fdatasync sockets after send

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

diff --git a/libust/lttng-ust-comm.c b/libust/lttng-ust-comm.c
index 2fd165730abef212696e2fcc0c1aa3071a25d036..1507a65fcd588a3276d863d87f36a1ff1c22d295 100644 (file)
--- a/libust/lttng-ust-comm.c
+++ b/libust/lttng-ust-comm.c
@@ -184,6 +184,10 @@ int register_app_to_sessiond(int socket)
        ret = lttcomm_send_unix_sock(socket, &reg_msg, sizeof(reg_msg));
        if (ret >= 0 && ret != sizeof(reg_msg))
                return -EIO;
+       ret = fdatasync(socket);
+       if (ret) {
+               return -errno;
+       }
        return ret;
 }
 
@@ -191,11 +195,17 @@ static
 int send_reply(int sock, struct lttcomm_ust_reply *lur)
 {
        ssize_t len;
+       int ret;
 
        len = lttcomm_send_unix_sock(sock, lur, sizeof(*lur));
        switch (len) {
        case sizeof(*lur):
                DBG("message successfully sent");
+               ret = fdatasync(sock);
+               if (ret) {
+                       DBG("fdatasync error");
+                       return -1;
+               }
                return 0;
        case -1:
                if (errno == ECONNRESET) {
@@ -217,6 +227,9 @@ int handle_register_done(struct sock_info *sock_info)
        if (sock_info->constructor_sem_posted)
                return 0;
        sock_info->constructor_sem_posted = 1;
+       if (uatomic_read(&sem_count) <= 0) {
+               return 0;
+       }
        ret = uatomic_add_return(&sem_count, -1);
        if (ret == 0) {
                ret = sem_post(&constructor_wait);
@@ -513,7 +526,6 @@ void wait_for_sessiond(struct sock_info *sock_info)
 "mainline). LTTng-UST will use polling mode fallback.");
                        }
                        PERROR("futex");
-                       sleep(5);
                }
        }
        return;
@@ -524,8 +536,6 @@ quit:
 
 error:
        ust_unlock();
-       /* Error handling: fallback on a 5 seconds sleep. */
-       sleep(5);
        return;
 }
 
@@ -540,10 +550,25 @@ static
 void *ust_listener_thread(void *arg)
 {
        struct sock_info *sock_info = arg;
-       int sock, ret;
+       int sock, ret, prev_connect_failed = 0, has_waited = 0;
 
        /* Restart trying to connect to the session daemon */
 restart:
+       if (prev_connect_failed) {
+               /* Wait for sessiond availability with pipe */
+               wait_for_sessiond(sock_info);
+               if (has_waited) {
+                       has_waited = 0;
+                       /*
+                        * Sleep for 5 seconds before retrying after a
+                        * sequence of failure / wait / failure. This
+                        * deals with a killed or broken session daemon.
+                        */
+                       sleep(5);
+               }
+               has_waited = 1;
+               prev_connect_failed = 0;
+       }
        ust_lock();
 
        if (lttng_ust_comm_should_quit) {
@@ -563,6 +588,7 @@ restart:
        ret = lttcomm_connect_unix_sock(sock_info->sock_path);
        if (ret < 0) {
                ERR("Error connecting to %s apps socket", sock_info->name);
+               prev_connect_failed = 1;
                /*
                 * If we cannot find the sessiond daemon, don't delay
                 * constructor execution.
@@ -570,9 +596,6 @@ restart:
                ret = handle_register_done(sock_info);
                assert(!ret);
                ust_unlock();
-
-               /* Wait for sessiond availability with pipe */
-               wait_for_sessiond(sock_info);
                goto restart;
        }
 
@@ -595,6 +618,7 @@ restart:
        ret = register_app_to_sessiond(sock);
        if (ret < 0) {
                ERR("Error registering to %s apps socket", sock_info->name);
+               prev_connect_failed = 1;
                /*
                 * If we cannot register to the sessiond daemon, don't
                 * delay constructor execution.
@@ -602,7 +626,6 @@ restart:
                ret = handle_register_done(sock_info);
                assert(!ret);
                ust_unlock();
-               wait_for_sessiond(sock_info);
                goto restart;
        }
        ust_unlock();