Re: [PATCH 1/1] btrfs: add integer overflow protection to flush_dir_items_batch allocation

From: kmpfqgdwxucqz9
Date: Wed Jul 30 2025 - 02:58:43 EST


From: KernelKraze <admin@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>

Hi Johannes,

Thanks for the review.

On 7/30/25 6:35 AM, Johannes Thumshirn wrote:
> Where does this number come from?

It's from log_delayed_insertion_items() at line 6111:

/* 195 (4095 bytes of keys and sizes) fits in a single 4K page. */
const int max_batch_size = 195;

I reused this limit for consistency across btrfs batch operations.

> Wouldn't kcalloc() or kmalloc_array() be the better choice here?
> kcalloc() calls kmalloc_array() which in term does overflow checking.

Good point. The issue is we're allocating a mixed buffer:

[u32 sizes array][struct btrfs_key keys array]

kmalloc_array() handles single-type arrays, but we need:
- ins_sizes = (u32 *)ins_data
- ins_keys = (struct btrfs_key *)(ins_data + sizes_size)

Two options:
1. Keep current approach with manual overflow checks
2. Split into separate kmalloc_array() calls (potential cache miss cost)

Which would you prefer? I'm happy to rework it either way.

Thanks,
KernelKraze