/*
- * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
+ * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
*
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License, version 2 only,
- * as published by the Free Software Foundation.
+ * SPDX-License-Identifier: GPL-2.0-only
*
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License along
- * with this program; if not, write to the Free Software Foundation, Inc.,
- * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
-#define _GNU_SOURCE
#define _LGPL_SOURCE
-#include <errno.h>
#include <fcntl.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <inttypes.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/event-rule.h>
+#include <lttng/condition/event-rule-internal.h>
+#include <lttng/event-rule/event-rule.h>
+#include <lttng/event-rule/event-rule-internal.h>
+#include <lttng/event-rule/uprobe-internal.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 "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
+ * is only read and updated with the session_list lock held.
+ */
+static uint64_t next_kernel_channel_key;
+
+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;
/*
* Add context on a kernel channel.
+ *
+ * Assumes the ownership of ctx.
*/
int kernel_add_channel_context(struct ltt_kernel_channel *chan,
struct ltt_kernel_context *ctx)
DBG("Adding context to channel %s", chan->channel->name);
ret = kernctl_add_context(chan->fd, &ctx->ctx);
if (ret < 0) {
- if (errno != EEXIST) {
- PERROR("add context ioctl");
- } else {
+ switch (-ret) {
+ case ENOSYS:
+ /* Exists but not available for this kernel */
+ ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
+ goto error;
+ case EEXIST:
/* If EEXIST, we just ignore the error */
ret = 0;
+ goto end;
+ default:
+ PERROR("add context ioctl");
+ ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
+ goto error;
}
- goto error;
}
+ ret = 0;
+end:
cds_list_add_tail(&ctx->list, &chan->ctx_list);
-
- return 0;
-
+ ctx->in_list = true;
+ ctx = NULL;
error:
+ if (ctx) {
+ trace_kernel_destroy_context(ctx);
+ }
return ret;
}
* Create a new kernel session, register it to the kernel tracer and add it to
* the session daemon session.
*/
-int kernel_create_session(struct ltt_session *session, int tracer_fd)
+int kernel_create_session(struct ltt_session *session)
{
int ret;
struct ltt_kernel_session *lks;
}
/* Kernel tracer session creation */
- ret = kernctl_create_session(tracer_fd);
+ ret = kernctl_create_session(kernel_tracer_fd);
if (ret < 0) {
PERROR("ioctl kernel create session");
goto error;
DBG("Kernel session created (fd: %d)", lks->fd);
+ /*
+ * This is necessary since the creation time is present in the session
+ * name when it is generated.
+ */
+ if (session->has_auto_generated_name) {
+ ret = kernctl_session_set_name(lks->fd, DEFAULT_SESSION_NAME);
+ } else {
+ ret = kernctl_session_set_name(lks->fd, session->name);
+ }
+ if (ret) {
+ WARN("Could not set kernel session name for session %" PRIu64 " name: %s",
+ session->id, session->name);
+ }
+
+ ret = kernctl_session_set_creation_time(lks->fd, session->creation_time);
+ if (ret) {
+ WARN("Could not set kernel session creation time for session %" PRIu64 " name: %s",
+ session->id, session->name);
+ }
+
return 0;
error:
if (lks) {
trace_kernel_destroy_session(lks);
+ trace_kernel_free_session(lks);
}
return ret;
}
cds_list_add(&lkc->list, &session->channel_list.head);
session->channel_count++;
lkc->session = session;
+ lkc->key = ++next_kernel_channel_key;
- DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
+ DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
+ lkc->channel->name, lkc->fd, lkc->key);
return 0;
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.
+ *
+ * Returns 0 on success and set the offset out parameter to the offset of the
+ * elf symbol
+ * Returns -1 on error
+ */
+static
+int extract_userspace_probe_offset_function_elf(
+ const struct lttng_userspace_probe_location *probe_location,
+ uid_t uid, gid_t gid, uint64_t *offset)
+{
+ int fd;
+ int ret = 0;
+ const char *symbol = NULL;
+ const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
+ enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
+
+ assert(lttng_userspace_probe_location_get_type(probe_location) ==
+ LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION);
+
+ lookup = lttng_userspace_probe_location_get_lookup_method(
+ probe_location);
+ if (!lookup) {
+ ret = -1;
+ goto end;
+ }
+
+ lookup_method_type =
+ lttng_userspace_probe_location_lookup_method_get_type(lookup);
+
+ assert(lookup_method_type ==
+ LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF);
+
+ symbol = lttng_userspace_probe_location_function_get_function_name(
+ probe_location);
+ if (!symbol) {
+ ret = -1;
+ goto end;
+ }
+
+ fd = lttng_userspace_probe_location_function_get_binary_fd(probe_location);
+ if (fd < 0) {
+ ret = -1;
+ goto end;
+ }
+
+ 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);
+ goto end;
+ }
+
+ DBG("userspace probe elf offset for %s is 0x%jd", symbol, (intmax_t)(*offset));
+end:
+ return ret;
+}
+
+/*
+ * Compute the offsets of the instrumentation bytes in the binary based on the
+ * tracepoint probe location using the SDT lookup method. This function
+ * allocates the offsets buffer, the caller must free it.
+ *
+ * Returns 0 on success and set the offset out parameter to the offsets of the
+ * SDT tracepoint.
+ * Returns -1 on error.
+ */
+static
+int extract_userspace_probe_offset_tracepoint_sdt(
+ const struct lttng_userspace_probe_location *probe_location,
+ uid_t uid, gid_t gid, uint64_t **offsets,
+ uint32_t *offsets_count)
+{
+ enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
+ const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
+ const char *probe_name = NULL, *provider_name = NULL;
+ int ret = 0;
+ int fd, i;
+
+ assert(lttng_userspace_probe_location_get_type(probe_location) ==
+ LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT);
+
+ lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
+ if (!lookup) {
+ ret = -1;
+ goto end;
+ }
+
+ lookup_method_type =
+ lttng_userspace_probe_location_lookup_method_get_type(lookup);
+
+ assert(lookup_method_type ==
+ LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT);
+
+
+ probe_name = lttng_userspace_probe_location_tracepoint_get_probe_name(
+ probe_location);
+ if (!probe_name) {
+ ret = -1;
+ goto end;
+ }
+
+ provider_name = lttng_userspace_probe_location_tracepoint_get_provider_name(
+ probe_location);
+ if (!provider_name) {
+ ret = -1;
+ goto end;
+ }
+
+ fd = lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location);
+ if (fd < 0) {
+ ret = -1;
+ goto end;
+ }
+
+ ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
+ uid, gid, offsets, offsets_count);
+ if (ret < 0) {
+ DBG("userspace probe offset calculation failed for sdt "
+ "probe %s:%s", provider_name, probe_name);
+ goto end;
+ }
+
+ if (*offsets_count == 0) {
+ DBG("no userspace probe offset found");
+ goto end;
+ }
+
+ DBG("%u userspace probe SDT offsets found for %s:%s at:",
+ *offsets_count, provider_name, probe_name);
+ for (i = 0; i < *offsets_count; i++) {
+ DBG("\t0x%jd", (intmax_t)((*offsets)[i]));
+ }
+end:
+ return ret;
+}
+
+static
+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;
+ int ret;
+
+ lookup_method = lttng_userspace_probe_location_get_lookup_method(location);
+ if (!lookup_method) {
+ ret = -1;
+ goto end;
+ }
+
+ type = lttng_userspace_probe_location_lookup_method_get_type(lookup_method);
+ switch (type) {
+ case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
+ {
+ struct lttng_kernel_event_callsite callsite;
+ uint64_t 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("Failed to add callsite to ELF userspace probe.");
+ ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ goto end;
+ }
+ break;
+ }
+ case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
+ {
+ int i;
+ uint64_t *offsets = NULL;
+ uint32_t offsets_count;
+ struct lttng_kernel_event_callsite callsite;
+
+ /*
+ * This call allocates the offsets buffer. This buffer must be freed
+ * by the caller
+ */
+ ret = extract_userspace_probe_offset_tracepoint_sdt(location,
+ uid, gid, &offsets, &offsets_count);
+ if (ret) {
+ ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
+ goto end;
+ }
+ for (i = 0; i < offsets_count; i++) {
+ callsite.u.uprobe.offset = offsets[i];
+ ret = kernctl_add_callsite(fd, &callsite);
+ if (ret) {
+ WARN("Failed to add callsite to SDT userspace probe");
+ ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ free(offsets);
+ goto end;
+ }
+ }
+ free(offsets);
+ break;
+ }
+ default:
+ ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
+ goto end;
+ }
+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_UPROBE);
+
+ status = lttng_event_rule_uprobe_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.
* We own filter_expression and filter.
*/
int kernel_create_event(struct lttng_event *ev,
- struct ltt_kernel_channel *channel)
+ struct ltt_kernel_channel *channel,
+ char *filter_expression,
+ struct lttng_filter_bytecode *filter)
{
- int ret;
+ int err, fd;
+ enum lttng_error_code ret;
struct ltt_kernel_event *event;
assert(ev);
assert(channel);
- event = trace_kernel_create_event(ev);
- if (event == NULL) {
- ret = -1;
+ /* We pass ownership of filter_expression and filter */
+ ret = trace_kernel_create_event(ev, filter_expression,
+ filter, &event);
+ if (ret != LTTNG_OK) {
goto error;
}
- ret = kernctl_create_event(channel->fd, event->event);
- if (ret < 0) {
- switch (errno) {
+ fd = kernctl_create_event(channel->fd, event->event);
+ if (fd < 0) {
+ switch (-fd) {
case EEXIST:
+ ret = LTTNG_ERR_KERN_EVENT_EXIST;
break;
case ENOSYS:
WARN("Event type not implemented");
+ ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
break;
case ENOENT:
WARN("Event %s not found!", ev->name);
+ ret = LTTNG_ERR_KERN_ENABLE_FAIL;
break;
default:
+ ret = LTTNG_ERR_KERN_ENABLE_FAIL;
PERROR("create event ioctl");
}
- ret = -errno;
goto free_event;
}
- /*
- * LTTNG_KERNEL_SYSCALL event creation will return 0 on success.
- */
- if (ret == 0 && event->event->instrumentation == LTTNG_KERNEL_SYSCALL) {
- DBG2("Kernel event syscall creation success");
- /*
- * We use fd == -1 to ensure that we never trigger a close of fd
- * 0.
- */
- event->fd = -1;
- goto add_list;
- }
-
- event->fd = ret;
+ event->type = ev->type;
+ event->fd = fd;
/* Prevent fd duplication after execlp() */
- ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
- if (ret < 0) {
+ err = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
+ if (err < 0) {
PERROR("fcntl session fd");
}
-add_list:
+ if (filter) {
+ err = kernctl_filter(event->fd, filter);
+ if (err < 0) {
+ switch (-err) {
+ case ENOMEM:
+ ret = LTTNG_ERR_FILTER_NOMEM;
+ break;
+ default:
+ ret = LTTNG_ERR_FILTER_INVAL;
+ break;
+ }
+ goto filter_error;
+ }
+ }
+
+ if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
+ ret = userspace_probe_event_add_callsites(ev, channel->session,
+ event->fd);
+ if (ret) {
+ goto add_callsite_error;
+ }
+ }
+
+ err = kernctl_enable(event->fd);
+ if (err < 0) {
+ switch (-err) {
+ case EEXIST:
+ ret = LTTNG_ERR_KERN_EVENT_EXIST;
+ break;
+ default:
+ PERROR("enable kernel event");
+ ret = LTTNG_ERR_KERN_ENABLE_FAIL;
+ break;
+ }
+ goto enable_error;
+ }
+
/* Add event to event list */
cds_list_add(&event->list, &channel->events_list.head);
channel->event_count++;
return 0;
+add_callsite_error:
+enable_error:
+filter_error:
+ {
+ int closeret;
+
+ closeret = close(event->fd);
+ if (closeret) {
+ PERROR("close event fd");
+ }
+ }
free_event:
free(event);
error:
ret = kernctl_disable(chan->fd);
if (ret < 0) {
PERROR("disable chan ioctl");
- ret = errno;
goto error;
}
chan->enabled = 0;
- DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
+ DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
+ chan->channel->name, chan->fd, chan->key);
return 0;
assert(chan);
ret = kernctl_enable(chan->fd);
- if (ret < 0 && errno != EEXIST) {
+ if (ret < 0 && ret != -EEXIST) {
PERROR("Enable kernel chan");
goto error;
}
chan->enabled = 1;
- DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
+ DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
+ chan->channel->name, chan->fd, chan->key);
return 0;
ret = kernctl_enable(event->fd);
if (ret < 0) {
- switch (errno) {
+ switch (-ret) {
case EEXIST:
ret = LTTNG_ERR_KERN_EVENT_EXIST;
break;
ret = kernctl_disable(event->fd);
if (ret < 0) {
- switch (errno) {
+ switch (-ret) {
case EEXIST:
ret = LTTNG_ERR_KERN_EVENT_EXIST;
break;
return ret;
}
-int kernel_enable_syscall(const char *syscall_name,
- struct ltt_kernel_channel *channel)
-{
- return kernctl_enable_syscall(channel->fd, syscall_name);
-}
-
-int kernel_disable_syscall(const char *syscall_name,
- struct ltt_kernel_channel *channel)
-{
- return kernctl_disable_syscall(channel->fd, syscall_name);
-}
-
/*
- * Create kernel metadata, open from the kernel tracer and add it to the
- * kernel session.
+ * Disable a kernel event notifier.
*/
-int kernel_open_metadata(struct ltt_kernel_session *session)
+static
+int kernel_disable_event_notifier_rule(struct ltt_kernel_event_notifier_rule *event)
{
int ret;
- struct ltt_kernel_metadata *lkm = NULL;
- assert(session);
+ assert(event);
- /* Allocate kernel metadata */
- lkm = trace_kernel_create_metadata();
- if (lkm == NULL) {
- goto error;
- }
+ rcu_read_lock();
+ cds_lfht_del(kernel_token_to_event_notifier_rule_ht, &event->ht_node);
+ rcu_read_unlock();
- /* Kernel tracer metadata creation */
- ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
+ ret = kernctl_disable(event->fd);
if (ret < 0) {
- goto error_open;
+ 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;
}
- lkm->fd = ret;
- /* Prevent fd duplication after execlp() */
- ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
- if (ret < 0) {
- PERROR("fcntl session fd");
- }
+ event->enabled = 0;
+ DBG("Disabled kernel event notifier: fd = %d, token = %" PRIu64,
+ event->fd, event->token);
- session->metadata = lkm;
+error:
+ return ret;
+}
- DBG("Kernel metadata opened (fd: %d)", lkm->fd);
+static
+struct process_attr_tracker *_kernel_get_process_attr_tracker(
+ struct ltt_kernel_session *session,
+ enum lttng_process_attr process_attr)
+{
+ switch (process_attr) {
+ case LTTNG_PROCESS_ATTR_PROCESS_ID:
+ return session->tracker_pid;
+ case LTTNG_PROCESS_ATTR_VIRTUAL_PROCESS_ID:
+ return session->tracker_vpid;
+ case LTTNG_PROCESS_ATTR_USER_ID:
+ return session->tracker_uid;
+ case LTTNG_PROCESS_ATTR_VIRTUAL_USER_ID:
+ return session->tracker_vuid;
+ case LTTNG_PROCESS_ATTR_GROUP_ID:
+ return session->tracker_gid;
+ case LTTNG_PROCESS_ATTR_VIRTUAL_GROUP_ID:
+ return session->tracker_vgid;
+ default:
+ return NULL;
+ }
+}
+
+const struct process_attr_tracker *kernel_get_process_attr_tracker(
+ struct ltt_kernel_session *session,
+ enum lttng_process_attr process_attr)
+{
+ return (const struct process_attr_tracker *)
+ _kernel_get_process_attr_tracker(session, process_attr);
+}
+
+enum lttng_error_code kernel_process_attr_tracker_set_tracking_policy(
+ struct ltt_kernel_session *session,
+ enum lttng_process_attr process_attr,
+ enum lttng_tracking_policy policy)
+{
+ int ret;
+ enum lttng_error_code ret_code = LTTNG_OK;
+ struct process_attr_tracker *tracker =
+ _kernel_get_process_attr_tracker(session, process_attr);
+ enum lttng_tracking_policy previous_policy;
+
+ if (!tracker) {
+ ret_code = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ previous_policy = process_attr_tracker_get_tracking_policy(tracker);
+ ret = process_attr_tracker_set_tracking_policy(tracker, policy);
+ if (ret) {
+ ret_code = LTTNG_ERR_UNK;
+ goto end;
+ }
+
+ if (previous_policy == policy) {
+ goto end;
+ }
+
+ switch (policy) {
+ case LTTNG_TRACKING_POLICY_INCLUDE_ALL:
+ if (process_attr == LTTNG_PROCESS_ATTR_PROCESS_ID) {
+ /*
+ * Maintain a special case for the process ID process
+ * attribute tracker as it was the only supported
+ * attribute prior to 2.12.
+ */
+ ret = kernctl_track_pid(session->fd, -1);
+ } else {
+ ret = kernctl_track_id(session->fd, process_attr, -1);
+ }
+ break;
+ case LTTNG_TRACKING_POLICY_EXCLUDE_ALL:
+ case LTTNG_TRACKING_POLICY_INCLUDE_SET:
+ /* fall-through. */
+ if (process_attr == LTTNG_PROCESS_ATTR_PROCESS_ID) {
+ /*
+ * Maintain a special case for the process ID process
+ * attribute tracker as it was the only supported
+ * attribute prior to 2.12.
+ */
+ ret = kernctl_untrack_pid(session->fd, -1);
+ } else {
+ ret = kernctl_untrack_id(session->fd, process_attr, -1);
+ }
+ break;
+ default:
+ abort();
+ }
+ /* kern-ctl error handling */
+ switch (-ret) {
+ case 0:
+ ret_code = LTTNG_OK;
+ break;
+ case EINVAL:
+ ret_code = LTTNG_ERR_INVALID;
+ break;
+ case ENOMEM:
+ ret_code = LTTNG_ERR_NOMEM;
+ break;
+ case EEXIST:
+ ret_code = LTTNG_ERR_PROCESS_ATTR_EXISTS;
+ break;
+ default:
+ ret_code = LTTNG_ERR_UNK;
+ break;
+ }
+end:
+ return ret_code;
+}
+
+enum lttng_error_code kernel_process_attr_tracker_inclusion_set_add_value(
+ struct ltt_kernel_session *session,
+ enum lttng_process_attr process_attr,
+ const struct process_attr_value *value)
+{
+ int ret, integral_value;
+ enum lttng_error_code ret_code;
+ struct process_attr_tracker *tracker;
+ enum process_attr_tracker_status status;
+
+ /*
+ * Convert process attribute tracker value to the integral
+ * representation required by the kern-ctl API.
+ */
+ switch (process_attr) {
+ case LTTNG_PROCESS_ATTR_PROCESS_ID:
+ case LTTNG_PROCESS_ATTR_VIRTUAL_PROCESS_ID:
+ integral_value = (int) value->value.pid;
+ break;
+ case LTTNG_PROCESS_ATTR_USER_ID:
+ case LTTNG_PROCESS_ATTR_VIRTUAL_USER_ID:
+ if (value->type == LTTNG_PROCESS_ATTR_VALUE_TYPE_USER_NAME) {
+ uid_t uid;
+
+ ret_code = utils_user_id_from_name(
+ value->value.user_name, &uid);
+ if (ret_code != LTTNG_OK) {
+ goto end;
+ }
+ integral_value = (int) uid;
+ } else {
+ integral_value = (int) value->value.uid;
+ }
+ break;
+ case LTTNG_PROCESS_ATTR_GROUP_ID:
+ case LTTNG_PROCESS_ATTR_VIRTUAL_GROUP_ID:
+ if (value->type == LTTNG_PROCESS_ATTR_VALUE_TYPE_GROUP_NAME) {
+ gid_t gid;
+
+ ret_code = utils_group_id_from_name(
+ value->value.group_name, &gid);
+ if (ret_code != LTTNG_OK) {
+ goto end;
+ }
+ integral_value = (int) gid;
+ } else {
+ integral_value = (int) value->value.gid;
+ }
+ break;
+ default:
+ ret_code = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ tracker = _kernel_get_process_attr_tracker(session, process_attr);
+ if (!tracker) {
+ ret_code = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ status = process_attr_tracker_inclusion_set_add_value(tracker, value);
+ if (status != PROCESS_ATTR_TRACKER_STATUS_OK) {
+ switch (status) {
+ case PROCESS_ATTR_TRACKER_STATUS_EXISTS:
+ ret_code = LTTNG_ERR_PROCESS_ATTR_EXISTS;
+ break;
+ case PROCESS_ATTR_TRACKER_STATUS_INVALID_TRACKING_POLICY:
+ ret_code = LTTNG_ERR_PROCESS_ATTR_TRACKER_INVALID_TRACKING_POLICY;
+ break;
+ case PROCESS_ATTR_TRACKER_STATUS_ERROR:
+ default:
+ ret_code = LTTNG_ERR_UNK;
+ break;
+ }
+ goto end;
+ }
+
+ DBG("Kernel track %s %d for session id %" PRIu64,
+ lttng_process_attr_to_string(process_attr),
+ integral_value, session->id);
+ if (process_attr == LTTNG_PROCESS_ATTR_PROCESS_ID) {
+ /*
+ * Maintain a special case for the process ID process attribute
+ * tracker as it was the only supported attribute prior to 2.12.
+ */
+ ret = kernctl_track_pid(session->fd, integral_value);
+ } else {
+ ret = kernctl_track_id(
+ session->fd, process_attr, integral_value);
+ }
+ if (ret == 0) {
+ ret_code = LTTNG_OK;
+ goto end;
+ }
+
+ kernel_wait_quiescent();
+
+ /* kern-ctl error handling */
+ switch (-ret) {
+ case 0:
+ ret_code = LTTNG_OK;
+ break;
+ case EINVAL:
+ ret_code = LTTNG_ERR_INVALID;
+ break;
+ case ENOMEM:
+ ret_code = LTTNG_ERR_NOMEM;
+ break;
+ case EEXIST:
+ ret_code = LTTNG_ERR_PROCESS_ATTR_EXISTS;
+ break;
+ default:
+ ret_code = LTTNG_ERR_UNK;
+ break;
+ }
+
+ /* Attempt to remove the value from the tracker. */
+ status = process_attr_tracker_inclusion_set_remove_value(
+ tracker, value);
+ if (status != PROCESS_ATTR_TRACKER_STATUS_OK) {
+ ERR("Failed to roll-back the tracking of kernel %s process attribute %d while handling a kern-ctl error",
+ lttng_process_attr_to_string(process_attr),
+ integral_value);
+ }
+end:
+ return ret_code;
+}
+
+enum lttng_error_code kernel_process_attr_tracker_inclusion_set_remove_value(
+ struct ltt_kernel_session *session,
+ enum lttng_process_attr process_attr,
+ const struct process_attr_value *value)
+{
+ int ret, integral_value;
+ enum lttng_error_code ret_code;
+ struct process_attr_tracker *tracker;
+ enum process_attr_tracker_status status;
+
+ /*
+ * Convert process attribute tracker value to the integral
+ * representation required by the kern-ctl API.
+ */
+ switch (process_attr) {
+ case LTTNG_PROCESS_ATTR_PROCESS_ID:
+ case LTTNG_PROCESS_ATTR_VIRTUAL_PROCESS_ID:
+ integral_value = (int) value->value.pid;
+ break;
+ case LTTNG_PROCESS_ATTR_USER_ID:
+ case LTTNG_PROCESS_ATTR_VIRTUAL_USER_ID:
+ if (value->type == LTTNG_PROCESS_ATTR_VALUE_TYPE_USER_NAME) {
+ uid_t uid;
+
+ ret_code = utils_user_id_from_name(
+ value->value.user_name, &uid);
+ if (ret_code != LTTNG_OK) {
+ goto end;
+ }
+ integral_value = (int) uid;
+ } else {
+ integral_value = (int) value->value.uid;
+ }
+ break;
+ case LTTNG_PROCESS_ATTR_GROUP_ID:
+ case LTTNG_PROCESS_ATTR_VIRTUAL_GROUP_ID:
+ if (value->type == LTTNG_PROCESS_ATTR_VALUE_TYPE_GROUP_NAME) {
+ gid_t gid;
+
+ ret_code = utils_group_id_from_name(
+ value->value.group_name, &gid);
+ if (ret_code != LTTNG_OK) {
+ goto end;
+ }
+ integral_value = (int) gid;
+ } else {
+ integral_value = (int) value->value.gid;
+ }
+ break;
+ default:
+ ret_code = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ tracker = _kernel_get_process_attr_tracker(session, process_attr);
+ if (!tracker) {
+ ret_code = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ status = process_attr_tracker_inclusion_set_remove_value(
+ tracker, value);
+ if (status != PROCESS_ATTR_TRACKER_STATUS_OK) {
+ switch (status) {
+ case PROCESS_ATTR_TRACKER_STATUS_MISSING:
+ ret_code = LTTNG_ERR_PROCESS_ATTR_MISSING;
+ break;
+ case PROCESS_ATTR_TRACKER_STATUS_INVALID_TRACKING_POLICY:
+ ret_code = LTTNG_ERR_PROCESS_ATTR_TRACKER_INVALID_TRACKING_POLICY;
+ break;
+ case PROCESS_ATTR_TRACKER_STATUS_ERROR:
+ default:
+ ret_code = LTTNG_ERR_UNK;
+ break;
+ }
+ goto end;
+ }
+
+ DBG("Kernel track %s %d for session id %" PRIu64,
+ lttng_process_attr_to_string(process_attr),
+ integral_value, session->id);
+ if (process_attr == LTTNG_PROCESS_ATTR_PROCESS_ID) {
+ /*
+ * Maintain a special case for the process ID process attribute
+ * tracker as it was the only supported attribute prior to 2.12.
+ */
+ ret = kernctl_untrack_pid(session->fd, integral_value);
+ } else {
+ ret = kernctl_untrack_id(
+ session->fd, process_attr, integral_value);
+ }
+ if (ret == 0) {
+ ret_code = LTTNG_OK;
+ goto end;
+ }
+ kernel_wait_quiescent();
+
+ /* kern-ctl error handling */
+ switch (-ret) {
+ case 0:
+ ret_code = LTTNG_OK;
+ break;
+ case EINVAL:
+ ret_code = LTTNG_ERR_INVALID;
+ break;
+ case ENOMEM:
+ ret_code = LTTNG_ERR_NOMEM;
+ break;
+ case ENOENT:
+ ret_code = LTTNG_ERR_PROCESS_ATTR_MISSING;
+ break;
+ default:
+ ret_code = LTTNG_ERR_UNK;
+ break;
+ }
+
+ /* Attempt to add the value to the tracker. */
+ status = process_attr_tracker_inclusion_set_add_value(
+ tracker, value);
+ if (status != PROCESS_ATTR_TRACKER_STATUS_OK) {
+ ERR("Failed to roll-back the tracking of kernel %s process attribute %d while handling a kern-ctl error",
+ lttng_process_attr_to_string(process_attr),
+ integral_value);
+ }
+end:
+ return ret_code;
+}
+
+/*
+ * Create kernel metadata, open from the kernel tracer and add it to the
+ * kernel session.
+ */
+int kernel_open_metadata(struct ltt_kernel_session *session)
+{
+ int ret;
+ struct ltt_kernel_metadata *lkm = NULL;
+
+ assert(session);
+
+ /* Allocate kernel metadata */
+ lkm = trace_kernel_create_metadata();
+ if (lkm == NULL) {
+ goto error;
+ }
+
+ /* Kernel tracer metadata creation */
+ ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
+ if (ret < 0) {
+ goto error_open;
+ }
+
+ lkm->fd = ret;
+ lkm->key = ++next_kernel_channel_key;
+ /* Prevent fd duplication after execlp() */
+ ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
+ if (ret < 0) {
+ PERROR("fcntl session fd");
+ }
+
+ session->metadata = lkm;
+
+ DBG("Kernel metadata opened (fd: %d)", lkm->fd);
return 0;
/*
* Make a kernel wait to make sure in-flight probe have completed.
*/
-void kernel_wait_quiescent(int fd)
+void kernel_wait_quiescent(void)
{
int ret;
+ int fd = kernel_tracer_fd;
DBG("Kernel quiescent wait on %d", fd);
}
}
-/*
- * Kernel calibrate
- */
-int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
-{
- int ret;
-
- assert(calibrate);
-
- ret = kernctl_calibrate(fd, calibrate);
- if (ret < 0) {
- PERROR("calibrate ioctl");
- return -1;
- }
-
- return 0;
-}
-
-
/*
* Force flush buffer of metadata.
*/
* Open stream of channel, register it to the kernel tracer and add it
* to the stream list of the channel.
*
+ * Note: given that the streams may appear in random order wrt CPU
+ * number (e.g. cpu hotplug), the index value of the stream number in
+ * the stream name is not necessarily linked to the CPU number.
+ *
* Return the number of created stream. Else, a negative value.
*/
int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
{
- int ret, count = 0;
+ int ret;
struct ltt_kernel_stream *lks;
assert(channel);
while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
- lks = trace_kernel_create_stream(channel->channel->name, count);
+ lks = trace_kernel_create_stream(channel->channel->name,
+ channel->stream_count);
if (lks == NULL) {
ret = close(ret);
if (ret) {
lks->tracefile_size = channel->channel->attr.tracefile_size;
lks->tracefile_count = channel->channel->attr.tracefile_count;
- /* Add stream to channe stream list */
+ /* Add stream to channel stream list */
cds_list_add(&lks->list, &channel->stream_list.head);
channel->stream_count++;
- /* Increment counter which represent CPU number. */
- count++;
-
DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
lks->state);
}
/*
* Get the event list from the kernel tracer and return the number of elements.
*/
-ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
+ssize_t kernel_list_events(struct lttng_event **events)
{
int fd, ret;
char *event;
assert(events);
- fd = kernctl_tracepoint_list(tracer_fd);
+ fd = kernctl_tracepoint_list(kernel_tracer_fd);
if (fd < 0) {
PERROR("kernel tracepoint list");
goto error;
/*
* Get kernel version and validate it.
*/
-int kernel_validate_version(int tracer_fd)
+int kernel_validate_version(struct lttng_kernel_tracer_version *version,
+ struct lttng_kernel_tracer_abi_version *abi_version)
{
int ret;
- struct lttng_kernel_tracer_version version;
- struct lttng_kernel_tracer_abi_version abi_version;
- ret = kernctl_tracer_version(tracer_fd, &version);
+ ret = kernctl_tracer_version(kernel_tracer_fd, version);
if (ret < 0) {
- ERR("Failed at getting the lttng-modules version");
+ ERR("Failed to retrieve the lttng-modules version");
goto error;
}
/* Validate version */
- if (version.major != VERSION_MAJOR) {
+ if (version->major != VERSION_MAJOR) {
ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
- version.major, VERSION_MAJOR);
+ version->major, VERSION_MAJOR);
goto error_version;
}
- ret = kernctl_tracer_abi_version(tracer_fd, &abi_version);
+ ret = kernctl_tracer_abi_version(kernel_tracer_fd, abi_version);
if (ret < 0) {
- ERR("Failed at getting lttng-modules ABI version");
+ ERR("Failed to retrieve lttng-modules ABI version");
goto error;
}
- if (abi_version.major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
- ERR("Kernel tracer ABI version (%d.%d) is not compatible with expected ABI major version (%d.*)",
- abi_version.major, abi_version.minor,
+ if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
+ ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
+ abi_version->major, abi_version->minor,
LTTNG_MODULES_ABI_MAJOR_VERSION);
goto error;
}
DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
- version.major, version.minor,
- abi_version.major, abi_version.minor);
+ version->major, version->minor,
+ abi_version->major, abi_version->minor);
return 0;
error_version:
ret = -1;
error:
+ ERR("Kernel tracer version check failed; kernel tracing will not be available");
return ret;
}
}
/*
- * Complete teardown of a kernel session.
+ * Teardown of a kernel session, keeping data required by destroy notifiers.
*/
void kernel_destroy_session(struct ltt_kernel_session *ksess)
{
+ struct lttng_trace_chunk *trace_chunk;
+
if (ksess == NULL) {
DBG3("No kernel session when tearing down session");
return;
}
DBG("Tearing down kernel session");
+ trace_chunk = ksess->current_trace_chunk;
/*
* Destroy channels on the consumer if at least one FD has been sent and we
struct lttng_ht_iter iter;
/* For each consumer socket. */
+ rcu_read_lock();
cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
socket, node.node) {
struct ltt_kernel_channel *chan;
}
}
}
+ rcu_read_unlock();
}
/* Close any relayd session */
consumer_output_send_destroy_relayd(ksess->consumer);
trace_kernel_destroy_session(ksess);
+ lttng_trace_chunk_put(trace_chunk);
+}
+
+/* Teardown of data required by destroy notifiers. */
+void kernel_free_session(struct ltt_kernel_session *ksess)
+{
+ if (ksess == NULL) {
+ return;
+ }
+ trace_kernel_free_session(ksess);
}
/*
/*
* Take a snapshot for a given kernel session.
*
- * Return 0 on success or else return a LTTNG_ERR code.
+ * Return LTTNG_OK on success or else return a LTTNG_ERR code.
*/
-int kernel_snapshot_record(struct ltt_kernel_session *ksess,
- struct snapshot_output *output, int wait, uint64_t max_size_per_stream)
+enum lttng_error_code kernel_snapshot_record(
+ struct ltt_kernel_session *ksess,
+ const struct consumer_output *output, int wait,
+ uint64_t nb_packets_per_stream)
{
int err, ret, saved_metadata_fd;
+ enum lttng_error_code status = LTTNG_OK;
struct consumer_socket *socket;
struct lttng_ht_iter iter;
struct ltt_kernel_metadata *saved_metadata;
+ char *trace_path = NULL;
+ size_t consumer_path_offset = 0;
assert(ksess);
assert(ksess->consumer);
ret = kernel_open_metadata(ksess);
if (ret < 0) {
- ret = LTTNG_ERR_KERN_META_FAIL;
+ status = LTTNG_ERR_KERN_META_FAIL;
goto error;
}
ret = kernel_open_metadata_stream(ksess);
if (ret < 0) {
- ret = LTTNG_ERR_KERN_META_FAIL;
+ status = LTTNG_ERR_KERN_META_FAIL;
goto error_open_stream;
}
+ trace_path = setup_channel_trace_path(ksess->consumer,
+ DEFAULT_KERNEL_TRACE_DIR, &consumer_path_offset);
+ if (!trace_path) {
+ status = LTTNG_ERR_INVALID;
+ goto error;
+ }
/* Send metadata to consumer and snapshot everything. */
- cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
+ cds_lfht_for_each_entry(output->socks->ht, &iter.iter,
socket, node.node) {
- struct consumer_output *saved_output;
struct ltt_kernel_channel *chan;
- /*
- * Temporarly switch consumer output for our snapshot output. As long
- * as the session lock is taken, this is safe.
- */
- saved_output = ksess->consumer;
- ksess->consumer = output->consumer;
-
pthread_mutex_lock(socket->lock);
/* This stream must not be monitored by the consumer. */
ret = kernel_consumer_add_metadata(socket, ksess, 0);
pthread_mutex_unlock(socket->lock);
- /* Put back the saved consumer output into the session. */
- ksess->consumer = saved_output;
if (ret < 0) {
- ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
+ status = LTTNG_ERR_KERN_META_FAIL;
goto error_consumer;
}
/* For each channel, ask the consumer to snapshot it. */
cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
- pthread_mutex_lock(socket->lock);
- ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
+ status = consumer_snapshot_channel(socket, chan->key, output, 0,
ksess->uid, ksess->gid,
- DEFAULT_KERNEL_TRACE_DIR, wait,
- max_size_per_stream);
- pthread_mutex_unlock(socket->lock);
- if (ret < 0) {
- ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
+ &trace_path[consumer_path_offset], wait,
+ nb_packets_per_stream);
+ if (status != LTTNG_OK) {
(void) kernel_consumer_destroy_metadata(socket,
ksess->metadata);
goto error_consumer;
}
/* Snapshot metadata, */
- pthread_mutex_lock(socket->lock);
- ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
- 1, ksess->uid, ksess->gid,
- DEFAULT_KERNEL_TRACE_DIR, wait, max_size_per_stream);
- pthread_mutex_unlock(socket->lock);
- if (ret < 0) {
- ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
+ status = consumer_snapshot_channel(socket, ksess->metadata->key, output,
+ 1, ksess->uid, ksess->gid, &trace_path[consumer_path_offset],
+ wait, 0);
+ if (status != LTTNG_OK) {
goto error_consumer;
}
(void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
}
- ret = LTTNG_OK;
-
error_consumer:
/* Close newly opened metadata stream. It's now on the consumer side. */
err = close(ksess->metadata_stream_fd);
/* Restore metadata state.*/
ksess->metadata = saved_metadata;
ksess->metadata_stream_fd = saved_metadata_fd;
-
rcu_read_unlock();
- return ret;
+ free(trace_path);
+ return status;
}
/*
return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
}
+
+static
+int kernel_tracer_abi_greater_or_equal(unsigned int major, unsigned int minor)
+{
+ int ret;
+ 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;
+ }
+
+ 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
+ */
+ return kernel_tracer_abi_greater_or_equal(2, 3);
+}
+
+/*
+ * Check for the support of the packet sequence number 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_packet_sequence_number(void)
+{
+ /*
+ * Packet sequence number was introduced in LTTng 2.8,
+ * lttng-modules ABI 2.1.
+ */
+ 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);
+}
+
+/*
+ * Rotate a kernel session.
+ *
+ * Return LTTNG_OK on success or else an LTTng error code.
+ */
+enum lttng_error_code kernel_rotate_session(struct ltt_session *session)
+{
+ int ret;
+ enum lttng_error_code status = LTTNG_OK;
+ struct consumer_socket *socket;
+ struct lttng_ht_iter iter;
+ struct ltt_kernel_session *ksess = session->kernel_session;
+
+ assert(ksess);
+ assert(ksess->consumer);
+
+ DBG("Rotate kernel session %s started (session %" PRIu64 ")",
+ session->name, session->id);
+
+ rcu_read_lock();
+
+ /*
+ * Note that this loop will end after one iteration given that there is
+ * only one kernel consumer.
+ */
+ cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
+ socket, node.node) {
+ struct ltt_kernel_channel *chan;
+
+ /* 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);
+ ret = consumer_rotate_channel(socket, chan->key,
+ ksess->uid, ksess->gid, ksess->consumer,
+ /* is_metadata_channel */ false);
+ if (ret < 0) {
+ status = LTTNG_ERR_ROTATION_FAIL_CONSUMER;
+ goto error;
+ }
+ }
+
+ /*
+ * Rotate the metadata channel.
+ */
+ ret = consumer_rotate_channel(socket, ksess->metadata->key,
+ ksess->uid, ksess->gid, ksess->consumer,
+ /* is_metadata_channel */ true);
+ if (ret < 0) {
+ status = LTTNG_ERR_ROTATION_FAIL_CONSUMER;
+ goto error;
+ }
+ }
+
+error:
+ rcu_read_unlock();
+ return status;
+}
+
+enum lttng_error_code kernel_create_channel_subdirectories(
+ const struct ltt_kernel_session *ksess)
+{
+ enum lttng_error_code ret = LTTNG_OK;
+ enum lttng_trace_chunk_status chunk_status;
+
+ rcu_read_lock();
+ assert(ksess->current_trace_chunk);
+
+ /*
+ * Create the index subdirectory which will take care
+ * of implicitly creating the channel's path.
+ */
+ chunk_status = lttng_trace_chunk_create_subdirectory(
+ ksess->current_trace_chunk,
+ DEFAULT_KERNEL_TRACE_DIR "/" DEFAULT_INDEX_DIR);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ret = LTTNG_ERR_CREATE_DIR_FAIL;
+ goto error;
+ }
+error:
+ rcu_read_unlock();
+ return ret;
+}
+
+/*
+ * Setup necessary data for kernel tracer action.
+ */
+LTTNG_HIDDEN
+int init_kernel_tracer(void)
+{
+ int ret;
+ bool is_root = !getuid();
+
+ /* Modprobe lttng kernel modules */
+ ret = modprobe_lttng_control();
+ if (ret < 0) {
+ goto error;
+ }
+
+ /* Open debugfs lttng */
+ kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
+ if (kernel_tracer_fd < 0) {
+ DBG("Failed to open %s", module_proc_lttng);
+ goto error_open;
+ }
+
+ /* Validate kernel version */
+ ret = kernel_validate_version(&kernel_tracer_version,
+ &kernel_tracer_abi_version);
+ if (ret < 0) {
+ goto error_version;
+ }
+
+ ret = modprobe_lttng_data();
+ if (ret < 0) {
+ goto error_modules;
+ }
+
+ ret = kernel_supports_ring_buffer_snapshot_sample_positions();
+ 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).");
+ }
+
+ 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 {
+ const 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;
+ }
+
+ 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("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("Failed to close kernel tracer file descriptor: fd = %d",
+ kernel_tracer_fd);
+ }
+
+error_open:
+ modprobe_remove_lttng_control();
+
+error:
+ WARN("No kernel tracer available");
+ kernel_tracer_fd = -1;
+ if (!is_root) {
+ return LTTNG_ERR_NEED_ROOT_SESSIOND;
+ } else {
+ return LTTNG_ERR_KERN_NA;
+ }
+}
+
+LTTNG_HIDDEN
+void cleanup_kernel_tracer(void)
+{
+ 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) {
+ const int ret = close(kernel_tracer_fd);
+
+ if (ret) {
+ 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);
+}
+
+LTTNG_HIDDEN
+bool kernel_tracer_is_initialized(void)
+{
+ 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)
+{
+ int ret;
+ enum lttng_error_code status = LTTNG_OK;
+ struct consumer_socket *socket;
+ struct lttng_ht_iter iter;
+ 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;
+ }
+
+ /*
+ * Note that this loop will end after one iteration given that there is
+ * only one kernel consumer.
+ */
+ cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
+ socket, node.node) {
+ struct ltt_kernel_channel *chan;
+
+ /* For each channel, ask the consumer to clear it. */
+ cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
+ DBG("Clear kernel channel %" PRIu64 ", session %s",
+ chan->key, session->name);
+ ret = consumer_clear_channel(socket, chan->key);
+ if (ret < 0) {
+ goto error;
+ }
+ }
+
+ if (!ksess->metadata) {
+ /*
+ * Nothing to do for the metadata.
+ * This is a snapshot session.
+ * The metadata is genererated on the fly.
+ */
+ continue;
+ }
+
+ /*
+ * Clear the metadata channel.
+ * Metadata channel is not cleared per se but we still need to
+ * perform a rotation operation on it behind the scene.
+ */
+ ret = consumer_clear_channel(socket, ksess->metadata->key);
+ if (ret < 0) {
+ goto error;
+ }
+ }
+
+ goto end;
+error:
+ switch (-ret) {
+ case LTTCOMM_CONSUMERD_RELAYD_CLEAR_DISALLOWED:
+ status = LTTNG_ERR_CLEAR_RELAY_DISALLOWED;
+ break;
+ default:
+ status = LTTNG_ERR_CLEAR_FAIL_CONSUMER;
+ break;
+ }
+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 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 = {};
+ const struct lttng_condition *condition = NULL;
+ const struct lttng_event_rule *event_rule = NULL;
+
+ assert(trigger);
+
+ condition = lttng_trigger_get_const_condition(trigger);
+ assert(condition);
+
+ condition_type = lttng_condition_get_type(condition);
+ assert(condition_type == LTTNG_CONDITION_TYPE_EVENT_RULE_HIT);
+
+ /* Does not acquire a reference. */
+ condition_status = lttng_condition_event_rule_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,
+ &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;
+
+ 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_UPROBE) {
+ 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;
+ }
+ }
+
+ 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_event_rule_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;
+}
+
+int kernel_get_notification_fd(void)
+{
+ return kernel_tracer_event_notifier_group_notification_fd;
+}