On May 8, 2019, at 12:13 AM, Sahitya Tummala <stummala@xxxxxxxxxxxxxx> wrote:
>
> The buffer_head (frames[0].bh) and it's corresping page can be
> potentially free'd once brelse() is done inside the for loop
> but before the for loop exits in dx_release(). It can be free'd
> in another context, when the page cache is flushed via
> drop_caches_sysctl_handler(). This results into below data abort
> when accessing info->indirect_levels in dx_release().
>
> Unable to handle kernel paging request at virtual address ffffffc17ac3e01e
> Call trace:
> dx_release+0x70/0x90
> ext4_htree_fill_tree+0x2d4/0x300
> ext4_readdir+0x244/0x6f8
> iterate_dir+0xbc/0x160
> SyS_getdents64+0x94/0x174
>
> Signed-off-by: Sahitya Tummala <stummala@xxxxxxxxxxxxxx>
The patch looks reasonable, but there is a danger that it may be
"optimized" back to the pre-patch form again. It probably makes
sense to include a comment like:
/* save local copy, "info" may be freed after brelse() */
Looks fine otherwise.
Reviewed-by: Andreas Dilger <adilger@xxxxxxxxx>
> ---
> fs/ext4/namei.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
> index 4181c9c..7e6c298 100644
> --- a/fs/ext4/namei.c
> +++ b/fs/ext4/namei.c
> @@ -871,12 +871,14 @@ static void dx_release(struct dx_frame *frames)
> {
> struct dx_root_info *info;
> int i;
> + unsigned int indirect_levels;
>
> if (frames[0].bh == NULL)
> return;
>
> info = &((struct dx_root *)frames[0].bh->b_data)->info;
> - for (i = 0; i <= info->indirect_levels; i++) {
> + indirect_levels = info->indirect_levels;
> + for (i = 0; i <= indirect_levels; i++) {
> if (frames[i].bh == NULL)
> break;
> brelse(frames[i].bh);
> --
> Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc.
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.
>
Cheers, Andreas
Attachment:
signature.asc
Description: Message signed with OpenPGP
