In the definition of ll_acl_access_xattr_handler we use the xattr name ("system.posix_acl_access") to initialize the prefix member:
static const struct xattr_handler ll_acl_access_xattr_handler = {
.prefix = XATTR_NAME_POSIX_ACL_ACCESS,
.flags = XATTR_ACL_ACCESS_T,
#if defined(HAVE_XATTR_HANDLER_SIMPLIFIED)
.get = ll_xattr_get_common_4_3,
.set = ll_xattr_set_common_4_3,
#elif !defined(HAVE_XATTR_HANDLER_INODE_PARAM)
.get = ll_xattr_get_common_3_11,
.set = ll_xattr_set_common_3_11,
#else
.get = ll_xattr_get_common,
.set = ll_xattr_set_common,
#endif
};
We should leave prefix unassigned and use the name member instead. This causes getfacl() (which uses getxattr()) to fail for some kernel versions. Similarly for ll_acl_default_xattr_handler.
There also may be conflicts between Lustre and upstream regarding the use of the flags member of struct xattr_handler. Lustre uses
#define XATTR_ACL_ACCESS_T 4
#define XATTR_ACL_DEFAULT_T 5
Upstream uses:
#define ACL_TYPE_ACCESS (0x8000)
#define ACL_TYPE_DEFAULT (0x4000)
I did not try to determine how the kernel uses the flags member.
Other filesystems use kernel defined handers for ACLs:
static const struct xattr_handler *ext4_xattr_handler_map[] = {
[EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
#ifdef CONFIG_EXT4_FS_POSIX_ACL
[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
#endif
[EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
#ifdef CONFIG_EXT4_FS_SECURITY
[EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
#endif
};
We should see if these can be used by Lustre as well.
Some auditing is needed.
Bug
Minor
LU-9679 Prepare lustre for adoption into the linux kernel
