1 This is some background information about the Linux Auditing Framework.
5 The audit daemon is released as GPL'd code. The audit daemon's libraries
6 libaudit.* and libauparse.* are released under LGPL so that it may be
7 linked with 3rd party software.
11 See the README-install File.
15 See the man pages for audit, auditctl, audit.rules, ausearch, and aureport.
19 Original lkml thread(s):
20 http://marc.theaimsgroup.com/?t=107815888100001&r=1&w=2
21 http://marc.theaimsgroup.com/?t=107901570800002&r=1&w=2
23 There is a linux audit mail list where any question whether kernel design,
24 setup and configuration, or usage can be discussed:
25 http://www.redhat.com/mailman/listinfo/linux-audit
28 DESIGN INFO (Very old)
30 The main goals were to provide system call auditing with 1) as low
31 overhead as possible, and 2) without duplicating functionality that is
32 already provided by SELinux (and/or other security infrastructures).
33 This framework will work "stand-alone", but is not designed to provide,
34 e.g., CAPP functionality without another security component in place.
36 There are two main parts, one that is always on (generic logging in
37 audit.c) and one that you can disable at boot- or run-time
38 (per-system-call auditing in auditsc.c). The patch includes changes to
39 security/selinux/avc.c as an example of how system-call auditing can be
40 integrated with other code that identifies auditable events.
43 1) Uses a netlink socket for communication with user-space. All
44 messages are logged via the netlink socket if a user-space daemon
45 is listening. If not, the messages are logged via printk to the
46 syslog daemon (by default).
47 2) Messages can be dropped (optionally) based on message rate or
48 memory use (this isn't fully integrated into the selinux/avc.c
49 part of the patch: the avc.c code that currently does this can be
51 3) When some part of the kernel generates part of an audit record,
52 the partial record is sent immediately to user-space, AND the
53 system call "auditable" flag is automatically set for that call
54 -- thereby producing extra information at syscall exit (if
55 syscall auditing is enabled).
58 1) At task-creation time, an audit context is allocated and linked
59 off the task structure.
60 2) At syscall entry time, if the audit context exists, information
61 is filled in (syscall number, timestamp; but not arguments).
62 3) During the system call, calls to getname() and path_lookup() are
63 intercepted. These routines are called when the kernel is
64 actually looking up information that will be used to make the
65 decision about whether the syscall will succeed or fail. An
66 effort has been made to avoid copying the information that
67 getname generates, since getname is already making a
68 kernel-private copy of the information. [Note that storing
69 copies of all syscall arguments requires complexity and overhead
70 that arguably isn't needed. With this patch, for example, if
71 chroot("foo") fails because you are not root, "foo" will not
72 appear in the audit record because the kernel determined the
73 syscall cannot proceed before it ever needed to look up "foo".
74 This approach avoids storing user-supplied information that could
75 be misleading or unreliable (e.g., due to a cooperative
76 shared-memory attack) in favor of reporting information actually
78 4) At syscall exit time, if the "auditable" flag has been set (e.g.,
79 because SELinux generated an avc record; or some other part of
80 the kernel detected an auditable event), the syscall-part of the
81 audit record is generated, including file names and inode numbers
82 (if available). Some of this information is currently
83 complementary to the information that selinux/avc.c generates
84 (e.g., file names and some inode numbers), but some is less
85 complete (e.g., getname doesn't return a fully-qualified path,
86 and this patch does not add the overhead of determining one).
87 [Note that the complete audit record comes to userspace in
88 pieces, which eliminates the need to store messages for
89 arbitrarily long periods inside the kernel.]
90 5) At task-exit time, the audit context is destroyed.
92 At steps 1, 2, and 4, simple filtering can be done (e.g., a database
93 role uid might have syscall auditing disabled for performance
94 reasons). The filtering is simple and could be made more complex.
95 However, I tried to implement as much filtering as possible without
96 adding significant overhead (e.g., d_path()). In general, the audit
97 framework should rely on some other kernel component (e.g., SELinux)
98 to make the majority of the decisions about what is and is not