Re: [PATCH v4 08/17] xfs: Factor out xfs_attr_leaf_addname helper

[Allison Collins <allison.henderson@oracle.com>]

On 11/8/19 1:57 PM, Darrick J. Wong wrote:
> On Wed, Nov 06, 2019 at 06:27:52PM -0700, Allison Collins wrote:
>> Factor out new helper function xfs_attr_leaf_try_add.
>> Because new delayed attribute routines cannot roll
>> transactions, we carve off the parts of
>> xfs_attr_leaf_addname that we can use.  This will help
>> to reduce repetitive code later when we introduce
>> delayed attributes.
>>
>> Signed-off-by: Allison Collins <allison.henderson@oracle.com>
>> ---
>>   fs/xfs/libxfs/xfs_attr.c | 84 +++++++++++++++++++++++++++++-------------------
>>   1 file changed, 51 insertions(+), 33 deletions(-)
>>
>> diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
>> index 212995f..dda2eba 100644
>> --- a/fs/xfs/libxfs/xfs_attr.c
>> +++ b/fs/xfs/libxfs/xfs_attr.c
>> @@ -305,10 +305,33 @@ xfs_attr_set_args(
>>   		}
>>   	}
>>   
>> -	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
>> +	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
>>   		error = xfs_attr_leaf_addname(args);
>> -	else
>> +		if (error == -ENOSPC) {
>> +			/*
>> +			 * Commit that transaction so that the node_addname()
>> +			 * call can manage its own transactions.
>> +			 */
>> +			error = xfs_defer_finish(&args->trans);
>> +			if (error)
>> +				return error;
>> +
>> +			/*
>> +			 * Commit the current trans (including the inode) and
>> +			 * start a new one.
>> +			 */
>> +			error = xfs_trans_roll_inode(&args->trans, dp);
>> +			if (error)
>> +				return error;
>> +
>> +			/*
>> +			 * Fob the rest of the problem off on the Btree code.
>> +			 */
>> +			error = xfs_attr_node_addname(args);
>> +		}
>> +	} else {
>>   		error = xfs_attr_node_addname(args);
>> +	}
>>   	return error;
> 
> 
>>   }
>>   
>> @@ -601,21 +624,12 @@ xfs_attr_shortform_addname(xfs_da_args_t *args)
>>    * External routines when attribute list is one block
>>    *========================================================================*/
>>   
>> -/*
>> - * Add a name to the leaf attribute list structure
>> - *
>> - * This leaf block cannot have a "remote" value, we only call this routine
>> - * if bmap_one_block() says there is only one block (ie: no remote blks).
>> - */
>>   STATIC int
>> -xfs_attr_leaf_addname(
>> -	struct xfs_da_args	*args)
>> +xfs_attr_leaf_try_add(
> 
> (total stream of consciousness here...)
> 
> AFAICT the old _addname function's responsibilities were:
> 
> 1 Try to add a new attr key entry to the leaf block, with INCOMPLETE set
>    if it's a rename op or we need to set a remote value.
> 2 If there wasn't space in the leaf block, convert to node format, call
>    the node version of this function, and exit.
> 3 Allocating blocks for the remote attr value and writing them, if
>    applicable
> 4 If it's a rename operation, clearing the INCOMPLETE flag on the new
>    entry; setting it on the old entry; and then removing the old entry.
> 5 Clearing the INCOMPLETE flag on the new entry when we're done writing
>    a remote value (if applicable)
> 
> I think we arrive at this split so that we don't have a transaction roll
> in the middle of the function, right?  And also to make the "convert to
> node format and roll" bits go elsewhere?
> 
> The way I'm thinking about how to accomplish this is...
> 
> xfs_attr_leaf_addname should be renamed xfs_attr_leaf_setname, and then
> hoist (1) into a separate function, move (2) into xfs_attr_set_args, and
> hoist (4) into a separate function.
> 
> ...ok, so let's test how closely my understanding fits the changes made
> in this patch:
> 
> _try_add is basically (1).
> 
> Most of (2) happened, though the call to xfs_attr3_leaf_to_node ought to
> go into the caller so that the conversion stays with the defer_finish
> and roll.
> 
> (4) could still be done, maybe as a separate prep patch.
> 

I think you're on the right track.  Perhaps the diff made it look a 
little crazier than it seems. Maybe its easier to describe it as this:

I renamed xfs_attr_leaf_addname to xfs_attr_leaf_try_add,
and then deleted (2) out of the body of the function.  The upper half 
(1) became the helper function, and lower half (3, 4, and 5) became the 
new xfs_attr_leaf_addname, which now calls the helper at the start of 
the routine. Finally (2) got factored up into the caller.

Hope that helps some :-)  The point of all this being to jigsaw out (1) 
into a stand alone routine that we can use later in delayed attrs.


> Hm, ok, I think I understand what this patch does.  The call site in
> xfs_attr_set_args would be clearer (and less indenty) if it looked like:
> 
> 	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
> 		error = xfs_attr_leaf_addname(args);
> 		if (error == 0 || error != -ENOSPC)
> 			return error;
> 
> 		/* Promote the attribute list to node format. */
> 		error = xfs_attr3_leaf_to_node(args);
> 		if (error)
> 			return error;
I think this part here is already taken care of in 
xfs_attr_leaf_addname, but otherwise, I think the rest of this is 
equivalent.  Will update in the next set.

Thanks for the reviews!
Allison



> 
> 		/*
> 		 * Commit that transaction so that the node_addname()
> 		 * call can manage its own transactions.
> 		 */
> 		error = xfs_defer_finish(&args->trans);
> 		if (error)
> 			return error;
> 
> 		/*
> 		 * Commit the current trans (including the inode) and
> 		 * start a new one.
> 		 */
> 		error = xfs_trans_roll_inode(&args->trans, dp);
> 		if (error)
> 			return error;
> 	}
> 
> 	return xfs_attr_node_addname(args);
> 
> But otherwise it looks decent, assuming I understood any of it. :)
> 
> --D
> 
>> +	struct xfs_da_args	*args,
>> +	struct xfs_buf		*bp)
>>   {
>> -	struct xfs_buf		*bp;
>> -	int			retval, error, forkoff;
>> -	struct xfs_inode	*dp = args->dp;
>> -
>> -	trace_xfs_attr_leaf_addname(args);
>> +	int			retval, error;
>>   
>>   	/*
>>   	 * Look up the given attribute in the leaf block.  Figure out if
>> @@ -661,31 +675,35 @@ xfs_attr_leaf_addname(
>>   	retval = xfs_attr3_leaf_add(bp, args);
>>   	if (retval == -ENOSPC) {
>>   		/*
>> -		 * Promote the attribute list to the Btree format, then
>> -		 * Commit that transaction so that the node_addname() call
>> -		 * can manage its own transactions.
>> +		 * Promote the attribute list to the Btree format.
>> +		 * Unless an error occurs, retain the -ENOSPC retval
>>   		 */
>>   		error = xfs_attr3_leaf_to_node(args);
>>   		if (error)
>>   			return error;
>> -		error = xfs_defer_finish(&args->trans);
>> -		if (error)
>> -			return error;
>> +	}
>> +	return retval;
>> +}
>>   
>> -		/*
>> -		 * Commit the current trans (including the inode) and start
>> -		 * a new one.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, dp);
>> -		if (error)
>> -			return error;
>>   
>> -		/*
>> -		 * Fob the whole rest of the problem off on the Btree code.
>> -		 */
>> -		error = xfs_attr_node_addname(args);
>> +/*
>> + * Add a name to the leaf attribute list structure
>> + *
>> + * This leaf block cannot have a "remote" value, we only call this routine
>> + * if bmap_one_block() says there is only one block (ie: no remote blks).
>> + */
>> +STATIC int
>> +xfs_attr_leaf_addname(struct xfs_da_args	*args)
>> +{
>> +	int			error, forkoff;
>> +	struct xfs_buf		*bp = NULL;
>> +	struct xfs_inode	*dp = args->dp;
>> +
>> +	trace_xfs_attr_leaf_addname(args);
>> +
>> +	error = xfs_attr_leaf_try_add(args, bp);
>> +	if (error)
>>   		return error;
>> -	}
>>   
>>   	/*
>>   	 * Commit the transaction that added the attr name so that
>> -- 
>> 2.7.4
>>