X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fust-app.c;h=22b395773b2efe922ef8f1e579fd71e07161bb04;hp=3accde269509e0e8b66c3b25c35b86a90289fb5b;hb=5922e6c21e0ad1a5050b2440bcc14f15abd9c5f4;hpb=d42f20df7f4ef656485ebb19c94a35152cb386f7 diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c index 3accde269..22b395773 100644 --- a/src/bin/lttng-sessiond/ust-app.c +++ b/src/bin/lttng-sessiond/ust-app.c @@ -721,9 +721,9 @@ error: } /* - * Create stream onto the UST tracer for a UST session. + * Create metadata stream onto the UST tracer for a given session. */ -static int create_ust_stream(struct ust_app *app, +static int create_ust_metadata_stream(struct ust_app *app, struct ust_app_session *ua_sess) { int ret; @@ -1030,17 +1030,27 @@ error: } /* - * Create a UST session onto the tracer of app and add it the session - * hashtable. + * Create a session on the tracer side for the given app. * - * Return ust app session or NULL on error. + * On success, ua_sess_ptr is populated with the session pointer or else left + * untouched. If the session was created, is_created is set to 1. On error, + * it's left untouched. Note that ua_sess_ptr is mandatory but is_created can + * be NULL. + * + * Returns 0 on success or else a negative code which is either -ENOMEM or + * -ENOTCONN which is the default code if the ustctl_create_session fails. */ -static struct ust_app_session *create_ust_app_session( - struct ltt_ust_session *usess, struct ust_app *app) +static int create_ust_app_session(struct ltt_ust_session *usess, + struct ust_app *app, struct ust_app_session **ua_sess_ptr, + int *is_created) { - int ret; + int ret, created = 0; struct ust_app_session *ua_sess; + assert(usess); + assert(app); + assert(ua_sess_ptr); + health_code_update(&health_thread_cmd); ua_sess = lookup_session_by_app(usess, app); @@ -1050,9 +1060,11 @@ static struct ust_app_session *create_ust_app_session( ua_sess = alloc_ust_app_session(); if (ua_sess == NULL) { /* Only malloc can failed so something is really wrong */ - goto end; + ret = -ENOMEM; + goto error; } shadow_copy_session(ua_sess, usess, app); + created = 1; } health_code_update(&health_thread_cmd); @@ -1062,9 +1074,14 @@ static struct ust_app_session *create_ust_app_session( if (ret < 0) { ERR("Creating session for app pid %d", app->pid); delete_ust_app_session(-1, ua_sess); - /* This means that the tracer is gone... */ - ua_sess = (void*) -1UL; - goto end; + if (ret != -ENOMEM) { + /* + * Tracer is probably gone or got an internal error so let's + * behave like it will soon unregister or not usable. + */ + ret = -ENOTCONN; + } + goto error; } ua_sess->handle = ret; @@ -1076,9 +1093,16 @@ static struct ust_app_session *create_ust_app_session( DBG2("UST app session created successfully with handle %d", ret); } -end: + *ua_sess_ptr = ua_sess; + if (is_created) { + *is_created = created; + } + /* Everything went well. */ + ret = 0; + +error: health_code_update(&health_thread_cmd); - return ua_sess; + return ret; } /* @@ -1212,11 +1236,12 @@ error: } /* - * Create UST app channel and create it on the tracer. + * Create UST app channel and create it on the tracer. Set ua_chanp of the + * newly created channel if not NULL. */ -static struct ust_app_channel *create_ust_app_channel( - struct ust_app_session *ua_sess, struct ltt_ust_channel *uchan, - struct ust_app *app) +static int create_ust_app_channel(struct ust_app_session *ua_sess, + struct ltt_ust_channel *uchan, struct ust_app *app, + struct ust_app_channel **ua_chanp) { int ret = 0; struct lttng_ht_iter iter; @@ -1234,6 +1259,7 @@ static struct ust_app_channel *create_ust_app_channel( ua_chan = alloc_ust_app_channel(uchan->name, &uchan->attr); if (ua_chan == NULL) { /* Only malloc can fail here */ + ret = -ENOMEM; goto error; } shadow_copy_channel(ua_chan, uchan); @@ -1245,17 +1271,23 @@ static struct ust_app_channel *create_ust_app_channel( goto error; } + /* Only add the channel if successful on the tracer side. */ lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node); DBG2("UST app create channel %s for PID %d completed", ua_chan->name, app->pid); end: - return ua_chan; + if (ua_chanp) { + *ua_chanp = ua_chan; + } + + /* Everything went well. */ + return 0; error: delete_ust_app_channel(-1, ua_chan); - return NULL; + return ret; } /* @@ -1343,7 +1375,7 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, /* Open UST metadata stream */ if (ua_sess->metadata->stream_obj == NULL) { - ret = create_ust_stream(app, ua_sess); + ret = create_ust_metadata_stream(app, ua_sess); if (ret < 0) { goto error; } @@ -1580,11 +1612,11 @@ int ust_app_list_events(struct lttng_event **events) size_t nbmem, count = 0; struct lttng_ht_iter iter; struct ust_app *app; - struct lttng_event *tmp; + struct lttng_event *tmp_event; nbmem = UST_APP_EVENT_LIST_SIZE; - tmp = zmalloc(nbmem * sizeof(struct lttng_event)); - if (tmp == NULL) { + tmp_event = zmalloc(nbmem * sizeof(struct lttng_event)); + if (tmp_event == NULL) { PERROR("zmalloc ust app events"); ret = -ENOMEM; goto error; @@ -1616,29 +1648,31 @@ int ust_app_list_events(struct lttng_event **events) health_code_update(&health_thread_cmd); if (count >= nbmem) { /* In case the realloc fails, we free the memory */ - void *tmp_ptr = (void *) tmp; + void *ptr; + DBG2("Reallocating event list from %zu to %zu entries", nbmem, 2 * nbmem); nbmem *= 2; - tmp = realloc(tmp, nbmem * sizeof(struct lttng_event)); - if (tmp == NULL) { + ptr = realloc(tmp_event, nbmem * sizeof(struct lttng_event)); + if (ptr == NULL) { PERROR("realloc ust app events"); - free(tmp_ptr); + free(tmp_event); ret = -ENOMEM; goto rcu_error; } + tmp_event = ptr; } - memcpy(tmp[count].name, uiter.name, LTTNG_UST_SYM_NAME_LEN); - tmp[count].loglevel = uiter.loglevel; - tmp[count].type = (enum lttng_event_type) LTTNG_UST_TRACEPOINT; - tmp[count].pid = app->pid; - tmp[count].enabled = -1; + memcpy(tmp_event[count].name, uiter.name, LTTNG_UST_SYM_NAME_LEN); + tmp_event[count].loglevel = uiter.loglevel; + tmp_event[count].type = (enum lttng_event_type) LTTNG_UST_TRACEPOINT; + tmp_event[count].pid = app->pid; + tmp_event[count].enabled = -1; count++; } } ret = count; - *events = tmp; + *events = tmp_event; DBG2("UST app list events done (%zu events)", count); @@ -1658,11 +1692,11 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) size_t nbmem, count = 0; struct lttng_ht_iter iter; struct ust_app *app; - struct lttng_event_field *tmp; + struct lttng_event_field *tmp_event; nbmem = UST_APP_EVENT_LIST_SIZE; - tmp = zmalloc(nbmem * sizeof(struct lttng_event_field)); - if (tmp == NULL) { + tmp_event = zmalloc(nbmem * sizeof(struct lttng_event_field)); + if (tmp_event == NULL) { PERROR("zmalloc ust app event fields"); ret = -ENOMEM; goto error; @@ -1694,34 +1728,36 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) health_code_update(&health_thread_cmd); if (count >= nbmem) { /* In case the realloc fails, we free the memory */ - void *tmp_ptr = (void *) tmp; + void *ptr; + DBG2("Reallocating event field list from %zu to %zu entries", nbmem, 2 * nbmem); nbmem *= 2; - tmp = realloc(tmp, nbmem * sizeof(struct lttng_event_field)); - if (tmp == NULL) { + ptr = realloc(tmp_event, nbmem * sizeof(struct lttng_event_field)); + if (ptr == NULL) { PERROR("realloc ust app event fields"); - free(tmp_ptr); + free(tmp_event); ret = -ENOMEM; goto rcu_error; } + tmp_event = ptr; } - memcpy(tmp[count].field_name, uiter.field_name, LTTNG_UST_SYM_NAME_LEN); - tmp[count].type = uiter.type; - tmp[count].nowrite = uiter.nowrite; + memcpy(tmp_event[count].field_name, uiter.field_name, LTTNG_UST_SYM_NAME_LEN); + tmp_event[count].type = uiter.type; + tmp_event[count].nowrite = uiter.nowrite; - memcpy(tmp[count].event.name, uiter.event_name, LTTNG_UST_SYM_NAME_LEN); - tmp[count].event.loglevel = uiter.loglevel; - tmp[count].event.type = LTTNG_UST_TRACEPOINT; - tmp[count].event.pid = app->pid; - tmp[count].event.enabled = -1; + memcpy(tmp_event[count].event.name, uiter.event_name, LTTNG_UST_SYM_NAME_LEN); + tmp_event[count].event.loglevel = uiter.loglevel; + tmp_event[count].event.type = LTTNG_UST_TRACEPOINT; + tmp_event[count].event.pid = app->pid; + tmp_event[count].event.enabled = -1; count++; } } ret = count; - *fields = tmp; + *fields = tmp_event; DBG2("UST app list event fields done (%zu events)", count); @@ -2017,11 +2053,10 @@ int ust_app_disable_all_event_glb(struct ltt_ust_session *usess, int ust_app_create_channel_glb(struct ltt_ust_session *usess, struct ltt_ust_channel *uchan) { - int ret = 0; + int ret = 0, created; struct lttng_ht_iter iter; struct ust_app *app; - struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; + struct ust_app_session *ua_sess = NULL; /* Very wrong code flow */ assert(usess); @@ -2046,29 +2081,39 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, * that if session exist, it will simply return a pointer to the ust * app session. */ - ua_sess = create_ust_app_session(usess, app); - if (ua_sess == NULL) { - /* The malloc() failed. */ - ret = -1; - goto error; - } else if (ua_sess == (void *) -1UL) { - /* The application's socket is not valid. Contiuing */ - ret = -1; - continue; + ret = create_ust_app_session(usess, app, &ua_sess, &created); + if (ret < 0) { + switch (ret) { + case -ENOTCONN: + /* + * The application's socket is not valid. Either a bad socket + * or a timeout on it. We can't inform the caller that for a + * specific app, the session failed so lets continue here. + */ + continue; + case -ENOMEM: + default: + goto error_rcu_unlock; + } } + assert(ua_sess); - /* Create channel onto application */ - ua_chan = create_ust_app_channel(ua_sess, uchan, app); - if (ua_chan == NULL) { - /* Major problem here and it's maybe the tracer or malloc() */ - ret = -1; - goto error; + /* Create channel onto application. We don't need the chan ref. */ + ret = create_ust_app_channel(ua_sess, uchan, app, NULL); + if (ret < 0) { + if (ret == -ENOMEM) { + /* No more memory is a fatal error. Stop right now. */ + goto error_rcu_unlock; + } + /* Cleanup the created session if it's the case. */ + if (created) { + delete_ust_app_session(app->sock, ua_sess); + } } } +error_rcu_unlock: rcu_read_unlock(); - -error: return ret; } @@ -2244,9 +2289,6 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) } } - /* Indicate that the session has been started once */ - ua_sess->started = 1; - ret = create_ust_app_metadata(ua_sess, usess->pathname, app); if (ret < 0) { ret = LTTNG_ERR_UST_META_FAIL; @@ -2278,11 +2320,16 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) ret = ustctl_create_stream(app->sock, ua_chan->obj, &ustream->obj); if (ret < 0) { - /* Got all streams */ - lttng_fd_put(LTTNG_FD_APPS, 2); + /* Free unused memory and reset FD states. */ free(ustream); + lttng_fd_put(LTTNG_FD_APPS, 2); + if (ret == -ENOENT) { + /* Got all streams. Continue normal execution. */ + break; + } + /* Error at this point. Stop everything. */ ret = LTTNG_ERR_UST_STREAM_FAIL; - break; + goto error_rcu_unlock; } ustream->handle = ustream->obj->handle; @@ -2340,10 +2387,13 @@ skip_setup: /* This start the UST tracing */ ret = ustctl_start_session(app->sock, ua_sess->handle); if (ret < 0) { - ERR("Error starting tracing for app pid: %d", app->pid); + ERR("Error starting tracing for app pid: %d (ret: %d)", app->pid, ret); goto error_rcu_unlock; } + /* Indicate that the session has been started once */ + ua_sess->started = 1; + health_code_update(&health_thread_cmd); /* Quiescent wait after starting trace */ @@ -2385,17 +2435,20 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) /* * If started = 0, it means that stop trace has been called for a session - * that was never started. This is a code flow error and should never - * happen. + * that was never started. It's possible since we can have a fail start + * from either the application manager thread or the command thread. Simply + * indicate that this is a stop error. */ - assert(ua_sess->started == 1); + if (!ua_sess->started) { + goto error_rcu_unlock; + } health_code_update(&health_thread_cmd); /* This inhibits UST tracing */ ret = ustctl_stop_session(app->sock, ua_sess->handle); if (ret < 0) { - ERR("Error stopping tracing for app pid: %d", app->pid); + ERR("Error stopping tracing for app pid: %d (ret: %d)", app->pid, ret); goto error_rcu_unlock; } @@ -2442,7 +2495,7 @@ error_rcu_unlock: /* * Destroy a specific UST session in apps. */ -int ust_app_destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) +static int destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) { struct ust_app_session *ua_sess; struct lttng_ust_object_data obj; @@ -2556,7 +2609,7 @@ int ust_app_destroy_trace_all(struct ltt_ust_session *usess) rcu_read_lock(); cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = ust_app_destroy_trace(usess, app); + ret = destroy_trace(usess, app); if (ret < 0) { /* Continue to next apps even on error */ continue; @@ -2576,15 +2629,12 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) int ret = 0; struct lttng_ht_iter iter, uiter, iter_ctx; struct ust_app *app; - struct ust_app_session *ua_sess; + struct ust_app_session *ua_sess = NULL; struct ust_app_channel *ua_chan; struct ust_app_event *ua_event; struct ust_app_ctx *ua_ctx; - if (usess == NULL) { - ERR("No UST session on global update. Returning"); - goto error; - } + assert(usess); DBG2("UST app global update for app sock %d for session id %d", sock, usess->id); @@ -2601,11 +2651,12 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) goto error; } - ua_sess = create_ust_app_session(usess, app); - if (ua_sess == NULL || ua_sess == (void *) -1UL) { - /* Tracer is gone for this session and has been freed */ + ret = create_ust_app_session(usess, app, &ua_sess, NULL); + if (ret < 0) { + /* Tracer is probably gone or ENOMEM. */ goto error; } + assert(ua_sess); /* * We can iterate safely here over all UST app session sicne the create ust @@ -2638,12 +2689,6 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) /* FIXME: Should we quit here or continue... */ continue; } - - ret = set_ust_event_filter(ua_event, app); - if (ret < 0) { - /* FIXME: Should we quit here or continue... */ - continue; - } } }