Re: [PATCH 8/8] ext4: optimize inflight dio tracking for zeroout

[sashiko-bot@kernel.org]

> Earlier zeroout used inode->i_dio_count to detect in-flight DIOs
> that might have valid data underneath an unwritten extent. However,
> i_dio_count accounts for every DIO on the inode, including reads and writes
> that never touch an unwritten extent, which makes it prone to false
> positives and needlessly suppresses the zeroout fallback.
> 
> Optimize this by moving DIO accounting to i_conv_pending_ios, which is only
> incremented for DIOs that actually encounter an unwritten extent, and
> decremented once at endio after the conversion is done.
> 
> The main issue is that ext4_iomap_begin() doesn't have any way to track
> something only once per dio, hence, we use a custom ext4_iomap_next() that
> utilizes iter->private to ensure we increment the i_conv_pending_ios only
> once per a dio that encounters unwritten extents.
> 
> Now that dio has moved to i_conv_pending_ios, we need to pass NO_CONV_ACC
> get blocks flag to ext4_ext_convert_to_initialized(). This path is
> reachable from dio but doesn't need to account in i_conv_pending_ios
> because we don't issue any IO before conversion, hence the flag helps
> zeroout to make the right call.
> 
> Signed-off-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261102-ext4-endio-zero-bug-v1-0-266d2bb9e6dc@linux.ibm.com?part=8