[PATCH] xfsdump: don't put null inode

[Eric Sandeen <sandeen@sandeen.net>]

John Willis pointed this out:

$ sudo xfs_db -x /dev/evms/RAID5-SATA-1 
cache_node_purge: refcount was 1, not zero (node=0x80dd660)
xfs_db: cannot read root inode (22)
cache_node_purge: refcount was 1, not zero (node=0x80e0008)
xfs_db: cannot read realtime bitmap inode (22)
Segmentation fault

it dies when pthread code gets a null ptr, like this:

(gdb) bt
#0  0x0000003ef1c08309 in pthread_mutex_lock () from /lib64/libpthread.so.0
#1  0x000000000044f5af in cache_node_put (cache=0x25a2030, node=0x0) at cache.c:368
#2  0x000000000041ff7e in libxfs_mount (mp=0x0, sb=<value optimized out>, dev=<value optimized out>, logdev=<value optimized out>, 
    rtdev=<value optimized out>, flags=1) at init.c:680
#3  0x0000000000416157 in init (argc=2, argv=0x7fffd9e19798) at init.c:125
#4  0x000000000041635a in main (argc=48, argv=0x7fffd9e19798) at init.c:153
(gdb) 

when reading the root inode fails, and we try to libxfs_iput it anyway.

The following seems to fix it ...

-Eric

Index: xfsprogs/libxfs/init.c
===================================================================
--- xfsprogs.orig/libxfs/init.c
+++ xfsprogs/libxfs/init.c
@@ -677,7 +677,8 @@ libxfs_mount(
 		ASSERT(mp->m_rootip != NULL);
 	}
 	if ((flags & LIBXFS_MOUNT_ROOTINOS) && rtmount_inodes(mp)) {
-		libxfs_iput(mp->m_rootip, 0);
+		if (mp->m_rootip)
+			libxfs_iput(mp->m_rootip, 0);
 		return NULL;
 	}