summaryrefslogtreecommitdiffstats
path: root/fs/overlayfs
diff options
context:
space:
mode:
authorVivek Goyal <vgoyal@redhat.com>2018-05-11 11:49:32 -0400
committerMiklos Szeredi <mszeredi@redhat.com>2018-07-20 09:56:15 +0200
commit4120fe64dce4f73d1a595253568d9f27674f2071 (patch)
tree2880d6fc0099c1e3c69a4a090c137f48f8a171c3 /fs/overlayfs
parent7bb083837d1b5244a809ff1483408992c0e05ca6 (diff)
downloadlinux-4120fe64dce4f73d1a595253568d9f27674f2071.tar.gz
linux-4120fe64dce4f73d1a595253568d9f27674f2071.tar.bz2
linux-4120fe64dce4f73d1a595253568d9f27674f2071.zip
ovl: Set redirect on upper inode when it is linked
When we create a hardlink to a metacopy upper file, first the redirect on that inode. Path based lookup will not work with newly created link and redirect will solve that issue. Also use absolute redirect as two hardlinks could be in different directores and relative redirect will not work. I have not put any additional locking around setting redirects while introducing redirects for non-dir files. For now it feels like existing locking is sufficient. If that's not the case, we will have add more locking. Following is my rationale about why do I think current locking seems ok. Basic problem for non-dir files is that more than on dentry could be pointing to same inode and in theory only relying on dentry based locks (d->d_lock) did not seem sufficient. We set redirect upon rename and upon link creation. In both the paths for non-dir file, VFS locks both source and target inodes (->i_rwsem). That means vfs rename and link operations on same source and target can't he happening in parallel (Even if there are multiple dentries pointing to same inode). So that probably means that at a time on an inode, only one call of ovl_set_redirect() could be working and we don't need additional locking in ovl_set_redirect(). ovl_inode->redirect is initialized only when inode is created new. That means it should not race with any other path and setting ovl_inode->redirect should be fine. Reading of ovl_inode->redirect happens in ovl_get_redirect() path. And this called only in ovl_set_redirect(). And ovl_set_redirect() already seemed to be protected using ->i_rwsem. That means ovl_set_redirect() and ovl_get_redirect() on source/target inode should not make progress in parallel and is mutually exclusive. Hence no additional locking required. Now, only case where ovl_set_redirect() and ovl_get_redirect() could race seems to be case of absolute redirects where ovl_get_redirect() has to travel up the tree. In that case we already take d->d_lock and that should be sufficient as directories will not have multiple dentries pointing to same inode. So given VFS locking and current usage of redirect, current locking around redirect seems to be ok for non-dir as well. Once we have the logic to remove redirect when metacopy file gets copied up, then we probably will need additional locking. Signed-off-by: Vivek Goyal <vgoyal@redhat.com> Reviewed-by: Amir Goldstein <amir73il@gmail.com> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Diffstat (limited to 'fs/overlayfs')
-rw-r--r--fs/overlayfs/dir.c8
1 files changed, 8 insertions, 0 deletions
diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c
index ba59081394f1..ec350d4d921c 100644
--- a/fs/overlayfs/dir.c
+++ b/fs/overlayfs/dir.c
@@ -24,6 +24,8 @@ module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
MODULE_PARM_DESC(ovl_redirect_max,
"Maximum length of absolute redirect xattr value");
+static int ovl_set_redirect(struct dentry *dentry, bool samedir);
+
int ovl_cleanup(struct inode *wdir, struct dentry *wdentry)
{
int err;
@@ -657,6 +659,12 @@ static int ovl_link(struct dentry *old, struct inode *newdir,
if (err)
goto out_drop_write;
+ if (ovl_is_metacopy_dentry(old)) {
+ err = ovl_set_redirect(old, false);
+ if (err)
+ goto out_drop_write;
+ }
+
err = ovl_nlink_start(old, &locked);
if (err)
goto out_drop_write;