Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.debian.kernel > #79655 > unrolled thread

Bug#1041552: HFS/HFS+ are insecure

Back to article view | Back to linux.debian.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Bug#1041552: HFS/HFS+ are insecure Marco d'Itri <md@Linux.IT> - 2023-07-20 20:10 +0200
    Re: HFS/HFS+ are insecure Marco d'Itri <md@Linux.IT> - 2023-07-21 11:00 +0200
      Re: HFS/HFS+ are insecure Magissia <debianlist@magissia.com> - 2023-07-21 11:10 +0200
      Re: HFS/HFS+ are insecure Martin Steigerwald <martin@lichtvoll.de> - 2023-07-21 12:10 +0200
      Re: HFS/HFS+ are insecure Bastien Roucariès <rouca@debian.org> - 2023-07-21 13:00 +0200
        Re: HFS/HFS+ are insecure Bastien Roucariès <rouca@debian.org> - 2023-07-21 13:10 +0200
      Re: HFS/HFS+ are insecure Matthew Garrett <mjg59@srcf.ucam.org> - 2023-07-21 19:40 +0200
        Re: HFS/HFS+ are insecure Ben Hutchings <ben@decadent.org.uk> - 2023-07-23 02:40 +0200
    Processed: Re: HFS/HFS+ are insecure "Debian Bug Tracking System" <owner@bugs.debian.org> - 2023-08-27 02:40 +0200
    Bug#1041552: HFS/HFS+ are insecure Marco d'Itri <md@Linux.IT> - 2023-08-27 02:40 +0200

#79655 — Bug#1041552: HFS/HFS+ are insecure

[Multipart message — attachments visible in raw view] — view raw

Package: src:linux
Severity: normal

You are totally correct.
Kernel team, please blacklist HFS/HFS+ for automounting.

On Jul 20, John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> wrote:

> Hello!
> 
> On Thu, 2023-07-20 at 18:30 +0100, Matthew Wilcox wrote:
> > On Thu, Jul 20, 2023 at 05:27:57PM +0200, Dmitry Vyukov wrote:
> > > On Thu, 5 Jan 2023 at 17:45, Viacheslav Dubeyko <slava@dubeyko.com> wrote:
> > > > > On Wed, Jan 04, 2023 at 08:37:16PM -0800, Viacheslav Dubeyko wrote:
> > > > > > Also, as far as I can see, available volume in report (mount_0.gz) somehow corrupted already:
> > > > > 
> > > > > Syzbot generates deliberately-corrupted (aka fuzzed) filesystem images.
> > > > > So basically, you can't trust anything you read from the disc.
> > > > > 
> > > > 
> > > > If the volume has been deliberately corrupted, then no guarantee that file system
> > > > driver will behave nicely. Technically speaking, inode write operation should never
> > > > happened for corrupted volume because the corruption should be detected during
> > > > b-tree node initialization time. If we would like to achieve such nice state of HFS/HFS+
> > > > drivers, then it requires a lot of refactoring/implementation efforts. I am not sure that
> > > > it is worth to do because not so many guys really use HFS/HFS+ as the main file
> > > > system under Linux.
> > > 
> > > 
> > > Most popular distros will happily auto-mount HFS/HFS+ from anything
> > > inserted into USB (e.g. what one may think is a charger). This creates
> > > interesting security consequences for most Linux users.
> > > An image may also be corrupted non-deliberately, which will lead to
> > > random memory corruptions if the kernel trusts it blindly.
> > 
> > Then we should delete the HFS/HFS+ filesystems.  They're orphaned in
> > MAINTAINERS and if distros are going to do such a damnfool thing,
> > then we must stop them.
> 
> Both HFS and HFS+ work perfectly fine. And if distributions or users are so
> sensitive about security, it's up to them to blacklist individual features
> in the kernel.
> 
> Both HFS and HFS+ have been the default filesystem on MacOS for 30 years
> and I don't think it's justified to introduce such a hard compatibility
> breakage just because some people are worried about theoretical evil
> maid attacks.
> 
> HFS/HFS+ mandatory if you want to boot Linux on a classic Mac or PowerMac
> and I don't think it's okay to break all these systems running Linux.
> 
> Thanks,
> Adrian
> 
> -- 
>  .''`.  John Paul Adrian Glaubitz
> : :' :  Debian Developer
> `. `'   Physicist
>   `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

-- 
ciao,
Marco

[toc] | [next] | [standalone]

#79658 — Re: HFS/HFS+ are insecure

[Multipart message — attachments visible in raw view] — view raw

On Jul 21, Matthew Garrett <mjg59@srcf.ucam.org> wrote:

> > You are totally correct.
> > Kernel team, please blacklist HFS/HFS+ for automounting.
> Isn't this a userland policy decision? udisks will happily trigger a 
> module load for hfsplus if udev has identified it, and I don't think 
> there's a trivial mechanism for the kernel to disable that. I believe 
Yes, I was also thinking about this and I believe that you are right.
The kernel team did this in the past for some uncommon network 
protocols, but they could do it themselves because these modules are 
autoloaded using aliases.

Since I happen to be the kmod maintainer it looks like that solving this
is on me. :-)

Unless somebody has a better idea then then my plan is to ship in the 
next upload of kmod a file in /etc/modprobe.d/ which uses the blacklist 
directive to prevent automatically loading some file system modules.

By looking at the MAINTAINERS file I have identified these file systems 
marked as "orphan" and "odd fixes":

efs
hfs
hfaplus
qnx6
sysv

affs
ecryptfs
jffs2
jfs

And I think that I can also safely add a few more which while actively 
maintained I believe are only used in a retrocomputing context or are 
generally uncommon anyway:

befs
bfs
hpfs
omfs
qnx4
reiserfs
spu
ufs

Did i miss anything?

I think that all of these have enough of a niche usage that it would not 
be an unreasonable burden for the affected users to manually load the 
modules when needed (ad hoc or using /etc/modules-load.d/).

-- 
ciao,
Marco

[toc] | [prev] | [next] | [standalone]

#79659 — Re: HFS/HFS+ are insecure

Looks reasonable.

Le vendredi 21 juillet 2023 à 10:55 +0200, Marco d'Itri a écrit :
> On Jul 21, Matthew Garrett <mjg59@srcf.ucam.org> wrote:
> 
> > > You are totally correct.
> > > Kernel team, please blacklist HFS/HFS+ for automounting.
> > 
> > Isn't this a userland policy decision? udisks will happily trigger
> > a 
> > module load for hfsplus if udev has identified it, and I don't
> > think 
> > there's a trivial mechanism for the kernel to disable that. I
> > believe 
> 
> Yes, I was also thinking about this and I believe that you are right.
> The kernel team did this in the past for some uncommon network 
> protocols, but they could do it themselves because these modules are 
> autoloaded using aliases.
> 
> Since I happen to be the kmod maintainer it looks like that solving
> this
> is on me. :-)
> 
> Unless somebody has a better idea then then my plan is to ship in
> the 
> next upload of kmod a file in /etc/modprobe.d/ which uses the
> blacklist 
> directive to prevent automatically loading some file system modules.
> 
> By looking at the MAINTAINERS file I have identified these file
> systems 
> marked as "orphan" and "odd fixes":
> 
> efs
> hfs
> hfaplus
> qnx6
> sysv
> 
> affs
> ecryptfs
> jffs2
> jfs
> 
> And I think that I can also safely add a few more which while
> actively 
> maintained I believe are only used in a retrocomputing context or
> are 
> generally uncommon anyway:
> 
> befs
> bfs
> hpfs
> omfs
> qnx4
> reiserfs
> spu
> ufs
> 
> Did i miss anything?
> 
> I think that all of these have enough of a niche usage that it would
> not 
> be an unreasonable burden for the affected users to manually load
> the 
> modules when needed (ad hoc or using /etc/modules-load.d/).
> 

[toc] | [prev] | [next] | [standalone]

#79660 — Re: HFS/HFS+ are insecure

Hi Marco, hi,

Marco d'Itri - 21.07.23, 10:55:39 CEST:
> On Jul 21, Matthew Garrett <mjg59@srcf.ucam.org> wrote:
> > > You are totally correct.
> > > Kernel team, please blacklist HFS/HFS+ for automounting.
> > 
> > Isn't this a userland policy decision? udisks will happily trigger a
> > module load for hfsplus if udev has identified it, and I don't think
> > there's a trivial mechanism for the kernel to disable that. I
> > believe
> 
> Yes, I was also thinking about this and I believe that you are right.
> The kernel team did this in the past for some uncommon network
> protocols, but they could do it themselves because these modules are
> autoloaded using aliases.
> 
> Since I happen to be the kmod maintainer it looks like that solving
> this is on me. :-)
> 
> Unless somebody has a better idea then then my plan is to ship in the
> next upload of kmod a file in /etc/modprobe.d/ which uses the
> blacklist directive to prevent automatically loading some file system
> modules.
[…]
> I think that all of these have enough of a niche usage that it would
> not be an unreasonable burden for the affected users to manually load
> the modules when needed (ad hoc or using /etc/modules-load.d/).

In case you do this, I'd like there to be a NEWS.Debian entry about 
this, explaining both the justification behind it and how people can work 
around it. It could go like this:

-----------------------------------------------------------------------

Since version xyz of kmod the file /etc/modprobe.d/block-unsafe-
filesystems.conf prevents loading of several filesystem modules that are 
automatically loaded by udev when inserting a medium that contains one 
of them. These filesystems are either known to be unsafe or are not 
maintained actively anymore. A deliberately corrupted filesystem 
structure could trigger the filesystem driver in the module to crash, 
corrupt memory of other kernel components or to cause other problems. 
[Adjust to whatever risks are most likely to occur] [Add some links here 
for the discussion about that]

In case you rely on using one or more of these filesystems you can either 
edit the file /etc/modprobe.d/block-unsafe-filesystems.conf and put a 
comment sign before the filesystem in question or add the filesystem to a 
file to a file in /etc/modules-load.d/. [Please clarify here as needed]

Please take care not to plug in any device that you do not trust.

-----------------------------------------------------------------------

This is just a rough idea it probably needs several iterations to obtain 
a good wording that balances on assessing the risk correctly (without 
over or under estimating it). Also the method of circumventing the 
blocking may need further explanations. I am not using systemd, so I can 
not describe exactly how modules-load.d works. In case you like to use 
any of the above wording, feel free to use it under the license of the 
packaging of kmod.


I wonder about other kernel modules in other areas of the kernel that 
may be automatically loaded when connecting some hardware… especially 
some random USB device, but… that appears to me like opening a huge can 
of worms. I bet the Linux kernel has more than several hundreds of 
specialized USB drivers maybe even more than thousand meanwhile despite 
all the USB standards out there?

Linux is not a micro kernel. It was not designed to run drivers in a 
restricted and (somewhat) safe environment to begin with. That means 
ideally you'd have to audit all the drivers for security issues 
regularly or at least after a certain amount of changes made to it. In 
case you do not, for some random driver it will be difficult to know for 
sure whether it is safe or unsafe to use. Maybe some small filesystem 
driver like affs that still receives a patch every now and then is safer 
to use than the much more complex BTRFS filesystem driver.¹ Who knows? Of 
cause some fuzzing may really help. But it is not a guarantee either.

And then what about other kernel functionally that is loaded as module 
on demand that is only rarely used by some people?

So I wonder what body of evidence there is to base a policy decision on 
which driver to load or not to load automatically. Without a reliable 
body of evidence there is always the risk to either over or under policy 
the users of Debian and derived distributions (whose maintainers do not 
decide to change such a policy decision again). So I'd argue against 
taking the quick route on that to allow the time for a more informed 
decision. Maybe start with clear-cut cases likely probably HFS/HFS+ 
instead of just adding all kinds of other filesystems without even know 
whether there is a known exploit. Of course you could go by maintenance 
status, however, this can be inaccurate. How to do accurately determine 
maintenance status, especially as MAINTAINERS file may not be accurate or 
up-to-date at all times? And how many specialized USB drivers are there 
that are compiled as modules on Debian kernels that may not be 
maintained as well? Disable all of them by default?

Risk assessment is very challenging here, if you ask me.


[1] After a long time finally some RDB partitioning overflow fixes went 
into the kernel. The overflow bug could have caused overwriting of data 
at a different place of the disk on a different partition than intended 
which could mean a two-way data loss of both the written data and the 
data that was accidentally overwritten. While this theoretically could 
have been used for an exploit by deliberately causing the kernel to 
overwrite data at a certain location of the disk, I am not aware of any 
existing exploit for this. Such an exploit would only target a quite low 
amount of users I bet. See:

Partitions: Amiga RDB partition on 2 TB disk way too big, while OK in 
AmigaOS 4.1

https://bugzilla.kernel.org/show_bug.cgi?id=43511

Plus a ton of discussions on various mailing lists, hopefully all linked 
in above bug report.

Thanks,
-- 
Martin

[toc] | [prev] | [next] | [standalone]

#79661 — Re: HFS/HFS+ are insecure

[Multipart message — attachments visible in raw view] — view raw

Le vendredi 21 juillet 2023, 08:55:39 UTC Marco d'Itri a écrit :
> efs
https://pypi.org/project/qnxmount/ claim to mount it. Check
> hfs
https://github.com/0x09/hfsfuse
> hfaplus
https://github.com/0x09/hfsfuse
> qnx6
Fuse ro filesystem https://pypi.org/project/qnxmount/ better support then kernel one
> sysv
no fuse equivalent may be easier to port from (net)?bsd source to fuse or use the method of qnxmount
> 
> affs
no fuse equivalent but a grub filesystem may be ported to fuse
> ecryptfs
no fuse equivalent
> jffs2
no fuse equivalent
> jfs
no fuse equivalent but a grub filesystem may be ported to fuse
> 
> And I think that I can also safely add a few more which while actively 
> maintained I believe are only used in a retrocomputing context or are 
> generally uncommon anyway:
> 
> befs
A fuse module seems to be avalaible https://www.haiku-os.org/guides/daily-tasks/access_bfs_with_fuse/
> bfs
no fuse equivalent
> hpfs
no fuse
> omfs
https://github.com/bcopeland/omfs_fuse/
> qnx4
Maybe Fuse ro filesystem https://pypi.org/project/qnxmount/
> reiserfs
so incomplete work using grub filesystem https://github.com/albertz/reiserfs-fuse
> spu
This one is a virtual fs for powerpc should be dropped of the list
> ufs
https://github.com/mkatiyar/fuse-ufs2
> 

[toc] | [prev] | [next] | [standalone]

#79662 — Re: HFS/HFS+ are insecure

[Multipart message — attachments visible in raw view] — view raw

Le vendredi 21 juillet 2023, 10:52:17 UTC Bastien Roucariès a écrit :
> Le vendredi 21 juillet 2023, 08:55:39 UTC Marco d'Itri a écrit :
> > efs
> https://pypi.org/project/qnxmount/ claim to mount it. Check
> > hfs
> https://github.com/0x09/hfsfuse
Corrected not supported by this package may be emulated by using user space hfs tools ?
> > hfaplus
> https://github.com/0x09/hfsfuse
> > qnx6
> Fuse ro filesystem https://pypi.org/project/qnxmount/ better support then kernel one
> > sysv
> no fuse equivalent may be easier to port from (net)?bsd source to fuse or use the method of qnxmount
> > 
> > affs
> no fuse equivalent but a grub filesystem may be ported to fuse
> > ecryptfs
> no fuse equivalent
> > jffs2
> no fuse equivalent
> > jfs
> no fuse equivalent but a grub filesystem may be ported to fuse
> > 
> > And I think that I can also safely add a few more which while actively 
> > maintained I believe are only used in a retrocomputing context or are 
> > generally uncommon anyway:
> > 
> > befs
> A fuse module seems to be avalaible https://www.haiku-os.org/guides/daily-tasks/access_bfs_with_fuse/
> > bfs
> no fuse equivalent
> > hpfs
> no fuse
> > omfs
> https://github.com/bcopeland/omfs_fuse/
> > qnx4
> Maybe Fuse ro filesystem https://pypi.org/project/qnxmount/
> > reiserfs
> so incomplete work using grub filesystem https://github.com/albertz/reiserfs-fuse
> > spu
> This one is a virtual fs for powerpc should be dropped of the list
> > ufs
> https://github.com/mkatiyar/fuse-ufs2
> > 
> 
> 

[toc] | [prev] | [next] | [standalone]

#79665 — Re: HFS/HFS+ are insecure

On Fri, Jul 21, 2023 at 10:55:39AM +0200, Marco d'Itri wrote:

> Unless somebody has a better idea then then my plan is to ship in the 
> next upload of kmod a file in /etc/modprobe.d/ which uses the blacklist 
> directive to prevent automatically loading some file system modules.

I think this would break any existing fstab entries that reference hfs 
and hfsplus, and the convenient way to integrate Linux boot with x86 
Macs is certainly to have an hfsplus EFI partition so this may be a 
legitimate use-case. It also means that anyone who has a need to use one 
of these filesystems in a static manner is vulnerable to automount 
attacks using them.

Completely untested, but I think something along the lines of:

SUBSYSTEM!="block", GOTO="udisks_insecure_fs_end"
ENV{ID_FS_TYPE}=="hfs", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="hfsplus", ENV{UDISKS_AUTO}="0"
LABEL="udisks_insecure_fs_end"

in a udev fragment should work? Any static fstab or mount units should 
still work, but it should disable udisks automounting regardless of the 
desktop agent involved, even if the fs modules are already loaded.

[toc] | [prev] | [next] | [standalone]

#79694 — Re: HFS/HFS+ are insecure

[Multipart message — attachments visible in raw view] — view raw

On Fri, 2023-07-21 at 18:35 +0100, Matthew Garrett wrote:
> On Fri, Jul 21, 2023 at 10:55:39AM +0200, Marco d'Itri wrote:
> 
> > Unless somebody has a better idea then then my plan is to ship in the 
> > next upload of kmod a file in /etc/modprobe.d/ which uses the blacklist 
> > directive to prevent automatically loading some file system modules.
> 
> I think this would break any existing fstab entries that reference hfs 
> and hfsplus, and the convenient way to integrate Linux boot with x86 
> Macs is certainly to have an hfsplus EFI partition so this may be a 
> legitimate use-case. It also means that anyone who has a need to use one 
> of these filesystems in a static manner is vulnerable to automount 
> attacks using them.

Right, auto-loading of filesystems has to keep working.  And since
mount() of arbitrary filesystems is restricted to root (CAP_NET_ADMIN
in the initial namespace), we should let the callers apply a block- or
allow-list.

The reason we have to disable auto-loading of network protocols is that
socket creation is generally an unprivileged operation, so there's no
trusted user-space that can apply the policy (besides kmod).

> Completely untested, but I think something along the lines of:
> 
> SUBSYSTEM!="block", GOTO="udisks_insecure_fs_end"
> ENV{ID_FS_TYPE}=="hfs", ENV{UDISKS_AUTO}="0"
> ENV{ID_FS_TYPE}=="hfsplus", ENV{UDISKS_AUTO}="0"
> LABEL="udisks_insecure_fs_end"
> 
> in a udev fragment should work? Any static fstab or mount units should 
> still work, but it should disable udisks automounting regardless of the 
> desktop agent involved, even if the fs modules are already loaded.

I agree we should not have UDisks probing for any of the (many) kernel
filesystems that aren't being actively maintained including responding
to security issues.

Beyond that, I would also like to see libmount limiting the filesystems
that it will probe when the fstab type is "auto".  But since UDisks
normally handles mounting for unprivileged users, that's probably less
of a concern.

Ben.

-- 
Ben Hutchings
If you seem to know what you are doing, you'll be given more to do.

[toc] | [prev] | [next] | [standalone]

#80195 — Processed: Re: HFS/HFS+ are insecure

Processing control commands:

> reassign -1 udisks2
Bug #1041552 [src:linux] HFS/HFS+ are insecure
Bug reassigned from package 'src:linux' to 'udisks2'.
Ignoring request to alter found versions of bug #1041552 to the same values previously set
Ignoring request to alter fixed versions of bug #1041552 to the same values previously set
> retitle -1 do not mount automatically unmaintained file systems
Bug #1041552 [udisks2] HFS/HFS+ are insecure
Changed Bug title to 'do not mount automatically unmaintained file systems' from 'HFS/HFS+ are insecure'.

-- 
1041552: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1041552
Debian Bug Tracking System
Contact owner@bugs.debian.org with problems

[toc] | [prev] | [next] | [standalone]

#80196

[Multipart message — attachments visible in raw view] — view raw

Control: reassign -1 udisks2
Control: retitle -1 do not mount automatically unmaintained file systems

On Jul 20, md wrote:

> You are totally correct.
> Kernel team, please blacklist HFS/HFS+ for automounting.
As discussed on debian-devel@, this policy should not be handled by the 
kernel because modules autoloading of file systems drivers should not be
disabled.

So I propose this content for a file like
/usr/lib/udev/rules.d/75-insecure-fs.rules:

# Do not automatically mount these file systems because their drivers are
# marked as "orphan" or "odd fixes" in the kernel MAINTAINERS file and so
# are more at risk of having security-sensitive defects which could be
# exploited by a crafted file system.
SUBSYSTEM!="block", GOTO="udisks_insecure_fs_end"

ENV{ID_FS_TYPE}=="affs", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="ecryptfs", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="efs", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="hfs", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="hfsplus", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="jffs2", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="jfs", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="qnx6", ENV{UDISKS_AUTO}="0"
ENV{ID_FS_TYPE}=="sysv", ENV{UDISKS_AUTO}="0"

LABEL="udisks_insecure_fs_end"

-- 
ciao,
Marco

[toc] | [prev] | [standalone]

Back to top | Article view | linux.debian.kernel

csiph-web