*/
#define _LGPL_SOURCE
+#define _GNU_SOURCE
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/mman.h>
#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-events.h>
#include <lttng/ust-abi.h>
#include <lttng/ust.h>
+#include <lttng/ust-error.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 "ltt-tracer-core.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 ?
* The ust_lock/ust_unlock lock is used as a communication thread mutex.
* Held when handling a command, also held by fork() to deal with
* removal of threads, and by exit path.
+ *
+ * The UST lock is the centralized mutex across UST tracing control and
+ * probe registration.
+ *
+ * ust_exit_mutex must never nest in ust_mutex.
+ *
+ * ust_fork_mutex must never nest in ust_mutex.
+ *
+ * ust_mutex_nest is a per-thread nesting counter, allowing the perf
+ * counter lazy initialization called by events within the statedump,
+ * which traces while the ust_mutex is held.
+ *
+ * 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;
+
+/* Allow nesting the ust_mutex within the same thread. */
+static DEFINE_URCU_TLS(int, ust_mutex_nest);
+
+/*
+ * ust_exit_mutex protects thread_active variable wrt thread exit. It
+ * cannot be done by ust_mutex because pthread_cancel(), which takes an
+ * internal libc lock, cannot nest within ust_mutex.
+ *
+ * It never nests within a ust_mutex.
+ */
+static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+/*
+ * ust_fork_mutex protects base address statedump tracing against forks. It
+ * prevents the dynamic loader lock to be taken (by base address statedump
+ * tracing) while a fork is happening, thus preventing deadlock issues with
+ * the dynamic loader lock.
+ */
+static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
/* Should the ust comm thread quit ? */
static int lttng_ust_comm_should_quit;
+/*
+ * This variable can be tested by applications to check whether
+ * lttng-ust is loaded. They simply have to define their own
+ * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
+ * library constructor.
+ */
+int lttng_ust_loaded __attribute__((weak));
+
+/*
+ * Return 0 on success, -1 if should quit.
+ * The lock is taken in both cases.
+ * Signal-safe.
+ */
+int ust_lock(void)
+{
+ sigset_t sig_all_blocked, orig_mask;
+ int ret, oldstate;
+
+ ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
+ if (ret) {
+ ERR("pthread_setcancelstate: %s", strerror(ret));
+ }
+ if (oldstate != PTHREAD_CANCEL_ENABLE) {
+ ERR("pthread_setcancelstate: unexpected oldstate");
+ }
+ sigfillset(&sig_all_blocked);
+ ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
+ if (ret) {
+ ERR("pthread_sigmask: %s", strerror(ret));
+ }
+ if (!URCU_TLS(ust_mutex_nest)++)
+ pthread_mutex_lock(&ust_mutex);
+ ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
+ if (ret) {
+ ERR("pthread_sigmask: %s", strerror(ret));
+ }
+ if (lttng_ust_comm_should_quit) {
+ return -1;
+ } else {
+ return 0;
+ }
+}
+
+/*
+ * ust_lock_nocheck() can be used in constructors/destructors, because
+ * they are already nested within the dynamic loader lock, and therefore
+ * have exclusive access against execution of liblttng-ust destructor.
+ * Signal-safe.
+ */
+void ust_lock_nocheck(void)
+{
+ sigset_t sig_all_blocked, orig_mask;
+ int ret, oldstate;
+
+ ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
+ if (ret) {
+ ERR("pthread_setcancelstate: %s", strerror(ret));
+ }
+ if (oldstate != PTHREAD_CANCEL_ENABLE) {
+ ERR("pthread_setcancelstate: unexpected oldstate");
+ }
+ sigfillset(&sig_all_blocked);
+ ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
+ if (ret) {
+ ERR("pthread_sigmask: %s", strerror(ret));
+ }
+ if (!URCU_TLS(ust_mutex_nest)++)
+ pthread_mutex_lock(&ust_mutex);
+ ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
+ if (ret) {
+ ERR("pthread_sigmask: %s", strerror(ret));
+ }
+}
+
+/*
+ * Signal-safe.
+ */
+void ust_unlock(void)
+{
+ sigset_t sig_all_blocked, orig_mask;
+ int ret, oldstate;
+
+ sigfillset(&sig_all_blocked);
+ ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
+ if (ret) {
+ ERR("pthread_sigmask: %s", strerror(ret));
+ }
+ if (!--URCU_TLS(ust_mutex_nest))
+ pthread_mutex_unlock(&ust_mutex);
+ ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
+ if (ret) {
+ ERR("pthread_sigmask: %s", strerror(ret));
+ }
+ ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
+ if (ret) {
+ ERR("pthread_setcancelstate: %s", strerror(ret));
+ }
+ if (oldstate != PTHREAD_CANCEL_DISABLE) {
+ ERR("pthread_setcancelstate: unexpected oldstate");
+ }
+}
+
/*
* Wait for either of these before continuing to the main
* program:
/*
* 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()
* from liblttng-ust does not execute the pre/post fork handlers.
*/
-static int __thread lttng_ust_nest_count;
+static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
/*
* Info about socket and associated listener thread.
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 sock_path[PATH_MAX];
int socket;
+ int notify_socket;
char wait_shm_path[PATH_MAX];
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 = DEFAULT_GLOBAL_APPS_UNIX_SOCK,
+ .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
.socket = -1,
+ .notify_socket = -1,
+
+ .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
- .wait_shm_path = DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH,
+ .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,
.socket = -1,
+ .notify_socket = -1,
+
+ .statedump_pending = 0,
+ .initial_statedump_done = 0,
};
static int wait_poll_fallback;
-extern void ltt_ring_buffer_client_overwrite_init(void);
-extern void ltt_ring_buffer_client_discard_init(void);
-extern void ltt_ring_buffer_metadata_client_init(void);
-extern void ltt_ring_buffer_client_overwrite_exit(void);
-extern void ltt_ring_buffer_client_discard_exit(void);
-extern void ltt_ring_buffer_metadata_client_exit(void);
+static const char *cmd_name_mapping[] = {
+ [ LTTNG_UST_RELEASE ] = "Release",
+ [ LTTNG_UST_SESSION ] = "Create Session",
+ [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
+
+ [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
+ [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
+ [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
+ [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
+
+ /* Session FD commands */
+ [ LTTNG_UST_CHANNEL ] = "Create Channel",
+ [ LTTNG_UST_SESSION_START ] = "Start Session",
+ [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
+
+ /* Channel FD commands */
+ [ LTTNG_UST_STREAM ] = "Create Stream",
+ [ LTTNG_UST_EVENT ] = "Create Event",
+
+ /* Event and Channel FD commands */
+ [ LTTNG_UST_CONTEXT ] = "Create Context",
+ [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
+
+ /* Event, Channel and Session commands */
+ [ LTTNG_UST_ENABLE ] = "Enable",
+ [ LTTNG_UST_DISABLE ] = "Disable",
+
+ /* Tracepoint list commands */
+ [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
+ [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
+
+ /* Event FD commands */
+ [ LTTNG_UST_FILTER ] = "Create Filter",
+ [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
+};
+
+static const char *str_timeout;
+static int got_timeout_env;
+
+extern void lttng_ring_buffer_client_overwrite_init(void);
+extern void lttng_ring_buffer_client_overwrite_rt_init(void);
+extern void lttng_ring_buffer_client_discard_init(void);
+extern void lttng_ring_buffer_client_discard_rt_init(void);
+extern void lttng_ring_buffer_metadata_client_init(void);
+extern void lttng_ring_buffer_client_overwrite_exit(void);
+extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
+extern void lttng_ring_buffer_client_discard_exit(void);
+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;
+ size_t copied = 0, to_copy = len;
+
+ do {
+ ret = read(fd, buf + copied, to_copy);
+ if (ret > 0) {
+ copied += ret;
+ to_copy -= ret;
+ }
+ } while ((ret > 0 && to_copy > 0)
+ || (ret < 0 && errno == EINTR));
+ if (ret > 0) {
+ ret = copied;
+ }
+ return ret;
+}
+/*
+ * Returns the HOME directory path. Caller MUST NOT free(3) the returned
+ * pointer.
+ */
+static
+const char *get_lttng_home_dir(void)
+{
+ const char *val;
+
+ val = (const char *) lttng_getenv("LTTNG_HOME");
+ if (val != NULL) {
+ return val;
+ }
+ return (const char *) lttng_getenv("HOME");
+}
/*
* Force a read (imply TLS fixup for dlopen) of TLS variables.
static
void lttng_fixup_nest_count_tls(void)
{
- asm volatile ("" : : "m" (lttng_ust_nest_count));
+ asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
}
+static
+void lttng_fixup_ust_mutex_nest_tls(void)
+{
+ asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
+}
+
+/*
+ * Fixup urcu bp TLS.
+ */
+static
+void lttng_fixup_urcu_bp_tls(void)
+{
+ rcu_read_lock();
+ rcu_read_unlock();
+}
+
+void lttng_ust_fixup_tls(void)
+{
+ lttng_fixup_urcu_bp_tls();
+ lttng_fixup_ringbuffer_tls();
+ lttng_fixup_vtid_tls();
+ 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)
+{
+ struct sock_info *info = owner;
+
+ return info->notify_socket;
+}
+
+static
+void print_cmd(int cmd, int handle)
+{
+ const char *cmd_name = "Unknown";
+
+ if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
+ && cmd_name_mapping[cmd]) {
+ cmd_name = cmd_name_mapping[cmd];
+ }
+ DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
+ cmd_name, cmd,
+ 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);
+ 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);
+ ret = -ENOENT;
+ goto end;
+ }
+ local_apps.allowed = 1;
+ snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
+ home_dir,
+ LTTNG_DEFAULT_HOME_RUNDIR,
+ LTTNG_UST_SOCK_FILENAME);
+ snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
+ LTTNG_UST_WAIT_FILENAME,
+ uid);
+
+ 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 socket timeout, in ms.
+ * -1: wait forever. 0: don't wait. >0: timeout, in ms.
+ */
+static
+long get_timeout(void)
+{
+ long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
+
+ if (!got_timeout_env) {
+ str_timeout = lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
+ got_timeout_env = 1;
+ }
+ if (str_timeout)
+ constructor_delay_ms = strtol(str_timeout, NULL, 10);
+ /* All negative values are considered as "-1". */
+ if (constructor_delay_ms < -1)
+ constructor_delay_ms = -1;
+ 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.
+ */
+static
+int get_constructor_timeout(struct timespec *constructor_timeout)
+{
+ long constructor_delay_ms;
+ int ret;
+
+ constructor_delay_ms = get_timeout();
+
+ switch (constructor_delay_ms) {
+ case -1:/* fall-through */
+ case 0:
+ return constructor_delay_ms;
+ default:
+ break;
+ }
+
+ /*
+ * If we are unable to find the current time, don't wait.
+ */
+ ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
+ if (ret) {
+ /* Don't wait. */
return 0;
- } else {
- local_apps.allowed = 1;
- }
- home_dir = (const char *) getenv("HOME");
- if (!home_dir)
- return -ENOENT;
- snprintf(local_apps.sock_path, PATH_MAX,
- DEFAULT_HOME_APPS_UNIX_SOCK, home_dir);
- snprintf(local_apps.wait_shm_path, PATH_MAX,
- DEFAULT_HOME_APPS_WAIT_SHM_PATH, uid);
- return 0;
+ }
+ constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
+ constructor_timeout->tv_nsec +=
+ (constructor_delay_ms % 1000UL) * 1000000UL;
+ if (constructor_timeout->tv_nsec >= 1000000000UL) {
+ constructor_timeout->tv_sec++;
+ constructor_timeout->tv_nsec -= 1000000000UL;
+ }
+ /* Timeout wait (constructor_delay_ms). */
+ return 1;
}
static
-int register_app_to_sessiond(int socket)
+void get_allow_blocking(void)
{
- ssize_t ret;
- struct {
- uint32_t major;
- uint32_t minor;
- pid_t pid;
- pid_t ppid;
- uid_t uid;
- gid_t gid;
- uint32_t bits_per_long;
- char name[16]; /* process name */
- } reg_msg;
-
- reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR;
- reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR;
- reg_msg.pid = getpid();
- reg_msg.ppid = getppid();
- reg_msg.uid = getuid();
- reg_msg.gid = getgid();
- reg_msg.bits_per_long = CAA_BITS_PER_LONG;
- lttng_ust_getprocname(reg_msg.name);
-
- ret = ustcomm_send_unix_sock(socket, ®_msg, sizeof(reg_msg));
- if (ret >= 0 && ret != sizeof(reg_msg))
- return -EIO;
- return ret;
+ 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)
+{
+ return ustcomm_send_reg_msg(socket,
+ type,
+ CAA_BITS_PER_LONG,
+ lttng_alignof(uint8_t) * CHAR_BIT,
+ lttng_alignof(uint16_t) * CHAR_BIT,
+ lttng_alignof(uint32_t) * CHAR_BIT,
+ lttng_alignof(uint64_t) * CHAR_BIT,
+ lttng_alignof(unsigned long) * CHAR_BIT);
}
static
case sizeof(*lur):
DBG("message successfully sent");
return 0;
- case -1:
- if (errno == ECONNRESET) {
- printf("remote end closed connection\n");
+ default:
+ if (len == -ECONNRESET) {
+ DBG("remote end closed connection");
return 0;
}
- return -1;
- default:
- printf("incorrect message size: %zd\n", len);
- return -1;
+ if (len < 0)
+ return len;
+ DBG("incorrect message size: %zd", len);
+ return -EINVAL;
}
}
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 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)
+{
+ 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);
+ }
+ }
+}
+
static
int handle_message(struct sock_info *sock_info,
int sock, struct ustcomm_ust_msg *lum)
int ret = 0;
const struct lttng_ust_objd_ops *ops;
struct ustcomm_ust_reply lur;
- int shm_fd, wait_fd;
union ust_args args;
+ char ctxstr[LTTNG_UST_SYM_NAME_LEN]; /* App context string. */
ssize_t len;
- ust_lock();
-
memset(&lur, 0, sizeof(lur));
- if (lttng_ust_comm_should_quit) {
- ret = -EPERM;
- goto end;
+ if (ust_lock()) {
+ ret = -LTTNG_UST_ERR_EXITING;
+ goto error;
}
ops = objd_ops(lum->handle);
if (!ops) {
ret = -ENOENT;
- goto end;
+ goto error;
}
switch (lum->cmd) {
if (lum->handle == LTTNG_UST_ROOT_HANDLE)
ret = -EPERM;
else
- ret = lttng_ust_objd_unref(lum->handle);
+ ret = lttng_ust_objd_unref(lum->handle, 1);
break;
case LTTNG_UST_FILTER:
{
/* Receive filter data */
- struct lttng_ust_filter_bytecode *bytecode;
+ struct lttng_ust_filter_bytecode_node *bytecode;
if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
- ERR("Filter data size is too large: %u bytes\n",
+ ERR("Filter data size is too large: %u bytes",
lum->u.filter.data_size);
ret = -EINVAL;
goto error;
}
+
+ if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
+ ERR("Filter reloc offset %u is not within data",
+ lum->u.filter.reloc_offset);
+ ret = -EINVAL;
+ goto error;
+ }
+
bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
if (!bytecode) {
ret = -ENOMEM;
goto error;
}
- len = ustcomm_recv_unix_sock(sock, bytecode->data,
+ len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
lum->u.filter.data_size);
switch (len) {
case 0: /* orderly shutdown */
ret = 0;
free(bytecode);
goto error;
- case -1:
- DBG("Receive failed from lttng-sessiond with errno %d", errno);
- if (errno == ECONNRESET) {
- ERR("%s remote end closed connection\n", sock_info->name);
- ret = -EINVAL;
- free(bytecode);
- goto error;
- }
- ret = -EINVAL;
- goto end;
default:
if (len == lum->u.filter.data_size) {
- DBG("filter data received\n");
+ DBG("filter data received");
break;
+ } else if (len < 0) {
+ DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
+ if (len == -ECONNRESET) {
+ ERR("%s remote end closed connection", sock_info->name);
+ ret = len;
+ free(bytecode);
+ goto error;
+ }
+ ret = len;
+ free(bytecode);
+ goto error;
} else {
- ERR("incorrect filter data message size: %zd\n", len);
+ DBG("incorrect filter data message size: %zd", len);
ret = -EINVAL;
free(bytecode);
- goto end;
+ goto error;
}
}
- bytecode->len = lum->u.filter.data_size;
- bytecode->reloc_offset = lum->u.filter.reloc_offset;
+ bytecode->bc.len = lum->u.filter.data_size;
+ bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
+ bytecode->bc.seqnum = lum->u.filter.seqnum;
if (ops->cmd) {
ret = ops->cmd(lum->handle, lum->cmd,
(unsigned long) bytecode,
- &args);
+ &args, sock_info);
if (ret) {
free(bytecode);
}
}
break;
}
+ case LTTNG_UST_EXCLUSION:
+ {
+ /* Receive exclusion names */
+ struct lttng_ust_excluder_node *node;
+ unsigned int count;
+
+ count = lum->u.exclusion.count;
+ if (count == 0) {
+ /* There are no names to read */
+ ret = 0;
+ goto error;
+ }
+ node = zmalloc(sizeof(*node) +
+ count * LTTNG_UST_SYM_NAME_LEN);
+ if (!node) {
+ ret = -ENOMEM;
+ goto error;
+ }
+ node->excluder.count = count;
+ len = ustcomm_recv_unix_sock(sock, node->excluder.names,
+ count * LTTNG_UST_SYM_NAME_LEN);
+ switch (len) {
+ case 0: /* orderly shutdown */
+ ret = 0;
+ free(node);
+ goto error;
+ default:
+ if (len == count * LTTNG_UST_SYM_NAME_LEN) {
+ DBG("Exclusion data received");
+ break;
+ } else if (len < 0) {
+ DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
+ if (len == -ECONNRESET) {
+ ERR("%s remote end closed connection", sock_info->name);
+ ret = len;
+ free(node);
+ goto error;
+ }
+ ret = len;
+ free(node);
+ goto error;
+ } else {
+ DBG("Incorrect exclusion data message size: %zd", len);
+ ret = -EINVAL;
+ free(node);
+ goto error;
+ }
+ }
+ if (ops->cmd) {
+ ret = ops->cmd(lum->handle, lum->cmd,
+ (unsigned long) node,
+ &args, sock_info);
+ if (ret) {
+ free(node);
+ }
+ /* Don't free exclusion data if everything went fine. */
+ } else {
+ ret = -ENOSYS;
+ free(node);
+ }
+ break;
+ }
+ case LTTNG_UST_CHANNEL:
+ {
+ void *chan_data;
+ int wakeup_fd;
+
+ len = ustcomm_recv_channel_from_sessiond(sock,
+ &chan_data, lum->u.channel.len,
+ &wakeup_fd);
+ switch (len) {
+ case 0: /* orderly shutdown */
+ ret = 0;
+ goto error;
+ default:
+ if (len == lum->u.channel.len) {
+ DBG("channel data received");
+ break;
+ } else if (len < 0) {
+ DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
+ if (len == -ECONNRESET) {
+ ERR("%s remote end closed connection", sock_info->name);
+ ret = len;
+ goto error;
+ }
+ ret = len;
+ goto error;
+ } else {
+ DBG("incorrect channel data message size: %zd", len);
+ ret = -EINVAL;
+ goto error;
+ }
+ }
+ args.channel.chan_data = chan_data;
+ args.channel.wakeup_fd = wakeup_fd;
+ if (ops->cmd)
+ ret = ops->cmd(lum->handle, lum->cmd,
+ (unsigned long) &lum->u,
+ &args, sock_info);
+ else
+ ret = -ENOSYS;
+ break;
+ }
+ case LTTNG_UST_STREAM:
+ {
+ /* Receive shm_fd, wakeup_fd */
+ ret = ustcomm_recv_stream_from_sessiond(sock,
+ 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,
+ &args, sock_info);
+ else
+ ret = -ENOSYS;
+ break;
+ }
+ case LTTNG_UST_CONTEXT:
+ switch (lum->u.context.ctx) {
+ case LTTNG_UST_CONTEXT_APP_CONTEXT:
+ {
+ char *p;
+ size_t ctxlen, recvlen;
+
+ ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
+ + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
+ if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
+ ERR("Application context string length size is too large: %zu bytes",
+ ctxlen);
+ ret = -EINVAL;
+ goto error;
+ }
+ strcpy(ctxstr, "$app.");
+ p = &ctxstr[strlen("$app.")];
+ recvlen = ctxlen - strlen("$app.");
+ len = ustcomm_recv_unix_sock(sock, p, recvlen);
+ switch (len) {
+ case 0: /* orderly shutdown */
+ ret = 0;
+ goto error;
+ default:
+ if (len == recvlen) {
+ DBG("app context data received");
+ break;
+ } else if (len < 0) {
+ DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
+ if (len == -ECONNRESET) {
+ ERR("%s remote end closed connection", sock_info->name);
+ ret = len;
+ goto error;
+ }
+ ret = len;
+ goto error;
+ } else {
+ DBG("incorrect app context data message size: %zd", len);
+ ret = -EINVAL;
+ goto error;
+ }
+ }
+ /* Put : between provider and ctxname. */
+ p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
+ args.app_context.ctxname = ctxstr;
+ break;
+ }
+ default:
+ break;
+ }
+ if (ops->cmd) {
+ ret = ops->cmd(lum->handle, lum->cmd,
+ (unsigned long) &lum->u,
+ &args, sock_info);
+ } else {
+ ret = -ENOSYS;
+ }
+ break;
default:
if (ops->cmd)
ret = ops->cmd(lum->handle, lum->cmd,
(unsigned long) &lum->u,
- &args);
+ &args, sock_info);
else
ret = -ENOSYS;
break;
}
-end:
lur.handle = lum->handle;
lur.cmd = lum->cmd;
lur.ret_val = ret;
if (ret >= 0) {
- lur.ret_code = USTCOMM_OK;
+ lur.ret_code = LTTNG_UST_OK;
} else {
- //lur.ret_code = USTCOMM_SESSION_FAIL;
- lur.ret_code = ret;
+ /*
+ * Use -LTTNG_UST_ERR as wildcard for UST internal
+ * error that are not caused by the transport, except if
+ * we already have a more precise error message to
+ * report.
+ */
+ if (ret > -LTTNG_UST_ERR) {
+ /* Translate code to UST error. */
+ switch (ret) {
+ case -EEXIST:
+ lur.ret_code = -LTTNG_UST_ERR_EXIST;
+ break;
+ case -EINVAL:
+ lur.ret_code = -LTTNG_UST_ERR_INVAL;
+ break;
+ case -ENOENT:
+ lur.ret_code = -LTTNG_UST_ERR_NOENT;
+ break;
+ case -EPERM:
+ lur.ret_code = -LTTNG_UST_ERR_PERM;
+ break;
+ case -ENOSYS:
+ lur.ret_code = -LTTNG_UST_ERR_NOSYS;
+ break;
+ default:
+ lur.ret_code = -LTTNG_UST_ERR;
+ break;
+ }
+ } else {
+ lur.ret_code = ret;
+ }
}
if (ret >= 0) {
switch (lum->cmd) {
- case LTTNG_UST_STREAM:
- /*
- * Special-case reply to send stream info.
- * Use lum.u output.
- */
- lur.u.stream.memory_map_size = *args.stream.memory_map_size;
- shm_fd = *args.stream.shm_fd;
- wait_fd = *args.stream.wait_fd;
- break;
- case LTTNG_UST_METADATA:
- case LTTNG_UST_CHANNEL:
- lur.u.channel.memory_map_size = *args.channel.memory_map_size;
- shm_fd = *args.channel.shm_fd;
- wait_fd = *args.channel.wait_fd;
- break;
case LTTNG_UST_TRACER_VERSION:
lur.u.version = lum->u.version;
break;
break;
}
}
+ 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) {
- perror("error sending reply");
+ DBG("error sending reply");
goto error;
}
- if ((lum->cmd == LTTNG_UST_STREAM
- || lum->cmd == LTTNG_UST_CHANNEL
- || lum->cmd == LTTNG_UST_METADATA)
- && lur.ret_code == USTCOMM_OK) {
- int sendret = 0;
-
- /* we also need to send the file descriptors. */
- ret = ustcomm_send_fds_unix_sock(sock,
- &shm_fd, &shm_fd,
- 1, sizeof(int));
- if (ret < 0) {
- perror("send shm_fd");
- sendret = ret;
- }
- /*
- * The sessiond expects 2 file descriptors, even upon
- * error.
- */
- ret = ustcomm_send_fds_unix_sock(sock,
- &wait_fd, &wait_fd,
- 1, sizeof(int));
- if (ret < 0) {
- perror("send wait_fd");
- goto error;
- }
- if (sendret) {
- ret = sendret;
- goto error;
- }
- }
/*
* LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
- * after the reply.
- */
- if (lur.ret_code == USTCOMM_OK) {
- switch (lum->cmd) {
- case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
- len = ustcomm_send_unix_sock(sock,
- &args.field_list.entry,
- sizeof(args.field_list.entry));
- if (len != sizeof(args.field_list.entry)) {
- ret = -1;
- goto error;
- }
- }
- }
- /*
- * We still have the memory map reference, and the fds have been
- * sent to the sessiond. We can therefore close those fds. Note
- * that we keep the write side of the wait_fd open, but close
- * the read side.
+ * after the reply.
*/
- if (lur.ret_code == USTCOMM_OK) {
+ if (lur.ret_code == LTTNG_UST_OK) {
switch (lum->cmd) {
- case LTTNG_UST_STREAM:
- if (shm_fd >= 0) {
- ret = close(shm_fd);
- if (ret) {
- PERROR("Error closing stream shm_fd");
- }
- *args.stream.shm_fd = -1;
- }
- if (wait_fd >= 0) {
- ret = close(wait_fd);
- if (ret) {
- PERROR("Error closing stream wait_fd");
- }
- *args.stream.wait_fd = -1;
- }
- break;
- case LTTNG_UST_METADATA:
- case LTTNG_UST_CHANNEL:
- if (shm_fd >= 0) {
- ret = close(shm_fd);
- if (ret) {
- PERROR("Error closing channel shm_fd");
- }
- *args.channel.shm_fd = -1;
+ case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
+ len = ustcomm_send_unix_sock(sock,
+ &args.field_list.entry,
+ sizeof(args.field_list.entry));
+ if (len < 0) {
+ ret = len;
+ goto error;
}
- if (wait_fd >= 0) {
- ret = close(wait_fd);
- if (ret) {
- PERROR("Error closing channel wait_fd");
- }
- *args.channel.wait_fd = -1;
+ if (len != sizeof(args.field_list.entry)) {
+ ret = -EINVAL;
+ goto error;
}
- break;
}
}
error:
ust_unlock();
+
return ret;
}
{
int ret;
- if (sock_info->socket != -1) {
- ret = ustcomm_close_unix_sock(sock_info->socket);
- if (ret) {
- ERR("Error closing apps socket");
- }
- sock_info->socket = -1;
- }
if (sock_info->root_handle != -1) {
- ret = lttng_ust_objd_unref(sock_info->root_handle);
+ ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
if (ret) {
ERR("Error unref root handle");
}
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 is used by listener threads outside of the
- * ust lock, so we cannot tear it down ourselves, because we
- * cannot join on these threads. Leave this task to the OS
+ * wait_shm_mmap, socket and notify socket are used by listener
+ * threads outside of the ust lock, so we cannot tear them down
+ * ourselves, because we cannot join on these threads. Leave
+ * responsibility of cleaning up these resources to the OS
* process exit.
*/
- if (!exiting && sock_info->wait_shm_mmap) {
- ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
+ if (exiting)
+ return;
+
+ if (sock_info->socket != -1) {
+ ret = ustcomm_close_unix_sock(sock_info->socket);
+ if (ret) {
+ ERR("Error closing ust cmd socket");
+ }
+ sock_info->socket = -1;
+ }
+ if (sock_info->notify_socket != -1) {
+ ret = ustcomm_close_unix_sock(sock_info->notify_socket);
if (ret) {
- ERR("Error unmapping wait shm");
+ ERR("Error closing ust notify socket");
+ }
+ sock_info->notify_socket = -1;
+ }
+ if (sock_info->wait_shm_mmap) {
+ long page_size;
+
+ page_size = sysconf(_SC_PAGE_SIZE);
+ if (page_size <= 0) {
+ if (!page_size) {
+ errno = EINVAL;
+ }
+ PERROR("Error in sysconf(_SC_PAGE_SIZE)");
+ } else {
+ ret = munmap(sock_info->wait_shm_mmap, page_size);
+ if (ret) {
+ ERR("Error unmapping wait shm");
+ }
}
sock_info->wait_shm_mmap = NULL;
}
*/
wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
if (wait_shm_fd >= 0) {
+ int32_t tmp_read;
+ ssize_t len;
+ size_t bytes_read = 0;
+
+ /*
+ * Try to read the fd. If unable to do so, try opening
+ * it in write mode.
+ */
+ do {
+ len = read(wait_shm_fd,
+ &((char *) &tmp_read)[bytes_read],
+ sizeof(tmp_read) - bytes_read);
+ if (len > 0) {
+ bytes_read += len;
+ }
+ } while ((len < 0 && errno == EINTR)
+ || (len > 0 && bytes_read < sizeof(tmp_read)));
+ if (bytes_read != sizeof(tmp_read)) {
+ ret = close(wait_shm_fd);
+ if (ret) {
+ ERR("close wait_shm_fd");
+ }
+ goto open_write;
+ }
goto end;
} else if (wait_shm_fd < 0 && errno != ENOENT) {
/*
ERR("Error opening shm %s", sock_info->wait_shm_path);
goto end;
}
+
+open_write:
/*
- * If the open failed because the file did not exist, try
- * creating it ourself.
+ * If the open failed because the file did not exist, or because
+ * the file was not truncated yet, try creating it ourself.
*/
- lttng_ust_nest_count++;
+ URCU_TLS(lttng_ust_nest_count)++;
pid = fork();
- lttng_ust_nest_count--;
+ URCU_TLS(lttng_ust_nest_count)--;
if (pid > 0) {
int status;
ret = ftruncate(wait_shm_fd, mmap_size);
if (ret) {
PERROR("ftruncate");
- exit(EXIT_FAILURE);
+ _exit(EXIT_FAILURE);
}
- exit(EXIT_SUCCESS);
+ _exit(EXIT_SUCCESS);
}
/*
* For local shm, we need to have rw access to accept
*/
if (!sock_info->global && errno != EACCES) {
ERR("Error opening shm %s", sock_info->wait_shm_path);
- exit(EXIT_FAILURE);
+ _exit(EXIT_FAILURE);
}
/*
* The shm exists, but we cannot open it RW. Report
* success.
*/
- exit(EXIT_SUCCESS);
+ _exit(EXIT_SUCCESS);
} else {
return -1;
}
static
char *get_map_shm(struct sock_info *sock_info)
{
- size_t mmap_size = sysconf(_SC_PAGE_SIZE);
+ long page_size;
int wait_shm_fd, ret;
char *wait_shm_mmap;
- wait_shm_fd = get_wait_shm(sock_info, mmap_size);
+ page_size = sysconf(_SC_PAGE_SIZE);
+ if (page_size <= 0) {
+ if (!page_size) {
+ errno = EINVAL;
+ }
+ PERROR("Error in sysconf(_SC_PAGE_SIZE)");
+ 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_mmap = mmap(NULL, mmap_size, PROT_READ,
+
+ 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)
{
- int ret;
-
- ust_lock();
- if (lttng_ust_comm_should_quit) {
+ /* 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) == 0) {
- ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
- FUTEX_WAIT, 0, NULL, NULL, 0);
- if (ret < 0) {
- if (errno == EFAULT) {
- wait_poll_fallback = 1;
- DBG(
+ if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
+ goto end_wait;
+
+ while (futex_async((int32_t *) sock_info->wait_shm_mmap,
+ FUTEX_WAIT, 0, NULL, NULL, 0)) {
+ switch (errno) {
+ case EWOULDBLOCK:
+ /* Value already changed. */
+ goto end_wait;
+ case EINTR:
+ /* Retry if interrupted by signal. */
+ break; /* Get out of switch. */
+ case EFAULT:
+ wait_poll_fallback = 1;
+ DBG(
"Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
"do not support FUTEX_WAKE on read-only memory mappings correctly. "
"Please upgrade your kernel "
"(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
"mainline). LTTng-UST will use polling mode fallback.");
- if (ust_debug())
- PERROR("futex");
- }
+ if (ust_debug())
+ PERROR("futex");
+ goto end_wait;
}
}
+end_wait:
return;
quit:
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();
+ /*
+ * If available, add '-ust' to the end of this thread's
+ * process name
+ */
+ ret = lttng_ust_setustprocname();
+ if (ret) {
+ ERR("Unable to set UST process name");
+ }
/* Restart trying to connect to the session daemon */
restart:
* deals with a killed or broken session daemon.
*/
sleep(5);
+ } else {
+ has_waited = 1;
}
- has_waited = 1;
prev_connect_failed = 0;
}
- ust_lock();
- if (lttng_ust_comm_should_quit) {
- ust_unlock();
+ 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 apps socket", sock_info->name);
+ ERR("Error closing %s ust cmd socket",
+ sock_info->name);
}
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->name);
+ }
+ sock_info->notify_socket = -1;
+ }
- /* Register */
- ret = ustcomm_connect_unix_sock(sock_info->sock_path);
+
+ /*
+ * Register. We need to perform both connect and sending
+ * registration message before doing the next connect otherwise
+ * we may reach unix socket connect queue max limits and block
+ * on the 2nd connect while the session daemon is awaiting the
+ * first connect registration message.
+ */
+ /* Connect cmd socket */
+ 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 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 = sock = ret;
+ 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;
+ }
/*
* Create only one root handle per listener thread for the whole
- * process lifetime.
+ * process lifetime, so we ensure we get ID which is statically
+ * assigned to the root handle.
*/
if (sock_info->root_handle == -1) {
ret = lttng_abi_create_root_handle();
if (ret < 0) {
ERR("Error creating root handle");
- ust_unlock();
goto quit;
}
sock_info->root_handle = ret;
}
- ret = register_app_to_sessiond(sock);
+ ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
+ if (ret < 0) {
+ ERR("Error registering to %s ust cmd socket",
+ sock_info->name);
+ prev_connect_failed = 1;
+ /*
+ * If we cannot register to the sessiond daemon, don't
+ * delay constructor execution.
+ */
+ 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 */
+ 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 we cannot find the sessiond daemon, don't delay
+ * constructor execution.
+ */
+ 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) {
+ /*
+ * Give at least 10ms to sessiond to reply to
+ * notifications.
+ */
+ if (timeout < 10)
+ timeout = 10;
+ ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
+ timeout);
+ if (ret < 0) {
+ WARN("Error setting socket receive timeout");
+ }
+ ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
+ timeout);
+ if (ret < 0) {
+ WARN("Error setting socket send timeout");
+ }
+ } else if (timeout < -1) {
+ WARN("Unsupported timeout value %ld", timeout);
+ }
+
+ ret = register_to_sessiond(sock_info->notify_socket,
+ USTCTL_SOCKET_NOTIFY);
if (ret < 0) {
- ERR("Error registering to %s apps socket", sock_info->name);
+ ERR("Error registering to %s ust notify socket",
+ sock_info->name);
prev_connect_failed = 1;
/*
* 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;
}
+ sock = sock_info->socket;
+
ust_unlock();
for (;;) {
len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
switch (len) {
case 0: /* orderly shutdown */
- DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name);
- ust_lock();
+ DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
+ if (ust_lock()) {
+ goto quit;
+ }
/*
* Either sessiond has shutdown or refused us by closing the socket.
* In either case, we don't want to delay construction execution,
* 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;
case sizeof(lum):
- DBG("message received\n");
+ print_cmd(lum.cmd, lum.handle);
ret = handle_message(sock_info, sock, &lum);
- if (ret < 0) {
- ERR("Error handling message for %s socket", sock_info->name);
+ if (ret) {
+ ERR("Error handling message for %s socket",
+ sock_info->name);
+ /*
+ * Close socket if protocol error is
+ * detected.
+ */
+ goto end;
}
continue;
- case -1:
- DBG("Receive failed from lttng-sessiond with errno %d", errno);
- if (errno == ECONNRESET) {
- ERR("%s remote end closed connection\n", sock_info->name);
+ default:
+ if (len < 0) {
+ DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
+ } else {
+ DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
+ }
+ if (len == -ECONNRESET) {
+ DBG("%s remote end closed connection", sock_info->name);
goto end;
}
goto end;
- default:
- ERR("incorrect message size (%s socket): %zd\n", sock_info->name, len);
- continue;
}
}
end:
+ if (ust_lock()) {
+ goto quit;
+ }
+ /* Cleanup socket handles before trying to reconnect */
+ lttng_ust_objd_table_owner_cleanup(sock_info);
+ ust_unlock();
goto restart; /* try to reconnect */
+
quit:
+ ust_unlock();
+
+ pthread_mutex_lock(&ust_exit_mutex);
+ sock_info->thread_active = 0;
+ pthread_mutex_unlock(&ust_exit_mutex);
return NULL;
}
/*
- * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
+ * Weak symbol to call when the ust malloc wrapper is not loaded.
*/
-static
-int get_timeout(struct timespec *constructor_timeout)
+__attribute__((weak))
+void lttng_ust_malloc_wrapper_init(void)
{
- long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
- char *str_delay;
- int ret;
-
- str_delay = getenv("LTTNG_UST_REGISTER_TIMEOUT");
- if (str_delay) {
- constructor_delay_ms = strtol(str_delay, NULL, 10);
- }
-
- switch (constructor_delay_ms) {
- case -1:/* fall-through */
- case 0:
- return constructor_delay_ms;
- default:
- break;
- }
-
- /*
- * If we are unable to find the current time, don't wait.
- */
- ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
- if (ret) {
- return -1;
- }
- constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
- constructor_timeout->tv_nsec +=
- (constructor_delay_ms % 1000UL) * 1000000UL;
- if (constructor_timeout->tv_nsec >= 1000000000UL) {
- constructor_timeout->tv_sec++;
- constructor_timeout->tv_nsec -= 1000000000UL;
- }
- return 1;
}
/*
* sessiond monitoring thread: monitor presence of global and per-user
* sessiond by polling the application common named pipe.
*/
-/* TODO */
-
void __attribute__((constructor)) lttng_ust_init(void)
{
struct timespec constructor_timeout;
sigset_t sig_all_blocked, orig_parent_mask;
+ pthread_attr_t thread_attr;
int timeout_mode;
int ret;
+ void *handle;
if (uatomic_xchg(&initialized, 1) == 1)
return;
* to be the dynamic linker mutex) and ust_lock, taken within
* the ust lock.
*/
- lttng_fixup_event_tls();
- lttng_fixup_ringbuffer_tls();
- lttng_fixup_vtid_tls();
- lttng_fixup_nest_count_tls();
+ lttng_ust_fixup_tls();
+
+ 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
* sessiond before the init functions are completed).
*/
init_usterr();
+ lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */
init_tracepoint();
- ltt_ring_buffer_metadata_client_init();
- ltt_ring_buffer_client_overwrite_init();
- ltt_ring_buffer_client_discard_init();
+ lttng_ust_init_fd_tracker();
+ lttng_ust_clock_init();
+ lttng_ust_getcpu_init();
+ lttng_ust_statedump_init();
+ lttng_ring_buffer_metadata_client_init();
+ lttng_ring_buffer_client_overwrite_init();
+ lttng_ring_buffer_client_overwrite_rt_init();
+ lttng_ring_buffer_client_discard_init();
+ lttng_ring_buffer_client_discard_rt_init();
+ lttng_perf_counter_init();
+ /*
+ * Invoke ust malloc wrapper init before starting other threads.
+ */
+ lttng_ust_malloc_wrapper_init();
+
+ timeout_mode = get_constructor_timeout(&constructor_timeout);
- timeout_mode = get_timeout(&constructor_timeout);
+ get_allow_blocking();
ret = sem_init(&constructor_wait, 0, 0);
- assert(!ret);
+ 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) {
- ERR("Error setting up to local apps");
+ assert(local_apps.allowed == 0);
+ DBG("local apps setup returned %d", ret);
}
/* A new thread created by pthread_create inherits the signal mask
ERR("pthread_sigmask: %s", strerror(ret));
}
- ret = pthread_create(&global_apps.ust_listener, NULL,
- ust_listener_thread, &global_apps);
+ ret = pthread_attr_init(&thread_attr);
+ if (ret) {
+ ERR("pthread_attr_init: %s", strerror(ret));
+ }
+ ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
if (ret) {
- ERR("pthread_create global: %s", strerror(ret));
+ ERR("pthread_attr_setdetachstate: %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);
}
+
if (local_apps.allowed) {
- ret = pthread_create(&local_apps.ust_listener, NULL,
+ pthread_mutex_lock(&ust_exit_mutex);
+ ret = pthread_create(&local_apps.ust_listener, &thread_attr,
ust_listener_thread, &local_apps);
if (ret) {
ERR("pthread_create local: %s", strerror(ret));
}
+ local_apps.thread_active = 1;
+ pthread_mutex_unlock(&ust_exit_mutex);
} else {
handle_register_done(&local_apps);
}
+ ret = pthread_attr_destroy(&thread_attr);
+ if (ret) {
+ ERR("pthread_attr_destroy: %s", strerror(ret));
+ }
/* Restore original signal mask in parent */
ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
ret = sem_timedwait(&constructor_wait,
&constructor_timeout);
} while (ret < 0 && errno == EINTR);
- if (ret < 0 && errno == ETIMEDOUT) {
- ERR("Timed out waiting for ltt-sessiond");
- } else {
- assert(!ret);
+ if (ret < 0) {
+ switch (errno) {
+ case ETIMEDOUT:
+ ERR("Timed out waiting for lttng-sessiond");
+ break;
+ case EINVAL:
+ PERROR("sem_timedwait");
+ break;
+ default:
+ ERR("Unexpected error \"%s\" returned by sem_timedwait",
+ strerror(errno));
+ }
}
break;
case -1:/* wait forever */
do {
ret = sem_wait(&constructor_wait);
} while (ret < 0 && errno == EINTR);
- assert(!ret);
+ if (ret < 0) {
+ switch (errno) {
+ case EINVAL:
+ PERROR("sem_wait");
+ break;
+ default:
+ ERR("Unexpected error \"%s\" returned by sem_wait",
+ strerror(errno));
+ }
+ }
break;
case 0: /* no timeout */
break;
void lttng_ust_cleanup(int exiting)
{
cleanup_sock_info(&global_apps, exiting);
- if (local_apps.allowed) {
- cleanup_sock_info(&local_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
*/
lttng_ust_abi_exit();
lttng_ust_events_exit();
- ltt_ring_buffer_client_discard_exit();
- ltt_ring_buffer_client_overwrite_exit();
- ltt_ring_buffer_metadata_client_exit();
+ lttng_perf_counter_exit();
+ lttng_ring_buffer_client_discard_rt_exit();
+ lttng_ring_buffer_client_discard_exit();
+ lttng_ring_buffer_client_overwrite_rt_exit();
+ lttng_ring_buffer_client_overwrite_exit();
+ lttng_ring_buffer_metadata_client_exit();
+ lttng_ust_statedump_destroy();
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;
}
* mutexes to ensure it is not in a mutex critical section when
* pthread_cancel is later called.
*/
- ust_lock();
+ ust_lock_nocheck();
lttng_ust_comm_should_quit = 1;
ust_unlock();
+ pthread_mutex_lock(&ust_exit_mutex);
/* cancel threads */
- ret = pthread_cancel(global_apps.ust_listener);
- if (ret) {
- ERR("Error cancelling global ust listener thread: %s",
- strerror(ret));
+ if (global_apps.thread_active) {
+ ret = pthread_cancel(global_apps.ust_listener);
+ if (ret) {
+ ERR("Error cancelling global ust listener thread: %s",
+ strerror(ret));
+ } else {
+ global_apps.thread_active = 0;
+ }
}
- if (local_apps.allowed) {
+ if (local_apps.thread_active) {
ret = pthread_cancel(local_apps.ust_listener);
if (ret) {
ERR("Error cancelling local ust listener thread: %s",
strerror(ret));
+ } else {
+ local_apps.thread_active = 0;
}
}
+ pthread_mutex_unlock(&ust_exit_mutex);
+
/*
* Do NOT join threads: use of sys_futex makes it impossible to
* join the threads without using async-cancel, but async-cancel
sigset_t all_sigs;
int ret;
- if (lttng_ust_nest_count)
+ /* Fixup lttng-ust TLS. */
+ lttng_ust_fixup_tls();
+
+ if (URCU_TLS(lttng_ust_nest_count))
return;
/* Disable signals */
sigfillset(&all_sigs);
if (ret == -1) {
PERROR("sigprocmask");
}
- ust_lock();
- rcu_bp_before_fork();
+
+ pthread_mutex_lock(&ust_fork_mutex);
+
+ ust_lock_nocheck();
+ 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);
+
/* Restore signals */
ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
if (ret == -1) {
void ust_after_fork_parent(sigset_t *restore_sigset)
{
- if (lttng_ust_nest_count)
+ 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);
}
*/
void ust_after_fork_child(sigset_t *restore_sigset)
{
- if (lttng_ust_nest_count)
+ 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);
- lttng_context_vtid_reset();
/* Release mutexes and reenable signals */
ust_after_fork_common(restore_sigset);
lttng_ust_init();
}
+
+void lttng_ust_sockinfo_session_enabled(void *owner)
+{
+ struct sock_info *sock_info = owner;
+ sock_info->statedump_pending = 1;
+}
projects / lttng-ust.git / blobdiff