Re: [PATCH v2 2/5] ext4: Add unit test of ext4_mb_generate_buddy

[Guenter Roeck]

On 3/22/24 02:27, Kemeng Shi wrote:
> on 3/21/2024 3:16 PM, Kemeng Shi wrote:
> > on 3/21/2024 12:23 AM, Guenter Roeck wrote:
> > > Hi,
> > >
> > > On Wed, Jan 03, 2024 at 06:48:57PM +0800, Kemeng Shi wrote:
> > > > Add unit test of ext4_mb_generate_buddy
> > > >
> > > > Signed-off-by: Kemeng Shi <shikemeng@xxxxxxxxxxxxxxx>
> > >
> > > With this and other new ext4 tests test in the tree, I see a variety
> > > of backtraces in the upstream kernel if debug options are enabled.
> > > An example is
> > >
> > > [    6.821447]         KTAP version 1
> > > [    6.821769]         # Subtest: test_mb_generate_buddy
> > > [    6.824787] =============================================================================
> > > [    6.825568] BUG inode_cache (Tainted: G                 N): Padding overwritten. 0xfffff80006223f68-0xfffff80006223f6f @offset=16232
> > > ...
> > > [    6.894341] ok 7 ext4_inode_test
> > > [    6.895411] =============================================================================
> > > [    6.895777] BUG inode_cache (Tainted: G    B            N): Padding overwritten. 0xfffff80006223f68-0xfffff80006223f6f @offset=16232
> > >
> > > Another example, from another test run, is
> > >
> > > [    3.938551]         # Subtest: test_new_blocks_simple
> > > [    3.947171]         ok 1 block_bits=10 cluster_bits=3 blocks_per_group=8192 group_count=4 desc_size=64
> > > [    3.952988]         ok 2 block_bits=12 cluster_bits=3 blocks_per_group=8192 group_count=4 desc_size=64
> > > [    3.958403]         ok 3 block_bits=16 cluster_bits=3 blocks_per_group=8192 group_count=4 desc_size=64
> > > [    3.958890] =============================================================================
> > > [    3.959159] BUG inode_cache (Tainted: G                 N): Padding overwritten. 0xffff8de881adbf68-0xffff8de881adbf6f @offset=16232
> > >
> > > Another one:
> > >
> > > [   18.730473]         # Subtest: test_new_blocks_simple
> > > [   18.760547]         ok 1 block_bits=10 cluster_bits=3 blocks_per_group=8192 group_count=4 desc_size=64
> > > [   18.778477] ==================================================================
> > > [   18.778950] BUG: KFENCE: out-of-bounds write in ext4_mb_init+0x5d7/0xa60
> > >
> > > This is just a sample, taken from a quick look at test results.
> > >
> > > Are those backtraces expected ? If so, would it be possible to execute the
> > > tests without generating such backtraces ? The backtraces, if intentional,
> > > hide real problems in the noise.
> > Thanks for the report. The backtrace is not expected, I will look into this. Thansk!
> > >
> Hi Guenter, I could not reproduce this in my local vm. From the reported backtraces, it's
> likely there is a out-of-bounds write to sbi->s_buddy_cache. I try to fix this in [1] and
> it works fine in my local vm. I wish this work for you to elimate the reported nosie.
> Look forward to your reply, Thanks.

You would need to have CONFIG_SLUB_DEBUG=y, CONFIG_SLUB_DEBUG_ON=y, and CONFIG_KFENCE=y
to see the problems.

Guenter