto error; + } + + tmp_events = zmalloc(nb_event * sizeof(*tmp_events)); + if (!tmp_events) { + PERROR("zmalloc JUL events session"); + ret = -LTTNG_ERR_FATAL; + goto error; + } + + rcu_read_lock(); + cds_lfht_for_each_entry(dom->events->ht, &iter.iter, event, node.node) { + strncpy(tmp_events[i].name, event->name, sizeof(tmp_events[i].name)); + tmp_events[i].name[sizeof(tmp_events[i].name) - 1] = '\0'; + tmp_events[i].enabled = event->enabled; + i++; + } + rcu_read_unlock(); + + *events = tmp_events; + ret = nb_event; + +error: + assert(nb_event == i); + return ret; +} + /* * Create a list of ust global domain events. */ @@ -256,6 +306,9 @@ static int list_lttng_ust_global_events(char *channel_name, if (uevent->filter) { tmp[i].filter = 1; } + if (uevent->exclusion) { + tmp[i].exclusion = 1; + } i++; } @@ -406,6 +459,7 @@ static int add_uri_to_consumer(struct consumer_output *consumer, * URI was the same in the consumer so we do not append the subdir * again so to not duplicate output dir. */ + ret = LTTNG_OK; goto error; } @@ -462,9 +516,6 @@ static int init_kernel_tracing(struct ltt_kernel_session *session) if (session->consumer_fds_sent == 0 && session->consumer != NULL) { cds_lfht_for_each_entry(session->consumer->socks->ht, &iter.iter, socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - pthread_mutex_lock(socket->lock); ret = kernel_consumer_send_session(socket, session); pthread_mutex_unlock(socket->lock); @@ -550,7 +601,8 @@ error: */ static int send_consumer_relayd_socket(int domain, unsigned int session_id, struct lttng_uri *relayd_uri, struct consumer_output *consumer, - struct consumer_socket *consumer_sock) + struct consumer_socket *consumer_sock, + char *session_name, char *hostname, int session_live_timer) { int ret; struct lttcomm_relayd_sock *rsock = NULL; @@ -576,7 +628,8 @@ static int send_consumer_relayd_socket(int domain, unsigned int session_id, /* Send relayd socket to consumer. */ ret = consumer_send_relayd_socket(consumer_sock, rsock, consumer, - relayd_uri->stype, session_id); + relayd_uri->stype, session_id, + session_name, hostname, session_live_timer); if (ret < 0) { ret = LTTNG_ERR_ENABLE_CONSUMER_FAIL; goto close_sock; @@ -618,7 +671,8 @@ error: * session. */ static int send_consumer_relayd_sockets(int domain, unsigned int session_id, - struct consumer_output *consumer, struct consumer_socket *sock) + struct consumer_output *consumer, struct consumer_socket *sock, + char *session_name, char *hostname, int session_live_timer) { int ret = LTTNG_OK; @@ -628,7 +682,8 @@ static int send_consumer_relayd_sockets(int domain, unsigned int session_id, /* Sending control relayd socket. */ if (!sock->control_sock_sent) { ret = send_consumer_relayd_socket(domain, session_id, - &consumer->dst.net.control, consumer, sock); + &consumer->dst.net.control, consumer, sock, + session_name, hostname, session_live_timer); if (ret != LTTNG_OK) { goto error; } @@ -637,7 +692,8 @@ static int send_consumer_relayd_sockets(int domain, unsigned int session_id, /* Sending data relayd socket. */ if (!sock->data_sock_sent) { ret = send_consumer_relayd_socket(domain, session_id, - &consumer->dst.net.data, consumer, sock); + &consumer->dst.net.data, consumer, sock, + session_name, hostname, session_live_timer); if (ret != LTTNG_OK) { goto error; } @@ -674,12 +730,11 @@ int cmd_setup_relayd(struct ltt_session *session) /* For each consumer socket, send relayd sockets */ cds_lfht_for_each_entry(usess->consumer->socks->ht, &iter.iter, socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - pthread_mutex_lock(socket->lock); ret = send_consumer_relayd_sockets(LTTNG_DOMAIN_UST, session->id, - usess->consumer, socket); + usess->consumer, socket, + session->name, session->hostname, + session->live_timer); pthread_mutex_unlock(socket->lock); if (ret != LTTNG_OK) { goto error; @@ -693,12 +748,11 @@ int cmd_setup_relayd(struct ltt_session *session) && ksess->consumer->enabled) { cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - pthread_mutex_lock(socket->lock); ret = send_consumer_relayd_sockets(LTTNG_DOMAIN_KERNEL, session->id, - ksess->consumer, socket); + ksess->consumer, socket, + session->name, session->hostname, + session->live_timer); pthread_mutex_unlock(socket->lock); if (ret != LTTNG_OK) { goto error; @@ -850,15 +904,54 @@ int cmd_enable_channel(struct ltt_session *session, int ret; struct ltt_ust_session *usess = session->ust_session; struct lttng_ht *chan_ht; + size_t len; assert(session); assert(attr); assert(domain); + len = strnlen(attr->name, sizeof(attr->name)); + + /* Validate channel name */ + if (attr->name[0] == '.' || + memchr(attr->name, '/', len) != NULL) { + ret = LTTNG_ERR_INVALID_CHANNEL_NAME; + goto end; + } + DBG("Enabling channel %s for session %s", attr->name, session->name); rcu_read_lock(); + /* + * Don't try to enable a channel if the session has been started at + * some point in time before. The tracer does not allow it. + */ + if (session->has_been_started) { + ret = LTTNG_ERR_TRACE_ALREADY_STARTED; + goto error; + } + + /* + * If the session is a live session, remove the switch timer, the + * live timer does the same thing but sends also synchronisation + * beacons for inactive streams. + */ + if (session->live_timer > 0) { + attr->attr.live_timer_interval = session->live_timer; + attr->attr.switch_timer_interval = 0; + } + + /* + * The ringbuffer (both in user space and kernel) behave badly in overwrite + * mode and with less than 2 subbuf so block it right away and send back an + * invalid attribute error. + */ + if (attr->attr.overwrite && attr->attr.num_subbuf < 2) { + ret = LTTNG_ERR_INVALID; + goto error; + } + switch (domain->type) { case LTTNG_DOMAIN_KERNEL: { @@ -867,15 +960,6 @@ int cmd_enable_channel(struct ltt_session *session, kchan = trace_kernel_get_channel_by_name(attr->name, session->kernel_session); if (kchan == NULL) { - /* - * Don't try to create a channel if the session - * has been started at some point in time - * before. The tracer does not allow it. - */ - if (session->started) { - ret = LTTNG_ERR_TRACE_ALREADY_STARTED; - goto error; - } ret = channel_kernel_create(session->kernel_session, attr, wpipe); if (attr->name[0] != '\0') { session->kernel_session->has_non_default_channel = 1; @@ -899,15 +983,6 @@ int cmd_enable_channel(struct ltt_session *session, uchan = trace_ust_find_channel_by_name(chan_ht, attr->name); if (uchan == NULL) { - /* - * Don't try to create a channel if the session - * has been started at some point in time - * before. The tracer does not allow it. - */ - if (session->started) { - ret = LTTNG_ERR_TRACE_ALREADY_STARTED; - goto error; - } ret = channel_ust_create(usess, attr, domain->buf_type); if (attr->name[0] != '\0') { usess->has_non_default_channel = 1; @@ -924,6 +999,7 @@ int cmd_enable_channel(struct ltt_session *session, error: rcu_read_unlock(); +end: return ret; } @@ -1003,6 +1079,19 @@ int cmd_disable_event(struct ltt_session *session, int domain, channel_name); break; } + case LTTNG_DOMAIN_JUL: + { + struct ltt_ust_session *usess = session->ust_session; + + assert(usess); + + ret = event_jul_disable(usess, event_name); + if (ret != LTTNG_OK) { + goto error; + } + + break; + } #if 0 case LTTNG_DOMAIN_UST_EXEC_NAME: case LTTNG_DOMAIN_UST_PID: @@ -1093,6 +1182,19 @@ int cmd_disable_event_all(struct ltt_session *session, int domain, DBG3("Disable all UST events in channel %s completed", channel_name); + break; + } + case LTTNG_DOMAIN_JUL: + { + struct ltt_ust_session *usess = session->ust_session; + + assert(usess); + + ret = event_jul_disable_all(usess); + if (ret != LTTNG_OK) { + goto error; + } + break; } #if 0 @@ -1212,7 +1314,9 @@ error: */ int cmd_enable_event(struct ltt_session *session, struct lttng_domain *domain, char *channel_name, struct lttng_event *event, - struct lttng_filter_bytecode *filter, int wpipe) + struct lttng_filter_bytecode *filter, + struct lttng_event_exclusion *exclusion, + int wpipe) { int ret, channel_created = 0; struct lttng_channel *attr; @@ -1325,10 +1429,57 @@ int cmd_enable_event(struct ltt_session *session, struct lttng_domain *domain, } /* At this point, the session and channel exist on the tracer */ - ret = event_ust_enable_tracepoint(usess, uchan, event, filter); + ret = event_ust_enable_tracepoint(usess, uchan, event, filter, exclusion); + if (ret != LTTNG_OK) { + goto error; + } + break; + } + case LTTNG_DOMAIN_JUL: + { + struct lttng_event uevent; + struct lttng_domain tmp_dom; + struct ltt_ust_session *usess = session->ust_session; + + assert(usess); + + /* Create the default JUL tracepoint. */ + memset(&uevent, 0, sizeof(uevent)); + uevent.type = LTTNG_EVENT_TRACEPOINT; + uevent.loglevel_type = LTTNG_EVENT_LOGLEVEL_ALL; + if (is_root) { + strncpy(uevent.name, DEFAULT_SYS_JUL_EVENT_NAME, + sizeof(uevent.name)); + } else { + strncpy(uevent.name, DEFAULT_USER_JUL_EVENT_NAME, + sizeof(uevent.name)); + } + uevent.name[sizeof(uevent.name) - 1] = '\0'; + + /* + * The domain type is changed because we are about to enable the + * default channel and event for the JUL domain that are hardcoded. + * This happens in the UST domain. + */ + memcpy(&tmp_dom, domain, sizeof(tmp_dom)); + tmp_dom.type = LTTNG_DOMAIN_UST; + + ret = cmd_enable_event(session, &tmp_dom, DEFAULT_JUL_CHANNEL_NAME, + &uevent, filter, NULL, wpipe); + if (ret != LTTNG_OK && ret != LTTNG_ERR_UST_EVENT_ENABLED) { + goto error; + } + + /* The wild card * means that everything should be enabled. */ + if (strncmp(event->name, "*", 1) == 0 && strlen(event->name) == 1) { + ret = event_jul_enable_all(usess, event, filter); + } else { + ret = event_jul_enable(usess, event, filter); + } if (ret != LTTNG_OK) { goto error; } + break; } #if 0 @@ -1501,6 +1652,52 @@ int cmd_enable_event_all(struct ltt_session *session, goto error; } + break; + } + case LTTNG_DOMAIN_JUL: + { + struct lttng_event uevent, event; + struct lttng_domain tmp_dom; + struct ltt_ust_session *usess = session->ust_session; + + assert(usess); + + /* Create the default JUL tracepoint. */ + uevent.type = LTTNG_EVENT_TRACEPOINT; + uevent.loglevel_type = LTTNG_EVENT_LOGLEVEL_ALL; + if (is_root) { + strncpy(uevent.name, DEFAULT_SYS_JUL_EVENT_NAME, + sizeof(uevent.name)); + } else { + strncpy(uevent.name, DEFAULT_USER_JUL_EVENT_NAME, + sizeof(uevent.name)); + } + uevent.name[sizeof(uevent.name) - 1] = '\0'; + + /* + * The domain type is changed because we are about to enable the + * default channel and event for the JUL domain that are hardcoded. + * This happens in the UST domain. + */ + memcpy(&tmp_dom, domain, sizeof(tmp_dom)); + tmp_dom.type = LTTNG_DOMAIN_UST; + + ret = cmd_enable_event(session, &tmp_dom, DEFAULT_JUL_CHANNEL_NAME, + &uevent, NULL, NULL, wpipe); + if (ret != LTTNG_OK && ret != LTTNG_ERR_UST_EVENT_ENABLED) { + goto error; + } + + event.loglevel = LTTNG_LOGLEVEL_JUL_ALL; + event.loglevel_type = LTTNG_EVENT_LOGLEVEL_ALL; + strncpy(event.name, "*", sizeof(event.name)); + event.name[sizeof(event.name) - 1] = '\0'; + + ret = event_jul_enable_all(usess, &event, filter); + if (ret != LTTNG_OK) { + goto error; + } + break; } #if 0 @@ -1544,6 +1741,13 @@ ssize_t cmd_list_tracepoints(int domain, struct lttng_event **events) goto error; } break; + case LTTNG_DOMAIN_JUL: + nb_events = jul_list_events(events); + if (nb_events < 0) { + ret = LTTNG_ERR_UST_LIST_FAIL; + goto error; + } + break; default: ret = LTTNG_ERR_UND; goto error; @@ -1592,6 +1796,7 @@ error: int cmd_start_trace(struct ltt_session *session) { int ret; + unsigned long nb_chan = 0; struct ltt_kernel_session *ksession; struct ltt_ust_session *usess; @@ -1601,13 +1806,28 @@ int cmd_start_trace(struct ltt_session *session) ksession = session->kernel_session; usess = session->ust_session; - if (session->enabled) { - /* Already started. */ + /* Is the session already started? */ + if (session->active) { ret = LTTNG_ERR_TRACE_ALREADY_STARTED; goto error; } - session->enabled = 1; + /* + * Starting a session without channel is useless since after that it's not + * possible to enable channel thus inform the client. + */ + if (usess && usess->domain_global.channels) { + rcu_read_lock(); + nb_chan += lttng_ht_get_count(usess->domain_global.channels); + rcu_read_unlock(); + } + if (ksession) { + nb_chan += ksession->channel_count; + } + if (!nb_chan) { + ret = LTTNG_ERR_NO_CHANNEL; + goto error; + } /* Kernel tracing */ if (ksession != NULL) { @@ -1628,7 +1848,9 @@ int cmd_start_trace(struct ltt_session *session) } } - session->started = 1; + /* Flag this after a successful start. */ + session->has_been_started = 1; + session->active = 1; ret = LTTNG_OK; @@ -1652,13 +1874,12 @@ int cmd_stop_trace(struct ltt_session *session) ksession = session->kernel_session; usess = session->ust_session; - if (!session->enabled) { + /* Session is not active. Skip everythong and inform the client. */ + if (!session->active) { ret = LTTNG_ERR_TRACE_ALREADY_STOPPED; goto error; } - session->enabled = 0; - /* Kernel tracer */ if (ksession && ksession->started) { DBG("Stop kernel tracing"); @@ -1700,6 +1921,8 @@ int cmd_stop_trace(struct ltt_session *session) } } + /* Flag inactive after a successful stop. */ + session->active = 0; ret = LTTNG_OK; error: @@ -1721,8 +1944,8 @@ int cmd_set_consumer_uri(int domain, struct ltt_session *session, assert(uris); assert(nb_uri > 0); - /* Can't enable consumer after session started. */ - if (session->enabled) { + /* Can't set consumer URI if the session is active. */ + if (session->active) { ret = LTTNG_ERR_TRACE_ALREADY_STARTED; goto error; } @@ -1762,6 +1985,17 @@ int cmd_set_consumer_uri(int domain, struct ltt_session *session, } } + /* + * Make sure to set the session in output mode after we set URI since a + * session can be created without URL (thus flagged in no output mode). + */ + session->output_traces = 1; + if (ksess) { + ksess->output_traces = 1; + } else if (usess) { + usess->output_traces = 1; + } + /* All good! */ ret = LTTNG_OK; @@ -1773,7 +2007,7 @@ error: * Command LTTNG_CREATE_SESSION processed by the client thread. */ int cmd_create_session_uri(char *name, struct lttng_uri *uris, - size_t nb_uri, lttng_sock_cred *creds) + size_t nb_uri, lttng_sock_cred *creds, unsigned int live_timer) { int ret; struct ltt_session *session; @@ -1811,6 +2045,7 @@ int cmd_create_session_uri(char *name, struct lttng_uri *uris, session = session_find_by_name(name); assert(session); + session->live_timer = live_timer; /* Create default consumer output for the session not yet created. */ session->consumer = consumer_create_output(CONSUMER_DST_LOCAL); if (session->consumer == NULL) { @@ -1857,7 +2092,7 @@ int cmd_create_session_snapshot(char *name, struct lttng_uri *uris, * Create session in no output mode with URIs set to NULL. The uris we've * received are for a default snapshot output if one. */ - ret = cmd_create_session_uri(name, NULL, 0, creds); + ret = cmd_create_session_uri(name, NULL, 0, creds, -1); if (ret != LTTNG_OK) { goto error; } @@ -1965,7 +2200,14 @@ int cmd_calibrate(int domain, struct lttng_calibrate *calibrate) { struct lttng_kernel_calibrate kcalibrate; - kcalibrate.type = calibrate->type; + switch (calibrate->type) { + case LTTNG_CALIBRATE_FUNCTION: + default: + /* Default and only possible calibrate option. */ + kcalibrate.type = LTTNG_KERNEL_CALIBRATE_KRETPROBE; + break; + } + ret = kernel_calibrate(kernel_tracer_fd, &kcalibrate); if (ret < 0) { ret = LTTNG_ERR_KERN_ENABLE_FAIL; @@ -1977,7 +2219,14 @@ int cmd_calibrate(int domain, struct lttng_calibrate *calibrate) { struct lttng_ust_calibrate ucalibrate; - ucalibrate.type = calibrate->type; + switch (calibrate->type) { + case LTTNG_CALIBRATE_FUNCTION: + default: + /* Default and only possible calibrate option. */ + ucalibrate.type = LTTNG_UST_CALIBRATE_TRACEPOINT; + break; + } + ret = ust_app_calibrate_glb(&ucalibrate); if (ret < 0) { ret = LTTNG_ERR_UST_CALIBRATE_FAIL; @@ -2027,13 +2276,15 @@ int cmd_register_consumer(struct ltt_session *session, int domain, ret = LTTNG_ERR_CONNECT_FAIL; goto error; } + cdata->cmd_sock = sock; - socket = consumer_allocate_socket(sock); + socket = consumer_allocate_socket(&cdata->cmd_sock); if (socket == NULL) { ret = close(sock); if (ret < 0) { PERROR("close register consumer"); } + cdata->cmd_sock = -1; ret = LTTNG_ERR_FATAL; goto error; } @@ -2089,6 +2340,10 @@ ssize_t cmd_list_domains(struct ltt_session *session, if (session->ust_session != NULL) { DBG3("Listing domains found UST global domain"); nb_dom++; + + if (session->ust_session->domain_jul.being_used) { + nb_dom++; + } } *domains = zmalloc(nb_dom * sizeof(struct lttng_domain)); @@ -2106,6 +2361,12 @@ ssize_t cmd_list_domains(struct ltt_session *session, (*domains)[index].type = LTTNG_DOMAIN_UST; (*domains)[index].buf_type = session->ust_session->buffer_type; index++; + + if (session->ust_session->domain_jul.being_used) { + (*domains)[index].type = LTTNG_DOMAIN_JUL; + (*domains)[index].buf_type = session->ust_session->buffer_type; + index++; + } } return nb_dom; @@ -2137,16 +2398,18 @@ ssize_t cmd_list_channels(int domain, struct ltt_session *session, break; case LTTNG_DOMAIN_UST: if (session->ust_session != NULL) { + rcu_read_lock(); nb_chan = lttng_ht_get_count( - session->ust_session->domain_global.channels); + session->ust_session->domain_global.channels); + rcu_read_unlock(); } DBG3("Number of UST global channels %zd", nb_chan); - if (nb_chan <= 0) { + if (nb_chan < 0) { ret = LTTNG_ERR_UST_CHAN_NOT_FOUND; + goto error; } break; default: - *channels = NULL; ret = LTTNG_ERR_UND; goto error; } @@ -2159,10 +2422,6 @@ ssize_t cmd_list_channels(int domain, struct ltt_session *session, } list_lttng_channels(domain, session, *channels); - } else { - *channels = NULL; - /* Ret value was set in the domain switch case */ - goto error; } return nb_chan; @@ -2196,6 +2455,12 @@ ssize_t cmd_list_events(int domain, struct ltt_session *session, } break; } + case LTTNG_DOMAIN_JUL: + if (session->ust_session) { + nb_event = list_lttng_jul_events( + &session->ust_session->domain_jul, events); + } + break; default: ret = LTTNG_ERR_UND; goto error; @@ -2256,8 +2521,9 @@ void cmd_list_lttng_sessions(struct lttng_session *sessions, uid_t uid, strncpy(sessions[i].name, session->name, NAME_MAX); sessions[i].name[NAME_MAX - 1] = '\0'; - sessions[i].enabled = session->enabled; + sessions[i].enabled = session->active; sessions[i].snapshot_mode = session->snapshot_mode; + sessions[i].live_timer_interval = session->live_timer; i++; } } @@ -2275,9 +2541,24 @@ int cmd_data_pending(struct ltt_session *session) assert(session); /* Session MUST be stopped to ask for data availability. */ - if (session->enabled) { + if (session->active) { ret = LTTNG_ERR_SESSION_STARTED; goto error; + } else { + /* + * If stopped, just make sure we've started before else the above call + * will always send that there is data pending. + * + * The consumer assumes that when the data pending command is received, + * the trace has been started before or else no output data is written + * by the streams which is a condition for data pending. So, this is + * *VERY* important that we don't ask the consumer before a start + * trace. + */ + if (!session->has_been_started) { + ret = 0; + goto error; + } } if (ksess && ksess->consumer) { @@ -2426,7 +2707,7 @@ ssize_t cmd_snapshot_list_outputs(struct ltt_session *session, struct lttng_snapshot_output **outputs) { int ret, idx = 0; - struct lttng_snapshot_output *list; + struct lttng_snapshot_output *list = NULL; struct lttng_ht_iter iter; struct snapshot_output *output; @@ -2440,7 +2721,7 @@ ssize_t cmd_snapshot_list_outputs(struct ltt_session *session, * set in no output mode. */ if (session->output_traces) { - ret = LTTNG_ERR_EPERM; + ret = -LTTNG_ERR_EPERM; goto error; } @@ -2451,11 +2732,12 @@ ssize_t cmd_snapshot_list_outputs(struct ltt_session *session, list = zmalloc(session->snapshot.nb_output * sizeof(*list)); if (!list) { - ret = LTTNG_ERR_NOMEM; + ret = -LTTNG_ERR_NOMEM; goto error; } /* Copy list from session to the new list object. */ + rcu_read_lock(); cds_lfht_for_each_entry(session->snapshot.output_ht->ht, &iter.iter, output, node.node) { assert(output->consumer); @@ -2470,28 +2752,28 @@ ssize_t cmd_snapshot_list_outputs(struct ltt_session *session, ret = uri_to_str_url(&output->consumer->dst.net.control, list[idx].ctrl_url, sizeof(list[idx].ctrl_url)); if (ret < 0) { - ret = LTTNG_ERR_NOMEM; - goto free_error; + ret = -LTTNG_ERR_NOMEM; + goto error; } /* Data URI. */ ret = uri_to_str_url(&output->consumer->dst.net.data, list[idx].data_url, sizeof(list[idx].data_url)); if (ret < 0) { - ret = LTTNG_ERR_NOMEM; - goto free_error; + ret = -LTTNG_ERR_NOMEM; + goto error; } } idx++; } *outputs = list; - return session->snapshot.nb_output; - -free_error: - free(list); + list = NULL; + ret = session->snapshot.nb_output; error: - return -ret; + free(list); + rcu_read_unlock(); + return ret; } /* @@ -2526,7 +2808,9 @@ static int set_relayd_for_snapshot(struct consumer_output *consumer, cds_lfht_for_each_entry(snap_output->consumer->socks->ht, &iter.iter, socket, node.node) { ret = send_consumer_relayd_sockets(0, session->id, - snap_output->consumer, socket); + snap_output->consumer, socket, + session->name, session->hostname, + session->live_timer); if (ret != LTTNG_OK) { rcu_read_unlock(); goto error; @@ -2541,11 +2825,11 @@ error: /* * Record a kernel snapshot. * - * Return 0 on success or a LTTNG_ERR code. + * Return LTTNG_OK on success or a LTTNG_ERR code. */ static int record_kernel_snapshot(struct ltt_kernel_session *ksess, struct snapshot_output *output, struct ltt_session *session, - int wait, int nb_streams) + int wait, uint64_t nb_packets_per_stream) { int ret; @@ -2576,23 +2860,19 @@ static int record_kernel_snapshot(struct ltt_kernel_session *ksess, goto error_snapshot; } - ret = kernel_snapshot_record(ksess, output, wait, nb_streams); - if (ret < 0) { - if (ret == -EINVAL) { - ret = LTTNG_ERR_INVALID; - goto error_snapshot; - } - - ret = LTTNG_ERR_SNAPSHOT_FAIL; + ret = kernel_snapshot_record(ksess, output, wait, nb_packets_per_stream); + if (ret != LTTNG_OK) { goto error_snapshot; } ret = LTTNG_OK; + goto end; error_snapshot: /* Clean up copied sockets so this output can use some other later on. */ consumer_destroy_output_sockets(output->consumer); error: +end: return ret; } @@ -2603,7 +2883,7 @@ error: */ static int record_ust_snapshot(struct ltt_ust_session *usess, struct snapshot_output *output, struct ltt_session *session, - int wait, int nb_streams) + int wait, uint64_t nb_packets_per_stream) { int ret; @@ -2634,14 +2914,19 @@ static int record_ust_snapshot(struct ltt_ust_session *usess, goto error_snapshot; } - ret = ust_app_snapshot_record(usess, output, wait, nb_streams); + ret = ust_app_snapshot_record(usess, output, wait, nb_packets_per_stream); if (ret < 0) { - if (ret == -EINVAL) { + switch (-ret) { + case EINVAL: ret = LTTNG_ERR_INVALID; - goto error_snapshot; + break; + case ENODATA: + ret = LTTNG_ERR_SNAPSHOT_NODATA; + break; + default: + ret = LTTNG_ERR_SNAPSHOT_FAIL; + break; } - - ret = LTTNG_ERR_SNAPSHOT_FAIL; goto error_snapshot; } @@ -2654,27 +2939,90 @@ error: return ret; } -/* - * Returns the total number of streams for a session or a negative value - * on error. - */ -static unsigned int get_total_nb_stream(struct ltt_session *session) +static +uint64_t get_session_size_one_more_packet_per_stream(struct ltt_session *session, + uint64_t cur_nr_packets) { - unsigned int total_streams = 0; + uint64_t tot_size = 0; if (session->kernel_session) { + struct ltt_kernel_channel *chan; struct ltt_kernel_session *ksess = session->kernel_session; - total_streams += ksess->stream_count_global; + cds_list_for_each_entry(chan, &ksess->channel_list.head, list) { + if (cur_nr_packets >= chan->channel->attr.num_subbuf) { + /* + * Don't take channel into account if we + * already grab all its packets. + */ + continue; + } + tot_size += chan->channel->attr.subbuf_size + * chan->stream_count; + } } if (session->ust_session) { struct ltt_ust_session *usess = session->ust_session; - total_streams += ust_app_get_nb_stream(usess); + tot_size += ust_app_get_size_one_more_packet_per_stream(usess, + cur_nr_packets); } - return total_streams; + return tot_size; +} + +/* + * Calculate the number of packets we can grab from each stream that + * fits within the overall snapshot max size. + * + * Returns -1 on error, 0 means infinite number of packets, else > 0 is + * the number of packets per stream. + * + * TODO: this approach is not perfect: we consider the worse case + * (packet filling the sub-buffers) as an upper bound, but we could do + * better if we do this calculation while we actually grab the packet + * content: we would know how much padding we don't actually store into + * the file. + * + * This algorithm is currently bounded by the number of packets per + * stream. + * + * Since we call this algorithm before actually grabbing the data, it's + * an approximation: for instance, applications could appear/disappear + * in between this call and actually grabbing data. + */ +static +int64_t get_session_nb_packets_per_stream(struct ltt_session *session, uint64_t max_size) +{ + int64_t size_left; + uint64_t cur_nb_packets = 0; + + if (!max_size) { + return 0; /* Infinite */ + } + + size_left = max_size; + for (;;) { + uint64_t one_more_packet_tot_size; + + one_more_packet_tot_size = get_session_size_one_more_packet_per_stream(session, + cur_nb_packets); + if (!one_more_packet_tot_size) { + /* We are already grabbing all packets. */ + break; + } + size_left -= one_more_packet_tot_size; + if (size_left < 0) { + break; + } + cur_nb_packets++; + } + if (!cur_nb_packets) { + /* Not enough room to grab one packet of each stream, error. */ + return -1; + } + return cur_nb_packets; } /* @@ -2691,7 +3039,7 @@ int cmd_snapshot_record(struct ltt_session *session, int ret = LTTNG_OK; unsigned int use_tmp_output = 0; struct snapshot_output tmp_output; - unsigned int nb_streams, snapshot_success = 0; + unsigned int snapshot_success = 0; assert(session); @@ -2707,7 +3055,7 @@ int cmd_snapshot_record(struct ltt_session *session, } /* The session needs to be started at least once. */ - if (!session->started) { + if (!session->has_been_started) { ret = LTTNG_ERR_START_SESSION_ONCE; goto error; } @@ -2730,18 +3078,20 @@ int cmd_snapshot_record(struct ltt_session *session, use_tmp_output = 1; } - /* - * Get the total number of stream of that session which is used by the - * maximum size of the snapshot feature. - */ - nb_streams = get_total_nb_stream(session); - if (session->kernel_session) { struct ltt_kernel_session *ksess = session->kernel_session; if (use_tmp_output) { + int64_t nb_packets_per_stream; + + nb_packets_per_stream = get_session_nb_packets_per_stream(session, + tmp_output.max_size); + if (nb_packets_per_stream < 0) { + ret = LTTNG_ERR_INVALID; + goto error; + } ret = record_kernel_snapshot(ksess, &tmp_output, session, - wait, nb_streams); + wait, nb_packets_per_stream); if (ret != LTTNG_OK) { goto error; } @@ -2753,6 +3103,8 @@ int cmd_snapshot_record(struct ltt_session *session, rcu_read_lock(); cds_lfht_for_each_entry(session->snapshot.output_ht->ht, &iter.iter, sout, node.node) { + int64_t nb_packets_per_stream; + /* * Make a local copy of the output and assign the possible * temporary value given by the caller. @@ -2760,11 +3112,17 @@ int cmd_snapshot_record(struct ltt_session *session, memset(&tmp_output, 0, sizeof(tmp_output)); memcpy(&tmp_output, sout, sizeof(tmp_output)); - /* Use temporary max size. */ if (output->max_size != (uint64_t) -1ULL) { tmp_output.max_size = output->max_size; } + nb_packets_per_stream = get_session_nb_packets_per_stream(session, + tmp_output.max_size); + if (nb_packets_per_stream < 0) { + ret = LTTNG_ERR_INVALID; + goto error; + } + /* Use temporary name. */ if (*output->name != '\0') { strncpy(tmp_output.name, output->name, @@ -2774,7 +3132,7 @@ int cmd_snapshot_record(struct ltt_session *session, tmp_output.nb_snapshot = session->snapshot.nb_snapshot; ret = record_kernel_snapshot(ksess, &tmp_output, - session, wait, nb_streams); + session, wait, nb_packets_per_stream); if (ret != LTTNG_OK) { rcu_read_unlock(); goto error; @@ -2789,8 +3147,16 @@ int cmd_snapshot_record(struct ltt_session *session, struct ltt_ust_session *usess = session->ust_session; if (use_tmp_output) { + int64_t nb_packets_per_stream; + + nb_packets_per_stream = get_session_nb_packets_per_stream(session, + tmp_output.max_size); + if (nb_packets_per_stream < 0) { + ret = LTTNG_ERR_INVALID; + goto error; + } ret = record_ust_snapshot(usess, &tmp_output, session, - wait, nb_streams); + wait, nb_packets_per_stream); if (ret != LTTNG_OK) { goto error; } @@ -2802,6 +3168,8 @@ int cmd_snapshot_record(struct ltt_session *session, rcu_read_lock(); cds_lfht_for_each_entry(session->snapshot.output_ht->ht, &iter.iter, sout, node.node) { + int64_t nb_packets_per_stream; + /* * Make a local copy of the output and assign the possible * temporary value given by the caller. @@ -2809,11 +3177,18 @@ int cmd_snapshot_record(struct ltt_session *session, memset(&tmp_output, 0, sizeof(tmp_output)); memcpy(&tmp_output, sout, sizeof(tmp_output)); - /* Use temporary max size. */ if (output->max_size != (uint64_t) -1ULL) { tmp_output.max_size = output->max_size; } + nb_packets_per_stream = get_session_nb_packets_per_stream(session, + tmp_output.max_size); + if (nb_packets_per_stream < 0) { + ret = LTTNG_ERR_INVALID; + rcu_read_unlock(); + goto error; + } + /* Use temporary name. */ if (*output->name != '\0') { strncpy(tmp_output.name, output->name, @@ -2823,7 +3198,7 @@ int cmd_snapshot_record(struct ltt_session *session, tmp_output.nb_snapshot = session->snapshot.nb_snapshot; ret = record_ust_snapshot(usess, &tmp_output, session, - wait, nb_streams); + wait, nb_packets_per_stream); if (ret != LTTNG_OK) { rcu_read_unlock(); goto error; @@ -2836,6 +3211,8 @@ int cmd_snapshot_record(struct ltt_session *session, if (snapshot_success) { session->snapshot.nb_snapshot++; + } else { + ret = LTTNG_ERR_SNAPSHOT_FAIL; } error: