SGI 932676 link_path_walk refcount problem allows umount of active filesystem
Subsystems (3)
PATCH 1/1 SGI 932676 link_path_walk refcount problem allows umount of active filesystem
G'day,
The attached patch fixes a bug in the VFS code which causes "Busy inodes after unmount" and a subsequent oops.
Greg.
Signature · 2 lines
Greg Banks, R&D Software Engineer, SGI Australian Software Group. I don't speak for SGI.
Following an absolute symlink opens a window during which the filesystem containing the symlink has an outstanding dentry count and no outstanding vfsmount count. A umount() of the filesystem can (incorrectly) proceed, resulting in the "Busy inodes after unmount" message and an oops shortly thereafter.
Systems using autofs-controlled NFS mounts are especially vulnerable, as autofs both increases the number of unmounts happening and does NFS mounting in response to lookups which can result in multiple-second vulnerability windows. However the bug could happen on any filesystem.
This patch adds a mntget()/mntput() pair around the link following code (as the 2.6 code does). Attempts to umount() during link following now return EBUSY.
--- linux/linux/fs/namei.c | 7 +++++++ 1 files changed, 7 insertions(+)
linux/linux/fs/namei.c +7 -0
@@ -541,8 +541,10 @@ goto out_dput;if (inode->i_op->follow_link) { +struct vfsmount *mnt =mntget (nd->mnt); err =do_follow_link (dentry, nd);dput (dentry); +mntput (mnt);if (err) goto return_err; err = -ENOENT ;
@@ -596,8 +598,10 @@ inode = dentry->d_inode;if ((lookup_flags &LOOKUP_FOLLOW ) && inode && inode->i_op && inode->i_op->follow_link) { +struct vfsmount *mnt =mntget (nd->mnt); err =do_follow_link (dentry, nd);dput (dentry); +mntput (mnt);if (err) goto return_err; inode = nd->dentry->d_inode;
@@ -1002,6 +1006,7 @@int acc_mode, error =0 ;struct inode *inode;struct dentry *dentry; +struct vfsmount *mnt;struct dentry *dir;int count =0 ;
@@ -1185,8 +1190,10 @@* are done. Procfs-like symlinks just set LAST_BIND. */ UPDATE_ATIME (dentry->d_inode); + mnt =mntget (nd->mnt); error = dentry->d_inode->i_op->follow_link (dentry, nd);dput (dentry); +mntput (mnt);if (error)return error;if (nd->last_type ==LAST_BIND ) {
Marcelo Tosatti
AppliedQuoted 4 lines · Greg Banks
G'day,
The attached patch fixes a bug in the VFS code which causes "Busy inodes after unmount" and a subsequent oops.
Applied, thanks.
Quoted 24 lines
Greg.
Signature · 21 lines
Greg Banks, R&D Software Engineer, SGI Australian Software Group. I don't speak for SGI. Following an absolute symlink opens a window during which the filesystem containing the symlink has an outstanding dentry count and no outstanding vfsmount count. A umount() of the filesystem can (incorrectly) proceed, resulting in the "Busy inodes after unmount" message and an oops shortly thereafter. Systems using autofs-controlled NFS mounts are especially vulnerable, as autofs both increases the number of unmounts happening and does NFS mounting in response to lookups which can result in multiple-second vulnerability windows. However the bug could happen on any filesystem. This patch adds a mntget()/mntput() pair around the link following code (as the 2.6 code does). Attempts to umount() during link following now return EBUSY. Signed-off-by: Greg Banks <[email protected]>