[PATCH dhowells/mount-api] fs: fix checking an error code of security_fs_context_parse_param

[PATCH dhowells/mount-api] fs: fix checking an error code of security_fs_context_parse_param

[Andrei Vagin]

From: Andrei Vagin <avagin@gmail.com>

security_fs_context_parse_param() returns 0 if everything is okay.

Signed-off-by: Andrei Vagin <avagin@gmail.com>
---
 fs/fs_context.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/fs_context.c b/fs/fs_context.c
index a82679441031..16bf2cb57534 100644
--- a/fs/fs_context.c
+++ b/fs/fs_context.c
@@ -146,7 +146,7 @@ int vfs_parse_fs_param(struct fs_context *fc, struct fs_parameter *param)
 		return ret;
 
 	ret = security_fs_context_parse_param(fc, param);
-	if (ret != -ENOPARAM)
+	if (ret)
 		/* Param belongs to the LSM or is disallowed by the LSM; so
 		 * don't pass to the FS.
 		 */
-- 
2.17.1

Re: [PATCH dhowells/mount-api] fs: fix checking an error code of security_fs_context_parse_param

[David Howells]

Andrei Vagin <avagin@openvz.org> wrote:

> security_fs_context_parse_param() returns 0 if everything is okay.

Not quite.  It returns 0 to indicate that it processed the argument and that
the argument doesn't belong to the filesystem.  It returns -ENOPARAM to
indicate that the argument should be passed along to the filesystem.  Any
other error is an actual error.

>  	ret = security_fs_context_parse_param(fc, param);
> -	if (ret != -ENOPARAM)
> +	if (ret)

So this change is wrong.

The change that needs making is to the documentation - in particular
lsm_hooks.h.

David

Re: [PATCH dhowells/mount-api] fs: fix checking an error code of security_fs_context_parse_param

[David Howells]

> The change that needs making is to the documentation - in particular
> lsm_hooks.h.

And also the default for security_fs_context_parse_param() needs changing to
-ENOPARAM.

David

Re: [PATCH dhowells/mount-api] fs: fix checking an error code of security_fs_context_parse_param

[Andrei Vagin]

On Thu, Sep 06, 2018 at 11:18:52AM +0100, David Howells wrote:
> > The change that needs making is to the documentation - in particular
> > lsm_hooks.h.
> 
> And also the default for security_fs_context_parse_param() needs changing to
> -ENOPARAM.

apparmor_fs_context_parse_param() should be fixed too, it never returns
-ENOPARAM.

> 
> David

Re: [PATCH dhowells/mount-api] fs: fix checking an error code of security_fs_context_parse_param

[David Howells]

Andrei Vagin <avagin@gmail.com> wrote:

> apparmor_fs_context_parse_param() should be fixed too, it never returns
> -ENOPARAM.

Fixed, thanks.

David

[PATCH dhowells/mount-api] fs: return -ENOPARAM from security_fs_context_parse_param by default

[Andrei Vagin]

From: Andrei Vagin <avagin@gmail.com>

It returns 0 to indicate that it processed the argument and that
the argument doesn't belong to the filesystem.  It returns -ENOPARAM to
indicate that the argument should be passed along to the filesystem.  Any
other error is an actual error. // David Howells <dhowells@redhat.com>

Signed-off-by: Andrei Vagin <avagin@gmail.com>
---
 include/linux/lsm_hooks.h | 4 ++++
 include/linux/security.h  | 2 +-
 security/security.c       | 2 +-
 3 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/include/linux/lsm_hooks.h b/include/linux/lsm_hooks.h
index ceb487c3fbc7..d333ceebf056 100644
--- a/include/linux/lsm_hooks.h
+++ b/include/linux/lsm_hooks.h
@@ -103,6 +103,10 @@
  *	Validate the filesystem context preparatory to applying it.  This is
  *	done after all the options have been parsed.
  *	@fc indicates the filesystem context.
+ *	Returns 0 to indicate that it processed the argument and that the argument
+ *	doesn't belong to the filesystem. It returns -ENOPARAM to indicate that
+ *	the argument should be passed along to the filesystem. Any other error is
+ *	an actual error.
  * @sb_get_tree:
  *	Assign the security to a newly created superblock.
  *	@fc indicates the filesystem context.
diff --git a/include/linux/security.h b/include/linux/security.h
index dbed5326b45c..bae191a96c73 100644
--- a/include/linux/security.h
+++ b/include/linux/security.h
@@ -576,7 +576,7 @@ static inline void security_fs_context_free(struct fs_context *fc)
 static inline int security_fs_context_parse_param(struct fs_context *fc,
 						  struct fs_parameter *param)
 {
-	return 0;
+	return -ENOPARAM;
 }
 static inline int security_fs_context_validate(struct fs_context *fc)
 {
diff --git a/security/security.c b/security/security.c
index 94420f3616a7..64304d20aae1 100644
--- a/security/security.c
+++ b/security/security.c
@@ -380,7 +380,7 @@ void security_fs_context_free(struct fs_context *fc)
 
 int security_fs_context_parse_param(struct fs_context *fc, struct fs_parameter *param)
 {
-	return call_int_hook(fs_context_parse_param, 0, fc, param);
+	return call_int_hook(fs_context_parse_param, -ENOPARAM, fc, param);
 }
 
 int security_fs_context_validate(struct fs_context *fc)
-- 
2.17.1

Re: [PATCH dhowells/mount-api] fs: return -ENOPARAM from security_fs_context_parse_param by default

[David Howells]

Andrei Vagin <avagin@openvz.org> wrote:

> It returns 0 to indicate that it processed the argument and that
> the argument doesn't belong to the filesystem.  It returns -ENOPARAM to
> indicate that the argument should be passed along to the filesystem.  Any
> other error is an actual error. // David Howells <dhowells@redhat.com>

I made these changes already, thanks.

David

Re: [PATCH dhowells/mount-api] fs: fix checking an error code of security_fs_context_parse_param

[Andrei Vagin]

On Thu, Sep 06, 2018 at 11:16:24AM +0100, David Howells wrote:
> Andrei Vagin <avagin@openvz.org> wrote:
> 
> > security_fs_context_parse_param() returns 0 if everything is okay.
> 
> Not quite.  It returns 0 to indicate that it processed the argument and that
> the argument doesn't belong to the filesystem.  It returns -ENOPARAM to
> indicate that the argument should be passed along to the filesystem.  Any
> other error is an actual error.
> 
> >  	ret = security_fs_context_parse_param(fc, param);
> > -	if (ret != -ENOPARAM)
> > +	if (ret)
> 
> So this change is wrong.

It depends from what point to see on it. Without this patch, the
linux-next kernel doesn't boot:

VFS: Cannot open root device "vda2" or unknown-block(253,2): error -2
Please append a correct "root=" boot option; here are the available partitions:
fd00        25165824 vda 
 driver: virtio_blk
  fd01          512000 vda1 5065bd2c-01

  fd02        23601152 vda2 5065bd2c-02

  fd03         1044480 vda3 5065bd2c-03

Kernel panic - not syncing: VFS: Unable to mount root fs on unknown-block(253,2)
CPU: 1 PID: 1 Comm: swapper/0 Not tainted 4.19.0-rc1-00106-g679a89690477 #38
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20180531_142017-buildhw-08.phx2.fedoraproject.org-1.fc28 04/01/2014
Call Trace:
 dump_stack+0x85/0xc0
 panic+0xec/0x251
 mount_block_root+0x23e/0x2e7
 ? do_early_param+0x8e/0x8e
 prepare_namespace+0x135/0x16b
 kernel_init_freeable+0x28b/0x2af
 ? rest_init+0xb9/0xb9
 kernel_init+0xa/0x10e
 ret_from_fork+0x3a/0x50
Kernel Offset: 0x10000000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
---[ end Kernel panic - not syncing: VFS: Unable to mount root fs on unknown-block(253,2) ]---

I agree that I haven't understood the idea of this code. I found that
security_fs_context_parse_param returns 0 by default and decided that this
patch would fix the problem. I have sent an update version of the patch.

Pls, take a look at it. And we need to push this patch to linux-next. We can't
run CRIU tests for linux-next without it.

Thanks,
Andrei

> 
> The change that needs making is to the documentation - in particular
> lsm_hooks.h.
> 
> David