table of contents
PRIV(9) | Kernel Developer's Manual | PRIV(9) |
NAME¶
priv
— kernel
privilege checking API
SYNOPSIS¶
#include
<sys/priv.h>
int
priv_check
(struct
thread *td, int
priv);
int
priv_check_cred
(struct
ucred *cred, int
priv, int
flags);
DESCRIPTION¶
The priv
interfaces check to see if
specific system privileges are granted to the passed thread,
td, or credential, cred. This
interface replaces the now removed suser(9) privilege
checking interface. Privileges typically represent rights in one of two
categories: the right to manage a particular component of the system, or an
exemption to a specific policy or access control list. The caller identifies
the desired privilege via the priv argument. The
optional flags argument, flags, is currently
unused.
Privilege Policies¶
Privileges are typically granted based on one of two base system policies: the superuser policy, which grants privilege based on the effective (or sometimes real) UID having a value of 0, and the jail(2) policy, which permits only certain privileges to be granted to processes in a jail. The set of available privileges may also be influenced by the TrustedBSD MAC Framework, described in mac(9).
IMPLEMENTATION NOTES¶
When adding a new privilege check to a code path, first check the
complete list of current privileges in sys/priv.h to
see if one already exists for the class of privilege required. Only if there
is not an exact match should a new privilege be added to the privilege list.
As privilege numbers becomes encoded in the kernel module ABI, privilege
constants must not be changed as any kernel modules depending on privileges
will then need to be recompiled. When adding a new privilege, be certain to
also determine whether it should be listed in
prison_priv_check
(), which includes a complete list
of privileges granted to the root user in jail(2).
Certain catch-all privileges exist, such as
PRIV_DRIVER
, intended to be used by device drivers,
rather than adding a new driver-specific privilege.
RETURN VALUES¶
Typically, 0 will be returned for success, and
EPERM
will be returned on failure. Most consumers of
priv
will wish to directly return the error code
from a failed privilege check to user space; a small number will wish to
translate it to another error code appropriate to a specific context.
When designing new APIs, it is preferable to return explicit errors from a call if privilege is not granted rather than changing the semantics of the call but returning success. For example, the behavior exhibited by stat(2), in which the generation field is optionally zero'd out when there is insufficient privilege is highly undesirable, as it results in frequent privilege checks, and the caller is unable to tell if an access control failure occurred.
SEE ALSO¶
AUTHORS¶
The priv
API and implementation were
created by Robert Watson under contract to nCircle
Network Security, Inc.
August 30, 2006 | Debian |