Re: [PATCH v2] ntfs: don't let a fatal signal interrupt metadata reads

From: xkeyC

Date: Thu Nov 12 2026 - 02:06:51 EST


Hi Baolin,

On 2026/11/12 14:49, liubaolin wrote:
>> + folio_wait_locked(folio);
>> + read_failed = !folio_test_uptodate(folio);
>> + folio_put(folio);
>> + goto repeat;
>
> Could we keep the reference here and recheck uptodate under the
> folio lock? We're dropping it even when the read succeeded, so reclaim
> could remove the folio before the next lookup and we'd have to read it
> again. We'd still need to check the mapping after taking the lock.

Good point, thanks. I'll do that in v3.

Thanks,
xkeyC