#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
+#include <dlfcn.h>
#include <fcntl.h>
#include <unistd.h>
#include <errno.h>
#include <time.h>
#include <assert.h>
#include <signal.h>
+#include <limits.h>
#include <urcu/uatomic.h>
#include <urcu/futex.h>
#include <urcu/compiler.h>
#include <lttng/ust-ctl.h>
#include <urcu/tls-compat.h>
#include <ust-comm.h>
+#include <ust-fd.h>
#include <usterr-signal-safe.h>
#include <helper.h>
#include "tracepoint-internal.h"
#include "lttng-tracer-core.h"
#include "compat.h"
-#include "../libringbuffer/tlsfixup.h"
+#include "../libringbuffer/rb-init.h"
#include "lttng-ust-statedump.h"
#include "clock.h"
#include "../libringbuffer/getcpu.h"
#include "getenv.h"
+/* Concatenate lttng ust shared library name with its major version number. */
+#define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." __ust_stringify(CONFIG_LTTNG_UST_LIBRARY_VERSION_MAJOR)
+
/*
* Has lttng ust comm constructor been called ?
*/
*
* ust_lock nests within the dynamic loader lock (within glibc) because
* it is taken within the library constructor.
+ *
+ * The ust fd tracker lock nests within the ust_mutex.
*/
static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
/*
* Doing this for both the global and local sessiond.
*/
-static int sem_count = { 2 };
+enum {
+ sem_count_initial_value = 4,
+};
+
+static int sem_count = sem_count_initial_value;
/*
* Counting nesting within lttng-ust. Used to ensure that calling fork()
const char *name;
pthread_t ust_listener; /* listener thread */
int root_handle;
- int constructor_sem_posted;
+ int registration_done;
int allowed;
int global;
int thread_active;
char *wait_shm_mmap;
/* Keep track of lazy state dump not performed yet. */
int statedump_pending;
+ int initial_statedump_done;
};
/* Socket from app (connect) to session daemon (listen) for communication */
.global = 1,
.root_handle = -1,
- .allowed = 1,
+ .registration_done = 0,
+ .allowed = 0,
.thread_active = 0,
.sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
.wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
.statedump_pending = 0,
+ .initial_statedump_done = 0,
};
/* TODO: allow global_apps_sock_path override */
.name = "local",
.global = 0,
.root_handle = -1,
+ .registration_done = 0,
.allowed = 0, /* Check setuid bit first */
.thread_active = 0,
.notify_socket = -1,
.statedump_pending = 0,
+ .initial_statedump_done = 0,
};
static int wait_poll_fallback;
extern void lttng_ring_buffer_client_discard_rt_exit(void);
extern void lttng_ring_buffer_metadata_client_exit(void);
+static char *get_map_shm(struct sock_info *sock_info);
+
ssize_t lttng_ust_read(int fd, void *buf, size_t len)
{
ssize_t ret;
{
const char *val;
- val = (const char *) lttng_secure_getenv("LTTNG_HOME");
+ val = (const char *) lttng_getenv("LTTNG_HOME");
if (val != NULL) {
return val;
}
- return (const char *) lttng_secure_getenv("HOME");
+ return (const char *) lttng_getenv("HOME");
}
/*
lttng_fixup_nest_count_tls();
lttng_fixup_procname_tls();
lttng_fixup_ust_mutex_nest_tls();
+ lttng_ust_fixup_fd_tracker_tls();
}
int lttng_get_notify_socket(void *owner)
lttng_ust_obj_get_name(handle), handle);
}
+static
+int setup_global_apps(void)
+{
+ int ret = 0;
+ assert(!global_apps.wait_shm_mmap);
+
+ global_apps.wait_shm_mmap = get_map_shm(&global_apps);
+ if (!global_apps.wait_shm_mmap) {
+ WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
+ global_apps.allowed = 0;
+ ret = -EIO;
+ goto error;
+ }
+
+ global_apps.allowed = 1;
+error:
+ return ret;
+}
static
int setup_local_apps(void)
{
+ int ret = 0;
const char *home_dir;
uid_t uid;
+ assert(!local_apps.wait_shm_mmap);
+
uid = getuid();
/*
* Disallow per-user tracing for setuid binaries.
*/
if (uid != geteuid()) {
assert(local_apps.allowed == 0);
- return 0;
+ ret = 0;
+ goto end;
}
home_dir = get_lttng_home_dir();
if (!home_dir) {
WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
assert(local_apps.allowed == 0);
- return -ENOENT;
+ ret = -ENOENT;
+ goto end;
}
local_apps.allowed = 1;
snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
LTTNG_UST_WAIT_FILENAME,
uid);
- return 0;
+
+ local_apps.wait_shm_mmap = get_map_shm(&local_apps);
+ if (!local_apps.wait_shm_mmap) {
+ WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
+ local_apps.allowed = 0;
+ ret = -EIO;
+ goto end;
+ }
+end:
+ return ret;
}
/*
- * Get notify_sock timeout, in ms.
+ * Get socket timeout, in ms.
* -1: wait forever. 0: don't wait. >0: timeout, in ms.
*/
static
long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
if (!got_timeout_env) {
- str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT");
+ str_timeout = lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
got_timeout_env = 1;
}
if (str_timeout)
return constructor_delay_ms;
}
+/* Timeout for notify socket send and recv. */
static
long get_notify_sock_timeout(void)
{
return get_timeout();
}
+/* Timeout for connecting to cmd and notify sockets. */
+static
+long get_connect_sock_timeout(void)
+{
+ return get_timeout();
+}
+
/*
* Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
*/
return 1;
}
+static
+void get_allow_blocking(void)
+{
+ const char *str_allow_blocking =
+ lttng_getenv("LTTNG_UST_ALLOW_BLOCKING");
+
+ if (str_allow_blocking) {
+ DBG("%s environment variable is set",
+ "LTTNG_UST_ALLOW_BLOCKING");
+ lttng_ust_ringbuffer_set_allow_blocking();
+ }
+}
+
static
int register_to_sessiond(int socket, enum ustctl_socket_type type)
{
}
static
-int handle_register_done(struct sock_info *sock_info)
+void decrement_sem_count(unsigned int count)
{
int ret;
- if (sock_info->constructor_sem_posted)
- return 0;
- sock_info->constructor_sem_posted = 1;
+ assert(uatomic_read(&sem_count) >= count);
+
if (uatomic_read(&sem_count) <= 0) {
- return 0;
+ return;
}
- ret = uatomic_add_return(&sem_count, -1);
+
+ ret = uatomic_add_return(&sem_count, -count);
if (ret == 0) {
ret = sem_post(&constructor_wait);
assert(!ret);
}
+}
+
+static
+int handle_register_done(struct sock_info *sock_info)
+{
+ if (sock_info->registration_done)
+ return 0;
+ sock_info->registration_done = 1;
+
+ decrement_sem_count(1);
+ if (!sock_info->statedump_pending) {
+ sock_info->initial_statedump_done = 1;
+ decrement_sem_count(1);
+ }
+
+ return 0;
+}
+
+static
+int handle_register_failed(struct sock_info *sock_info)
+{
+ if (sock_info->registration_done)
+ return 0;
+ sock_info->registration_done = 1;
+ sock_info->initial_statedump_done = 1;
+
+ decrement_sem_count(2);
+
return 0;
}
/*
* Only execute pending statedump after the constructor semaphore has
- * been posted by each listener thread. This means statedump will only
- * be performed after the "registration done" command is received from
- * each session daemon the application is connected to.
+ * been posted by the current listener thread. This means statedump will
+ * only be performed after the "registration done" command is received
+ * from this thread's session daemon.
*
* This ensures we don't run into deadlock issues with the dynamic
* loader mutex, which is held while the constructor is called and
* waiting on the constructor semaphore. All operations requiring this
* dynamic loader lock need to be postponed using this mechanism.
+ *
+ * In a scenario with two session daemons connected to the application,
+ * it is possible that the first listener thread which receives the
+ * registration done command issues its statedump while the dynamic
+ * loader lock is still held by the application constructor waiting on
+ * the semaphore. It will however be allowed to proceed when the
+ * second session daemon sends the registration done command to the
+ * second listener thread. This situation therefore does not produce
+ * a deadlock.
*/
static
void handle_pending_statedump(struct sock_info *sock_info)
{
- int ctor_passed = sock_info->constructor_sem_posted;
-
- if (ctor_passed && sock_info->statedump_pending) {
+ if (sock_info->registration_done && sock_info->statedump_pending) {
sock_info->statedump_pending = 0;
pthread_mutex_lock(&ust_fork_mutex);
lttng_handle_pending_statedump(sock_info);
pthread_mutex_unlock(&ust_fork_mutex);
+
+ if (!sock_info->initial_statedump_done) {
+ sock_info->initial_statedump_done = 1;
+ decrement_sem_count(1);
+ }
}
}
{
/* Receive shm_fd, wakeup_fd */
ret = ustcomm_recv_stream_from_sessiond(sock,
- &lum->u.stream.len,
+ NULL,
&args.stream.shm_fd,
&args.stream.wakeup_fd);
if (ret) {
goto error;
}
+
if (ops->cmd)
ret = ops->cmd(lum->handle, lum->cmd,
(unsigned long) &lum->u,
}
}
DBG("Return value: %d", lur.ret_val);
+
+ ust_unlock();
+
+ /*
+ * Performed delayed statedump operations outside of the UST
+ * lock. We need to take the dynamic loader lock before we take
+ * the UST lock internally within handle_pending_statedump().
+ */
+ handle_pending_statedump(sock_info);
+
+ if (ust_lock()) {
+ ret = -LTTNG_UST_ERR_EXITING;
+ goto error;
+ }
+
ret = send_reply(sock, &lur);
if (ret < 0) {
DBG("error sending reply");
error:
ust_unlock();
- /*
- * Performed delayed statedump operations outside of the UST
- * lock. We need to take the dynamic loader lock before we take
- * the UST lock internally within handle_pending_statedump().
- */
- handle_pending_statedump(sock_info);
-
return ret;
}
}
sock_info->root_handle = -1;
}
- sock_info->constructor_sem_posted = 0;
+ sock_info->registration_done = 0;
+ sock_info->initial_statedump_done = 0;
/*
* wait_shm_mmap, socket and notify socket are used by listener
goto error;
}
+ lttng_ust_lock_fd_tracker();
wait_shm_fd = get_wait_shm(sock_info, page_size);
if (wait_shm_fd < 0) {
+ lttng_ust_unlock_fd_tracker();
goto error;
}
+
+ ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
+ if (ret < 0) {
+ ret = close(wait_shm_fd);
+ if (!ret) {
+ PERROR("Error closing fd");
+ }
+ lttng_ust_unlock_fd_tracker();
+ goto error;
+ }
+
+ wait_shm_fd = ret;
+ lttng_ust_unlock_fd_tracker();
+
wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
MAP_SHARED, wait_shm_fd, 0);
+
/* close shm fd immediately after taking the mmap reference */
+ lttng_ust_lock_fd_tracker();
ret = close(wait_shm_fd);
- if (ret) {
+ if (!ret) {
+ lttng_ust_delete_fd_from_tracker(wait_shm_fd);
+ } else {
PERROR("Error closing fd");
}
+ lttng_ust_unlock_fd_tracker();
+
if (wait_shm_mmap == MAP_FAILED) {
DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
goto error;
static
void wait_for_sessiond(struct sock_info *sock_info)
{
+ /* Use ust_lock to check if we should quit. */
if (ust_lock()) {
goto quit;
}
if (wait_poll_fallback) {
goto error;
}
- if (!sock_info->wait_shm_mmap) {
- sock_info->wait_shm_mmap = get_map_shm(sock_info);
- if (!sock_info->wait_shm_mmap)
- goto error;
- }
ust_unlock();
+ assert(sock_info->wait_shm_mmap);
+
DBG("Waiting for %s apps sessiond", sock_info->name);
/* Wait for futex wakeup */
if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
void *ust_listener_thread(void *arg)
{
struct sock_info *sock_info = arg;
- int sock, ret, prev_connect_failed = 0, has_waited = 0;
+ int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
long timeout;
lttng_ust_fixup_tls();
prev_connect_failed = 0;
}
+ if (ust_lock()) {
+ goto quit;
+ }
+
if (sock_info->socket != -1) {
+ /* FD tracker is updated by ustcomm_close_unix_sock() */
ret = ustcomm_close_unix_sock(sock_info->socket);
if (ret) {
ERR("Error closing %s ust cmd socket",
sock_info->socket = -1;
}
if (sock_info->notify_socket != -1) {
+ /* FD tracker is updated by ustcomm_close_unix_sock() */
ret = ustcomm_close_unix_sock(sock_info->notify_socket);
if (ret) {
ERR("Error closing %s ust notify socket",
sock_info->notify_socket = -1;
}
+
/*
* Register. We need to perform both connect and sending
* registration message before doing the next connect otherwise
* first connect registration message.
*/
/* Connect cmd socket */
- ret = ustcomm_connect_unix_sock(sock_info->sock_path);
+ lttng_ust_lock_fd_tracker();
+ ret = ustcomm_connect_unix_sock(sock_info->sock_path,
+ get_connect_sock_timeout());
if (ret < 0) {
+ lttng_ust_unlock_fd_tracker();
DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
prev_connect_failed = 1;
- if (ust_lock()) {
- goto quit;
- }
-
/*
* If we cannot find the sessiond daemon, don't delay
* constructor execution.
*/
- ret = handle_register_done(sock_info);
+ ret = handle_register_failed(sock_info);
assert(!ret);
ust_unlock();
goto restart;
}
+ fd = ret;
+ ret = lttng_ust_add_fd_to_tracker(fd);
+ if (ret < 0) {
+ ret = close(fd);
+ if (ret) {
+ PERROR("close on sock_info->socket");
+ }
+ ret = -1;
+ lttng_ust_unlock_fd_tracker();
+ ust_unlock();
+ goto quit;
+ }
+
sock_info->socket = ret;
+ lttng_ust_unlock_fd_tracker();
+ ust_unlock();
+ /*
+ * Unlock/relock ust lock because connect is blocking (with
+ * timeout). Don't delay constructors on the ust lock for too
+ * long.
+ */
if (ust_lock()) {
goto quit;
}
* If we cannot register to the sessiond daemon, don't
* delay constructor execution.
*/
- ret = handle_register_done(sock_info);
+ ret = handle_register_failed(sock_info);
assert(!ret);
ust_unlock();
goto restart;
}
ust_unlock();
+ /*
+ * Unlock/relock ust lock because connect is blocking (with
+ * timeout). Don't delay constructors on the ust lock for too
+ * long.
+ */
+ if (ust_lock()) {
+ goto quit;
+ }
/* Connect notify socket */
- ret = ustcomm_connect_unix_sock(sock_info->sock_path);
+ lttng_ust_lock_fd_tracker();
+ ret = ustcomm_connect_unix_sock(sock_info->sock_path,
+ get_connect_sock_timeout());
if (ret < 0) {
+ lttng_ust_unlock_fd_tracker();
DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
prev_connect_failed = 1;
- if (ust_lock()) {
- goto quit;
- }
-
/*
* If we cannot find the sessiond daemon, don't delay
* constructor execution.
*/
- ret = handle_register_done(sock_info);
+ ret = handle_register_failed(sock_info);
assert(!ret);
ust_unlock();
goto restart;
}
+
+ fd = ret;
+ ret = lttng_ust_add_fd_to_tracker(fd);
+ if (ret < 0) {
+ ret = close(fd);
+ if (ret) {
+ PERROR("close on sock_info->notify_socket");
+ }
+ ret = -1;
+ lttng_ust_unlock_fd_tracker();
+ ust_unlock();
+ goto quit;
+ }
+
sock_info->notify_socket = ret;
+ lttng_ust_unlock_fd_tracker();
+
+ ust_unlock();
+ /*
+ * Unlock/relock ust lock because connect is blocking (with
+ * timeout). Don't delay constructors on the ust lock for too
+ * long.
+ */
+ if (ust_lock()) {
+ goto quit;
+ }
timeout = get_notify_sock_timeout();
if (timeout >= 0) {
WARN("Unsupported timeout value %ld", timeout);
}
- if (ust_lock()) {
- goto quit;
- }
-
ret = register_to_sessiond(sock_info->notify_socket,
USTCTL_SOCKET_NOTIFY);
if (ret < 0) {
* If we cannot register to the sessiond daemon, don't
* delay constructor execution.
*/
- ret = handle_register_done(sock_info);
+ ret = handle_register_failed(sock_info);
assert(!ret);
ust_unlock();
goto restart;
* If we cannot register to the sessiond daemon, don't
* delay constructor execution.
*/
- ret = handle_register_done(sock_info);
+ ret = handle_register_failed(sock_info);
assert(!ret);
ust_unlock();
goto end;
pthread_attr_t thread_attr;
int timeout_mode;
int ret;
+ void *handle;
if (uatomic_xchg(&initialized, 1) == 1)
return;
lttng_ust_loaded = 1;
+ /*
+ * We need to ensure that the liblttng-ust library is not unloaded to avoid
+ * the unloading of code used by the ust_listener_threads as we can not
+ * reliably know when they exited. To do that, manually load
+ * liblttng-ust.so to increment the dynamic loader's internal refcount for
+ * this library so it never becomes zero, thus never gets unloaded from the
+ * address space of the process. Since we are already running in the
+ * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
+ * simply increment the refcount and no additionnal work is needed by the
+ * dynamic loader as the shared library is already loaded in the address
+ * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
+ * unloading of the UST library if its refcount becomes zero (which should
+ * never happen). Do the return value check but discard the handle at the
+ * end of the function as it's not needed.
+ */
+ handle = dlopen(LTTNG_UST_LIB_SO_NAME, RTLD_LAZY | RTLD_NODELETE);
+ if (!handle) {
+ ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME);
+ }
+
/*
* We want precise control over the order in which we construct
* our sub-libraries vs starting to receive commands from
* sessiond before the init functions are completed).
*/
init_usterr();
+ lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */
init_tracepoint();
+ lttng_ust_init_fd_tracker();
lttng_ust_clock_init();
lttng_ust_getcpu_init();
lttng_ust_statedump_init();
timeout_mode = get_constructor_timeout(&constructor_timeout);
+ get_allow_blocking();
+
ret = sem_init(&constructor_wait, 0, 0);
if (ret) {
PERROR("sem_init");
}
+ ret = setup_global_apps();
+ if (ret) {
+ assert(global_apps.allowed == 0);
+ DBG("global apps setup returned %d", ret);
+ }
+
ret = setup_local_apps();
if (ret) {
+ assert(local_apps.allowed == 0);
DBG("local apps setup returned %d", ret);
}
ERR("pthread_attr_setdetachstate: %s", strerror(ret));
}
- pthread_mutex_lock(&ust_exit_mutex);
- ret = pthread_create(&global_apps.ust_listener, &thread_attr,
- ust_listener_thread, &global_apps);
- if (ret) {
- ERR("pthread_create global: %s", strerror(ret));
+ if (global_apps.allowed) {
+ pthread_mutex_lock(&ust_exit_mutex);
+ ret = pthread_create(&global_apps.ust_listener, &thread_attr,
+ ust_listener_thread, &global_apps);
+ if (ret) {
+ ERR("pthread_create global: %s", strerror(ret));
+ }
+ global_apps.thread_active = 1;
+ pthread_mutex_unlock(&ust_exit_mutex);
+ } else {
+ handle_register_done(&global_apps);
}
- global_apps.thread_active = 1;
- pthread_mutex_unlock(&ust_exit_mutex);
if (local_apps.allowed) {
pthread_mutex_lock(&ust_exit_mutex);
cleanup_sock_info(&global_apps, exiting);
cleanup_sock_info(&local_apps, exiting);
local_apps.allowed = 0;
+ global_apps.allowed = 0;
/*
* The teardown in this function all affect data structures
* accessed under the UST lock by the listener thread. This
exit_tracepoint();
if (!exiting) {
/* Reinitialize values for fork */
- sem_count = 2;
+ sem_count = sem_count_initial_value;
lttng_ust_comm_should_quit = 0;
initialized = 0;
}
pthread_mutex_lock(&ust_fork_mutex);
ust_lock_nocheck();
- rcu_bp_before_fork();
+ urcu_bp_before_fork();
+ lttng_ust_lock_fd_tracker();
}
static void ust_after_fork_common(sigset_t *restore_sigset)
int ret;
DBG("process %d", getpid());
+ lttng_ust_unlock_fd_tracker();
ust_unlock();
pthread_mutex_unlock(&ust_fork_mutex);
if (URCU_TLS(lttng_ust_nest_count))
return;
DBG("process %d", getpid());
- rcu_bp_after_fork_parent();
+ urcu_bp_after_fork_parent();
/* Release mutexes and reenable signals */
ust_after_fork_common(restore_sigset);
}
{
if (URCU_TLS(lttng_ust_nest_count))
return;
+ lttng_context_vpid_reset();
lttng_context_vtid_reset();
+ lttng_context_procname_reset();
DBG("process %d", getpid());
/* Release urcu mutexes */
- rcu_bp_after_fork_child();
+ urcu_bp_after_fork_child();
lttng_ust_cleanup(0);
/* Release mutexes and reenable signals */
ust_after_fork_common(restore_sigset);
projects / lttng-ust.git / blobdiff