Modifying a file and then executing it can fail on Linux #58964
Comments
|
I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label. |
dotnet-issue-labeler
bot
added
the
untriaged
|
Tagging subscribers to this area: @dotnet/area-system-io Issue Detailsgolang/go#22315 sets out the issue. Fundamentally, this seems like an inescapable race in multi-threaded programs on Linux. My understanding of the failure is this:
I don't see a way to completely prevent this race, as there's no way to avoid executing
|
We do that already. We use
This is just a tip of the iceberg that we happened to hit in CI. There are likely many other APIs that can fail in a bad way in this situation... |
|
It may be worth double checking that the detection of |
I checked, and it works as expected.
There should be no issue on Linux, since that supports Have you seen the issue occur? On what platform? |
adamsitnik
removed
the
untriaged
|
Thanks all -- I looked at the pal and did indeed see CLOEXEC but I wasn't certain that code was a bottleneck for the entire product.
Yes, this is happening regularly in the Linux installer tests. To be clear, CLOEXEC does not solve the problem, only minimizes the race. As far as I can tell, there is no solution. The race is an inherent design flaw in the Linux process model. |
|
Got it. The issue is that the file handle is still referenced by processes being started which are between The unix process implementation has a When That should be a way to fix this. You can assign this to me. |
ghost
added
the
in-pr
agocke
added a commit
that referenced
this issue
Sep 29, 2021
ghost
removed
the
in-pr
agocke
added a commit
that referenced
this issue
Jan 5, 2022
This was referenced Apr 25, 2022
agocke
changed the title
brauner
added a commit
to brauner/linux
that referenced
this issue
Jun 3, 2024
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/[email protected]
Reviewed-by: Josef Bacik <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
brauner
added a commit
to brauner/linux
that referenced
this issue
Jun 10, 2024
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/[email protected]
Reviewed-by: Josef Bacik <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
josefbacik
pushed a commit
to josefbacik/linux
that referenced
this issue
Jul 12, 2024
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
Signed-off-by: Christian Brauner <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
golang/go#22315 sets out the issue.
Fundamentally, this seems like an inescapable race in multi-threaded programs on Linux.
My understanding of the failure is this:
forkcall.I don't see a way to completely prevent this race, as there's no way to avoid executing
forkto create a new process on Linux. However, there are two mitigations we could consider:exec. Since the "new process" procedure in Unix isfork+exec, this would narrow the window of the race to just afterforkand just beforeexec. It may be possible to always set this flag when opening file descriptors on Linux from managed code -- we don't supportforkwithoutexecand inheriting non-tty open file descriptors does not appear to be in any .NET contract.ETXTBSY. This could workaround the issue, at the expense of taking longer to fail in a real contention situation.The text was updated successfully, but these errors were encountered: