1.. SPDX-License-Identifier: GPL-2.0
2.. Copyright © 2017-2020 Mickaël Salaün <mic@digikod.net>
3.. Copyright © 2019-2020 ANSSI
4
5==================================
6Landlock LSM: kernel documentation
7==================================
8
9:Author: Mickaël Salaün
10:Date: December 2022
11
12Landlock's goal is to create scoped access-control (i.e. sandboxing).  To
13harden a whole system, this feature should be available to any process,
14including unprivileged ones.  Because such a process may be compromised or
15backdoored (i.e. untrusted), Landlock's features must be safe to use from the
16kernel and other processes point of view.  Landlock's interface must therefore
17expose a minimal attack surface.
18
19Landlock is designed to be usable by unprivileged processes while following the
20system security policy enforced by other access control mechanisms (e.g. DAC,
21LSM).  A Landlock rule shall not interfere with other access-controls enforced
22on the system, only add more restrictions.
23
24Any user can enforce Landlock rulesets on their processes.  They are merged and
25evaluated against inherited rulesets in a way that ensures that only more
26constraints can be added.
27
28User space documentation can be found here:
29Documentation/userspace-api/landlock.rst.
30
31Guiding principles for safe access controls
32===========================================
33
34* A Landlock rule shall be focused on access control on kernel objects instead
35  of syscall filtering (i.e. syscall arguments), which is the purpose of
36  seccomp-bpf.
37* To avoid multiple kinds of side-channel attacks (e.g. leak of security
38  policies, CPU-based attacks), Landlock rules shall not be able to
39  programmatically communicate with user space.
40* Kernel access check shall not slow down access request from unsandboxed
41  processes.
42* Computation related to Landlock operations (e.g. enforcing a ruleset) shall
43  only impact the processes requesting them.
44* Resources (e.g. file descriptors) directly obtained from the kernel by a
45  sandboxed process shall retain their scoped accesses (at the time of resource
46  acquisition) whatever process uses them.
47  Cf. `File descriptor access rights`_.
48
49Design choices
50==============
51
52Inode access rights
53-------------------
54
55All access rights are tied to an inode and what can be accessed through it.
56Reading the content of a directory does not imply to be allowed to read the
57content of a listed inode.  Indeed, a file name is local to its parent
58directory, and an inode can be referenced by multiple file names thanks to
59(hard) links.  Being able to unlink a file only has a direct impact on the
60directory, not the unlinked inode.  This is the reason why
61``LANDLOCK_ACCESS_FS_REMOVE_FILE`` or ``LANDLOCK_ACCESS_FS_REFER`` are not
62allowed to be tied to files but only to directories.
63
64File descriptor access rights
65-----------------------------
66
67Access rights are checked and tied to file descriptors at open time.  The
68underlying principle is that equivalent sequences of operations should lead to
69the same results, when they are executed under the same Landlock domain.
70
71Taking the ``LANDLOCK_ACCESS_FS_TRUNCATE`` right as an example, it may be
72allowed to open a file for writing without being allowed to
73:manpage:`ftruncate` the resulting file descriptor if the related file
74hierarchy doesn't grant that access right.  The following sequences of
75operations have the same semantic and should then have the same result:
76
77* ``truncate(path);``
78* ``int fd = open(path, O_WRONLY); ftruncate(fd); close(fd);``
79
80Similarly to file access modes (e.g. ``O_RDWR``), Landlock access rights
81attached to file descriptors are retained even if they are passed between
82processes (e.g. through a Unix domain socket).  Such access rights will then be
83enforced even if the receiving process is not sandboxed by Landlock.  Indeed,
84this is required to keep access controls consistent over the whole system, and
85this avoids unattended bypasses through file descriptor passing (i.e. confused
86deputy attack).
87
88Tests
89=====
90
91Userspace tests for backward compatibility, ptrace restrictions and filesystem
92support can be found here: `tools/testing/selftests/landlock/`_.
93
94Kernel structures
95=================
96
97Object
98------
99
100.. kernel-doc:: security/landlock/object.h
101    :identifiers:
102
103Filesystem
104----------
105
106.. kernel-doc:: security/landlock/fs.h
107    :identifiers:
108
109Ruleset and domain
110------------------
111
112A domain is a read-only ruleset tied to a set of subjects (i.e. tasks'
113credentials).  Each time a ruleset is enforced on a task, the current domain is
114duplicated and the ruleset is imported as a new layer of rules in the new
115domain.  Indeed, once in a domain, each rule is tied to a layer level.  To
116grant access to an object, at least one rule of each layer must allow the
117requested action on the object.  A task can then only transit to a new domain
118that is the intersection of the constraints from the current domain and those
119of a ruleset provided by the task.
120
121The definition of a subject is implicit for a task sandboxing itself, which
122makes the reasoning much easier and helps avoid pitfalls.
123
124.. kernel-doc:: security/landlock/ruleset.h
125    :identifiers:
126
127.. Links
128.. _tools/testing/selftests/landlock/:
129   https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/tools/testing/selftests/landlock/
130