*/
#define _LGPL_SOURCE
-#include <errno.h>
#include <fcntl.h>
#include <stdlib.h>
#include <stdio.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 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;
}
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;
}
+/*
+ * 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(
+ struct lttng_userspace_probe_location *probe_location,
+ struct ltt_kernel_session *session, uint64_t *offset)
+{
+ int fd;
+ int ret = 0;
+ const char *symbol = NULL;
+ 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(
+ 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;
+ 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)
+{
+ struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
+ enum lttng_userspace_probe_location_lookup_method_type type;
+ 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.
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);
/* We pass ownership of filter_expression and filter */
- event = trace_kernel_create_event(ev, filter_expression,
- filter);
- if (event == NULL) {
- ret = -1;
+ 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;
}
event->type = ev->type;
- event->fd = ret;
+ 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");
}
if (filter) {
- ret = kernctl_filter(event->fd, filter);
- if (ret) {
+ 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;
}
}
- ret = kernctl_enable(event->fd);
- if (ret < 0) {
- switch (errno) {
+ 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;
return 0;
+add_callsite_error:
enable_error:
filter_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;
if (!ret) {
return LTTNG_OK;
}
- switch (errno) {
+ switch (-ret) {
case EINVAL:
return LTTNG_ERR_INVALID;
case ENOMEM:
if (!ret) {
return LTTNG_OK;
}
- switch (errno) {
+ switch (-ret) {
case EINVAL:
return LTTNG_ERR_INVALID;
case ENOMEM:
}
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) {
}
}
-/*
- * 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 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;
- struct lttng_kernel_tracer_abi_version abi_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 != 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(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;
}
struct consumer_socket *socket;
struct lttng_ht_iter iter;
struct ltt_kernel_metadata *saved_metadata;
+ struct ltt_session *session;
+ uint64_t trace_archive_id;
assert(ksess);
assert(ksess->consumer);
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;
/* 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,
+ ret = consumer_snapshot_channel(socket, chan->key, output, 0,
ksess->uid, ksess->gid,
DEFAULT_KERNEL_TRACE_DIR, wait,
- nb_packets_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,
}
/* 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, 0);
- 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;
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 0 on success or else return a LTTNG_ERR code.
+ */
+int kernel_rotate_session(struct ltt_session *session)
+{
+ int ret;
+ 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;
+
+ /*
+ * Account the metadata channel first to make sure the
+ * number of channels waiting for a rotation cannot
+ * reach 0 before we complete the iteration over all
+ * the channels.
+ */
+ ret = rotate_add_channel_pending(ksess->metadata->key,
+ LTTNG_DOMAIN_KERNEL, session);
+ if (ret < 0) {
+ ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
+ goto error;
+ }
+
+ /* For each channel, ask the consumer to rotate it. */
+ cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
+ ret = rotate_add_channel_pending(chan->key,
+ LTTNG_DOMAIN_KERNEL, session);
+ if (ret < 0) {
+ ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
+ goto error;
+ }
+
+ DBG("Rotate 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,
+ &session->rotate_pending_relay);
+ if (ret < 0) {
+ ret = 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,
+ &session->rotate_pending_relay);
+ if (ret < 0) {
+ ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
+ goto error;
+ }
+ }
+
+ ret = LTTNG_OK;
+
+error:
+ rcu_read_unlock();
+ return ret;
+}