X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fkernel.c;h=4327db24bdb3291f575da0d90c143e280c7bdbfa;hp=b6b24a5a2b2b426fc78845bb8ca4bf7b1f16770a;hb=d5a1b7aa06b4c924b1cd30623758343c74ecab5c;hpb=53efb85a242809ed5ed21e9ab40effa696ecbc6f

diff --git a/src/bin/lttng-sessiond/kernel.c b/src/bin/lttng-sessiond/kernel.c
index b6b24a5a2..4327db24b 100644
--- a/src/bin/lttng-sessiond/kernel.c
+++ b/src/bin/lttng-sessiond/kernel.c
@@ -15,8 +15,7 @@
  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  */
 
-#define _GNU_SOURCE
-#include <errno.h>
+#define _LGPL_SOURCE
 #include <fcntl.h>
 #include <stdlib.h>
 #include <stdio.h>
@@ -26,18 +25,31 @@
 
 #include <common/common.h>
 #include <common/kernel-ctl/kernel-ctl.h>
+#include <common/kernel-ctl/kernel-ioctl.h>
 #include <common/sessiond-comm/sessiond-comm.h>
 
 #include "consumer.h"
 #include "kernel.h"
 #include "kernel-consumer.h"
 #include "kern-modules.h"
+#include "utils.h"
+#include "rotate.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;
+
+#include <lttng/userspace-probe.h>
+#include <lttng/userspace-probe-internal.h>
 /*
  * Add context on a kernel channel.
+ *
+ * Assumes the ownership of ctx.
  */
 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
-		struct lttng_kernel_context *ctx)
+		struct ltt_kernel_context *ctx)
 {
 	int ret;
 
@@ -45,28 +57,33 @@ int kernel_add_channel_context(struct ltt_kernel_channel *chan,
 	assert(ctx);
 
 	DBG("Adding context to channel %s", chan->channel->name);
-	ret = kernctl_add_context(chan->fd, ctx);
+	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;
-	}
-
-	chan->ctx = zmalloc(sizeof(struct lttng_kernel_context));
-	if (chan->ctx == NULL) {
-		PERROR("zmalloc event context");
-		goto error;
 	}
+	ret = 0;
 
-	memcpy(chan->ctx, ctx, sizeof(struct lttng_kernel_context));
-
-	return 0;
-
+end:
+	cds_list_add_tail(&ctx->list, &chan->ctx_list);
+	ctx->in_list = true;
+	ctx = NULL;
 error:
+	if (ctx) {
+		trace_kernel_destroy_context(ctx);
+	}
 	return ret;
 }
 
@@ -161,8 +178,10 @@ int kernel_create_channel(struct ltt_kernel_session *session,
 	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;
 
@@ -174,61 +193,325 @@ error:
 	return -1;
 }
 
+/*
+ * 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,
+		struct ltt_kernel_session *session, 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, session->uid,
+			session->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,
+		struct ltt_kernel_session *session, 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,
+			session->uid, session->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;
+}
+
+/*
+ * 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)
+{
+	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);
+	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, session, &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);
+			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, session,
+				&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("Adding callsite to userspace probe "
+						"event %s failed.", ev->name);
+				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;
+}
+
 /*
  * 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_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++;
@@ -237,6 +520,17 @@ add_list:
 
 	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:
@@ -255,12 +549,12 @@ int kernel_disable_channel(struct ltt_kernel_channel *chan)
 	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;
 
@@ -278,13 +572,14 @@ int kernel_enable_channel(struct ltt_kernel_channel *chan)
 	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;
 
@@ -303,7 +598,7 @@ int kernel_enable_event(struct ltt_kernel_event *event)
 
 	ret = kernctl_enable(event->fd);
 	if (ret < 0) {
-		switch (errno) {
+		switch (-ret) {
 		case EEXIST:
 			ret = LTTNG_ERR_KERN_EVENT_EXIST;
 			break;
@@ -334,7 +629,7 @@ int kernel_disable_event(struct ltt_kernel_event *event)
 
 	ret = kernctl_disable(event->fd);
 	if (ret < 0) {
-		switch (errno) {
+		switch (-ret) {
 		case EEXIST:
 			ret = LTTNG_ERR_KERN_EVENT_EXIST;
 			break;
@@ -354,6 +649,122 @@ error:
 	return ret;
 }
 
+
+int kernel_track_pid(struct ltt_kernel_session *session, int pid)
+{
+	int ret;
+
+	DBG("Kernel track PID %d for session id %" PRIu64 ".",
+			pid, session->id);
+	ret = kernctl_track_pid(session->fd, pid);
+	if (!ret) {
+		return LTTNG_OK;
+	}
+	switch (-ret) {
+	case EINVAL:
+		return LTTNG_ERR_INVALID;
+	case ENOMEM:
+		return LTTNG_ERR_NOMEM;
+	case EEXIST:
+		return LTTNG_ERR_PID_TRACKED;
+	default:
+		return LTTNG_ERR_UNK;
+	}
+}
+
+int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
+{
+	int ret;
+
+	DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
+			pid, session->id);
+	ret = kernctl_untrack_pid(session->fd, pid);
+	if (!ret) {
+		return LTTNG_OK;
+	}
+	switch (-ret) {
+	case EINVAL:
+		return LTTNG_ERR_INVALID;
+	case ENOMEM:
+		return LTTNG_ERR_NOMEM;
+	case ENOENT:
+		return LTTNG_ERR_PID_NOT_TRACKED;
+	default:
+		return LTTNG_ERR_UNK;
+	}
+}
+
+ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
+		int **_pids)
+{
+	int fd, ret;
+	int pid;
+	ssize_t nbmem, count = 0;
+	FILE *fp;
+	int *pids;
+
+	fd = kernctl_list_tracker_pids(session->fd);
+	if (fd < 0) {
+		PERROR("kernel tracker pids list");
+		goto error;
+	}
+
+	fp = fdopen(fd, "r");
+	if (fp == NULL) {
+		PERROR("kernel tracker pids list fdopen");
+		goto error_fp;
+	}
+
+	nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
+	pids = zmalloc(sizeof(*pids) * nbmem);
+	if (pids == NULL) {
+		PERROR("alloc list pids");
+		count = -ENOMEM;
+		goto end;
+	}
+
+	while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
+		if (count >= nbmem) {
+			int *new_pids;
+			size_t new_nbmem;
+
+			new_nbmem = nbmem << 1;
+			DBG("Reallocating pids list from %zu to %zu entries",
+					nbmem, new_nbmem);
+			new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
+			if (new_pids == NULL) {
+				PERROR("realloc list events");
+				free(pids);
+				count = -ENOMEM;
+				goto end;
+			}
+			/* Zero the new memory */
+			memset(new_pids + nbmem, 0,
+				(new_nbmem - nbmem) * sizeof(*new_pids));
+			nbmem = new_nbmem;
+			pids = new_pids;
+		}
+		pids[count++] = pid;
+	}
+
+	*_pids = pids;
+	DBG("Kernel list tracker pids done (%zd pids)", count);
+end:
+	ret = fclose(fp);	/* closes both fp and fd */
+	if (ret) {
+		PERROR("fclose");
+	}
+	return count;
+
+error_fp:
+	ret = close(fd);
+	if (ret) {
+		PERROR("close");
+	}
+error:
+	return -1;
+}
+
 /*
  * Create kernel metadata, open from the kernel tracer and add it to the
  * kernel session.
@@ -378,6 +789,7 @@ int kernel_open_metadata(struct ltt_kernel_session *session)
 	}
 
 	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) {
@@ -435,25 +847,6 @@ void kernel_wait_quiescent(int 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.
  */
@@ -522,17 +915,22 @@ error:
  * 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) {
@@ -551,13 +949,10 @@ int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 		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);
 	}
@@ -684,32 +1079,45 @@ error:
 /*
  * Get kernel version and validate it.
  */
-int kernel_validate_version(int tracer_fd)
+int kernel_validate_version(int tracer_fd,
+		struct lttng_kernel_tracer_version *version,
+		struct lttng_kernel_tracer_abi_version *abi_version)
 {
 	int ret;
-	struct lttng_kernel_tracer_version version;
 
-	ret = kernctl_tracer_version(tracer_fd, &version);
+	ret = kernctl_tracer_version(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 != KERN_MODULES_PRE_MAJOR
-		&& version.major != KERN_MODULES_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);
 		goto error_version;
 	}
-
-	DBG2("Kernel tracer version validated (major version %d)", version.major);
+	ret = kernctl_tracer_abi_version(tracer_fd, abi_version);
+	if (ret < 0) {
+		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) 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);
 	return 0;
 
 error_version:
-	ERR("Kernel major version %d is not compatible (supporting <= %d)",
-			version.major, KERN_MODULES_MAJOR)
 	ret = -1;
 
 error:
+	ERR("Kernel tracer version check failed; kernel tracing will not be available");
 	return ret;
 }
 
@@ -759,16 +1167,17 @@ void kernel_destroy_session(struct ltt_kernel_session *ksess)
 	DBG("Tearing down kernel session");
 
 	/*
-	 * Destroy channels on the consumer if in no output mode because the
-	 * streams are in *no* monitor mode so we have to send a command to clean
-	 * them up or else they leaked.
+	 * Destroy channels on the consumer if at least one FD has been sent and we
+	 * are in no output mode because the streams are in *no* monitor mode so we
+	 * have to send a command to clean them up or else they leaked.
 	 */
-	if (!ksess->output_traces) {
+	if (!ksess->output_traces && ksess->consumer_fds_sent) {
 		int ret;
 		struct consumer_socket *socket;
 		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;
@@ -782,6 +1191,7 @@ void kernel_destroy_session(struct ltt_kernel_session *ksess)
 				}
 			}
 		}
+		rcu_read_unlock();
 	}
 
 	/* Close any relayd session */
@@ -826,13 +1236,15 @@ void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
  * Return 0 on success or else return a LTTNG_ERR code.
  */
 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
-		struct snapshot_output *output, int wait, unsigned int nb_streams)
+		struct snapshot_output *output, int wait,
+		uint64_t nb_packets_per_stream)
 {
 	int err, ret, saved_metadata_fd;
 	struct consumer_socket *socket;
 	struct lttng_ht_iter iter;
 	struct ltt_kernel_metadata *saved_metadata;
-	uint64_t max_size_per_stream = 0;
+	struct ltt_session *session;
+	uint64_t trace_archive_id;
 
 	assert(ksess);
 	assert(ksess->consumer);
@@ -840,6 +1252,12 @@ int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 
 	DBG("Kernel snapshot record started");
 
+	session = session_find_by_id(ksess->id);
+	assert(session);
+	assert(pthread_mutex_trylock(&session->lock));
+	assert(session_trylock_list());
+	trace_archive_id = session->current_archive_id;
+
 	/* Save current metadata since the following calls will change it. */
 	saved_metadata = ksess->metadata;
 	saved_metadata_fd = ksess->metadata_stream_fd;
@@ -858,10 +1276,6 @@ int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 		goto error_open_stream;
 	}
 
-	if (output->max_size > 0 && nb_streams > 0) {
-		max_size_per_stream = output->max_size / nb_streams;
-	}
-
 	/* Send metadata to consumer and snapshot everything. */
 	cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 			socket, node.node) {
@@ -888,23 +1302,11 @@ int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 
 		/* For each channel, ask the consumer to snapshot it. */
 		cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
-			if (max_size_per_stream &&
-					chan->channel->attr.subbuf_size > max_size_per_stream) {
-				ret = LTTNG_ERR_INVALID;
-				DBG3("Kernel snapshot record maximum stream size %" PRIu64
-						" is smaller than subbuffer size of %" PRIu64,
-						max_size_per_stream, chan->channel->attr.subbuf_size);
-				(void) kernel_consumer_destroy_metadata(socket,
-						ksess->metadata);
-				goto error_consumer;
-			}
-
-			pthread_mutex_lock(socket->lock);
-			ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
+			ret = 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);
+					nb_packets_per_stream,
+					trace_archive_id);
 			if (ret < 0) {
 				ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 				(void) kernel_consumer_destroy_metadata(socket,
@@ -914,11 +1316,10 @@ int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 		}
 
 		/* Snapshot metadata, */
-		pthread_mutex_lock(socket->lock);
-		ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
+		ret = consumer_snapshot_channel(socket, ksess->metadata->key, output,
 				1, ksess->uid, ksess->gid,
-				DEFAULT_KERNEL_TRACE_DIR, wait, max_size_per_stream);
-		pthread_mutex_unlock(socket->lock);
+				DEFAULT_KERNEL_TRACE_DIR, wait, 0,
+				trace_archive_id);
 		if (ret < 0) {
 			ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 			goto error_consumer;
@@ -950,3 +1351,112 @@ error:
 	rcu_read_unlock();
 	return ret;
 }
+
+/*
+ * Get the syscall mask array from the kernel tracer.
+ *
+ * Return 0 on success else a negative value. In both case, syscall_mask should
+ * be freed.
+ */
+int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
+{
+	assert(syscall_mask);
+	assert(nr_bits);
+
+	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(int tracer_fd)
+{
+	int ret = 0; // Not supported by default
+	struct lttng_kernel_tracer_abi_version abi;
+
+	ret = kernctl_tracer_abi_version(tracer_fd, &abi);
+	if (ret < 0) {
+		ERR("Failed to retrieve lttng-modules ABI version");
+		goto error;
+	}
+
+	/*
+	 * 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;
+}
+
+/*
+ * 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,
+					ksess->consumer->subdir,
+					/* is_metadata_channel */ false,
+					session->current_archive_id);
+			if (ret < 0) {
+				status = LTTNG_ERR_KERN_CONSUMER_FAIL;
+				goto error;
+			}
+		}
+
+		/*
+		 * Rotate the metadata channel.
+		 */
+		ret = consumer_rotate_channel(socket, ksess->metadata->key,
+				ksess->uid, ksess->gid, ksess->consumer,
+				ksess->consumer->subdir,
+				/* is_metadata_channel */ true,
+				session->current_archive_id);
+		if (ret < 0) {
+			status = LTTNG_ERR_KERN_CONSUMER_FAIL;
+			goto error;
+		}
+	}
+
+error:
+	rcu_read_unlock();
+	return status;
+}