security: new security_file_ioctl_compat() hook

Commit Message

Alfred Piccioni Dec. 19, 2023, 9:09 a.m. UTC

Some ioctl commands do not require ioctl permission, but are routed to
other permissions such as FILE_GETATTR or FILE_SETATTR. This routing is
done by comparing the ioctl cmd to a set of 64-bit flags (FS_IOC_*).

However, if a 32-bit process is running on a 64-bit kernel, it emits
32-bit flags (FS_IOC32_*) for certain ioctl operations. These flags are
being checked erroneously, which leads to these ioctl operations being
routed to the ioctl permission, rather than the correct file
permissions.

This was also noted in a RED-PEN finding from a while back -
"/* RED-PEN how should LSM module know it's handling 32bit? */".

This patch introduces a new hook, security_file_ioctl_compat, that is
called from the compat ioctl syscall. All current LSMs have been changed
to support this hook.

Reviewing the three places where we are currently using
security_file_ioctl, it appears that only SELinux needs a dedicated
compat change; TOMOYO and SMACK appear to be functional without any
change.

Fixes: 0b24dcb7f2f7 ("Revert "selinux: simplify ioctl checking"")
Signed-off-by: Alfred Piccioni <[e-mail]>
Cc: [e-mail]
---
 fs/ioctl.c                    |  3 +--
 include/linux/lsm_hook_defs.h |  2 ++
 include/linux/security.h      |  7 +++++++
 security/security.c           | 17 +++++++++++++++++
 security/selinux/hooks.c      | 28 ++++++++++++++++++++++++++++
 security/smack/smack_lsm.c    |  1 +
 security/tomoyo/tomoyo.c      |  1 +
 7 files changed, 57 insertions(+), 2 deletions(-)


base-commit: 196e95aa8305aecafc4e1857b7d3eff200d953b6

Comments

Alfred Piccioni Dec. 19, 2023, 9:10 a.m. UTC | #1

Thanks for taking the time to review! Apologies for the number of
small mistakes.

> s/syscal/syscall/
> Might to consider checking using codespell to catch such things
> although it is imperfect.

Fixed, loaded codespell.

> Paul doesn't like C++-style comments so rewrite using kernel coding
> style for multi-line comments or drop.
> I don't think kernel coding style strictly prohibits use for
> single-line comments and it isn't detected by checkpatch.pl but he has
> previously
> raised this on other patches. I actually like the C++-style comments
> for one-liners especially for comments at the end of a line of code
> but Paul is the maintainer so he gets the final word.

Changed to /**/ style comments. No particular preference on my side
for comment structure, just used to C++/Java style.

> Sorry, missed this the first time but cut-and-paste error above:
> s/GETFLAGS/SETFLAGS/

Egads. Fixed.

> Also, IIRC, Paul prefers putting a pair of parentheses after function
> names to distinguish them, so in the subject line
> and description it should be security_file_ioctl_compat() and
> security_file_ioctl(), and you should put a patch version
> in the [PATCH] prefix e.g. [PATCH v3] to make clear that it is a later
> version, and usually one doesn't capitalize SELinux
> or the leading verb in the subject line (just "selinux: introduce").

Changed title to lower-case, prefixed with security, changed slightly
to fit in summary with new parentheses. Added [PATCH V3] to the
subject.

> Actually, since this spans more than just SELinux, the prefix likely
> needs to reflect that (e.g. security: introduce ...)
> and the patch should go to the linux-security-module mailing list too
> and perhaps linux-fsdevel for the ioctl change.

Added cc '[e-mail]' and cc
'[e-mail]'. Thanks!

> I didn't do an audit but does anything need to be updated for the BPF
> LSM or does it auto-magically pick up new hooks?

I'm unsure. I looked through the BPF LSM and I can't see any way it's
picking up the file_ioctl hook to begin with. It appears to me
skimming through the code that it automagically picks it up, but I'm
not willing to bet the kernel on it.

Do you know who would be a good person to ask about this to make sure?

Alfred Piccioni Dec. 20, 2023, 2:38 p.m. UTC | #2

>> By the way, for extra credit, you could augment the ioctl tests in the
>> selinux-testsuite to also exercise this new hook and confirm that it
>> works correctly. See
>> https://github.com/SELinuxProject/selinux-testsuite particularly
>> tests/ioctl and policy/test_ioctl.te. Feel free to ask for help on
>> that.

> I do like extra credit. I'll take a look and see if it's something I
> can tackle. I'm primarily doing ad hoc checks on Android devices, so
> I'm unsure how easy it will be for me to run the suite. I'll get back
> to you shortly on that.

In response to myself, I unfortunately won't have time to do the
testing updates this year. If someone else wants to help, that'd be
great! Otherwise, I'll take a look next year after vacation and see if
I can take a crack at it. Thanks!

Stephen Smalley Dec. 20, 2023, 3:34 p.m. UTC | #3

On Tue, Dec 19, 2023 at 4:11 AM Alfred Piccioni <[e-mail]> wrote:
>
> Thanks for taking the time to review! Apologies for the number of
> small mistakes.

NP.

> > Also, IIRC, Paul prefers putting a pair of parentheses after function
> > names to distinguish them, so in the subject line
> > and description it should be security_file_ioctl_compat() and
> > security_file_ioctl(), and you should put a patch version
> > in the [PATCH] prefix e.g. [PATCH v3] to make clear that it is a later
> > version, and usually one doesn't capitalize SELinux
> > or the leading verb in the subject line (just "selinux: introduce").
>
> Changed title to lower-case, prefixed with security, changed slightly
> to fit in summary with new parentheses. Added [PATCH V3] to the
> subject.

Patch description still doesn't include the parentheses after each
function name but probably not worth re-spinning unless Paul says to
do so. I don't see the v3 in the subject line. Seemingly that in
combination with the fact that you replied to the original thread
confuses the b4 tool (b4.docs.kernel.org) such that b4 mbox/am/shazam
ends up selecting the v2 patch instead by default.

> > Actually, since this spans more than just SELinux, the prefix likely
> > needs to reflect that (e.g. security: introduce ...)
> > and the patch should go to the linux-security-module mailing list too
> > and perhaps linux-fsdevel for the ioctl change.
>
> Added cc '[e-mail]' and cc
> '[e-mail]'. Thanks!

Just FYI, scripts/get_maintainer.pl /path/to/patch will provide an
over-approximation of who to include on the distribution for patches
based on MAINTAINERS and recent committers. That said, I generally
prune the set it provides. More art than science.

> > I didn't do an audit but does anything need to be updated for the BPF
> > LSM or does it auto-magically pick up new hooks?
>
> I'm unsure. I looked through the BPF LSM and I can't see any way it's
> picking up the file_ioctl hook to begin with. It appears to me
> skimming through the code that it automagically picks it up, but I'm
> not willing to bet the kernel on it.
>
> Do you know who would be a good person to ask about this to make sure?

Looks like it inherited it via the lsm_hook_defs.h.
$ nm security/bpf/hooks.o | grep ioctl
                 U bpf_lsm_file_ioctl
                 U bpf_lsm_file_ioctl_compat

Stephen Smalley Dec. 20, 2023, 5:31 p.m. UTC | #4

On Tue, Dec 19, 2023 at 4:09 AM Alfred Piccioni <[e-mail]> wrote:
>
> Some ioctl commands do not require ioctl permission, but are routed to
> other permissions such as FILE_GETATTR or FILE_SETATTR. This routing is
> done by comparing the ioctl cmd to a set of 64-bit flags (FS_IOC_*).
>
> However, if a 32-bit process is running on a 64-bit kernel, it emits
> 32-bit flags (FS_IOC32_*) for certain ioctl operations. These flags are
> being checked erroneously, which leads to these ioctl operations being
> routed to the ioctl permission, rather than the correct file
> permissions.
>
> This was also noted in a RED-PEN finding from a while back -
> "/* RED-PEN how should LSM module know it's handling 32bit? */".
>
> This patch introduces a new hook, security_file_ioctl_compat, that is
> called from the compat ioctl syscall. All current LSMs have been changed
> to support this hook.
>
> Reviewing the three places where we are currently using
> security_file_ioctl, it appears that only SELinux needs a dedicated
> compat change; TOMOYO and SMACK appear to be functional without any
> change.
>
> Fixes: 0b24dcb7f2f7 ("Revert "selinux: simplify ioctl checking"")
> Signed-off-by: Alfred Piccioni <[e-mail]>
> Cc: [e-mail]
> ---

Reviewed-by: Stephen Smalley <[e-mail]>

Eric Biggers Dec. 20, 2023, 6:48 p.m. UTC | #5

On Tue, Dec 19, 2023 at 10:09:09AM +0100, Alfred Piccioni wrote:
> 
> base-commit: 196e95aa8305aecafc4e1857b7d3eff200d953b6
> -- 

Where can I get this base commit?

- Eric

Paul Moore Dec. 23, 2023, 1:23 a.m. UTC | #6

On Tue, Dec 19, 2023 at 4:09 AM Alfred Piccioni <[e-mail]> wrote:
>
> Some ioctl commands do not require ioctl permission, but are routed to
> other permissions such as FILE_GETATTR or FILE_SETATTR. This routing is
> done by comparing the ioctl cmd to a set of 64-bit flags (FS_IOC_*).
>
> However, if a 32-bit process is running on a 64-bit kernel, it emits
> 32-bit flags (FS_IOC32_*) for certain ioctl operations. These flags are
> being checked erroneously, which leads to these ioctl operations being
> routed to the ioctl permission, rather than the correct file
> permissions.
>
> This was also noted in a RED-PEN finding from a while back -
> "/* RED-PEN how should LSM module know it's handling 32bit? */".
>
> This patch introduces a new hook, security_file_ioctl_compat, that is
> called from the compat ioctl syscall. All current LSMs have been changed
> to support this hook.
>
> Reviewing the three places where we are currently using
> security_file_ioctl, it appears that only SELinux needs a dedicated
> compat change; TOMOYO and SMACK appear to be functional without any
> change.
>
> Fixes: 0b24dcb7f2f7 ("Revert "selinux: simplify ioctl checking"")
> Signed-off-by: Alfred Piccioni <[e-mail]>
> Cc: [e-mail]
> ---
>  fs/ioctl.c                    |  3 +--
>  include/linux/lsm_hook_defs.h |  2 ++
>  include/linux/security.h      |  7 +++++++
>  security/security.c           | 17 +++++++++++++++++
>  security/selinux/hooks.c      | 28 ++++++++++++++++++++++++++++
>  security/smack/smack_lsm.c    |  1 +
>  security/tomoyo/tomoyo.c      |  1 +
>  7 files changed, 57 insertions(+), 2 deletions(-)
>
> diff --git a/fs/ioctl.c b/fs/ioctl.c
> index f5fd99d6b0d4..76cf22ac97d7 100644
> --- a/fs/ioctl.c
> +++ b/fs/ioctl.c
> @@ -920,8 +920,7 @@ COMPAT_SYSCALL_DEFINE3(ioctl, unsigned int, fd, unsigned int, cmd,
>         if (!f.file)
>                 return -EBADF;
>
> -       /* RED-PEN how should LSM module know it's handling 32bit? */
> -       error = security_file_ioctl(f.file, cmd, arg);
> +       error = security_file_ioctl_compat(f.file, cmd, arg);
>         if (error)
>                 goto out;

This is interesting ... if you look at the normal ioctl() syscall
definition in the kernel you see 'ioctl(unsigned int fd, unsigned int
cmd, unsigned long arg)' and if you look at the compat definition you
see 'ioctl(unsigned int fd, unsigned int cmd, compat_ulong_t arg)'.  I
was expecting the second parameter, @cmd, to be a long type in the
normal definition, but it is an int type in both cases.  It looks like
it has been that way long enough that it is correct, but I'm a little
lost ...

> diff --git a/include/linux/lsm_hook_defs.h b/include/linux/lsm_hook_defs.h
> index ac962c4cb44b..626aa8cf930d 100644
> --- a/include/linux/lsm_hook_defs.h
> +++ b/include/linux/lsm_hook_defs.h
> @@ -171,6 +171,8 @@ LSM_HOOK(int, 0, file_alloc_security, struct file *file)
>  LSM_HOOK(void, LSM_RET_VOID, file_free_security, struct file *file)
>  LSM_HOOK(int, 0, file_ioctl, struct file *file, unsigned int cmd,
>          unsigned long arg)
> +LSM_HOOK(int, 0, file_ioctl_compat, struct file *file, unsigned int cmd,
> +        unsigned long arg)
>  LSM_HOOK(int, 0, mmap_addr, unsigned long addr)
>  LSM_HOOK(int, 0, mmap_file, struct file *file, unsigned long reqprot,
>          unsigned long prot, unsigned long flags)
> diff --git a/include/linux/security.h b/include/linux/security.h
> index 5f16eecde00b..22a82b7c59f1 100644
> --- a/include/linux/security.h
> +++ b/include/linux/security.h
> @@ -389,6 +389,7 @@ int security_file_permission(struct file *file, int mask);
>  int security_file_alloc(struct file *file);
>  void security_file_free(struct file *file);
>  int security_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
> +int security_file_ioctl_compat(struct file *file, unsigned int cmd, unsigned long arg);
>  int security_mmap_file(struct file *file, unsigned long prot,
>                         unsigned long flags);
>  int security_mmap_addr(unsigned long addr);
> @@ -987,6 +988,12 @@ static inline int security_file_ioctl(struct file *file, unsigned int cmd,
>         return 0;
>  }
>
> +static inline int security_file_ioctl_compat(struct file *file, unsigned int cmd,
> +                                     unsigned long arg)
> +{
> +       return 0;
> +}
> +
>  static inline int security_mmap_file(struct file *file, unsigned long prot,
>                                      unsigned long flags)
>  {
> diff --git a/security/security.c b/security/security.c
> index 23b129d482a7..5c16ffc99b1e 100644
> --- a/security/security.c
> +++ b/security/security.c
> @@ -2648,6 +2648,23 @@ int security_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
>  }
>  EXPORT_SYMBOL_GPL(security_file_ioctl);
>
> +/**
> + * security_file_ioctl_compat() - Check if an ioctl is allowed in 32-bit compat mode
> + * @file: associated file
> + * @cmd: ioctl cmd
> + * @arg: ioctl arguments
> + *
> + * Compat version of security_file_ioctl() that correctly handles 32-bit processes
> + * running on 64-bit kernels.
> + *
> + * Return: Returns 0 if permission is granted.
> + */
> +int security_file_ioctl_compat(struct file *file, unsigned int cmd, unsigned long arg)
> +{
> +       return call_int_hook(file_ioctl_compat, 0, file, cmd, arg);
> +}
> +EXPORT_SYMBOL_GPL(security_file_ioctl_compat);
> +
>  static inline unsigned long mmap_prot(struct file *file, unsigned long prot)
>  {
>         /*
> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
> index 2aa0e219d721..c617ae21dba8 100644
> --- a/security/selinux/hooks.c
> +++ b/security/selinux/hooks.c
> @@ -3731,6 +3731,33 @@ static int selinux_file_ioctl(struct file *file, unsigned int cmd,
>         return error;
>  }
>
> +static int selinux_file_ioctl_compat(struct file *file, unsigned int cmd,
> +                             unsigned long arg)
> +{
> +       /*
> +        * If we are in a 64-bit kernel running 32-bit userspace, we need to make
> +        * sure we don't compare 32-bit flags to 64-bit flags.
> +        */
> +       switch (cmd) {
> +       case FS_IOC32_GETFLAGS:
> +               cmd = FS_IOC_GETFLAGS;
> +               break;
> +       case FS_IOC32_SETFLAGS:
> +               cmd = FS_IOC_SETFLAGS;
> +               break;
> +       case FS_IOC32_GETVERSION:
> +               cmd = FS_IOC_GETVERSION;
> +               break;
> +       case FS_IOC32_SETVERSION:
> +               cmd = FS_IOC_SETVERSION;
> +               break;
> +       default:
> +               break;
> +       }
> +
> +       return selinux_file_ioctl(file, cmd, arg);
> +}

Is it considered valid for a native 64-bit task to use 32-bit
FS_IO32_XXX flags?  If not, do we want to remove the FS_IO32_XXX flag
checks in selinux_file_ioctl()?

>  static int default_noexec __ro_after_init;
>
>  static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
> @@ -7036,6 +7063,7 @@ static struct security_hook_list selinux_hooks[] __ro_after_init = {
>         LSM_HOOK_INIT(file_permission, selinux_file_permission),
>         LSM_HOOK_INIT(file_alloc_security, selinux_file_alloc_security),
>         LSM_HOOK_INIT(file_ioctl, selinux_file_ioctl),
> +       LSM_HOOK_INIT(file_ioctl_compat, selinux_file_ioctl_compat),
>         LSM_HOOK_INIT(mmap_file, selinux_mmap_file),
>         LSM_HOOK_INIT(mmap_addr, selinux_mmap_addr),
>         LSM_HOOK_INIT(file_mprotect, selinux_file_mprotect),
> diff --git a/security/smack/smack_lsm.c b/security/smack/smack_lsm.c
> index 65130a791f57..1f1ea8529421 100644
> --- a/security/smack/smack_lsm.c
> +++ b/security/smack/smack_lsm.c
> @@ -4973,6 +4973,7 @@ static struct security_hook_list smack_hooks[] __ro_after_init = {
>
>         LSM_HOOK_INIT(file_alloc_security, smack_file_alloc_security),
>         LSM_HOOK_INIT(file_ioctl, smack_file_ioctl),
> +       LSM_HOOK_INIT(file_ioctl_compat, smack_file_ioctl),
>         LSM_HOOK_INIT(file_lock, smack_file_lock),
>         LSM_HOOK_INIT(file_fcntl, smack_file_fcntl),
>         LSM_HOOK_INIT(mmap_file, smack_mmap_file),
> diff --git a/security/tomoyo/tomoyo.c b/security/tomoyo/tomoyo.c
> index 25006fddc964..298d182759c2 100644
> --- a/security/tomoyo/tomoyo.c
> +++ b/security/tomoyo/tomoyo.c
> @@ -568,6 +568,7 @@ static struct security_hook_list tomoyo_hooks[] __ro_after_init = {
>         LSM_HOOK_INIT(path_rename, tomoyo_path_rename),
>         LSM_HOOK_INIT(inode_getattr, tomoyo_inode_getattr),
>         LSM_HOOK_INIT(file_ioctl, tomoyo_file_ioctl),
> +       LSM_HOOK_INIT(file_ioctl_compat, tomoyo_file_ioctl),
>         LSM_HOOK_INIT(path_chmod, tomoyo_path_chmod),
>         LSM_HOOK_INIT(path_chown, tomoyo_path_chown),
>         LSM_HOOK_INIT(path_chroot, tomoyo_path_chroot),

I agree that it looks like Smack and TOMOYO should be fine, but I
would like to hear from Casey and Tetsuo to confirm.

Tetsuo Handa Dec. 23, 2023, 10:48 a.m. UTC | #7

On 2023/12/23 10:23, Paul Moore wrote:
>> -       /* RED-PEN how should LSM module know it's handling 32bit? */
>> -       error = security_file_ioctl(f.file, cmd, arg);
>> +       error = security_file_ioctl_compat(f.file, cmd, arg);
>>         if (error)
>>                 goto out;
> 
> This is interesting ... if you look at the normal ioctl() syscall
> definition in the kernel you see 'ioctl(unsigned int fd, unsigned int
> cmd, unsigned long arg)' and if you look at the compat definition you
> see 'ioctl(unsigned int fd, unsigned int cmd, compat_ulong_t arg)'.  I
> was expecting the second parameter, @cmd, to be a long type in the
> normal definition, but it is an int type in both cases.  It looks like
> it has been that way long enough that it is correct, but I'm a little
> lost ...

Since @arg might be a pointer to some struct, @arg needs to use a long type.
But @cmd can remain 32bits for both 32bits/64bits kernels because @cmd is not
a pointer, can't it?

> I agree that it looks like Smack and TOMOYO should be fine, but I
> would like to hear from Casey and Tetsuo to confirm.

Fine for TOMOYO part, for TOMOYO treats @cmd as an integer.

Tetsuo Handa Dec. 23, 2023, 2:41 p.m. UTC | #8

Adding BPF.

On 2023/12/19 18:10, Alfred Piccioni wrote:
>> I didn't do an audit but does anything need to be updated for the BPF
>> LSM or does it auto-magically pick up new hooks?
> 
> I'm unsure. I looked through the BPF LSM and I can't see any way it's
> picking up the file_ioctl hook to begin with. It appears to me
> skimming through the code that it automagically picks it up, but I'm
> not willing to bet the kernel on it.

If BPF LSM silently picks up security_file_ioctl_compat() hook, I worry
that some existing BPF programs which check ioctl() using BPF LSM fail to
understand that such BPF programs need to be updated.

We basically don't care about out-of-tree kernel code. But does that rule
apply to BPF programs? Since BPF programs are out-of-tree, are BPF programs
which depend on BPF LSM considered as "we don't care about" rule?
Or is breakage of existing BPF programs considered as a regression?
(Note that this patch is CC:ed for stable kernels.)

Maybe BPF LSM should at least emit warning if the loaded BPF program defined
security_file_ioctl() hook and did not define security_file_ioctl_compat() hook?

We could use a struct where undefined hooks needs to be manually filled with
a dummy pointer, so that we can catch erroneously undefined hooks (detected by
being automatically filled with a NULL pointer) at load time?

Eric Biggers Dec. 23, 2023, 3:34 p.m. UTC | #9

On Fri, Dec 22, 2023 at 08:23:26PM -0500, Paul Moore wrote:
> 
> Is it considered valid for a native 64-bit task to use 32-bit
> FS_IO32_XXX flags?

No, that's not valid.

> If not, do we want to remove the FS_IO32_XXX flag
> checks in selinux_file_ioctl()?

I don't see any such flag checks in selinux_file_ioctl().

Is there something else you have in mind?

- Eric

Casey Schaufler Dec. 23, 2023, 5:54 p.m. UTC | #10

On 12/22/2023 5:23 PM, Paul Moore wrote:
> On Tue, Dec 19, 2023 at 4:09 AM Alfred Piccioni <[e-mail]> wrote:
>> Some ioctl commands do not require ioctl permission, but are routed to
>> other permissions such as FILE_GETATTR or FILE_SETATTR. This routing is
>> done by comparing the ioctl cmd to a set of 64-bit flags (FS_IOC_*).
>>
>> However, if a 32-bit process is running on a 64-bit kernel, it emits
>> 32-bit flags (FS_IOC32_*) for certain ioctl operations. These flags are
>> being checked erroneously, which leads to these ioctl operations being
>> routed to the ioctl permission, rather than the correct file
>> permissions.
>>
>> This was also noted in a RED-PEN finding from a while back -
>> "/* RED-PEN how should LSM module know it's handling 32bit? */".
>>
>> This patch introduces a new hook, security_file_ioctl_compat, that is
>> called from the compat ioctl syscall. All current LSMs have been changed
>> to support this hook.
>>
>> Reviewing the three places where we are currently using
>> security_file_ioctl, it appears that only SELinux needs a dedicated
>> compat change; TOMOYO and SMACK appear to be functional without any
>> change.
>>
>> Fixes: 0b24dcb7f2f7 ("Revert "selinux: simplify ioctl checking"")
>> Signed-off-by: Alfred Piccioni <[e-mail]>
>> Cc: [e-mail]
>> ---
>>  fs/ioctl.c                    |  3 +--
>>  include/linux/lsm_hook_defs.h |  2 ++
>>  include/linux/security.h      |  7 +++++++
>>  security/security.c           | 17 +++++++++++++++++
>>  security/selinux/hooks.c      | 28 ++++++++++++++++++++++++++++
>>  security/smack/smack_lsm.c    |  1 +
>>  security/tomoyo/tomoyo.c      |  1 +
>>  7 files changed, 57 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/ioctl.c b/fs/ioctl.c
>> index f5fd99d6b0d4..76cf22ac97d7 100644
>> --- a/fs/ioctl.c
>> +++ b/fs/ioctl.c
>> @@ -920,8 +920,7 @@ COMPAT_SYSCALL_DEFINE3(ioctl, unsigned int, fd, unsigned int, cmd,
>>         if (!f.file)
>>                 return -EBADF;
>>
>> -       /* RED-PEN how should LSM module know it's handling 32bit? */
>> -       error = security_file_ioctl(f.file, cmd, arg);
>> +       error = security_file_ioctl_compat(f.file, cmd, arg);
>>         if (error)
>>                 goto out;
> This is interesting ... if you look at the normal ioctl() syscall
> definition in the kernel you see 'ioctl(unsigned int fd, unsigned int
> cmd, unsigned long arg)' and if you look at the compat definition you
> see 'ioctl(unsigned int fd, unsigned int cmd, compat_ulong_t arg)'.  I
> was expecting the second parameter, @cmd, to be a long type in the
> normal definition, but it is an int type in both cases.  It looks like
> it has been that way long enough that it is correct, but I'm a little
> lost ...
>
>> diff --git a/include/linux/lsm_hook_defs.h b/include/linux/lsm_hook_defs.h
>> index ac962c4cb44b..626aa8cf930d 100644
>> --- a/include/linux/lsm_hook_defs.h
>> +++ b/include/linux/lsm_hook_defs.h
>> @@ -171,6 +171,8 @@ LSM_HOOK(int, 0, file_alloc_security, struct file *file)
>>  LSM_HOOK(void, LSM_RET_VOID, file_free_security, struct file *file)
>>  LSM_HOOK(int, 0, file_ioctl, struct file *file, unsigned int cmd,
>>          unsigned long arg)
>> +LSM_HOOK(int, 0, file_ioctl_compat, struct file *file, unsigned int cmd,
>> +        unsigned long arg)
>>  LSM_HOOK(int, 0, mmap_addr, unsigned long addr)
>>  LSM_HOOK(int, 0, mmap_file, struct file *file, unsigned long reqprot,
>>          unsigned long prot, unsigned long flags)
>> diff --git a/include/linux/security.h b/include/linux/security.h
>> index 5f16eecde00b..22a82b7c59f1 100644
>> --- a/include/linux/security.h
>> +++ b/include/linux/security.h
>> @@ -389,6 +389,7 @@ int security_file_permission(struct file *file, int mask);
>>  int security_file_alloc(struct file *file);
>>  void security_file_free(struct file *file);
>>  int security_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
>> +int security_file_ioctl_compat(struct file *file, unsigned int cmd, unsigned long arg);
>>  int security_mmap_file(struct file *file, unsigned long prot,
>>                         unsigned long flags);
>>  int security_mmap_addr(unsigned long addr);
>> @@ -987,6 +988,12 @@ static inline int security_file_ioctl(struct file *file, unsigned int cmd,
>>         return 0;
>>  }
>>
>> +static inline int security_file_ioctl_compat(struct file *file, unsigned int cmd,
>> +                                     unsigned long arg)
>> +{
>> +       return 0;
>> +}
>> +
>>  static inline int security_mmap_file(struct file *file, unsigned long prot,
>>                                      unsigned long flags)
>>  {
>> diff --git a/security/security.c b/security/security.c
>> index 23b129d482a7..5c16ffc99b1e 100644
>> --- a/security/security.c
>> +++ b/security/security.c
>> @@ -2648,6 +2648,23 @@ int security_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
>>  }
>>  EXPORT_SYMBOL_GPL(security_file_ioctl);
>>
>> +/**
>> + * security_file_ioctl_compat() - Check if an ioctl is allowed in 32-bit compat mode
>> + * @file: associated file
>> + * @cmd: ioctl cmd
>> + * @arg: ioctl arguments
>> + *
>> + * Compat version of security_file_ioctl() that correctly handles 32-bit processes
>> + * running on 64-bit kernels.
>> + *
>> + * Return: Returns 0 if permission is granted.
>> + */
>> +int security_file_ioctl_compat(struct file *file, unsigned int cmd, unsigned long arg)
>> +{
>> +       return call_int_hook(file_ioctl_compat, 0, file, cmd, arg);
>> +}
>> +EXPORT_SYMBOL_GPL(security_file_ioctl_compat);
>> +
>>  static inline unsigned long mmap_prot(struct file *file, unsigned long prot)
>>  {
>>         /*
>> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
>> index 2aa0e219d721..c617ae21dba8 100644
>> --- a/security/selinux/hooks.c
>> +++ b/security/selinux/hooks.c
>> @@ -3731,6 +3731,33 @@ static int selinux_file_ioctl(struct file *file, unsigned int cmd,
>>         return error;
>>  }
>>
>> +static int selinux_file_ioctl_compat(struct file *file, unsigned int cmd,
>> +                             unsigned long arg)
>> +{
>> +       /*
>> +        * If we are in a 64-bit kernel running 32-bit userspace, we need to make
>> +        * sure we don't compare 32-bit flags to 64-bit flags.
>> +        */
>> +       switch (cmd) {
>> +       case FS_IOC32_GETFLAGS:
>> +               cmd = FS_IOC_GETFLAGS;
>> +               break;
>> +       case FS_IOC32_SETFLAGS:
>> +               cmd = FS_IOC_SETFLAGS;
>> +               break;
>> +       case FS_IOC32_GETVERSION:
>> +               cmd = FS_IOC_GETVERSION;
>> +               break;
>> +       case FS_IOC32_SETVERSION:
>> +               cmd = FS_IOC_SETVERSION;
>> +               break;
>> +       default:
>> +               break;
>> +       }
>> +
>> +       return selinux_file_ioctl(file, cmd, arg);
>> +}
> Is it considered valid for a native 64-bit task to use 32-bit
> FS_IO32_XXX flags?  If not, do we want to remove the FS_IO32_XXX flag
> checks in selinux_file_ioctl()?
>
>>  static int default_noexec __ro_after_init;
>>
>>  static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
>> @@ -7036,6 +7063,7 @@ static struct security_hook_list selinux_hooks[] __ro_after_init = {
>>         LSM_HOOK_INIT(file_permission, selinux_file_permission),
>>         LSM_HOOK_INIT(file_alloc_security, selinux_file_alloc_security),
>>         LSM_HOOK_INIT(file_ioctl, selinux_file_ioctl),
>> +       LSM_HOOK_INIT(file_ioctl_compat, selinux_file_ioctl_compat),
>>         LSM_HOOK_INIT(mmap_file, selinux_mmap_file),
>>         LSM_HOOK_INIT(mmap_addr, selinux_mmap_addr),
>>         LSM_HOOK_INIT(file_mprotect, selinux_file_mprotect),
>> diff --git a/security/smack/smack_lsm.c b/security/smack/smack_lsm.c
>> index 65130a791f57..1f1ea8529421 100644
>> --- a/security/smack/smack_lsm.c
>> +++ b/security/smack/smack_lsm.c
>> @@ -4973,6 +4973,7 @@ static struct security_hook_list smack_hooks[] __ro_after_init = {
>>
>>         LSM_HOOK_INIT(file_alloc_security, smack_file_alloc_security),
>>         LSM_HOOK_INIT(file_ioctl, smack_file_ioctl),
>> +       LSM_HOOK_INIT(file_ioctl_compat, smack_file_ioctl),
>>         LSM_HOOK_INIT(file_lock, smack_file_lock),
>>         LSM_HOOK_INIT(file_fcntl, smack_file_fcntl),
>>         LSM_HOOK_INIT(mmap_file, smack_mmap_file),
>> diff --git a/security/tomoyo/tomoyo.c b/security/tomoyo/tomoyo.c
>> index 25006fddc964..298d182759c2 100644
>> --- a/security/tomoyo/tomoyo.c
>> +++ b/security/tomoyo/tomoyo.c
>> @@ -568,6 +568,7 @@ static struct security_hook_list tomoyo_hooks[] __ro_after_init = {
>>         LSM_HOOK_INIT(path_rename, tomoyo_path_rename),
>>         LSM_HOOK_INIT(inode_getattr, tomoyo_inode_getattr),
>>         LSM_HOOK_INIT(file_ioctl, tomoyo_file_ioctl),
>> +       LSM_HOOK_INIT(file_ioctl_compat, tomoyo_file_ioctl),
>>         LSM_HOOK_INIT(path_chmod, tomoyo_path_chmod),
>>         LSM_HOOK_INIT(path_chown, tomoyo_path_chown),
>>         LSM_HOOK_INIT(path_chroot, tomoyo_path_chroot),
> I agree that it looks like Smack and TOMOYO should be fine, but I
> would like to hear from Casey and Tetsuo to confirm.

Smack should be OK.

Paul Moore Dec. 24, 2023, 7:58 p.m. UTC | #11

On Sat, Dec 23, 2023 at 5:49 AM Tetsuo Handa
<[e-mail]> wrote:
> On 2023/12/23 10:23, Paul Moore wrote:
> >> -       /* RED-PEN how should LSM module know it's handling 32bit? */
> >> -       error = security_file_ioctl(f.file, cmd, arg);
> >> +       error = security_file_ioctl_compat(f.file, cmd, arg);
> >>         if (error)
> >>                 goto out;
> >
> > This is interesting ... if you look at the normal ioctl() syscall
> > definition in the kernel you see 'ioctl(unsigned int fd, unsigned int
> > cmd, unsigned long arg)' and if you look at the compat definition you
> > see 'ioctl(unsigned int fd, unsigned int cmd, compat_ulong_t arg)'.  I
> > was expecting the second parameter, @cmd, to be a long type in the
> > normal definition, but it is an int type in both cases.  It looks like
> > it has been that way long enough that it is correct, but I'm a little
> > lost ...
>
> Since @arg might be a pointer to some struct, @arg needs to use a long type.
> But @cmd can remain 32bits for both 32bits/64bits kernels because @cmd is not
> a pointer, can't it?

I'm not worried about @arg, I'm worried about @cmd, the second
parameter to the syscall.  I was looking at the manpage and it is
specified as an unsigned long, which would be a size mismatch on a
64-bit system, although now that I'm reading further into the manpage
I see that the command is specified as a 32-bit value so an int
shouldn't be a problem.  I'm guessing the unsigned long type persists
from the days before 64-bit systems.

> > I agree that it looks like Smack and TOMOYO should be fine, but I
> > would like to hear from Casey and Tetsuo to confirm.
>
> Fine for TOMOYO part, for TOMOYO treats @cmd as an integer.

Great, thank you.

Paul Moore Dec. 24, 2023, 8 p.m. UTC | #12

On Sat, Dec 23, 2023 at 10:34 AM Eric Biggers <[e-mail]> wrote:
> On Fri, Dec 22, 2023 at 08:23:26PM -0500, Paul Moore wrote:
> > Is it considered valid for a native 64-bit task to use 32-bit
> > FS_IO32_XXX flags?
>
> No, that's not valid.

Excellent, thank you.

> > If not, do we want to remove the FS_IO32_XXX flag
> > checks in selinux_file_ioctl()?
>
> I don't see any such flag checks in selinux_file_ioctl().

Neither do I ... I'm not sure what I was looking at when I made that
comment, I'm going to chalk that up to a bit of holiday fog.  Sorry
for the noise.

> Is there something else you have in mind?

Nope.

Paul Moore Dec. 24, 2023, 8:09 p.m. UTC | #13

On Sun, Dec 24, 2023 at 3:00 PM Paul Moore <[e-mail]> wrote:
> On Sat, Dec 23, 2023 at 10:34 AM Eric Biggers <[e-mail]> wrote:
> > On Fri, Dec 22, 2023 at 08:23:26PM -0500, Paul Moore wrote:
> > > Is it considered valid for a native 64-bit task to use 32-bit
> > > FS_IO32_XXX flags?
> >
> > No, that's not valid.
>
> Excellent, thank you.
>
> > > If not, do we want to remove the FS_IO32_XXX flag
> > > checks in selinux_file_ioctl()?
> >
> > I don't see any such flag checks in selinux_file_ioctl().
>
> Neither do I ... I'm not sure what I was looking at when I made that
> comment, I'm going to chalk that up to a bit of holiday fog.  Sorry
> for the noise.

Ah ha, I think I found the problem - the tools I use to pull in
patches for review seemed to have grabbed an old version of the patch
that *did* as the 32-bit ioctl commands to selinux_file_ioctl().

https://lore.kernel.org/selinux/[e-mail]/

Paul Moore Dec. 24, 2023, 8:53 p.m. UTC | #14

On Tue, Dec 19, 2023 at 4:09 AM Alfred Piccioni <[e-mail]> wrote:
>
> Some ioctl commands do not require ioctl permission, but are routed to
> other permissions such as FILE_GETATTR or FILE_SETATTR. This routing is
> done by comparing the ioctl cmd to a set of 64-bit flags (FS_IOC_*).
>
> However, if a 32-bit process is running on a 64-bit kernel, it emits
> 32-bit flags (FS_IOC32_*) for certain ioctl operations. These flags are
> being checked erroneously, which leads to these ioctl operations being
> routed to the ioctl permission, rather than the correct file
> permissions.
>
> This was also noted in a RED-PEN finding from a while back -
> "/* RED-PEN how should LSM module know it's handling 32bit? */".
>
> This patch introduces a new hook, security_file_ioctl_compat, that is
> called from the compat ioctl syscall. All current LSMs have been changed
> to support this hook.
>
> Reviewing the three places where we are currently using
> security_file_ioctl, it appears that only SELinux needs a dedicated
> compat change; TOMOYO and SMACK appear to be functional without any
> change.
>
> Fixes: 0b24dcb7f2f7 ("Revert "selinux: simplify ioctl checking"")
> Signed-off-by: Alfred Piccioni <[e-mail]>
> Cc: [e-mail]
> ---
>  fs/ioctl.c                    |  3 +--
>  include/linux/lsm_hook_defs.h |  2 ++
>  include/linux/security.h      |  7 +++++++
>  security/security.c           | 17 +++++++++++++++++
>  security/selinux/hooks.c      | 28 ++++++++++++++++++++++++++++
>  security/smack/smack_lsm.c    |  1 +
>  security/tomoyo/tomoyo.c      |  1 +
>  7 files changed, 57 insertions(+), 2 deletions(-)

I made some minor style tweaks around line length and alignment, but
otherwise this looked good to me.  Thanks all!

While I agree this is definitely stable kernel material, given where
we are at in the current kernel cycle, and with the end-of-year
holidays in full swing, I'm going to merge this into lsm/dev and send
it up to Linus during the next merge window.  The stable tag will
remain intact, so it will end up trickling down into the stable
kernels, it will just take an extra week or so (which I think will be
good from a testing perspective).

Eric Biggers Dec. 27, 2023, 4:43 a.m. UTC | #15

On Sun, Dec 24, 2023 at 03:53:16PM -0500, Paul Moore wrote:
> On Tue, Dec 19, 2023 at 4:09 AM Alfred Piccioni <[e-mail]> wrote:
> >
> > Some ioctl commands do not require ioctl permission, but are routed to
> > other permissions such as FILE_GETATTR or FILE_SETATTR. This routing is
> > done by comparing the ioctl cmd to a set of 64-bit flags (FS_IOC_*).
> >
> > However, if a 32-bit process is running on a 64-bit kernel, it emits
> > 32-bit flags (FS_IOC32_*) for certain ioctl operations. These flags are
> > being checked erroneously, which leads to these ioctl operations being
> > routed to the ioctl permission, rather than the correct file
> > permissions.
> >
> > This was also noted in a RED-PEN finding from a while back -
> > "/* RED-PEN how should LSM module know it's handling 32bit? */".
> >
> > This patch introduces a new hook, security_file_ioctl_compat, that is
> > called from the compat ioctl syscall. All current LSMs have been changed
> > to support this hook.
> >
> > Reviewing the three places where we are currently using
> > security_file_ioctl, it appears that only SELinux needs a dedicated
> > compat change; TOMOYO and SMACK appear to be functional without any
> > change.
> >
> > Fixes: 0b24dcb7f2f7 ("Revert "selinux: simplify ioctl checking"")
> > Signed-off-by: Alfred Piccioni <[e-mail]>
> > Cc: [e-mail]
> > ---
> >  fs/ioctl.c                    |  3 +--
> >  include/linux/lsm_hook_defs.h |  2 ++
> >  include/linux/security.h      |  7 +++++++
> >  security/security.c           | 17 +++++++++++++++++
> >  security/selinux/hooks.c      | 28 ++++++++++++++++++++++++++++
> >  security/smack/smack_lsm.c    |  1 +
> >  security/tomoyo/tomoyo.c      |  1 +
> >  7 files changed, 57 insertions(+), 2 deletions(-)
> 
> I made some minor style tweaks around line length and alignment, but
> otherwise this looked good to me.  Thanks all!
> 

Reviewed-by: Eric Biggers <[e-mail]>

(I reviewed the version in branch "next" of
https://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/lsm.git)

- Eric

Patch

diff --git a/fs/ioctl.c b/fs/ioctl.c
index f5fd99d6b0d4..76cf22ac97d7 100644
--- a/fs/ioctl.c
+++ b/fs/ioctl.c
@@ -920,8 +920,7 @@  COMPAT_SYSCALL_DEFINE3(ioctl, unsigned int, fd, unsigned int, cmd,
 	if (!f.file)
 		return -EBADF;
 
-	/* RED-PEN how should LSM module know it's handling 32bit? */
-	error = security_file_ioctl(f.file, cmd, arg);
+	error = security_file_ioctl_compat(f.file, cmd, arg);
 	if (error)
 		goto out;
 
diff --git a/include/linux/lsm_hook_defs.h b/include/linux/lsm_hook_defs.h
index ac962c4cb44b..626aa8cf930d 100644
--- a/include/linux/lsm_hook_defs.h
+++ b/include/linux/lsm_hook_defs.h
@@ -171,6 +171,8 @@  LSM_HOOK(int, 0, file_alloc_security, struct file *file)
 LSM_HOOK(void, LSM_RET_VOID, file_free_security, struct file *file)
 LSM_HOOK(int, 0, file_ioctl, struct file *file, unsigned int cmd,
 	 unsigned long arg)
+LSM_HOOK(int, 0, file_ioctl_compat, struct file *file, unsigned int cmd,
+	 unsigned long arg)
 LSM_HOOK(int, 0, mmap_addr, unsigned long addr)
 LSM_HOOK(int, 0, mmap_file, struct file *file, unsigned long reqprot,
 	 unsigned long prot, unsigned long flags)
diff --git a/include/linux/security.h b/include/linux/security.h
index 5f16eecde00b..22a82b7c59f1 100644
--- a/include/linux/security.h
+++ b/include/linux/security.h
@@ -389,6 +389,7 @@  int security_file_permission(struct file *file, int mask);
 int security_file_alloc(struct file *file);
 void security_file_free(struct file *file);
 int security_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
+int security_file_ioctl_compat(struct file *file, unsigned int cmd, unsigned long arg);
 int security_mmap_file(struct file *file, unsigned long prot,
 			unsigned long flags);
 int security_mmap_addr(unsigned long addr);
@@ -987,6 +988,12 @@  static inline int security_file_ioctl(struct file *file, unsigned int cmd,
 	return 0;
 }
 
+static inline int security_file_ioctl_compat(struct file *file, unsigned int cmd,
+				      unsigned long arg)
+{
+	return 0;
+}
+
 static inline int security_mmap_file(struct file *file, unsigned long prot,
 				     unsigned long flags)
 {
diff --git a/security/security.c b/security/security.c
index 23b129d482a7..5c16ffc99b1e 100644
--- a/security/security.c
+++ b/security/security.c
@@ -2648,6 +2648,23 @@  int security_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 }
 EXPORT_SYMBOL_GPL(security_file_ioctl);
 
+/**
+ * security_file_ioctl_compat() - Check if an ioctl is allowed in 32-bit compat mode
+ * @file: associated file
+ * @cmd: ioctl cmd
+ * @arg: ioctl arguments
+ *
+ * Compat version of security_file_ioctl() that correctly handles 32-bit processes
+ * running on 64-bit kernels.
+ *
+ * Return: Returns 0 if permission is granted.
+ */
+int security_file_ioctl_compat(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	return call_int_hook(file_ioctl_compat, 0, file, cmd, arg);
+}
+EXPORT_SYMBOL_GPL(security_file_ioctl_compat);
+
 static inline unsigned long mmap_prot(struct file *file, unsigned long prot)
 {
 	/*
diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
index 2aa0e219d721..c617ae21dba8 100644
--- a/security/selinux/hooks.c
+++ b/security/selinux/hooks.c
@@ -3731,6 +3731,33 @@  static int selinux_file_ioctl(struct file *file, unsigned int cmd,
 	return error;
 }
 
+static int selinux_file_ioctl_compat(struct file *file, unsigned int cmd,
+			      unsigned long arg)
+{
+	/*
+	 * If we are in a 64-bit kernel running 32-bit userspace, we need to make
+	 * sure we don't compare 32-bit flags to 64-bit flags.
+	 */
+	switch (cmd) {
+	case FS_IOC32_GETFLAGS:
+		cmd = FS_IOC_GETFLAGS;
+		break;
+	case FS_IOC32_SETFLAGS:
+		cmd = FS_IOC_SETFLAGS;
+		break;
+	case FS_IOC32_GETVERSION:
+		cmd = FS_IOC_GETVERSION;
+		break;
+	case FS_IOC32_SETVERSION:
+		cmd = FS_IOC_SETVERSION;
+		break;
+	default:
+		break;
+	}
+
+	return selinux_file_ioctl(file, cmd, arg);
+}
+
 static int default_noexec __ro_after_init;
 
 static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
@@ -7036,6 +7063,7 @@  static struct security_hook_list selinux_hooks[] __ro_after_init = {
 	LSM_HOOK_INIT(file_permission, selinux_file_permission),
 	LSM_HOOK_INIT(file_alloc_security, selinux_file_alloc_security),
 	LSM_HOOK_INIT(file_ioctl, selinux_file_ioctl),
+	LSM_HOOK_INIT(file_ioctl_compat, selinux_file_ioctl_compat),
 	LSM_HOOK_INIT(mmap_file, selinux_mmap_file),
 	LSM_HOOK_INIT(mmap_addr, selinux_mmap_addr),
 	LSM_HOOK_INIT(file_mprotect, selinux_file_mprotect),
diff --git a/security/smack/smack_lsm.c b/security/smack/smack_lsm.c
index 65130a791f57..1f1ea8529421 100644
--- a/security/smack/smack_lsm.c
+++ b/security/smack/smack_lsm.c
@@ -4973,6 +4973,7 @@  static struct security_hook_list smack_hooks[] __ro_after_init = {
 
 	LSM_HOOK_INIT(file_alloc_security, smack_file_alloc_security),
 	LSM_HOOK_INIT(file_ioctl, smack_file_ioctl),
+	LSM_HOOK_INIT(file_ioctl_compat, smack_file_ioctl),
 	LSM_HOOK_INIT(file_lock, smack_file_lock),
 	LSM_HOOK_INIT(file_fcntl, smack_file_fcntl),
 	LSM_HOOK_INIT(mmap_file, smack_mmap_file),
diff --git a/security/tomoyo/tomoyo.c b/security/tomoyo/tomoyo.c
index 25006fddc964..298d182759c2 100644
--- a/security/tomoyo/tomoyo.c
+++ b/security/tomoyo/tomoyo.c
@@ -568,6 +568,7 @@  static struct security_hook_list tomoyo_hooks[] __ro_after_init = {
 	LSM_HOOK_INIT(path_rename, tomoyo_path_rename),
 	LSM_HOOK_INIT(inode_getattr, tomoyo_inode_getattr),
 	LSM_HOOK_INIT(file_ioctl, tomoyo_file_ioctl),
+	LSM_HOOK_INIT(file_ioctl_compat, tomoyo_file_ioctl),
 	LSM_HOOK_INIT(path_chmod, tomoyo_path_chmod),
 	LSM_HOOK_INIT(path_chown, tomoyo_path_chown),
 	LSM_HOOK_INIT(path_chroot, tomoyo_path_chroot),