Fix: perform TLS fixup of nest count outside of UST mutex

[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 1815801a8fe08a06df22b6d1614efc969ca75b4d..4104a3f563fd400709e7e7d3e8a73a1437321d88 100644 (file)
--- a/liblttng-ust/lttng-ust-comm.c
+++ b/liblttng-ust/lttng-ust-comm.c
@@ -78,6 +78,12 @@ static sem_t constructor_wait;
  */
 static int sem_count = { 2 };
 
+/*
+ * 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;
+
 /*
  * Info about socket and associated listener thread.
  */
@@ -130,6 +136,15 @@ 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);
 
+/*
+ * 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));
+}
+
 static
 int setup_local_apps(void)
 {
@@ -236,6 +251,7 @@ int handle_message(struct sock_info *sock_info,
        struct ustcomm_ust_reply lur;
        int shm_fd, wait_fd;
        union ust_args args;
+       ssize_t len;
 
        ust_lock();
 
@@ -320,14 +336,20 @@ end:
             || 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");
-                       goto error;
+                       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));
@@ -335,6 +357,26 @@ end:
                        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
@@ -391,7 +433,7 @@ void cleanup_sock_info(struct sock_info *sock_info, int exiting)
        int ret;
 
        if (sock_info->socket != -1) {
-               ret = close(sock_info->socket);
+               ret = ustcomm_close_unix_sock(sock_info->socket);
                if (ret) {
                        ERR("Error closing apps socket");
                }
@@ -454,7 +496,9 @@ int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
         * If the open failed because the file did not exist, try
         * creating it ourself.
         */
+       lttng_ust_nest_count++;
        pid = fork();
+       lttng_ust_nest_count--;
        if (pid > 0) {
                int status;
 
@@ -671,7 +715,7 @@ restart:
        }
 
        if (sock_info->socket != -1) {
-               ret = close(sock_info->socket);
+               ret = ustcomm_close_unix_sock(sock_info->socket);
                if (ret) {
                        ERR("Error closing %s apps socket", sock_info->name);
                }
@@ -822,6 +866,7 @@ int get_timeout(struct timespec *constructor_timeout)
 void __attribute__((constructor)) lttng_ust_init(void)
 {
        struct timespec constructor_timeout;
+       sigset_t sig_all_blocked, orig_parent_mask;
        int timeout_mode;
        int ret;
 
@@ -836,6 +881,7 @@ void __attribute__((constructor)) lttng_ust_init(void)
        lttng_fixup_event_tls();
        lttng_fixup_ringbuffer_tls();
        lttng_fixup_vtid_tls();
+       lttng_fixup_nest_count_tls();
 
        /*
         * We want precise control over the order in which we construct
@@ -858,16 +904,39 @@ void __attribute__((constructor)) lttng_ust_init(void)
        if (ret) {
                ERR("Error setting up to local apps");
        }
-       ret = pthread_create(&local_apps.ust_listener, NULL,
-                       ust_listener_thread, &local_apps);
 
+       /* A new thread created by pthread_create inherits the signal mask
+        * from the parent. To avoid any signal being received by the
+        * listener thread, we block all signals temporarily in the parent,
+        * while we create the listener thread.
+        */
+       sigfillset(&sig_all_blocked);
+       ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
+       if (ret) {
+               ERR("pthread_sigmask: %s", strerror(ret));
+       }
+
+       ret = pthread_create(&global_apps.ust_listener, NULL,
+                       ust_listener_thread, &global_apps);
+       if (ret) {
+               ERR("pthread_create global: %s", strerror(ret));
+       }
        if (local_apps.allowed) {
-               ret = pthread_create(&global_apps.ust_listener, NULL,
-                               ust_listener_thread, &global_apps);
+               ret = pthread_create(&local_apps.ust_listener, NULL,
+                               ust_listener_thread, &local_apps);
+               if (ret) {
+                       ERR("pthread_create local: %s", strerror(ret));
+               }
        } else {
                handle_register_done(&local_apps);
        }
 
+       /* Restore original signal mask in parent */
+       ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
+       if (ret) {
+               ERR("pthread_sigmask: %s", strerror(ret));
+       }
+
        switch (timeout_mode) {
        case 1: /* timeout wait */
                do {
@@ -941,12 +1010,14 @@ void __attribute__((destructor)) lttng_ust_exit(void)
        /* cancel threads */
        ret = pthread_cancel(global_apps.ust_listener);
        if (ret) {
-               ERR("Error cancelling global ust listener thread");
+               ERR("Error cancelling global ust listener thread: %s",
+                       strerror(ret));
        }
        if (local_apps.allowed) {
                ret = pthread_cancel(local_apps.ust_listener);
                if (ret) {
-                       ERR("Error cancelling local ust listener thread");
+                       ERR("Error cancelling local ust listener thread: %s",
+                               strerror(ret));
                }
        }
        /*
@@ -978,6 +1049,8 @@ void ust_before_fork(sigset_t *save_sigset)
        sigset_t all_sigs;
        int ret;
 
+       if (lttng_ust_nest_count)
+               return;
        /* Disable signals */
        sigfillset(&all_sigs);
        ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
@@ -1003,6 +1076,8 @@ static void ust_after_fork_common(sigset_t *restore_sigset)
 
 void ust_after_fork_parent(sigset_t *restore_sigset)
 {
+       if (lttng_ust_nest_count)
+               return;
        DBG("process %d", getpid());
        rcu_bp_after_fork_parent();
        /* Release mutexes and reenable signals */
@@ -1020,6 +1095,8 @@ void ust_after_fork_parent(sigset_t *restore_sigset)
  */
 void ust_after_fork_child(sigset_t *restore_sigset)
 {
+       if (lttng_ust_nest_count)
+               return;
        DBG("process %d", getpid());
        /* Release urcu mutexes */
        rcu_bp_after_fork_child();