* 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>
#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;
/*
* 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;
/*
* 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;
struct ltt_kernel_event *event;
assert(ev);
assert(channel);
- event = trace_kernel_create_event(ev);
+ /* We pass ownership of filter_expression and filter */
+ event = trace_kernel_create_event(ev, filter_expression,
+ filter);
if (event == NULL) {
ret = -1;
goto error;
ret = kernctl_create_event(channel->fd, event->event);
if (ret < 0) {
- switch (errno) {
+ switch (-ret) {
case EEXIST:
break;
case ENOSYS:
default:
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->type = ev->type;
event->fd = ret;
/* Prevent fd duplication after execlp() */
ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
PERROR("fcntl session fd");
}
-add_list:
+ if (filter) {
+ ret = kernctl_filter(event->fd, filter);
+ if (ret) {
+ goto filter_error;
+ }
+ }
+
+ ret = kernctl_enable(event->fd);
+ if (ret < 0) {
+ switch (-ret) {
+ case EEXIST:
+ ret = LTTNG_ERR_KERN_EVENT_EXIST;
+ break;
+ default:
+ PERROR("enable kernel event");
+ break;
+ }
+ goto enable_error;
+ }
+
/* Add event to event list */
cds_list_add(&event->list, &channel->events_list.head);
channel->event_count++;
return 0;
+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)
+
+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)
{
- return kernctl_enable_syscall(channel->fd, syscall_name);
+ 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;
+ }
}
-int kernel_disable_syscall(const char *syscall_name,
- struct ltt_kernel_channel *channel)
+ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
+ int **_pids)
{
- return kernctl_disable_syscall(channel->fd, syscall_name);
+ 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;
}
/*
}
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;
- 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;
}
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 */
* 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, uint64_t max_size_per_stream)
+ 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;
+ 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,
- 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,
}
/* 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;
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 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;
+}