Re: [PATCH 1/5] xfs: make attr lookup returns consistent

[Christoph Hellwig <hch@infradead.org>]

On Thu, Aug 29, 2019 at 09:35:01PM +1000, Dave Chinner wrote:
> +/*
> + * Retrieve an extended attribute and its value.  Must have ilock.
> + * Returns 0 on successful retrieval, otherwise an error.
> + */
>  int
>  xfs_attr_get_ilocked(

This just updates a comment on xfs_attr_get_ilocked, no other change,
so looks good.

>
>  	struct xfs_inode	*ip,
> @@ -147,7 +150,7 @@ xfs_attr_get(
>  	xfs_iunlock(ip, lock_mode);
>  
>  	*valuelenp = args.valuelen;
> -	return error == -EEXIST ? 0 : error;
> +	return error;
>  }

This drops a conversion from -EEXIST to 0 at the very top of the
callchain.

> -	if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
> -		error = xfs_attr_rmtval_get(args);
> -	}
> -	return error;
> +	if (error)
> +		return error;
> +
> +	/* check if we have to retrieve a remote attribute to get the value */
> +	if (args->flags & ATTR_KERNOVAL)
> +		return 0;
> +	if (!args->rmtblkno)
> +		return 0;
> +	return xfs_attr_rmtval_get(args);

No change at all here, just a more verbose style.

>  STATIC int
>  xfs_attr_node_get(xfs_da_args_t *args)
> @@ -1294,24 +1306,27 @@ xfs_attr_node_get(xfs_da_args_t *args)
>  	error = xfs_da3_node_lookup_int(state, &retval);
>  	if (error) {
>  		retval = error;
> -	} else if (retval == -EEXIST) {
> -		blk = &state->path.blk[ state->path.active-1 ];
> -		ASSERT(blk->bp != NULL);
> -		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
> -
> -		/*
> -		 * Get the value, local or "remote"
> -		 */
> -		retval = xfs_attr3_leaf_getvalue(blk->bp, args);
> -		if (!retval && (args->rmtblkno > 0)
> -		    && !(args->flags & ATTR_KERNOVAL)) {
> -			retval = xfs_attr_rmtval_get(args);
> -		}
> +		goto out_release;
>  	}
> +	if (retval != -EEXIST)
> +		goto out_release;
> +
> +	/*
> +	 * Get the value, local or "remote"
> +	 */
> +	blk = &state->path.blk[state->path.active - 1];
> +	retval = xfs_attr3_leaf_getvalue(blk->bp, args);
> +	if (retval)
> +		goto out_release;
> +	if (args->flags & ATTR_KERNOVAL)
> +		goto out_release;
> +	if (args->rmtblkno > 0)
> +		retval = xfs_attr_rmtval_get(args);
>  
>  	/*
>  	 * If not in a transaction, we have to release all the buffers.
>  	 */
> +out_release:

No change at all here as far as I can tell, just using a goto to
unwind error handling, and rewriting some code to be less
dense/obsfucated.

>  int
>  xfs_attr_shortform_getvalue(xfs_da_args_t *args)
>  {
> @@ -743,7 +746,7 @@ xfs_attr_shortform_getvalue(xfs_da_args_t *args)
>  			continue;
>  		if (args->flags & ATTR_KERNOVAL) {
>  			args->valuelen = sfe->valuelen;
> -			return -EEXIST;
> +			return 0;
>  		}
>  		if (args->valuelen < sfe->valuelen) {
>  			args->valuelen = sfe->valuelen;
> @@ -752,7 +755,7 @@ xfs_attr_shortform_getvalue(xfs_da_args_t *args)
>  		args->valuelen = sfe->valuelen;
>  		memcpy(args->value, &sfe->nameval[args->namelen],
>  						    args->valuelen);
> -		return -EEXIST;
> +		return 0;

And here we have the first change - xfs_attr_shortform_getvalue now
returns 0 instead of -EEXIST when finding the xattr.  The old calling
conventions does indeed seem very odd.  As xfs_attr_shortform_getvalue
is directly called by the small xfs_attr_get_ilocked wrapper, which
only has two calers that fix up -EEXIST to 0 this looks safe.

So overall this looks good, but I think having this split into two
pure cleanups and one to actually switch this single case away from
-EEXIST (and document the chain as I wrote above while following it)
would have been more helpful.

So what:

Reviewed-by: Christoph Hellwig <hch@lst.de>