[Bug 216127] sbin/restore doesn't honour extended attributes (extattr on ufs)

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=216127

Conrad Meyer <cem at freebsd.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mckusick at FreeBSD.org

--- Comment #3 from Conrad Meyer <cem at freebsd.org> ---
The kernel code (e.g. ffs_findextattr) indicates that EA names are not, in
fact, nul-terminated.  So the headers restore uses are wrong, as is the restore
code.  This piece of kernel code hasn't changed substantially since the early
2000s.

The misleading header comment and macros that reference ea_name as if it is a
nul-terminated string were introduced in 2007 in r167010.  It immediately
precedes r167011, which introduced extattr support to restore.

ea_name will *happen* to be nul-terminated when ea_namelength % 8 == 1, like in
your "m" example.  This is because `sizeof(uint32_t) + 3 * sizeof(uint8_t) +
strlen(ea_name)` is padded to multiples of 8 bytes, and the padding is always
zeroed.

Well, there's a long time between ~2002 and 2007.  Maybe Kirk just forgot that
ea_name wasn't nul-terminated, or didn't try any keys with length == 1 (mod 8).

I propose the following actions:

* Fix the ufs/extattr.h documentation for ea_name, and remove
EXTATTR_SET_LENGTHS macro (unused) entirely (it uses strlen(ea_name))
* Fix accompanying documentation in extattr.9, fs.5.
* Merge set_extattr_file(), set_extattr_link(), and set_extattr_fd() in restore
and fix uses of ea_name that assume nul-termination.
* Finally, perform the simplification that was proposed in the r167010 commit
log.

I'll draw up some patches.

-- 
You are receiving this mail because:
You are on the CC list for the bug.