1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
|
# Linux `SUID` Sandbox
With [r20110](https://crrev.com/20110), Chromium on Linux can now sandbox its
renderers using a `SUID` helper binary. This is one of
[our layer-1 sandboxing solutions](linux_sandboxing.md).
## `SUID` helper executable
The `SUID` helper binary is called `chrome_sandbox` and you must build it
separately from the main 'chrome' target. To use this sandbox, you have to
specify its path in the `linux_sandbox_path` GYP variable. When spawning the
[zygote process](linux_zygote.md), if the `SUID` sandbox is enabled, Chromium
will check for the sandbox binary at the location specified by
`linux_sandbox_path`. For Google Chrome, this is set to
`/opt/google/chrome/chrome-sandbox`, and early version had this value hard coded
in `chrome/browser/zygote_host_linux.cc`.
In order for the sandbox to be used, the following conditions must be met:
* The sandbox binary must be executable by the Chromium process.
* It must be `SUID` and executable by other.
If these conditions are met then the sandbox binary is used to launch the zygote
process. Once the zygote has started, it asks a helper process to chroot it to a
temp directory.
## `CLONE_NEWPID` method
The sandbox does three things to restrict the authority of a sandboxed process.
The `SUID` helper is responsible for the first two:
* The `SUID` helper chroots the process. This takes away access to the
filesystem namespace.
* The `SUID` helper puts the process in a PID namespace using the
`CLONE_NEWPID` option to
[clone()](http://www.kernel.org/doc/man-pages/online/pages/man2/clone.2.html).
This stops the sandboxed process from being able to `ptrace()` or `kill()`
unsandboxed processes.
In addition:
* The [Linux Zygote](linux_zygote.md) startup code sets the process to be
_undumpable_ using
[prctl()](http://www.kernel.org/doc/man-pages/online/pages/man2/prctl.2.html).
This stops sandboxed processes from being able to `ptrace()` each other.
More specifically, it stops the sandboxed process from being `ptrace()`'d by
any other process. This can be switched off with the
`--allow-sandbox-debugging` option.
Limitations:
* Not all kernel versions support `CLONE_NEWPID`. If the `SUID` helper is run
on a kernel that does not support `CLONE_NEWPID`, it will ignore the problem
without a warning, but the protection offered by the sandbox will be
substantially reduced. See LinuxPidNamespaceSupport for how to test whether
your system supports PID namespaces.
* This does not restrict network access.
* This does not prevent processes within a given sandbox from sending each
other signals or killing each other.
* Setting a process to be undumpable is not irreversible. A sandboxed process
can make itself dumpable again, opening itself up to being taken over by
another process (either unsandboxed or within the same sandbox).
* Breakpad (the crash reporting tool) makes use of this. If a process
crashes, Breakpad makes it dumpable in order to use ptrace() to halt
threads and capture the process's state at the time of the crash. This
opens a small window of vulnerability.
## `setuid()` method
_This is an alternative to the `CLONE_NEWPID` method; it is not currently
implemented in the Chromium codebase._
Instead of using `CLONE_NEWPID`, the `SUID` helper can use `setuid()` to put the
process into a currently-unused UID, which is allocated out of a range of UIDs.
In order to ensure that the `UID` has not been allocated for another sandbox,
the `SUID` helper uses
[getrlimit()](http://www.kernel.org/doc/man-pages/online/pages/man2/getrlimit.2.html)
to set `RLIMIT_NPROC` temporarily to a soft limit of 1. (Note that the docs
specify that [setuid()](http://www.kernel.org/doc/man-pages/online/pages/man2/setuid.2.html)
returns `EAGAIN` if `RLIMIT_NPROC` is exceeded.) We can reset `RLIMIT_NPROC`
afterwards in order to allow the sandboxed process to fork child processes.
As before, the `SUID` helper chroots the process.
As before, LinuxZygote can set itself to be undumpable to stop processes in the
sandbox from being able to `ptrace()` each other.
Limitations:
* It is not possible for an unsandboxed process to `ptrace()` a sandboxed
process because they run under different UIDs. This makes debugging harder.
There is no equivalent of the `--allow-sandbox-debugging` other than turning
the sandbox off with `--no-sandbox`.
* The `SUID` helper can check that a `UID` is unused before it uses it (hence
this is safe if the `SUID` helper is installed into multiple chroots), but
it cannot prevent other root processes from putting processes into this
`UID` after the sandbox has been started. This means we should make the
`UID` range configurable, or distributions should reserve a `UID` range.
## `CLONE_NEWNET` method
The `SUID` helper uses
[CLONE_NEWNET](http://www.kernel.org/doc/man-pages/online/pages/man2/clone.2.html)
to restrict network access.
## Future work
We are splitting the `SUID` sandbox into a separate project which will support
both the `CLONE_NEWNS` and `setuid()` methods:
http://code.google.com/p/setuid-sandbox/
Having the `SUID` helper as a separate project should make it easier for
distributions to review and package.
## Possible extensions
## History
Older versions of the sandbox helper process will _only_ run
`/opt/google/chrome/chrome`. This string is hard coded
(`sandbox/linux/suid/sandbox.cc`). If your package is going to place the
Chromium binary somewhere else you need to modify this string.
## See also
* [LinuxSUIDSandboxDevelopment](linux_suid_sandbox_development.md)
* [LinuxSandboxing](linux_sandboxing.md)
* General information on Chromium sandboxing:
http://dev.chromium.org/developers/design-documents/sandbox
|