[v4,17/17] xfs: Add delay ready attr set routines

Commit Message

Allison Henderson Nov. 7, 2019, 1:28 a.m. UTC

This patch modifies the attr set routines to be delay ready.
This means they no longer roll or commit transactions, but instead
return -EAGAIN to have the calling routine roll and refresh the
transaction.  In this series, xfs_attr_set_args has become
xfs_attr_set_later, which uses a state machine to keep track
of where it was when EAGAIN was returned.  Part of
xfs_attr_leaf_addname has been factored out into a new helper
function xfs_attr_leaf_try_add to allow transaction cycling between
the two routines, and the flipflags logic has been removed since we
can simply cancel the transaction upon error.  xfs_attr_set_args
consists of a simple loop to refresh the transaction until the
operation is completed.

Signed-off-by: Allison Collins <[e-mail]>
---
 fs/xfs/libxfs/xfs_attr.c | 435 +++++++++++++++++++++++------------------------
 fs/xfs/libxfs/xfs_attr.h |   1 +
 2 files changed, 211 insertions(+), 225 deletions(-)

Comments

Darrick J. Wong Nov. 8, 2019, 9:42 p.m. UTC | #1

On Wed, Nov 06, 2019 at 06:28:01PM -0700, Allison Collins wrote:
> This patch modifies the attr set routines to be delay ready.
> This means they no longer roll or commit transactions, but instead
> return -EAGAIN to have the calling routine roll and refresh the
> transaction.  In this series, xfs_attr_set_args has become
> xfs_attr_set_later, which uses a state machine to keep track
> of where it was when EAGAIN was returned.  Part of
> xfs_attr_leaf_addname has been factored out into a new helper
> function xfs_attr_leaf_try_add to allow transaction cycling between
> the two routines, and the flipflags logic has been removed since we
> can simply cancel the transaction upon error.  xfs_attr_set_args
> consists of a simple loop to refresh the transaction until the
> operation is completed.
> 
> Signed-off-by: Allison Collins <[e-mail]>
> ---
>  fs/xfs/libxfs/xfs_attr.c | 435 +++++++++++++++++++++++------------------------
>  fs/xfs/libxfs/xfs_attr.h |   1 +
>  2 files changed, 211 insertions(+), 225 deletions(-)
> 
> diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
> index 38d5c5c..97e5ae0 100644
> --- a/fs/xfs/libxfs/xfs_attr.c
> +++ b/fs/xfs/libxfs/xfs_attr.c
> @@ -58,6 +58,7 @@ STATIC int xfs_attr_node_hasname(xfs_da_args_t *args,
>  				 struct xfs_da_state **state);
>  STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
>  STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
> +STATIC int xfs_attr_leaf_try_add(struct xfs_da_args *args, struct xfs_buf *bp);
>  
>  
>  STATIC int
> @@ -250,9 +251,79 @@ int
>  xfs_attr_set_args(
>  	struct xfs_da_args	*args)
>  {
> +	int			error = 0;
> +	int			err2 = 0;
> +	struct xfs_buf		*leaf_bp = NULL;
> +
> +	do {
> +		error = xfs_attr_set_later(args, &leaf_bp);
> +		if (error && error != -EAGAIN)
> +			goto out;
> +
> +		xfs_trans_log_inode(args->trans, args->dp,
> +				    XFS_ILOG_CORE | XFS_ILOG_ADATA);

Same question as the last patch about ADATA.

> +
> +		err2 = xfs_trans_roll(&args->trans);
> +		if (err2) {
> +			error = err2;
> +			goto out;
> +		}
> +
> +		/* Rejoin inode and leaf if needed */
> +		xfs_trans_ijoin(args->trans, args->dp, 0);
> +		if (leaf_bp) {
> +			xfs_trans_bjoin(args->trans, leaf_bp);
> +			xfs_trans_bhold(args->trans, leaf_bp);
> +		}
> +
> +	} while (error == -EAGAIN);
> +
> +out:
> +	return error;
> +}
> +
> +/*
> + * Set the attribute specified in @args.
> + * This routine is meant to function as a delayed operation, and may return
> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
> + * to handle this, and recall the function until a successful error code is
> + * returned.
> + */
> +int
> +xfs_attr_set_later(
> +	struct xfs_da_args	*args,
> +	struct xfs_buf          **leaf_bp)
> +{
>  	struct xfs_inode	*dp = args->dp;
> -	struct xfs_buf          *leaf_bp = NULL;
> -	int			error, error2 = 0;;
> +	int			error = 0;
> +	int			sf_size;
> +
> +	/* State machine switch */
> +	switch (args->dc.dc_state) {
> +	case XFS_DC_SF_TO_LEAF:
> +		goto sf_to_leaf;
> +	case XFS_DC_ALLOC_LEAF:
> +	case XFS_DC_FOUND_LBLK:
> +		goto leaf;
> +	case XFS_DC_FOUND_NBLK:
> +	case XFS_DC_ALLOC_NODE:
> +	case XFS_DC_LEAF_TO_NODE:
> +		goto node;
> +	default:
> +		break;
> +	}
> +
> +	/*
> +	 * New inodes may not have an attribute fork yet. So set the attribute
> +	 * fork appropriately
> +	 */
> +	if (XFS_IFORK_Q((args->dp)) == 0) {
> +		sf_size = sizeof(struct xfs_attr_sf_hdr) +
> +		     XFS_ATTR_SF_ENTSIZE_BYNAME(args->name.len, args->valuelen);
> +		xfs_bmap_set_attrforkoff(args->dp, sf_size, NULL);
> +		args->dp->i_afp = kmem_zone_zalloc(xfs_ifork_zone, 0);
> +		args->dp->i_afp->if_flags = XFS_IFEXTENTS;
> +	}
>  
>  	/*
>  	 * If the attribute list is non-existent or a shortform list,
> @@ -272,21 +343,14 @@ xfs_attr_set_args(
>  		 * Try to add the attr to the attribute list in the inode.
>  		 */
>  		error = xfs_attr_try_sf_addname(dp, args);
> -		if (error != -ENOSPC) {
> -			if (dp->i_mount->m_flags & XFS_MOUNT_WSYNC)
> -				xfs_trans_set_sync(args->trans);

Where does the xfs_trans_set_sync call go?  Do we not need it anymore?

> -			error2 = xfs_trans_commit(args->trans);
> -			args->trans = NULL;
> -			return error ? error : error2;
> -		}
> -
> +		if (error != -ENOSPC)
> +			return error;
>  
>  		/*
>  		 * It won't fit in the shortform, transform to a leaf block.
>  		 * GROT: another possible req'mt for a double-split btree op.
>  		 */
> -		error = xfs_attr_shortform_to_leaf(args, &leaf_bp);
> +		error = xfs_attr_shortform_to_leaf(args, leaf_bp);
>  		if (error)
>  			return error;
>  
> @@ -294,43 +358,42 @@ xfs_attr_set_args(
>  		 * Prevent the leaf buffer from being unlocked so that a
>  		 * concurrent AIL push cannot grab the half-baked leaf
>  		 * buffer and run into problems with the write verifier.
> -		 * Once we're done rolling the transaction we can release
> -		 * the hold and add the attr to the leaf.
>  		 */
> -		xfs_trans_bhold(args->trans, leaf_bp);
> -		error = xfs_defer_finish(&args->trans);
> -		xfs_trans_bhold_release(args->trans, leaf_bp);
> -		if (error) {
> -			xfs_trans_brelse(args->trans, leaf_bp);
> -			return error;
> -		}
> +
> +		xfs_trans_bhold(args->trans, *leaf_bp);
> +		args->dc.dc_state = XFS_DC_SF_TO_LEAF;
> +		return -EAGAIN;
> +	}
> +sf_to_leaf:
> +
> +	/*
> +	 * After a shortform to leaf conversion, we need to hold the leaf and
> +	 * cylce out the transaction.  When we get back, we need to release

           "cycle"

> +	 * the leaf.
> +	 */
> +	if (*leaf_bp != NULL) {
> +		xfs_trans_brelse(args->trans, *leaf_bp);
> +		*leaf_bp = NULL;
>  	}
>  
>  	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
> +		error = xfs_attr_leaf_try_add(args, *leaf_bp);
> +		if (error == -ENOSPC)
> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
> +		else if (error)
> +			return error;
> +		else
> +			args->dc.dc_state = XFS_DC_FOUND_LBLK;
> +		return -EAGAIN;

Please use a switch statement here...

switch (error) {
case -ENOSPC:
	dc_state = LEAF_TO_NODE;
	return -EAGAIN;
case 0:
	dc_state = FOUND_LBLK;
	return -EAGAIN;
default:
	return error;
}

> +leaf:
>  		error = xfs_attr_leaf_addname(args);
>  		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);
> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
> +			return -EAGAIN;
>  		}
>  	} else {
> +		args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
> +node:
>  		error = xfs_attr_node_addname(args);
>  	}
>  	return error;
> @@ -764,27 +827,26 @@ xfs_attr_leaf_try_add(
>   *
>   * 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).
> + *
> + * This routine is meant to function as a delayed operation, and may return
> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
> + * to handle this, and recall the function until a successful error code is
> + * returned.
>   */
>  STATIC int
>  xfs_attr_leaf_addname(struct xfs_da_args	*args)
>  {
> -	int			error, forkoff;
> -	struct xfs_buf		*bp = NULL;
> +	int			error, nmap;
>  	struct xfs_inode	*dp = args->dp;
> +	struct xfs_bmbt_irec	*map = &args->dc.map;
>  
> -	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
> -	 * later routines can manage their own transactions.
> -	 */
> -	error = xfs_trans_roll_inode(&args->trans, dp);
> -	if (error)
> -		return error;
> +	/* State machine switch */
> +	switch (args->dc.dc_state) {
> +	case XFS_DC_ALLOC_LEAF:
> +		goto alloc_leaf;
> +	default:
> +		break;
> +	}
>  
>  	/*
>  	 * If there was an out-of-line value, allocate the blocks we
> @@ -793,90 +855,58 @@ xfs_attr_leaf_addname(struct xfs_da_args	*args)
>  	 * maximum size of a transaction and/or hit a deadlock.
>  	 */
>  	if (args->rmtblkno > 0) {
> -		error = xfs_attr_rmtval_set(args);
> -		if (error)
> -			return error;
> -	}
>  
> -	/*
> -	 * If this is an atomic rename operation, we must "flip" the
> -	 * incomplete flags on the "new" and "old" attribute/value pairs
> -	 * so that one disappears and one appears atomically.  Then we
> -	 * must remove the "old" attribute/value pair.
> -	 */
> -	if (args->op_flags & XFS_DA_OP_RENAME) {
> -		/*
> -		 * In a separate transaction, set the incomplete flag on the
> -		 * "old" attr and clear the incomplete flag on the "new" attr.
> -		 */
> -		error = xfs_attr3_leaf_flipflags(args);
> -		if (error)
> -			return error;
> -		/*
> -		 * Commit the flag value change and start the next trans in
> -		 * series.
> -		 */
> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
> -		if (error)
> -			return error;
> +		/* Open coded xfs_attr_rmtval_set without trans handling */

Can't we just fix that instead of reproducing it here?

Especially because I think I see it twice in this patch?

Also, do you have a git tree handy?  I /think/ I see how this works but
oh man is it difficult to see that from patches alone.

--D

>  
> -		/*
> -		 * Dismantle the "old" attribute/value pair by removing
> -		 * a "remote" value (if it exists).
> -		 */
> -		args->index = args->index2;
> -		args->blkno = args->blkno2;
> -		args->rmtblkno = args->rmtblkno2;
> -		args->rmtblkcnt = args->rmtblkcnt2;
> -		args->rmtvaluelen = args->rmtvaluelen2;
> -		if (args->rmtblkno) {
> -			error = xfs_attr_rmtval_remove(args);
> -			if (error)
> -				return error;
> -		}
> +		args->dc.lfileoff = 0;
> +		args->dc.lblkno = 0;
> +		args->dc.blkcnt = 0;
> +		args->rmtblkcnt = 0;
> +		args->rmtblkno = 0;
> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>  
> -		/*
> -		 * Read in the block containing the "old" attr, then
> -		 * remove the "old" attr from that block (neat, huh!)
> -		 */
> -		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno,
> -					   XFS_DABUF_MAP_NOMAPPING, &bp);
> +		error = xfs_attr_rmt_find_hole(args);
>  		if (error)
>  			return error;
>  
> -		xfs_attr3_leaf_remove(bp, args);
> +		args->dc.blkcnt = args->rmtblkcnt;
> +		args->dc.lblkno = args->rmtblkno;
>  
>  		/*
> -		 * If the result is small enough, shrink it all into the inode.
> +		 * Roll through the "value", allocating blocks on disk as
> +		 * required.
>  		 */
> -		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
> -			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
> -			/* bp is gone due to xfs_da_shrink_inode */
> -			if (error)
> -				return error;
> -			error = xfs_defer_finish(&args->trans);
> +alloc_leaf:
> +		while (args->dc.blkcnt > 0) {
> +			nmap = 1;
> +			error = xfs_bmapi_write(args->trans, dp,
> +				  (xfs_fileoff_t)args->dc.lblkno,
> +				  args->dc.blkcnt, XFS_BMAPI_ATTRFORK,
> +				  args->total, map, &nmap);
>  			if (error)
>  				return error;
> -		}
> +			ASSERT(nmap == 1);
> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
> +			       (map->br_startblock != HOLESTARTBLOCK));
>  
> -		/*
> -		 * Commit the remove and start the next trans in series.
> -		 */
> -		error = xfs_trans_roll_inode(&args->trans, dp);
> +			/* roll attribute extent map forwards */
> +			args->dc.lblkno += map->br_blockcount;
> +			args->dc.blkcnt -= map->br_blockcount;
>  
> -	} else if (args->rmtblkno > 0) {
> -		/*
> -		 * Added a "remote" value, just clear the incomplete flag.
> -		 */
> -		error = xfs_attr3_leaf_clearflag(args);
> +			args->dc.dc_state = XFS_DC_ALLOC_LEAF;
> +			return -EAGAIN;
> +		}
> +
> +		error = xfs_attr_rmtval_set_value(args);
>  		if (error)
>  			return error;
> +	}
>  
> +	if (args->rmtblkno > 0) {
>  		/*
> -		 * Commit the flag value change and start the next trans in
> -		 * series.
> +		 * Added a "remote" value, just clear the incomplete flag.
>  		 */
> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
> +		error = xfs_attr3_leaf_clearflag(args);
>  	}
>  	return error;
>  }
> @@ -1017,16 +1047,23 @@ xfs_attr_node_hasname(
>   *
>   * "Remote" attribute values confuse the issue and atomic rename operations
>   * add a whole extra layer of confusion on top of that.
> + *
> + * This routine is meant to function as a delayed operation, and may return
> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
> + * to handle this, and recall the function until a successful error code is
> + *returned.
>   */
>  STATIC int
>  xfs_attr_node_addname(
>  	struct xfs_da_args	*args)
>  {
> -	struct xfs_da_state	*state;
> +	struct xfs_da_state	*state = NULL;
>  	struct xfs_da_state_blk	*blk;
>  	struct xfs_inode	*dp;
> -	struct xfs_mount	*mp;
> -	int			retval, error;
> +	int			retval = 0;
> +	int			error = 0;
> +	int			nmap;
> +	struct xfs_bmbt_irec    *map = &args->dc.map;
>  
>  	trace_xfs_attr_node_addname(args);
>  
> @@ -1034,8 +1071,17 @@ xfs_attr_node_addname(
>  	 * Fill in bucket of arguments/results/context to carry around.
>  	 */
>  	dp = args->dp;
> -	mp = dp->i_mount;
> -restart:
> +
> +	/* State machine switch */
> +	switch (args->dc.dc_state) {
> +	case XFS_DC_FOUND_NBLK:
> +		goto found_nblk;
> +	case XFS_DC_ALLOC_NODE:
> +		goto alloc_node;
> +	default:
> +		break;
> +	}
> +
>  	/*
>  	 * Search to see if name already exists, and get back a pointer
>  	 * to where it should go.
> @@ -1085,19 +1131,12 @@ xfs_attr_node_addname(
>  			error = xfs_attr3_leaf_to_node(args);
>  			if (error)
>  				goto out;
> -			error = xfs_defer_finish(&args->trans);
> -			if (error)
> -				goto out;
>  
>  			/*
> -			 * Commit the node conversion and start the next
> -			 * trans in the chain.
> +			 * Restart routine from the top.  No need to set  the
> +			 * state
>  			 */
> -			error = xfs_trans_roll_inode(&args->trans, dp);
> -			if (error)
> -				goto out;
> -
> -			goto restart;
> +			return -EAGAIN;
>  		}
>  
>  		/*
> @@ -1109,9 +1148,6 @@ xfs_attr_node_addname(
>  		error = xfs_da3_split(state);
>  		if (error)
>  			goto out;
> -		error = xfs_defer_finish(&args->trans);
> -		if (error)
> -			goto out;
>  	} else {
>  		/*
>  		 * Addition succeeded, update Btree hashvals.
> @@ -1126,13 +1162,9 @@ xfs_attr_node_addname(
>  	xfs_da_state_free(state);
>  	state = NULL;
>  
> -	/*
> -	 * Commit the leaf addition or btree split and start the next
> -	 * trans in the chain.
> -	 */
> -	error = xfs_trans_roll_inode(&args->trans, dp);
> -	if (error)
> -		goto out;
> +	args->dc.dc_state = XFS_DC_FOUND_NBLK;
> +	return -EAGAIN;
> +found_nblk:
>  
>  	/*
>  	 * If there was an out-of-line value, allocate the blocks we
> @@ -1141,104 +1173,57 @@ xfs_attr_node_addname(
>  	 * maximum size of a transaction and/or hit a deadlock.
>  	 */
>  	if (args->rmtblkno > 0) {
> -		error = xfs_attr_rmtval_set(args);
> -		if (error)
> -			return error;
> -	}
> +		/* Open coded xfs_attr_rmtval_set without trans handling */
> +		args->dc.lblkno = 0;
> +		args->dc.lfileoff = 0;
> +		args->dc.blkcnt = 0;
> +		args->rmtblkcnt = 0;
> +		args->rmtblkno = 0;
> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>  
> -	/*
> -	 * If this is an atomic rename operation, we must "flip" the
> -	 * incomplete flags on the "new" and "old" attribute/value pairs
> -	 * so that one disappears and one appears atomically.  Then we
> -	 * must remove the "old" attribute/value pair.
> -	 */
> -	if (args->op_flags & XFS_DA_OP_RENAME) {
> -		/*
> -		 * In a separate transaction, set the incomplete flag on the
> -		 * "old" attr and clear the incomplete flag on the "new" attr.
> -		 */
> -		error = xfs_attr3_leaf_flipflags(args);
> -		if (error)
> -			goto out;
> -		/*
> -		 * Commit the flag value change and start the next trans in
> -		 * series
> -		 */
> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
> +		error = xfs_attr_rmt_find_hole(args);
>  		if (error)
> -			goto out;
> +			return error;
>  
> +		args->dc.blkcnt = args->rmtblkcnt;
> +		args->dc.lblkno = args->rmtblkno;
>  		/*
> -		 * Dismantle the "old" attribute/value pair by removing
> -		 * a "remote" value (if it exists).
> +		 * Roll through the "value", allocating blocks on disk as
> +		 * required.
>  		 */
> -		args->index = args->index2;
> -		args->blkno = args->blkno2;
> -		args->rmtblkno = args->rmtblkno2;
> -		args->rmtblkcnt = args->rmtblkcnt2;
> -		args->rmtvaluelen = args->rmtvaluelen2;
> -		if (args->rmtblkno) {
> -			error = xfs_attr_rmtval_remove(args);
> +alloc_node:
> +		while (args->dc.blkcnt > 0) {
> +			nmap = 1;
> +			error = xfs_bmapi_write(args->trans, dp,
> +				(xfs_fileoff_t)args->dc.lblkno, args->dc.blkcnt,
> +				XFS_BMAPI_ATTRFORK, args->total, map, &nmap);
>  			if (error)
>  				return error;
> -		}
>  
> -		/*
> -		 * Re-find the "old" attribute entry after any split ops.
> -		 * The INCOMPLETE flag means that we will find the "old"
> -		 * attr, not the "new" one.
> -		 */
> -		args->name.type |= XFS_ATTR_INCOMPLETE;
> -		state = xfs_da_state_alloc();
> -		state->args = args;
> -		state->mp = mp;
> -		state->inleaf = 0;
> -		error = xfs_da3_node_lookup_int(state, &retval);
> -		if (error)
> -			goto out;
> +			ASSERT(nmap == 1);
> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
> +			       (map->br_startblock != HOLESTARTBLOCK));
>  
> -		/*
> -		 * Remove the name and update the hashvals in the tree.
> -		 */
> -		blk = &state->path.blk[ state->path.active-1 ];
> -		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
> -		error = xfs_attr3_leaf_remove(blk->bp, args);
> -		xfs_da3_fixhashpath(state, &state->path);
> +			/* roll attribute extent map forwards */
> +			args->dc.lblkno += map->br_blockcount;
> +			args->dc.blkcnt -= map->br_blockcount;
>  
> -		/*
> -		 * Check to see if the tree needs to be collapsed.
> -		 */
> -		if (retval && (state->path.active > 1)) {
> -			error = xfs_da3_join(state);
> -			if (error)
> -				goto out;
> -			error = xfs_defer_finish(&args->trans);
> -			if (error)
> -				goto out;
> +			args->dc.dc_state = XFS_DC_ALLOC_NODE;
> +			return -EAGAIN;
>  		}
>  
> -		/*
> -		 * Commit and start the next trans in the chain.
> -		 */
> -		error = xfs_trans_roll_inode(&args->trans, dp);
> +		error = xfs_attr_rmtval_set_value(args);
>  		if (error)
> -			goto out;
> +			return error;
> +	}
>  
> -	} else if (args->rmtblkno > 0) {
> +	if (args->rmtblkno > 0) {
>  		/*
>  		 * Added a "remote" value, just clear the incomplete flag.
>  		 */
>  		error = xfs_attr3_leaf_clearflag(args);
>  		if (error)
>  			goto out;
> -
> -		 /*
> -		  * Commit the flag value change and start the next trans in
> -		  * series.
> -		  */
> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
> -		if (error)
> -			goto out;
>  	}
>  	retval = error = 0;
>  
> diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
> index fb8bf5b..c710387 100644
> --- a/fs/xfs/libxfs/xfs_attr.h
> +++ b/fs/xfs/libxfs/xfs_attr.h
> @@ -149,6 +149,7 @@ int xfs_attr_get(struct xfs_inode *ip, struct xfs_name *name,
>  int xfs_attr_set(struct xfs_inode *dp, struct xfs_name *name,
>  		 unsigned char *value, int valuelen, int flags);
>  int xfs_attr_set_args(struct xfs_da_args *args);
> +int xfs_attr_set_later(struct xfs_da_args *args, struct xfs_buf **leaf_bp);
>  int xfs_attr_remove(struct xfs_inode *dp, struct xfs_name *name, int flags);
>  int xfs_has_attr(struct xfs_da_args *args);
>  int xfs_attr_remove_args(struct xfs_da_args *args);
> -- 
> 2.7.4
>

Allison Henderson Nov. 8, 2019, 9:52 p.m. UTC | #2

On 11/8/19 2:42 PM, Darrick J. Wong wrote:
> On Wed, Nov 06, 2019 at 06:28:01PM -0700, Allison Collins wrote:
>> This patch modifies the attr set routines to be delay ready.
>> This means they no longer roll or commit transactions, but instead
>> return -EAGAIN to have the calling routine roll and refresh the
>> transaction.  In this series, xfs_attr_set_args has become
>> xfs_attr_set_later, which uses a state machine to keep track
>> of where it was when EAGAIN was returned.  Part of
>> xfs_attr_leaf_addname has been factored out into a new helper
>> function xfs_attr_leaf_try_add to allow transaction cycling between
>> the two routines, and the flipflags logic has been removed since we
>> can simply cancel the transaction upon error.  xfs_attr_set_args
>> consists of a simple loop to refresh the transaction until the
>> operation is completed.
>>
>> Signed-off-by: Allison Collins <[e-mail]>
>> ---
>>   fs/xfs/libxfs/xfs_attr.c | 435 +++++++++++++++++++++++------------------------
>>   fs/xfs/libxfs/xfs_attr.h |   1 +
>>   2 files changed, 211 insertions(+), 225 deletions(-)
>>
>> diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
>> index 38d5c5c..97e5ae0 100644
>> --- a/fs/xfs/libxfs/xfs_attr.c
>> +++ b/fs/xfs/libxfs/xfs_attr.c
>> @@ -58,6 +58,7 @@ STATIC int xfs_attr_node_hasname(xfs_da_args_t *args,
>>   				 struct xfs_da_state **state);
>>   STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
>>   STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
>> +STATIC int xfs_attr_leaf_try_add(struct xfs_da_args *args, struct xfs_buf *bp);
>>   
>>   
>>   STATIC int
>> @@ -250,9 +251,79 @@ int
>>   xfs_attr_set_args(
>>   	struct xfs_da_args	*args)
>>   {
>> +	int			error = 0;
>> +	int			err2 = 0;
>> +	struct xfs_buf		*leaf_bp = NULL;
>> +
>> +	do {
>> +		error = xfs_attr_set_later(args, &leaf_bp);
>> +		if (error && error != -EAGAIN)
>> +			goto out;
>> +
>> +		xfs_trans_log_inode(args->trans, args->dp,
>> +				    XFS_ILOG_CORE | XFS_ILOG_ADATA);
> 
> Same question as the last patch about ADATA.
> 
>> +
>> +		err2 = xfs_trans_roll(&args->trans);
>> +		if (err2) {
>> +			error = err2;
>> +			goto out;
>> +		}
>> +
>> +		/* Rejoin inode and leaf if needed */
>> +		xfs_trans_ijoin(args->trans, args->dp, 0);
>> +		if (leaf_bp) {
>> +			xfs_trans_bjoin(args->trans, leaf_bp);
>> +			xfs_trans_bhold(args->trans, leaf_bp);
>> +		}
>> +
>> +	} while (error == -EAGAIN);
>> +
>> +out:
>> +	return error;
>> +}
>> +
>> +/*
>> + * Set the attribute specified in @args.
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + * returned.
>> + */
>> +int
>> +xfs_attr_set_later(
>> +	struct xfs_da_args	*args,
>> +	struct xfs_buf          **leaf_bp)
>> +{
>>   	struct xfs_inode	*dp = args->dp;
>> -	struct xfs_buf          *leaf_bp = NULL;
>> -	int			error, error2 = 0;;
>> +	int			error = 0;
>> +	int			sf_size;
>> +
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_SF_TO_LEAF:
>> +		goto sf_to_leaf;
>> +	case XFS_DC_ALLOC_LEAF:
>> +	case XFS_DC_FOUND_LBLK:
>> +		goto leaf;
>> +	case XFS_DC_FOUND_NBLK:
>> +	case XFS_DC_ALLOC_NODE:
>> +	case XFS_DC_LEAF_TO_NODE:
>> +		goto node;
>> +	default:
>> +		break;
>> +	}
>> +
>> +	/*
>> +	 * New inodes may not have an attribute fork yet. So set the attribute
>> +	 * fork appropriately
>> +	 */
>> +	if (XFS_IFORK_Q((args->dp)) == 0) {
>> +		sf_size = sizeof(struct xfs_attr_sf_hdr) +
>> +		     XFS_ATTR_SF_ENTSIZE_BYNAME(args->name.len, args->valuelen);
>> +		xfs_bmap_set_attrforkoff(args->dp, sf_size, NULL);
>> +		args->dp->i_afp = kmem_zone_zalloc(xfs_ifork_zone, 0);
>> +		args->dp->i_afp->if_flags = XFS_IFEXTENTS;
>> +	}
>>   
>>   	/*
>>   	 * If the attribute list is non-existent or a shortform list,
>> @@ -272,21 +343,14 @@ xfs_attr_set_args(
>>   		 * Try to add the attr to the attribute list in the inode.
>>   		 */
>>   		error = xfs_attr_try_sf_addname(dp, args);
>> -		if (error != -ENOSPC) {
>> -			if (dp->i_mount->m_flags & XFS_MOUNT_WSYNC)
>> -				xfs_trans_set_sync(args->trans);
> 
> Where does the xfs_trans_set_sync call go?  Do we not need it anymore?
> 
>> -			error2 = xfs_trans_commit(args->trans);
>> -			args->trans = NULL;
>> -			return error ? error : error2;
>> -		}
>> -
>> +		if (error != -ENOSPC)
>> +			return error;
>>   
>>   		/*
>>   		 * It won't fit in the shortform, transform to a leaf block.
>>   		 * GROT: another possible req'mt for a double-split btree op.
>>   		 */
>> -		error = xfs_attr_shortform_to_leaf(args, &leaf_bp);
>> +		error = xfs_attr_shortform_to_leaf(args, leaf_bp);
>>   		if (error)
>>   			return error;
>>   
>> @@ -294,43 +358,42 @@ xfs_attr_set_args(
>>   		 * Prevent the leaf buffer from being unlocked so that a
>>   		 * concurrent AIL push cannot grab the half-baked leaf
>>   		 * buffer and run into problems with the write verifier.
>> -		 * Once we're done rolling the transaction we can release
>> -		 * the hold and add the attr to the leaf.
>>   		 */
>> -		xfs_trans_bhold(args->trans, leaf_bp);
>> -		error = xfs_defer_finish(&args->trans);
>> -		xfs_trans_bhold_release(args->trans, leaf_bp);
>> -		if (error) {
>> -			xfs_trans_brelse(args->trans, leaf_bp);
>> -			return error;
>> -		}
>> +
>> +		xfs_trans_bhold(args->trans, *leaf_bp);
>> +		args->dc.dc_state = XFS_DC_SF_TO_LEAF;
>> +		return -EAGAIN;
>> +	}
>> +sf_to_leaf:
>> +
>> +	/*
>> +	 * After a shortform to leaf conversion, we need to hold the leaf and
>> +	 * cylce out the transaction.  When we get back, we need to release
> 
>             "cycle"
> 
>> +	 * the leaf.
>> +	 */
>> +	if (*leaf_bp != NULL) {
>> +		xfs_trans_brelse(args->trans, *leaf_bp);
>> +		*leaf_bp = NULL;
>>   	}
>>   
>>   	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
>> +		error = xfs_attr_leaf_try_add(args, *leaf_bp);
>> +		if (error == -ENOSPC)
>> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +		else if (error)
>> +			return error;
>> +		else
>> +			args->dc.dc_state = XFS_DC_FOUND_LBLK;
>> +		return -EAGAIN;
> 
> Please use a switch statement here...
> 
> switch (error) {
> case -ENOSPC:
> 	dc_state = LEAF_TO_NODE;
> 	return -EAGAIN;
> case 0:
> 	dc_state = FOUND_LBLK;
> 	return -EAGAIN;
> default:
> 	return error;
> }
> 
>> +leaf:
>>   		error = xfs_attr_leaf_addname(args);
>>   		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);
>> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +			return -EAGAIN;
>>   		}
>>   	} else {
>> +		args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +node:
>>   		error = xfs_attr_node_addname(args);
>>   	}
>>   	return error;
>> @@ -764,27 +827,26 @@ xfs_attr_leaf_try_add(
>>    *
>>    * 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).
>> + *
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + * returned.
>>    */
>>   STATIC int
>>   xfs_attr_leaf_addname(struct xfs_da_args	*args)
>>   {
>> -	int			error, forkoff;
>> -	struct xfs_buf		*bp = NULL;
>> +	int			error, nmap;
>>   	struct xfs_inode	*dp = args->dp;
>> +	struct xfs_bmbt_irec	*map = &args->dc.map;
>>   
>> -	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
>> -	 * later routines can manage their own transactions.
>> -	 */
>> -	error = xfs_trans_roll_inode(&args->trans, dp);
>> -	if (error)
>> -		return error;
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_ALLOC_LEAF:
>> +		goto alloc_leaf;
>> +	default:
>> +		break;
>> +	}
>>   
>>   	/*
>>   	 * If there was an out-of-line value, allocate the blocks we
>> @@ -793,90 +855,58 @@ xfs_attr_leaf_addname(struct xfs_da_args	*args)
>>   	 * maximum size of a transaction and/or hit a deadlock.
>>   	 */
>>   	if (args->rmtblkno > 0) {
>> -		error = xfs_attr_rmtval_set(args);
>> -		if (error)
>> -			return error;
>> -	}
>>   
>> -	/*
>> -	 * If this is an atomic rename operation, we must "flip" the
>> -	 * incomplete flags on the "new" and "old" attribute/value pairs
>> -	 * so that one disappears and one appears atomically.  Then we
>> -	 * must remove the "old" attribute/value pair.
>> -	 */
>> -	if (args->op_flags & XFS_DA_OP_RENAME) {
>> -		/*
>> -		 * In a separate transaction, set the incomplete flag on the
>> -		 * "old" attr and clear the incomplete flag on the "new" attr.
>> -		 */
>> -		error = xfs_attr3_leaf_flipflags(args);
>> -		if (error)
>> -			return error;
>> -		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> -		if (error)
>> -			return error;
>> +		/* Open coded xfs_attr_rmtval_set without trans handling */
> 
> Can't we just fix that instead of reproducing it here?
> 
> Especially because I think I see it twice in this patch?
> 
> Also, do you have a git tree handy?  I /think/ I see how this works but
> oh man is it difficult to see that from patches alone.

Sure, I've pushed the branches here.  These ones go all the way through 
delayed attrs and ppts, but I usually just checkout the commit of the 
subset I need to work on.


https://github.com/allisonhenderson/xfs_work/tree/pptrs_restart38

https://github.com/allisonhenderson/xfs_work/tree/parent_pointers_xfsprogs_restart30

> 
> --D
> 
>>   
>> -		/*
>> -		 * Dismantle the "old" attribute/value pair by removing
>> -		 * a "remote" value (if it exists).
>> -		 */
>> -		args->index = args->index2;
>> -		args->blkno = args->blkno2;
>> -		args->rmtblkno = args->rmtblkno2;
>> -		args->rmtblkcnt = args->rmtblkcnt2;
>> -		args->rmtvaluelen = args->rmtvaluelen2;
>> -		if (args->rmtblkno) {
>> -			error = xfs_attr_rmtval_remove(args);
>> -			if (error)
>> -				return error;
>> -		}
>> +		args->dc.lfileoff = 0;
>> +		args->dc.lblkno = 0;
>> +		args->dc.blkcnt = 0;
>> +		args->rmtblkcnt = 0;
>> +		args->rmtblkno = 0;
>> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>>   
>> -		/*
>> -		 * Read in the block containing the "old" attr, then
>> -		 * remove the "old" attr from that block (neat, huh!)
>> -		 */
>> -		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno,
>> -					   XFS_DABUF_MAP_NOMAPPING, &bp);
>> +		error = xfs_attr_rmt_find_hole(args);
>>   		if (error)
>>   			return error;
>>   
>> -		xfs_attr3_leaf_remove(bp, args);
>> +		args->dc.blkcnt = args->rmtblkcnt;
>> +		args->dc.lblkno = args->rmtblkno;
>>   
>>   		/*
>> -		 * If the result is small enough, shrink it all into the inode.
>> +		 * Roll through the "value", allocating blocks on disk as
>> +		 * required.
>>   		 */
>> -		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
>> -			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
>> -			/* bp is gone due to xfs_da_shrink_inode */
>> -			if (error)
>> -				return error;
>> -			error = xfs_defer_finish(&args->trans);
>> +alloc_leaf:
>> +		while (args->dc.blkcnt > 0) {
>> +			nmap = 1;
>> +			error = xfs_bmapi_write(args->trans, dp,
>> +				  (xfs_fileoff_t)args->dc.lblkno,
>> +				  args->dc.blkcnt, XFS_BMAPI_ATTRFORK,
>> +				  args->total, map, &nmap);
>>   			if (error)
>>   				return error;
>> -		}
>> +			ASSERT(nmap == 1);
>> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
>> +			       (map->br_startblock != HOLESTARTBLOCK));
>>   
>> -		/*
>> -		 * Commit the remove and start the next trans in series.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, dp);
>> +			/* roll attribute extent map forwards */
>> +			args->dc.lblkno += map->br_blockcount;
>> +			args->dc.blkcnt -= map->br_blockcount;
>>   
>> -	} else if (args->rmtblkno > 0) {
>> -		/*
>> -		 * Added a "remote" value, just clear the incomplete flag.
>> -		 */
>> -		error = xfs_attr3_leaf_clearflag(args);
>> +			args->dc.dc_state = XFS_DC_ALLOC_LEAF;
>> +			return -EAGAIN;
>> +		}
>> +
>> +		error = xfs_attr_rmtval_set_value(args);
>>   		if (error)
>>   			return error;
>> +	}
>>   
>> +	if (args->rmtblkno > 0) {
>>   		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series.
>> +		 * Added a "remote" value, just clear the incomplete flag.
>>   		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> +		error = xfs_attr3_leaf_clearflag(args);
>>   	}
>>   	return error;
>>   }
>> @@ -1017,16 +1047,23 @@ xfs_attr_node_hasname(
>>    *
>>    * "Remote" attribute values confuse the issue and atomic rename operations
>>    * add a whole extra layer of confusion on top of that.
>> + *
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + *returned.
>>    */
>>   STATIC int
>>   xfs_attr_node_addname(
>>   	struct xfs_da_args	*args)
>>   {
>> -	struct xfs_da_state	*state;
>> +	struct xfs_da_state	*state = NULL;
>>   	struct xfs_da_state_blk	*blk;
>>   	struct xfs_inode	*dp;
>> -	struct xfs_mount	*mp;
>> -	int			retval, error;
>> +	int			retval = 0;
>> +	int			error = 0;
>> +	int			nmap;
>> +	struct xfs_bmbt_irec    *map = &args->dc.map;
>>   
>>   	trace_xfs_attr_node_addname(args);
>>   
>> @@ -1034,8 +1071,17 @@ xfs_attr_node_addname(
>>   	 * Fill in bucket of arguments/results/context to carry around.
>>   	 */
>>   	dp = args->dp;
>> -	mp = dp->i_mount;
>> -restart:
>> +
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_FOUND_NBLK:
>> +		goto found_nblk;
>> +	case XFS_DC_ALLOC_NODE:
>> +		goto alloc_node;
>> +	default:
>> +		break;
>> +	}
>> +
>>   	/*
>>   	 * Search to see if name already exists, and get back a pointer
>>   	 * to where it should go.
>> @@ -1085,19 +1131,12 @@ xfs_attr_node_addname(
>>   			error = xfs_attr3_leaf_to_node(args);
>>   			if (error)
>>   				goto out;
>> -			error = xfs_defer_finish(&args->trans);
>> -			if (error)
>> -				goto out;
>>   
>>   			/*
>> -			 * Commit the node conversion and start the next
>> -			 * trans in the chain.
>> +			 * Restart routine from the top.  No need to set  the
>> +			 * state
>>   			 */
>> -			error = xfs_trans_roll_inode(&args->trans, dp);
>> -			if (error)
>> -				goto out;
>> -
>> -			goto restart;
>> +			return -EAGAIN;
>>   		}
>>   
>>   		/*
>> @@ -1109,9 +1148,6 @@ xfs_attr_node_addname(
>>   		error = xfs_da3_split(state);
>>   		if (error)
>>   			goto out;
>> -		error = xfs_defer_finish(&args->trans);
>> -		if (error)
>> -			goto out;
>>   	} else {
>>   		/*
>>   		 * Addition succeeded, update Btree hashvals.
>> @@ -1126,13 +1162,9 @@ xfs_attr_node_addname(
>>   	xfs_da_state_free(state);
>>   	state = NULL;
>>   
>> -	/*
>> -	 * Commit the leaf addition or btree split and start the next
>> -	 * trans in the chain.
>> -	 */
>> -	error = xfs_trans_roll_inode(&args->trans, dp);
>> -	if (error)
>> -		goto out;
>> +	args->dc.dc_state = XFS_DC_FOUND_NBLK;
>> +	return -EAGAIN;
>> +found_nblk:
>>   
>>   	/*
>>   	 * If there was an out-of-line value, allocate the blocks we
>> @@ -1141,104 +1173,57 @@ xfs_attr_node_addname(
>>   	 * maximum size of a transaction and/or hit a deadlock.
>>   	 */
>>   	if (args->rmtblkno > 0) {
>> -		error = xfs_attr_rmtval_set(args);
>> -		if (error)
>> -			return error;
>> -	}
>> +		/* Open coded xfs_attr_rmtval_set without trans handling */
>> +		args->dc.lblkno = 0;
>> +		args->dc.lfileoff = 0;
>> +		args->dc.blkcnt = 0;
>> +		args->rmtblkcnt = 0;
>> +		args->rmtblkno = 0;
>> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>>   
>> -	/*
>> -	 * If this is an atomic rename operation, we must "flip" the
>> -	 * incomplete flags on the "new" and "old" attribute/value pairs
>> -	 * so that one disappears and one appears atomically.  Then we
>> -	 * must remove the "old" attribute/value pair.
>> -	 */
>> -	if (args->op_flags & XFS_DA_OP_RENAME) {
>> -		/*
>> -		 * In a separate transaction, set the incomplete flag on the
>> -		 * "old" attr and clear the incomplete flag on the "new" attr.
>> -		 */
>> -		error = xfs_attr3_leaf_flipflags(args);
>> -		if (error)
>> -			goto out;
>> -		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> +		error = xfs_attr_rmt_find_hole(args);
>>   		if (error)
>> -			goto out;
>> +			return error;
>>   
>> +		args->dc.blkcnt = args->rmtblkcnt;
>> +		args->dc.lblkno = args->rmtblkno;
>>   		/*
>> -		 * Dismantle the "old" attribute/value pair by removing
>> -		 * a "remote" value (if it exists).
>> +		 * Roll through the "value", allocating blocks on disk as
>> +		 * required.
>>   		 */
>> -		args->index = args->index2;
>> -		args->blkno = args->blkno2;
>> -		args->rmtblkno = args->rmtblkno2;
>> -		args->rmtblkcnt = args->rmtblkcnt2;
>> -		args->rmtvaluelen = args->rmtvaluelen2;
>> -		if (args->rmtblkno) {
>> -			error = xfs_attr_rmtval_remove(args);
>> +alloc_node:
>> +		while (args->dc.blkcnt > 0) {
>> +			nmap = 1;
>> +			error = xfs_bmapi_write(args->trans, dp,
>> +				(xfs_fileoff_t)args->dc.lblkno, args->dc.blkcnt,
>> +				XFS_BMAPI_ATTRFORK, args->total, map, &nmap);
>>   			if (error)
>>   				return error;
>> -		}
>>   
>> -		/*
>> -		 * Re-find the "old" attribute entry after any split ops.
>> -		 * The INCOMPLETE flag means that we will find the "old"
>> -		 * attr, not the "new" one.
>> -		 */
>> -		args->name.type |= XFS_ATTR_INCOMPLETE;
>> -		state = xfs_da_state_alloc();
>> -		state->args = args;
>> -		state->mp = mp;
>> -		state->inleaf = 0;
>> -		error = xfs_da3_node_lookup_int(state, &retval);
>> -		if (error)
>> -			goto out;
>> +			ASSERT(nmap == 1);
>> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
>> +			       (map->br_startblock != HOLESTARTBLOCK));
>>   
>> -		/*
>> -		 * Remove the name and update the hashvals in the tree.
>> -		 */
>> -		blk = &state->path.blk[ state->path.active-1 ];
>> -		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
>> -		error = xfs_attr3_leaf_remove(blk->bp, args);
>> -		xfs_da3_fixhashpath(state, &state->path);
>> +			/* roll attribute extent map forwards */
>> +			args->dc.lblkno += map->br_blockcount;
>> +			args->dc.blkcnt -= map->br_blockcount;
>>   
>> -		/*
>> -		 * Check to see if the tree needs to be collapsed.
>> -		 */
>> -		if (retval && (state->path.active > 1)) {
>> -			error = xfs_da3_join(state);
>> -			if (error)
>> -				goto out;
>> -			error = xfs_defer_finish(&args->trans);
>> -			if (error)
>> -				goto out;
>> +			args->dc.dc_state = XFS_DC_ALLOC_NODE;
>> +			return -EAGAIN;
>>   		}
>>   
>> -		/*
>> -		 * Commit and start the next trans in the chain.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, dp);
>> +		error = xfs_attr_rmtval_set_value(args);
>>   		if (error)
>> -			goto out;
>> +			return error;
>> +	}
>>   
>> -	} else if (args->rmtblkno > 0) {
>> +	if (args->rmtblkno > 0) {
>>   		/*
>>   		 * Added a "remote" value, just clear the incomplete flag.
>>   		 */
>>   		error = xfs_attr3_leaf_clearflag(args);
>>   		if (error)
>>   			goto out;
>> -
>> -		 /*
>> -		  * Commit the flag value change and start the next trans in
>> -		  * series.
>> -		  */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> -		if (error)
>> -			goto out;
>>   	}
>>   	retval = error = 0;
>>   
>> diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
>> index fb8bf5b..c710387 100644
>> --- a/fs/xfs/libxfs/xfs_attr.h
>> +++ b/fs/xfs/libxfs/xfs_attr.h
>> @@ -149,6 +149,7 @@ int xfs_attr_get(struct xfs_inode *ip, struct xfs_name *name,
>>   int xfs_attr_set(struct xfs_inode *dp, struct xfs_name *name,
>>   		 unsigned char *value, int valuelen, int flags);
>>   int xfs_attr_set_args(struct xfs_da_args *args);
>> +int xfs_attr_set_later(struct xfs_da_args *args, struct xfs_buf **leaf_bp);
>>   int xfs_attr_remove(struct xfs_inode *dp, struct xfs_name *name, int flags);
>>   int xfs_has_attr(struct xfs_da_args *args);
>>   int xfs_attr_remove_args(struct xfs_da_args *args);
>> -- 
>> 2.7.4
>>

Allison Henderson Nov. 9, 2019, 4:07 a.m. UTC | #3

On 11/8/19 2:42 PM, Darrick J. Wong wrote:
> On Wed, Nov 06, 2019 at 06:28:01PM -0700, Allison Collins wrote:
>> This patch modifies the attr set routines to be delay ready.
>> This means they no longer roll or commit transactions, but instead
>> return -EAGAIN to have the calling routine roll and refresh the
>> transaction.  In this series, xfs_attr_set_args has become
>> xfs_attr_set_later, which uses a state machine to keep track
>> of where it was when EAGAIN was returned.  Part of
>> xfs_attr_leaf_addname has been factored out into a new helper
>> function xfs_attr_leaf_try_add to allow transaction cycling between
>> the two routines, and the flipflags logic has been removed since we
>> can simply cancel the transaction upon error.  xfs_attr_set_args
>> consists of a simple loop to refresh the transaction until the
>> operation is completed.
>>
>> Signed-off-by: Allison Collins <[e-mail]>
>> ---
>>   fs/xfs/libxfs/xfs_attr.c | 435 +++++++++++++++++++++++------------------------
>>   fs/xfs/libxfs/xfs_attr.h |   1 +
>>   2 files changed, 211 insertions(+), 225 deletions(-)
>>
>> diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
>> index 38d5c5c..97e5ae0 100644
>> --- a/fs/xfs/libxfs/xfs_attr.c
>> +++ b/fs/xfs/libxfs/xfs_attr.c
>> @@ -58,6 +58,7 @@ STATIC int xfs_attr_node_hasname(xfs_da_args_t *args,
>>   				 struct xfs_da_state **state);
>>   STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
>>   STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
>> +STATIC int xfs_attr_leaf_try_add(struct xfs_da_args *args, struct xfs_buf *bp);
>>   
>>   
>>   STATIC int
>> @@ -250,9 +251,79 @@ int
>>   xfs_attr_set_args(
>>   	struct xfs_da_args	*args)
>>   {
>> +	int			error = 0;
>> +	int			err2 = 0;
>> +	struct xfs_buf		*leaf_bp = NULL;
>> +
>> +	do {
>> +		error = xfs_attr_set_later(args, &leaf_bp);
>> +		if (error && error != -EAGAIN)
>> +			goto out;
>> +
>> +		xfs_trans_log_inode(args->trans, args->dp,
>> +				    XFS_ILOG_CORE | XFS_ILOG_ADATA);
> 
> Same question as the last patch about ADATA.
> 
>> +
>> +		err2 = xfs_trans_roll(&args->trans);
>> +		if (err2) {
>> +			error = err2;
>> +			goto out;
>> +		}
>> +
>> +		/* Rejoin inode and leaf if needed */
>> +		xfs_trans_ijoin(args->trans, args->dp, 0);
>> +		if (leaf_bp) {
>> +			xfs_trans_bjoin(args->trans, leaf_bp);
>> +			xfs_trans_bhold(args->trans, leaf_bp);
>> +		}
>> +
>> +	} while (error == -EAGAIN);
>> +
>> +out:
>> +	return error;
>> +}
>> +
>> +/*
>> + * Set the attribute specified in @args.
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + * returned.
>> + */
>> +int
>> +xfs_attr_set_later(
>> +	struct xfs_da_args	*args,
>> +	struct xfs_buf          **leaf_bp)
>> +{
>>   	struct xfs_inode	*dp = args->dp;
>> -	struct xfs_buf          *leaf_bp = NULL;
>> -	int			error, error2 = 0;;
>> +	int			error = 0;
>> +	int			sf_size;
>> +
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_SF_TO_LEAF:
>> +		goto sf_to_leaf;
>> +	case XFS_DC_ALLOC_LEAF:
>> +	case XFS_DC_FOUND_LBLK:
>> +		goto leaf;
>> +	case XFS_DC_FOUND_NBLK:
>> +	case XFS_DC_ALLOC_NODE:
>> +	case XFS_DC_LEAF_TO_NODE:
>> +		goto node;
>> +	default:
>> +		break;
>> +	}
>> +
>> +	/*
>> +	 * New inodes may not have an attribute fork yet. So set the attribute
>> +	 * fork appropriately
>> +	 */
>> +	if (XFS_IFORK_Q((args->dp)) == 0) {
>> +		sf_size = sizeof(struct xfs_attr_sf_hdr) +
>> +		     XFS_ATTR_SF_ENTSIZE_BYNAME(args->name.len, args->valuelen);
>> +		xfs_bmap_set_attrforkoff(args->dp, sf_size, NULL);
>> +		args->dp->i_afp = kmem_zone_zalloc(xfs_ifork_zone, 0);
>> +		args->dp->i_afp->if_flags = XFS_IFEXTENTS;
>> +	}
>>   
>>   	/*
>>   	 * If the attribute list is non-existent or a shortform list,
>> @@ -272,21 +343,14 @@ xfs_attr_set_args(
>>   		 * Try to add the attr to the attribute list in the inode.
>>   		 */
>>   		error = xfs_attr_try_sf_addname(dp, args);
>> -		if (error != -ENOSPC) {
>> -			if (dp->i_mount->m_flags & XFS_MOUNT_WSYNC)
>> -				xfs_trans_set_sync(args->trans);
> 
> Where does the xfs_trans_set_sync call go?  Do we not need it anymore?
I may have lost it somewhere in all the re-refactoring.  I will add it 
back.  Sorry!

> 
>> -			error2 = xfs_trans_commit(args->trans);
>> -			args->trans = NULL;
>> -			return error ? error : error2;
>> -		}
>> -
>> +		if (error != -ENOSPC)
>> +			return error;
>>   
>>   		/*
>>   		 * It won't fit in the shortform, transform to a leaf block.
>>   		 * GROT: another possible req'mt for a double-split btree op.
>>   		 */
>> -		error = xfs_attr_shortform_to_leaf(args, &leaf_bp);
>> +		error = xfs_attr_shortform_to_leaf(args, leaf_bp);
>>   		if (error)
>>   			return error;
>>   
>> @@ -294,43 +358,42 @@ xfs_attr_set_args(
>>   		 * Prevent the leaf buffer from being unlocked so that a
>>   		 * concurrent AIL push cannot grab the half-baked leaf
>>   		 * buffer and run into problems with the write verifier.
>> -		 * Once we're done rolling the transaction we can release
>> -		 * the hold and add the attr to the leaf.
>>   		 */
>> -		xfs_trans_bhold(args->trans, leaf_bp);
>> -		error = xfs_defer_finish(&args->trans);
>> -		xfs_trans_bhold_release(args->trans, leaf_bp);
>> -		if (error) {
>> -			xfs_trans_brelse(args->trans, leaf_bp);
>> -			return error;
>> -		}
>> +
>> +		xfs_trans_bhold(args->trans, *leaf_bp);
>> +		args->dc.dc_state = XFS_DC_SF_TO_LEAF;
>> +		return -EAGAIN;
>> +	}
>> +sf_to_leaf:
>> +
>> +	/*
>> +	 * After a shortform to leaf conversion, we need to hold the leaf and
>> +	 * cylce out the transaction.  When we get back, we need to release
> 
>             "cycle"
> 
>> +	 * the leaf.
>> +	 */
>> +	if (*leaf_bp != NULL) {
>> +		xfs_trans_brelse(args->trans, *leaf_bp);
>> +		*leaf_bp = NULL;
>>   	}
>>   
>>   	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
>> +		error = xfs_attr_leaf_try_add(args, *leaf_bp);
>> +		if (error == -ENOSPC)
>> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +		else if (error)
>> +			return error;
>> +		else
>> +			args->dc.dc_state = XFS_DC_FOUND_LBLK;
>> +		return -EAGAIN;
> 
> Please use a switch statement here...
> 
> switch (error) {
> case -ENOSPC:
> 	dc_state = LEAF_TO_NODE;
> 	return -EAGAIN;
> case 0:
> 	dc_state = FOUND_LBLK;
> 	return -EAGAIN;
> default:
> 	return error;
> }
Sure, will do

> 
>> +leaf:
>>   		error = xfs_attr_leaf_addname(args);
>>   		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);
>> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +			return -EAGAIN;
>>   		}
>>   	} else {
>> +		args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +node:
>>   		error = xfs_attr_node_addname(args);
>>   	}
>>   	return error;
>> @@ -764,27 +827,26 @@ xfs_attr_leaf_try_add(
>>    *
>>    * 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).
>> + *
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + * returned.
>>    */
>>   STATIC int
>>   xfs_attr_leaf_addname(struct xfs_da_args	*args)
>>   {
>> -	int			error, forkoff;
>> -	struct xfs_buf		*bp = NULL;
>> +	int			error, nmap;
>>   	struct xfs_inode	*dp = args->dp;
>> +	struct xfs_bmbt_irec	*map = &args->dc.map;
>>   
>> -	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
>> -	 * later routines can manage their own transactions.
>> -	 */
>> -	error = xfs_trans_roll_inode(&args->trans, dp);
>> -	if (error)
>> -		return error;
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_ALLOC_LEAF:
>> +		goto alloc_leaf;
>> +	default:
>> +		break;
>> +	}
>>   
>>   	/*
>>   	 * If there was an out-of-line value, allocate the blocks we
>> @@ -793,90 +855,58 @@ xfs_attr_leaf_addname(struct xfs_da_args	*args)
>>   	 * maximum size of a transaction and/or hit a deadlock.
>>   	 */
>>   	if (args->rmtblkno > 0) {
>> -		error = xfs_attr_rmtval_set(args);
>> -		if (error)
>> -			return error;
>> -	}
>>   
>> -	/*
>> -	 * If this is an atomic rename operation, we must "flip" the
>> -	 * incomplete flags on the "new" and "old" attribute/value pairs
>> -	 * so that one disappears and one appears atomically.  Then we
>> -	 * must remove the "old" attribute/value pair.
>> -	 */
>> -	if (args->op_flags & XFS_DA_OP_RENAME) {
>> -		/*
>> -		 * In a separate transaction, set the incomplete flag on the
>> -		 * "old" attr and clear the incomplete flag on the "new" attr.
>> -		 */
>> -		error = xfs_attr3_leaf_flipflags(args);
>> -		if (error)
>> -			return error;
>> -		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> -		if (error)
>> -			return error;
>> +		/* Open coded xfs_attr_rmtval_set without trans handling */
> 
> Can't we just fix that instead of reproducing it here?
By fix, you mean to plumb in the state machine into xfs_attr_rmtval_set? 
  We can, it will just need a few more states I think.  I guess I was 
just trying to avoid that as much as possible by surfacing all the 
places that we need to return EAGAIN.  I think the idea Brian was aiming 
for in some of the earlier reviews was to refactor things such that we 
have the helper calls in between the EAGAINS so that the state machine 
hops are easier to see rather than being buried here and there in sub 
routines.

But spots like this are harder to do that with because of the while 
loop.  The loop has to come up, and then the transaction code 
disappears, but then there's not much left to modularize with in the 
loop body. It is hard to see though patch diff though, hopefully those 
links helped?

Allison

> 
> Especially because I think I see it twice in this patch?
> 
> Also, do you have a git tree handy?  I /think/ I see how this works but
> oh man is it difficult to see that from patches alone.
> 
> --D
> 
>>   
>> -		/*
>> -		 * Dismantle the "old" attribute/value pair by removing
>> -		 * a "remote" value (if it exists).
>> -		 */
>> -		args->index = args->index2;
>> -		args->blkno = args->blkno2;
>> -		args->rmtblkno = args->rmtblkno2;
>> -		args->rmtblkcnt = args->rmtblkcnt2;
>> -		args->rmtvaluelen = args->rmtvaluelen2;
>> -		if (args->rmtblkno) {
>> -			error = xfs_attr_rmtval_remove(args);
>> -			if (error)
>> -				return error;
>> -		}
>> +		args->dc.lfileoff = 0;
>> +		args->dc.lblkno = 0;
>> +		args->dc.blkcnt = 0;
>> +		args->rmtblkcnt = 0;
>> +		args->rmtblkno = 0;
>> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>>   
>> -		/*
>> -		 * Read in the block containing the "old" attr, then
>> -		 * remove the "old" attr from that block (neat, huh!)
>> -		 */
>> -		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno,
>> -					   XFS_DABUF_MAP_NOMAPPING, &bp);
>> +		error = xfs_attr_rmt_find_hole(args);
>>   		if (error)
>>   			return error;
>>   
>> -		xfs_attr3_leaf_remove(bp, args);
>> +		args->dc.blkcnt = args->rmtblkcnt;
>> +		args->dc.lblkno = args->rmtblkno;
>>   
>>   		/*
>> -		 * If the result is small enough, shrink it all into the inode.
>> +		 * Roll through the "value", allocating blocks on disk as
>> +		 * required.
>>   		 */
>> -		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
>> -			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
>> -			/* bp is gone due to xfs_da_shrink_inode */
>> -			if (error)
>> -				return error;
>> -			error = xfs_defer_finish(&args->trans);
>> +alloc_leaf:
>> +		while (args->dc.blkcnt > 0) {
>> +			nmap = 1;
>> +			error = xfs_bmapi_write(args->trans, dp,
>> +				  (xfs_fileoff_t)args->dc.lblkno,
>> +				  args->dc.blkcnt, XFS_BMAPI_ATTRFORK,
>> +				  args->total, map, &nmap);
>>   			if (error)
>>   				return error;
>> -		}
>> +			ASSERT(nmap == 1);
>> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
>> +			       (map->br_startblock != HOLESTARTBLOCK));
>>   
>> -		/*
>> -		 * Commit the remove and start the next trans in series.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, dp);
>> +			/* roll attribute extent map forwards */
>> +			args->dc.lblkno += map->br_blockcount;
>> +			args->dc.blkcnt -= map->br_blockcount;
>>   
>> -	} else if (args->rmtblkno > 0) {
>> -		/*
>> -		 * Added a "remote" value, just clear the incomplete flag.
>> -		 */
>> -		error = xfs_attr3_leaf_clearflag(args);
>> +			args->dc.dc_state = XFS_DC_ALLOC_LEAF;
>> +			return -EAGAIN;
>> +		}
>> +
>> +		error = xfs_attr_rmtval_set_value(args);
>>   		if (error)
>>   			return error;
>> +	}
>>   
>> +	if (args->rmtblkno > 0) {
>>   		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series.
>> +		 * Added a "remote" value, just clear the incomplete flag.
>>   		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> +		error = xfs_attr3_leaf_clearflag(args);
>>   	}
>>   	return error;
>>   }
>> @@ -1017,16 +1047,23 @@ xfs_attr_node_hasname(
>>    *
>>    * "Remote" attribute values confuse the issue and atomic rename operations
>>    * add a whole extra layer of confusion on top of that.
>> + *
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + *returned.
>>    */
>>   STATIC int
>>   xfs_attr_node_addname(
>>   	struct xfs_da_args	*args)
>>   {
>> -	struct xfs_da_state	*state;
>> +	struct xfs_da_state	*state = NULL;
>>   	struct xfs_da_state_blk	*blk;
>>   	struct xfs_inode	*dp;
>> -	struct xfs_mount	*mp;
>> -	int			retval, error;
>> +	int			retval = 0;
>> +	int			error = 0;
>> +	int			nmap;
>> +	struct xfs_bmbt_irec    *map = &args->dc.map;
>>   
>>   	trace_xfs_attr_node_addname(args);
>>   
>> @@ -1034,8 +1071,17 @@ xfs_attr_node_addname(
>>   	 * Fill in bucket of arguments/results/context to carry around.
>>   	 */
>>   	dp = args->dp;
>> -	mp = dp->i_mount;
>> -restart:
>> +
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_FOUND_NBLK:
>> +		goto found_nblk;
>> +	case XFS_DC_ALLOC_NODE:
>> +		goto alloc_node;
>> +	default:
>> +		break;
>> +	}
>> +
>>   	/*
>>   	 * Search to see if name already exists, and get back a pointer
>>   	 * to where it should go.
>> @@ -1085,19 +1131,12 @@ xfs_attr_node_addname(
>>   			error = xfs_attr3_leaf_to_node(args);
>>   			if (error)
>>   				goto out;
>> -			error = xfs_defer_finish(&args->trans);
>> -			if (error)
>> -				goto out;
>>   
>>   			/*
>> -			 * Commit the node conversion and start the next
>> -			 * trans in the chain.
>> +			 * Restart routine from the top.  No need to set  the
>> +			 * state
>>   			 */
>> -			error = xfs_trans_roll_inode(&args->trans, dp);
>> -			if (error)
>> -				goto out;
>> -
>> -			goto restart;
>> +			return -EAGAIN;
>>   		}
>>   
>>   		/*
>> @@ -1109,9 +1148,6 @@ xfs_attr_node_addname(
>>   		error = xfs_da3_split(state);
>>   		if (error)
>>   			goto out;
>> -		error = xfs_defer_finish(&args->trans);
>> -		if (error)
>> -			goto out;
>>   	} else {
>>   		/*
>>   		 * Addition succeeded, update Btree hashvals.
>> @@ -1126,13 +1162,9 @@ xfs_attr_node_addname(
>>   	xfs_da_state_free(state);
>>   	state = NULL;
>>   
>> -	/*
>> -	 * Commit the leaf addition or btree split and start the next
>> -	 * trans in the chain.
>> -	 */
>> -	error = xfs_trans_roll_inode(&args->trans, dp);
>> -	if (error)
>> -		goto out;
>> +	args->dc.dc_state = XFS_DC_FOUND_NBLK;
>> +	return -EAGAIN;
>> +found_nblk:
>>   
>>   	/*
>>   	 * If there was an out-of-line value, allocate the blocks we
>> @@ -1141,104 +1173,57 @@ xfs_attr_node_addname(
>>   	 * maximum size of a transaction and/or hit a deadlock.
>>   	 */
>>   	if (args->rmtblkno > 0) {
>> -		error = xfs_attr_rmtval_set(args);
>> -		if (error)
>> -			return error;
>> -	}
>> +		/* Open coded xfs_attr_rmtval_set without trans handling */
>> +		args->dc.lblkno = 0;
>> +		args->dc.lfileoff = 0;
>> +		args->dc.blkcnt = 0;
>> +		args->rmtblkcnt = 0;
>> +		args->rmtblkno = 0;
>> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>>   
>> -	/*
>> -	 * If this is an atomic rename operation, we must "flip" the
>> -	 * incomplete flags on the "new" and "old" attribute/value pairs
>> -	 * so that one disappears and one appears atomically.  Then we
>> -	 * must remove the "old" attribute/value pair.
>> -	 */
>> -	if (args->op_flags & XFS_DA_OP_RENAME) {
>> -		/*
>> -		 * In a separate transaction, set the incomplete flag on the
>> -		 * "old" attr and clear the incomplete flag on the "new" attr.
>> -		 */
>> -		error = xfs_attr3_leaf_flipflags(args);
>> -		if (error)
>> -			goto out;
>> -		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> +		error = xfs_attr_rmt_find_hole(args);
>>   		if (error)
>> -			goto out;
>> +			return error;
>>   
>> +		args->dc.blkcnt = args->rmtblkcnt;
>> +		args->dc.lblkno = args->rmtblkno;
>>   		/*
>> -		 * Dismantle the "old" attribute/value pair by removing
>> -		 * a "remote" value (if it exists).
>> +		 * Roll through the "value", allocating blocks on disk as
>> +		 * required.
>>   		 */
>> -		args->index = args->index2;
>> -		args->blkno = args->blkno2;
>> -		args->rmtblkno = args->rmtblkno2;
>> -		args->rmtblkcnt = args->rmtblkcnt2;
>> -		args->rmtvaluelen = args->rmtvaluelen2;
>> -		if (args->rmtblkno) {
>> -			error = xfs_attr_rmtval_remove(args);
>> +alloc_node:
>> +		while (args->dc.blkcnt > 0) {
>> +			nmap = 1;
>> +			error = xfs_bmapi_write(args->trans, dp,
>> +				(xfs_fileoff_t)args->dc.lblkno, args->dc.blkcnt,
>> +				XFS_BMAPI_ATTRFORK, args->total, map, &nmap);
>>   			if (error)
>>   				return error;
>> -		}
>>   
>> -		/*
>> -		 * Re-find the "old" attribute entry after any split ops.
>> -		 * The INCOMPLETE flag means that we will find the "old"
>> -		 * attr, not the "new" one.
>> -		 */
>> -		args->name.type |= XFS_ATTR_INCOMPLETE;
>> -		state = xfs_da_state_alloc();
>> -		state->args = args;
>> -		state->mp = mp;
>> -		state->inleaf = 0;
>> -		error = xfs_da3_node_lookup_int(state, &retval);
>> -		if (error)
>> -			goto out;
>> +			ASSERT(nmap == 1);
>> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
>> +			       (map->br_startblock != HOLESTARTBLOCK));
>>   
>> -		/*
>> -		 * Remove the name and update the hashvals in the tree.
>> -		 */
>> -		blk = &state->path.blk[ state->path.active-1 ];
>> -		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
>> -		error = xfs_attr3_leaf_remove(blk->bp, args);
>> -		xfs_da3_fixhashpath(state, &state->path);
>> +			/* roll attribute extent map forwards */
>> +			args->dc.lblkno += map->br_blockcount;
>> +			args->dc.blkcnt -= map->br_blockcount;
>>   
>> -		/*
>> -		 * Check to see if the tree needs to be collapsed.
>> -		 */
>> -		if (retval && (state->path.active > 1)) {
>> -			error = xfs_da3_join(state);
>> -			if (error)
>> -				goto out;
>> -			error = xfs_defer_finish(&args->trans);
>> -			if (error)
>> -				goto out;
>> +			args->dc.dc_state = XFS_DC_ALLOC_NODE;
>> +			return -EAGAIN;
>>   		}
>>   
>> -		/*
>> -		 * Commit and start the next trans in the chain.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, dp);
>> +		error = xfs_attr_rmtval_set_value(args);
>>   		if (error)
>> -			goto out;
>> +			return error;
>> +	}
>>   
>> -	} else if (args->rmtblkno > 0) {
>> +	if (args->rmtblkno > 0) {
>>   		/*
>>   		 * Added a "remote" value, just clear the incomplete flag.
>>   		 */
>>   		error = xfs_attr3_leaf_clearflag(args);
>>   		if (error)
>>   			goto out;
>> -
>> -		 /*
>> -		  * Commit the flag value change and start the next trans in
>> -		  * series.
>> -		  */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> -		if (error)
>> -			goto out;
>>   	}
>>   	retval = error = 0;
>>   
>> diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
>> index fb8bf5b..c710387 100644
>> --- a/fs/xfs/libxfs/xfs_attr.h
>> +++ b/fs/xfs/libxfs/xfs_attr.h
>> @@ -149,6 +149,7 @@ int xfs_attr_get(struct xfs_inode *ip, struct xfs_name *name,
>>   int xfs_attr_set(struct xfs_inode *dp, struct xfs_name *name,
>>   		 unsigned char *value, int valuelen, int flags);
>>   int xfs_attr_set_args(struct xfs_da_args *args);
>> +int xfs_attr_set_later(struct xfs_da_args *args, struct xfs_buf **leaf_bp);
>>   int xfs_attr_remove(struct xfs_inode *dp, struct xfs_name *name, int flags);
>>   int xfs_has_attr(struct xfs_da_args *args);
>>   int xfs_attr_remove_args(struct xfs_da_args *args);
>> -- 
>> 2.7.4
>>

Brian Foster Nov. 12, 2019, 1:37 p.m. UTC | #4

On Wed, Nov 06, 2019 at 06:28:01PM -0700, Allison Collins wrote:
> This patch modifies the attr set routines to be delay ready.
> This means they no longer roll or commit transactions, but instead
> return -EAGAIN to have the calling routine roll and refresh the
> transaction.  In this series, xfs_attr_set_args has become
> xfs_attr_set_later, which uses a state machine to keep track
> of where it was when EAGAIN was returned.  Part of
> xfs_attr_leaf_addname has been factored out into a new helper
> function xfs_attr_leaf_try_add to allow transaction cycling between
> the two routines, and the flipflags logic has been removed since we
> can simply cancel the transaction upon error.  xfs_attr_set_args
> consists of a simple loop to refresh the transaction until the
> operation is completed.
> 
> Signed-off-by: Allison Collins <[e-mail]>
> ---

Pretty much the same fundamental comments here as for the remove patch.
Can we break things down further to avoid much of the jumping into if
branch behavior and such? I also wonder if it might be possible to split
this one up into smaller patches to fix up the leaf/node paths
separately, but additional prep refactoring might simplify things enough
on its own to make that unnecessary..

Brian

>  fs/xfs/libxfs/xfs_attr.c | 435 +++++++++++++++++++++++------------------------
>  fs/xfs/libxfs/xfs_attr.h |   1 +
>  2 files changed, 211 insertions(+), 225 deletions(-)
> 
> diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
> index 38d5c5c..97e5ae0 100644
> --- a/fs/xfs/libxfs/xfs_attr.c
> +++ b/fs/xfs/libxfs/xfs_attr.c
> @@ -58,6 +58,7 @@ STATIC int xfs_attr_node_hasname(xfs_da_args_t *args,
>  				 struct xfs_da_state **state);
>  STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
>  STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
> +STATIC int xfs_attr_leaf_try_add(struct xfs_da_args *args, struct xfs_buf *bp);
>  
>  
>  STATIC int
> @@ -250,9 +251,79 @@ int
>  xfs_attr_set_args(
>  	struct xfs_da_args	*args)
>  {
> +	int			error = 0;
> +	int			err2 = 0;
> +	struct xfs_buf		*leaf_bp = NULL;
> +
> +	do {
> +		error = xfs_attr_set_later(args, &leaf_bp);
> +		if (error && error != -EAGAIN)
> +			goto out;
> +
> +		xfs_trans_log_inode(args->trans, args->dp,
> +				    XFS_ILOG_CORE | XFS_ILOG_ADATA);
> +
> +		err2 = xfs_trans_roll(&args->trans);
> +		if (err2) {
> +			error = err2;
> +			goto out;
> +		}
> +
> +		/* Rejoin inode and leaf if needed */
> +		xfs_trans_ijoin(args->trans, args->dp, 0);
> +		if (leaf_bp) {
> +			xfs_trans_bjoin(args->trans, leaf_bp);
> +			xfs_trans_bhold(args->trans, leaf_bp);
> +		}
> +
> +	} while (error == -EAGAIN);
> +
> +out:
> +	return error;
> +}
> +
> +/*
> + * Set the attribute specified in @args.
> + * This routine is meant to function as a delayed operation, and may return
> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
> + * to handle this, and recall the function until a successful error code is
> + * returned.
> + */
> +int
> +xfs_attr_set_later(
> +	struct xfs_da_args	*args,
> +	struct xfs_buf          **leaf_bp)
> +{
>  	struct xfs_inode	*dp = args->dp;
> -	struct xfs_buf          *leaf_bp = NULL;
> -	int			error, error2 = 0;;
> +	int			error = 0;
> +	int			sf_size;
> +
> +	/* State machine switch */
> +	switch (args->dc.dc_state) {
> +	case XFS_DC_SF_TO_LEAF:
> +		goto sf_to_leaf;
> +	case XFS_DC_ALLOC_LEAF:
> +	case XFS_DC_FOUND_LBLK:
> +		goto leaf;
> +	case XFS_DC_FOUND_NBLK:
> +	case XFS_DC_ALLOC_NODE:
> +	case XFS_DC_LEAF_TO_NODE:
> +		goto node;
> +	default:
> +		break;
> +	}
> +
> +	/*
> +	 * New inodes may not have an attribute fork yet. So set the attribute
> +	 * fork appropriately
> +	 */
> +	if (XFS_IFORK_Q((args->dp)) == 0) {
> +		sf_size = sizeof(struct xfs_attr_sf_hdr) +
> +		     XFS_ATTR_SF_ENTSIZE_BYNAME(args->name.len, args->valuelen);
> +		xfs_bmap_set_attrforkoff(args->dp, sf_size, NULL);
> +		args->dp->i_afp = kmem_zone_zalloc(xfs_ifork_zone, 0);
> +		args->dp->i_afp->if_flags = XFS_IFEXTENTS;
> +	}
>  
>  	/*
>  	 * If the attribute list is non-existent or a shortform list,
> @@ -272,21 +343,14 @@ xfs_attr_set_args(
>  		 * Try to add the attr to the attribute list in the inode.
>  		 */
>  		error = xfs_attr_try_sf_addname(dp, args);
> -		if (error != -ENOSPC) {
> -			if (dp->i_mount->m_flags & XFS_MOUNT_WSYNC)
> -				xfs_trans_set_sync(args->trans);
> -
> -			error2 = xfs_trans_commit(args->trans);
> -			args->trans = NULL;
> -			return error ? error : error2;
> -		}
> -
> +		if (error != -ENOSPC)
> +			return error;
>  
>  		/*
>  		 * It won't fit in the shortform, transform to a leaf block.
>  		 * GROT: another possible req'mt for a double-split btree op.
>  		 */
> -		error = xfs_attr_shortform_to_leaf(args, &leaf_bp);
> +		error = xfs_attr_shortform_to_leaf(args, leaf_bp);
>  		if (error)
>  			return error;
>  
> @@ -294,43 +358,42 @@ xfs_attr_set_args(
>  		 * Prevent the leaf buffer from being unlocked so that a
>  		 * concurrent AIL push cannot grab the half-baked leaf
>  		 * buffer and run into problems with the write verifier.
> -		 * Once we're done rolling the transaction we can release
> -		 * the hold and add the attr to the leaf.
>  		 */
> -		xfs_trans_bhold(args->trans, leaf_bp);
> -		error = xfs_defer_finish(&args->trans);
> -		xfs_trans_bhold_release(args->trans, leaf_bp);
> -		if (error) {
> -			xfs_trans_brelse(args->trans, leaf_bp);
> -			return error;
> -		}
> +
> +		xfs_trans_bhold(args->trans, *leaf_bp);
> +		args->dc.dc_state = XFS_DC_SF_TO_LEAF;
> +		return -EAGAIN;
> +	}
> +sf_to_leaf:
> +
> +	/*
> +	 * After a shortform to leaf conversion, we need to hold the leaf and
> +	 * cylce out the transaction.  When we get back, we need to release
> +	 * the leaf.
> +	 */
> +	if (*leaf_bp != NULL) {
> +		xfs_trans_brelse(args->trans, *leaf_bp);
> +		*leaf_bp = NULL;
>  	}
>  
>  	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
> +		error = xfs_attr_leaf_try_add(args, *leaf_bp);
> +		if (error == -ENOSPC)
> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
> +		else if (error)
> +			return error;
> +		else
> +			args->dc.dc_state = XFS_DC_FOUND_LBLK;
> +		return -EAGAIN;
> +leaf:
>  		error = xfs_attr_leaf_addname(args);
>  		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);
> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
> +			return -EAGAIN;
>  		}
>  	} else {
> +		args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
> +node:
>  		error = xfs_attr_node_addname(args);
>  	}
>  	return error;
> @@ -764,27 +827,26 @@ xfs_attr_leaf_try_add(
>   *
>   * 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).
> + *
> + * This routine is meant to function as a delayed operation, and may return
> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
> + * to handle this, and recall the function until a successful error code is
> + * returned.
>   */
>  STATIC int
>  xfs_attr_leaf_addname(struct xfs_da_args	*args)
>  {
> -	int			error, forkoff;
> -	struct xfs_buf		*bp = NULL;
> +	int			error, nmap;
>  	struct xfs_inode	*dp = args->dp;
> +	struct xfs_bmbt_irec	*map = &args->dc.map;
>  
> -	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
> -	 * later routines can manage their own transactions.
> -	 */
> -	error = xfs_trans_roll_inode(&args->trans, dp);
> -	if (error)
> -		return error;
> +	/* State machine switch */
> +	switch (args->dc.dc_state) {
> +	case XFS_DC_ALLOC_LEAF:
> +		goto alloc_leaf;
> +	default:
> +		break;
> +	}
>  
>  	/*
>  	 * If there was an out-of-line value, allocate the blocks we
> @@ -793,90 +855,58 @@ xfs_attr_leaf_addname(struct xfs_da_args	*args)
>  	 * maximum size of a transaction and/or hit a deadlock.
>  	 */
>  	if (args->rmtblkno > 0) {
> -		error = xfs_attr_rmtval_set(args);
> -		if (error)
> -			return error;
> -	}
>  
> -	/*
> -	 * If this is an atomic rename operation, we must "flip" the
> -	 * incomplete flags on the "new" and "old" attribute/value pairs
> -	 * so that one disappears and one appears atomically.  Then we
> -	 * must remove the "old" attribute/value pair.
> -	 */
> -	if (args->op_flags & XFS_DA_OP_RENAME) {
> -		/*
> -		 * In a separate transaction, set the incomplete flag on the
> -		 * "old" attr and clear the incomplete flag on the "new" attr.
> -		 */
> -		error = xfs_attr3_leaf_flipflags(args);
> -		if (error)
> -			return error;
> -		/*
> -		 * Commit the flag value change and start the next trans in
> -		 * series.
> -		 */
> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
> -		if (error)
> -			return error;
> +		/* Open coded xfs_attr_rmtval_set without trans handling */
>  
> -		/*
> -		 * Dismantle the "old" attribute/value pair by removing
> -		 * a "remote" value (if it exists).
> -		 */
> -		args->index = args->index2;
> -		args->blkno = args->blkno2;
> -		args->rmtblkno = args->rmtblkno2;
> -		args->rmtblkcnt = args->rmtblkcnt2;
> -		args->rmtvaluelen = args->rmtvaluelen2;
> -		if (args->rmtblkno) {
> -			error = xfs_attr_rmtval_remove(args);
> -			if (error)
> -				return error;
> -		}
> +		args->dc.lfileoff = 0;
> +		args->dc.lblkno = 0;
> +		args->dc.blkcnt = 0;
> +		args->rmtblkcnt = 0;
> +		args->rmtblkno = 0;
> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>  
> -		/*
> -		 * Read in the block containing the "old" attr, then
> -		 * remove the "old" attr from that block (neat, huh!)
> -		 */
> -		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno,
> -					   XFS_DABUF_MAP_NOMAPPING, &bp);
> +		error = xfs_attr_rmt_find_hole(args);
>  		if (error)
>  			return error;
>  
> -		xfs_attr3_leaf_remove(bp, args);
> +		args->dc.blkcnt = args->rmtblkcnt;
> +		args->dc.lblkno = args->rmtblkno;
>  
>  		/*
> -		 * If the result is small enough, shrink it all into the inode.
> +		 * Roll through the "value", allocating blocks on disk as
> +		 * required.
>  		 */
> -		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
> -			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
> -			/* bp is gone due to xfs_da_shrink_inode */
> -			if (error)
> -				return error;
> -			error = xfs_defer_finish(&args->trans);
> +alloc_leaf:
> +		while (args->dc.blkcnt > 0) {
> +			nmap = 1;
> +			error = xfs_bmapi_write(args->trans, dp,
> +				  (xfs_fileoff_t)args->dc.lblkno,
> +				  args->dc.blkcnt, XFS_BMAPI_ATTRFORK,
> +				  args->total, map, &nmap);
>  			if (error)
>  				return error;
> -		}
> +			ASSERT(nmap == 1);
> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
> +			       (map->br_startblock != HOLESTARTBLOCK));
>  
> -		/*
> -		 * Commit the remove and start the next trans in series.
> -		 */
> -		error = xfs_trans_roll_inode(&args->trans, dp);
> +			/* roll attribute extent map forwards */
> +			args->dc.lblkno += map->br_blockcount;
> +			args->dc.blkcnt -= map->br_blockcount;
>  
> -	} else if (args->rmtblkno > 0) {
> -		/*
> -		 * Added a "remote" value, just clear the incomplete flag.
> -		 */
> -		error = xfs_attr3_leaf_clearflag(args);
> +			args->dc.dc_state = XFS_DC_ALLOC_LEAF;
> +			return -EAGAIN;
> +		}
> +
> +		error = xfs_attr_rmtval_set_value(args);
>  		if (error)
>  			return error;
> +	}
>  
> +	if (args->rmtblkno > 0) {
>  		/*
> -		 * Commit the flag value change and start the next trans in
> -		 * series.
> +		 * Added a "remote" value, just clear the incomplete flag.
>  		 */
> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
> +		error = xfs_attr3_leaf_clearflag(args);
>  	}
>  	return error;
>  }
> @@ -1017,16 +1047,23 @@ xfs_attr_node_hasname(
>   *
>   * "Remote" attribute values confuse the issue and atomic rename operations
>   * add a whole extra layer of confusion on top of that.
> + *
> + * This routine is meant to function as a delayed operation, and may return
> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
> + * to handle this, and recall the function until a successful error code is
> + *returned.
>   */
>  STATIC int
>  xfs_attr_node_addname(
>  	struct xfs_da_args	*args)
>  {
> -	struct xfs_da_state	*state;
> +	struct xfs_da_state	*state = NULL;
>  	struct xfs_da_state_blk	*blk;
>  	struct xfs_inode	*dp;
> -	struct xfs_mount	*mp;
> -	int			retval, error;
> +	int			retval = 0;
> +	int			error = 0;
> +	int			nmap;
> +	struct xfs_bmbt_irec    *map = &args->dc.map;
>  
>  	trace_xfs_attr_node_addname(args);
>  
> @@ -1034,8 +1071,17 @@ xfs_attr_node_addname(
>  	 * Fill in bucket of arguments/results/context to carry around.
>  	 */
>  	dp = args->dp;
> -	mp = dp->i_mount;
> -restart:
> +
> +	/* State machine switch */
> +	switch (args->dc.dc_state) {
> +	case XFS_DC_FOUND_NBLK:
> +		goto found_nblk;
> +	case XFS_DC_ALLOC_NODE:
> +		goto alloc_node;
> +	default:
> +		break;
> +	}
> +
>  	/*
>  	 * Search to see if name already exists, and get back a pointer
>  	 * to where it should go.
> @@ -1085,19 +1131,12 @@ xfs_attr_node_addname(
>  			error = xfs_attr3_leaf_to_node(args);
>  			if (error)
>  				goto out;
> -			error = xfs_defer_finish(&args->trans);
> -			if (error)
> -				goto out;
>  
>  			/*
> -			 * Commit the node conversion and start the next
> -			 * trans in the chain.
> +			 * Restart routine from the top.  No need to set  the
> +			 * state
>  			 */
> -			error = xfs_trans_roll_inode(&args->trans, dp);
> -			if (error)
> -				goto out;
> -
> -			goto restart;
> +			return -EAGAIN;
>  		}
>  
>  		/*
> @@ -1109,9 +1148,6 @@ xfs_attr_node_addname(
>  		error = xfs_da3_split(state);
>  		if (error)
>  			goto out;
> -		error = xfs_defer_finish(&args->trans);
> -		if (error)
> -			goto out;
>  	} else {
>  		/*
>  		 * Addition succeeded, update Btree hashvals.
> @@ -1126,13 +1162,9 @@ xfs_attr_node_addname(
>  	xfs_da_state_free(state);
>  	state = NULL;
>  
> -	/*
> -	 * Commit the leaf addition or btree split and start the next
> -	 * trans in the chain.
> -	 */
> -	error = xfs_trans_roll_inode(&args->trans, dp);
> -	if (error)
> -		goto out;
> +	args->dc.dc_state = XFS_DC_FOUND_NBLK;
> +	return -EAGAIN;
> +found_nblk:
>  
>  	/*
>  	 * If there was an out-of-line value, allocate the blocks we
> @@ -1141,104 +1173,57 @@ xfs_attr_node_addname(
>  	 * maximum size of a transaction and/or hit a deadlock.
>  	 */
>  	if (args->rmtblkno > 0) {
> -		error = xfs_attr_rmtval_set(args);
> -		if (error)
> -			return error;
> -	}
> +		/* Open coded xfs_attr_rmtval_set without trans handling */
> +		args->dc.lblkno = 0;
> +		args->dc.lfileoff = 0;
> +		args->dc.blkcnt = 0;
> +		args->rmtblkcnt = 0;
> +		args->rmtblkno = 0;
> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>  
> -	/*
> -	 * If this is an atomic rename operation, we must "flip" the
> -	 * incomplete flags on the "new" and "old" attribute/value pairs
> -	 * so that one disappears and one appears atomically.  Then we
> -	 * must remove the "old" attribute/value pair.
> -	 */
> -	if (args->op_flags & XFS_DA_OP_RENAME) {
> -		/*
> -		 * In a separate transaction, set the incomplete flag on the
> -		 * "old" attr and clear the incomplete flag on the "new" attr.
> -		 */
> -		error = xfs_attr3_leaf_flipflags(args);
> -		if (error)
> -			goto out;
> -		/*
> -		 * Commit the flag value change and start the next trans in
> -		 * series
> -		 */
> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
> +		error = xfs_attr_rmt_find_hole(args);
>  		if (error)
> -			goto out;
> +			return error;
>  
> +		args->dc.blkcnt = args->rmtblkcnt;
> +		args->dc.lblkno = args->rmtblkno;
>  		/*
> -		 * Dismantle the "old" attribute/value pair by removing
> -		 * a "remote" value (if it exists).
> +		 * Roll through the "value", allocating blocks on disk as
> +		 * required.
>  		 */
> -		args->index = args->index2;
> -		args->blkno = args->blkno2;
> -		args->rmtblkno = args->rmtblkno2;
> -		args->rmtblkcnt = args->rmtblkcnt2;
> -		args->rmtvaluelen = args->rmtvaluelen2;
> -		if (args->rmtblkno) {
> -			error = xfs_attr_rmtval_remove(args);
> +alloc_node:
> +		while (args->dc.blkcnt > 0) {
> +			nmap = 1;
> +			error = xfs_bmapi_write(args->trans, dp,
> +				(xfs_fileoff_t)args->dc.lblkno, args->dc.blkcnt,
> +				XFS_BMAPI_ATTRFORK, args->total, map, &nmap);
>  			if (error)
>  				return error;
> -		}
>  
> -		/*
> -		 * Re-find the "old" attribute entry after any split ops.
> -		 * The INCOMPLETE flag means that we will find the "old"
> -		 * attr, not the "new" one.
> -		 */
> -		args->name.type |= XFS_ATTR_INCOMPLETE;
> -		state = xfs_da_state_alloc();
> -		state->args = args;
> -		state->mp = mp;
> -		state->inleaf = 0;
> -		error = xfs_da3_node_lookup_int(state, &retval);
> -		if (error)
> -			goto out;
> +			ASSERT(nmap == 1);
> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
> +			       (map->br_startblock != HOLESTARTBLOCK));
>  
> -		/*
> -		 * Remove the name and update the hashvals in the tree.
> -		 */
> -		blk = &state->path.blk[ state->path.active-1 ];
> -		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
> -		error = xfs_attr3_leaf_remove(blk->bp, args);
> -		xfs_da3_fixhashpath(state, &state->path);
> +			/* roll attribute extent map forwards */
> +			args->dc.lblkno += map->br_blockcount;
> +			args->dc.blkcnt -= map->br_blockcount;
>  
> -		/*
> -		 * Check to see if the tree needs to be collapsed.
> -		 */
> -		if (retval && (state->path.active > 1)) {
> -			error = xfs_da3_join(state);
> -			if (error)
> -				goto out;
> -			error = xfs_defer_finish(&args->trans);
> -			if (error)
> -				goto out;
> +			args->dc.dc_state = XFS_DC_ALLOC_NODE;
> +			return -EAGAIN;
>  		}
>  
> -		/*
> -		 * Commit and start the next trans in the chain.
> -		 */
> -		error = xfs_trans_roll_inode(&args->trans, dp);
> +		error = xfs_attr_rmtval_set_value(args);
>  		if (error)
> -			goto out;
> +			return error;
> +	}
>  
> -	} else if (args->rmtblkno > 0) {
> +	if (args->rmtblkno > 0) {
>  		/*
>  		 * Added a "remote" value, just clear the incomplete flag.
>  		 */
>  		error = xfs_attr3_leaf_clearflag(args);
>  		if (error)
>  			goto out;
> -
> -		 /*
> -		  * Commit the flag value change and start the next trans in
> -		  * series.
> -		  */
> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
> -		if (error)
> -			goto out;
>  	}
>  	retval = error = 0;
>  
> diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
> index fb8bf5b..c710387 100644
> --- a/fs/xfs/libxfs/xfs_attr.h
> +++ b/fs/xfs/libxfs/xfs_attr.h
> @@ -149,6 +149,7 @@ int xfs_attr_get(struct xfs_inode *ip, struct xfs_name *name,
>  int xfs_attr_set(struct xfs_inode *dp, struct xfs_name *name,
>  		 unsigned char *value, int valuelen, int flags);
>  int xfs_attr_set_args(struct xfs_da_args *args);
> +int xfs_attr_set_later(struct xfs_da_args *args, struct xfs_buf **leaf_bp);
>  int xfs_attr_remove(struct xfs_inode *dp, struct xfs_name *name, int flags);
>  int xfs_has_attr(struct xfs_da_args *args);
>  int xfs_attr_remove_args(struct xfs_da_args *args);
> -- 
> 2.7.4
>

Allison Henderson Nov. 13, 2019, 4:57 a.m. UTC | #5

On 11/12/19 6:37 AM, Brian Foster wrote:
> On Wed, Nov 06, 2019 at 06:28:01PM -0700, Allison Collins wrote:
>> This patch modifies the attr set routines to be delay ready.
>> This means they no longer roll or commit transactions, but instead
>> return -EAGAIN to have the calling routine roll and refresh the
>> transaction.  In this series, xfs_attr_set_args has become
>> xfs_attr_set_later, which uses a state machine to keep track
>> of where it was when EAGAIN was returned.  Part of
>> xfs_attr_leaf_addname has been factored out into a new helper
>> function xfs_attr_leaf_try_add to allow transaction cycling between
>> the two routines, and the flipflags logic has been removed since we
>> can simply cancel the transaction upon error.  xfs_attr_set_args
>> consists of a simple loop to refresh the transaction until the
>> operation is completed.
>>
>> Signed-off-by: Allison Collins <[e-mail]>
>> ---
> 
> Pretty much the same fundamental comments here as for the remove patch.
> Can we break things down further to avoid much of the jumping into if
> branch behavior and such? I also wonder if it might be possible to split
> this one up into smaller patches to fix up the leaf/node paths
> separately, but additional prep refactoring might simplify things enough
> on its own to make that unnecessary..
> 
> Brian

Sure, I'll see if I can get the branching out of the if statements. 
Maybe I can break this up into smaller patches too.  Thanks again for 
the reviews!

Allison

> 
>>   fs/xfs/libxfs/xfs_attr.c | 435 +++++++++++++++++++++++------------------------
>>   fs/xfs/libxfs/xfs_attr.h |   1 +
>>   2 files changed, 211 insertions(+), 225 deletions(-)
>>
>> diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
>> index 38d5c5c..97e5ae0 100644
>> --- a/fs/xfs/libxfs/xfs_attr.c
>> +++ b/fs/xfs/libxfs/xfs_attr.c
>> @@ -58,6 +58,7 @@ STATIC int xfs_attr_node_hasname(xfs_da_args_t *args,
>>   				 struct xfs_da_state **state);
>>   STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
>>   STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
>> +STATIC int xfs_attr_leaf_try_add(struct xfs_da_args *args, struct xfs_buf *bp);
>>   
>>   
>>   STATIC int
>> @@ -250,9 +251,79 @@ int
>>   xfs_attr_set_args(
>>   	struct xfs_da_args	*args)
>>   {
>> +	int			error = 0;
>> +	int			err2 = 0;
>> +	struct xfs_buf		*leaf_bp = NULL;
>> +
>> +	do {
>> +		error = xfs_attr_set_later(args, &leaf_bp);
>> +		if (error && error != -EAGAIN)
>> +			goto out;
>> +
>> +		xfs_trans_log_inode(args->trans, args->dp,
>> +				    XFS_ILOG_CORE | XFS_ILOG_ADATA);
>> +
>> +		err2 = xfs_trans_roll(&args->trans);
>> +		if (err2) {
>> +			error = err2;
>> +			goto out;
>> +		}
>> +
>> +		/* Rejoin inode and leaf if needed */
>> +		xfs_trans_ijoin(args->trans, args->dp, 0);
>> +		if (leaf_bp) {
>> +			xfs_trans_bjoin(args->trans, leaf_bp);
>> +			xfs_trans_bhold(args->trans, leaf_bp);
>> +		}
>> +
>> +	} while (error == -EAGAIN);
>> +
>> +out:
>> +	return error;
>> +}
>> +
>> +/*
>> + * Set the attribute specified in @args.
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + * returned.
>> + */
>> +int
>> +xfs_attr_set_later(
>> +	struct xfs_da_args	*args,
>> +	struct xfs_buf          **leaf_bp)
>> +{
>>   	struct xfs_inode	*dp = args->dp;
>> -	struct xfs_buf          *leaf_bp = NULL;
>> -	int			error, error2 = 0;;
>> +	int			error = 0;
>> +	int			sf_size;
>> +
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_SF_TO_LEAF:
>> +		goto sf_to_leaf;
>> +	case XFS_DC_ALLOC_LEAF:
>> +	case XFS_DC_FOUND_LBLK:
>> +		goto leaf;
>> +	case XFS_DC_FOUND_NBLK:
>> +	case XFS_DC_ALLOC_NODE:
>> +	case XFS_DC_LEAF_TO_NODE:
>> +		goto node;
>> +	default:
>> +		break;
>> +	}
>> +
>> +	/*
>> +	 * New inodes may not have an attribute fork yet. So set the attribute
>> +	 * fork appropriately
>> +	 */
>> +	if (XFS_IFORK_Q((args->dp)) == 0) {
>> +		sf_size = sizeof(struct xfs_attr_sf_hdr) +
>> +		     XFS_ATTR_SF_ENTSIZE_BYNAME(args->name.len, args->valuelen);
>> +		xfs_bmap_set_attrforkoff(args->dp, sf_size, NULL);
>> +		args->dp->i_afp = kmem_zone_zalloc(xfs_ifork_zone, 0);
>> +		args->dp->i_afp->if_flags = XFS_IFEXTENTS;
>> +	}
>>   
>>   	/*
>>   	 * If the attribute list is non-existent or a shortform list,
>> @@ -272,21 +343,14 @@ xfs_attr_set_args(
>>   		 * Try to add the attr to the attribute list in the inode.
>>   		 */
>>   		error = xfs_attr_try_sf_addname(dp, args);
>> -		if (error != -ENOSPC) {
>> -			if (dp->i_mount->m_flags & XFS_MOUNT_WSYNC)
>> -				xfs_trans_set_sync(args->trans);
>> -
>> -			error2 = xfs_trans_commit(args->trans);
>> -			args->trans = NULL;
>> -			return error ? error : error2;
>> -		}
>> -
>> +		if (error != -ENOSPC)
>> +			return error;
>>   
>>   		/*
>>   		 * It won't fit in the shortform, transform to a leaf block.
>>   		 * GROT: another possible req'mt for a double-split btree op.
>>   		 */
>> -		error = xfs_attr_shortform_to_leaf(args, &leaf_bp);
>> +		error = xfs_attr_shortform_to_leaf(args, leaf_bp);
>>   		if (error)
>>   			return error;
>>   
>> @@ -294,43 +358,42 @@ xfs_attr_set_args(
>>   		 * Prevent the leaf buffer from being unlocked so that a
>>   		 * concurrent AIL push cannot grab the half-baked leaf
>>   		 * buffer and run into problems with the write verifier.
>> -		 * Once we're done rolling the transaction we can release
>> -		 * the hold and add the attr to the leaf.
>>   		 */
>> -		xfs_trans_bhold(args->trans, leaf_bp);
>> -		error = xfs_defer_finish(&args->trans);
>> -		xfs_trans_bhold_release(args->trans, leaf_bp);
>> -		if (error) {
>> -			xfs_trans_brelse(args->trans, leaf_bp);
>> -			return error;
>> -		}
>> +
>> +		xfs_trans_bhold(args->trans, *leaf_bp);
>> +		args->dc.dc_state = XFS_DC_SF_TO_LEAF;
>> +		return -EAGAIN;
>> +	}
>> +sf_to_leaf:
>> +
>> +	/*
>> +	 * After a shortform to leaf conversion, we need to hold the leaf and
>> +	 * cylce out the transaction.  When we get back, we need to release
>> +	 * the leaf.
>> +	 */
>> +	if (*leaf_bp != NULL) {
>> +		xfs_trans_brelse(args->trans, *leaf_bp);
>> +		*leaf_bp = NULL;
>>   	}
>>   
>>   	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
>> +		error = xfs_attr_leaf_try_add(args, *leaf_bp);
>> +		if (error == -ENOSPC)
>> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +		else if (error)
>> +			return error;
>> +		else
>> +			args->dc.dc_state = XFS_DC_FOUND_LBLK;
>> +		return -EAGAIN;
>> +leaf:
>>   		error = xfs_attr_leaf_addname(args);
>>   		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);
>> +			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +			return -EAGAIN;
>>   		}
>>   	} else {
>> +		args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
>> +node:
>>   		error = xfs_attr_node_addname(args);
>>   	}
>>   	return error;
>> @@ -764,27 +827,26 @@ xfs_attr_leaf_try_add(
>>    *
>>    * 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).
>> + *
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + * returned.
>>    */
>>   STATIC int
>>   xfs_attr_leaf_addname(struct xfs_da_args	*args)
>>   {
>> -	int			error, forkoff;
>> -	struct xfs_buf		*bp = NULL;
>> +	int			error, nmap;
>>   	struct xfs_inode	*dp = args->dp;
>> +	struct xfs_bmbt_irec	*map = &args->dc.map;
>>   
>> -	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
>> -	 * later routines can manage their own transactions.
>> -	 */
>> -	error = xfs_trans_roll_inode(&args->trans, dp);
>> -	if (error)
>> -		return error;
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_ALLOC_LEAF:
>> +		goto alloc_leaf;
>> +	default:
>> +		break;
>> +	}
>>   
>>   	/*
>>   	 * If there was an out-of-line value, allocate the blocks we
>> @@ -793,90 +855,58 @@ xfs_attr_leaf_addname(struct xfs_da_args	*args)
>>   	 * maximum size of a transaction and/or hit a deadlock.
>>   	 */
>>   	if (args->rmtblkno > 0) {
>> -		error = xfs_attr_rmtval_set(args);
>> -		if (error)
>> -			return error;
>> -	}
>>   
>> -	/*
>> -	 * If this is an atomic rename operation, we must "flip" the
>> -	 * incomplete flags on the "new" and "old" attribute/value pairs
>> -	 * so that one disappears and one appears atomically.  Then we
>> -	 * must remove the "old" attribute/value pair.
>> -	 */
>> -	if (args->op_flags & XFS_DA_OP_RENAME) {
>> -		/*
>> -		 * In a separate transaction, set the incomplete flag on the
>> -		 * "old" attr and clear the incomplete flag on the "new" attr.
>> -		 */
>> -		error = xfs_attr3_leaf_flipflags(args);
>> -		if (error)
>> -			return error;
>> -		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> -		if (error)
>> -			return error;
>> +		/* Open coded xfs_attr_rmtval_set without trans handling */
>>   
>> -		/*
>> -		 * Dismantle the "old" attribute/value pair by removing
>> -		 * a "remote" value (if it exists).
>> -		 */
>> -		args->index = args->index2;
>> -		args->blkno = args->blkno2;
>> -		args->rmtblkno = args->rmtblkno2;
>> -		args->rmtblkcnt = args->rmtblkcnt2;
>> -		args->rmtvaluelen = args->rmtvaluelen2;
>> -		if (args->rmtblkno) {
>> -			error = xfs_attr_rmtval_remove(args);
>> -			if (error)
>> -				return error;
>> -		}
>> +		args->dc.lfileoff = 0;
>> +		args->dc.lblkno = 0;
>> +		args->dc.blkcnt = 0;
>> +		args->rmtblkcnt = 0;
>> +		args->rmtblkno = 0;
>> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>>   
>> -		/*
>> -		 * Read in the block containing the "old" attr, then
>> -		 * remove the "old" attr from that block (neat, huh!)
>> -		 */
>> -		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno,
>> -					   XFS_DABUF_MAP_NOMAPPING, &bp);
>> +		error = xfs_attr_rmt_find_hole(args);
>>   		if (error)
>>   			return error;
>>   
>> -		xfs_attr3_leaf_remove(bp, args);
>> +		args->dc.blkcnt = args->rmtblkcnt;
>> +		args->dc.lblkno = args->rmtblkno;
>>   
>>   		/*
>> -		 * If the result is small enough, shrink it all into the inode.
>> +		 * Roll through the "value", allocating blocks on disk as
>> +		 * required.
>>   		 */
>> -		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
>> -			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
>> -			/* bp is gone due to xfs_da_shrink_inode */
>> -			if (error)
>> -				return error;
>> -			error = xfs_defer_finish(&args->trans);
>> +alloc_leaf:
>> +		while (args->dc.blkcnt > 0) {
>> +			nmap = 1;
>> +			error = xfs_bmapi_write(args->trans, dp,
>> +				  (xfs_fileoff_t)args->dc.lblkno,
>> +				  args->dc.blkcnt, XFS_BMAPI_ATTRFORK,
>> +				  args->total, map, &nmap);
>>   			if (error)
>>   				return error;
>> -		}
>> +			ASSERT(nmap == 1);
>> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
>> +			       (map->br_startblock != HOLESTARTBLOCK));
>>   
>> -		/*
>> -		 * Commit the remove and start the next trans in series.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, dp);
>> +			/* roll attribute extent map forwards */
>> +			args->dc.lblkno += map->br_blockcount;
>> +			args->dc.blkcnt -= map->br_blockcount;
>>   
>> -	} else if (args->rmtblkno > 0) {
>> -		/*
>> -		 * Added a "remote" value, just clear the incomplete flag.
>> -		 */
>> -		error = xfs_attr3_leaf_clearflag(args);
>> +			args->dc.dc_state = XFS_DC_ALLOC_LEAF;
>> +			return -EAGAIN;
>> +		}
>> +
>> +		error = xfs_attr_rmtval_set_value(args);
>>   		if (error)
>>   			return error;
>> +	}
>>   
>> +	if (args->rmtblkno > 0) {
>>   		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series.
>> +		 * Added a "remote" value, just clear the incomplete flag.
>>   		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> +		error = xfs_attr3_leaf_clearflag(args);
>>   	}
>>   	return error;
>>   }
>> @@ -1017,16 +1047,23 @@ xfs_attr_node_hasname(
>>    *
>>    * "Remote" attribute values confuse the issue and atomic rename operations
>>    * add a whole extra layer of confusion on top of that.
>> + *
>> + * This routine is meant to function as a delayed operation, and may return
>> + * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
>> + * to handle this, and recall the function until a successful error code is
>> + *returned.
>>    */
>>   STATIC int
>>   xfs_attr_node_addname(
>>   	struct xfs_da_args	*args)
>>   {
>> -	struct xfs_da_state	*state;
>> +	struct xfs_da_state	*state = NULL;
>>   	struct xfs_da_state_blk	*blk;
>>   	struct xfs_inode	*dp;
>> -	struct xfs_mount	*mp;
>> -	int			retval, error;
>> +	int			retval = 0;
>> +	int			error = 0;
>> +	int			nmap;
>> +	struct xfs_bmbt_irec    *map = &args->dc.map;
>>   
>>   	trace_xfs_attr_node_addname(args);
>>   
>> @@ -1034,8 +1071,17 @@ xfs_attr_node_addname(
>>   	 * Fill in bucket of arguments/results/context to carry around.
>>   	 */
>>   	dp = args->dp;
>> -	mp = dp->i_mount;
>> -restart:
>> +
>> +	/* State machine switch */
>> +	switch (args->dc.dc_state) {
>> +	case XFS_DC_FOUND_NBLK:
>> +		goto found_nblk;
>> +	case XFS_DC_ALLOC_NODE:
>> +		goto alloc_node;
>> +	default:
>> +		break;
>> +	}
>> +
>>   	/*
>>   	 * Search to see if name already exists, and get back a pointer
>>   	 * to where it should go.
>> @@ -1085,19 +1131,12 @@ xfs_attr_node_addname(
>>   			error = xfs_attr3_leaf_to_node(args);
>>   			if (error)
>>   				goto out;
>> -			error = xfs_defer_finish(&args->trans);
>> -			if (error)
>> -				goto out;
>>   
>>   			/*
>> -			 * Commit the node conversion and start the next
>> -			 * trans in the chain.
>> +			 * Restart routine from the top.  No need to set  the
>> +			 * state
>>   			 */
>> -			error = xfs_trans_roll_inode(&args->trans, dp);
>> -			if (error)
>> -				goto out;
>> -
>> -			goto restart;
>> +			return -EAGAIN;
>>   		}
>>   
>>   		/*
>> @@ -1109,9 +1148,6 @@ xfs_attr_node_addname(
>>   		error = xfs_da3_split(state);
>>   		if (error)
>>   			goto out;
>> -		error = xfs_defer_finish(&args->trans);
>> -		if (error)
>> -			goto out;
>>   	} else {
>>   		/*
>>   		 * Addition succeeded, update Btree hashvals.
>> @@ -1126,13 +1162,9 @@ xfs_attr_node_addname(
>>   	xfs_da_state_free(state);
>>   	state = NULL;
>>   
>> -	/*
>> -	 * Commit the leaf addition or btree split and start the next
>> -	 * trans in the chain.
>> -	 */
>> -	error = xfs_trans_roll_inode(&args->trans, dp);
>> -	if (error)
>> -		goto out;
>> +	args->dc.dc_state = XFS_DC_FOUND_NBLK;
>> +	return -EAGAIN;
>> +found_nblk:
>>   
>>   	/*
>>   	 * If there was an out-of-line value, allocate the blocks we
>> @@ -1141,104 +1173,57 @@ xfs_attr_node_addname(
>>   	 * maximum size of a transaction and/or hit a deadlock.
>>   	 */
>>   	if (args->rmtblkno > 0) {
>> -		error = xfs_attr_rmtval_set(args);
>> -		if (error)
>> -			return error;
>> -	}
>> +		/* Open coded xfs_attr_rmtval_set without trans handling */
>> +		args->dc.lblkno = 0;
>> +		args->dc.lfileoff = 0;
>> +		args->dc.blkcnt = 0;
>> +		args->rmtblkcnt = 0;
>> +		args->rmtblkno = 0;
>> +		memset(map, 0, sizeof(struct xfs_bmbt_irec));
>>   
>> -	/*
>> -	 * If this is an atomic rename operation, we must "flip" the
>> -	 * incomplete flags on the "new" and "old" attribute/value pairs
>> -	 * so that one disappears and one appears atomically.  Then we
>> -	 * must remove the "old" attribute/value pair.
>> -	 */
>> -	if (args->op_flags & XFS_DA_OP_RENAME) {
>> -		/*
>> -		 * In a separate transaction, set the incomplete flag on the
>> -		 * "old" attr and clear the incomplete flag on the "new" attr.
>> -		 */
>> -		error = xfs_attr3_leaf_flipflags(args);
>> -		if (error)
>> -			goto out;
>> -		/*
>> -		 * Commit the flag value change and start the next trans in
>> -		 * series
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> +		error = xfs_attr_rmt_find_hole(args);
>>   		if (error)
>> -			goto out;
>> +			return error;
>>   
>> +		args->dc.blkcnt = args->rmtblkcnt;
>> +		args->dc.lblkno = args->rmtblkno;
>>   		/*
>> -		 * Dismantle the "old" attribute/value pair by removing
>> -		 * a "remote" value (if it exists).
>> +		 * Roll through the "value", allocating blocks on disk as
>> +		 * required.
>>   		 */
>> -		args->index = args->index2;
>> -		args->blkno = args->blkno2;
>> -		args->rmtblkno = args->rmtblkno2;
>> -		args->rmtblkcnt = args->rmtblkcnt2;
>> -		args->rmtvaluelen = args->rmtvaluelen2;
>> -		if (args->rmtblkno) {
>> -			error = xfs_attr_rmtval_remove(args);
>> +alloc_node:
>> +		while (args->dc.blkcnt > 0) {
>> +			nmap = 1;
>> +			error = xfs_bmapi_write(args->trans, dp,
>> +				(xfs_fileoff_t)args->dc.lblkno, args->dc.blkcnt,
>> +				XFS_BMAPI_ATTRFORK, args->total, map, &nmap);
>>   			if (error)
>>   				return error;
>> -		}
>>   
>> -		/*
>> -		 * Re-find the "old" attribute entry after any split ops.
>> -		 * The INCOMPLETE flag means that we will find the "old"
>> -		 * attr, not the "new" one.
>> -		 */
>> -		args->name.type |= XFS_ATTR_INCOMPLETE;
>> -		state = xfs_da_state_alloc();
>> -		state->args = args;
>> -		state->mp = mp;
>> -		state->inleaf = 0;
>> -		error = xfs_da3_node_lookup_int(state, &retval);
>> -		if (error)
>> -			goto out;
>> +			ASSERT(nmap == 1);
>> +			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
>> +			       (map->br_startblock != HOLESTARTBLOCK));
>>   
>> -		/*
>> -		 * Remove the name and update the hashvals in the tree.
>> -		 */
>> -		blk = &state->path.blk[ state->path.active-1 ];
>> -		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
>> -		error = xfs_attr3_leaf_remove(blk->bp, args);
>> -		xfs_da3_fixhashpath(state, &state->path);
>> +			/* roll attribute extent map forwards */
>> +			args->dc.lblkno += map->br_blockcount;
>> +			args->dc.blkcnt -= map->br_blockcount;
>>   
>> -		/*
>> -		 * Check to see if the tree needs to be collapsed.
>> -		 */
>> -		if (retval && (state->path.active > 1)) {
>> -			error = xfs_da3_join(state);
>> -			if (error)
>> -				goto out;
>> -			error = xfs_defer_finish(&args->trans);
>> -			if (error)
>> -				goto out;
>> +			args->dc.dc_state = XFS_DC_ALLOC_NODE;
>> +			return -EAGAIN;
>>   		}
>>   
>> -		/*
>> -		 * Commit and start the next trans in the chain.
>> -		 */
>> -		error = xfs_trans_roll_inode(&args->trans, dp);
>> +		error = xfs_attr_rmtval_set_value(args);
>>   		if (error)
>> -			goto out;
>> +			return error;
>> +	}
>>   
>> -	} else if (args->rmtblkno > 0) {
>> +	if (args->rmtblkno > 0) {
>>   		/*
>>   		 * Added a "remote" value, just clear the incomplete flag.
>>   		 */
>>   		error = xfs_attr3_leaf_clearflag(args);
>>   		if (error)
>>   			goto out;
>> -
>> -		 /*
>> -		  * Commit the flag value change and start the next trans in
>> -		  * series.
>> -		  */
>> -		error = xfs_trans_roll_inode(&args->trans, args->dp);
>> -		if (error)
>> -			goto out;
>>   	}
>>   	retval = error = 0;
>>   
>> diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
>> index fb8bf5b..c710387 100644
>> --- a/fs/xfs/libxfs/xfs_attr.h
>> +++ b/fs/xfs/libxfs/xfs_attr.h
>> @@ -149,6 +149,7 @@ int xfs_attr_get(struct xfs_inode *ip, struct xfs_name *name,
>>   int xfs_attr_set(struct xfs_inode *dp, struct xfs_name *name,
>>   		 unsigned char *value, int valuelen, int flags);
>>   int xfs_attr_set_args(struct xfs_da_args *args);
>> +int xfs_attr_set_later(struct xfs_da_args *args, struct xfs_buf **leaf_bp);
>>   int xfs_attr_remove(struct xfs_inode *dp, struct xfs_name *name, int flags);
>>   int xfs_has_attr(struct xfs_da_args *args);
>>   int xfs_attr_remove_args(struct xfs_da_args *args);
>> -- 
>> 2.7.4
>>
>

Patch

diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
index 38d5c5c..97e5ae0 100644
--- a/fs/xfs/libxfs/xfs_attr.c
+++ b/fs/xfs/libxfs/xfs_attr.c
@@ -58,6 +58,7 @@  STATIC int xfs_attr_node_hasname(xfs_da_args_t *args,
 				 struct xfs_da_state **state);
 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
+STATIC int xfs_attr_leaf_try_add(struct xfs_da_args *args, struct xfs_buf *bp);
 
 
 STATIC int
@@ -250,9 +251,79 @@  int
 xfs_attr_set_args(
 	struct xfs_da_args	*args)
 {
+	int			error = 0;
+	int			err2 = 0;
+	struct xfs_buf		*leaf_bp = NULL;
+
+	do {
+		error = xfs_attr_set_later(args, &leaf_bp);
+		if (error && error != -EAGAIN)
+			goto out;
+
+		xfs_trans_log_inode(args->trans, args->dp,
+				    XFS_ILOG_CORE | XFS_ILOG_ADATA);
+
+		err2 = xfs_trans_roll(&args->trans);
+		if (err2) {
+			error = err2;
+			goto out;
+		}
+
+		/* Rejoin inode and leaf if needed */
+		xfs_trans_ijoin(args->trans, args->dp, 0);
+		if (leaf_bp) {
+			xfs_trans_bjoin(args->trans, leaf_bp);
+			xfs_trans_bhold(args->trans, leaf_bp);
+		}
+
+	} while (error == -EAGAIN);
+
+out:
+	return error;
+}
+
+/*
+ * Set the attribute specified in @args.
+ * This routine is meant to function as a delayed operation, and may return
+ * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
+ * to handle this, and recall the function until a successful error code is
+ * returned.
+ */
+int
+xfs_attr_set_later(
+	struct xfs_da_args	*args,
+	struct xfs_buf          **leaf_bp)
+{
 	struct xfs_inode	*dp = args->dp;
-	struct xfs_buf          *leaf_bp = NULL;
-	int			error, error2 = 0;;
+	int			error = 0;
+	int			sf_size;
+
+	/* State machine switch */
+	switch (args->dc.dc_state) {
+	case XFS_DC_SF_TO_LEAF:
+		goto sf_to_leaf;
+	case XFS_DC_ALLOC_LEAF:
+	case XFS_DC_FOUND_LBLK:
+		goto leaf;
+	case XFS_DC_FOUND_NBLK:
+	case XFS_DC_ALLOC_NODE:
+	case XFS_DC_LEAF_TO_NODE:
+		goto node;
+	default:
+		break;
+	}
+
+	/*
+	 * New inodes may not have an attribute fork yet. So set the attribute
+	 * fork appropriately
+	 */
+	if (XFS_IFORK_Q((args->dp)) == 0) {
+		sf_size = sizeof(struct xfs_attr_sf_hdr) +
+		     XFS_ATTR_SF_ENTSIZE_BYNAME(args->name.len, args->valuelen);
+		xfs_bmap_set_attrforkoff(args->dp, sf_size, NULL);
+		args->dp->i_afp = kmem_zone_zalloc(xfs_ifork_zone, 0);
+		args->dp->i_afp->if_flags = XFS_IFEXTENTS;
+	}
 
 	/*
 	 * If the attribute list is non-existent or a shortform list,
@@ -272,21 +343,14 @@  xfs_attr_set_args(
 		 * Try to add the attr to the attribute list in the inode.
 		 */
 		error = xfs_attr_try_sf_addname(dp, args);
-		if (error != -ENOSPC) {
-			if (dp->i_mount->m_flags & XFS_MOUNT_WSYNC)
-				xfs_trans_set_sync(args->trans);
-
-			error2 = xfs_trans_commit(args->trans);
-			args->trans = NULL;
-			return error ? error : error2;
-		}
-
+		if (error != -ENOSPC)
+			return error;
 
 		/*
 		 * It won't fit in the shortform, transform to a leaf block.
 		 * GROT: another possible req'mt for a double-split btree op.
 		 */
-		error = xfs_attr_shortform_to_leaf(args, &leaf_bp);
+		error = xfs_attr_shortform_to_leaf(args, leaf_bp);
 		if (error)
 			return error;
 
@@ -294,43 +358,42 @@  xfs_attr_set_args(
 		 * Prevent the leaf buffer from being unlocked so that a
 		 * concurrent AIL push cannot grab the half-baked leaf
 		 * buffer and run into problems with the write verifier.
-		 * Once we're done rolling the transaction we can release
-		 * the hold and add the attr to the leaf.
 		 */
-		xfs_trans_bhold(args->trans, leaf_bp);
-		error = xfs_defer_finish(&args->trans);
-		xfs_trans_bhold_release(args->trans, leaf_bp);
-		if (error) {
-			xfs_trans_brelse(args->trans, leaf_bp);
-			return error;
-		}
+
+		xfs_trans_bhold(args->trans, *leaf_bp);
+		args->dc.dc_state = XFS_DC_SF_TO_LEAF;
+		return -EAGAIN;
+	}
+sf_to_leaf:
+
+	/*
+	 * After a shortform to leaf conversion, we need to hold the leaf and
+	 * cylce out the transaction.  When we get back, we need to release
+	 * the leaf.
+	 */
+	if (*leaf_bp != NULL) {
+		xfs_trans_brelse(args->trans, *leaf_bp);
+		*leaf_bp = NULL;
 	}
 
 	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
+		error = xfs_attr_leaf_try_add(args, *leaf_bp);
+		if (error == -ENOSPC)
+			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
+		else if (error)
+			return error;
+		else
+			args->dc.dc_state = XFS_DC_FOUND_LBLK;
+		return -EAGAIN;
+leaf:
 		error = xfs_attr_leaf_addname(args);
 		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);
+			args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
+			return -EAGAIN;
 		}
 	} else {
+		args->dc.dc_state = XFS_DC_LEAF_TO_NODE;
+node:
 		error = xfs_attr_node_addname(args);
 	}
 	return error;
@@ -764,27 +827,26 @@  xfs_attr_leaf_try_add(
  *
  * 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).
+ *
+ * This routine is meant to function as a delayed operation, and may return
+ * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
+ * to handle this, and recall the function until a successful error code is
+ * returned.
  */
 STATIC int
 xfs_attr_leaf_addname(struct xfs_da_args	*args)
 {
-	int			error, forkoff;
-	struct xfs_buf		*bp = NULL;
+	int			error, nmap;
 	struct xfs_inode	*dp = args->dp;
+	struct xfs_bmbt_irec	*map = &args->dc.map;
 
-	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
-	 * later routines can manage their own transactions.
-	 */
-	error = xfs_trans_roll_inode(&args->trans, dp);
-	if (error)
-		return error;
+	/* State machine switch */
+	switch (args->dc.dc_state) {
+	case XFS_DC_ALLOC_LEAF:
+		goto alloc_leaf;
+	default:
+		break;
+	}
 
 	/*
 	 * If there was an out-of-line value, allocate the blocks we
@@ -793,90 +855,58 @@  xfs_attr_leaf_addname(struct xfs_da_args	*args)
 	 * maximum size of a transaction and/or hit a deadlock.
 	 */
 	if (args->rmtblkno > 0) {
-		error = xfs_attr_rmtval_set(args);
-		if (error)
-			return error;
-	}
 
-	/*
-	 * If this is an atomic rename operation, we must "flip" the
-	 * incomplete flags on the "new" and "old" attribute/value pairs
-	 * so that one disappears and one appears atomically.  Then we
-	 * must remove the "old" attribute/value pair.
-	 */
-	if (args->op_flags & XFS_DA_OP_RENAME) {
-		/*
-		 * In a separate transaction, set the incomplete flag on the
-		 * "old" attr and clear the incomplete flag on the "new" attr.
-		 */
-		error = xfs_attr3_leaf_flipflags(args);
-		if (error)
-			return error;
-		/*
-		 * Commit the flag value change and start the next trans in
-		 * series.
-		 */
-		error = xfs_trans_roll_inode(&args->trans, args->dp);
-		if (error)
-			return error;
+		/* Open coded xfs_attr_rmtval_set without trans handling */
 
-		/*
-		 * Dismantle the "old" attribute/value pair by removing
-		 * a "remote" value (if it exists).
-		 */
-		args->index = args->index2;
-		args->blkno = args->blkno2;
-		args->rmtblkno = args->rmtblkno2;
-		args->rmtblkcnt = args->rmtblkcnt2;
-		args->rmtvaluelen = args->rmtvaluelen2;
-		if (args->rmtblkno) {
-			error = xfs_attr_rmtval_remove(args);
-			if (error)
-				return error;
-		}
+		args->dc.lfileoff = 0;
+		args->dc.lblkno = 0;
+		args->dc.blkcnt = 0;
+		args->rmtblkcnt = 0;
+		args->rmtblkno = 0;
+		memset(map, 0, sizeof(struct xfs_bmbt_irec));
 
-		/*
-		 * Read in the block containing the "old" attr, then
-		 * remove the "old" attr from that block (neat, huh!)
-		 */
-		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno,
-					   XFS_DABUF_MAP_NOMAPPING, &bp);
+		error = xfs_attr_rmt_find_hole(args);
 		if (error)
 			return error;
 
-		xfs_attr3_leaf_remove(bp, args);
+		args->dc.blkcnt = args->rmtblkcnt;
+		args->dc.lblkno = args->rmtblkno;
 
 		/*
-		 * If the result is small enough, shrink it all into the inode.
+		 * Roll through the "value", allocating blocks on disk as
+		 * required.
 		 */
-		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
-			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
-			/* bp is gone due to xfs_da_shrink_inode */
-			if (error)
-				return error;
-			error = xfs_defer_finish(&args->trans);
+alloc_leaf:
+		while (args->dc.blkcnt > 0) {
+			nmap = 1;
+			error = xfs_bmapi_write(args->trans, dp,
+				  (xfs_fileoff_t)args->dc.lblkno,
+				  args->dc.blkcnt, XFS_BMAPI_ATTRFORK,
+				  args->total, map, &nmap);
 			if (error)
 				return error;
-		}
+			ASSERT(nmap == 1);
+			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
+			       (map->br_startblock != HOLESTARTBLOCK));
 
-		/*
-		 * Commit the remove and start the next trans in series.
-		 */
-		error = xfs_trans_roll_inode(&args->trans, dp);
+			/* roll attribute extent map forwards */
+			args->dc.lblkno += map->br_blockcount;
+			args->dc.blkcnt -= map->br_blockcount;
 
-	} else if (args->rmtblkno > 0) {
-		/*
-		 * Added a "remote" value, just clear the incomplete flag.
-		 */
-		error = xfs_attr3_leaf_clearflag(args);
+			args->dc.dc_state = XFS_DC_ALLOC_LEAF;
+			return -EAGAIN;
+		}
+
+		error = xfs_attr_rmtval_set_value(args);
 		if (error)
 			return error;
+	}
 
+	if (args->rmtblkno > 0) {
 		/*
-		 * Commit the flag value change and start the next trans in
-		 * series.
+		 * Added a "remote" value, just clear the incomplete flag.
 		 */
-		error = xfs_trans_roll_inode(&args->trans, args->dp);
+		error = xfs_attr3_leaf_clearflag(args);
 	}
 	return error;
 }
@@ -1017,16 +1047,23 @@  xfs_attr_node_hasname(
  *
  * "Remote" attribute values confuse the issue and atomic rename operations
  * add a whole extra layer of confusion on top of that.
+ *
+ * This routine is meant to function as a delayed operation, and may return
+ * -EAGAIN when the transaction needs to be rolled.  Calling functions will need
+ * to handle this, and recall the function until a successful error code is
+ *returned.
  */
 STATIC int
 xfs_attr_node_addname(
 	struct xfs_da_args	*args)
 {
-	struct xfs_da_state	*state;
+	struct xfs_da_state	*state = NULL;
 	struct xfs_da_state_blk	*blk;
 	struct xfs_inode	*dp;
-	struct xfs_mount	*mp;
-	int			retval, error;
+	int			retval = 0;
+	int			error = 0;
+	int			nmap;
+	struct xfs_bmbt_irec    *map = &args->dc.map;
 
 	trace_xfs_attr_node_addname(args);
 
@@ -1034,8 +1071,17 @@  xfs_attr_node_addname(
 	 * Fill in bucket of arguments/results/context to carry around.
 	 */
 	dp = args->dp;
-	mp = dp->i_mount;
-restart:
+
+	/* State machine switch */
+	switch (args->dc.dc_state) {
+	case XFS_DC_FOUND_NBLK:
+		goto found_nblk;
+	case XFS_DC_ALLOC_NODE:
+		goto alloc_node;
+	default:
+		break;
+	}
+
 	/*
 	 * Search to see if name already exists, and get back a pointer
 	 * to where it should go.
@@ -1085,19 +1131,12 @@  xfs_attr_node_addname(
 			error = xfs_attr3_leaf_to_node(args);
 			if (error)
 				goto out;
-			error = xfs_defer_finish(&args->trans);
-			if (error)
-				goto out;
 
 			/*
-			 * Commit the node conversion and start the next
-			 * trans in the chain.
+			 * Restart routine from the top.  No need to set  the
+			 * state
 			 */
-			error = xfs_trans_roll_inode(&args->trans, dp);
-			if (error)
-				goto out;
-
-			goto restart;
+			return -EAGAIN;
 		}
 
 		/*
@@ -1109,9 +1148,6 @@  xfs_attr_node_addname(
 		error = xfs_da3_split(state);
 		if (error)
 			goto out;
-		error = xfs_defer_finish(&args->trans);
-		if (error)
-			goto out;
 	} else {
 		/*
 		 * Addition succeeded, update Btree hashvals.
@@ -1126,13 +1162,9 @@  xfs_attr_node_addname(
 	xfs_da_state_free(state);
 	state = NULL;
 
-	/*
-	 * Commit the leaf addition or btree split and start the next
-	 * trans in the chain.
-	 */
-	error = xfs_trans_roll_inode(&args->trans, dp);
-	if (error)
-		goto out;
+	args->dc.dc_state = XFS_DC_FOUND_NBLK;
+	return -EAGAIN;
+found_nblk:
 
 	/*
 	 * If there was an out-of-line value, allocate the blocks we
@@ -1141,104 +1173,57 @@  xfs_attr_node_addname(
 	 * maximum size of a transaction and/or hit a deadlock.
 	 */
 	if (args->rmtblkno > 0) {
-		error = xfs_attr_rmtval_set(args);
-		if (error)
-			return error;
-	}
+		/* Open coded xfs_attr_rmtval_set without trans handling */
+		args->dc.lblkno = 0;
+		args->dc.lfileoff = 0;
+		args->dc.blkcnt = 0;
+		args->rmtblkcnt = 0;
+		args->rmtblkno = 0;
+		memset(map, 0, sizeof(struct xfs_bmbt_irec));
 
-	/*
-	 * If this is an atomic rename operation, we must "flip" the
-	 * incomplete flags on the "new" and "old" attribute/value pairs
-	 * so that one disappears and one appears atomically.  Then we
-	 * must remove the "old" attribute/value pair.
-	 */
-	if (args->op_flags & XFS_DA_OP_RENAME) {
-		/*
-		 * In a separate transaction, set the incomplete flag on the
-		 * "old" attr and clear the incomplete flag on the "new" attr.
-		 */
-		error = xfs_attr3_leaf_flipflags(args);
-		if (error)
-			goto out;
-		/*
-		 * Commit the flag value change and start the next trans in
-		 * series
-		 */
-		error = xfs_trans_roll_inode(&args->trans, args->dp);
+		error = xfs_attr_rmt_find_hole(args);
 		if (error)
-			goto out;
+			return error;
 
+		args->dc.blkcnt = args->rmtblkcnt;
+		args->dc.lblkno = args->rmtblkno;
 		/*
-		 * Dismantle the "old" attribute/value pair by removing
-		 * a "remote" value (if it exists).
+		 * Roll through the "value", allocating blocks on disk as
+		 * required.
 		 */
-		args->index = args->index2;
-		args->blkno = args->blkno2;
-		args->rmtblkno = args->rmtblkno2;
-		args->rmtblkcnt = args->rmtblkcnt2;
-		args->rmtvaluelen = args->rmtvaluelen2;
-		if (args->rmtblkno) {
-			error = xfs_attr_rmtval_remove(args);
+alloc_node:
+		while (args->dc.blkcnt > 0) {
+			nmap = 1;
+			error = xfs_bmapi_write(args->trans, dp,
+				(xfs_fileoff_t)args->dc.lblkno, args->dc.blkcnt,
+				XFS_BMAPI_ATTRFORK, args->total, map, &nmap);
 			if (error)
 				return error;
-		}
 
-		/*
-		 * Re-find the "old" attribute entry after any split ops.
-		 * The INCOMPLETE flag means that we will find the "old"
-		 * attr, not the "new" one.
-		 */
-		args->name.type |= XFS_ATTR_INCOMPLETE;
-		state = xfs_da_state_alloc();
-		state->args = args;
-		state->mp = mp;
-		state->inleaf = 0;
-		error = xfs_da3_node_lookup_int(state, &retval);
-		if (error)
-			goto out;
+			ASSERT(nmap == 1);
+			ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
+			       (map->br_startblock != HOLESTARTBLOCK));
 
-		/*
-		 * Remove the name and update the hashvals in the tree.
-		 */
-		blk = &state->path.blk[ state->path.active-1 ];
-		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
-		error = xfs_attr3_leaf_remove(blk->bp, args);
-		xfs_da3_fixhashpath(state, &state->path);
+			/* roll attribute extent map forwards */
+			args->dc.lblkno += map->br_blockcount;
+			args->dc.blkcnt -= map->br_blockcount;
 
-		/*
-		 * Check to see if the tree needs to be collapsed.
-		 */
-		if (retval && (state->path.active > 1)) {
-			error = xfs_da3_join(state);
-			if (error)
-				goto out;
-			error = xfs_defer_finish(&args->trans);
-			if (error)
-				goto out;
+			args->dc.dc_state = XFS_DC_ALLOC_NODE;
+			return -EAGAIN;
 		}
 
-		/*
-		 * Commit and start the next trans in the chain.
-		 */
-		error = xfs_trans_roll_inode(&args->trans, dp);
+		error = xfs_attr_rmtval_set_value(args);
 		if (error)
-			goto out;
+			return error;
+	}
 
-	} else if (args->rmtblkno > 0) {
+	if (args->rmtblkno > 0) {
 		/*
 		 * Added a "remote" value, just clear the incomplete flag.
 		 */
 		error = xfs_attr3_leaf_clearflag(args);
 		if (error)
 			goto out;
-
-		 /*
-		  * Commit the flag value change and start the next trans in
-		  * series.
-		  */
-		error = xfs_trans_roll_inode(&args->trans, args->dp);
-		if (error)
-			goto out;
 	}
 	retval = error = 0;
 
diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
index fb8bf5b..c710387 100644
--- a/fs/xfs/libxfs/xfs_attr.h
+++ b/fs/xfs/libxfs/xfs_attr.h
@@ -149,6 +149,7 @@  int xfs_attr_get(struct xfs_inode *ip, struct xfs_name *name,
 int xfs_attr_set(struct xfs_inode *dp, struct xfs_name *name,
 		 unsigned char *value, int valuelen, int flags);
 int xfs_attr_set_args(struct xfs_da_args *args);
+int xfs_attr_set_later(struct xfs_da_args *args, struct xfs_buf **leaf_bp);
 int xfs_attr_remove(struct xfs_inode *dp, struct xfs_name *name, int flags);
 int xfs_has_attr(struct xfs_da_args *args);
 int xfs_attr_remove_args(struct xfs_da_args *args);