another small patch: get_attr_bulk

[Nickolai Zeldovich]

It looks like get_attr_bulk() uses conn around fcache.c:3393 (in the
call to update_attr-entry) after free_fs_server_context has released
it, which means conn can potentially be recycled already.

(In my case, this bug seems to have caused arlad to segfault, while
I was writing to AFS and running aklog at the same time; conn was
on connfreelist and zeroed out..)

I believe the patch below should fix this bug.

-- kolya

--- fcache.c	2000/11/28 01:48:42	1.307
+++ fcache.c	2000/12/29 07:03:18
@@ -3262,6 +3264,7 @@
     AFSCBs cbs;
     int i;
     int len;
+    u_int32_t host;
 
     arla_warnx (ADEBFCACHE, "get_attr_bulk");
 
@@ -3359,6 +3362,7 @@
 	    break;	
     }
 
+    host = rx_HostOf (rx_PeerOf (conn->connection));
     free_fs_server_context (&context);
 
     if (ret) {
@@ -3394,7 +3398,7 @@
 			       &stats.val[i],
 			       &cbs.val[i],
 			       &sync,
-			       rx_HostOf (rx_PeerOf (conn->connection)),
+			       host,
 			       ce->cred);
 	    e->parent		= parent_entry->fid;
 	    if (!(prefered_entry && i == 0)) {