Fix: Cleanup local_apps sock_info in lttng_ust_cleanup

[lttng-ust.git] / liblttng-ust / lttng-ust-comm.c

diff --git a/liblttng-ust/lttng-ust-comm.c b/liblttng-ust/lttng-ust-comm.c
index 5df33f5f789e529a86249899b0ff4a2655757c7c..049667bf298d2122ffb904fc554b96744eb0083f 100644 (file)
--- a/liblttng-ust/lttng-ust-comm.c
+++ b/liblttng-ust/lttng-ust-comm.c
@@ -52,6 +52,7 @@
 #include "compat.h"
 #include "../libringbuffer/tlsfixup.h"
 #include "lttng-ust-baddr.h"
+#include "getenv.h"
 
 /*
  * Has lttng ust comm constructor been called ?
@@ -68,9 +69,14 @@ static int initialized;
  *
  * 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.
  */
 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
 
@@ -86,6 +92,14 @@ static DEFINE_URCU_TLS(int, ust_mutex_nest);
  */
 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;
 
@@ -300,11 +314,11 @@ const char *get_lttng_home_dir(void)
 {
        const char *val;
 
-       val = (const char *) getenv("LTTNG_HOME");
+       val = (const char *) lttng_secure_getenv("LTTNG_HOME");
        if (val != NULL) {
                return val;
        }
-       return (const char *) getenv("HOME");
+       return (const char *) lttng_secure_getenv("HOME");
 }
 
 /*
@@ -322,6 +336,16 @@ 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();
+}
+
 int lttng_get_notify_socket(void *owner)
 {
        struct sock_info *info = owner;
@@ -376,7 +400,7 @@ int setup_local_apps(void)
 
 /*
  * Get notify_sock timeout, in ms.
- * -1: don't wait. 0: wait forever. >0: timeout, in ms.
+ * -1: wait forever. 0: don't wait. >0: timeout, in ms.
  */
 static
 long get_timeout(void)
@@ -399,7 +423,7 @@ long get_notify_sock_timeout(void)
 }
 
 /*
- * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
+ * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
  */
 static
 int get_constructor_timeout(struct timespec *constructor_timeout)
@@ -422,7 +446,8 @@ int get_constructor_timeout(struct timespec *constructor_timeout)
         */
        ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
        if (ret) {
-               return -1;
+               /* Don't wait. */
+               return 0;
        }
        constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
        constructor_timeout->tv_nsec +=
@@ -431,6 +456,7 @@ int get_constructor_timeout(struct timespec *constructor_timeout)
                constructor_timeout->tv_sec++;
                constructor_timeout->tv_nsec -= 1000000000UL;
        }
+       /* Timeout wait (constructor_delay_ms). */
        return 1;
 }
 
@@ -506,7 +532,9 @@ void handle_pending_statedump(struct sock_info *sock_info)
 
        if (ctor_passed && 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);
        }
 }
 
@@ -1089,8 +1117,6 @@ error:
 static
 void wait_for_sessiond(struct sock_info *sock_info)
 {
-       int ret;
-
        if (ust_lock()) {
                goto quit;
        }
@@ -1106,23 +1132,32 @@ void wait_for_sessiond(struct sock_info *sock_info)
 
        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:
@@ -1404,6 +1439,7 @@ void __attribute__((constructor)) lttng_ust_init(void)
         * to be the dynamic linker mutex) and ust_lock, taken within
         * the ust lock.
         */
+       lttng_fixup_urcu_bp_tls();
        lttng_fixup_ringbuffer_tls();
        lttng_fixup_vtid_tls();
        lttng_fixup_nest_count_tls();
@@ -1520,9 +1556,7 @@ static
 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);
        /*
         * The teardown in this function all affect data structures
         * accessed under the UST lock by the listener thread. This
@@ -1627,6 +1661,9 @@ void ust_before_fork(sigset_t *save_sigset)
        if (ret == -1) {
                PERROR("sigprocmask");
        }
+
+       pthread_mutex_lock(&ust_fork_mutex);
+
        ust_lock_nocheck();
        rcu_bp_before_fork();
 }
@@ -1637,6 +1674,9 @@ static void ust_after_fork_common(sigset_t *restore_sigset)
 
        DBG("process %d", getpid());
        ust_unlock();
+
+       pthread_mutex_unlock(&ust_fork_mutex);
+
        /* Restore signals */
        ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
        if (ret == -1) {