RFC: selinux: don't filter copy-up xattrs while uninitialized
PATCH 1/1 RFC: selinux: don't filter copy-up xattrs while uninitialized
Extended attribute copy-up functionality added via 19472b69d639d ("selinux: Implementation for inode_copy_up_xattr() hook") sees "security.selinux" contexts dropped, instead relying on contexts applied via the inode_copy_up() hook.
When copy-up takes place during early boot, prior to selinux initialization / policy load, the context stripping can be unwanted and unexpected. Make filtering dependent on selinux_initialized().
RFC: This changes user behaviour so is likely unacceptable. Still, I'd be interested in hearing other suggestions for how this could be addressed.
--- security/selinux/hooks.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
security/selinux/hooks.c +1 -1
@@ -3527,7 +3527,7 @@ static int selinux_inode_copy_up_xattr(const char *name)* don't then want to overwrite it by blindly copying all the lower * xattrs up. Instead, we have to filter out SELinux-related xattrs. */ -if (strcmp (name,XATTR_NAME_SELINUX ) ==0 ) +if (selinux_initialized () &&strcmp (name,XATTR_NAME_SELINUX ) ==0 )return 1 ;/* Discard */ /* * Any other attribute apart from SELINUX is not claimed, supported
Signature · 1 line
2.35.3
Stephen Smalley
maintainerQuoted 6 lines · David Disseldorp
Extended attribute copy-up functionality added via
19472b69d639d("selinux: Implementation forinode_copy_up_xattr()hook") sees "security.selinux" contexts dropped, instead relying on contexts applied via theinode_copy_up()hook.
Quoted 7 lines
When copy-up takes place during early boot, prior to selinux initialization / policy load, the context stripping can be unwanted and unexpected. Make filtering dependent on
selinux_initialized().RFC: This changes user behaviour so is likely unacceptable. Still, I'd be interested in hearing other suggestions for how this could be addressed.
IMHO, this is fixing a bug, only affects early userspace (pre policy load), and is likely acceptable.
But Paul will make the final call. We can't introduce and use a new policy capability here because this is before policy has been loaded.
Quoted 19 lines
--- security/selinux/hooks.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)security/selinux/hooks.c +1
-2@@ -3527,7 +3527,7 @@ static int selinux_inode_copy_up_xattr(const char *name)* don't then want to overwrite it by blindly copying all the lower * xattrs up. Instead, we have to filter out SELinux-related xattrs. */ -if (strcmp (name,XATTR_NAME_SELINUX ) ==0 ) +if (selinux_initialized () &&strcmp (name,XATTR_NAME_SELINUX ) ==0 )return 1 ;/* Discard */ /* * Any other attribute apart from SELINUX is not claimed, supported -- 2.35.3
Paul Moore
maintainerQuoted 11 lines · Stephen Smalley
Quoted 10 lines · David Disseldorp
Extended attribute copy-up functionality added via
19472b69d639d("selinux: Implementation forinode_copy_up_xattr()hook") sees "security.selinux" contexts dropped, instead relying on contexts applied via theinode_copy_up()hook.When copy-up takes place during early boot, prior to selinux initialization / policy load, the context stripping can be unwanted and unexpected. Make filtering dependent on
selinux_initialized().
Quoted 8 lines
Quoted 3 lines
RFC: This changes user behaviour so is likely unacceptable. Still, I'd be interested in hearing other suggestions for how this could be addressed.
IMHO, this is fixing a bug, only affects early userspace (pre policy load), and is likely acceptable.
But Paul will make the final call. We can't introduce and use a new policy capability here because this is before policy has been loaded.
I agree with Stephen, this is a bug fix so I wouldn't worry too much about user visible behavior. For better or worse, the SELinux-enabled-but-no-policy-loaded case has always been a bit awkward and has required multiple patches over the years to correct unwanted behaviors.
I'm open to comments on this, but I don't believe this is something we want to see backported to the stable kernels, and considering we are currently at v6.6-rc6, this isn't really a candidate for the upcoming merge window. This means we have a few more weeks to comment, test, etc. and one of the things I would like to see is a better description of before-and-after labeling in the commit description. This helps people who trip over this change, identify what changed, and helps them resolve the problem on their systems.
Does that sound good?
Signature · 1 line
paul-moore.com
Hi Paul and Stephen,
Quoted 22 lines · Paul Moore
Quoted 11 lines · Stephen Smalley
Quoted 10 lines · David Disseldorp
Extended attribute copy-up functionality added via
19472b69d639d("selinux: Implementation forinode_copy_up_xattr()hook") sees "security.selinux" contexts dropped, instead relying on contexts applied via theinode_copy_up()hook.When copy-up takes place during early boot, prior to selinux initialization / policy load, the context stripping can be unwanted and unexpected. Make filtering dependent on
selinux_initialized().Quoted 8 lines
Quoted 3 lines
RFC: This changes user behaviour so is likely unacceptable. Still, I'd be interested in hearing other suggestions for how this could be addressed.
IMHO, this is fixing a bug, only affects early userspace (pre policy load), and is likely acceptable.
But Paul will make the final call. We can't introduce and use a new policy capability here because this is before policy has been loaded.
Quoted 5 lines
I agree with Stephen, this is a bug fix so I wouldn't worry too much about user visible behavior. For better or worse, the SELinux-enabled-but-no-policy-loaded case has always been a bit awkward and has required multiple patches over the years to correct unwanted behaviors.
Understood.
Quoted 10 lines
I'm open to comments on this, but I don't believe this is something we want to see backported to the stable kernels, and considering we are currently at v6.6-rc6, this isn't really a candidate for the upcoming merge window. This means we have a few more weeks to comment, test, etc. and one of the things I would like to see is a better description of before-and-after labeling in the commit description. This helps people who trip over this change, identify what changed, and helps them resolve the problem on their systems.
Does that sound good?
That sounds good to me. I'll rework the commit description (and comment above this change), do some further testing and then submit a v2.
Thanks for your feedback,
David
Paul Moore
maintainerQuoted 44 lines · David Disseldorp
Hi Paul and Stephen,
Quoted 22 lines · Paul Moore
Quoted 11 lines · Stephen Smalley
Quoted 10 lines · David Disseldorp
Extended attribute copy-up functionality added via
19472b69d639d("selinux: Implementation forinode_copy_up_xattr()hook") sees "security.selinux" contexts dropped, instead relying on contexts applied via theinode_copy_up()hook.When copy-up takes place during early boot, prior to selinux initialization / policy load, the context stripping can be unwanted and unexpected. Make filtering dependent on
selinux_initialized().Quoted 8 lines
Quoted 3 lines
RFC: This changes user behaviour so is likely unacceptable. Still, I'd be interested in hearing other suggestions for how this could be addressed.
IMHO, this is fixing a bug, only affects early userspace (pre policy load), and is likely acceptable.
But Paul will make the final call. We can't introduce and use a new policy capability here because this is before policy has been loaded.Quoted 5 lines
I agree with Stephen, this is a bug fix so I wouldn't worry too much about user visible behavior. For better or worse, the SELinux-enabled-but-no-policy-loaded case has always been a bit awkward and has required multiple patches over the years to correct unwanted behaviors.
Understood.
Quoted 9 lines
I'm open to comments on this, but I don't believe this is something we want to see backported to the stable kernels, and considering we are currently at v6.6-rc6, this isn't really a candidate for the upcoming merge window. This means we have a few more weeks to comment, test, etc. and one of the things I would like to see is a better description of before-and-after labeling in the commit description. This helps people who trip over this change, identify what changed, and helps them resolve the problem on their systems.
Quoted 4 lines
Quoted 1 line
Does that sound good?
That sounds good to me. I'll rework the commit description (and comment above this change), do some further testing and then submit a v2.
Hi David,
No rush, I just wanted to check in on this and see how things were going?
Signature · 1 line
paul-moore.com