overlayfs, doc: Do not allow lower layer recreation with redirect_dir enabled
Subsystems (2)
PATCH 1/1 overlayfs, doc: Do not allow lower layer recreation with redirect_dir enabled
Currently we seem to support lower layer recreation and re-use with existing upper until and unless "index" or "metadata only copy up" feature is enabled.
If redirect_dir feature is enabled then re-creating/modifying lower layers will break things. For example.
- mkdir lower lower/foo upper work merged
- touch lower/foo/foo-child
- mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,redirect_dir=on none merged
- mv merged/foo merged/bar
- ls merged/bar/ (this should list foo-child)
- umount merged
- mv lower/foo lower/baz
- mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,redirect_dir=on none merged
- ls merged/bar/ (Now foo-child has disappeared)
IOW, modifying lower layers did not crash overlay but it resulted in directory contents being lost and that can be unexpected. So don't support lower layer recreation/modification when redirect_dir is enabled at any point of time.
--- Documentation/filesystems/overlayfs.rst | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)
Documentation/filesystems/overlayfs.rst +8 -4
@@ -371,8 +371,8 @@ conflict with metacopy=on, and will result in an error. [*] redirect_dir=follow only conflicts with metacopy=on if upperdir=... is given. -Sharing and copying layers --------------------------- +Sharing, copying and recreating lower layers +-------------------------------------------- Lower layers may be shared among several overlay mounts and that is indeed a very common practice. An overlay mount may use the same lower layer
@@ -388,8 +388,12 @@ though it will not result in a crash or deadlock. Mounting an overlay using an upper layer path, where the upper layer path was previously used by another mounted overlay in combination with a -different lower layer path, is allowed, unless the "inodes index" feature -or "metadata only copy up" feature is enabled. +different lower layer path, is allowed, unless any of the following features +is enabled at any point of time. + +- inode index +- metadata only copy up +- redirect_dir With the "inodes index" feature, on the first time mount, an NFS file handle of the lower layer root directory, along with the UUID of the lower
Signature · 1 line
2.25.4
Quoted 45 lines · Vivek Goyal
Currently we seem to support lower layer recreation and re-use with existing upper until and unless "index" or "metadata only copy up" feature is enabled.
If
redirect_dirfeature is enabled then re-creating/modifying lower layers will break things. For example.- mkdir lower lower/foo upper work merged
- touch lower/foo/foo-child
- mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,redirect_dir=on none merged
- mv merged/foo merged/bar
- ls merged/bar/ (this should list foo-child)- umount merged
- mv lower/foo lower/baz
- mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,redirect_dir=on none merged
- ls merged/bar/ (Now foo-child has disappeared)IOW, modifying lower layers did not crash overlay but it resulted in directory contents being lost and that can be unexpected. So don't support lower layer recreation/modification when
redirect_diris enabled at any point of time.Signed-off-by Vivek Goyal <[email protected]>--- Documentation/filesystems/overlayfs.rst | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)Documentation/filesystems/overlayfs.rst +2
-2@@ -371,8 +371,8 @@ conflict with metacopy=on, and will result in an error. [*] redirect_dir=follow only conflicts with metacopy=on if upperdir=... is given. -Sharing and copying layers --------------------------- +Sharing, copying and recreating lower layers +-------------------------------------------- Lower layers may be shared among several overlay mounts and that is indeed a very common practice. An overlay mount may use the same lower layer@@ -388,8 +388,12 @@ though it will not result in a crash or deadlock.
Documentation/filesystems/overlayfs.rst though it will not result in a crash or deadlock. · line 396
Quoted 10 lines
Documentation/filesystems/overlayfs.rst +6
-2Mounting an overlay using an upper layer path, where the upper layer path was previously used by another mounted overlay in combination with a -different lower layer path, is allowed, unless the "inodes index" feature -or "metadata only copy up" feature is enabled. +different lower layer path, is allowed, unless any of the following features +is enabled at any point of time. + +- inode index +- metadata only copy up +- redirect_dir
I probably should add "nfs_export" to the list as well. Though it is implicitly there as enabling nfs export requires to enable index. But saying it explicitly is even better.
Vivek
Quoted 6 lines
Documentation/filesystems/overlayfs.rst +0
-0With the "inodes index" feature, on the first time mount, an NFS file handle of the lower layer root directory, along with the UUID of the lowerSignature · 1 line
2.25.4
Amir Goldstein
maintainerQuoted 20 lines · Vivek Goyal
Quoted 18 lines · Vivek Goyal
Currently we seem to support lower layer recreation and re-use with existing upper until and unless "index" or "metadata only copy up" feature is enabled.
If
redirect_dirfeature is enabled then re-creating/modifying lower layers will break things. For example.- mkdir lower lower/foo upper work merged
- touch lower/foo/foo-child
- mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,redirect_dir=on none merged
- mv merged/foo merged/bar
- ls merged/bar/ (this should list foo-child)- umount merged
- mv lower/foo lower/baz
- mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,redirect_dir=on none merged
- ls merged/bar/ (Now foo-child has disappeared)
Quoted 4 lines
Quoted 4 lines
IOW, modifying lower layers did not crash overlay but it resulted in directory contents being lost and that can be unexpected. So don't support lower layer recreation/modification when
redirect_diris enabled at any point of time.
I don't understand why this has to do with redirect_dir. The same text holds also if you do not do mv merged/foo merged/bar
Quoted 23 lines
Quoted 23 lines
Signed-off-by Vivek Goyal <[email protected]>--- Documentation/filesystems/overlayfs.rst | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)Documentation/filesystems/overlayfs.rst +2
-2@@ -371,8 +371,8 @@ conflict with metacopy=on, and will result in an error. [*] redirect_dir=follow only conflicts with metacopy=on if upperdir=... is given. -Sharing and copying layers --------------------------- +Sharing, copying and recreating lower layers +-------------------------------------------- Lower layers may be shared among several overlay mounts and that is indeed a very common practice. An overlay mount may use the same lower layer@@ -388,8 +388,12 @@ though it will not result in a crash or deadlock.
Quoted 15 lines
Documentation/filesystems/overlayfs.rst though it will not result in a crash or deadlock. · line 396
Quoted 10 lines
Documentation/filesystems/overlayfs.rst +6
-2Mounting an overlay using an upper layer path, where the upper layer path was previously used by another mounted overlay in combination with a -different lower layer path, is allowed, unless the "inodes index" feature -or "metadata only copy up" feature is enabled. +different lower layer path, is allowed, unless any of the following features +is enabled at any point of time. + +- inode index +- metadata only copy up +- redirect_dirI probably should add "
nfs_export" to the list as well. Though it is implicitly there as enabling nfs export requires to enable index. But saying it explicitly is even better.
Ok you can mention it but xino is also relevant.
I wonder if we should define "legacy mode" where no options expect for lowerdir,upperdir,workdir,default_permissions are provided
Thanks,
Amir.
Quoted 21 lines · Amir Goldstein
Quoted 20 lines · Vivek Goyal
Quoted 18 lines · Vivek Goyal
Currently we seem to support lower layer recreation and re-use with existing upper until and unless "index" or "metadata only copy up" feature is enabled.
If
redirect_dirfeature is enabled then re-creating/modifying lower layers will break things. For example.- mkdir lower lower/foo upper work merged
- touch lower/foo/foo-child
- mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,redirect_dir=on none merged
- mv merged/foo merged/bar
- ls merged/bar/ (this should list foo-child)- umount merged
- mv lower/foo lower/baz
- mount -t overlay -o lowerdir=lower,upperdir=upper,workdir=work,redirect_dir=on none merged
- ls merged/bar/ (Now foo-child has disappeared)
Quoted 7 lines
Quoted 4 lines
Quoted 4 lines
IOW, modifying lower layers did not crash overlay but it resulted in directory contents being lost and that can be unexpected. So don't support lower layer recreation/modification when
redirect_diris enabled at any point of time.I don't understand why this has to do with
redirect_dir. The same text holds also if you do not do mv merged/foo merged/bar
It does not. "mv foo bar" gets -EXDEV and then mv copies up whole directory tree. So even if lower layer is modified, we still see "bar/foo-child".
So behavior is little different in two cases. In one case, we lose directory tree (as lower renamed it) and in another case we still see it(because we copied it up).
Now one can argue that probably user will expect it (as long as they know what redirect_dir does and all the semantics of copy up).
If our goal is that we need to allow this (as long as system does not crash and does not throw errors) and all the resulting behaviors can be explained away, then we probably don't need this patch.
I am not sure why do we need to allow updating lower and reuse with same upper. It is hard to support these configurations and resulting behavior. So with this patch I am trying to cut down the number of options which allow this configuraiton.
In general, if any overlay features stores metadata in upper about lower layer, then with lower layer change, that metadata can be broken. And we probably should not allow such configuraiton. I am approaching it from this angle. (until and unless there is a strong use case that why we need to continue to support it with newer overlay features).
Thanks
Vivek
Quoted 48 lines
Quoted 25 lines
Quoted 25 lines
Signed-off-by Vivek Goyal <[email protected]>--- Documentation/filesystems/overlayfs.rst | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-)Documentation/filesystems/overlayfs.rst +2
-2@@ -371,8 +371,8 @@ conflict with metacopy=on, and will result in an error. [*] redirect_dir=follow only conflicts with metacopy=on if upperdir=... is given. -Sharing and copying layers --------------------------- +Sharing, copying and recreating lower layers +-------------------------------------------- Lower layers may be shared among several overlay mounts and that is indeed a very common practice. An overlay mount may use the same lower layer@@ -388,8 +388,12 @@ though it will not result in a crash or deadlock.Quoted 15 lines
Documentation/filesystems/overlayfs.rst though it will not result in a crash or deadlock. · line 396
Quoted 10 lines
Documentation/filesystems/overlayfs.rst +6
-2Mounting an overlay using an upper layer path, where the upper layer path was previously used by another mounted overlay in combination with a -different lower layer path, is allowed, unless the "inodes index" feature -or "metadata only copy up" feature is enabled. +different lower layer path, is allowed, unless any of the following features +is enabled at any point of time. + +- inode index +- metadata only copy up +- redirect_dirI probably should add "
nfs_export" to the list as well. Though it is implicitly there as enabling nfs export requires to enable index. But saying it explicitly is even better.Ok you can mention it but xino is also relevant.
I wonder if we should define "legacy mode" where no options expect for lowerdir,upperdir,workdir,default_permissions are providedThanks,
Amir.