Never ever use uninitialised descriptors if pipe() fails

(because of maximum open count fd exceeded or so).

Better return busy message, so even clients fail in more controlled way.

Without patch,  clvmd crashes with various strange backtraces
if under heavy load of connection requests.

https://bugzilla.redhat.com/show_bug.cgi?id=698393
https://bugzilla.redhat.com/show_bug.cgi?id=697945

(It still can OOM but that's different issue.)

2 files changed, 13 insertions, 2 deletions

diff --git a/WHATS_NEW b/WHATS_NEW
index 85398adb..9c53e390 100644
--- a/WHATS_NEW
+++ b/WHATS_NEW
@@ -1,5 +1,6 @@
 Version 2.02.87 - 
 ===============================
+  If pipe in clvmd fails, return busy instead of using uninitialised descriptors.
   Add dmeventd monitoring shared library for RAID.
   Add RAID metadata devices to considered devices in _add_lv_to_dtree.
   Fix renaming of RAID logical volumes.
diff --git a/daemons/clvmd/clvmd.c b/daemons/clvmd/clvmd.c
index 07883eaa..cffb7456 100644
--- a/daemons/clvmd/clvmd.c
+++ b/daemons/clvmd/clvmd.c
@@ -1263,9 +1263,19 @@ static int read_from_local_sock(struct local_client *thisfd)
 		}
 
 		/* Create a pipe and add the reading end to our FD list */
-		if (pipe(comms_pipe))
+		if (pipe(comms_pipe)) {
+			struct clvm_header reply;
 			DEBUGLOG("creating pipe failed: %s\n", strerror(errno));
-		
+			reply.cmd = CLVMD_CMD_REPLY;
+			reply.status = EBUSY;
+			reply.arglen = 0;
+			reply.flags = 0;
+			send_message(&reply, sizeof(reply), our_csid,
+				     thisfd->fd,
+				     "Error sending EBUSY reply to local user");
+			return len;
+		}
+
 		newfd = malloc(sizeof(struct local_client));
 		if (!newfd) {
 			struct clvm_header reply;