Possibility of merge of disable icotl TIOCSTI patch

Possibility of merge of disable icotl TIOCSTI patch

[Simon Brand]

Hi,

in the past there have been attempts to restrict the TIOCSTI ioctl. [0, 1]
None of them are present in the current kernel.
Since those tries there have been some security issues (sandbox
escapes in flatpak (CVE-2019-10063) [2] and snap (CVE 2019-7303) [3],
runuser [4], su [5]).

I would provide a patch which leaves the current behavior as default,
but TIOCSTI can be disabled via Kconfig or cmdline switch.
Is there any chance this will get merged in 2022, since past
attempts failed?

Escapes can be reproduced easiliy (on archlinux) via a python script:
```
import fcntl
import termios
with open("/dev/tty", "w") as fd:
    for c in "id\n":
        fcntl.ioctl(fd, termios.TIOCSTI, c)
```
Now run as root:
# su user
$ python3 /path/to/script.py ; exit
uid=0(root) ...

Best,
Simon


[0] https://lkml.kernel.org/lkml/CAG48ez1NBnrsPnHN6D9nbOJP6+Q6zEV9vfx9q7ME4Eti-vRmhQ@mail.gmail.com/T/
[1] https://lkml.kernel.org/lkml/20170420174100.GA16822@mail.hallyn.com/T/
[2] https://github.com/flatpak/flatpak/issues/2782
[3] https://wiki.ubuntu.com/SecurityTeam/KnowledgeBase/SnapIoctlTIOCSTI
[4] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815922
[5] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=628843

_______________________________________________
Kernelnewbies mailing list
Kernelnewbies@kernelnewbies.org
https://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies

Re: Possibility of merge of disable icotl TIOCSTI patch

[Yann Droneaud]

Hi,

Le 24/05/2022 à 13:10, Simon Brand a écrit :
> Hi,
>
> in the past there have been attempts to restrict the TIOCSTI ioctl. [0, 1]
> None of them are present in the current kernel.
> Since those tries there have been some security issues (sandbox
> escapes in flatpak (CVE-2019-10063) [2] and snap (CVE 2019-7303) [3],
> runuser [4], su [5]).
>
> I would provide a patch which leaves the current behavior as default,
> but TIOCSTI can be disabled via Kconfig or cmdline switch.
> Is there any chance this will get merged in 2022, since past
> attempts failed?
>
> Escapes can be reproduced easiliy (on archlinux) via a python script:
> ```
> import fcntl
> import termios
> with open("/dev/tty", "w") as fd:
>      for c in "id\n":
>          fcntl.ioctl(fd, termios.TIOCSTI, c)
> ```
> Now run as root:
> # su user
> $ python3 /path/to/script.py ; exit
> uid=0(root) ...
>
> Best,
> Simon
>
>
> [0] https://lkml.kernel.org/lkml/CAG48ez1NBnrsPnHN6D9nbOJP6+Q6zEV9vfx9q7ME4Eti-vRmhQ@mail.gmail.com/T/
> [1] https://lkml.kernel.org/lkml/20170420174100.GA16822@mail.hallyn.com/T/
> [2] https://github.com/flatpak/flatpak/issues/2782
> [3] https://wiki.ubuntu.com/SecurityTeam/KnowledgeBase/SnapIoctlTIOCSTI
> [4] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815922
> [5] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=628843
>

This is probably some topic for (kernel|linux)-hardening@ mailing lists.


Regards.

-- 

Yann Droneaud

OPTEYA



_______________________________________________
Kernelnewbies mailing list
Kernelnewbies@kernelnewbies.org
https://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies

Re: Possibility of merge of disable icotl TIOCSTI patch

[Yann Droneaud]

Hi,

Le 24/05/2022 à 13:10, Simon Brand a écrit :
> Hi,
>
> in the past there have been attempts to restrict the TIOCSTI ioctl. [0, 1]
> None of them are present in the current kernel.
> Since those tries there have been some security issues (sandbox
> escapes in flatpak (CVE-2019-10063) [2] and snap (CVE 2019-7303) [3],
> runuser [4], su [5]).
>
> I would provide a patch which leaves the current behavior as default,
> but TIOCSTI can be disabled via Kconfig or cmdline switch.
> Is there any chance this will get merged in 2022, since past
> attempts failed?
>
> Escapes can be reproduced easiliy (on archlinux) via a python script:
> ```
> import fcntl
> import termios
> with open("/dev/tty", "w") as fd:
>      for c in "id\n":
>          fcntl.ioctl(fd, termios.TIOCSTI, c)
> ```
> Now run as root:
> # su user
> $ python3 /path/to/script.py ; exit
> uid=0(root) ...
>
> Best,
> Simon
>
>
> [0] https://lkml.kernel.org/lkml/CAG48ez1NBnrsPnHN6D9nbOJP6+Q6zEV9vfx9q7ME4Eti-vRmhQ@mail.gmail.com/T/
> [1] https://lkml.kernel.org/lkml/20170420174100.GA16822@mail.hallyn.com/T/
> [2] https://github.com/flatpak/flatpak/issues/2782
> [3] https://wiki.ubuntu.com/SecurityTeam/KnowledgeBase/SnapIoctlTIOCSTI
> [4] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=815922
> [5] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=628843
>

This is probably some topic for (kernel|linux)-hardening@ mailing lists.


Regards.

-- 

Yann Droneaud

OPTEYA

Re: Possibility of merge of disable icotl TIOCSTI patch

[Levente Polyak]

On 6/1/22 17:41, Yann Droneaud wrote:
>> I would provide a patch which leaves the current behavior as default,
>> but TIOCSTI can be disabled via Kconfig or cmdline switch.
>> Is there any chance this will get merged in 2022, since past
>> attempts failed?
>>

Small side note:

A complete version of Matt's initial patch has lived on in 
linux-hardened [0][1] with the `SECURITY_TIOCSTI_RESTRICT` Kconfig 
(default no) and a `tiocsti_restrict` sysctl.

If a re-attempt is feasible, both patchs [0][1] could potentially be 
re-proposed as is.

In linux-hardened we have an independent patch [2] which simply sets the 
default value of `SECURITY_TIOCSTI_RESTRICT` to `yes`, but that most 
likely is not desired.

cheers,
Levente


[0] 
https://github.com/anthraxx/linux-hardened/commit/d0e49deb1a39dc64e7c7db3340579cfc9ab1e0df
[1] 
https://github.com/anthraxx/linux-hardened/commit/ea8f20602a993c90125bf08da39894f01166dc73
[2] 
https://github.com/anthraxx/linux-hardened/commit/238551f7b6a138d6f9ba0d55fe70cf6ddc237f47