Re: [PATCH] ext4: enable mballoc kunit tests for blocksize > PAGE_SIZE

[Baokun Li <libaokun@linux.alibaba.com>]

Please ignore this patch.

During review, Sashiko noticed a potential out-of-bounds write when
CONFIG_TRANSPARENT_HUGEPAGE is disabled.

This has been fixed in v2 by adding a CONFIG_TRANSPARENT_HUGEPAGE check:

v2:
https://patch.msgid.link/20260507083754.1646636-1-libaokun@linux.alibaba.com


Regards,
Baokun

在 2026/5/6 15:59, Baokun Li 写道:
> With Large Block Size (LBS) support, ext4 can now use block sizes larger
> than PAGE_SIZE. The mballoc kunit tests previously skipped three test
> cases (test_mb_mark_used, test_mb_free_blocks, test_mb_mark_used_cost)
> under this configuration because the buddy cache inode's folio mapping
> order was never initialized in the test harness.
>
> The real mount path configures s_min_folio_order and s_max_folio_order
> in ext4_fill_super(), which allows ext4_set_inode_mapping_order() to
> set up the correct folio order for the buddy cache inode. The kunit
> test bypasses ext4_fill_super(), so the mapping order stayed at zero
> and __filemap_get_folio() allocated order-0 folios too small for LBS.
>
> Initialize s_min_folio_order and s_max_folio_order in mbt_init_sb_layout()
> to mirror ext4_fill_super() behavior, enabling properly sized folio
> allocations and removing the three blocksize > PAGE_SIZE skips.
>
> Signed-off-by: Baokun Li <libaokun@linux.alibaba.com>
> ---
>  fs/ext4/mballoc-test.c | 14 ++------------
>  1 file changed, 2 insertions(+), 12 deletions(-)
>
> diff --git a/fs/ext4/mballoc-test.c b/fs/ext4/mballoc-test.c
> index 90ed505fa4b1..04bc9f773d63 100644
> --- a/fs/ext4/mballoc-test.c
> +++ b/fs/ext4/mballoc-test.c
> @@ -206,6 +206,8 @@ static void mbt_init_sb_layout(struct super_block *sb,
>  	sbi->s_desc_per_block_bits =
>  		sb->s_blocksize_bits - (fls(layout->desc_size) - 1);
>  	sbi->s_desc_per_block = 1 << sbi->s_desc_per_block_bits;
> +	sbi->s_min_folio_order = get_order(sb->s_blocksize);
> +	sbi->s_max_folio_order = sbi->s_min_folio_order;
>  
>  	es->s_first_data_block = cpu_to_le32(0);
>  	es->s_blocks_count_lo = cpu_to_le32(layout->blocks_per_group *
> @@ -791,10 +793,6 @@ static void test_mb_mark_used(struct kunit *test)
>  	struct test_range ranges[TEST_RANGE_COUNT];
>  	int i;
>  
> -	/* buddy cache assumes that each page contains at least one block */
> -	if (sb->s_blocksize > PAGE_SIZE)
> -		kunit_skip(test, "blocksize exceeds pagesize");
> -
>  	bitmap = kunit_kzalloc(test, sb->s_blocksize, GFP_KERNEL);
>  	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, bitmap);
>  	buddy = kunit_kzalloc(test, sb->s_blocksize, GFP_KERNEL);
> @@ -858,10 +856,6 @@ static void test_mb_free_blocks(struct kunit *test)
>  	int i;
>  	struct test_range ranges[TEST_RANGE_COUNT];
>  
> -	/* buddy cache assumes that each page contains at least one block */
> -	if (sb->s_blocksize > PAGE_SIZE)
> -		kunit_skip(test, "blocksize exceeds pagesize");
> -
>  	bitmap = kunit_kzalloc(test, sb->s_blocksize, GFP_KERNEL);
>  	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, bitmap);
>  	buddy = kunit_kzalloc(test, sb->s_blocksize, GFP_KERNEL);
> @@ -905,10 +899,6 @@ static void test_mb_mark_used_cost(struct kunit *test)
>  	int i, j;
>  	unsigned long start, end, all = 0;
>  
> -	/* buddy cache assumes that each page contains at least one block */
> -	if (sb->s_blocksize > PAGE_SIZE)
> -		kunit_skip(test, "blocksize exceeds pagesize");
> -
>  	ret = ext4_mb_load_buddy_test(sb, TEST_GOAL_GROUP, &e4b);
>  	KUNIT_ASSERT_EQ(test, ret, 0);
>