[LTP] [PATCH] statx09: Reduce fs-verity blocksize to 1024

[LTP] [PATCH] statx09: Reduce fs-verity blocksize to 1024

[Martin Doucha]

The kernel requires that fs-verity blocksize must be at most equal
to the filesystem blocksize. Testing on small loop device means that
mkfs.ext4 will very likely default to blocksize of 1024. Set fs-verity
blocksize to the minimum possible value (1024) to avoid blocksize
mismatch.

Signed-off-by: Martin Doucha <mdoucha@suse.cz>
---

Alternatively, we could add "-b 4096" to .dev_fs_opts.

 testcases/kernel/syscalls/statx/statx09.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/testcases/kernel/syscalls/statx/statx09.c b/testcases/kernel/syscalls/statx/statx09.c
index c03d2c91e..dc9786c76 100644
--- a/testcases/kernel/syscalls/statx/statx09.c
+++ b/testcases/kernel/syscalls/statx/statx09.c
@@ -94,7 +94,7 @@ static void flag_setup(void)
 	memset(&enable, 0, sizeof(enable));
 	enable.version = 1;
 	enable.hash_algorithm = hash_algorithms[0];
-	enable.block_size = 4096;
+	enable.block_size = 1024;
 	enable.salt_size = 0;
 	enable.salt_ptr = (intptr_t)NULL;
 	enable.sig_size = 0;
-- 
2.41.0


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] [PATCH] statx09: Reduce fs-verity blocksize to 1024

[Eric Biggers]

Hi Martin,

On Thu, Jul 13, 2023 at 03:28:52PM +0200, Martin Doucha wrote:
> The kernel requires that fs-verity blocksize must be at most equal
> to the filesystem blocksize. Testing on small loop device means that
> mkfs.ext4 will very likely default to blocksize of 1024. Set fs-verity
> blocksize to the minimum possible value (1024) to avoid blocksize
> mismatch.
> 
> Signed-off-by: Martin Doucha <mdoucha@suse.cz>
> ---
> 
> Alternatively, we could add "-b 4096" to .dev_fs_opts.
> 
>  testcases/kernel/syscalls/statx/statx09.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/testcases/kernel/syscalls/statx/statx09.c b/testcases/kernel/syscalls/statx/statx09.c
> index c03d2c91e..dc9786c76 100644
> --- a/testcases/kernel/syscalls/statx/statx09.c
> +++ b/testcases/kernel/syscalls/statx/statx09.c
> @@ -94,7 +94,7 @@ static void flag_setup(void)
>  	memset(&enable, 0, sizeof(enable));
>  	enable.version = 1;
>  	enable.hash_algorithm = hash_algorithms[0];
> -	enable.block_size = 4096;
> +	enable.block_size = 1024;
>  	enable.salt_size = 0;
>  	enable.salt_ptr = (intptr_t)NULL;
>  	enable.sig_size = 0;
> -- 

Support for Merkle tree block sizes less than the page size didn't exist until
kernel version 6.3.  Please see the documentation for FS_IOC_ENABLE_VERITY
(https://www.kernel.org/doc/html/latest/filesystems/fsverity.html#fs-ioc-enable-verity):

    "block_size is the Merkle tree block size, in bytes. In Linux v6.3 and
    later, this can be any power of 2 between (inclusively) 1024 and the minimum
    of the system page size and the filesystem block size. In earlier versions,
    the page size was the only allowed value."

There are a few different approaches you could take.  One would be the one you
mentioned: just add "-b 4096".  Though, note that systems with non-4K pages and
kernel older than v6.3 can't mount an "-O verity -b 4096" filesystem.

Or you could query the filesystem block size (that resulted from mkfs.ext4 with
unspecified block size) and use that as the Merkle tree block size.

Or you could simply make the test depend on kernel v6.3, I suppose.

- Eric

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] [PATCH] statx09: Reduce fs-verity blocksize to 1024

[Petr Vorel]

Hi Eric, Martin,

Eric, thanks for your input, Martin, thanks for working on this.

> Hi Martin,

> On Thu, Jul 13, 2023 at 03:28:52PM +0200, Martin Doucha wrote:
> > The kernel requires that fs-verity blocksize must be at most equal
> > to the filesystem blocksize. Testing on small loop device means that
> > mkfs.ext4 will very likely default to blocksize of 1024. Set fs-verity
> > blocksize to the minimum possible value (1024) to avoid blocksize
> > mismatch.

Yes kernels < 6.3 TCONF on current master:
statx09.c:124: TCONF: fs-verity not supported on loopdev

And 6.3 fails:
statx09.c:109: TBROK: ioctl(3, FS_IOC_ENABLE_VERITY) failed: EINVAL (22)

> > Signed-off-by: Martin Doucha <mdoucha@suse.cz>
> > ---

> > Alternatively, we could add "-b 4096" to .dev_fs_opts.

> >  testcases/kernel/syscalls/statx/statx09.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)

> > diff --git a/testcases/kernel/syscalls/statx/statx09.c b/testcases/kernel/syscalls/statx/statx09.c
> > index c03d2c91e..dc9786c76 100644
> > --- a/testcases/kernel/syscalls/statx/statx09.c
> > +++ b/testcases/kernel/syscalls/statx/statx09.c
> > @@ -94,7 +94,7 @@ static void flag_setup(void)
> >  	memset(&enable, 0, sizeof(enable));
> >  	enable.version = 1;
> >  	enable.hash_algorithm = hash_algorithms[0];
> > -	enable.block_size = 4096;
> > +	enable.block_size = 1024;
> >  	enable.salt_size = 0;
> >  	enable.salt_ptr = (intptr_t)NULL;
> >  	enable.sig_size = 0;
> > -- 

> Support for Merkle tree block sizes less than the page size didn't exist until
> kernel version 6.3.  Please see the documentation for FS_IOC_ENABLE_VERITY
> (https://www.kernel.org/doc/html/latest/filesystems/fsverity.html#fs-ioc-enable-verity):

>     "block_size is the Merkle tree block size, in bytes. In Linux v6.3 and
>     later, this can be any power of 2 between (inclusively) 1024 and the minimum
>     of the system page size and the filesystem block size. In earlier versions,
>     the page size was the only allowed value."

> There are a few different approaches you could take.  One would be the one you
> mentioned: just add "-b 4096".  Though, note that systems with non-4K pages and
> kernel older than v6.3 can't mount an "-O verity -b 4096" filesystem.

I suppose the problem would be with e.g. ppc64le. Because otherwise
it works with this setup (untested on ppc64le, I believe it'd fail):
.dev_fs_opts = (const char *const []){"-O verity", "-b 4096", NULL},
and kept enable.block_size = 4096
the test passes on older kernels 6.2.12-1-default (openSUSE), 5.10.0-8-amd64
(Debian) as well as on 6.3 kernel.

> Or you could query the filesystem block size (that resulted from mkfs.ext4 with
> unspecified block size) and use that as the Merkle tree block size.

Unless it's not too complicated, it looks to me the best, as it'd support both
 >= 6.3 and older kernels.

> Or you could simply make the test depend on kernel v6.3, I suppose.

If possible I'd prefer to keep functionality also for older kernels.

Another (IMHO not really good option) would be to acquire device big enough via
.dev_min_size (it'd probably have to be too big device).

Kind regards,
Petr

> - Eric

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] [PATCH] statx09: Reduce fs-verity blocksize to 1024

[Petr Vorel]

Hi Eric, Martin,

> Yes kernels < 6.3 TCONF on current master:
> statx09.c:124: TCONF: fs-verity not supported on loopdev
I forget to write the behavior is the same on older kernels with Martin's
patch.

Kind regards,
Petr

> And 6.3 fails:
> statx09.c:109: TBROK: ioctl(3, FS_IOC_ENABLE_VERITY) failed: EINVAL (22)

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Re: [LTP] [PATCH] statx09: Reduce fs-verity blocksize to 1024

[Martin Doucha]

On 14. 07. 23 13:47, Petr Vorel wrote:
>> There are a few different approaches you could take.  One would be the one you
>> mentioned: just add "-b 4096".  Though, note that systems with non-4K pages and
>> kernel older than v6.3 can't mount an "-O verity -b 4096" filesystem.
> 
> I suppose the problem would be with e.g. ppc64le. Because otherwise
> it works with this setup (untested on ppc64le, I believe it'd fail):
> .dev_fs_opts = (const char *const []){"-O verity", "-b 4096", NULL},
> and kept enable.block_size = 4096
> the test passes on older kernels 6.2.12-1-default (openSUSE), 5.10.0-8-amd64
> (Debian) as well as on 6.3 kernel.
> 
>> Or you could query the filesystem block size (that resulted from mkfs.ext4 with
>> unspecified block size) and use that as the Merkle tree block size.
> 
> Unless it's not too complicated, it looks to me the best, as it'd support both
>   >= 6.3 and older kernels.

Querying the filesystem size doesn't help on older kernels because 
mkfs.ext4 still creates the filesystem with small blocksize and setup() 
exits with TCONF because it can't mount the loop device, as you've seen 
in your own experiments. But it'll work on kernels v6.3+ so I've 
submitted that as v2.

For the older kernels, we'll have to run mkfs.ext4 in setup() with 
dynamically constructed command line arguments and set blocksize 
explicitly to pagesize.

-- 
Martin Doucha   mdoucha@suse.cz
SW Quality Engineer
SUSE LINUX, s.r.o.
CORSO IIa
Krizikova 148/34
186 00 Prague 8
Czech Republic


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp