sbin/devd/devd.cc

	Increase the size of devd's client socket's send buffer from the
	default (8k) to 128k.  This prevents clients from getting
	POLLHUPped during event storms.  For example, during zpool creation,
	the kernel emits a resource.fs.zfs.statechange event for every vdev
	in the pool.  A 128k buffer is large enough to hold the statechange
	events for a pool with nearly 800 drives.

Reviewed by:	ian, imp
Approved by:	ken (mentor)
Sponsored by:	Spectra Logic Corp
MFC after:	4 weeks

1 files changed, 20 insertions, 1 deletions

diff --git a/sbin/devd/devd.cc b/sbin/devd/devd.cc
index 21816e5..b64c459 100644
--- a/sbin/devd/devd.cc
+++ b/sbin/devd/devd.cc
@@ -104,6 +104,19 @@ __FBSDID("$FreeBSD$");
 #define CF "/etc/devd.conf"
 #define SYSCTL "hw.bus.devctl_disable"
 
+/*
+ * Since the client socket is nonblocking, we must increase its send buffer to
+ * handle brief event storms.  On FreeBSD, AF_UNIX sockets don't have a receive
+ * buffer, so the client can't increate the buffersize by itself.
+ *
+ * For example, when creating a ZFS pool, devd emits one 165 character
+ * resource.fs.zfs.statechange message for each vdev in the pool.  A 64k
+ * buffer has enough space for almost 400 drives, which would be very large but
+ * not impossibly large pool.  A 128k buffer has enough space for 794 drives,
+ * which is more than can fit in a rack with modern technology.
+ */
+#define CLIENT_BUFSIZE 131072
+
 using namespace std;
 
 extern FILE *yyin;
@@ -892,6 +905,7 @@ void
 new_client(int fd)
 {
 	int s;
+	int sndbuf_size;
 
 	/*
 	 * First go reap any zombie clients, then accept the connection, and
@@ -901,10 +915,15 @@ new_client(int fd)
 	check_clients();
 	s = accept(fd, NULL, NULL);
 	if (s != -1) {
+		sndbuf_size = CLIENT_BUFSIZE;
+		if (setsockopt(s, SOL_SOCKET, SO_SNDBUF, &sndbuf_size,
+		    sizeof(sndbuf_size)))
+			err(1, "setsockopt");
 		shutdown(s, SHUT_RD);
 		clients.push_back(s);
 		++num_clients;
-	}
+	} else
+		err(1, "accept");
 }
 
 static void