On Wed 21-09-22 21:42:18, Baokun Li wrote:
> If the starting position of our insert range happens to be in the hole
> between the two ext4_extent_idx, because the lblk of the ext4_extent in
> the previous ext4_extent_idx is always less than the start, which leads
> to the "extent" variable access across the boundary, the following UAF is
> triggered:
> ==================================================================
> BUG: KASAN: use-after-free in ext4_ext_shift_extents+0x257/0x790
> Read of size 4 at addr ffff88819807a008 by task fallocate/8010
> CPU: 3 PID: 8010 Comm: fallocate Tainted: G E 5.10.0+ #492
> Call Trace:
> dump_stack+0x7d/0xa3
> print_address_description.constprop.0+0x1e/0x220
> kasan_report.cold+0x67/0x7f
> ext4_ext_shift_extents+0x257/0x790
> ext4_insert_range+0x5b6/0x700
> ext4_fallocate+0x39e/0x3d0
> vfs_fallocate+0x26f/0x470
> ksys_fallocate+0x3a/0x70
> __x64_sys_fallocate+0x4f/0x60
> do_syscall_64+0x33/0x40
> entry_SYSCALL_64_after_hwframe+0x44/0xa9
> ==================================================================
>
> To solve this issue, when the ee_block of the last extent is less than
> the start, exit the loop in advance to avoid UAF.
>
> Fixes: 331573febb6a ("ext4: Add support FALLOC_FL_INSERT_RANGE for fallocate")
> Signed-off-by: Baokun Li <libaokun1@xxxxxxxxxx>
> Signed-off-by: Zhihao Cheng <chengzhihao1@xxxxxxxxxx>
Nice catch. The fix looks mostly good, just one small thing noted below.
> diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> index c148bb97b527..25fc1f4b35a5 100644
> --- a/fs/ext4/extents.c
> +++ b/fs/ext4/extents.c
> @@ -5216,11 +5216,18 @@ ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
> }
>
> tmp = *iterator;
> + extent = EXT_LAST_EXTENT(path[depth].p_hdr);
> if (SHIFT == SHIFT_LEFT) {
> - extent = EXT_LAST_EXTENT(path[depth].p_hdr);
> *iterator = le32_to_cpu(extent->ee_block) +
> ext4_ext_get_actual_len(extent);
> } else {
> + /*
> + * start happens to be in the hole between
> + * the two ext4_extent_idx.
> + */
> + if (le32_to_cpu(extent->ee_block) < start)
> + break;
I think you need to initialize 'ret' somewhere (probably just after the
again: label would make most sense) so that we don't accidentally return
-EAGAIN here.
> +
> extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
> if (le32_to_cpu(extent->ee_block) > 0)
> *iterator = le32_to_cpu(extent->ee_block) - 1;
Honza
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR
