Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
fs: don't block i_writecount during exec
Back in 2021 we already discussed removing deny_write_access() for
executables. Back then I was hesistant because I thought that this might
cause issues in userspace. But even back then I had started taking some
notes on what could potentially depend on this and I didn't come up with
a lot so I've changed my mind and I would like to try this.
Here are some of the notes that I took:
(1) The deny_write_access() mechanism is causing really pointless issues
such as [1]. If a thread in a thread-group opens a file writable,
then writes some stuff, then closing the file descriptor and then
calling execve() they can fail the execve() with ETXTBUSY because
another thread in the thread-group could have concurrently called
fork(). Multi-threaded libraries such as go suffer from this.
(2) There are userspace attacks that rely on overwriting the binary of a
running process. These attacks are _mitigated_ but _not at all
prevented_ from ocurring by the deny_write_access() mechanism.
I'll go over some details. The clearest example of such attacks was
the attack against runC in CVE-2019-5736 (cf. [3]).
An attack could compromise the runC host binary from inside a
_privileged_ runC container. The malicious binary could then be used
to take over the host.
(It is crucial to note that this attack is _not_ possible with
unprivileged containers. IOW, the setup here is already insecure.)
The attack can be made when attaching to a running container or when
starting a container running a specially crafted image. For example,
when runC attaches to a container the attacker can trick it into
executing itself.
This could be done by replacing the target binary inside the
container with a custom binary pointing back at the runC binary
itself. As an example, if the target binary was /bin/bash, this
could be replaced with an executable script specifying the
interpreter path #!/proc/self/exe.
As such when /bin/bash is executed inside the container, instead the
target of /proc/self/exe will be executed. That magic link will
point to the runc binary on the host. The attacker can then proceed
to write to the target of /proc/self/exe to try and overwrite the
runC binary on the host.
However, this will not succeed because of deny_write_access(). Now,
one might think that this would prevent the attack but it doesn't.
To overcome this, the attacker has multiple ways:
* Open a file descriptor to /proc/self/exe using the O_PATH flag and
then proceed to reopen the binary as O_WRONLY through
/proc/self/fd/<nr> and try to write to it in a busy loop from a
separate process. Ultimately it will succeed when the runC binary
exits. After this the runC binary is compromised and can be used
to attack other containers or the host itself.
* Use a malicious shared library annotating a function in there with
the constructor attribute making the malicious function run as an
initializor. The malicious library will then open /proc/self/exe
for creating a new entry under /proc/self/fd/<nr>. It'll then call
exec to a) force runC to exit and b) hand the file descriptor off
to a program that then reopens /proc/self/fd/<nr> for writing
(which is now possible because runC has exited) and overwriting
that binary.
To sum up: the deny_write_access() mechanism doesn't prevent such
attacks in insecure setups. It just makes them minimally harder.
That's all.
The only way back then to prevent this is to create a temporary copy
of the calling binary itself when it starts or attaches to
containers. So what I did back then for LXC (and Aleksa for runC)
was to create an anonymous, in-memory file using the memfd_create()
system call and to copy itself into the temporary in-memory file,
which is then sealed to prevent further modifications. This sealed,
in-memory file copy is then executed instead of the original on-disk
binary.
Any compromising write operations from a privileged container to the
host binary will then write to the temporary in-memory binary and
not to the host binary on-disk, preserving the integrity of the host
binary. Also as the temporary, in-memory binary is sealed, writes to
this will also fail.
The point is that deny_write_access() is uselss to prevent these
attacks.
(3) Denying write access to an inode because it's currently used in an
exec path could easily be done on an LSM level. It might need an
additional hook but that should be about it.
(4) The MAP_DENYWRITE flag for mmap() has been deprecated a long time
ago so while we do protect the main executable the bigger portion of
the things you'd think need protecting such as the shared libraries
aren't. IOW, we let anyone happily overwrite shared libraries.
(5) We removed all remaining uses of VM_DENYWRITE in [2]. That means:
(5.1) We removed the legacy uselib() protection for preventing
overwriting of shared libraries. Nobody cared in 3 years.
(5.2) We allow write access to the elf interpreter after exec
completed treating it on a par with shared libraries.
Yes, someone in userspace could potentially be relying on this. It's not
completely out of the realm of possibility but let's find out if that's
actually the case and not guess.
Link: golang/go#22315 [1]
Link: 49624ef ("Merge tag 'denywrite-for-5.15' of git://github.com/davidhildenbrand/linux") [2]
Link: https://unit42.paloaltonetworks.com/breaking-docker-via-runc-explaining-cve-2019-5736 [3]
Link: https://lwn.net/Articles/866493
Link: golang/go#22220
Link: https://github.com/golang/go/blob/5bf8c0cf09ee5c7e5a37ab90afcce154ab716a97/src/cmd/go/internal/work/buildid.go#L724
Link: https://github.com/golang/go/blob/5bf8c0cf09ee5c7e5a37ab90afcce154ab716a97/src/cmd/go/internal/work/exec.go#L1493
Link: https://github.com/golang/go/blob/5bf8c0cf09ee5c7e5a37ab90afcce154ab716a97/src/cmd/go/internal/script/cmds.go#L457
Link: https://github.com/golang/go/blob/5bf8c0cf09ee5c7e5a37ab90afcce154ab716a97/src/cmd/go/internal/test/test.go#L1557
Link: https://github.com/golang/go/blob/5bf8c0cf09ee5c7e5a37ab90afcce154ab716a97/src/os/exec/lp_linux_test.go#L61
Link: buildkite/agent#2736
Link: rust-lang/rust#114554
Link: https://bugs.openjdk.org/browse/JDK-8068370
Link: dotnet/runtime#58964
Link: https://lore.kernel.org/r/20240531-vfs-i_writecount-v1-1-a17bea7ee36b@kernel.org
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>- Loading branch information
