Remove root handle asssert: can differ from 0 if many sessiond

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

diff --git a/libust/lttng-ust-comm.c b/libust/lttng-ust-comm.c
index 2fd165730abef212696e2fcc0c1aa3071a25d036..858e92c9b513c3bfb79a844e37d1cc315c1b968b 100644 (file)
--- a/libust/lttng-ust-comm.c
+++ b/libust/lttng-ust-comm.c
@@ -217,6 +217,9 @@ 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);
@@ -232,6 +235,7 @@ int handle_message(struct sock_info *sock_info,
        int ret = 0;
        const struct objd_ops *ops;
        struct lttcomm_ust_reply lur;
+       int shm_fd, wait_fd;
 
        ust_lock();
 
@@ -277,10 +281,53 @@ end:
        if (ret >= 0) {
                lur.ret_code = LTTCOMM_OK;
        } else {
-               lur.ret_code = LTTCOMM_SESSION_FAIL;
+               //lur.ret_code = LTTCOMM_SESSION_FAIL;
+               lur.ret_code = ret;
+       }
+       switch (lum->cmd) {
+       case LTTNG_UST_STREAM:
+               /*
+                * Special-case reply to send stream info.
+                * Use lum.u output.
+                */
+               lur.u.stream.memory_map_size = lum->u.stream.memory_map_size;
+               shm_fd = lum->u.stream.shm_fd;
+               wait_fd = lum->u.stream.wait_fd;
+               break;
+       case LTTNG_UST_METADATA:
+       case LTTNG_UST_CHANNEL:
+               lur.u.channel.memory_map_size = lum->u.channel.memory_map_size;
+               shm_fd = lum->u.channel.shm_fd;
+               wait_fd = lum->u.channel.wait_fd;
+               break;
        }
        ret = send_reply(sock, &lur);
+       if (ret < 0) {
+               perror("error sending reply");
+               goto error;
+       }
 
+       if ((lum->cmd == LTTNG_UST_STREAM
+            || lum->cmd == LTTNG_UST_CHANNEL
+            || lum->cmd == LTTNG_UST_METADATA)
+                       && lur.ret_code == LTTCOMM_OK) {
+               /* we also need to send the file descriptors. */
+               ret = lttcomm_send_fds_unix_sock(sock,
+                       &shm_fd, &shm_fd,
+                       1, sizeof(int));
+               if (ret < 0) {
+                       perror("send shm_fd");
+                       goto error;
+               }
+               ret = lttcomm_send_fds_unix_sock(sock,
+                       &wait_fd, &wait_fd,
+                       1, sizeof(int));
+               if (ret < 0) {
+                       perror("send wait_fd");
+                       goto error;
+               }
+       }
+error:
        ust_unlock();
        return ret;
 }
@@ -513,7 +560,6 @@ void wait_for_sessiond(struct sock_info *sock_info)
 "mainline). LTTng-UST will use polling mode fallback.");
                        }
                        PERROR("futex");
-                       sleep(5);
                }
        }
        return;
@@ -524,8 +570,6 @@ quit:
 
 error:
        ust_unlock();
-       /* Error handling: fallback on a 5 seconds sleep. */
-       sleep(5);
        return;
 }
 
@@ -540,10 +584,25 @@ 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 +622,7 @@ 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 +630,6 @@ 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 +652,7 @@ 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 +660,6 @@ restart:
                ret = handle_register_done(sock_info);
                assert(!ret);
                ust_unlock();
-               wait_for_sessiond(sock_info);
                goto restart;
        }
        ust_unlock();