*
*/
-#include "bin/lttng-sessiond/tracker.h"
-#include "common/tracker.h"
-#include "common/utils.h"
-#include "lttng/event.h"
-#include "lttng/lttng-error.h"
-#include "lttng/tracker.h"
#define _LGPL_SOURCE
#include <fcntl.h>
#include <stdlib.h>
#include <sys/types.h>
#include <common/common.h>
+#include <common/hashtable/utils.h>
#include <common/trace-chunk.h>
#include <common/kernel-ctl/kernel-ctl.h>
#include <common/kernel-ctl/kernel-ioctl.h>
#include <common/sessiond-comm/sessiond-comm.h>
+#include <common/tracker.h>
+#include <common/utils.h>
+#include <lttng/event.h>
+#include <lttng/lttng-error.h>
+#include <lttng/tracker.h>
+#include <lttng/userspace-probe.h>
+#include <lttng/userspace-probe-internal.h>
+#include <lttng/condition/on-event.h>
+#include <lttng/condition/on-event-internal.h>
+#include <lttng/event-rule/event-rule.h>
+#include <lttng/event-rule/event-rule-internal.h>
+#include <lttng/event-rule/userspace-probe-internal.h>
+#include <lttng/map/map.h>
+#include <lttng/map/map-internal.h>
+#include <lttng/map/map-query-internal.h>
+#include <lttng/map-key.h>
+#include <lttng/map-key-internal.h>
+
+#include "event-notifier-error-accounting.h"
#include "lttng-sessiond.h"
#include "lttng-syscall.h"
+#include "condition-internal.h"
#include "consumer.h"
#include "kernel.h"
#include "kernel-consumer.h"
#include "kern-modules.h"
+#include "map.h"
+#include "sessiond-config.h"
#include "utils.h"
#include "rotate.h"
#include "modprobe.h"
+#include "tracker.h"
+#include "notification-thread-commands.h"
/*
* Key used to reference a channel between the sessiond and the consumer. This
static const char *module_proc_lttng = "/proc/lttng";
static int kernel_tracer_fd = -1;
+static int kernel_tracer_event_notifier_group_fd = -1;
+static int kernel_tracer_event_notifier_group_notification_fd = -1;
+static struct cds_lfht *kernel_token_to_event_notifier_rule_ht;
-#include <lttng/userspace-probe.h>
-#include <lttng/userspace-probe-internal.h>
/*
* Add context on a kernel channel.
*
return -1;
}
+/*
+ * Create a kernel event notifier group, register it to the kernel tracer and
+ * add it to the kernel session.
+ */
+static int kernel_create_event_notifier_group(int *event_notifier_group_fd)
+{
+ int ret;
+ int local_fd = -1;
+
+ assert(event_notifier_group_fd);
+
+ /* Kernel event notifier group creation. */
+ ret = kernctl_create_event_notifier_group(kernel_tracer_fd);
+ if (ret < 0) {
+ PERROR("Failed to create kernel event notifier group");
+ ret = -1;
+ goto error;
+ }
+
+ local_fd = ret;
+
+ /* Prevent fd duplication after execlp(). */
+ ret = fcntl(local_fd, F_SETFD, FD_CLOEXEC);
+ if (ret < 0) {
+ PERROR("Failed to set FD_CLOEXEC on kernel event notifier group file descriptor: fd = %d",
+ local_fd);
+ goto error;
+ }
+
+ DBG("Created kernel event notifier group: fd = %d", local_fd);
+ *event_notifier_group_fd = local_fd;
+ local_fd = -1;
+ ret = 0;
+error:
+ if (local_fd >= 0) {
+ ret = close(local_fd);
+ if (ret) {
+ PERROR("Failed to close kernel event notifier group file descriptor: fd = %d",
+ local_fd);
+ }
+ }
+
+ return ret;
+}
+
/*
* Compute the offset of the instrumentation byte in the binary based on the
* function probe location using the ELF lookup method.
static
int extract_userspace_probe_offset_function_elf(
const struct lttng_userspace_probe_location *probe_location,
- struct ltt_kernel_session *session, uint64_t *offset)
+ uid_t uid, gid_t gid, uint64_t *offset)
{
int fd;
int ret = 0;
goto end;
}
- ret = run_as_extract_elf_symbol_offset(fd, symbol, session->uid,
- session->gid, offset);
+ ret = run_as_extract_elf_symbol_offset(fd, symbol, uid, gid, offset);
if (ret < 0) {
DBG("userspace probe offset calculation failed for "
"function %s", symbol);
static
int extract_userspace_probe_offset_tracepoint_sdt(
const struct lttng_userspace_probe_location *probe_location,
- struct ltt_kernel_session *session, uint64_t **offsets,
+ uid_t uid, gid_t gid, uint64_t **offsets,
uint32_t *offsets_count)
{
enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
}
ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
- session->uid, session->gid, offsets, offsets_count);
+ uid, gid, offsets, offsets_count);
if (ret < 0) {
DBG("userspace probe offset calculation failed for sdt "
"probe %s:%s", provider_name, probe_name);
return ret;
}
-/*
- * Extract the offsets of the instrumentation point for the different lookup
- * methods.
- */
static
-int userspace_probe_add_callsites(struct lttng_event *ev,
- struct ltt_kernel_session *session, int fd)
+int userspace_probe_add_callsite(
+ const struct lttng_userspace_probe_location *location,
+ uid_t uid, gid_t gid, int fd)
{
const struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
enum lttng_userspace_probe_location_lookup_method_type type;
- const struct lttng_userspace_probe_location *location = NULL;
int ret;
- assert(ev);
- assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
-
- location = lttng_event_get_userspace_probe_location(ev);
- if (!location) {
- ret = -1;
- goto end;
- }
- lookup_method =
- lttng_userspace_probe_location_get_lookup_method(location);
+ lookup_method = lttng_userspace_probe_location_get_lookup_method(location);
if (!lookup_method) {
ret = -1;
goto end;
struct lttng_kernel_event_callsite callsite;
uint64_t offset;
- ret = extract_userspace_probe_offset_function_elf(location, session, &offset);
+ ret = extract_userspace_probe_offset_function_elf(location,
+ uid, gid, &offset);
if (ret) {
ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
goto end;
callsite.u.uprobe.offset = offset;
ret = kernctl_add_callsite(fd, &callsite);
if (ret) {
- WARN("Adding callsite to userspace probe "
- "event %s failed.", ev->name);
+ WARN("Failed to add callsite to ELF userspace probe.");
ret = LTTNG_ERR_KERN_ENABLE_FAIL;
goto end;
}
* This call allocates the offsets buffer. This buffer must be freed
* by the caller
*/
- ret = extract_userspace_probe_offset_tracepoint_sdt(location, session,
- &offsets, &offsets_count);
+ ret = extract_userspace_probe_offset_tracepoint_sdt(location,
+ uid, gid, &offsets, &offsets_count);
if (ret) {
ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
goto end;
callsite.u.uprobe.offset = offsets[i];
ret = kernctl_add_callsite(fd, &callsite);
if (ret) {
- WARN("Adding callsite to userspace probe "
- "event %s failed.", ev->name);
+ WARN("Failed to add callsite to SDT userspace probe");
ret = LTTNG_ERR_KERN_ENABLE_FAIL;
free(offsets);
goto end;
return ret;
}
+/*
+ * Extract the offsets of the instrumentation point for the different lookup
+ * methods.
+ */
+static
+int userspace_probe_event_add_callsites(struct lttng_event *ev,
+ struct ltt_kernel_session *session, int fd)
+{
+ int ret;
+ const struct lttng_userspace_probe_location *location = NULL;
+
+ assert(ev);
+ assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
+
+ location = lttng_event_get_userspace_probe_location(ev);
+ if (!location) {
+ ret = -1;
+ goto end;
+ }
+
+ ret = userspace_probe_add_callsite(location, session->uid, session->gid,
+ fd);
+ if (ret) {
+ WARN("Failed to add callsite to userspace probe event '%s'",
+ ev->name);
+ }
+
+end:
+ return ret;
+}
+
+/*
+ * Extract the offsets of the instrumentation point for the different look-up
+ * methods.
+ */
+static int userspace_probe_event_rule_add_callsites(
+ const struct lttng_event_rule *rule,
+ const struct lttng_credentials *creds,
+ int fd)
+{
+ int ret;
+ enum lttng_event_rule_status status;
+ enum lttng_event_rule_type event_rule_type;
+ const struct lttng_userspace_probe_location *location = NULL;
+
+ assert(rule);
+ assert(creds);
+
+ event_rule_type = lttng_event_rule_get_type(rule);
+ assert(event_rule_type == LTTNG_EVENT_RULE_TYPE_USERSPACE_PROBE);
+
+ status = lttng_event_rule_userspace_probe_get_location(rule, &location);
+ if (status != LTTNG_EVENT_RULE_STATUS_OK || !location) {
+ ret = -1;
+ goto end;
+ }
+
+ ret = userspace_probe_add_callsite(location,
+ lttng_credentials_get_uid(creds),
+ lttng_credentials_get_gid(creds), fd);
+ if (ret) {
+ WARN("Failed to add callsite to user space probe object: fd = %d",
+ fd);
+ }
+
+end:
+ return ret;
+}
+
/*
* Create a kernel event, enable it to the kernel tracer and add it to the
* channel event list of the kernel session.
int kernel_create_event(struct lttng_event *ev,
struct ltt_kernel_channel *channel,
char *filter_expression,
- struct lttng_filter_bytecode *filter)
+ struct lttng_bytecode *filter)
{
int err, fd;
enum lttng_error_code ret;
}
if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
- ret = userspace_probe_add_callsites(ev, channel->session, event->fd);
+ ret = userspace_probe_event_add_callsites(ev, channel->session,
+ event->fd);
if (ret) {
goto add_callsite_error;
}
return ret;
}
+/*
+ * Disable a kernel event notifier.
+ */
+static
+int kernel_disable_event_notifier_rule(struct ltt_kernel_event_notifier_rule *event)
+{
+ int ret;
+
+ assert(event);
+
+ rcu_read_lock();
+ cds_lfht_del(kernel_token_to_event_notifier_rule_ht, &event->ht_node);
+ rcu_read_unlock();
+
+ ret = kernctl_disable(event->fd);
+ if (ret < 0) {
+ switch (-ret) {
+ case EEXIST:
+ ret = LTTNG_ERR_KERN_EVENT_EXIST;
+ break;
+ default:
+ PERROR("Failed to disable kernel event notifier: fd = %d, token = %" PRIu64,
+ event->fd, event->token);
+ break;
+ }
+ goto error;
+ }
+
+ event->enabled = 0;
+ DBG("Disabled kernel event notifier: fd = %d, token = %" PRIu64,
+ event->fd, event->token);
+
+error:
+ return ret;
+}
+
static
struct process_attr_tracker *_kernel_get_process_attr_tracker(
struct ltt_kernel_session *session,
return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
}
-/*
- * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
- * version number.
- *
- * Return 1 on success, 0 when feature is not supported, negative value in case
- * of errors.
- */
-int kernel_supports_ring_buffer_snapshot_sample_positions(void)
+static
+int kernel_tracer_abi_greater_or_equal(unsigned int major, unsigned int minor)
{
- int ret = 0; // Not supported by default
+ int ret;
struct lttng_kernel_tracer_abi_version abi;
ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
goto error;
}
+ ret = abi.major > major || (abi.major == major && abi.minor >= minor);
+error:
+ return ret;
+}
+
+/*
+ * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
+ * version number.
+ *
+ * Return 1 on success, 0 when feature is not supported, negative value in case
+ * of errors.
+ */
+int kernel_supports_ring_buffer_snapshot_sample_positions(void)
+{
/*
* RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
*/
- if (abi.major >= 2 && abi.minor >= 3) {
- /* Supported */
- ret = 1;
- } else {
- /* Not supported */
- ret = 0;
- }
-error:
- return ret;
+ return kernel_tracer_abi_greater_or_equal(2, 3);
}
/*
*/
int kernel_supports_ring_buffer_packet_sequence_number(void)
{
- int ret = 0; // Not supported by default
- struct lttng_kernel_tracer_abi_version abi;
-
- ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
- if (ret < 0) {
- ERR("Failed to retrieve lttng-modules ABI version");
- goto error;
- }
-
/*
* Packet sequence number was introduced in LTTng 2.8,
* lttng-modules ABI 2.1.
*/
- if (abi.major >= 2 && abi.minor >= 1) {
- /* Supported */
- ret = 1;
- } else {
- /* Not supported */
- ret = 0;
- }
-error:
- return ret;
+ return kernel_tracer_abi_greater_or_equal(2, 1);
+}
+
+/*
+ * Check for the support of event notifiers via abi version number.
+ *
+ * Return 1 on success, 0 when feature is not supported, negative value in case
+ * of errors.
+ */
+int kernel_supports_event_notifiers(void)
+{
+ /*
+ * Event notifiers were introduced in LTTng 2.13, lttng-modules ABI 2.6.
+ */
+ return kernel_tracer_abi_greater_or_equal(2, 6);
}
/*
socket, node.node) {
struct ltt_kernel_channel *chan;
- /* For each channel, ask the consumer to rotate it. */
+ /* For each channel, ask the consumer to rotate it. */
cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
DBG("Rotate kernel channel %" PRIu64 ", session %s",
chan->key, session->name);
if (ret < 0) {
goto error_modules;
}
-
if (ret < 1) {
WARN("Kernel tracer does not support buffer monitoring. "
"The monitoring timer of channels in the kernel domain "
"will be set to 0 (disabled).");
}
- DBG("Kernel tracer fd %d", kernel_tracer_fd);
+ ret = kernel_supports_event_notifiers();
+ if (ret < 0) {
+ ERR("Failed to check for kernel tracer event notifier support");
+ goto error_modules;
+ }
+ ret = kernel_create_event_notifier_group(&kernel_tracer_event_notifier_group_fd);
+ if (ret < 0) {
+ /* This is not fatal. */
+ WARN("Failed to create kernel event notifier group");
+ kernel_tracer_event_notifier_group_fd = -1;
+ } else {
+ enum event_notifier_error_accounting_status error_accounting_status;
+ enum lttng_error_code error_code_ret =
+ kernel_create_event_notifier_group_notification_fd(
+ &kernel_tracer_event_notifier_group_notification_fd);
+
+ if (error_code_ret != LTTNG_OK) {
+ goto error_modules;
+ }
+
+ error_accounting_status = event_notifier_error_accounting_register_kernel(
+ kernel_tracer_event_notifier_group_fd);
+ if (error_accounting_status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) {
+ ERR("Error initializing event notifier error accounting for kernel tracer.");
+ error_code_ret = LTTNG_ERR_EVENT_NOTIFIER_ERROR_ACCOUNTING;
+ goto error_modules;
+ }
+
+ kernel_token_to_event_notifier_rule_ht = cds_lfht_new(
+ DEFAULT_HT_SIZE, 1, 0,
+ CDS_LFHT_AUTO_RESIZE | CDS_LFHT_ACCOUNTING,
+ NULL);
+ if (!kernel_token_to_event_notifier_rule_ht) {
+ goto error_token_ht;
+ }
+ }
+
+ DBG("Kernel tracer initialized: kernel tracer fd = %d, event notifier group fd = %d, event notifier group notification fd = %d",
+ kernel_tracer_fd, kernel_tracer_event_notifier_group_fd,
+ kernel_tracer_event_notifier_group_notification_fd);
ret = syscall_init_table(kernel_tracer_fd);
if (ret < 0) {
ERR("Unable to populate syscall table. Syscall tracing won't "
"work for this session daemon.");
}
+
return 0;
error_version:
modprobe_remove_lttng_control();
ret = close(kernel_tracer_fd);
if (ret) {
- PERROR("close");
+ PERROR("Failed to close kernel tracer file descriptor: fd = %d",
+ kernel_tracer_fd);
}
kernel_tracer_fd = -1;
return LTTNG_ERR_KERN_VERSION;
+
+error_token_ht:
+ ret = close(kernel_tracer_event_notifier_group_notification_fd);
+ if (ret) {
+ PERROR("Failed to close kernel tracer event notifier group notification file descriptor: fd = %d",
+ kernel_tracer_event_notifier_group_notification_fd);
+ }
+
error_modules:
+ ret = close(kernel_tracer_event_notifier_group_fd);
+ if (ret) {
+ PERROR("Failed to close kernel tracer event notifier group file descriptor: fd = %d",
+ kernel_tracer_event_notifier_group_fd);
+ }
+
ret = close(kernel_tracer_fd);
if (ret) {
- PERROR("close");
+ PERROR("Failed to close kernel tracer file descriptor: fd = %d",
+ kernel_tracer_fd);
}
error_open:
LTTNG_HIDDEN
void cleanup_kernel_tracer(void)
{
- int ret;
+ DBG2("Closing kernel event notifier group notification file descriptor");
+ if (kernel_tracer_event_notifier_group_notification_fd >= 0) {
+ int ret = notification_thread_command_remove_tracer_event_source(
+ notification_thread_handle,
+ kernel_tracer_event_notifier_group_notification_fd);
+ if (ret != LTTNG_OK) {
+ ERR("Failed to remove kernel event notifier notification from notification thread");
+ }
+
+ ret = close(kernel_tracer_event_notifier_group_notification_fd);
+ if (ret) {
+ PERROR("Failed to close kernel event notifier group notification file descriptor: fd = %d",
+ kernel_tracer_event_notifier_group_notification_fd);
+ }
+
+ kernel_tracer_event_notifier_group_notification_fd = -1;
+ }
+
+ if (kernel_token_to_event_notifier_rule_ht) {
+ const int ret = cds_lfht_destroy(
+ kernel_token_to_event_notifier_rule_ht, NULL);
+ assert(ret == 0);
+ }
+
+ DBG2("Closing kernel event notifier group file descriptor");
+ if (kernel_tracer_event_notifier_group_fd >= 0) {
+ const int ret = close(kernel_tracer_event_notifier_group_fd);
+
+ if (ret) {
+ PERROR("Failed to close kernel event notifier group file descriptor: fd = %d",
+ kernel_tracer_event_notifier_group_fd);
+ }
+
+ kernel_tracer_event_notifier_group_fd = -1;
+ }
DBG2("Closing kernel fd");
if (kernel_tracer_fd >= 0) {
- ret = close(kernel_tracer_fd);
+ const int ret = close(kernel_tracer_fd);
+
if (ret) {
- PERROR("close");
+ PERROR("Failed to close kernel tracer file descriptor: fd = %d",
+ kernel_tracer_fd);
}
+
kernel_tracer_fd = -1;
}
- DBG("Unloading kernel modules");
- modprobe_remove_lttng_all();
+
free(syscall_table);
}
{
return kernel_tracer_fd >= 0;
}
-
-/*
- * Clear a kernel session.
- *
- * Return LTTNG_OK on success or else an LTTng error code.
- */
-enum lttng_error_code kernel_clear_session(struct ltt_session *session)
+static
+enum lttng_error_code kernel_clear_session_channels(struct ltt_session *session)
{
int ret;
enum lttng_error_code status = LTTNG_OK;
assert(ksess);
assert(ksess->consumer);
- DBG("Clear kernel session %s (session %" PRIu64 ")",
- session->name, session->id);
-
rcu_read_lock();
if (ksess->active) {
}
}
+
goto end;
error:
switch (-ret) {
rcu_read_unlock();
return status;
}
+
+static
+enum lttng_error_code kernel_map_clear_all(struct ltt_kernel_map *map)
+{
+ enum lttng_error_code status;
+ uint64_t descr_count, i;
+ int ret;
+
+ assert(map);
+
+ ret = kernctl_counter_map_descriptor_count(map->fd, &descr_count);
+ if (ret) {
+ ERR("Error getting map descriptor count");
+ status = LTTNG_ERR_MAP_VALUES_LIST_FAIL;
+ goto end;
+ }
+
+ for(i = 0; i < descr_count; i++) {
+ struct lttng_kernel_counter_map_descriptor descriptor = {0};
+ struct lttng_kernel_counter_clear counter_clear = {0};
+
+ descriptor.descriptor_index = i;
+
+ ret = kernctl_counter_map_descriptor(map->fd, &descriptor);
+ if (ret) {
+ ERR("Error getting map descriptor %"PRIu64, i);
+ status = LTTNG_ERR_MAP_VALUES_LIST_FAIL;
+ goto end;
+ }
+
+ counter_clear.index.number_dimensions = 1;
+ counter_clear.index.dimension_indexes[0] = descriptor.array_index;
+
+ ret = kernctl_counter_clear(map->fd, &counter_clear);
+ if (ret) {
+ ERR("Error clearing value of map descriptor %"PRIu64, i);
+ status = LTTNG_ERR_MAP_VALUES_LIST_FAIL;
+ goto end;
+ }
+ }
+
+ status = LTTNG_OK;
+end:
+ return status;
+}
+
+static
+enum lttng_error_code kernel_clear_session_maps(struct ltt_session *session)
+{
+ enum lttng_error_code status = LTTNG_OK;
+ struct ltt_kernel_map *map;
+ struct ltt_kernel_session *ksess = session->kernel_session;
+
+ assert(ksess);
+
+ cds_list_for_each_entry(map, &ksess->map_list.head, list) {
+ DBG("Clear kernel map %" PRIu64 ", session %s",
+ map->key, session->name);
+ status = kernel_map_clear_all(map);
+ if (status != LTTNG_OK) {
+ ERR("Clearing all values of map");
+ goto end;
+ }
+ }
+
+end:
+ return status;
+}
+
+/*
+ * Clear a kernel session.
+ *
+ * Return LTTNG_OK on success or else an LTTng error code.
+ */
+enum lttng_error_code kernel_clear_session(struct ltt_session *session)
+{
+ enum lttng_error_code status = LTTNG_OK;
+ struct ltt_kernel_session *ksess = session->kernel_session;
+
+ assert(ksess);
+ assert(ksess->consumer);
+
+ DBG("Clear kernel session %s (session %" PRIu64 ")",
+ session->name, session->id);
+
+ rcu_read_lock();
+
+ if (ksess->active) {
+ ERR("Expecting inactive session %s (%" PRIu64 ")", session->name, session->id);
+ status = LTTNG_ERR_FATAL;
+ goto end;
+ }
+
+ status = kernel_clear_session_channels(session);
+ if (status != LTTNG_OK) {
+ goto end;
+ }
+ /*
+ * Iterate and clear all kernel maps.
+ */
+ status = kernel_clear_session_maps(session);
+ if (status != LTTNG_OK) {
+ goto end;
+ }
+
+
+end:
+ rcu_read_unlock();
+ return status;
+}
+
+enum lttng_error_code kernel_create_event_notifier_group_notification_fd(
+ int *event_notifier_group_notification_fd)
+{
+ int local_fd = -1, ret;
+ enum lttng_error_code error_code_ret;
+
+ assert(event_notifier_group_notification_fd);
+
+ ret = kernctl_create_event_notifier_group_notification_fd(
+ kernel_tracer_event_notifier_group_fd);
+ if (ret < 0) {
+ PERROR("Failed to create kernel event notifier group notification file descriptor");
+ error_code_ret = LTTNG_ERR_EVENT_NOTIFIER_GROUP_NOTIFICATION_FD;
+ goto error;
+ }
+
+ local_fd = ret;
+
+ /* Prevent fd duplication after execlp(). */
+ ret = fcntl(local_fd, F_SETFD, FD_CLOEXEC);
+ if (ret < 0) {
+ PERROR("Failed to set FD_CLOEXEC on kernel event notifier group notification file descriptor: fd = %d",
+ local_fd);
+ error_code_ret = LTTNG_ERR_EVENT_NOTIFIER_GROUP_NOTIFICATION_FD;
+ goto error;
+ }
+
+ DBG("Created kernel notifier group notification file descriptor: fd = %d",
+ local_fd);
+ error_code_ret = LTTNG_OK;
+ *event_notifier_group_notification_fd = local_fd;
+ local_fd = -1;
+
+error:
+ if (local_fd >= 0) {
+ ret = close(local_fd);
+ if (ret) {
+ PERROR("Failed to close kernel event notifier group notification file descriptor: fd = %d",
+ local_fd);
+ }
+ }
+
+ return error_code_ret;
+}
+
+enum lttng_error_code kernel_destroy_event_notifier_group_notification_fd(
+ int event_notifier_group_notification_fd)
+{
+ enum lttng_error_code ret_code = LTTNG_OK;
+
+ DBG("Closing event notifier group notification file descriptor: fd = %d",
+ event_notifier_group_notification_fd);
+ if (event_notifier_group_notification_fd >= 0) {
+ const int ret = close(event_notifier_group_notification_fd);
+ if (ret) {
+ PERROR("Failed to close event notifier group notification file descriptor: fd = %d",
+ event_notifier_group_notification_fd);
+ }
+ }
+
+ return ret_code;
+}
+
+static
+unsigned long hash_trigger(const struct lttng_trigger *trigger)
+{
+ const struct lttng_condition *condition =
+ lttng_trigger_get_const_condition(trigger);
+
+ return lttng_condition_hash(condition);
+}
+
+static
+int match_trigger(struct cds_lfht_node *node, const void *key)
+{
+ const struct ltt_kernel_event_notifier_rule *event_notifier_rule;
+ const struct lttng_trigger *trigger = key;
+
+ event_notifier_rule = caa_container_of(node,
+ const struct ltt_kernel_event_notifier_rule, ht_node);
+
+ return lttng_trigger_is_equal(trigger, event_notifier_rule->trigger);
+}
+
+static
+int add_key_token(struct lttng_kernel_key_token *kernel_key_token,
+ const struct lttng_map_key_token *key_token)
+{
+ int ret;
+ switch (key_token->type) {
+ case LTTNG_MAP_KEY_TOKEN_TYPE_STRING:
+ {
+ const struct lttng_map_key_token_string *str_token;
+ str_token = (typeof(str_token)) key_token;
+
+ kernel_key_token->type = LTTNG_KERNEL_KEY_TOKEN_STRING;
+ kernel_key_token->arg.string_ptr = (uint64_t) str_token->string;
+
+ break;
+ }
+ case LTTNG_MAP_KEY_TOKEN_TYPE_VARIABLE:
+ {
+ const struct lttng_map_key_token_variable *var_token;
+ var_token = (typeof(var_token)) key_token;
+ switch (var_token->type) {
+ case LTTNG_MAP_KEY_TOKEN_VARIABLE_TYPE_EVENT_NAME:
+ kernel_key_token->type = LTTNG_KERNEL_KEY_TOKEN_EVENT_NAME;
+ break;
+ case LTTNG_MAP_KEY_TOKEN_VARIABLE_TYPE_PROVIDER_NAME:
+ /* The kernel events don't have providers */
+ ERR("Provider variable token type not supported for kernel tracer");
+ ret = -1;
+ goto end;
+ default:
+ abort();
+ }
+
+ break;
+ }
+ default:
+ abort();
+ }
+ ret = 0;
+end:
+ return ret;
+}
+
+enum lttng_error_code kernel_create_event_counter(
+ struct ltt_kernel_map *kmap,
+ const struct lttng_credentials *creds,
+ uint64_t action_tracer_token,
+ const struct lttng_event_rule *event_rule,
+ struct lttng_map_key *key)
+{
+ int err, fd, ret = 0;
+ unsigned int i, key_token_count;
+ enum lttng_error_code error_code_ret;
+ enum lttng_map_key_status status;
+ struct ltt_kernel_event_counter *event_counter;
+ struct lttng_kernel_counter_event k_counter_event = {};
+
+
+ event_counter = zmalloc(sizeof(*event_counter));
+ if (!event_counter) {
+ error_code_ret = LTTNG_ERR_NOMEM;
+ goto error;
+ }
+
+ trace_kernel_init_event_counter_from_event_rule(event_rule,
+ &k_counter_event);
+ event_counter->fd = -1;
+ event_counter->enabled = 1;
+ event_counter->action_tracer_token = action_tracer_token;
+ event_counter->filter = lttng_event_rule_get_filter_bytecode(event_rule);
+
+ k_counter_event.event.token = action_tracer_token;
+
+ /* Set the key pattern for this event counter. */
+ k_counter_event.key.nr_dimensions = 1;
+
+ status = lttng_map_key_get_token_count(key, &key_token_count);
+ if (status != LTTNG_MAP_KEY_STATUS_OK) {
+ error_code_ret = LTTNG_ERR_UNK;
+ goto error;
+ }
+
+ assert(key_token_count > 0);
+
+ k_counter_event.key.key_dimensions[0].nr_key_tokens = key_token_count;
+
+ for (i = 0; i < key_token_count; i++) {
+ const struct lttng_map_key_token *token =
+ lttng_map_key_get_token_at_index(key, i);
+
+ ret = add_key_token(&k_counter_event.key.key_dimensions[0].key_tokens[i],
+ token);
+ if (ret) {
+ ERR("Error appending map key token");
+ error_code_ret = LTTNG_ERR_INVALID;
+ goto error;
+ }
+ }
+
+ fd = kernctl_create_counter_event(kmap->fd, &k_counter_event);
+ if (fd < 0) {
+ switch (-fd) {
+ case EEXIST:
+ error_code_ret = LTTNG_ERR_KERN_EVENT_EXIST;
+ break;
+ case ENOSYS:
+ WARN("Event counter type not implemented");
+ error_code_ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
+ break;
+ case ENOENT:
+ WARN("Event counter %s not found!", k_counter_event.event.name);
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ break;
+ default:
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ PERROR("create event counter ioctl");
+ }
+ }
+
+ event_counter->fd = fd;
+ event_counter->enabled = true;
+
+ /* Prevent fd duplication after execlp() */
+ err = fcntl(fd, F_SETFD, FD_CLOEXEC);
+ if (err < 0) {
+ PERROR("fcntl session fd");
+ }
+
+ if (event_counter->filter) {
+ err = kernctl_filter(event_counter->fd, event_counter->filter);
+ if (err < 0) {
+ switch (-err) {
+ case ENOMEM:
+ error_code_ret = LTTNG_ERR_FILTER_NOMEM;
+ break;
+ default:
+ error_code_ret = LTTNG_ERR_FILTER_INVAL;
+ break;
+ }
+ goto filter_error;
+ }
+ }
+ if (lttng_event_rule_get_type(event_rule) ==
+ LTTNG_EVENT_RULE_TYPE_USERSPACE_PROBE) {
+ ret = userspace_probe_event_rule_add_callsites(
+ event_rule, creds, event_counter->fd);
+ if (ret) {
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ goto add_callsite_error;
+ }
+ }
+
+ err = kernctl_enable(event_counter->fd);
+ if (err < 0) {
+ switch (-err) {
+ case EEXIST:
+ error_code_ret = LTTNG_ERR_KERN_EVENT_EXIST;
+ break;
+ default:
+ PERROR("enable kernel counter event");
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ break;
+ }
+ goto enable_error;
+ }
+
+ /* Add event to event list */
+ rcu_read_lock();
+ lttng_ht_node_init_u64(&event_counter->ht_node,
+ event_counter->action_tracer_token);
+ lttng_ht_add_unique_u64(kmap->event_counters_ht,
+ &event_counter->ht_node);
+ rcu_read_unlock();
+ kmap->event_count++;
+
+ DBG("Kernel event counter %s created (fd: %d)",
+ event_counter->event->name,
+ event_counter->fd);
+ error_code_ret = LTTNG_OK;
+
+add_callsite_error:
+filter_error:
+enable_error:
+error:
+ return error_code_ret;
+}
+
+static enum lttng_error_code kernel_create_event_notifier_rule(
+ struct lttng_trigger *trigger,
+ const struct lttng_credentials *creds, uint64_t token)
+{
+ int err, fd, ret = 0;
+ enum lttng_error_code error_code_ret;
+ enum lttng_condition_status condition_status;
+ enum lttng_condition_type condition_type;
+ enum lttng_event_rule_type event_rule_type;
+ struct ltt_kernel_event_notifier_rule *event_notifier_rule;
+ struct lttng_kernel_event_notifier kernel_event_notifier = {};
+ unsigned int capture_bytecode_count = 0, i;
+ const struct lttng_condition *condition = NULL;
+ const struct lttng_event_rule *event_rule = NULL;
+ enum lttng_condition_status cond_status;
+
+ assert(trigger);
+
+ condition = lttng_trigger_get_const_condition(trigger);
+ assert(condition);
+
+ condition_type = lttng_condition_get_type(condition);
+ assert(condition_type == LTTNG_CONDITION_TYPE_ON_EVENT);
+
+ /* Does not acquire a reference. */
+ condition_status = lttng_condition_on_event_get_rule(
+ condition, &event_rule);
+ assert(condition_status == LTTNG_CONDITION_STATUS_OK);
+ assert(event_rule);
+
+ event_rule_type = lttng_event_rule_get_type(event_rule);
+ assert(event_rule_type != LTTNG_EVENT_RULE_TYPE_UNKNOWN);
+
+ error_code_ret = trace_kernel_create_event_notifier_rule(trigger, token,
+ lttng_condition_on_event_get_error_counter_index(condition),
+ &event_notifier_rule);
+ if (error_code_ret != LTTNG_OK) {
+ goto error;
+ }
+
+ error_code_ret = trace_kernel_init_event_notifier_from_event_rule(
+ event_rule, &kernel_event_notifier);
+ if (error_code_ret != LTTNG_OK) {
+ goto free_event;
+ }
+
+ kernel_event_notifier.event.token = event_notifier_rule->token;
+ kernel_event_notifier.error_counter_idx =
+ lttng_condition_on_event_get_error_counter_index(condition);
+
+ fd = kernctl_create_event_notifier(
+ kernel_tracer_event_notifier_group_fd,
+ &kernel_event_notifier);
+ if (fd < 0) {
+ switch (-fd) {
+ case EEXIST:
+ error_code_ret = LTTNG_ERR_KERN_EVENT_EXIST;
+ break;
+ case ENOSYS:
+ WARN("Failed to create kernel event notifier: not notifier type not implemented");
+ error_code_ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
+ break;
+ case ENOENT:
+ WARN("Failed to create kernel event notifier: not found: name = '%s'",
+ kernel_event_notifier.event.name);
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ break;
+ default:
+ PERROR("Failed to create kernel event notifier: error code = %d, name = '%s'",
+ fd, kernel_event_notifier.event.name);
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ }
+ goto free_event;
+ }
+
+ event_notifier_rule->fd = fd;
+ /* Prevent fd duplication after execlp(). */
+ err = fcntl(event_notifier_rule->fd, F_SETFD, FD_CLOEXEC);
+ if (err < 0) {
+ PERROR("Failed to set FD_CLOEXEC on kernel event notifier file descriptor: fd = %d",
+ fd);
+ error_code_ret = LTTNG_ERR_FATAL;
+ goto set_cloexec_error;
+ }
+
+ if (event_notifier_rule->filter) {
+ err = kernctl_filter(event_notifier_rule->fd, event_notifier_rule->filter);
+ if (err < 0) {
+ switch (-err) {
+ case ENOMEM:
+ error_code_ret = LTTNG_ERR_FILTER_NOMEM;
+ break;
+ default:
+ error_code_ret = LTTNG_ERR_FILTER_INVAL;
+ break;
+ }
+ goto filter_error;
+ }
+ }
+
+ if (lttng_event_rule_get_type(event_rule) ==
+ LTTNG_EVENT_RULE_TYPE_USERSPACE_PROBE) {
+ ret = userspace_probe_event_rule_add_callsites(
+ event_rule, creds, event_notifier_rule->fd);
+ if (ret) {
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ goto add_callsite_error;
+ }
+ }
+
+ /* Set the capture bytecode if any */
+ cond_status = lttng_condition_on_event_get_capture_descriptor_count(condition, &capture_bytecode_count);
+ assert(cond_status == LTTNG_CONDITION_STATUS_OK);
+ for (i = 0; i < capture_bytecode_count; i++) {
+ const struct lttng_bytecode *capture_bytecode =
+ lttng_condition_on_event_get_capture_bytecode_at_index(
+ condition, i);
+ if (capture_bytecode == NULL) {
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ goto error;
+ }
+
+ ret = kernctl_capture(event_notifier_rule->fd, capture_bytecode);
+ if (ret < 0) {
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ goto error;
+ }
+ }
+
+ err = kernctl_enable(event_notifier_rule->fd);
+ if (err < 0) {
+ switch (-err) {
+ case EEXIST:
+ error_code_ret = LTTNG_ERR_KERN_EVENT_EXIST;
+ break;
+ default:
+ PERROR("enable kernel event notifier");
+ error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ break;
+ }
+ goto enable_error;
+ }
+
+ /* Add trigger to kernel token mapping in the hash table. */
+ rcu_read_lock();
+ cds_lfht_add(kernel_token_to_event_notifier_rule_ht, hash_trigger(trigger),
+ &event_notifier_rule->ht_node);
+ rcu_read_unlock();
+
+ DBG("Created kernel event notifier: name = '%s', fd = %d",
+ kernel_event_notifier.event.name,
+ event_notifier_rule->fd);
+
+ return LTTNG_OK;
+
+add_callsite_error:
+enable_error:
+set_cloexec_error:
+filter_error:
+ {
+ const int close_ret = close(event_notifier_rule->fd);
+
+ if (close_ret) {
+ PERROR("Failed to close kernel event notifier file descriptor: fd = %d",
+ event_notifier_rule->fd);
+ }
+ }
+free_event:
+ free(event_notifier_rule);
+error:
+ return error_code_ret;
+}
+
+enum lttng_error_code kernel_register_event_notifier(
+ struct lttng_trigger *trigger,
+ const struct lttng_credentials *cmd_creds)
+{
+ enum lttng_error_code ret;
+ enum lttng_condition_status status;
+ enum lttng_domain_type domain_type;
+ const struct lttng_event_rule *event_rule;
+ const struct lttng_condition *const condition =
+ lttng_trigger_get_const_condition(trigger);
+ const uint64_t token = lttng_trigger_get_tracer_token(trigger);
+
+ assert(condition);
+
+ /* Does not acquire a reference to the event rule. */
+ status = lttng_condition_on_event_get_rule(
+ condition, &event_rule);
+ assert(status == LTTNG_CONDITION_STATUS_OK);
+
+ domain_type = lttng_event_rule_get_domain_type(event_rule);
+ assert(domain_type == LTTNG_DOMAIN_KERNEL);
+
+ ret = kernel_create_event_notifier_rule(trigger, cmd_creds, token);
+ if (ret != LTTNG_OK) {
+ ERR("Failed to create kernel event notifier rule");
+ }
+
+ return ret;
+}
+
+enum lttng_error_code kernel_unregister_event_notifier(
+ const struct lttng_trigger *trigger)
+{
+ struct ltt_kernel_event_notifier_rule *token_event_rule_element;
+ struct cds_lfht_node *node;
+ struct cds_lfht_iter iter;
+ enum lttng_error_code error_code_ret;
+ int ret;
+
+ rcu_read_lock();
+
+ cds_lfht_lookup(kernel_token_to_event_notifier_rule_ht,
+ hash_trigger(trigger), match_trigger, trigger, &iter);
+
+ node = cds_lfht_iter_get_node(&iter);
+ if (!node) {
+ error_code_ret = LTTNG_ERR_TRIGGER_NOT_FOUND;
+ goto error;
+ }
+
+ token_event_rule_element = caa_container_of(node,
+ struct ltt_kernel_event_notifier_rule, ht_node);
+
+ ret = kernel_disable_event_notifier_rule(token_event_rule_element);
+ if (ret) {
+ error_code_ret = LTTNG_ERR_FATAL;
+ goto error;
+ }
+
+ trace_kernel_destroy_event_notifier_rule(token_event_rule_element);
+ error_code_ret = LTTNG_OK;
+
+error:
+ rcu_read_unlock();
+
+ return error_code_ret;
+}
+
+struct key_ht_entry {
+ char *key;
+ struct lttng_ht_node_str node;
+};
+
+enum lttng_error_code kernel_list_map_values(const struct ltt_kernel_map *map,
+ const struct lttng_map_query *query,
+ struct lttng_map_content **map_content)
+{
+ enum lttng_map_status map_status;
+ enum lttng_error_code ret_code;
+ const char *map_name = NULL;
+ uint64_t descr_count, descr_idx, cpu_idx;
+ struct lttng_map_content *local_map_content;
+ struct lttng_ht *key_ht;
+ struct lttng_ht *values = NULL;
+ struct lttng_ht_node_str *node;
+ struct key_ht_entry *ht_entry;
+ struct lttng_ht_iter iter;
+ enum lttng_map_query_status map_query_status;
+ const char *key_filter;
+ bool sum_cpus = lttng_map_query_get_config_sum_by_cpu(query);
+ enum lttng_map_query_config_cpu config_cpu;
+ int ret;
+ int selected_cpu;
+
+
+ local_map_content = lttng_map_content_create(LTTNG_BUFFER_GLOBAL);
+ if (!local_map_content) {
+ ERR("Error creating map content");
+ ret_code = LTTNG_ERR_NOMEM;
+ goto end;
+ }
+
+ map_query_status = lttng_map_query_get_key_filter(query, &key_filter);
+ if (map_query_status == LTTNG_MAP_QUERY_STATUS_NONE) {
+ key_filter = NULL;
+ } else if (map_query_status != LTTNG_MAP_QUERY_STATUS_OK) {
+ ret_code = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ config_cpu = lttng_map_query_get_config_cpu(query);
+ if (config_cpu == LTTNG_MAP_QUERY_CONFIG_CPU_SUBSET) {
+ unsigned int count;
+ map_query_status = lttng_map_query_get_cpu_count(query, &count);
+ assert(map_query_status == LTTNG_MAP_QUERY_STATUS_OK);
+ assert(count == 1);
+
+ map_query_status = lttng_map_query_get_cpu_at_index(query, 0,
+ &selected_cpu);
+ assert(map_query_status == LTTNG_MAP_QUERY_STATUS_OK);
+ }
+
+ map_status = lttng_map_get_name(map->map, &map_name);
+ assert(map_status == LTTNG_MAP_STATUS_OK);
+
+ DBG("Listing kernel map values: map-name = '%s'", map_name);
+
+ ret = kernctl_counter_map_descriptor_count(map->fd, &descr_count);
+ if (ret) {
+ ERR("Error getting map descriptor count");
+ ret_code = LTTNG_ERR_MAP_VALUES_LIST_FAIL;
+ goto end;
+ }
+
+ /*
+ * The kernel tracer sends us descriptors that may be identical aside
+ * from their user token field. This ABI was design this way to cover a
+ * potential use case where the user wants to know what enabler might
+ * have contributed to a specific bucket.
+ *
+ * We use this hashtable to de-duplicate keys.
+ */
+ if (sum_cpus) {
+ values = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
+ if (!values) {
+ ret_code = LTTNG_ERR_NOMEM;
+ goto end;
+ }
+ }
+
+ DBG("Querying kernel for all map values: "
+ "map-name = '%s', key-value count = %"PRIu64,
+ map_name, descr_count);
+ for (cpu_idx = 0; cpu_idx < utils_get_number_of_possible_cpus(); cpu_idx++) {
+ struct lttng_kernel_counter_read value = {0};
+
+ if (config_cpu == LTTNG_MAP_QUERY_CONFIG_CPU_SUBSET) {
+ if (selected_cpu != cpu_idx) {
+ continue;
+ }
+ }
+
+ if (!sum_cpus) {
+ values = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
+ assert(values);
+ }
+
+ key_ht = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
+ if (!key_ht) {
+ ret_code = LTTNG_ERR_NOMEM;
+ goto end;
+ }
+
+ for(descr_idx = 0; descr_idx < descr_count; descr_idx++) {
+ struct lttng_kernel_counter_map_descriptor descriptor = {0};
+
+ DBG("Querying kernel for map key-value descriptor: "
+ "map-name = '%s', descriptor = %"PRIu64,
+ map_name, descr_idx);
+ descriptor.descriptor_index = descr_idx;
+
+ ret = kernctl_counter_map_descriptor(map->fd, &descriptor);
+ if (ret) {
+ ERR("Error getting map descriptor %"PRIu64, descr_idx);
+ ret_code = LTTNG_ERR_MAP_VALUES_LIST_FAIL;
+ goto end;
+ }
+
+ if (key_filter && strcmp(key_filter, descriptor.key) != 0) {
+ continue;
+ }
+
+ lttng_ht_lookup(key_ht, descriptor.key, &iter);
+ node = lttng_ht_iter_get_node_str(&iter);
+ if (node) {
+ /* This key was already appended to the list. */
+ continue;
+ }
+
+
+ value.index.number_dimensions = 1;
+ value.index.dimension_indexes[0] = descriptor.array_index;
+ value.cpu = cpu_idx;
+
+ DBG("Querying kernel for map descriptor value: "
+ "map-name = '%s', counter-index = %"PRIu64,
+ map_name, descriptor.array_index);
+ ret = kernctl_counter_read_value(map->fd, &value);
+ if (ret) {
+ ERR("Error getting value of map descriptor %"PRIu64, descr_idx);
+ ret_code = LTTNG_ERR_MAP_VALUES_LIST_FAIL;
+ goto end;
+ }
+
+ map_add_or_increment_map_values(values, descriptor.key,
+ value.value.value, value.value.underflow,
+ value.value.overflow);
+
+ ht_entry = zmalloc(sizeof(*ht_entry));
+ assert(ht_entry);
+ ht_entry->key = strdup(descriptor.key);
+ lttng_ht_node_init_str(&ht_entry->node, ht_entry->key);
+ lttng_ht_add_unique_str(key_ht, &ht_entry->node);
+ }
+
+ if (!sum_cpus) {
+ ret = map_new_content_section(local_map_content,
+ LTTNG_MAP_KEY_VALUE_PAIR_LIST_TYPE_KERNEL,
+ sum_cpus, 0,
+ cpu_idx, values);
+ if (ret) {
+ abort();
+ }
+
+ lttng_ht_destroy(values);
+ }
+
+ /*
+ * Remove all the keys before destroying the hashtable.
+ */
+ cds_lfht_for_each_entry(key_ht->ht, &iter.iter, ht_entry, node.node) {
+ struct lttng_ht_iter entry_iter;
+
+ entry_iter.iter.node = &ht_entry->node.node;
+ lttng_ht_del(key_ht, &entry_iter);
+
+ free(ht_entry);
+ }
+
+ lttng_ht_destroy(key_ht);
+ }
+
+ if (sum_cpus) {
+ ret = map_new_content_section(local_map_content,
+ LTTNG_MAP_KEY_VALUE_PAIR_LIST_TYPE_KERNEL,
+ sum_cpus, 0, 0, values);
+ if (ret) {
+ abort();
+ }
+ lttng_ht_destroy(values);
+ }
+
+
+ *map_content = local_map_content;
+ local_map_content = NULL;
+ ret_code = LTTNG_OK;
+
+end:
+ lttng_map_content_destroy(local_map_content);
+ return ret_code;
+}
+
+int kernel_get_notification_fd(void)
+{
+ return kernel_tracer_event_notifier_group_notification_fd;
+}