Re: [PATCH v5] NFSD: Track SCSI Persistent Registration Fencing per Client with xarray

["Chuck Lever" <cel@kernel.org>]

On Wed, Jan 7, 2026, at 11:17 AM, Christoph Hellwig wrote:
> On Fri, Dec 26, 2025 at 11:24:37PM -0500, Chuck Lever wrote:
>
>> @@ -342,6 +343,20 @@ nfsd4_block_get_device_info_scsi(struct super_block *sb,
>>  		goto out_free_dev;
>>  	}
>>  
>> +	/*
>> +	 * xa_store() is idempotent -- the device is added exactly once
>> +	 * to a client's cl_dev_fences no matter how many times
>> +	 * nfsd4_block_get_device_info_scsi() is invoked. This prevents
>> +	 * adding more entries to cl_dev_fences than there are devices on
>> +	 * the server. XA_MARK_0 tracks whether the device has been fenced.
>> +	 */
>
> Only in the most abstract way, xa_store actually stores the new
> entry/value and returns the old one.  In other words, this does
> quite a bit of work for the non-initial GETDEVICEINFO.

The comment is misleading then, and should be clarified.

But should the code use xa_insert instead of xa_store ?


-- 
Chuck Lever