Re: [PATCH v2 1/7] block: fix zone write plug removal

[Damien Le Moal <dlemoal@kernel.org>]

On 2/27/26 01:10, Christoph Hellwig wrote:
> On Thu, Feb 26, 2026 at 01:10:18PM +0900, Damien Le Moal wrote:
>>  		/*
>> -		 * Check that a BIO completion or a zone reset or finish
>> -		 * operation has not already removed the zone write plug from
>> -		 * the hash table and dropped its reference count. In such case,
>> -		 * we need to get a new plug so start over from the beginning.
>> +		 * We already have a zone write plug. If it is flagged as dead,
>> +		 * its initial reference count was already dropped. In
>> +		 * this case, increment the reference count and clear the dead
>> +		 * flag to restore the plug to a state similar to a newly
>> +		 * allocated zone write plug.
> 
> This comment, both in the old and new version describes what happens
> here, but it really fails to explain why.  The plug is scheduled
> for removal if we either filled up the zone or reset it, so any new
> bio coming in would result in an I/O error anyway.  So why do we try
> to resurrect the plug instead of just failing the I/O?

Hmm, yes, I guess we could do that indeed.

This case can happen only if the user starts doing stupid things like issuing a
zone reset and writes from the beginning of the zone without waiting for the
reset first. So yes, I guess we can fail the BIO if it hits a dead zone, since
that necessarily mean that the user did not synchronize something that should be.

I will drop that bit and instead print a rate limited error message signaling
the bad usage.


-- 
Damien Le Moal
Western Digital Research