Re: [kernel-hardening] [PATCH] fs: Use CAP_DAC_OVERRIDE to allow for file dedupe

On Sat, Oct 21, 2017 at 12:40 PM, Nicolas Belouin [off-list ref] wrote:

On October 21, 2017 4:08:31 PM GMT+02:00, Nick Kralevich [off-list ref] wrote:

quoted

On Sat, Oct 21, 2017 at 6:28 AM, Nicolas Belouin [off-list ref]
wrote:

quoted

In its current implementation the check is against CAP_SYS_ADMIN,
however this capability is bloated and inapropriate for this use.
Indeed the check aims to avoid dedupe against non writable files,
falling directly in the use case of CAP_DAC_OVERRIDE.

Signed-off-by: Nicolas Belouin <redacted>
---
 fs/read_write.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/read_write.c b/fs/read_write.c
index f0d4b16873e8..43cc7e84e29e 100644
--- a/fs/read_write.c
+++ b/fs/read_write.c

@@ -1965,7 +1965,7 @@ int vfs_dedupe_file_range(struct file *file,

struct file_dedupe_range *same)

quoted

        u64 len;
        int i;
        int ret;
-       bool is_admin = capable(CAP_SYS_ADMIN);
+       bool is_admin = capable(CAP_SYS_ADMIN) ||

capable(CAP_DAC_OVERRIDE);

Can you please reverse the order of the checks? In particular, on an
SELinux based system, a capable() call generates an SELinux denial,
and people often instinctively allow the first operation performed.
Reordering the elements will ensure that the CAP_DAC_OVERRIDE denial
(least permissive) is generated first.

Will do in the v2 of every concerned patch.

That's still a bit wrong because of how audit works.  What you really want is:

bool have_either_global_cap(int cap1, int cap2);

where, if neither cap is available, the audit message references cap1
and not cap2.  Ditto for have_either_ns_cap().

--Andy